blob: df8b26a425248ac4aa87a4ef9772b9700b417209 [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/fs/locks.c
4 *
J. Bruce Fieldse9728cc2021-10-19 13:38:35 -04005 * We implement four types of file locks: BSD locks, posix locks, open
6 * file description locks, and leases. For details about BSD locks,
7 * see the flock(2) man page; for details about the other three, see
8 * fcntl(2).
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
NeilBrownfd7732e2018-11-30 10:04:08 +110010 *
11 * Locking conflicts and dependencies:
12 * If multiple threads attempt to lock the same byte (or flock the same file)
13 * only one can be granted the lock, and other must wait their turn.
14 * The first lock has been "applied" or "granted", the others are "waiting"
15 * and are "blocked" by the "applied" lock..
16 *
17 * Waiting and applied locks are all kept in trees whose properties are:
18 *
19 * - the root of a tree may be an applied or waiting lock.
20 * - every other node in the tree is a waiting lock that
21 * conflicts with every ancestor of that node.
22 *
23 * Every such tree begins life as a waiting singleton which obviously
24 * satisfies the above properties.
25 *
26 * The only ways we modify trees preserve these properties:
27 *
28 * 1. We may add a new leaf node, but only after first verifying that it
29 * conflicts with all of its ancestors.
30 * 2. We may remove the root of a tree, creating a new singleton
31 * tree from the root and N new trees rooted in the immediate
32 * children.
33 * 3. If the root of a tree is not currently an applied lock, we may
34 * apply it (if possible).
35 * 4. We may upgrade the root of the tree (either extend its range,
36 * or upgrade its entire range from read to write).
37 *
38 * When an applied lock is modified in a way that reduces or downgrades any
39 * part of its range, we remove all its children (2 above). This particularly
40 * happens when a lock is unlocked.
41 *
42 * For each of those child trees we "wake up" the thread which is
43 * waiting for the lock so it can continue handling as follows: if the
44 * root of the tree applies, we do so (3). If it doesn't, it must
45 * conflict with some applied lock. We remove (wake up) all of its children
46 * (2), and add it is a new leaf to the tree rooted in the applied
47 * lock (1). We then repeat the process recursively with those
48 * children.
49 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 */
51
52#include <linux/capability.h>
53#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040054#include <linux/fdtable.h>
Jeff Layton5970e152022-11-20 09:15:34 -050055#include <linux/filelock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#include <linux/fs.h>
57#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#include <linux/security.h>
59#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <linux/syscalls.h>
61#include <linux/time.h>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -070062#include <linux/rcupdate.h>
Vitaliy Gusevab1f1612008-01-17 00:07:08 +000063#include <linux/pid_namespace.h>
Jeff Layton48f74182013-06-21 08:58:18 -040064#include <linux/hashtable.h>
Jeff Layton7012b022013-06-21 08:58:22 -040065#include <linux/percpu.h>
Luis Chamberlaindd81faa2022-01-21 22:13:10 -080066#include <linux/sysctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
Jeff Layton62af4f12014-05-09 14:13:05 -040068#define CREATE_TRACE_POINTS
69#include <trace/events/filelock.h>
70
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080071#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
74#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
Christoph Hellwig11afe9f2015-01-21 19:17:03 +010075#define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
Jeff Laytoncff2fce2014-04-22 08:24:32 -040076#define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
Benjamin Coddington9d5b86a2017-07-16 10:28:22 -040077#define IS_REMOTELCK(fl) (fl->fl_pid <= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -040079static bool lease_breaking(struct file_lock *fl)
80{
J. Bruce Fields778fc542011-07-26 18:25:49 -040081 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
82}
83
84static int target_leasetype(struct file_lock *fl)
85{
86 if (fl->fl_flags & FL_UNLOCK_PENDING)
87 return F_UNLCK;
88 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
89 return F_RDLCK;
90 return fl->fl_type;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -040091}
92
Luis Chamberlaindd81faa2022-01-21 22:13:10 -080093static int leases_enable = 1;
94static int lease_break_time = 45;
95
96#ifdef CONFIG_SYSCTL
97static struct ctl_table locks_sysctls[] = {
98 {
99 .procname = "leases-enable",
100 .data = &leases_enable,
101 .maxlen = sizeof(int),
102 .mode = 0644,
103 .proc_handler = proc_dointvec,
104 },
105#ifdef CONFIG_MMU
106 {
107 .procname = "lease-break-time",
108 .data = &lease_break_time,
109 .maxlen = sizeof(int),
110 .mode = 0644,
111 .proc_handler = proc_dointvec,
112 },
113#endif /* CONFIG_MMU */
114 {}
115};
116
117static int __init init_fs_locks_sysctls(void)
118{
119 register_sysctl_init("fs", locks_sysctls);
120 return 0;
121}
122early_initcall(init_fs_locks_sysctls);
123#endif /* CONFIG_SYSCTL */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Jeff Layton1c8c6012013-06-21 08:58:15 -0400125/*
Jeff Layton7012b022013-06-21 08:58:22 -0400126 * The global file_lock_list is only used for displaying /proc/locks, so we
Peter Zijlstra7c3f6542015-06-22 14:16:34 +0200127 * keep a list on each CPU, with each list protected by its own spinlock.
128 * Global serialization is done using file_rwsem.
129 *
130 * Note that alterations to the list also require that the relevant flc_lock is
131 * held.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400132 */
Peter Zijlstra7c3f6542015-06-22 14:16:34 +0200133struct file_lock_list_struct {
134 spinlock_t lock;
135 struct hlist_head hlist;
136};
137static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
Peter Zijlstraaba37662015-06-22 14:16:33 +0200138DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
Jeff Layton88974692013-06-21 08:58:14 -0400139
Jeff Laytoneb82dd32019-08-18 14:18:53 -0400140
Jeff Layton1c8c6012013-06-21 08:58:15 -0400141/*
Jeff Layton48f74182013-06-21 08:58:18 -0400142 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
Jeff Layton7b2296a2013-06-21 08:58:20 -0400143 * It is protected by blocked_lock_lock.
Jeff Layton48f74182013-06-21 08:58:18 -0400144 *
145 * We hash locks by lockowner in order to optimize searching for the lock a
146 * particular lockowner is waiting on.
147 *
148 * FIXME: make this value scale via some heuristic? We generally will want more
149 * buckets when we have more lockowners holding locks, but that's a little
150 * difficult to determine without knowing what the workload will look like.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400151 */
Jeff Layton48f74182013-06-21 08:58:18 -0400152#define BLOCKED_HASH_BITS 7
153static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
Jeff Layton88974692013-06-21 08:58:14 -0400154
Jeff Layton1c8c6012013-06-21 08:58:15 -0400155/*
Jeff Layton7b2296a2013-06-21 08:58:20 -0400156 * This lock protects the blocked_hash. Generally, if you're accessing it, you
157 * want to be holding this lock.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400158 *
NeilBrownada5c1d2018-11-30 10:04:08 +1100159 * In addition, it also protects the fl->fl_blocked_requests list, and the
160 * fl->fl_blocker pointer for file_lock structures that are acting as lock
161 * requests (in contrast to those that are acting as records of acquired locks).
Jeff Layton1c8c6012013-06-21 08:58:15 -0400162 *
163 * Note that when we acquire this lock in order to change the above fields,
Jeff Layton6109c852015-01-16 15:05:57 -0500164 * we often hold the flc_lock as well. In certain cases, when reading the fields
Jeff Layton1c8c6012013-06-21 08:58:15 -0400165 * protected by this lock, we can skip acquiring it iff we already hold the
Jeff Layton6109c852015-01-16 15:05:57 -0500166 * flc_lock.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400167 */
Jeff Layton7b2296a2013-06-21 08:58:20 -0400168static DEFINE_SPINLOCK(blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
Jeff Layton4a075e32015-01-16 15:05:54 -0500170static struct kmem_cache *flctx_cache __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -0800171static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Jeff Layton4a075e32015-01-16 15:05:54 -0500173static struct file_lock_context *
Jeff Layton5c1c6692015-04-03 09:04:03 -0400174locks_get_lock_context(struct inode *inode, int type)
Jeff Layton4a075e32015-01-16 15:05:54 -0500175{
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200176 struct file_lock_context *ctx;
Jeff Layton4a075e32015-01-16 15:05:54 -0500177
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200178 /* paired with cmpxchg() below */
Jeff Layton401a8b82022-11-16 09:02:30 -0500179 ctx = locks_inode_context(inode);
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200180 if (likely(ctx) || type == F_UNLCK)
Jeff Layton4a075e32015-01-16 15:05:54 -0500181 goto out;
182
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200183 ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
184 if (!ctx)
Jeff Layton4a075e32015-01-16 15:05:54 -0500185 goto out;
186
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200187 spin_lock_init(&ctx->flc_lock);
188 INIT_LIST_HEAD(&ctx->flc_flock);
189 INIT_LIST_HEAD(&ctx->flc_posix);
190 INIT_LIST_HEAD(&ctx->flc_lease);
Jeff Layton4a075e32015-01-16 15:05:54 -0500191
192 /*
193 * Assign the pointer if it's not already assigned. If it is, then
194 * free the context we just allocated.
