blob: 3ebaafb4c5872470f7637602f0b9b23b58c27690 [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{
208 struct file_lock_context *new;
209
Jeff Layton5c1c6692015-04-03 09:04:03 -0400210 if (likely(inode->i_flctx) || type == F_UNLCK)
Jeff Layton4a075e32015-01-16 15:05:54 -0500211 goto out;
212
213 new = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
214 if (!new)
215 goto out;
216
Jeff Layton6109c852015-01-16 15:05:57 -0500217 spin_lock_init(&new->flc_lock);
Jeff Layton4a075e32015-01-16 15:05:54 -0500218 INIT_LIST_HEAD(&new->flc_flock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500219 INIT_LIST_HEAD(&new->flc_posix);
Jeff Layton8634b512015-01-16 15:05:55 -0500220 INIT_LIST_HEAD(&new->flc_lease);
Jeff Layton4a075e32015-01-16 15:05:54 -0500221
222 /*
223 * Assign the pointer if it's not already assigned. If it is, then
224 * free the context we just allocated.
225 */
226 spin_lock(&inode->i_lock);
227 if (likely(!inode->i_flctx)) {
228 inode->i_flctx = new;
229 new = NULL;
230 }
231 spin_unlock(&inode->i_lock);
232
233 if (new)
234 kmem_cache_free(flctx_cache, new);
235out:
236 return inode->i_flctx;
237}
238
239void
240locks_free_lock_context(struct file_lock_context *ctx)
241{
242 if (ctx) {
243 WARN_ON_ONCE(!list_empty(&ctx->flc_flock));
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500244 WARN_ON_ONCE(!list_empty(&ctx->flc_posix));
Jeff Layton8634b512015-01-16 15:05:55 -0500245 WARN_ON_ONCE(!list_empty(&ctx->flc_lease));
Jeff Layton4a075e32015-01-16 15:05:54 -0500246 kmem_cache_free(flctx_cache, ctx);
247 }
248}
249
Miklos Szerediee19cc42011-07-07 13:06:09 +0200250static void locks_init_lock_heads(struct file_lock *fl)
Miklos Szeredia51cb912011-07-06 12:33:55 +0200251{
Jeff Layton139ca042013-06-21 08:58:17 -0400252 INIT_HLIST_NODE(&fl->fl_link);
Jeff Layton6dee60f2015-01-16 15:05:54 -0500253 INIT_LIST_HEAD(&fl->fl_list);
Miklos Szerediee19cc42011-07-07 13:06:09 +0200254 INIT_LIST_HEAD(&fl->fl_block);
255 init_waitqueue_head(&fl->fl_wait);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200256}
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258/* Allocate an empty lock structure. */
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200259struct file_lock *locks_alloc_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200261 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200262
263 if (fl)
Miklos Szerediee19cc42011-07-07 13:06:09 +0200264 locks_init_lock_heads(fl);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200265
266 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267}
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200268EXPORT_SYMBOL_GPL(locks_alloc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500270void locks_release_private(struct file_lock *fl)
Trond Myklebust47831f32006-03-20 13:44:05 -0500271{
272 if (fl->fl_ops) {
273 if (fl->fl_ops->fl_release_private)
274 fl->fl_ops->fl_release_private(fl);
275 fl->fl_ops = NULL;
276 }
Trond Myklebust47831f32006-03-20 13:44:05 -0500277
Kinglong Mee5c97d7b2014-08-22 10:18:43 -0400278 if (fl->fl_lmops) {
Jeff Laytoncae80b32015-04-03 09:04:04 -0400279 if (fl->fl_lmops->lm_put_owner) {
280 fl->fl_lmops->lm_put_owner(fl->fl_owner);
281 fl->fl_owner = NULL;
282 }
Kinglong Mee5c97d7b2014-08-22 10:18:43 -0400283 fl->fl_lmops = NULL;
284 }
Trond Myklebust47831f32006-03-20 13:44:05 -0500285}
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500286EXPORT_SYMBOL_GPL(locks_release_private);
Trond Myklebust47831f32006-03-20 13:44:05 -0500287
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288/* Free a lock which is not in use. */
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400289void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800291 BUG_ON(waitqueue_active(&fl->fl_wait));
Jeff Layton6dee60f2015-01-16 15:05:54 -0500292 BUG_ON(!list_empty(&fl->fl_list));
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800293 BUG_ON(!list_empty(&fl->fl_block));
Jeff Layton139ca042013-06-21 08:58:17 -0400294 BUG_ON(!hlist_unhashed(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
Trond Myklebust47831f32006-03-20 13:44:05 -0500296 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 kmem_cache_free(filelock_cache, fl);
298}
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400299EXPORT_SYMBOL(locks_free_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
Jeff Laytoned9814d2014-08-11 14:20:31 -0400301static void
302locks_dispose_list(struct list_head *dispose)
303{
304 struct file_lock *fl;
305
306 while (!list_empty(dispose)) {
Jeff Layton6dee60f2015-01-16 15:05:54 -0500307 fl = list_first_entry(dispose, struct file_lock, fl_list);
308 list_del_init(&fl->fl_list);
Jeff Laytoned9814d2014-08-11 14:20:31 -0400309 locks_free_lock(fl);
310 }
311}
312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313void locks_init_lock(struct file_lock *fl)
314{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200315 memset(fl, 0, sizeof(struct file_lock));
316 locks_init_lock_heads(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317}
318
319EXPORT_SYMBOL(locks_init_lock);
320
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321/*
322 * Initialize a new lock from an existing file_lock structure.
323 */
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400324void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 new->fl_owner = fl->fl_owner;
327 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500328 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 new->fl_flags = fl->fl_flags;
330 new->fl_type = fl->fl_type;
331 new->fl_start = fl->fl_start;
332 new->fl_end = fl->fl_end;
Kinglong Meef3282962014-08-22 10:18:43 -0400333 new->fl_lmops = fl->fl_lmops;
Trond Myklebust09969052006-03-20 13:44:38 -0500334 new->fl_ops = NULL;
Kinglong Meef3282962014-08-22 10:18:43 -0400335
336 if (fl->fl_lmops) {
337 if (fl->fl_lmops->lm_get_owner)
Jeff Laytoncae80b32015-04-03 09:04:04 -0400338 fl->fl_lmops->lm_get_owner(fl->fl_owner);
Kinglong Meef3282962014-08-22 10:18:43 -0400339 }
Trond Myklebust09969052006-03-20 13:44:38 -0500340}
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400341EXPORT_SYMBOL(locks_copy_conflock);
Trond Myklebust09969052006-03-20 13:44:38 -0500342
343void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
344{
Jeff Layton566709b2014-08-11 14:09:35 -0400345 /* "new" must be a freshly-initialized lock */
346 WARN_ON_ONCE(new->fl_ops);
Trond Myklebust09969052006-03-20 13:44:38 -0500347
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400348 locks_copy_conflock(new, fl);
Kinglong Meef3282962014-08-22 10:18:43 -0400349
Trond Myklebust09969052006-03-20 13:44:38 -0500350 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 new->fl_ops = fl->fl_ops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500352
Kinglong Meef3282962014-08-22 10:18:43 -0400353 if (fl->fl_ops) {
354 if (fl->fl_ops->fl_copy_lock)
355 fl->fl_ops->fl_copy_lock(new, fl);
356 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357}
358
359EXPORT_SYMBOL(locks_copy_lock);
360
361static inline int flock_translate_cmd(int cmd) {
362 if (cmd & LOCK_MAND)
363 return cmd & (LOCK_MAND | LOCK_RW);
364 switch (cmd) {
365 case LOCK_SH:
366 return F_RDLCK;
367 case LOCK_EX:
368 return F_WRLCK;
369 case LOCK_UN:
370 return F_UNLCK;
371 }
372 return -EINVAL;
373}
374
375/* Fill in a file_lock structure with an appropriate FLOCK lock. */
Jeff Layton6e129d002014-09-04 10:25:06 -0400376static struct file_lock *
377flock_make_lock(struct file *filp, unsigned int cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378{
379 struct file_lock *fl;
380 int type = flock_translate_cmd(cmd);
Jeff Layton6e129d002014-09-04 10:25:06 -0400381
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 if (type < 0)
Jeff Layton6e129d002014-09-04 10:25:06 -0400383 return ERR_PTR(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
385 fl = locks_alloc_lock();
386 if (fl == NULL)
Jeff Layton6e129d002014-09-04 10:25:06 -0400387 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389 fl->fl_file = filp;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +0200390 fl->fl_owner = filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 fl->fl_pid = current->tgid;
392 fl->fl_flags = FL_FLOCK;
393 fl->fl_type = type;
394 fl->fl_end = OFFSET_MAX;
395
Jeff Layton6e129d002014-09-04 10:25:06 -0400396 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397}
398
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400399static int assign_type(struct file_lock *fl, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400{
401 switch (type) {
402 case F_RDLCK:
403 case F_WRLCK:
404 case F_UNLCK:
405 fl->fl_type = type;
406 break;
407 default:
408 return -EINVAL;
409 }
410 return 0;
411}
412
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500413static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
414 struct flock64 *l)
415{
416 switch (l->l_whence) {
417 case SEEK_SET:
418 fl->fl_start = 0;
419 break;
420 case SEEK_CUR:
421 fl->fl_start = filp->f_pos;
422 break;
423 case SEEK_END:
424 fl->fl_start = i_size_read(file_inode(filp));
425 break;
426 default:
427 return -EINVAL;
428 }
429 if (l->l_start > OFFSET_MAX - fl->fl_start)
430 return -EOVERFLOW;
431 fl->fl_start += l->l_start;
432 if (fl->fl_start < 0)
433 return -EINVAL;
434
435 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
436 POSIX-2001 defines it. */
437 if (l->l_len > 0) {
438 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
439 return -EOVERFLOW;
440 fl->fl_end = fl->fl_start + l->l_len - 1;
441
442 } else if (l->l_len < 0) {
443 if (fl->fl_start + l->l_len < 0)
444 return -EINVAL;
445 fl->fl_end = fl->fl_start - 1;
446 fl->fl_start += l->l_len;
447 } else
448 fl->fl_end = OFFSET_MAX;
449
450 fl->fl_owner = current->files;
451 fl->fl_pid = current->tgid;
452 fl->fl_file = filp;
453 fl->fl_flags = FL_POSIX;
454 fl->fl_ops = NULL;
455 fl->fl_lmops = NULL;
456
457 return assign_type(fl, l->l_type);
458}
459
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
461 * style lock.
462 */
463static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
464 struct flock *l)
465{
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500466 struct flock64 ll = {
467 .l_type = l->l_type,
468 .l_whence = l->l_whence,
469 .l_start = l->l_start,
470 .l_len = l->l_len,
471 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500473 return flock64_to_posix_lock(filp, fl, &ll);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474}
475
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476/* default lease lock manager operations */
Jeff Layton4d01b7f2014-09-01 15:06:54 -0400477static bool
478lease_break_callback(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479{
480 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
Jeff Layton4d01b7f2014-09-01 15:06:54 -0400481 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
Jeff Layton1c7dd2f2014-08-22 10:55:47 -0400484static void
485lease_setup(struct file_lock *fl, void **priv)
486{
487 struct file *filp = fl->fl_file;
488 struct fasync_struct *fa = *priv;
489
490 /*
491 * fasync_insert_entry() returns the old entry if any. If there was no
492 * old entry, then it used "priv" and inserted it into the fasync list.