195 */
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200196 if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
197 kmem_cache_free(flctx_cache, ctx);
Jeff Layton401a8b82022-11-16 09:02:30 -0500198 ctx = locks_inode_context(inode);
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200199 }
Jeff Layton4a075e32015-01-16 15:05:54 -0500200out:
Jeff Layton18909102016-01-06 21:26:10 -0500201 trace_locks_get_lock_context(inode, type, ctx);
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200202 return ctx;
Jeff Layton4a075e32015-01-16 15:05:54 -0500203}
204
Jeff Laytone24dada2016-01-06 21:28:41 -0500205static void
206locks_dump_ctx_list(struct list_head *list, char *list_type)
207{
208 struct file_lock *fl;
209
210 list_for_each_entry(fl, list, fl_list) {
211 pr_warn("%s: fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n", list_type, fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
212 }
213}
214
215static void
216locks_check_ctx_lists(struct inode *inode)
217{
218 struct file_lock_context *ctx = inode->i_flctx;
219
220 if (unlikely(!list_empty(&ctx->flc_flock) ||
221 !list_empty(&ctx->flc_posix) ||
222 !list_empty(&ctx->flc_lease))) {
223 pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
224 MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
225 inode->i_ino);
226 locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
227 locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
228 locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
229 }
230}
231
Benjamin Coddington39537042017-07-21 13:36:25 -0400232static void
233locks_check_ctx_file_list(struct file *filp, struct list_head *list,
234 char *list_type)
235{
236 struct file_lock *fl;
Jeff Laytonc65454a2022-11-25 08:48:37 -0500237 struct inode *inode = file_inode(filp);
Benjamin Coddington39537042017-07-21 13:36:25 -0400238
239 list_for_each_entry(fl, list, fl_list)
240 if (fl->fl_file == filp)
241 pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
242 " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
243 list_type, MAJOR(inode->i_sb->s_dev),
244 MINOR(inode->i_sb->s_dev), inode->i_ino,
245 fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
246}
247
Jeff Layton4a075e32015-01-16 15:05:54 -0500248void
Jeff Laytonf27a0fe2016-01-07 15:08:51 -0500249locks_free_lock_context(struct inode *inode)
Jeff Layton4a075e32015-01-16 15:05:54 -0500250{
Jeff Layton401a8b82022-11-16 09:02:30 -0500251 struct file_lock_context *ctx = locks_inode_context(inode);
Jeff Laytonf27a0fe2016-01-07 15:08:51 -0500252
Jeff Laytone24dada2016-01-06 21:28:41 -0500253 if (unlikely(ctx)) {
254 locks_check_ctx_lists(inode);
Jeff Layton4a075e32015-01-16 15:05:54 -0500255 kmem_cache_free(flctx_cache, ctx);
256 }
257}
258
Miklos Szerediee19cc42011-07-07 13:06:09 +0200259static void locks_init_lock_heads(struct file_lock *fl)
Miklos Szeredia51cb91d2011-07-06 12:33:55 +0200260{
Jeff Layton139ca042013-06-21 08:58:17 -0400261 INIT_HLIST_NODE(&fl->fl_link);
Jeff Layton6dee60f2015-01-16 15:05:54 -0500262 INIT_LIST_HEAD(&fl->fl_list);
NeilBrownada5c1d2018-11-30 10:04:08 +1100263 INIT_LIST_HEAD(&fl->fl_blocked_requests);
264 INIT_LIST_HEAD(&fl->fl_blocked_member);
Miklos Szerediee19cc42011-07-07 13:06:09 +0200265 init_waitqueue_head(&fl->fl_wait);
Miklos Szeredia51cb91d2011-07-06 12:33:55 +0200266}
267
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268/* Allocate an empty lock structure. */
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200269struct file_lock *locks_alloc_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200271 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
Miklos Szeredia51cb91d2011-07-06 12:33:55 +0200272
273 if (fl)
Miklos Szerediee19cc42011-07-07 13:06:09 +0200274 locks_init_lock_heads(fl);
Miklos Szeredia51cb91d2011-07-06 12:33:55 +0200275
276 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277}
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200278EXPORT_SYMBOL_GPL(locks_alloc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500280void locks_release_private(struct file_lock *fl)
Trond Myklebust47831f32006-03-20 13:44:05 -0500281{
NeilBrown59264592019-04-24 12:00:08 +1000282 BUG_ON(waitqueue_active(&fl->fl_wait));
283 BUG_ON(!list_empty(&fl->fl_list));
284 BUG_ON(!list_empty(&fl->fl_blocked_requests));
285 BUG_ON(!list_empty(&fl->fl_blocked_member));
286 BUG_ON(!hlist_unhashed(&fl->fl_link));
287
Trond Myklebust47831f32006-03-20 13:44:05 -0500288 if (fl->fl_ops) {
289 if (fl->fl_ops->fl_release_private)
290 fl->fl_ops->fl_release_private(fl);
291 fl->fl_ops = NULL;
292 }
Trond Myklebust47831f32006-03-20 13:44:05 -0500293
Kinglong Mee5c97d7b2014-08-22 10:18:43 -0400294 if (fl->fl_lmops) {
Jeff Laytoncae80b32015-04-03 09:04:04 -0400295 if (fl->fl_lmops->lm_put_owner) {
296 fl->fl_lmops->lm_put_owner(fl->fl_owner);
297 fl->fl_owner = NULL;
298 }
Kinglong Mee5c97d7b2014-08-22 10:18:43 -0400299 fl->fl_lmops = NULL;
300 }
Trond Myklebust47831f32006-03-20 13:44:05 -0500301}
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500302EXPORT_SYMBOL_GPL(locks_release_private);
Trond Myklebust47831f32006-03-20 13:44:05 -0500303
Dai Ngo591502c52022-05-02 14:19:24 -0700304/**
305 * locks_owner_has_blockers - Check for blocking lock requests
306 * @flctx: file lock context
307 * @owner: lock owner
308 *
309 * Return values:
310 * %true: @owner has at least one blocker
311 * %false: @owner has no blockers
312 */
313bool locks_owner_has_blockers(struct file_lock_context *flctx,
314 fl_owner_t owner)
315{
316 struct file_lock *fl;
317
318 spin_lock(&flctx->flc_lock);
319 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
320 if (fl->fl_owner != owner)
321 continue;
322 if (!list_empty(&fl->fl_blocked_requests)) {
323 spin_unlock(&flctx->flc_lock);
324 return true;
325 }
326 }
327 spin_unlock(&flctx->flc_lock);
328 return false;
329}
330EXPORT_SYMBOL_GPL(locks_owner_has_blockers);
331
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332/* Free a lock which is not in use. */
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400333void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
Trond Myklebust47831f32006-03-20 13:44:05 -0500335 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 kmem_cache_free(filelock_cache, fl);
337}
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400338EXPORT_SYMBOL(locks_free_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Jeff Laytoned9814d2014-08-11 14:20:31 -0400340static void
341locks_dispose_list(struct list_head *dispose)
342{
343 struct file_lock *fl;
344
345 while (!list_empty(dispose)) {
Jeff Layton6dee60f2015-01-16 15:05:54 -0500346 fl = list_first_entry(dispose, struct file_lock, fl_list);
347 list_del_init(&fl->fl_list);
Jeff Laytoned9814d2014-08-11 14:20:31 -0400348 locks_free_lock(fl);
349 }
350}
351
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352void locks_init_lock(struct file_lock *fl)
353{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200354 memset(fl, 0, sizeof(struct file_lock));
355 locks_init_lock_heads(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357EXPORT_SYMBOL(locks_init_lock);
358
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359/*
360 * Initialize a new lock from an existing file_lock structure.
361 */
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400362void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363{
364 new->fl_owner = fl->fl_owner;
365 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500366 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 new->fl_flags = fl->fl_flags;
368 new->fl_type = fl->fl_type;
369 new->fl_start = fl->fl_start;
370 new->fl_end = fl->fl_end;
Kinglong Meef3282962014-08-22 10:18:43 -0400371 new->fl_lmops = fl->fl_lmops;
Trond Myklebust09969052006-03-20 13:44:38 -0500372 new->fl_ops = NULL;
Kinglong Meef3282962014-08-22 10:18:43 -0400373
374 if (fl->fl_lmops) {
375 if (fl->fl_lmops->lm_get_owner)
Jeff Laytoncae80b32015-04-03 09:04:04 -0400376 fl->fl_lmops->lm_get_owner(fl->fl_owner);
Kinglong Meef3282962014-08-22 10:18:43 -0400377 }
Trond Myklebust09969052006-03-20 13:44:38 -0500378}
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400379EXPORT_SYMBOL(locks_copy_conflock);
Trond Myklebust09969052006-03-20 13:44:38 -0500380
381void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
382{
Jeff Layton566709b2014-08-11 14:09:35 -0400383 /* "new" must be a freshly-initialized lock */
384 WARN_ON_ONCE(new->fl_ops);
Trond Myklebust09969052006-03-20 13:44:38 -0500385
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400386 locks_copy_conflock(new, fl);
Kinglong Meef3282962014-08-22 10:18:43 -0400387
Trond Myklebust09969052006-03-20 13:44:38 -0500388 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 new->fl_ops = fl->fl_ops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500390
Kinglong Meef3282962014-08-22 10:18:43 -0400391 if (fl->fl_ops) {
392 if (fl->fl_ops->fl_copy_lock)
393 fl->fl_ops->fl_copy_lock(new, fl);
394 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396EXPORT_SYMBOL(locks_copy_lock);
397
NeilBrown5946c432018-11-30 10:04:08 +1100398static void locks_move_blocks(struct file_lock *new, struct file_lock *fl)
399{
400 struct file_lock *f;
401
402 /*
403 * As ctx->flc_lock is held, new requests cannot be added to
404 * ->fl_blocked_requests, so we don't need a lock to check if it
405 * is empty.
406 */
407 if (list_empty(&fl->fl_blocked_requests))
408 return;
409 spin_lock(&blocked_lock_lock);
410 list_splice_init(&fl->fl_blocked_requests, &new->fl_blocked_requests);
NeilBrownbf77ae42019-01-03 11:04:08 +1100411 list_for_each_entry(f, &new->fl_blocked_requests, fl_blocked_member)
NeilBrown5946c432018-11-30 10:04:08 +1100412 f->fl_blocker = new;
413 spin_unlock(&blocked_lock_lock);
414}
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416static inline int flock_translate_cmd(int cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 switch (cmd) {
418 case LOCK_SH:
419 return F_RDLCK;
420 case LOCK_EX:
421 return F_WRLCK;
422 case LOCK_UN:
423 return F_UNLCK;
424 }
425 return -EINVAL;
426}
427
428/* Fill in a file_lock structure with an appropriate FLOCK lock. */
Kuniyuki Iwashima4149be72022-07-16 21:35:31 -0700429static void flock_make_lock(struct file *filp, struct file_lock *fl, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430{
Kuniyuki Iwashima4149be72022-07-16 21:35:31 -0700431 locks_init_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
433 fl->fl_file = filp;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +0200434 fl->fl_owner = filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 fl->fl_pid = current->tgid;
436 fl->fl_flags = FL_FLOCK;
437 fl->fl_type = type;
438 fl->fl_end = OFFSET_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439}
440
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400441static int assign_type(struct file_lock *fl, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442{
443 switch (type) {
444 case F_RDLCK:
445 case F_WRLCK:
446 case F_UNLCK:
447 fl->fl_type = type;
448 break;
449 default:
450 return -EINVAL;
451 }
452 return 0;
453}
454
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500455static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
456 struct flock64 *l)
457{
458 switch (l->l_whence) {
459 case SEEK_SET:
460 fl->fl_start = 0;
461 break;
462 case SEEK_CUR:
463 fl->fl_start = filp->f_pos;
464 break;
465 case SEEK_END:
466 fl->fl_start = i_size_read(file_inode(filp));
467 break;
468 default:
469 return -EINVAL;
470 }
471 if (l->l_start > OFFSET_MAX - fl->fl_start)
472 return -EOVERFLOW;
473 fl->fl_start += l->l_start;
474 if (fl->fl_start < 0)
475 return -EINVAL;
476
477 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
478 POSIX-2001 defines it. */
479 if (l->l_len > 0) {
480 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
481 return -EOVERFLOW;
Luo Meng16238412020-10-23 14:20:05 +0800482 fl->fl_end = fl->fl_start + (l->l_len - 1);
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500483
484 } else if (l->l_len < 0) {
485 if (fl->fl_start + l->l_len < 0)
486 return -EINVAL;
487 fl->fl_end = fl->fl_start - 1;
488 fl->fl_start += l->l_len;
489 } else
490 fl->fl_end = OFFSET_MAX;
491
492 fl->fl_owner = current->files;
493 fl->fl_pid = current->tgid;
494 fl->fl_file = filp;
495 fl->fl_flags = FL_POSIX;
496 fl->fl_ops = NULL;
497 fl->fl_lmops = NULL;
498
499 return assign_type(fl, l->l_type);
500}
501
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
503 * style lock.
504 */
505static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
506 struct flock *l)
507{
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500508 struct flock64 ll = {
509 .l_type = l->l_type,
510 .l_whence = l->l_whence,
511 .l_start = l->l_start,
512 .l_len = l->l_len,
513 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500515 return flock64_to_posix_lock(filp, fl, &ll);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516}
517
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518/* default lease lock manager operations */
Jeff Layton4d01b7f2014-09-01 15:06:54 -0400519static bool
520lease_break_callback(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521{
522 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
Jeff Layton4d01b7f2014-09-01 15:06:54 -0400523 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524}
525
Jeff Layton1c7dd2f2014-08-22 10:55:47 -0400526static void
527lease_setup(struct file_lock *fl, void **priv)
528{
529 struct file *filp = fl->fl_file;
530 struct fasync_struct *fa = *priv;
531
532 /*
533 * fasync_insert_entry() returns the old entry if any. If there was no
534 * old entry, then it used "priv" and inserted it into the fasync list.
535 * Clear the pointer to indicate that it shouldn't be freed.