493 * Clear the pointer to indicate that it shouldn't be freed.
494 */
495 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
496 *priv = NULL;
497
498 __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
499}
500
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700501static const struct lock_manager_operations lease_manager_ops = {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400502 .lm_break = lease_break_callback,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400503 .lm_change = lease_modify,
Jeff Layton1c7dd2f2014-08-22 10:55:47 -0400504 .lm_setup = lease_setup,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505};
506
507/*
508 * Initialize a lease, use the default lock manager operations
509 */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400510static int lease_init(struct file *filp, long type, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400512 if (assign_type(fl, type) != 0)
513 return -EINVAL;
514
Jeff Layton7ca76312014-09-01 19:04:48 -0400515 fl->fl_owner = filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 fl->fl_pid = current->tgid;
517
518 fl->fl_file = filp;
519 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 fl->fl_start = 0;
521 fl->fl_end = OFFSET_MAX;
522 fl->fl_ops = NULL;
523 fl->fl_lmops = &lease_manager_ops;
524 return 0;
525}
526
527/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400528static struct file_lock *lease_alloc(struct file *filp, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529{
530 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400531 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
533 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500534 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
536 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400537 if (error) {
538 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500539 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400540 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500541 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542}
543
544/* Check if two locks overlap each other.
545 */
546static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
547{
548 return ((fl1->fl_end >= fl2->fl_start) &&
549 (fl2->fl_end >= fl1->fl_start));
550}
551
552/*
553 * Check whether two locks have the same owner.
554 */
Matt Mackall33443c42006-01-08 01:05:22 -0800555static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400557 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 return fl2->fl_lmops == fl1->fl_lmops &&
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400559 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 return fl1->fl_owner == fl2->fl_owner;
561}
562
Jeff Layton6109c852015-01-16 15:05:57 -0500563/* Must be called with the flc_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500564static void locks_insert_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400565{
Jeff Layton7012b022013-06-21 08:58:22 -0400566 lg_local_lock(&file_lock_lglock);
567 fl->fl_link_cpu = smp_processor_id();
568 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
569 lg_local_unlock(&file_lock_lglock);
Jeff Layton88974692013-06-21 08:58:14 -0400570}
571
Jeff Layton6109c852015-01-16 15:05:57 -0500572/* Must be called with the flc_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500573static void locks_delete_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400574{
Jeff Layton7012b022013-06-21 08:58:22 -0400575 /*
576 * Avoid taking lock if already unhashed. This is safe since this check
Jeff Layton6109c852015-01-16 15:05:57 -0500577 * is done while holding the flc_lock, and new insertions into the list
Jeff Layton7012b022013-06-21 08:58:22 -0400578 * also require that it be held.
579 */
580 if (hlist_unhashed(&fl->fl_link))
581 return;
582 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
Jeff Layton139ca042013-06-21 08:58:17 -0400583 hlist_del_init(&fl->fl_link);
Jeff Layton7012b022013-06-21 08:58:22 -0400584 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
Jeff Layton88974692013-06-21 08:58:14 -0400585}
586
Jeff Layton3999e492013-06-21 08:58:19 -0400587static unsigned long
588posix_owner_key(struct file_lock *fl)
589{
590 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
591 return fl->fl_lmops->lm_owner_key(fl);
592 return (unsigned long)fl->fl_owner;
593}
594
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500595static void locks_insert_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400596{
Daniel Wagner663d5af2015-04-03 09:04:03 -0400597 lockdep_assert_held(&blocked_lock_lock);
598
Jeff Layton3999e492013-06-21 08:58:19 -0400599 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
Jeff Layton88974692013-06-21 08:58:14 -0400600}
601
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500602static void locks_delete_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400603{
Daniel Wagner663d5af2015-04-03 09:04:03 -0400604 lockdep_assert_held(&blocked_lock_lock);
605
Jeff Layton48f74182013-06-21 08:58:18 -0400606 hash_del(&waiter->fl_link);
Jeff Layton88974692013-06-21 08:58:14 -0400607}
608
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609/* Remove waiter from blocker's block list.
610 * When blocker ends up pointing to itself then the list is empty.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400611 *
Jeff Layton7b2296a2013-06-21 08:58:20 -0400612 * Must be called with blocked_lock_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 */
Matt Mackall33443c42006-01-08 01:05:22 -0800614static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615{
Jeff Layton88974692013-06-21 08:58:14 -0400616 locks_delete_global_blocked(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 list_del_init(&waiter->fl_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 waiter->fl_next = NULL;
619}
620
Jeff Layton1a9e64a2013-06-21 08:58:10 -0400621static void locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400623 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 __locks_delete_block(waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400625 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626}
627
628/* Insert waiter into blocker's block list.
629 * We use a circular list so that processes can be easily woken up in
630 * the order they blocked. The documentation doesn't require this but
631 * it seems like the reasonable thing to do.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400632 *
Jeff Layton6109c852015-01-16 15:05:57 -0500633 * Must be called with both the flc_lock and blocked_lock_lock held. The
634 * fl_block list itself is protected by the blocked_lock_lock, but by ensuring
635 * that the flc_lock is also held on insertions we can avoid taking the
636 * blocked_lock_lock in some cases when we see that the fl_block list is empty.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 */
Jeff Layton1c8c6012013-06-21 08:58:15 -0400638static void __locks_insert_block(struct file_lock *blocker,
639 struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800641 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 waiter->fl_next = blocker;
Jeff Layton88974692013-06-21 08:58:14 -0400643 list_add_tail(&waiter->fl_block, &blocker->fl_block);
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400644 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
Jeff Layton1c8c6012013-06-21 08:58:15 -0400645 locks_insert_global_blocked(waiter);
646}
647
Jeff Layton6109c852015-01-16 15:05:57 -0500648/* Must be called with flc_lock held. */
Jeff Layton1c8c6012013-06-21 08:58:15 -0400649static void locks_insert_block(struct file_lock *blocker,
650 struct file_lock *waiter)
651{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400652 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400653 __locks_insert_block(blocker, waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400654 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655}
656
Jeff Layton1cb36012013-06-21 08:58:12 -0400657/*
658 * Wake up processes blocked waiting for blocker.
659 *
Jeff Layton6109c852015-01-16 15:05:57 -0500660 * Must be called with the inode->flc_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 */
662static void locks_wake_up_blocks(struct file_lock *blocker)
663{
Jeff Layton4e8c7652013-06-21 08:58:16 -0400664 /*
665 * Avoid taking global lock if list is empty. This is safe since new
Jeff Layton6109c852015-01-16 15:05:57 -0500666 * blocked requests are only added to the list under the flc_lock, and
667 * the flc_lock is always held here. Note that removal from the fl_block
668 * list does not require the flc_lock, so we must recheck list_empty()
Jeff Layton7b2296a2013-06-21 08:58:20 -0400669 * after acquiring the blocked_lock_lock.
Jeff Layton4e8c7652013-06-21 08:58:16 -0400670 */
671 if (list_empty(&blocker->fl_block))
672 return;
673
Jeff Layton7b2296a2013-06-21 08:58:20 -0400674 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400676 struct file_lock *waiter;
677
678 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 struct file_lock, fl_block);
680 __locks_delete_block(waiter);
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400681 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
682 waiter->fl_lmops->lm_notify(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 else
684 wake_up(&waiter->fl_wait);
685 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400686 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687}
688
Jeff Layton5263e312015-01-16 15:05:55 -0500689static void
Jeff Laytone084c1b2015-02-16 14:32:03 -0500690locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
Jeff Layton5263e312015-01-16 15:05:55 -0500691{
692 fl->fl_nspid = get_pid(task_tgid(current));
693 list_add_tail(&fl->fl_list, before);
694 locks_insert_global_locks(fl);
695}
696
Jeff Layton8634b512015-01-16 15:05:55 -0500697static void
Jeff Laytone084c1b2015-02-16 14:32:03 -0500698locks_unlink_lock_ctx(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
Jeff Layton88974692013-06-21 08:58:14 -0400700 locks_delete_global_locks(fl);
Jeff Layton8634b512015-01-16 15:05:55 -0500701 list_del_init(&fl->fl_list);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000702 if (fl->fl_nspid) {
703 put_pid(fl->fl_nspid);
704 fl->fl_nspid = NULL;
705 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 locks_wake_up_blocks(fl);
Jeff Layton24cbe782014-02-03 12:13:06 -0500707}
708
Jeff Layton5263e312015-01-16 15:05:55 -0500709static void
Jeff Laytone084c1b2015-02-16 14:32:03 -0500710locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
Jeff Layton5263e312015-01-16 15:05:55 -0500711{
Jeff Laytone084c1b2015-02-16 14:32:03 -0500712 locks_unlink_lock_ctx(fl);
Jeff Layton8634b512015-01-16 15:05:55 -0500713 if (dispose)
714 list_add(&fl->fl_list, dispose);
715 else
716 locks_free_lock(fl);
Jeff Layton5263e312015-01-16 15:05:55 -0500717}
718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
720 * checks for shared/exclusive status of overlapping locks.
721 */
722static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
723{
724 if (sys_fl->fl_type == F_WRLCK)
725 return 1;
726 if (caller_fl->fl_type == F_WRLCK)
727 return 1;
728 return 0;
729}
730
731/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
732 * checking before calling the locks_conflict().
733 */
734static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
735{
736 /* POSIX locks owned by the same process do not conflict with
737 * each other.
738 */
Jeff Layton9b8c8692015-04-03 09:04:02 -0400739 if (posix_same_owner(caller_fl, sys_fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 return (0);
741
742 /* Check whether they overlap */
743 if (!locks_overlap(caller_fl, sys_fl))
744 return 0;
745
746 return (locks_conflict(caller_fl, sys_fl));
747}
748
749/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
750 * checking before calling the locks_conflict().
751 */
752static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
753{
754 /* FLOCK locks referring to the same filp do not conflict with
755 * each other.
756 */
Jeff Layton9b8c8692015-04-03 09:04:02 -0400757 if (caller_fl->fl_file == sys_fl->fl_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 return (0);
759 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
760 return 0;
761
762 return (locks_conflict(caller_fl, sys_fl));
763}
764
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400765void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500766posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767{
768 struct file_lock *cfl;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500769 struct file_lock_context *ctx;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400770 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500772 ctx = inode->i_flctx;
773 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400774 fl->fl_type = F_UNLCK;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500775 return;
776 }
777
Jeff Layton6109c852015-01-16 15:05:57 -0500778 spin_lock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500779 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
780 if (posix_locks_conflict(fl, cfl)) {
781 locks_copy_conflock(fl, cfl);
782 if (cfl->fl_nspid)
783 fl->fl_pid = pid_vnr(cfl->fl_nspid);
784 goto out;
785 }
786 }
787 fl->fl_type = F_UNLCK;
788out:
Jeff Layton6109c852015-01-16 15:05:57 -0500789 spin_unlock(&ctx->flc_lock);
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400790 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792EXPORT_SYMBOL(posix_test_lock);
793
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400794/*
795 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400797 * We attempt to detect deadlocks that are due purely to posix file
798 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400800 * We assume that a task can be waiting for at most one lock at a time.
801 * So for any acquired lock, the process holding that lock may be
802 * waiting on at most one other lock. That lock in turns may be held by
803 * someone waiting for at most one other lock. Given a requested lock
804 * caller_fl which is about to wait for a conflicting lock block_fl, we
805 * follow this chain of waiters to ensure we are not about to create a
806 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400807 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400808 * Since we do this before we ever put a process to sleep on a lock, we
809 * are ensured that there is never a cycle; that is what guarantees that
810 * the while() loop in posix_locks_deadlock() eventually completes.
811 *
812 * Note: the above assumption may not be true when handling lock
813 * requests from a broken NFS client. It may also fail in the presence
814 * of tasks (such as posix threads) sharing the same open file table.
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400815 * To handle those cases, we just bail out after a few iterations.
Jeff Layton57b65322014-02-03 12:13:09 -0500816 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400817 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
Jeff Layton57b65322014-02-03 12:13:09 -0500818 * Because the owner is not even nominally tied to a thread of
819 * execution, the deadlock detection below can't reasonably work well. Just
820 * skip it for those.
821 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400822 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
Jeff Layton57b65322014-02-03 12:13:09 -0500823 * locks that just checks for the case where two tasks are attempting to
824 * upgrade from read to write locks on the same inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400826
827#define MAX_DEADLK_ITERATIONS 10
828
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400829/* Find a lock that the owner of the given block_fl is blocking on. */
830static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
831{
832 struct file_lock *fl;
833
Jeff Layton3999e492013-06-21 08:58:19 -0400834 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400835 if (posix_same_owner(fl, block_fl))
836 return fl->fl_next;
837 }
838 return NULL;
839}
840
Jeff Layton7b2296a2013-06-21 08:58:20 -0400841/* Must be called with the blocked_lock_lock held! */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700842static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 struct file_lock *block_fl)
844{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400845 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Daniel Wagner663d5af2015-04-03 09:04:03 -0400847 lockdep_assert_held(&blocked_lock_lock);
848
Jeff Layton57b65322014-02-03 12:13:09 -0500849 /*
850 * This deadlock detector can't reasonably detect deadlocks with
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400851 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
Jeff Layton57b65322014-02-03 12:13:09 -0500852 */
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400853 if (IS_OFDLCK(caller_fl))
Jeff Layton57b65322014-02-03 12:13:09 -0500854 return 0;
855
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400856 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
857 if (i++ > MAX_DEADLK_ITERATIONS)
858 return 0;
859 if (posix_same_owner(caller_fl, block_fl))
860 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 }
862 return 0;
863}
864
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400866 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400867 *
868 * Note that if called with an FL_EXISTS argument, the caller may determine
869 * whether or not a lock was successfully freed by testing the return
870 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800872static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800874 struct file_lock *new_fl = NULL;
Jeff Layton5263e312015-01-16 15:05:55 -0500875 struct file_lock *fl;
876 struct file_lock_context *ctx;
877 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 int error = 0;
Jeff Layton5263e312015-01-16 15:05:55 -0500879 bool found = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400880 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881
Jeff Layton5c1c6692015-04-03 09:04:03 -0400882 ctx = locks_get_lock_context(inode, request->fl_type);
883 if (!ctx) {
884 if (request->fl_type != F_UNLCK)
885 return -ENOMEM;
886 return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
887 }
Jeff Layton5263e312015-01-16 15:05:55 -0500888
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200889 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
890 new_fl = locks_alloc_lock();
891 if (!new_fl)
892 return -ENOMEM;
893 }
894
Jeff Layton6109c852015-01-16 15:05:57 -0500895 spin_lock(&ctx->flc_lock);
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400896 if (request->fl_flags & FL_ACCESS)
897 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400898
Jeff Layton5263e312015-01-16 15:05:55 -0500899 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 if (filp != fl->fl_file)
901 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800902 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 goto out;
Jeff Layton5263e312015-01-16 15:05:55 -0500904 found = true;
Jeff Laytone084c1b2015-02-16 14:32:03 -0500905 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 break;
907 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
Trond Myklebustf475ae92006-06-29 16:38:32 -0400909 if (request->fl_type == F_UNLCK) {
910 if ((request->fl_flags & FL_EXISTS) && !found)
911 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800912 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400913 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400915find_conflict:
Jeff Layton5263e312015-01-16 15:05:55 -0500916 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
Trond Myklebust993dfa82006-03-31 02:30:55 -0800917 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 continue;
919 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700920 if (!(request->fl_flags & FL_SLEEP))
921 goto out;
922 error = FILE_LOCK_DEFERRED;
923 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 goto out;
925 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400926 if (request->fl_flags & FL_ACCESS)
927 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800928 locks_copy_lock(new_fl, request);
Jeff Laytone084c1b2015-02-16 14:32:03 -0500929 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800930 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400931 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
933out:
Jeff Layton6109c852015-01-16 15:05:57 -0500934 spin_unlock(&ctx->flc_lock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800935 if (new_fl)
936 locks_free_lock(new_fl);
Jeff Laytoned9814d2014-08-11 14:20:31 -0400937 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 return error;
939}
940
Marc Eshel150b3932007-01-18 16:15:35 -0500941static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942{
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500943 struct file_lock *fl, *tmp;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700944 struct file_lock *new_fl = NULL;
945 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 struct file_lock *left = NULL;
947 struct file_lock *right = NULL;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500948 struct file_lock_context *ctx;
Jeff Laytonb9746ef2013-06-21 08:58:13 -0400949 int error;
950 bool added = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400951 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
Jeff Layton5c1c6692015-04-03 09:04:03 -0400953 ctx = locks_get_lock_context(inode, request->fl_type);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500954 if (!ctx)
Jeff Layton5c1c6692015-04-03 09:04:03 -0400955 return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500956
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 /*
958 * We may need two file_lock structures for this operation,
959 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700960 *
961 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700963 if (!(request->fl_flags & FL_ACCESS) &&
964 (request->fl_type != F_UNLCK ||
965 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
966 new_fl = locks_alloc_lock();
967 new_fl2 = locks_alloc_lock();
968 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
Jeff Layton6109c852015-01-16 15:05:57 -0500970 spin_lock(&ctx->flc_lock);
Jeff Layton1cb36012013-06-21 08:58:12 -0400971 /*
972 * New lock request. Walk all POSIX locks and look for conflicts. If
973 * there are any, either return error or put the request on the
Jeff Layton48f74182013-06-21 08:58:18 -0400974 * blocker's list of waiters and the global blocked_hash.
Jeff Layton1cb36012013-06-21 08:58:12 -0400975 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 if (request->fl_type != F_UNLCK) {
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500977 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 if (!posix_locks_conflict(request, fl))
979 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800980 if (conflock)
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400981 locks_copy_conflock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 error = -EAGAIN;
983 if (!(request->fl_flags & FL_SLEEP))
984 goto out;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400985 /*
986 * Deadlock detection and insertion into the blocked
987 * locks list must be done while holding the same lock!
988 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 error = -EDEADLK;
Jeff Layton7b2296a2013-06-21 08:58:20 -0400990 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400991 if (likely(!posix_locks_deadlock(request, fl))) {
992 error = FILE_LOCK_DEFERRED;
993 __locks_insert_block(fl, request);
994 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400995 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 goto out;
997 }
998 }
999
1000 /* If we're just looking for a conflict, we're done. */
1001 error = 0;
1002 if (request->fl_flags & FL_ACCESS)
1003 goto out;
1004
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001005 /* Find the first old lock with the same owner as the new lock */
1006 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1007 if (posix_same_owner(request, fl))
1008 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 }
1010
Jeff Layton1cb36012013-06-21 08:58:12 -04001011 /* Process locks with this owner. */
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001012 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1013 if (!posix_same_owner(request, fl))
1014 break;
1015
1016 /* Detect adjacent or overlapping regions (if same lock type) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -07001018 /* In all comparisons of start vs end, use
1019 * "start - 1" rather than "end + 1". If end
1020 * is OFFSET_MAX, end + 1 will become negative.
1021 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 if (fl->fl_end < request->fl_start - 1)
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001023 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 /* If the next lock in the list has entirely bigger
1025 * addresses than the new one, insert the lock here.
1026 */
Olaf Kirch449231d2005-08-25 16:25:35 -07001027 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 break;
1029
1030 /* If we come here, the new and old lock are of the
1031 * same type and adjacent or overlapping. Make one
1032 * lock yielding from the lower start address of both
1033 * locks to the higher end address.
1034 */
1035 if (fl->fl_start > request->fl_start)
1036 fl->fl_start = request->fl_start;
1037 else
1038 request->fl_start = fl->fl_start;
1039 if (fl->fl_end < request->fl_end)
1040 fl->fl_end = request->fl_end;
1041 else
1042 request->fl_end = fl->fl_end;
1043 if (added) {
Jeff Laytone084c1b2015-02-16 14:32:03 -05001044 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 continue;
1046 }
1047 request = fl;
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001048 added = true;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001049 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 /* Processing for different lock types is a bit
1051 * more complex.
1052 */
1053 if (fl->fl_end < request->fl_start)
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001054 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 if (fl->fl_start > request->fl_end)
1056 break;
1057 if (request->fl_type == F_UNLCK)
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001058 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 if (fl->fl_start < request->fl_start)
1060 left = fl;
1061 /* If the next lock in the list has a higher end
1062 * address than the new one, insert the new one here.
1063 */
1064 if (fl->fl_end > request->fl_end) {
1065 right = fl;
1066 break;
1067 }
1068 if (fl->fl_start >= request->fl_start) {
1069 /* The new lock completely replaces an old
1070 * one (This may happen several times).
1071 */
1072 if (added) {
Jeff Laytone084c1b2015-02-16 14:32:03 -05001073 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 continue;
1075 }
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001076 /*
1077 * Replace the old lock with new_fl, and
1078 * remove the old one. It's safe to do the
1079 * insert here since we know that we won't be
1080 * using new_fl later, and that the lock is
1081 * just replacing an existing lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 */
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001083 error = -ENOLCK;
1084 if (!new_fl)
1085 goto out;
1086 locks_copy_lock(new_fl, request);
1087 request = new_fl;
1088 new_fl = NULL;
Jeff Laytone084c1b2015-02-16 14:32:03 -05001089 locks_insert_lock_ctx(request, &fl->fl_list);
1090 locks_delete_lock_ctx(fl, &dispose);
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001091 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 }
1093 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 }
1095
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001096 /*
Jeff Layton1cb36012013-06-21 08:58:12 -04001097 * The above code only modifies existing locks in case of merging or
1098 * replacing. If new lock(s) need to be inserted all modifications are
1099 * done below this, so it's safe yet to bail out.