536 */
537 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
538 *priv = NULL;
539
Eric W. Biederman01919132017-07-16 22:05:57 -0500540 __f_setown(filp, task_pid(current), PIDTYPE_TGID, 0);
Jeff Layton1c7dd2f2014-08-22 10:55:47 -0400541}
542
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700543static const struct lock_manager_operations lease_manager_ops = {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400544 .lm_break = lease_break_callback,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400545 .lm_change = lease_modify,
Jeff Layton1c7dd2f2014-08-22 10:55:47 -0400546 .lm_setup = lease_setup,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547};
548
549/*
550 * Initialize a lease, use the default lock manager operations
551 */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400552static int lease_init(struct file *filp, long type, struct file_lock *fl)
Joe Perches447a5642018-03-21 15:09:32 -0700553{
Trond Myklebust75dff552006-05-07 23:02:42 -0400554 if (assign_type(fl, type) != 0)
555 return -EINVAL;
556
Jeff Layton7ca76312014-09-01 19:04:48 -0400557 fl->fl_owner = filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 fl->fl_pid = current->tgid;
559
560 fl->fl_file = filp;
561 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 fl->fl_start = 0;
563 fl->fl_end = OFFSET_MAX;
564 fl->fl_ops = NULL;
565 fl->fl_lmops = &lease_manager_ops;
566 return 0;
567}
568
569/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400570static struct file_lock *lease_alloc(struct file *filp, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571{
572 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400573 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
575 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500576 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400579 if (error) {
580 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500581 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400582 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500583 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584}
585
586/* Check if two locks overlap each other.
587 */
588static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
589{
590 return ((fl1->fl_end >= fl2->fl_start) &&
591 (fl2->fl_end >= fl1->fl_start));
592}
593
594/*
595 * Check whether two locks have the same owner.
596 */
Matt Mackall33443c42006-01-08 01:05:22 -0800597static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 return fl1->fl_owner == fl2->fl_owner;
600}
601
Jeff Layton6109c852015-01-16 15:05:57 -0500602/* Must be called with the flc_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500603static void locks_insert_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400604{
Peter Zijlstra7c3f6542015-06-22 14:16:34 +0200605 struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
606
Peter Zijlstraaba37662015-06-22 14:16:33 +0200607 percpu_rwsem_assert_held(&file_rwsem);
608
Peter Zijlstra7c3f6542015-06-22 14:16:34 +0200609 spin_lock(&fll->lock);
Jeff Layton7012b022013-06-21 08:58:22 -0400610 fl->fl_link_cpu = smp_processor_id();
Peter Zijlstra7c3f6542015-06-22 14:16:34 +0200611 hlist_add_head(&fl->fl_link, &fll->hlist);
612 spin_unlock(&fll->lock);
Jeff Layton88974692013-06-21 08:58:14 -0400613}
614
Jeff Layton6109c852015-01-16 15:05:57 -0500615/* Must be called with the flc_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500616static void locks_delete_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400617{
Peter Zijlstra7c3f6542015-06-22 14:16:34 +0200618 struct file_lock_list_struct *fll;
619
Peter Zijlstraaba37662015-06-22 14:16:33 +0200620 percpu_rwsem_assert_held(&file_rwsem);
621
Jeff Layton7012b022013-06-21 08:58:22 -0400622 /*
623 * Avoid taking lock if already unhashed. This is safe since this check
Jeff Layton6109c852015-01-16 15:05:57 -0500624 * is done while holding the flc_lock, and new insertions into the list
Jeff Layton7012b022013-06-21 08:58:22 -0400625 * also require that it be held.
626 */
627 if (hlist_unhashed(&fl->fl_link))
628 return;
Peter Zijlstra7c3f6542015-06-22 14:16:34 +0200629
630 fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
631 spin_lock(&fll->lock);
Jeff Layton139ca042013-06-21 08:58:17 -0400632 hlist_del_init(&fl->fl_link);
Peter Zijlstra7c3f6542015-06-22 14:16:34 +0200633 spin_unlock(&fll->lock);
Jeff Layton88974692013-06-21 08:58:14 -0400634}
635
Jeff Layton3999e492013-06-21 08:58:19 -0400636static unsigned long
637posix_owner_key(struct file_lock *fl)
638{
Jeff Layton3999e492013-06-21 08:58:19 -0400639 return (unsigned long)fl->fl_owner;
640}
641
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500642static void locks_insert_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400643{
Daniel Wagner663d5af2015-04-03 09:04:03 -0400644 lockdep_assert_held(&blocked_lock_lock);
645
Jeff Layton3999e492013-06-21 08:58:19 -0400646 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
Jeff Layton88974692013-06-21 08:58:14 -0400647}
648
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500649static void locks_delete_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400650{
Daniel Wagner663d5af2015-04-03 09:04:03 -0400651 lockdep_assert_held(&blocked_lock_lock);
652
Jeff Layton48f74182013-06-21 08:58:18 -0400653 hash_del(&waiter->fl_link);
Jeff Layton88974692013-06-21 08:58:14 -0400654}
655
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656/* Remove waiter from blocker's block list.
657 * When blocker ends up pointing to itself then the list is empty.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400658 *
Jeff Layton7b2296a2013-06-21 08:58:20 -0400659 * Must be called with blocked_lock_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 */
Matt Mackall33443c42006-01-08 01:05:22 -0800661static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662{
Jeff Layton88974692013-06-21 08:58:14 -0400663 locks_delete_global_blocked(waiter);
NeilBrownada5c1d2018-11-30 10:04:08 +1100664 list_del_init(&waiter->fl_blocked_member);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
666
NeilBrownad6bbd82018-11-30 10:04:08 +1100667static void __locks_wake_up_blocks(struct file_lock *blocker)
668{
669 while (!list_empty(&blocker->fl_blocked_requests)) {
670 struct file_lock *waiter;
671
672 waiter = list_first_entry(&blocker->fl_blocked_requests,
673 struct file_lock, fl_blocked_member);
674 __locks_delete_block(waiter);
675 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
676 waiter->fl_lmops->lm_notify(waiter);
677 else
678 wake_up(&waiter->fl_wait);
Linus Torvaldsdcf23ac2020-03-18 07:52:21 -0400679
680 /*
681 * The setting of fl_blocker to NULL marks the "done"
682 * point in deleting a block. Paired with acquire at the top
683 * of locks_delete_block().
684 */
685 smp_store_release(&waiter->fl_blocker, NULL);
NeilBrownad6bbd82018-11-30 10:04:08 +1100686 }
687}
688
NeilBrowncb03f942018-11-30 10:04:08 +1100689/**
Mauro Carvalho Chehab529adfe82020-10-23 18:33:23 +0200690 * locks_delete_block - stop waiting for a file lock
NeilBrowncb03f942018-11-30 10:04:08 +1100691 * @waiter: the lock which was waiting
692 *
693 * lockd/nfsd need to disconnect the lock while working on it.
694 */
695int locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
NeilBrowncb03f942018-11-30 10:04:08 +1100697 int status = -ENOENT;
698
Linus Torvaldsdcf23ac2020-03-18 07:52:21 -0400699 /*
700 * If fl_blocker is NULL, it won't be set again as this thread "owns"
701 * the lock and is the only one that might try to claim the lock.
702 *
703 * We use acquire/release to manage fl_blocker so that we can
704 * optimize away taking the blocked_lock_lock in many cases.
705 *
706 * The smp_load_acquire guarantees two things:
707 *
708 * 1/ that fl_blocked_requests can be tested locklessly. If something
709 * was recently added to that list it must have been in a locked region
710 * *before* the locked region when fl_blocker was set to NULL.
711 *
712 * 2/ that no other thread is accessing 'waiter', so it is safe to free
713 * it. __locks_wake_up_blocks is careful not to touch waiter after
714 * fl_blocker is released.
715 *
716 * If a lockless check of fl_blocker shows it to be NULL, we know that
717 * no new locks can be inserted into its fl_blocked_requests list, and
718 * can avoid doing anything further if the list is empty.
719 */
720 if (!smp_load_acquire(&waiter->fl_blocker) &&
721 list_empty(&waiter->fl_blocked_requests))
722 return status;
723
Jeff Layton7b2296a2013-06-21 08:58:20 -0400724 spin_lock(&blocked_lock_lock);
NeilBrowncb03f942018-11-30 10:04:08 +1100725 if (waiter->fl_blocker)
726 status = 0;
NeilBrown5946c432018-11-30 10:04:08 +1100727 __locks_wake_up_blocks(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 __locks_delete_block(waiter);
Linus Torvaldsdcf23ac2020-03-18 07:52:21 -0400729
730 /*
731 * The setting of fl_blocker to NULL marks the "done" point in deleting
732 * a block. Paired with acquire at the top of this function.
733 */
734 smp_store_release(&waiter->fl_blocker, NULL);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400735 spin_unlock(&blocked_lock_lock);
NeilBrowncb03f942018-11-30 10:04:08 +1100736 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737}
NeilBrowncb03f942018-11-30 10:04:08 +1100738EXPORT_SYMBOL(locks_delete_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
740/* Insert waiter into blocker's block list.
741 * We use a circular list so that processes can be easily woken up in
742 * the order they blocked. The documentation doesn't require this but
743 * it seems like the reasonable thing to do.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400744 *
Jeff Layton6109c852015-01-16 15:05:57 -0500745 * Must be called with both the flc_lock and blocked_lock_lock held. The
NeilBrownada5c1d2018-11-30 10:04:08 +1100746 * fl_blocked_requests list itself is protected by the blocked_lock_lock,
747 * but by ensuring that the flc_lock is also held on insertions we can avoid
748 * taking the blocked_lock_lock in some cases when we see that the
749 * fl_blocked_requests list is empty.
NeilBrownfd7732e2018-11-30 10:04:08 +1100750 *
751 * Rather than just adding to the list, we check for conflicts with any existing
752 * waiters, and add beneath any waiter that blocks the new waiter.
753 * Thus wakeups don't happen until needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 */
Jeff Layton1c8c6012013-06-21 08:58:15 -0400755static void __locks_insert_block(struct file_lock *blocker,
NeilBrownfd7732e2018-11-30 10:04:08 +1100756 struct file_lock *waiter,
757 bool conflict(struct file_lock *,
758 struct file_lock *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759{
NeilBrownfd7732e2018-11-30 10:04:08 +1100760 struct file_lock *fl;
NeilBrownada5c1d2018-11-30 10:04:08 +1100761 BUG_ON(!list_empty(&waiter->fl_blocked_member));
NeilBrownfd7732e2018-11-30 10:04:08 +1100762
763new_blocker:
764 list_for_each_entry(fl, &blocker->fl_blocked_requests, fl_blocked_member)
765 if (conflict(fl, waiter)) {
766 blocker = fl;
767 goto new_blocker;
768 }
NeilBrownada5c1d2018-11-30 10:04:08 +1100769 waiter->fl_blocker = blocker;
770 list_add_tail(&waiter->fl_blocked_member, &blocker->fl_blocked_requests);
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400771 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
Jeff Layton1c8c6012013-06-21 08:58:15 -0400772 locks_insert_global_blocked(waiter);
NeilBrown5946c432018-11-30 10:04:08 +1100773
774 /* The requests in waiter->fl_blocked are known to conflict with
775 * waiter, but might not conflict with blocker, or the requests
776 * and lock which block it. So they all need to be woken.
777 */
778 __locks_wake_up_blocks(waiter);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400779}
780
Jeff Layton6109c852015-01-16 15:05:57 -0500781/* Must be called with flc_lock held. */
Jeff Layton1c8c6012013-06-21 08:58:15 -0400782static void locks_insert_block(struct file_lock *blocker,
NeilBrownfd7732e2018-11-30 10:04:08 +1100783 struct file_lock *waiter,
784 bool conflict(struct file_lock *,
785 struct file_lock *))
Jeff Layton1c8c6012013-06-21 08:58:15 -0400786{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400787 spin_lock(&blocked_lock_lock);
NeilBrownfd7732e2018-11-30 10:04:08 +1100788 __locks_insert_block(blocker, waiter, conflict);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400789 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790}
791
Jeff Layton1cb36012013-06-21 08:58:12 -0400792/*
793 * Wake up processes blocked waiting for blocker.