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001100 */
1101 error = -ENOLCK; /* "no luck" */
1102 if (right && left == right && !new_fl2)
1103 goto out;
1104
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 error = 0;
1106 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -04001107 if (request->fl_type == F_UNLCK) {
1108 if (request->fl_flags & FL_EXISTS)
1109 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -04001111 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001112
1113 if (!new_fl) {
1114 error = -ENOLCK;
1115 goto out;
1116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 locks_copy_lock(new_fl, request);
Jeff Laytone084c1b2015-02-16 14:32:03 -05001118 locks_insert_lock_ctx(new_fl, &fl->fl_list);
Jeff Layton2e2f756f2015-02-17 17:08:23 -05001119 fl = new_fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 new_fl = NULL;
1121 }
1122 if (right) {
1123 if (left == right) {
1124 /* The new lock breaks the old one in two pieces,
1125 * so we have to use the second new lock.
1126 */
1127 left = new_fl2;
1128 new_fl2 = NULL;
1129 locks_copy_lock(left, right);
Jeff Laytone084c1b2015-02-16 14:32:03 -05001130 locks_insert_lock_ctx(left, &fl->fl_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 }
1132 right->fl_start = request->fl_end + 1;
1133 locks_wake_up_blocks(right);
1134 }
1135 if (left) {
1136 left->fl_end = request->fl_start - 1;
1137 locks_wake_up_blocks(left);
1138 }
1139 out:
Jeff Layton6109c852015-01-16 15:05:57 -05001140 spin_unlock(&ctx->flc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 /*
1142 * Free any unused locks.
1143 */
1144 if (new_fl)
1145 locks_free_lock(new_fl);
1146 if (new_fl2)
1147 locks_free_lock(new_fl2);
Jeff Laytoned9814d2014-08-11 14:20:31 -04001148 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 return error;
1150}
1151
1152/**
1153 * posix_lock_file - Apply a POSIX-style lock to a file
1154 * @filp: The file to apply the lock to
1155 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001156 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 *
1158 * Add a POSIX style lock to a file.
1159 * We merge adjacent & overlapping locks whenever possible.
1160 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001161 *
1162 * Note that if called with an FL_EXISTS argument, the caller may determine
1163 * whether or not a lock was successfully freed by testing the return
1164 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 */
Marc Eshel150b3932007-01-18 16:15:35 -05001166int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001167 struct file_lock *conflock)
1168{
Al Viro496ad9a2013-01-23 17:07:38 -05001169 return __posix_lock_file(file_inode(filp), fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001170}
Marc Eshel150b3932007-01-18 16:15:35 -05001171EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
1173/**
1174 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1175 * @filp: The file to apply the lock to
1176 * @fl: The lock to be applied
1177 *
1178 * Add a POSIX style lock to a file.
1179 * We merge adjacent & overlapping locks whenever possible.
1180 * POSIX locks are sorted by owner task, then by starting address
1181 */
1182int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1183{
1184 int error;
1185 might_sleep ();
1186 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001187 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001188 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 break;
1190 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1191 if (!error)
1192 continue;
1193
1194 locks_delete_block(fl);
1195 break;
1196 }
1197 return error;
1198}
1199EXPORT_SYMBOL(posix_lock_file_wait);
1200
1201/**
1202 * locks_mandatory_locked - Check for an active lock
Jeff Laytond7a06982014-03-10 09:54:15 -04001203 * @file: the file to check
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 *
1205 * Searches the inode's list of locks to find any POSIX locks which conflict.
1206 * This function is called from locks_verify_locked() only.
1207 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001208int locks_mandatory_locked(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209{
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001210 int ret;
Jeff Laytond7a06982014-03-10 09:54:15 -04001211 struct inode *inode = file_inode(file);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001212 struct file_lock_context *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 struct file_lock *fl;
1214
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001215 ctx = inode->i_flctx;
1216 if (!ctx || list_empty_careful(&ctx->flc_posix))
1217 return 0;
1218
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 /*
1220 * Search the lock list for this inode for any POSIX locks.
1221 */
Jeff Layton6109c852015-01-16 15:05:57 -05001222 spin_lock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001223 ret = 0;
1224 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001225 if (fl->fl_owner != current->files &&
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001226 fl->fl_owner != file) {
1227 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 break;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001229 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 }
Jeff Layton6109c852015-01-16 15:05:57 -05001231 spin_unlock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001232 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233}
1234
1235/**
1236 * locks_mandatory_area - Check for a conflicting lock
1237 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1238 * for shared
1239 * @inode: the file to check
1240 * @filp: how the file was opened (if it was)
1241 * @offset: start of area to check
1242 * @count: length of area to check
1243 *
1244 * Searches the inode's list of locks to find any POSIX locks which conflict.
1245 * This function is called from rw_verify_area() and
1246 * locks_verify_truncate().
1247 */
1248int locks_mandatory_area(int read_write, struct inode *inode,
1249 struct file *filp, loff_t offset,
1250 size_t count)
1251{
1252 struct file_lock fl;
1253 int error;
Jeff Layton29723ad2014-03-10 09:54:19 -04001254 bool sleep = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
1256 locks_init_lock(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 fl.fl_pid = current->tgid;
1258 fl.fl_file = filp;
1259 fl.fl_flags = FL_POSIX | FL_ACCESS;
1260 if (filp && !(filp->f_flags & O_NONBLOCK))
Jeff Layton29723ad2014-03-10 09:54:19 -04001261 sleep = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1263 fl.fl_start = offset;
1264 fl.fl_end = offset + count - 1;
1265
1266 for (;;) {
Jeff Layton29723ad2014-03-10 09:54:19 -04001267 if (filp) {
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001268 fl.fl_owner = filp;
Jeff Layton29723ad2014-03-10 09:54:19 -04001269 fl.fl_flags &= ~FL_SLEEP;
1270 error = __posix_lock_file(inode, &fl, NULL);
1271 if (!error)
1272 break;
1273 }
1274
1275 if (sleep)
1276 fl.fl_flags |= FL_SLEEP;
1277 fl.fl_owner = current->files;
Marc Eshel150b3932007-01-18 16:15:35 -05001278 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001279 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 break;
1281 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1282 if (!error) {
1283 /*
1284 * If we've been sleeping someone might have
1285 * changed the permissions behind our back.
1286 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001287 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 continue;
1289 }
1290
1291 locks_delete_block(&fl);
1292 break;
1293 }
1294
1295 return error;
1296}
1297
1298EXPORT_SYMBOL(locks_mandatory_area);
1299
J. Bruce Fields778fc542011-07-26 18:25:49 -04001300static void lease_clear_pending(struct file_lock *fl, int arg)
1301{
1302 switch (arg) {
1303 case F_UNLCK:
1304 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1305 /* fall through: */
1306 case F_RDLCK:
1307 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1308 }
1309}
1310
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311/* We already had a lease on this file; just change its type */
Jeff Layton7448cc32015-01-16 15:05:57 -05001312int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 int error = assign_type(fl, arg);
1315
1316 if (error)
1317 return error;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001318 lease_clear_pending(fl, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 locks_wake_up_blocks(fl);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001320 if (arg == F_UNLCK) {
1321 struct file *filp = fl->fl_file;
1322
1323 f_delown(filp);
1324 filp->f_owner.signum = 0;
J. Bruce Fields96d6d592012-07-27 16:18:00 -04001325 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1326 if (fl->fl_fasync != NULL) {
1327 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1328 fl->fl_fasync = NULL;
1329 }
Jeff Laytone084c1b2015-02-16 14:32:03 -05001330 locks_delete_lock_ctx(fl, dispose);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001331 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 return 0;
1333}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334EXPORT_SYMBOL(lease_modify);
1335
J. Bruce Fields778fc542011-07-26 18:25:49 -04001336static bool past_time(unsigned long then)
1337{
1338 if (!then)
1339 /* 0 is a special value meaning "this never expires": */
1340 return false;
1341 return time_after(jiffies, then);
1342}
1343
Jeff Laytonc45198e2014-09-01 07:12:07 -04001344static void time_out_leases(struct inode *inode, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345{
Jeff Layton8634b512015-01-16 15:05:55 -05001346 struct file_lock_context *ctx = inode->i_flctx;
1347 struct file_lock *fl, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
Jeff Layton6109c852015-01-16 15:05:57 -05001349 lockdep_assert_held(&ctx->flc_lock);
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001350
Jeff Layton8634b512015-01-16 15:05:55 -05001351 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001352 trace_time_out_leases(inode, fl);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001353 if (past_time(fl->fl_downgrade_time))
Jeff Layton7448cc32015-01-16 15:05:57 -05001354 lease_modify(fl, F_RDLCK, dispose);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001355 if (past_time(fl->fl_break_time))
Jeff Layton7448cc32015-01-16 15:05:57 -05001356 lease_modify(fl, F_UNLCK, dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 }
1358}
1359
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001360static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1361{
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001362 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT))
1363 return false;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001364 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1365 return false;
1366 return locks_conflict(breaker, lease);
1367}
1368
Jeff Layton03d12dd2014-09-01 14:53:41 -04001369static bool
1370any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1371{
Jeff Layton8634b512015-01-16 15:05:55 -05001372 struct file_lock_context *ctx = inode->i_flctx;
Jeff Layton03d12dd2014-09-01 14:53:41 -04001373 struct file_lock *fl;
1374
Jeff Layton6109c852015-01-16 15:05:57 -05001375 lockdep_assert_held(&ctx->flc_lock);
Jeff Layton03d12dd2014-09-01 14:53:41 -04001376
Jeff Layton8634b512015-01-16 15:05:55 -05001377 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Jeff Layton03d12dd2014-09-01 14:53:41 -04001378 if (leases_conflict(fl, breaker))
1379 return true;
1380 }
1381 return false;
1382}
1383
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384/**
1385 * __break_lease - revoke all outstanding leases on file
1386 * @inode: the inode of the file to return
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001387 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1388 * break all leases
1389 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1390 * only delegations
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 *
david m. richter87250dd2007-05-09 16:10:27 -04001392 * break_lease (inlined for speed) has checked there already is at least
1393 * some kind of lock (maybe a lease) on this file. Leases are broken on
1394 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 * specified %O_NONBLOCK to your open().
1396 */
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001397int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398{
J. Bruce Fields778fc542011-07-26 18:25:49 -04001399 int error = 0;
Jeff Layton8634b512015-01-16 15:05:55 -05001400 struct file_lock_context *ctx = inode->i_flctx;
Yan, Zhenga9011252015-03-27 10:34:20 +08001401 struct file_lock *new_fl, *fl, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 unsigned long break_time;
Al Viro8737c932009-12-24 06:47:55 -05001403 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001404 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
Al Viro8737c932009-12-24 06:47:55 -05001406 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001407 if (IS_ERR(new_fl))
1408 return PTR_ERR(new_fl);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001409 new_fl->fl_flags = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
Jeff Layton8634b512015-01-16 15:05:55 -05001411 /* typically we will check that ctx is non-NULL before calling */
1412 if (!ctx) {
1413 WARN_ON_ONCE(1);
1414 return error;
1415 }
1416
Jeff Layton6109c852015-01-16 15:05:57 -05001417 spin_lock(&ctx->flc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Jeff Laytonc45198e2014-09-01 07:12:07 -04001419 time_out_leases(inode, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
Jeff Layton03d12dd2014-09-01 14:53:41 -04001421 if (!any_leases_conflict(inode, new_fl))
J. Bruce Fields778fc542011-07-26 18:25:49 -04001422 goto out;
1423
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 break_time = 0;
1425 if (lease_break_time > 0) {
1426 break_time = jiffies + lease_break_time * HZ;
1427 if (break_time == 0)
1428 break_time++; /* so that 0 means no break time */
1429 }
1430
Yan, Zhenga9011252015-03-27 10:34:20 +08001431 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001432 if (!leases_conflict(fl, new_fl))
1433 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001434 if (want_write) {
1435 if (fl->fl_flags & FL_UNLOCK_PENDING)
1436 continue;
1437 fl->fl_flags |= FL_UNLOCK_PENDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 fl->fl_break_time = break_time;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001439 } else {
Jeff Layton8634b512015-01-16 15:05:55 -05001440 if (lease_breaking(fl))
J. Bruce Fields778fc542011-07-26 18:25:49 -04001441 continue;
1442 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1443 fl->fl_downgrade_time = break_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 }
Jeff Layton4d01b7f2014-09-01 15:06:54 -04001445 if (fl->fl_lmops->lm_break(fl))
Jeff Laytone084c1b2015-02-16 14:32:03 -05001446 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 }
1448
Jeff Layton8634b512015-01-16 15:05:55 -05001449 if (list_empty(&ctx->flc_lease))
Jeff Layton4d01b7f2014-09-01 15:06:54 -04001450 goto out;
1451
Jeff Layton843c6b22014-09-01 14:27:43 -04001452 if (mode & O_NONBLOCK) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001453 trace_break_lease_noblock(inode, new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 error = -EWOULDBLOCK;
1455 goto out;
1456 }
1457
1458restart:
Jeff Layton8634b512015-01-16 15:05:55 -05001459 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1460 break_time = fl->fl_break_time;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001461 if (break_time != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 break_time -= jiffies;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001463 if (break_time == 0)
1464 break_time++;
Jeff Layton8634b512015-01-16 15:05:55 -05001465 locks_insert_block(fl, new_fl);
Jeff Layton62af4f12014-05-09 14:13:05 -04001466 trace_break_lease_block(inode, new_fl);
Jeff Layton6109c852015-01-16 15:05:57 -05001467 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001468 locks_dispose_list(&dispose);
Matthew Wilcox4321e012008-01-14 21:28:30 -07001469 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1470 !new_fl->fl_next, break_time);
Jeff Layton6109c852015-01-16 15:05:57 -05001471 spin_lock(&ctx->flc_lock);
Jeff Layton62af4f12014-05-09 14:13:05 -04001472 trace_break_lease_unblock(inode, new_fl);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001473 locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 if (error >= 0) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001475 /*
1476 * Wait for the next conflicting lease that has not been
1477 * broken yet
1478 */
Jeff Layton03d12dd2014-09-01 14:53:41 -04001479 if (error == 0)
1480 time_out_leases(inode, &dispose);
1481 if (any_leases_conflict(inode, new_fl))
1482 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 error = 0;
1484 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485out:
Jeff Layton6109c852015-01-16 15:05:57 -05001486 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001487 locks_dispose_list(&dispose);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001488 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 return error;
1490}
1491
1492EXPORT_SYMBOL(__break_lease);
1493
1494/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001495 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 * @inode: the inode
1497 * @time: pointer to a timespec which will contain the last modified time
1498 *
1499 * This is to force NFS clients to flush their caches for files with
1500 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001501 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 */
1503void lease_get_mtime(struct inode *inode, struct timespec *time)
1504{
Jeff Laytonbfe86022014-08-22 10:18:44 -04001505 bool has_lease = false;
Jeff Layton8634b512015-01-16 15:05:55 -05001506 struct file_lock_context *ctx = inode->i_flctx;
1507 struct file_lock *fl;
Jeff Laytonbfe86022014-08-22 10:18:44 -04001508
Jeff Layton8634b512015-01-16 15:05:55 -05001509 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
Jeff Layton6109c852015-01-16 15:05:57 -05001510 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001511 if (!list_empty(&ctx->flc_lease)) {
1512 fl = list_first_entry(&ctx->flc_lease,
1513 struct file_lock, fl_list);
1514 if (fl->fl_type == F_WRLCK)
1515 has_lease = true;
1516 }
Jeff Layton6109c852015-01-16 15:05:57 -05001517 spin_unlock(&ctx->flc_lock);
Jeff Laytonbfe86022014-08-22 10:18:44 -04001518 }
1519
1520 if (has_lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 *time = current_fs_time(inode->i_sb);
1522 else
1523 *time = inode->i_mtime;
1524}
1525
1526EXPORT_SYMBOL(lease_get_mtime);
1527
1528/**
1529 * fcntl_getlease - Enquire what lease is currently active
1530 * @filp: the file
1531 *
1532 * The value returned by this function will be one of
1533 * (if no lease break is pending):
1534 *
1535 * %F_RDLCK to indicate a shared lease is held.
1536 *
1537 * %F_WRLCK to indicate an exclusive lease is held.
1538 *
1539 * %F_UNLCK to indicate no lease is held.
1540 *
1541 * (if a lease break is pending):
1542 *
1543 * %F_RDLCK to indicate an exclusive lease needs to be
1544 * changed to a shared lease (or removed).
1545 *
1546 * %F_UNLCK to indicate the lease needs to be removed.
1547 *
1548 * XXX: sfr & willy disagree over whether F_INPROGRESS
1549 * should be returned to userspace.
1550 */
1551int fcntl_getlease(struct file *filp)
1552{
1553 struct file_lock *fl;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001554 struct inode *inode = file_inode(filp);
Jeff Layton8634b512015-01-16 15:05:55 -05001555 struct file_lock_context *ctx = inode->i_flctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 int type = F_UNLCK;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001557 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558
Jeff Layton8634b512015-01-16 15:05:55 -05001559 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
Jeff Layton6109c852015-01-16 15:05:57 -05001560 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001561 time_out_leases(file_inode(filp), &dispose);
1562 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1563 if (fl->fl_file != filp)
1564 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001565 type = target_leasetype(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 break;
1567 }
Jeff Layton6109c852015-01-16 15:05:57 -05001568 spin_unlock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001569 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 return type;
1572}
1573
Jeff Layton24cbe782014-02-03 12:13:06 -05001574/**
1575 * check_conflicting_open - see if the given dentry points to a file that has
1576 * an existing open that would conflict with the
1577 * desired lease.
1578 * @dentry: dentry to check
1579 * @arg: type of lease that we're trying to acquire
1580 *
1581 * Check to see if there's an existing open fd on this file that would
1582 * conflict with the lease we're trying to set.
1583 */
1584static int
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001585check_conflicting_open(const struct dentry *dentry, const long arg, int flags)
Jeff Layton24cbe782014-02-03 12:13:06 -05001586{
1587 int ret = 0;
1588 struct inode *inode = dentry->d_inode;
1589
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001590 if (flags & FL_LAYOUT)
1591 return 0;
1592
Jeff Layton24cbe782014-02-03 12:13:06 -05001593 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1594 return -EAGAIN;
1595
1596 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1597 (atomic_read(&inode->i_count) > 1)))
1598 ret = -EAGAIN;
1599
1600 return ret;
1601}
1602
Jeff Laytone6f5c782014-08-22 10:40:25 -04001603static int
1604generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605{
Jeff Layton8634b512015-01-16 15:05:55 -05001606 struct file_lock *fl, *my_fl = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001607 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 struct inode *inode = dentry->d_inode;
Jeff Layton8634b512015-01-16 15:05:55 -05001609 struct file_lock_context *ctx;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001610 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001611 int error;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001612 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
J. Bruce Fields096657b2010-10-30 17:31:14 -04001614 lease = *flp;
Jeff Layton62af4f12014-05-09 14:13:05 -04001615 trace_generic_add_lease(inode, lease);
1616
Jeff Layton5c1c6692015-04-03 09:04:03 -04001617 /* Note that arg is never F_UNLCK here */
1618 ctx = locks_get_lock_context(inode, arg);
Jeff Layton8634b512015-01-16 15:05:55 -05001619 if (!ctx)
1620 return -ENOMEM;
1621
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001622 /*
1623 * In the delegation case we need mutual exclusion with
1624 * a number of operations that take the i_mutex. We trylock
1625 * because delegations are an optional optimization, and if
1626 * there's some chance of a conflict--we'd rather not
1627 * bother, maybe that's a sign this just isn't a good file to
1628 * hand out a delegation on.
1629 */
1630 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1631 return -EAGAIN;
1632
1633 if (is_deleg && arg == F_WRLCK) {
1634 /* Write delegations are not currently supported: */
Dan Carpenter4fdb7932013-11-13 10:56:27 +03001635 mutex_unlock(&inode->i_mutex);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001636 WARN_ON_ONCE(1);
1637 return -EINVAL;
1638 }
J. Bruce Fields096657b2010-10-30 17:31:14 -04001639
Jeff Layton6109c852015-01-16 15:05:57 -05001640 spin_lock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001641 time_out_leases(inode, &dispose);
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001642 error = check_conflicting_open(dentry, arg, lease->fl_flags);
Jeff Layton24cbe782014-02-03 12:13:06 -05001643 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001644 goto out;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001645
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 /*
1647 * At this point, we know that if there is an exclusive
1648 * lease on this file, then we hold it on this filp
1649 * (otherwise our open of this file would have blocked).
1650 * And if we are trying to acquire an exclusive lease,
1651 * then the file is not open by anyone (including us)
1652 * except for this filp.