794 *
Jeff Layton6109c852015-01-16 15:05:57 -0500795 * Must be called with the inode->flc_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 */
797static void locks_wake_up_blocks(struct file_lock *blocker)
798{
Jeff Layton4e8c7652013-06-21 08:58:16 -0400799 /*
800 * Avoid taking global lock if list is empty. This is safe since new
Jeff Layton6109c852015-01-16 15:05:57 -0500801 * blocked requests are only added to the list under the flc_lock, and
NeilBrownada5c1d2018-11-30 10:04:08 +1100802 * the flc_lock is always held here. Note that removal from the
803 * fl_blocked_requests list does not require the flc_lock, so we must
804 * recheck list_empty() after acquiring the blocked_lock_lock.
Jeff Layton4e8c7652013-06-21 08:58:16 -0400805 */
NeilBrownada5c1d2018-11-30 10:04:08 +1100806 if (list_empty(&blocker->fl_blocked_requests))
Jeff Layton4e8c7652013-06-21 08:58:16 -0400807 return;
808
Jeff Layton7b2296a2013-06-21 08:58:20 -0400809 spin_lock(&blocked_lock_lock);
NeilBrownad6bbd82018-11-30 10:04:08 +1100810 __locks_wake_up_blocks(blocker);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400811 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812}
813
Jeff Layton5263e312015-01-16 15:05:55 -0500814static void
Jeff Laytone084c1b2015-02-16 14:32:03 -0500815locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
Jeff Layton5263e312015-01-16 15:05:55 -0500816{
Jeff Layton5263e312015-01-16 15:05:55 -0500817 list_add_tail(&fl->fl_list, before);
818 locks_insert_global_locks(fl);
819}
820
Jeff Layton8634b512015-01-16 15:05:55 -0500821static void
Jeff Laytone084c1b2015-02-16 14:32:03 -0500822locks_unlink_lock_ctx(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823{
Jeff Layton88974692013-06-21 08:58:14 -0400824 locks_delete_global_locks(fl);
Jeff Layton8634b512015-01-16 15:05:55 -0500825 list_del_init(&fl->fl_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 locks_wake_up_blocks(fl);
Jeff Layton24cbe782014-02-03 12:13:06 -0500827}
828
Jeff Layton5263e312015-01-16 15:05:55 -0500829static void
Jeff Laytone084c1b2015-02-16 14:32:03 -0500830locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
Jeff Layton5263e312015-01-16 15:05:55 -0500831{
Jeff Laytone084c1b2015-02-16 14:32:03 -0500832 locks_unlink_lock_ctx(fl);
Jeff Layton8634b512015-01-16 15:05:55 -0500833 if (dispose)
834 list_add(&fl->fl_list, dispose);
835 else
836 locks_free_lock(fl);
Jeff Layton5263e312015-01-16 15:05:55 -0500837}
838
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
840 * checks for shared/exclusive status of overlapping locks.
841 */
NeilBrownc0e15902018-11-30 10:04:08 +1100842static bool locks_conflict(struct file_lock *caller_fl,
843 struct file_lock *sys_fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844{
845 if (sys_fl->fl_type == F_WRLCK)
NeilBrownc0e15902018-11-30 10:04:08 +1100846 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 if (caller_fl->fl_type == F_WRLCK)
NeilBrownc0e15902018-11-30 10:04:08 +1100848 return true;
849 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850}
851
852/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
853 * checking before calling the locks_conflict().
854 */
NeilBrownc0e15902018-11-30 10:04:08 +1100855static bool posix_locks_conflict(struct file_lock *caller_fl,
856 struct file_lock *sys_fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857{
858 /* POSIX locks owned by the same process do not conflict with
859 * each other.
860 */
Jeff Layton9b8c8692015-04-03 09:04:02 -0400861 if (posix_same_owner(caller_fl, sys_fl))
NeilBrownc0e15902018-11-30 10:04:08 +1100862 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864 /* Check whether they overlap */
865 if (!locks_overlap(caller_fl, sys_fl))
NeilBrownc0e15902018-11-30 10:04:08 +1100866 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
NeilBrownc0e15902018-11-30 10:04:08 +1100868 return locks_conflict(caller_fl, sys_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869}
870
871/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
872 * checking before calling the locks_conflict().
873 */
NeilBrownc0e15902018-11-30 10:04:08 +1100874static bool flock_locks_conflict(struct file_lock *caller_fl,
875 struct file_lock *sys_fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876{
877 /* FLOCK locks referring to the same filp do not conflict with
878 * each other.
879 */
Jeff Layton9b8c8692015-04-03 09:04:02 -0400880 if (caller_fl->fl_file == sys_fl->fl_file)
NeilBrownc0e15902018-11-30 10:04:08 +1100881 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
NeilBrownc0e15902018-11-30 10:04:08 +1100883 return locks_conflict(caller_fl, sys_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884}
885
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400886void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500887posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888{
889 struct file_lock *cfl;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500890 struct file_lock_context *ctx;
Jeff Laytonc65454a2022-11-25 08:48:37 -0500891 struct inode *inode = file_inode(filp);
Dai Ngo2443da22022-05-02 14:19:25 -0700892 void *owner;
893 void (*func)(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
Jeff Layton401a8b82022-11-16 09:02:30 -0500895 ctx = locks_inode_context(inode);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500896 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400897 fl->fl_type = F_UNLCK;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500898 return;
899 }
900
Dai Ngo2443da22022-05-02 14:19:25 -0700901retry:
Jeff Layton6109c852015-01-16 15:05:57 -0500902 spin_lock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500903 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
Dai Ngo2443da22022-05-02 14:19:25 -0700904 if (!posix_locks_conflict(fl, cfl))
905 continue;
906 if (cfl->fl_lmops && cfl->fl_lmops->lm_lock_expirable
907 && (*cfl->fl_lmops->lm_lock_expirable)(cfl)) {
908 owner = cfl->fl_lmops->lm_mod_owner;
909 func = cfl->fl_lmops->lm_expire_lock;
910 __module_get(owner);
911 spin_unlock(&ctx->flc_lock);
912 (*func)();
913 module_put(owner);
914 goto retry;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500915 }
Dai Ngo2443da22022-05-02 14:19:25 -0700916 locks_copy_conflock(fl, cfl);
917 goto out;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500918 }
919 fl->fl_type = F_UNLCK;
920out:
Jeff Layton6109c852015-01-16 15:05:57 -0500921 spin_unlock(&ctx->flc_lock);
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400922 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924EXPORT_SYMBOL(posix_test_lock);
925
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400926/*
927 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400929 * We attempt to detect deadlocks that are due purely to posix file
930 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400932 * We assume that a task can be waiting for at most one lock at a time.
933 * So for any acquired lock, the process holding that lock may be
934 * waiting on at most one other lock. That lock in turns may be held by
935 * someone waiting for at most one other lock. Given a requested lock
936 * caller_fl which is about to wait for a conflicting lock block_fl, we
937 * follow this chain of waiters to ensure we are not about to create a
938 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400939 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400940 * Since we do this before we ever put a process to sleep on a lock, we
941 * are ensured that there is never a cycle; that is what guarantees that
942 * the while() loop in posix_locks_deadlock() eventually completes.
943 *
944 * Note: the above assumption may not be true when handling lock
945 * requests from a broken NFS client. It may also fail in the presence
946 * of tasks (such as posix threads) sharing the same open file table.
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400947 * To handle those cases, we just bail out after a few iterations.
Jeff Layton57b65322014-02-03 12:13:09 -0500948 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400949 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
Jeff Layton57b65322014-02-03 12:13:09 -0500950 * Because the owner is not even nominally tied to a thread of
951 * execution, the deadlock detection below can't reasonably work well. Just
952 * skip it for those.
953 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400954 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
Jeff Layton57b65322014-02-03 12:13:09 -0500955 * locks that just checks for the case where two tasks are attempting to
956 * upgrade from read to write locks on the same inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400958
959#define MAX_DEADLK_ITERATIONS 10
960
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400961/* Find a lock that the owner of the given block_fl is blocking on. */
962static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
963{
964 struct file_lock *fl;
965
Jeff Layton3999e492013-06-21 08:58:19 -0400966 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
NeilBrown5946c432018-11-30 10:04:08 +1100967 if (posix_same_owner(fl, block_fl)) {
968 while (fl->fl_blocker)
969 fl = fl->fl_blocker;
970 return fl;
971 }
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400972 }
973 return NULL;
974}
975
Jeff Layton7b2296a2013-06-21 08:58:20 -0400976/* Must be called with the blocked_lock_lock held! */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700977static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 struct file_lock *block_fl)
979{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400980 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
Daniel Wagner663d5af2015-04-03 09:04:03 -0400982 lockdep_assert_held(&blocked_lock_lock);
983
Jeff Layton57b65322014-02-03 12:13:09 -0500984 /*
985 * This deadlock detector can't reasonably detect deadlocks with
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400986 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
Jeff Layton57b65322014-02-03 12:13:09 -0500987 */
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400988 if (IS_OFDLCK(caller_fl))
Jeff Layton57b65322014-02-03 12:13:09 -0500989 return 0;
990
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400991 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
992 if (i++ > MAX_DEADLK_ITERATIONS)
993 return 0;
994 if (posix_same_owner(caller_fl, block_fl))
995 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 }
997 return 0;
998}
999
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -04001001 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -04001002 *
1003 * Note that if called with an FL_EXISTS argument, the caller may determine
1004 * whether or not a lock was successfully freed by testing the return
1005 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 */
Jeff Laytonbcd7f78d2015-07-11 06:43:02 -04001007static int flock_lock_inode(struct inode *inode, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008{
Trond Myklebust993dfa82006-03-31 02:30:55 -08001009 struct file_lock *new_fl = NULL;
Jeff Layton5263e312015-01-16 15:05:55 -05001010 struct file_lock *fl;
1011 struct file_lock_context *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 int error = 0;
Jeff Layton5263e312015-01-16 15:05:55 -05001013 bool found = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -04001014 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
Jeff Layton5c1c6692015-04-03 09:04:03 -04001016 ctx = locks_get_lock_context(inode, request->fl_type);
1017 if (!ctx) {
1018 if (request->fl_type != F_UNLCK)
1019 return -ENOMEM;
1020 return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
1021 }
Jeff Layton5263e312015-01-16 15:05:55 -05001022
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001023 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
1024 new_fl = locks_alloc_lock();
1025 if (!new_fl)
1026 return -ENOMEM;
1027 }
1028
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001029 percpu_down_read(&file_rwsem);
Jeff Layton6109c852015-01-16 15:05:57 -05001030 spin_lock(&ctx->flc_lock);
Trond Myklebustf07f18d2006-06-29 16:38:37 -04001031 if (request->fl_flags & FL_ACCESS)
1032 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +04001033
Jeff Layton5263e312015-01-16 15:05:55 -05001034 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
Jeff Laytonbcd7f78d2015-07-11 06:43:02 -04001035 if (request->fl_file != fl->fl_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -08001037 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 goto out;
Jeff Layton5263e312015-01-16 15:05:55 -05001039 found = true;
Jeff Laytone084c1b2015-02-16 14:32:03 -05001040 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 break;
1042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Trond Myklebustf475ae92006-06-29 16:38:32 -04001044 if (request->fl_type == F_UNLCK) {
1045 if ((request->fl_flags & FL_EXISTS) && !found)
1046 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -08001047 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -04001048 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
Trond Myklebustf07f18d2006-06-29 16:38:37 -04001050find_conflict:
Jeff Layton5263e312015-01-16 15:05:55 -05001051 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
Trond Myklebust993dfa82006-03-31 02:30:55 -08001052 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 continue;
1054 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -07001055 if (!(request->fl_flags & FL_SLEEP))
1056 goto out;
1057 error = FILE_LOCK_DEFERRED;
NeilBrownfd7732e2018-11-30 10:04:08 +11001058 locks_insert_block(fl, request, flock_locks_conflict);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 goto out;
1060 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -04001061 if (request->fl_flags & FL_ACCESS)
1062 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -08001063 locks_copy_lock(new_fl, request);
NeilBrown5946c432018-11-30 10:04:08 +11001064 locks_move_blocks(new_fl, request);
Jeff Laytone084c1b2015-02-16 14:32:03 -05001065 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
Trond Myklebust993dfa82006-03-31 02:30:55 -08001066 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +04001067 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
1069out:
Jeff Layton6109c852015-01-16 15:05:57 -05001070 spin_unlock(&ctx->flc_lock);
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001071 percpu_up_read(&file_rwsem);
Trond Myklebust993dfa82006-03-31 02:30:55 -08001072 if (new_fl)
1073 locks_free_lock(new_fl);
Jeff Laytoned9814d2014-08-11 14:20:31 -04001074 locks_dispose_list(&dispose);
Jeff Laytonc883da32018-07-30 07:54:56 -04001075 trace_flock_lock_inode(inode, request, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 return error;
1077}
1078
Jeff Laytonb4d629a2016-01-07 18:27:42 -05001079static int posix_lock_inode(struct inode *inode, struct file_lock *request,
1080 struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081{
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001082 struct file_lock *fl, *tmp;
Miklos Szeredi39005d02006-06-23 02:05:10 -07001083 struct file_lock *new_fl = NULL;
1084 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 struct file_lock *left = NULL;
1086 struct file_lock *right = NULL;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001087 struct file_lock_context *ctx;
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001088 int error;
1089 bool added = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -04001090 LIST_HEAD(dispose);
Dai Ngo2443da22022-05-02 14:19:25 -07001091 void *owner;
1092 void (*func)(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Jeff Layton5c1c6692015-04-03 09:04:03 -04001094 ctx = locks_get_lock_context(inode, request->fl_type);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001095 if (!ctx)
Jeff Layton5c1c6692015-04-03 09:04:03 -04001096 return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001097
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 /*
1099 * We may need two file_lock structures for this operation,
1100 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -07001101 *
1102 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 */
Miklos Szeredi39005d02006-06-23 02:05:10 -07001104 if (!(request->fl_flags & FL_ACCESS) &&
1105 (request->fl_type != F_UNLCK ||
1106 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1107 new_fl = locks_alloc_lock();
1108 new_fl2 = locks_alloc_lock();
1109 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110
Dai Ngo2443da22022-05-02 14:19:25 -07001111retry:
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001112 percpu_down_read(&file_rwsem);
Jeff Layton6109c852015-01-16 15:05:57 -05001113 spin_lock(&ctx->flc_lock);
Jeff Layton1cb36012013-06-21 08:58:12 -04001114 /*
1115 * New lock request. Walk all POSIX locks and look for conflicts. If
1116 * there are any, either return error or put the request on the
Jeff Layton48f74182013-06-21 08:58:18 -04001117 * blocker's list of waiters and the global blocked_hash.