1653 */
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001654 error = -EAGAIN;
Jeff Layton8634b512015-01-16 15:05:55 -05001655 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001656 if (fl->fl_file == filp &&
1657 fl->fl_owner == lease->fl_owner) {
Jeff Layton8634b512015-01-16 15:05:55 -05001658 my_fl = fl;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001659 continue;
1660 }
Jeff Layton8634b512015-01-16 15:05:55 -05001661
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001662 /*
1663 * No exclusive leases if someone else has a lease on
1664 * this file:
1665 */
1666 if (arg == F_WRLCK)
1667 goto out;
1668 /*
1669 * Modifying our existing lease is OK, but no getting a
1670 * new lease if someone else is opening for write:
1671 */
1672 if (fl->fl_flags & FL_UNLOCK_PENDING)
1673 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 }
1675
Jeff Layton8634b512015-01-16 15:05:55 -05001676 if (my_fl != NULL) {
Jeff Layton0164bf02015-03-04 17:34:32 -05001677 lease = my_fl;
1678 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001679 if (error)
1680 goto out;
1681 goto out_setup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 }
1683
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 error = -EINVAL;
1685 if (!leases_enable)
1686 goto out;
1687
Jeff Laytone084c1b2015-02-16 14:32:03 -05001688 locks_insert_lock_ctx(lease, &ctx->flc_lease);
Jeff Layton24cbe782014-02-03 12:13:06 -05001689 /*
1690 * The check in break_lease() is lockless. It's possible for another
1691 * open to race in after we did the earlier check for a conflicting
1692 * open but before the lease was inserted. Check again for a
1693 * conflicting open and cancel the lease if there is one.
1694 *
1695 * We also add a barrier here to ensure that the insertion of the lock
1696 * precedes these checks.
1697 */
1698 smp_mb();
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001699 error = check_conflicting_open(dentry, arg, lease->fl_flags);
Jeff Layton8634b512015-01-16 15:05:55 -05001700 if (error) {
Jeff Laytone084c1b2015-02-16 14:32:03 -05001701 locks_unlink_lock_ctx(lease);
Jeff Layton8634b512015-01-16 15:05:55 -05001702 goto out;
1703 }
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001704
1705out_setup:
1706 if (lease->fl_lmops->lm_setup)
1707 lease->fl_lmops->lm_setup(lease, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708out:
Jeff Layton6109c852015-01-16 15:05:57 -05001709 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001710 locks_dispose_list(&dispose);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001711 if (is_deleg)
1712 mutex_unlock(&inode->i_mutex);
Jeff Layton8634b512015-01-16 15:05:55 -05001713 if (!error && !my_fl)
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001714 *flp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 return error;
1716}
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001717
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001718static int generic_delete_lease(struct file *filp, void *owner)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001719{
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001720 int error = -EAGAIN;
Jeff Layton8634b512015-01-16 15:05:55 -05001721 struct file_lock *fl, *victim = NULL;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001722 struct dentry *dentry = filp->f_path.dentry;
1723 struct inode *inode = dentry->d_inode;
Jeff Layton8634b512015-01-16 15:05:55 -05001724 struct file_lock_context *ctx = inode->i_flctx;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001725 LIST_HEAD(dispose);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001726
Jeff Layton8634b512015-01-16 15:05:55 -05001727 if (!ctx) {
1728 trace_generic_delete_lease(inode, NULL);
1729 return error;
1730 }
1731
Jeff Layton6109c852015-01-16 15:05:57 -05001732 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001733 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001734 if (fl->fl_file == filp &&
1735 fl->fl_owner == owner) {
Jeff Layton8634b512015-01-16 15:05:55 -05001736 victim = fl;
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001737 break;
Jeff Layton8634b512015-01-16 15:05:55 -05001738 }
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001739 }
Jeff Laytona9b1b452015-03-14 09:45:35 -04001740 trace_generic_delete_lease(inode, victim);
Jeff Layton8634b512015-01-16 15:05:55 -05001741 if (victim)
Jeff Layton7448cc32015-01-16 15:05:57 -05001742 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
Jeff Layton6109c852015-01-16 15:05:57 -05001743 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001744 locks_dispose_list(&dispose);
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001745 return error;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001746}
1747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748/**
1749 * generic_setlease - sets a lease on an open file
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001750 * @filp: file pointer
1751 * @arg: type of lease to obtain
1752 * @flp: input - file_lock to use, output - file_lock inserted
1753 * @priv: private data for lm_setup (may be NULL if lm_setup
1754 * doesn't require it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 *
1756 * The (input) flp->fl_lmops->lm_break function is required
1757 * by break_lease().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 */
Jeff Laytone6f5c782014-08-22 10:40:25 -04001759int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1760 void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 struct dentry *dentry = filp->f_path.dentry;
1763 struct inode *inode = dentry->d_inode;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001764 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001766 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001767 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 if (!S_ISREG(inode->i_mode))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001769 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 error = security_file_lock(filp, arg);
1771 if (error)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001772 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001774 switch (arg) {
1775 case F_UNLCK:
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001776 return generic_delete_lease(filp, *priv);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001777 case F_RDLCK:
1778 case F_WRLCK:
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001779 if (!(*flp)->fl_lmops->lm_break) {
1780 WARN_ON_ONCE(1);
1781 return -ENOLCK;
1782 }
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001783
Jeff Laytone6f5c782014-08-22 10:40:25 -04001784 return generic_add_lease(filp, arg, flp, priv);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001785 default:
Dave Jones8d657eb2012-07-13 13:35:36 -04001786 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001789EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001791/**
Jeff Laytone51673a2014-08-22 18:13:28 -04001792 * vfs_setlease - sets a lease on an open file
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001793 * @filp: file pointer
1794 * @arg: type of lease to obtain
1795 * @lease: file_lock to use when adding a lease
1796 * @priv: private info for lm_setup when adding a lease (may be
1797 * NULL if lm_setup doesn't require it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 *
Jeff Laytone51673a2014-08-22 18:13:28 -04001799 * Call this to establish a lease on the file. The "lease" argument is not
1800 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1801 * an existing lease, the (*lease)->fl_lmops->lm_break operation must be set;
1802 * if not, this function will return -ENOLCK (and generate a scary-looking
1803 * stack trace).
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001804 *
1805 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1806 * may be NULL if the lm_setup operation doesn't require it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 */
Jeff Laytone6f5c782014-08-22 10:40:25 -04001808int
1809vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810{
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001811 if (filp->f_op->setlease)
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001812 return filp->f_op->setlease(filp, arg, lease, priv);
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001813 else
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001814 return generic_setlease(filp, arg, lease, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001816EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001818static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819{
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001820 struct file_lock *fl;
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001821 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 int error;
1823
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001824 fl = lease_alloc(filp, arg);
1825 if (IS_ERR(fl))
1826 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001828 new = fasync_alloc();
1829 if (!new) {
1830 locks_free_lock(fl);
1831 return -ENOMEM;
1832 }
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001833 new->fa_fd = fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001835 error = vfs_setlease(filp, arg, &fl, (void **)&new);
Jeff Layton2dfb9282014-08-11 18:14:12 -04001836 if (fl)
1837 locks_free_lock(fl);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001838 if (new)
1839 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 return error;
1841}
1842
1843/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001844 * fcntl_setlease - sets a lease on an open file
1845 * @fd: open file descriptor
1846 * @filp: file pointer
1847 * @arg: type of lease to obtain
1848 *
1849 * Call this fcntl to establish a lease on the file.
1850 * Note that you also need to call %F_SETSIG to
1851 * receive a signal when the lease is broken.
1852 */
1853int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1854{
1855 if (arg == F_UNLCK)
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001856 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001857 return do_fcntl_add_lease(fd, filp, arg);
1858}
1859
1860/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1862 * @filp: The file to apply the lock to
1863 * @fl: The lock to be applied
1864 *
1865 * Add a FLOCK style lock to a file.
1866 */
1867int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1868{
1869 int error;
1870 might_sleep();
1871 for (;;) {
1872 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001873 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 break;
1875 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1876 if (!error)
1877 continue;
1878
1879 locks_delete_block(fl);
1880 break;
1881 }
1882 return error;
1883}
1884
1885EXPORT_SYMBOL(flock_lock_file_wait);
1886
1887/**
1888 * sys_flock: - flock() system call.
1889 * @fd: the file descriptor to lock.
1890 * @cmd: the type of lock to apply.
1891 *
1892 * Apply a %FL_FLOCK style lock to an open file descriptor.
1893 * The @cmd can be one of
1894 *
1895 * %LOCK_SH -- a shared lock.
1896 *
1897 * %LOCK_EX -- an exclusive lock.
1898 *
1899 * %LOCK_UN -- remove an existing lock.
1900 *
1901 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1902 *
1903 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1904 * processes read and write access respectively.
1905 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001906SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907{
Al Viro2903ff02012-08-28 12:52:22 -04001908 struct fd f = fdget(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 struct file_lock *lock;
1910 int can_sleep, unlock;
1911 int error;
1912
1913 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001914 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 goto out;
1916
1917 can_sleep = !(cmd & LOCK_NB);
1918 cmd &= ~LOCK_NB;
1919 unlock = (cmd == LOCK_UN);
1920
Al Viroaeb5d722008-09-02 15:28:45 -04001921 if (!unlock && !(cmd & LOCK_MAND) &&
Al Viro2903ff02012-08-28 12:52:22 -04001922 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 goto out_putf;
1924
Jeff Layton6e129d002014-09-04 10:25:06 -04001925 lock = flock_make_lock(f.file, cmd);
1926 if (IS_ERR(lock)) {
1927 error = PTR_ERR(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 goto out_putf;
Jeff Layton6e129d002014-09-04 10:25:06 -04001929 }
1930
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 if (can_sleep)
1932 lock->fl_flags |= FL_SLEEP;
1933
Al Viro2903ff02012-08-28 12:52:22 -04001934 error = security_file_lock(f.file, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 if (error)
1936 goto out_free;
1937
Al Viro72c2d532013-09-22 16:27:52 -04001938 if (f.file->f_op->flock)
Al Viro2903ff02012-08-28 12:52:22 -04001939 error = f.file->f_op->flock(f.file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 (can_sleep) ? F_SETLKW : F_SETLK,
1941 lock);
1942 else
Al Viro2903ff02012-08-28 12:52:22 -04001943 error = flock_lock_file_wait(f.file, lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
1945 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001946 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
1948 out_putf:
Al Viro2903ff02012-08-28 12:52:22 -04001949 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 out:
1951 return error;
1952}
1953
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001954/**
1955 * vfs_test_lock - test file byte range lock
1956 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001957 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001958 *
1959 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1960 * setting conf->fl_type to something other than F_UNLCK.
1961 */
1962int vfs_test_lock(struct file *filp, struct file_lock *fl)
1963{
Al Viro72c2d532013-09-22 16:27:52 -04001964 if (filp->f_op->lock)
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001965 return filp->f_op->lock(filp, F_GETLK, fl);
1966 posix_test_lock(filp, fl);
1967 return 0;
1968}
1969EXPORT_SYMBOL_GPL(vfs_test_lock);
1970
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001971static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1972{
Jeff Laytoncff2fce2014-04-22 08:24:32 -04001973 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001974#if BITS_PER_LONG == 32
1975 /*
1976 * Make sure we can represent the posix lock via
1977 * legacy 32bit flock.