Jeff Layton1cb36012013-06-21 08:58:12 -04001118 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 if (request->fl_type != F_UNLCK) {
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001120 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 if (!posix_locks_conflict(request, fl))
1122 continue;
Dai Ngo2443da22022-05-02 14:19:25 -07001123 if (fl->fl_lmops && fl->fl_lmops->lm_lock_expirable
1124 && (*fl->fl_lmops->lm_lock_expirable)(fl)) {
1125 owner = fl->fl_lmops->lm_mod_owner;
1126 func = fl->fl_lmops->lm_expire_lock;
1127 __module_get(owner);
1128 spin_unlock(&ctx->flc_lock);
1129 percpu_up_read(&file_rwsem);
1130 (*func)();
1131 module_put(owner);
1132 goto retry;
1133 }
Andy Adamson5842add2006-03-26 01:37:26 -08001134 if (conflock)
Kinglong Mee3fe0fff12014-08-22 10:18:42 -04001135 locks_copy_conflock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 error = -EAGAIN;
1137 if (!(request->fl_flags & FL_SLEEP))
1138 goto out;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001139 /*
1140 * Deadlock detection and insertion into the blocked
1141 * locks list must be done while holding the same lock!
1142 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 error = -EDEADLK;
Jeff Layton7b2296a2013-06-21 08:58:20 -04001144 spin_lock(&blocked_lock_lock);
Jeff Layton945ab8f2019-03-25 08:15:14 -04001145 /*
1146 * Ensure that we don't find any locks blocked on this
1147 * request during deadlock detection.
1148 */
1149 __locks_wake_up_blocks(request);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001150 if (likely(!posix_locks_deadlock(request, fl))) {
1151 error = FILE_LOCK_DEFERRED;
NeilBrownfd7732e2018-11-30 10:04:08 +11001152 __locks_insert_block(fl, request,
1153 posix_locks_conflict);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001154 }
Jeff Layton7b2296a2013-06-21 08:58:20 -04001155 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 goto out;
NeilBrown7bbd1fc2018-11-30 10:04:08 +11001157 }
1158 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
1160 /* If we're just looking for a conflict, we're done. */
1161 error = 0;
1162 if (request->fl_flags & FL_ACCESS)
1163 goto out;
1164
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001165 /* Find the first old lock with the same owner as the new lock */
1166 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1167 if (posix_same_owner(request, fl))
1168 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 }
1170
Jeff Layton1cb36012013-06-21 08:58:12 -04001171 /* Process locks with this owner. */
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001172 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1173 if (!posix_same_owner(request, fl))
1174 break;
1175
1176 /* Detect adjacent or overlapping regions (if same lock type) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -07001178 /* In all comparisons of start vs end, use
1179 * "start - 1" rather than "end + 1". If end
1180 * is OFFSET_MAX, end + 1 will become negative.
1181 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 if (fl->fl_end < request->fl_start - 1)
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001183 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 /* If the next lock in the list has entirely bigger
1185 * addresses than the new one, insert the lock here.
1186 */
Olaf Kirch449231d2005-08-25 16:25:35 -07001187 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 break;
1189
1190 /* If we come here, the new and old lock are of the
1191 * same type and adjacent or overlapping. Make one
1192 * lock yielding from the lower start address of both
1193 * locks to the higher end address.
1194 */
1195 if (fl->fl_start > request->fl_start)
1196 fl->fl_start = request->fl_start;
1197 else
1198 request->fl_start = fl->fl_start;
1199 if (fl->fl_end < request->fl_end)
1200 fl->fl_end = request->fl_end;
1201 else
1202 request->fl_end = fl->fl_end;
1203 if (added) {
Jeff Laytone084c1b2015-02-16 14:32:03 -05001204 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 continue;
1206 }
1207 request = fl;
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001208 added = true;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001209 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 /* Processing for different lock types is a bit
1211 * more complex.
1212 */
1213 if (fl->fl_end < request->fl_start)
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001214 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 if (fl->fl_start > request->fl_end)
1216 break;
1217 if (request->fl_type == F_UNLCK)
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001218 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 if (fl->fl_start < request->fl_start)
1220 left = fl;
1221 /* If the next lock in the list has a higher end
1222 * address than the new one, insert the new one here.
1223 */
1224 if (fl->fl_end > request->fl_end) {
1225 right = fl;
1226 break;
1227 }
1228 if (fl->fl_start >= request->fl_start) {
1229 /* The new lock completely replaces an old
1230 * one (This may happen several times).
1231 */
1232 if (added) {
Jeff Laytone084c1b2015-02-16 14:32:03 -05001233 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 continue;
1235 }
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001236 /*
1237 * Replace the old lock with new_fl, and
1238 * remove the old one. It's safe to do the
1239 * insert here since we know that we won't be
1240 * using new_fl later, and that the lock is
1241 * just replacing an existing lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 */
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001243 error = -ENOLCK;
1244 if (!new_fl)
1245 goto out;
1246 locks_copy_lock(new_fl, request);
yangerkun5ef15962020-06-01 17:16:16 +08001247 locks_move_blocks(new_fl, request);
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001248 request = new_fl;
1249 new_fl = NULL;
Jeff Laytone084c1b2015-02-16 14:32:03 -05001250 locks_insert_lock_ctx(request, &fl->fl_list);
1251 locks_delete_lock_ctx(fl, &dispose);
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001252 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 }
1254 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
1256
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001257 /*
Jeff Layton1cb36012013-06-21 08:58:12 -04001258 * The above code only modifies existing locks in case of merging or
1259 * replacing. If new lock(s) need to be inserted all modifications are
1260 * done below this, so it's safe yet to bail out.
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001261 */
1262 error = -ENOLCK; /* "no luck" */
1263 if (right && left == right && !new_fl2)
1264 goto out;
1265
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 error = 0;
1267 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -04001268 if (request->fl_type == F_UNLCK) {
1269 if (request->fl_flags & FL_EXISTS)
1270 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -04001272 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001273
1274 if (!new_fl) {
1275 error = -ENOLCK;
1276 goto out;
1277 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 locks_copy_lock(new_fl, request);
NeilBrown5946c432018-11-30 10:04:08 +11001279 locks_move_blocks(new_fl, request);
Jeff Laytone084c1b2015-02-16 14:32:03 -05001280 locks_insert_lock_ctx(new_fl, &fl->fl_list);
Jeff Layton2e2f756f2015-02-17 17:08:23 -05001281 fl = new_fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 new_fl = NULL;
1283 }
1284 if (right) {
1285 if (left == right) {
1286 /* The new lock breaks the old one in two pieces,
1287 * so we have to use the second new lock.
1288 */
1289 left = new_fl2;
1290 new_fl2 = NULL;
1291 locks_copy_lock(left, right);
Jeff Laytone084c1b2015-02-16 14:32:03 -05001292 locks_insert_lock_ctx(left, &fl->fl_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 }
1294 right->fl_start = request->fl_end + 1;
1295 locks_wake_up_blocks(right);
1296 }
1297 if (left) {
1298 left->fl_end = request->fl_start - 1;
1299 locks_wake_up_blocks(left);
1300 }
1301 out:
Jeff Layton6109c852015-01-16 15:05:57 -05001302 spin_unlock(&ctx->flc_lock);
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001303 percpu_up_read(&file_rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 /*
1305 * Free any unused locks.
1306 */
1307 if (new_fl)
1308 locks_free_lock(new_fl);
1309 if (new_fl2)
1310 locks_free_lock(new_fl2);
Jeff Laytoned9814d2014-08-11 14:20:31 -04001311 locks_dispose_list(&dispose);
Jeff Layton18909102016-01-06 21:26:10 -05001312 trace_posix_lock_inode(inode, request, error);
1313
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 return error;
1315}
1316
1317/**
1318 * posix_lock_file - Apply a POSIX-style lock to a file
1319 * @filp: The file to apply the lock to
1320 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001321 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 *
1323 * Add a POSIX style lock to a file.
1324 * We merge adjacent & overlapping locks whenever possible.
1325 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001326 *
1327 * Note that if called with an FL_EXISTS argument, the caller may determine
1328 * whether or not a lock was successfully freed by testing the return
1329 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 */
Marc Eshel150b3932007-01-18 16:15:35 -05001331int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001332 struct file_lock *conflock)
1333{
Jeff Laytonc65454a2022-11-25 08:48:37 -05001334 return posix_lock_inode(file_inode(filp), fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001335}
Marc Eshel150b3932007-01-18 16:15:35 -05001336EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
1338/**
Jeff Layton29d01b22015-07-11 06:43:02 -04001339 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1340 * @inode: inode of file to which lock request should be applied
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 * @fl: The lock to be applied
1342 *
Benjamin Coddington616fb382015-10-22 13:38:15 -04001343 * Apply a POSIX style lock request to an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 */
Benjamin Coddington616fb382015-10-22 13:38:15 -04001345static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346{
1347 int error;
1348 might_sleep ();
1349 for (;;) {
Jeff Laytonb4d629a2016-01-07 18:27:42 -05001350 error = posix_lock_inode(inode, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001351 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 break;
Linus Torvaldsdcf23ac2020-03-18 07:52:21 -04001353 error = wait_event_interruptible(fl->fl_wait,
1354 list_empty(&fl->fl_blocked_member));
NeilBrown16306a62018-11-30 10:04:08 +11001355 if (error)
1356 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 }
NeilBrown16306a62018-11-30 10:04:08 +11001358 locks_delete_block(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 return error;
1360}
Jeff Layton29d01b22015-07-11 06:43:02 -04001361
J. Bruce Fields778fc542011-07-26 18:25:49 -04001362static void lease_clear_pending(struct file_lock *fl, int arg)
1363{
1364 switch (arg) {
1365 case F_UNLCK:
1366 fl->fl_flags &= ~FL_UNLOCK_PENDING;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001367 fallthrough;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001368 case F_RDLCK:
1369 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1370 }
1371}
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373/* We already had a lease on this file; just change its type */
Jeff Layton7448cc32015-01-16 15:05:57 -05001374int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 int error = assign_type(fl, arg);
1377
1378 if (error)
1379 return error;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001380 lease_clear_pending(fl, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 locks_wake_up_blocks(fl);
Filipe Brandenburger3b6e27232012-07-27 00:42:52 -04001382 if (arg == F_UNLCK) {
1383 struct file *filp = fl->fl_file;
1384
1385 f_delown(filp);
1386 filp->f_owner.signum = 0;
J. Bruce Fields96d6d592012-07-27 16:18:00 -04001387 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1388 if (fl->fl_fasync != NULL) {
1389 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1390 fl->fl_fasync = NULL;
1391 }
Jeff Laytone084c1b2015-02-16 14:32:03 -05001392 locks_delete_lock_ctx(fl, dispose);
Filipe Brandenburger3b6e27232012-07-27 00:42:52 -04001393 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 return 0;
1395}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396EXPORT_SYMBOL(lease_modify);
1397
J. Bruce Fields778fc542011-07-26 18:25:49 -04001398static bool past_time(unsigned long then)
1399{
1400 if (!then)
1401 /* 0 is a special value meaning "this never expires": */
1402 return false;
1403 return time_after(jiffies, then);
1404}
1405
Jeff Laytonc45198e2014-09-01 07:12:07 -04001406static void time_out_leases(struct inode *inode, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407{
Jeff Layton8634b512015-01-16 15:05:55 -05001408 struct file_lock_context *ctx = inode->i_flctx;
1409 struct file_lock *fl, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
Jeff Layton6109c852015-01-16 15:05:57 -05001411 lockdep_assert_held(&ctx->flc_lock);
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001412
Jeff Layton8634b512015-01-16 15:05:55 -05001413 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001414 trace_time_out_leases(inode, fl);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001415 if (past_time(fl->fl_downgrade_time))
Jeff Layton7448cc32015-01-16 15:05:57 -05001416 lease_modify(fl, F_RDLCK, dispose);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001417 if (past_time(fl->fl_break_time))
Jeff Layton7448cc32015-01-16 15:05:57 -05001418 lease_modify(fl, F_UNLCK, dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 }
1420}
1421
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001422static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1423{
Ira Weinyd51f5272019-06-05 18:45:34 -07001424 bool rc;
1425
J. Bruce Fields28df3d12017-07-28 16:35:15 -04001426 if (lease->fl_lmops->lm_breaker_owns_lease
1427 && lease->fl_lmops->lm_breaker_owns_lease(lease))
1428 return false;
Ira Weinyd51f5272019-06-05 18:45:34 -07001429 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) {
1430 rc = false;
1431 goto trace;
1432 }
1433 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) {
1434 rc = false;
1435 goto trace;
1436 }
1437
1438 rc = locks_conflict(breaker, lease);
1439trace:
1440 trace_leases_conflict(rc, lease, breaker);
1441 return rc;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001442}
1443
Jeff Layton03d12dd2014-09-01 14:53:41 -04001444static bool
1445any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1446{
Jeff Layton8634b512015-01-16 15:05:55 -05001447 struct file_lock_context *ctx = inode->i_flctx;
Jeff Layton03d12dd2014-09-01 14:53:41 -04001448 struct file_lock *fl;
1449
Jeff Layton6109c852015-01-16 15:05:57 -05001450 lockdep_assert_held(&ctx->flc_lock);
Jeff Layton03d12dd2014-09-01 14:53:41 -04001451
Jeff Layton8634b512015-01-16 15:05:55 -05001452 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Jeff Layton03d12dd2014-09-01 14:53:41 -04001453 if (leases_conflict(fl, breaker))
1454 return true;
1455 }
1456 return false;
1457}
1458
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459/**
1460 * __break_lease - revoke all outstanding leases on file
1461 * @inode: the inode of the file to return
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001462 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1463 * break all leases
1464 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1465 * only delegations
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 *
david m. richter87250dd2007-05-09 16:10:27 -04001467 * break_lease (inlined for speed) has checked there already is at least
1468 * some kind of lock (maybe a lease) on this file. Leases are broken on
1469 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 * specified %O_NONBLOCK to your open().
1471 */
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001472int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473{
J. Bruce Fields778fc542011-07-26 18:25:49 -04001474 int error = 0;
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001475 struct file_lock_context *ctx;
Yan, Zhenga9011252015-03-27 10:34:20 +08001476 struct file_lock *new_fl, *fl, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 unsigned long break_time;
Al Viro8737c932009-12-24 06:47:55 -05001478 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001479 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
Al Viro8737c932009-12-24 06:47:55 -05001481 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001482 if (IS_ERR(new_fl))
1483 return PTR_ERR(new_fl);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001484 new_fl->fl_flags = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
Jeff Layton8634b512015-01-16 15:05:55 -05001486 /* typically we will check that ctx is non-NULL before calling */
Jeff Layton401a8b82022-11-16 09:02:30 -05001487 ctx = locks_inode_context(inode);
Jeff Layton8634b512015-01-16 15:05:55 -05001488 if (!ctx) {
1489 WARN_ON_ONCE(1);
Wenwen Wangcfddf9f2019-08-19 18:47:34 -05001490 goto free_lock;
Jeff Layton8634b512015-01-16 15:05:55 -05001491 }
1492
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001493 percpu_down_read(&file_rwsem);
Jeff Layton6109c852015-01-16 15:05:57 -05001494 spin_lock(&ctx->flc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
Jeff Laytonc45198e2014-09-01 07:12:07 -04001496 time_out_leases(inode, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
Jeff Layton03d12dd2014-09-01 14:53:41 -04001498 if (!any_leases_conflict(inode, new_fl))
J. Bruce Fields778fc542011-07-26 18:25:49 -04001499 goto out;
1500
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 break_time = 0;
1502 if (lease_break_time > 0) {
1503 break_time = jiffies + lease_break_time * HZ;
1504 if (break_time == 0)
1505 break_time++; /* so that 0 means no break time */
1506 }
1507
Yan, Zhenga9011252015-03-27 10:34:20 +08001508 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001509 if (!leases_conflict(fl, new_fl))
1510 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001511 if (want_write) {
1512 if (fl->fl_flags & FL_UNLOCK_PENDING)
1513 continue;
1514 fl->fl_flags |= FL_UNLOCK_PENDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 fl->fl_break_time = break_time;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001516 } else {
Jeff Layton8634b512015-01-16 15:05:55 -05001517 if (lease_breaking(fl))
J. Bruce Fields778fc542011-07-26 18:25:49 -04001518 continue;
1519 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1520 fl->fl_downgrade_time = break_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 }
Jeff Layton4d01b7f2014-09-01 15:06:54 -04001522 if (fl->fl_lmops->lm_break(fl))
Jeff Laytone084c1b2015-02-16 14:32:03 -05001523 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 }
1525
Jeff Layton8634b512015-01-16 15:05:55 -05001526 if (list_empty(&ctx->flc_lease))
Jeff Layton4d01b7f2014-09-01 15:06:54 -04001527 goto out;
1528
Jeff Layton843c6b22014-09-01 14:27:43 -04001529 if (mode & O_NONBLOCK) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001530 trace_break_lease_noblock(inode, new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 error = -EWOULDBLOCK;
1532 goto out;
1533 }
1534
1535restart:
Jeff Layton8634b512015-01-16 15:05:55 -05001536 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1537 break_time = fl->fl_break_time;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001538 if (break_time != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 break_time -= jiffies;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001540 if (break_time == 0)
1541 break_time++;
NeilBrownfd7732e2018-11-30 10:04:08 +11001542 locks_insert_block(fl, new_fl, leases_conflict);
Jeff Layton62af4f12014-05-09 14:13:05 -04001543 trace_break_lease_block(inode, new_fl);
Jeff Layton6109c852015-01-16 15:05:57 -05001544 spin_unlock(&ctx->flc_lock);
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001545 percpu_up_read(&file_rwsem);
Peter Zijlstraaba37662015-06-22 14:16:33 +02001546
Jeff Laytonc45198e2014-09-01 07:12:07 -04001547 locks_dispose_list(&dispose);
Matthew Wilcox4321e012008-01-14 21:28:30 -07001548 error = wait_event_interruptible_timeout(new_fl->fl_wait,
Linus Torvaldsdcf23ac2020-03-18 07:52:21 -04001549 list_empty(&new_fl->fl_blocked_member),
1550 break_time);
Peter Zijlstraaba37662015-06-22 14:16:33 +02001551
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001552 percpu_down_read(&file_rwsem);
Jeff Layton6109c852015-01-16 15:05:57 -05001553 spin_lock(&ctx->flc_lock);
Jeff Layton62af4f12014-05-09 14:13:05 -04001554 trace_break_lease_unblock(inode, new_fl);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001555 locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 if (error >= 0) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001557 /*
1558 * Wait for the next conflicting lease that has not been
1559 * broken yet
1560 */
Jeff Layton03d12dd2014-09-01 14:53:41 -04001561 if (error == 0)
1562 time_out_leases(inode, &dispose);
1563 if (any_leases_conflict(inode, new_fl))
1564 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 error = 0;
1566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567out:
Jeff Layton6109c852015-01-16 15:05:57 -05001568 spin_unlock(&ctx->flc_lock);
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001569 percpu_up_read(&file_rwsem);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001570 locks_dispose_list(&dispose);
Wenwen Wangcfddf9f2019-08-19 18:47:34 -05001571free_lock:
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001572 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 return error;
1574}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575EXPORT_SYMBOL(__break_lease);
1576
1577/**
Amir Goldstein76c47942018-01-03 17:14:34 +02001578 * lease_get_mtime - update modified time of an inode with exclusive lease
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 * @inode: the inode
Amir Goldstein76c47942018-01-03 17:14:34 +02001580 * @time: pointer to a timespec which contains the last modified time
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 *
1582 * This is to force NFS clients to flush their caches for files with
1583 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001584 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 */
Deepa Dinamani95582b02018-05-08 19:36:02 -07001586void lease_get_mtime(struct inode *inode, struct timespec64 *time)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587{
Jeff Laytonbfe86022014-08-22 10:18:44 -04001588 bool has_lease = false;
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001589 struct file_lock_context *ctx;
Jeff Layton8634b512015-01-16 15:05:55 -05001590 struct file_lock *fl;
Jeff Laytonbfe86022014-08-22 10:18:44 -04001591
Jeff Layton401a8b82022-11-16 09:02:30 -05001592 ctx = locks_inode_context(inode);
Jeff Layton8634b512015-01-16 15:05:55 -05001593 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
Jeff Layton6109c852015-01-16 15:05:57 -05001594 spin_lock(&ctx->flc_lock);
Geliang Tang8ace5df2015-11-18 21:40:33 +08001595 fl = list_first_entry_or_null(&ctx->flc_lease,
1596 struct file_lock, fl_list);
1597 if (fl && (fl->fl_type == F_WRLCK))
1598 has_lease = true;
Jeff Layton6109c852015-01-16 15:05:57 -05001599 spin_unlock(&ctx->flc_lock);
Jeff Laytonbfe86022014-08-22 10:18:44 -04001600 }
1601
1602 if (has_lease)
Deepa Dinamanic2050a42016-09-14 07:48:06 -07001603 *time = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605EXPORT_SYMBOL(lease_get_mtime);
1606
1607/**
1608 * fcntl_getlease - Enquire what lease is currently active
1609 * @filp: the file
1610 *
1611 * The value returned by this function will be one of
1612 * (if no lease break is pending):
1613 *
1614 * %F_RDLCK to indicate a shared lease is held.