1978 */
1979 if (fl->fl_start > OFFT_OFFSET_MAX)
1980 return -EOVERFLOW;
1981 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1982 return -EOVERFLOW;
1983#endif
1984 flock->l_start = fl->fl_start;
1985 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1986 fl->fl_end - fl->fl_start + 1;
1987 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001988 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001989 return 0;
1990}
1991
1992#if BITS_PER_LONG == 32
1993static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1994{
Jeff Laytoncff2fce2014-04-22 08:24:32 -04001995 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001996 flock->l_start = fl->fl_start;
1997 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1998 fl->fl_end - fl->fl_start + 1;
1999 flock->l_whence = 0;
2000 flock->l_type = fl->fl_type;
2001}
2002#endif
2003
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004/* Report the first existing lock that would conflict with l.
2005 * This implements the F_GETLK command of fcntl().
2006 */
Jeff Laytonc1e62b82014-02-03 12:13:09 -05002007int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002009 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 struct flock flock;
2011 int error;
2012
2013 error = -EFAULT;
2014 if (copy_from_user(&flock, l, sizeof(flock)))
2015 goto out;
2016 error = -EINVAL;
2017 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2018 goto out;
2019
2020 error = flock_to_posix_lock(filp, &file_lock, &flock);
2021 if (error)
2022 goto out;
2023
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002024 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05002025 error = -EINVAL;
2026 if (flock.l_pid != 0)
2027 goto out;
2028
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002029 cmd = F_GETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002030 file_lock.fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002031 file_lock.fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002032 }
2033
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002034 error = vfs_test_lock(filp, &file_lock);
2035 if (error)
2036 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002038 flock.l_type = file_lock.fl_type;
2039 if (file_lock.fl_type != F_UNLCK) {
2040 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05002041 if (error)
Kinglong Meef3282962014-08-22 10:18:43 -04002042 goto rel_priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 }
2044 error = -EFAULT;
2045 if (!copy_to_user(l, &flock, sizeof(flock)))
2046 error = 0;
Kinglong Meef3282962014-08-22 10:18:43 -04002047rel_priv:
2048 locks_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049out:
2050 return error;
2051}
2052
Marc Eshel7723ec92007-01-18 15:08:55 -05002053/**
2054 * vfs_lock_file - file byte range lock
2055 * @filp: The file to apply the lock to
2056 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2057 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05002058 * @conf: Place to return a copy of the conflicting lock, if found.
2059 *
2060 * A caller that doesn't care about the conflicting lock may pass NULL
2061 * as the final argument.
2062 *
2063 * If the filesystem defines a private ->lock() method, then @conf will
2064 * be left unchanged; so a caller that cares should initialize it to
2065 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05002066 *
2067 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2068 * locks, the ->lock() interface may return asynchronously, before the lock has
2069 * been granted or denied by the underlying filesystem, if (and only if)
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002070 * lm_grant is set. Callers expecting ->lock() to return asynchronously
Marc Eshel2beb6612006-12-05 23:31:28 -05002071 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2072 * the request is for a blocking lock. When ->lock() does return asynchronously,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002073 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05002074 * request completes.
2075 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07002076 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2077 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05002078 * nonzero return code and the file system should release the lock. The file
2079 * system is also responsible to keep a corresponding posix lock when it
2080 * grants a lock so the VFS can find out which locks are locally held and do
2081 * the correct lock cleanup when required.
2082 * The underlying filesystem must not drop the kernel lock or call
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002083 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05002084 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05002085 */
Marc Eshel150b3932007-01-18 16:15:35 -05002086int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05002087{
Al Viro72c2d532013-09-22 16:27:52 -04002088 if (filp->f_op->lock)
Marc Eshel7723ec92007-01-18 15:08:55 -05002089 return filp->f_op->lock(filp, cmd, fl);
2090 else
Marc Eshel150b3932007-01-18 16:15:35 -05002091 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05002092}
2093EXPORT_SYMBOL_GPL(vfs_lock_file);
2094
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002095static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2096 struct file_lock *fl)
2097{
2098 int error;
2099
2100 error = security_file_lock(filp, fl->fl_type);
2101 if (error)
2102 return error;
2103
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002104 for (;;) {
2105 error = vfs_lock_file(filp, cmd, fl, NULL);
2106 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002107 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002108 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2109 if (!error)
2110 continue;
2111
2112 locks_delete_block(fl);
2113 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002114 }
2115
2116 return error;
2117}
2118
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002119/* Ensure that fl->fl_filp has compatible f_mode for F_SETLK calls */
2120static int
2121check_fmode_for_setlk(struct file_lock *fl)
2122{
2123 switch (fl->fl_type) {
2124 case F_RDLCK:
2125 if (!(fl->fl_file->f_mode & FMODE_READ))
2126 return -EBADF;
2127 break;
2128 case F_WRLCK:
2129 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2130 return -EBADF;
2131 }
2132 return 0;
2133}
2134
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135/* Apply the lock described by l to an open file descriptor.
2136 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2137 */
Peter Staubachc2936212005-07-27 11:45:09 -07002138int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2139 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140{
2141 struct file_lock *file_lock = locks_alloc_lock();
2142 struct flock flock;
2143 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002144 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 int error;
2146
2147 if (file_lock == NULL)
2148 return -ENOLCK;
2149
2150 /*
2151 * This might block, so we do it before checking the inode.
2152 */
2153 error = -EFAULT;
2154 if (copy_from_user(&flock, l, sizeof(flock)))
2155 goto out;
2156
Al Viro496ad9a2013-01-23 17:07:38 -05002157 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158
2159 /* Don't allow mandatory locks on files that may be memory mapped
2160 * and shared.
2161 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002162 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 error = -EAGAIN;
2164 goto out;
2165 }
2166
Peter Staubachc2936212005-07-27 11:45:09 -07002167again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 error = flock_to_posix_lock(filp, file_lock, &flock);
2169 if (error)
2170 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002171
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002172 error = check_fmode_for_setlk(file_lock);
2173 if (error)
2174 goto out;
2175
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002176 /*
2177 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002178 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002179 */
2180 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002181 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002182 error = -EINVAL;
2183 if (flock.l_pid != 0)
2184 goto out;
2185
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002186 cmd = F_SETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002187 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002188 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002189 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002190 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002191 error = -EINVAL;
2192 if (flock.l_pid != 0)
2193 goto out;
2194
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002195 cmd = F_SETLKW;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002196 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002197 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002198 /* Fallthrough */
2199 case F_SETLKW:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 file_lock->fl_flags |= FL_SLEEP;
2201 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002202
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002203 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204
Peter Staubachc2936212005-07-27 11:45:09 -07002205 /*
2206 * Attempt to detect a close/fcntl race and recover by
2207 * releasing the lock that was just acquired.
2208 */
Al Viro0b2bac22008-05-06 13:58:34 -04002209 /*
2210 * we need that spin_lock here - it prevents reordering between
Jeff Layton8116bf42015-01-21 20:44:01 -05002211 * update of i_flctx->flc_posix and check for it done in close().
Al Viro0b2bac22008-05-06 13:58:34 -04002212 * rcu_read_lock() wouldn't do.
2213 */
2214 spin_lock(&current->files->file_lock);
2215 f = fcheck(fd);
2216 spin_unlock(&current->files->file_lock);
2217 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002218 flock.l_type = F_UNLCK;
2219 goto again;
2220 }
2221
2222out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 locks_free_lock(file_lock);
2224 return error;
2225}
2226
2227#if BITS_PER_LONG == 32
2228/* Report the first existing lock that would conflict with l.
2229 * This implements the F_GETLK command of fcntl().
2230 */
Jeff Laytonc1e62b82014-02-03 12:13:09 -05002231int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002233 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 struct flock64 flock;
2235 int error;
2236
2237 error = -EFAULT;
2238 if (copy_from_user(&flock, l, sizeof(flock)))
2239 goto out;
2240 error = -EINVAL;
2241 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2242 goto out;
2243
2244 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2245 if (error)
2246 goto out;
2247
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002248 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05002249 error = -EINVAL;
2250 if (flock.l_pid != 0)
2251 goto out;
2252
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002253 cmd = F_GETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002254 file_lock.fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002255 file_lock.fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002256 }
2257
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002258 error = vfs_test_lock(filp, &file_lock);
2259 if (error)
2260 goto out;
2261
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002262 flock.l_type = file_lock.fl_type;
2263 if (file_lock.fl_type != F_UNLCK)
2264 posix_lock_to_flock64(&flock, &file_lock);
2265
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 error = -EFAULT;
2267 if (!copy_to_user(l, &flock, sizeof(flock)))
2268 error = 0;
Kinglong Meef3282962014-08-22 10:18:43 -04002269
2270 locks_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271out:
2272 return error;
2273}
2274
2275/* Apply the lock described by l to an open file descriptor.
2276 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2277 */
Peter Staubachc2936212005-07-27 11:45:09 -07002278int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2279 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280{
2281 struct file_lock *file_lock = locks_alloc_lock();
2282 struct flock64 flock;
2283 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002284 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 int error;
2286
2287 if (file_lock == NULL)
2288 return -ENOLCK;
2289
2290 /*
2291 * This might block, so we do it before checking the inode.
2292 */
2293 error = -EFAULT;
2294 if (copy_from_user(&flock, l, sizeof(flock)))
2295 goto out;
2296
Al Viro496ad9a2013-01-23 17:07:38 -05002297 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
2299 /* Don't allow mandatory locks on files that may be memory mapped
2300 * and shared.
2301 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002302 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 error = -EAGAIN;
2304 goto out;
2305 }
2306
Peter Staubachc2936212005-07-27 11:45:09 -07002307again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 error = flock64_to_posix_lock(filp, file_lock, &flock);
2309 if (error)
2310 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002311
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002312 error = check_fmode_for_setlk(file_lock);
2313 if (error)
2314 goto out;
2315
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002316 /*
2317 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002318 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002319 */
2320 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002321 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002322 error = -EINVAL;
2323 if (flock.l_pid != 0)
2324 goto out;
2325
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002326 cmd = F_SETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002327 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002328 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002329 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002330 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002331 error = -EINVAL;
2332 if (flock.l_pid != 0)
2333 goto out;
2334
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002335 cmd = F_SETLKW64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002336 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002337 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002338 /* Fallthrough */
2339 case F_SETLKW64:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 file_lock->fl_flags |= FL_SLEEP;
2341 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002342
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002343 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344
Peter Staubachc2936212005-07-27 11:45:09 -07002345 /*
2346 * Attempt to detect a close/fcntl race and recover by
2347 * releasing the lock that was just acquired.
2348 */
Al Viro0b2bac22008-05-06 13:58:34 -04002349 spin_lock(&current->files->file_lock);
2350 f = fcheck(fd);
2351 spin_unlock(&current->files->file_lock);
2352 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002353 flock.l_type = F_UNLCK;
2354 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 }
2356
2357out:
2358 locks_free_lock(file_lock);
2359 return error;
2360}
2361#endif /* BITS_PER_LONG == 32 */
2362
2363/*
2364 * This function is called when the file is being removed
2365 * from the task's fd array. POSIX locks belonging to this task
2366 * are deleted at this time.
2367 */
2368void locks_remove_posix(struct file *filp, fl_owner_t owner)
2369{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002370 struct file_lock lock;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05002371 struct file_lock_context *ctx = file_inode(filp)->i_flctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372
2373 /*
2374 * If there are no locks held on this file, we don't need to call
2375 * posix_lock_file(). Another process could be setting a lock on this
2376 * file at the same time, but we wouldn't remove that lock anyway.
2377 */
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05002378 if (!ctx || list_empty(&ctx->flc_posix))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 return;
2380
2381 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002382 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 lock.fl_start = 0;
2384 lock.fl_end = OFFSET_MAX;
2385 lock.fl_owner = owner;
2386 lock.fl_pid = current->tgid;
2387 lock.fl_file = filp;
2388 lock.fl_ops = NULL;
2389 lock.fl_lmops = NULL;
2390
Marc Eshel150b3932007-01-18 16:15:35 -05002391 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2394 lock.fl_ops->fl_release_private(&lock);
2395}
2396
2397EXPORT_SYMBOL(locks_remove_posix);
2398
Jeff Layton3d8e5602015-01-16 15:05:58 -05002399/* The i_flctx must be valid when calling into here */
Jeff Laytondd459bb2015-01-16 15:05:54 -05002400static void
2401locks_remove_flock(struct file *filp)
2402{
2403 struct file_lock fl = {
2404 .fl_owner = filp,
2405 .fl_pid = current->tgid,
2406 .fl_file = filp,
2407 .fl_flags = FL_FLOCK,
2408 .fl_type = F_UNLCK,
2409 .fl_end = OFFSET_MAX,
2410 };
Jeff Layton5263e312015-01-16 15:05:55 -05002411 struct file_lock_context *flctx = file_inode(filp)->i_flctx;
Jeff Laytondd459bb2015-01-16 15:05:54 -05002412
Jeff Layton3d8e5602015-01-16 15:05:58 -05002413 if (list_empty(&flctx->flc_flock))
Jeff Laytondd459bb2015-01-16 15:05:54 -05002414 return;
2415
2416 if (filp->f_op->flock)
2417 filp->f_op->flock(filp, F_SETLKW, &fl);
2418 else
2419 flock_lock_file(filp, &fl);
2420
2421 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2422 fl.fl_ops->fl_release_private(&fl);
2423}
2424
Jeff Layton3d8e5602015-01-16 15:05:58 -05002425/* The i_flctx must be valid when calling into here */
Jeff Layton8634b512015-01-16 15:05:55 -05002426static void
2427locks_remove_lease(struct file *filp)
2428{
2429 struct inode *inode = file_inode(filp);
2430 struct file_lock_context *ctx = inode->i_flctx;
2431 struct file_lock *fl, *tmp;
2432 LIST_HEAD(dispose);
2433
Jeff Layton3d8e5602015-01-16 15:05:58 -05002434 if (list_empty(&ctx->flc_lease))
Jeff Layton8634b512015-01-16 15:05:55 -05002435 return;
2436
Jeff Layton6109c852015-01-16 15:05:57 -05002437 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05002438 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
Jeff Laytonc4e136c2015-02-16 19:37:42 -05002439 if (filp == fl->fl_file)
2440 lease_modify(fl, F_UNLCK, &dispose);
Jeff Layton6109c852015-01-16 15:05:57 -05002441 spin_unlock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05002442 locks_dispose_list(&dispose);
2443}
2444
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445/*
2446 * This function is called on the last close of an open file.
2447 */
Jeff Layton78ed8a12014-02-03 12:13:08 -05002448void locks_remove_file(struct file *filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449{
Jeff Layton3d8e5602015-01-16 15:05:58 -05002450 if (!file_inode(filp)->i_flctx)
2451 return;
2452
Jeff Laytondd459bb2015-01-16 15:05:54 -05002453 /* remove any OFD locks */
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002454 locks_remove_posix(filp, filp);
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002455
Jeff Laytondd459bb2015-01-16 15:05:54 -05002456 /* remove flock locks */
2457 locks_remove_flock(filp);
2458
Jeff Layton8634b512015-01-16 15:05:55 -05002459 /* remove any leases */
2460 locks_remove_lease(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461}
2462
2463/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 * posix_unblock_lock - stop waiting for a file lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 * @waiter: the lock which was waiting
2466 *
2467 * lockd needs to block waiting for locks.
2468 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002469int
Jeff Laytonf891a292013-06-21 08:58:09 -04002470posix_unblock_lock(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002472 int status = 0;
2473
Jeff Layton7b2296a2013-06-21 08:58:20 -04002474 spin_lock(&blocked_lock_lock);
J. Bruce Fields5996a292006-01-03 09:55:44 +01002475 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002477 else
2478 status = -ENOENT;
Jeff Layton7b2296a2013-06-21 08:58:20 -04002479 spin_unlock(&blocked_lock_lock);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002480 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482EXPORT_SYMBOL(posix_unblock_lock);
2483
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002484/**
2485 * vfs_cancel_lock - file byte range unblock lock
2486 * @filp: The file to apply the unblock to
2487 * @fl: The lock to be unblocked
2488 *
2489 * Used by lock managers to cancel blocked requests
2490 */
2491int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2492{
Al Viro72c2d532013-09-22 16:27:52 -04002493 if (filp->f_op->lock)
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002494 return filp->f_op->lock(filp, F_CANCELLK, fl);
2495 return 0;
2496}
2497
2498EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2499
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002500#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002501#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002502#include <linux/seq_file.h>
2503
Jeff Layton7012b022013-06-21 08:58:22 -04002504struct locks_iterator {
2505 int li_cpu;
2506 loff_t li_pos;
2507};
2508
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002509static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002510 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511{
2512 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002513 unsigned int fl_pid;
2514
2515 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002516 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002517 else
2518 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519
2520 if (fl->fl_file != NULL)
Al Viro496ad9a2013-01-23 17:07:38 -05002521 inode = file_inode(fl->fl_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522
Jerome Marchand99dc8292010-10-26 14:22:33 -07002523 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 if (IS_POSIX(fl)) {
Jeff Laytonc918d422014-02-03 12:13:09 -05002525 if (fl->fl_flags & FL_ACCESS)
Fabian Frederick5315c262014-05-09 14:13:05 -04002526 seq_puts(f, "ACCESS");
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002527 else if (IS_OFDLCK(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002528 seq_puts(f, "OFDLCK");
Jeff Laytonc918d422014-02-03 12:13:09 -05002529 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002530 seq_puts(f, "POSIX ");
Jeff Laytonc918d422014-02-03 12:13:09 -05002531
2532 seq_printf(f, " %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002534 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 } else if (IS_FLOCK(fl)) {
2536 if (fl->fl_type & LOCK_MAND) {
Fabian Frederick5315c262014-05-09 14:13:05 -04002537 seq_puts(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002539 seq_puts(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 }
2541 } else if (IS_LEASE(fl)) {
Jeff Layton8144f1f2014-08-11 13:36:54 -04002542 if (fl->fl_flags & FL_DELEG)
2543 seq_puts(f, "DELEG ");
2544 else
2545 seq_puts(f, "LEASE ");
2546
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002547 if (lease_breaking(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002548 seq_puts(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 else if (fl->fl_file)
Fabian Frederick5315c262014-05-09 14:13:05 -04002550 seq_puts(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002552 seq_puts(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002554 seq_puts(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 }
2556 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002557 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 (fl->fl_type & LOCK_READ)
2559 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2560 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2561 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002562 seq_printf(f, "%s ",
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002563 (lease_breaking(fl))
Jeff Layton0ee5c6d2012-08-02 15:46:30 -04002564 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2565 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 }
2567 if (inode) {
2568#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002569 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 inode->i_sb->s_id, inode->i_ino);
2571#else
2572 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002573 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 MAJOR(inode->i_sb->s_dev),
2575 MINOR(inode->i_sb->s_dev), inode->i_ino);
2576#endif
2577 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002578 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 }
2580 if (IS_POSIX(fl)) {
2581 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002582 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002584 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002586 seq_puts(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 }
2588}
2589
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002590static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591{
Jeff Layton7012b022013-06-21 08:58:22 -04002592 struct locks_iterator *iter = f->private;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002593 struct file_lock *fl, *bfl;
2594
Jeff Layton139ca042013-06-21 08:58:17 -04002595 fl = hlist_entry(v, struct file_lock, fl_link);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002596
Jeff Layton7012b022013-06-21 08:58:22 -04002597 lock_get_status(f, fl, iter->li_pos, "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002598
2599 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jeff Layton7012b022013-06-21 08:58:22 -04002600 lock_get_status(f, bfl, iter->li_pos, " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002601
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002602 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603}
2604
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002605static void *locks_start(struct seq_file *f, loff_t *pos)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002606 __acquires(&blocked_lock_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607{
Jeff Layton7012b022013-06-21 08:58:22 -04002608 struct locks_iterator *iter = f->private;
Jerome Marchand99dc8292010-10-26 14:22:33 -07002609
Jeff Layton7012b022013-06-21 08:58:22 -04002610 iter->li_pos = *pos + 1;
2611 lg_global_lock(&file_lock_lglock);
Jeff Layton7b2296a2013-06-21 08:58:20 -04002612 spin_lock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002613 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614}
2615
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002616static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2617{
Jeff Layton7012b022013-06-21 08:58:22 -04002618 struct locks_iterator *iter = f->private;
2619
2620 ++iter->li_pos;
2621 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002622}
2623
2624static void locks_stop(struct seq_file *f, void *v)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002625 __releases(&blocked_lock_lock)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002626{
Jeff Layton7b2296a2013-06-21 08:58:20 -04002627 spin_unlock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002628 lg_global_unlock(&file_lock_lglock);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002629}
2630
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002631static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002632 .start = locks_start,
2633 .next = locks_next,
2634 .stop = locks_stop,
2635 .show = locks_show,
2636};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002637
2638static int locks_open(struct inode *inode, struct file *filp)
2639{
Jeff Layton7012b022013-06-21 08:58:22 -04002640 return seq_open_private(filp, &locks_seq_operations,
2641 sizeof(struct locks_iterator));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002642}
2643
2644static const struct file_operations proc_locks_operations = {
2645 .open = locks_open,
2646 .read = seq_read,
2647 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002648 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002649};
2650
2651static int __init proc_locks_init(void)
2652{
2653 proc_create("locks", 0, NULL, &proc_locks_operations);
2654 return 0;
2655}
2656module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002657#endif
2658
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659static int __init filelock_init(void)
2660{
Jeff Layton7012b022013-06-21 08:58:22 -04002661 int i;
2662
Jeff Layton4a075e32015-01-16 15:05:54 -05002663 flctx_cache = kmem_cache_create("file_lock_ctx",
2664 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2665
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 filelock_cache = kmem_cache_create("file_lock_cache",
Miklos Szerediee19cc42011-07-07 13:06:09 +02002667 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2668
Jeff Layton7012b022013-06-21 08:58:22 -04002669 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2670
2671 for_each_possible_cpu(i)
2672 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2673
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 return 0;
2675}
2676
2677core_initcall(filelock_init);