1615 *
1616 * %F_WRLCK to indicate an exclusive lease is held.
1617 *
1618 * %F_UNLCK to indicate no lease is held.
1619 *
1620 * (if a lease break is pending):
1621 *
1622 * %F_RDLCK to indicate an exclusive lease needs to be
1623 * changed to a shared lease (or removed).
1624 *
1625 * %F_UNLCK to indicate the lease needs to be removed.
1626 *
1627 * XXX: sfr & willy disagree over whether F_INPROGRESS
1628 * should be returned to userspace.
1629 */
1630int fcntl_getlease(struct file *filp)
1631{
1632 struct file_lock *fl;
Jeff Laytonc65454a2022-11-25 08:48:37 -05001633 struct inode *inode = file_inode(filp);
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001634 struct file_lock_context *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 int type = F_UNLCK;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001636 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637
Jeff Layton401a8b82022-11-16 09:02:30 -05001638 ctx = locks_inode_context(inode);
Jeff Layton8634b512015-01-16 15:05:55 -05001639 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001640 percpu_down_read(&file_rwsem);
Jeff Layton6109c852015-01-16 15:05:57 -05001641 spin_lock(&ctx->flc_lock);
Miklos Szeredic568d682016-09-16 12:44:20 +02001642 time_out_leases(inode, &dispose);
Jeff Layton8634b512015-01-16 15:05:55 -05001643 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1644 if (fl->fl_file != filp)
1645 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001646 type = target_leasetype(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 break;
1648 }
Jeff Layton6109c852015-01-16 15:05:57 -05001649 spin_unlock(&ctx->flc_lock);
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001650 percpu_up_read(&file_rwsem);
Peter Zijlstra5f430862016-10-08 10:12:28 +02001651
Jeff Layton8634b512015-01-16 15:05:55 -05001652 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 return type;
1655}
1656
Jeff Layton24cbe782014-02-03 12:13:06 -05001657/**
Amir Goldstein387e3742019-06-07 17:24:38 +03001658 * check_conflicting_open - see if the given file points to an inode that has
NeilBrown7bbd1fc2018-11-30 10:04:08 +11001659 * an existing open that would conflict with the
1660 * desired lease.
Amir Goldstein387e3742019-06-07 17:24:38 +03001661 * @filp: file to check
Jeff Layton24cbe782014-02-03 12:13:06 -05001662 * @arg: type of lease that we're trying to acquire
Randy Dunlap7fadc592015-08-09 18:43:17 -07001663 * @flags: current lock flags
Jeff Layton24cbe782014-02-03 12:13:06 -05001664 *
1665 * Check to see if there's an existing open fd on this file that would
1666 * conflict with the lease we're trying to set.
1667 */
1668static int
Amir Goldstein387e3742019-06-07 17:24:38 +03001669check_conflicting_open(struct file *filp, const long arg, int flags)
Jeff Layton24cbe782014-02-03 12:13:06 -05001670{
Jeff Laytonc65454a2022-11-25 08:48:37 -05001671 struct inode *inode = file_inode(filp);
Amir Goldstein387e3742019-06-07 17:24:38 +03001672 int self_wcount = 0, self_rcount = 0;
Jeff Layton24cbe782014-02-03 12:13:06 -05001673
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001674 if (flags & FL_LAYOUT)
1675 return 0;
J. Bruce Fieldsaba20722021-04-16 14:00:18 -04001676 if (flags & FL_DELEG)
1677 /* We leave these checks to the caller */
1678 return 0;
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001679
Amir Goldstein387e3742019-06-07 17:24:38 +03001680 if (arg == F_RDLCK)
1681 return inode_is_open_for_write(inode) ? -EAGAIN : 0;
1682 else if (arg != F_WRLCK)
1683 return 0;
1684
1685 /*
1686 * Make sure that only read/write count is from lease requestor.
1687 * Note that this will result in denying write leases when i_writecount
1688 * is negative, which is what we want. (We shouldn't grant write leases
1689 * on files open for execution.)
1690 */
1691 if (filp->f_mode & FMODE_WRITE)
1692 self_wcount = 1;
1693 else if (filp->f_mode & FMODE_READ)
1694 self_rcount = 1;
1695
1696 if (atomic_read(&inode->i_writecount) != self_wcount ||
1697 atomic_read(&inode->i_readcount) != self_rcount)
Jeff Layton24cbe782014-02-03 12:13:06 -05001698 return -EAGAIN;
1699
Amir Goldstein387e3742019-06-07 17:24:38 +03001700 return 0;
Jeff Layton24cbe782014-02-03 12:13:06 -05001701}
1702
Jeff Laytone6f5c782014-08-22 10:40:25 -04001703static int
1704generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705{
Jeff Layton8634b512015-01-16 15:05:55 -05001706 struct file_lock *fl, *my_fl = NULL, *lease;
Jeff Laytonc65454a2022-11-25 08:48:37 -05001707 struct inode *inode = file_inode(filp);
Jeff Layton8634b512015-01-16 15:05:55 -05001708 struct file_lock_context *ctx;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001709 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001710 int error;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001711 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712
J. Bruce Fields096657b2010-10-30 17:31:14 -04001713 lease = *flp;
Jeff Layton62af4f12014-05-09 14:13:05 -04001714 trace_generic_add_lease(inode, lease);
1715
Jeff Layton5c1c6692015-04-03 09:04:03 -04001716 /* Note that arg is never F_UNLCK here */
1717 ctx = locks_get_lock_context(inode, arg);
Jeff Layton8634b512015-01-16 15:05:55 -05001718 if (!ctx)
1719 return -ENOMEM;
1720
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001721 /*
1722 * In the delegation case we need mutual exclusion with
1723 * a number of operations that take the i_mutex. We trylock
1724 * because delegations are an optional optimization, and if
1725 * there's some chance of a conflict--we'd rather not
1726 * bother, maybe that's a sign this just isn't a good file to
1727 * hand out a delegation on.
1728 */
Al Viro59551022016-01-22 15:40:57 -05001729 if (is_deleg && !inode_trylock(inode))
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001730 return -EAGAIN;
1731
1732 if (is_deleg && arg == F_WRLCK) {
1733 /* Write delegations are not currently supported: */
Al Viro59551022016-01-22 15:40:57 -05001734 inode_unlock(inode);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001735 WARN_ON_ONCE(1);
1736 return -EINVAL;
1737 }
J. Bruce Fields096657b2010-10-30 17:31:14 -04001738
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001739 percpu_down_read(&file_rwsem);
Jeff Layton6109c852015-01-16 15:05:57 -05001740 spin_lock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001741 time_out_leases(inode, &dispose);
Amir Goldstein387e3742019-06-07 17:24:38 +03001742 error = check_conflicting_open(filp, arg, lease->fl_flags);
Jeff Layton24cbe782014-02-03 12:13:06 -05001743 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001744 goto out;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001745
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 /*
1747 * At this point, we know that if there is an exclusive
1748 * lease on this file, then we hold it on this filp
1749 * (otherwise our open of this file would have blocked).
1750 * And if we are trying to acquire an exclusive lease,
1751 * then the file is not open by anyone (including us)
1752 * except for this filp.
1753 */
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001754 error = -EAGAIN;
Jeff Layton8634b512015-01-16 15:05:55 -05001755 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001756 if (fl->fl_file == filp &&
1757 fl->fl_owner == lease->fl_owner) {
Jeff Layton8634b512015-01-16 15:05:55 -05001758 my_fl = fl;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001759 continue;
1760 }
Jeff Layton8634b512015-01-16 15:05:55 -05001761
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001762 /*
1763 * No exclusive leases if someone else has a lease on
1764 * this file:
1765 */
1766 if (arg == F_WRLCK)
1767 goto out;
1768 /*
1769 * Modifying our existing lease is OK, but no getting a
1770 * new lease if someone else is opening for write:
1771 */
1772 if (fl->fl_flags & FL_UNLOCK_PENDING)
1773 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 }
1775
Jeff Layton8634b512015-01-16 15:05:55 -05001776 if (my_fl != NULL) {
Jeff Layton0164bf02015-03-04 17:34:32 -05001777 lease = my_fl;
1778 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001779 if (error)
1780 goto out;
1781 goto out_setup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 }
1783
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 error = -EINVAL;
1785 if (!leases_enable)
1786 goto out;
1787
Jeff Laytone084c1b2015-02-16 14:32:03 -05001788 locks_insert_lock_ctx(lease, &ctx->flc_lease);
Jeff Layton24cbe782014-02-03 12:13:06 -05001789 /*
1790 * The check in break_lease() is lockless. It's possible for another
1791 * open to race in after we did the earlier check for a conflicting
1792 * open but before the lease was inserted. Check again for a
1793 * conflicting open and cancel the lease if there is one.
1794 *
1795 * We also add a barrier here to ensure that the insertion of the lock
1796 * precedes these checks.
1797 */
1798 smp_mb();
Amir Goldstein387e3742019-06-07 17:24:38 +03001799 error = check_conflicting_open(filp, arg, lease->fl_flags);
Jeff Layton8634b512015-01-16 15:05:55 -05001800 if (error) {
Jeff Laytone084c1b2015-02-16 14:32:03 -05001801 locks_unlink_lock_ctx(lease);
Jeff Layton8634b512015-01-16 15:05:55 -05001802 goto out;
1803 }
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001804
1805out_setup:
1806 if (lease->fl_lmops->lm_setup)
1807 lease->fl_lmops->lm_setup(lease, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808out:
Jeff Layton6109c852015-01-16 15:05:57 -05001809 spin_unlock(&ctx->flc_lock);
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001810 percpu_up_read(&file_rwsem);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001811 locks_dispose_list(&dispose);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001812 if (is_deleg)
Al Viro59551022016-01-22 15:40:57 -05001813 inode_unlock(inode);
Jeff Layton8634b512015-01-16 15:05:55 -05001814 if (!error && !my_fl)
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001815 *flp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 return error;
1817}
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001818
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001819static int generic_delete_lease(struct file *filp, void *owner)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001820{
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001821 int error = -EAGAIN;
Jeff Layton8634b512015-01-16 15:05:55 -05001822 struct file_lock *fl, *victim = NULL;
Jeff Laytonc65454a2022-11-25 08:48:37 -05001823 struct inode *inode = file_inode(filp);
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001824 struct file_lock_context *ctx;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001825 LIST_HEAD(dispose);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001826
Jeff Layton401a8b82022-11-16 09:02:30 -05001827 ctx = locks_inode_context(inode);
Jeff Layton8634b512015-01-16 15:05:55 -05001828 if (!ctx) {
1829 trace_generic_delete_lease(inode, NULL);
1830 return error;
1831 }
1832
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001833 percpu_down_read(&file_rwsem);
Jeff Layton6109c852015-01-16 15:05:57 -05001834 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001835 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001836 if (fl->fl_file == filp &&
1837 fl->fl_owner == owner) {
Jeff Layton8634b512015-01-16 15:05:55 -05001838 victim = fl;
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001839 break;
Jeff Layton8634b512015-01-16 15:05:55 -05001840 }
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001841 }
Jeff Laytona9b1b452015-03-14 09:45:35 -04001842 trace_generic_delete_lease(inode, victim);
Jeff Layton8634b512015-01-16 15:05:55 -05001843 if (victim)
Jeff Layton7448cc32015-01-16 15:05:57 -05001844 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
Jeff Layton6109c852015-01-16 15:05:57 -05001845 spin_unlock(&ctx->flc_lock);
Peter Zijlstra02e525b22019-02-21 15:38:40 +01001846 percpu_up_read(&file_rwsem);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001847 locks_dispose_list(&dispose);
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001848 return error;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001849}
1850
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851/**
1852 * generic_setlease - sets a lease on an open file
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001853 * @filp: file pointer
1854 * @arg: type of lease to obtain
1855 * @flp: input - file_lock to use, output - file_lock inserted
1856 * @priv: private data for lm_setup (may be NULL if lm_setup
1857 * doesn't require it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 *
1859 * The (input) flp->fl_lmops->lm_break function is required
1860 * by break_lease().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 */
Jeff Laytone6f5c782014-08-22 10:40:25 -04001862int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1863 void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864{
Jeff Laytonc65454a2022-11-25 08:48:37 -05001865 struct inode *inode = file_inode(filp);
Seth Forshee42d0c4b2023-03-09 14:39:09 -06001866 vfsuid_t vfsuid = i_uid_into_vfsuid(file_mnt_idmap(filp), inode);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001867 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868
Seth Forshee42d0c4b2023-03-09 14:39:09 -06001869 if ((!vfsuid_eq_kuid(vfsuid, current_fsuid())) && !capable(CAP_LEASE))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001870 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 if (!S_ISREG(inode->i_mode))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001872 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 error = security_file_lock(filp, arg);
1874 if (error)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001875 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001877 switch (arg) {
1878 case F_UNLCK:
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001879 return generic_delete_lease(filp, *priv);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001880 case F_RDLCK:
1881 case F_WRLCK:
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001882 if (!(*flp)->fl_lmops->lm_break) {
1883 WARN_ON_ONCE(1);
1884 return -ENOLCK;
1885 }
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001886
Jeff Laytone6f5c782014-08-22 10:40:25 -04001887 return generic_add_lease(filp, arg, flp, priv);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001888 default:
Dave Jones8d657eb2012-07-13 13:35:36 -04001889 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001892EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
Jeff Layton18f66222019-08-18 14:18:45 -04001894/*
1895 * Kernel subsystems can register to be notified on any attempt to set
1896 * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
1897 * to close files that it may have cached when there is an attempt to set a
1898 * conflicting lease.
1899 */
1900static struct srcu_notifier_head lease_notifier_chain;
1901
1902static inline void
1903lease_notifier_chain_init(void)
1904{
1905 srcu_init_notifier_head(&lease_notifier_chain);
1906}
1907
1908static inline void
1909setlease_notifier(long arg, struct file_lock *lease)
1910{
1911 if (arg != F_UNLCK)
1912 srcu_notifier_call_chain(&lease_notifier_chain, arg, lease);
1913}
1914
1915int lease_register_notifier(struct notifier_block *nb)
1916{
1917 return srcu_notifier_chain_register(&lease_notifier_chain, nb);
1918}
1919EXPORT_SYMBOL_GPL(lease_register_notifier);
1920
1921void lease_unregister_notifier(struct notifier_block *nb)
1922{
1923 srcu_notifier_chain_unregister(&lease_notifier_chain, nb);
1924}
1925EXPORT_SYMBOL_GPL(lease_unregister_notifier);
1926
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001927/**
Jeff Laytone51673a2014-08-22 18:13:28 -04001928 * vfs_setlease - sets a lease on an open file
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001929 * @filp: file pointer
1930 * @arg: type of lease to obtain
1931 * @lease: file_lock to use when adding a lease
1932 * @priv: private info for lm_setup when adding a lease (may be
NeilBrown7bbd1fc2018-11-30 10:04:08 +11001933 * NULL if lm_setup doesn't require it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 *
Jeff Laytone51673a2014-08-22 18:13:28 -04001935 * Call this to establish a lease on the file. The "lease" argument is not
1936 * used for F_UNLCK requests and may be NULL. For commands that set or alter
Mauro Carvalho Chehab80b79dd2017-05-27 06:07:18 -04001937 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
1938 * set; if not, this function will return -ENOLCK (and generate a scary-looking
Jeff Laytone51673a2014-08-22 18:13:28 -04001939 * stack trace).
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001940 *
1941 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1942 * may be NULL if the lm_setup operation doesn't require it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 */
Jeff Laytone6f5c782014-08-22 10:40:25 -04001944int
1945vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946{
Jeff Layton18f66222019-08-18 14:18:45 -04001947 if (lease)
1948 setlease_notifier(arg, *lease);
Miklos Szeredide2a4a52018-07-18 15:44:43 +02001949 if (filp->f_op->setlease)
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001950 return filp->f_op->setlease(filp, arg, lease, priv);
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001951 else
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001952 return generic_setlease(filp, arg, lease, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001954EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001956static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957{
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001958 struct file_lock *fl;
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001959 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 int error;
1961
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001962 fl = lease_alloc(filp, arg);
1963 if (IS_ERR(fl))
1964 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001966 new = fasync_alloc();
1967 if (!new) {
1968 locks_free_lock(fl);
1969 return -ENOMEM;
1970 }
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001971 new->fa_fd = fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001973 error = vfs_setlease(filp, arg, &fl, (void **)&new);
Jeff Layton2dfb9282014-08-11 18:14:12 -04001974 if (fl)
1975 locks_free_lock(fl);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001976 if (new)
1977 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 return error;
1979}
1980
1981/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001982 * fcntl_setlease - sets a lease on an open file
1983 * @fd: open file descriptor
1984 * @filp: file pointer
1985 * @arg: type of lease to obtain
1986 *
1987 * Call this fcntl to establish a lease on the file.
1988 * Note that you also need to call %F_SETSIG to
1989 * receive a signal when the lease is broken.
1990 */
1991int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1992{
1993 if (arg == F_UNLCK)
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001994 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001995 return do_fcntl_add_lease(fd, filp, arg);
1996}
1997
1998/**
Jeff Layton29d01b22015-07-11 06:43:02 -04001999 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2000 * @inode: inode of the file to apply to
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 * @fl: The lock to be applied
2002 *
Jeff Layton29d01b22015-07-11 06:43:02 -04002003 * Apply a FLOCK style lock request to an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 */
Benjamin Coddington616fb382015-10-22 13:38:15 -04002005static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006{
2007 int error;
2008 might_sleep();
2009 for (;;) {
Jeff Layton29d01b22015-07-11 06:43:02 -04002010 error = flock_lock_inode(inode, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07002011 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 break;
Linus Torvaldsdcf23ac2020-03-18 07:52:21 -04002013 error = wait_event_interruptible(fl->fl_wait,
2014 list_empty(&fl->fl_blocked_member));
NeilBrown16306a62018-11-30 10:04:08 +11002015 if (error)
2016 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 }
NeilBrown16306a62018-11-30 10:04:08 +11002018 locks_delete_block(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 return error;
2020}
2021
Jeff Layton29d01b22015-07-11 06:43:02 -04002022/**
Benjamin Coddingtone55c34a2015-10-22 13:38:13 -04002023 * locks_lock_inode_wait - Apply a lock to an inode
2024 * @inode: inode of the file to apply to
2025 * @fl: The lock to be applied
2026 *
2027 * Apply a POSIX or FLOCK style lock request to an inode.
2028 */
2029int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2030{
2031 int res = 0;
2032 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
2033 case FL_POSIX:
2034 res = posix_lock_inode_wait(inode, fl);
2035 break;
2036 case FL_FLOCK:
2037 res = flock_lock_inode_wait(inode, fl);
2038 break;
2039 default:
2040 BUG();
2041 }
2042 return res;
2043}
2044EXPORT_SYMBOL(locks_lock_inode_wait);
2045
2046/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 * sys_flock: - flock() system call.
2048 * @fd: the file descriptor to lock.
2049 * @cmd: the type of lock to apply.
2050 *
2051 * Apply a %FL_FLOCK style lock to an open file descriptor.
Mauro Carvalho Chehab80b79dd2017-05-27 06:07:18 -04002052 * The @cmd can be one of:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 *
Mauro Carvalho Chehab80b79dd2017-05-27 06:07:18 -04002054 * - %LOCK_SH -- a shared lock.
2055 * - %LOCK_EX -- an exclusive lock.
2056 * - %LOCK_UN -- remove an existing lock.
Jeff Layton90f7d7a2021-09-10 15:36:29 -04002057 * - %LOCK_MAND -- a 'mandatory' flock. (DEPRECATED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 *
Jeff Layton90f7d7a2021-09-10 15:36:29 -04002059 * %LOCK_MAND support has been removed from the kernel.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 */
Heiko Carstens002c8972009-01-14 14:14:18 +01002061SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062{
Kuniyuki Iwashimadb4abb42022-07-16 21:35:32 -07002063 int can_sleep, error, type;
Kuniyuki Iwashima4149be72022-07-16 21:35:31 -07002064 struct file_lock fl;
Kuniyuki Iwashimadb4abb42022-07-16 21:35:32 -07002065 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066
Jeff Layton90f7d7a2021-09-10 15:36:29 -04002067 /*
2068 * LOCK_MAND locks were broken for a long time in that they never
2069 * conflicted with one another and didn't prevent any sort of open,
2070 * read or write activity.
2071 *
2072 * Just ignore these requests now, to preserve legacy behavior, but
2073 * throw a warning to let people know that they don't actually work.
2074 */
2075 if (cmd & LOCK_MAND) {
Andi Kleenf2f2494c2022-11-18 15:43:57 -08002076 pr_warn_once("%s(%d): Attempt to set a LOCK_MAND lock via flock(2). This support has been removed and the request ignored.\n", current->comm, current->pid);
Kuniyuki Iwashimadb4abb42022-07-16 21:35:32 -07002077 return 0;
Jeff Layton90f7d7a2021-09-10 15:36:29 -04002078 }
2079
Kuniyuki Iwashimadb4abb42022-07-16 21:35:32 -07002080 type = flock_translate_cmd(cmd & ~LOCK_NB);
2081 if (type < 0)
2082 return type;
2083
2084 error = -EBADF;
2085 f = fdget(fd);
2086 if (!f.file)
2087 return error;
2088
2089 if (type != F_UNLCK && !(f.file->f_mode & (FMODE_READ | FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 goto out_putf;
Jeff Layton6e129d002014-09-04 10:25:06 -04002091
Kuniyuki Iwashima4149be72022-07-16 21:35:31 -07002092 flock_make_lock(f.file, &fl, type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
Kuniyuki Iwashima4149be72022-07-16 21:35:31 -07002094 error = security_file_lock(f.file, fl.fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 if (error)
Kuniyuki Iwashima4149be72022-07-16 21:35:31 -07002096 goto out_putf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097
Kuniyuki Iwashimadb4abb42022-07-16 21:35:32 -07002098 can_sleep = !(cmd & LOCK_NB);
2099 if (can_sleep)
2100 fl.fl_flags |= FL_SLEEP;
2101
Miklos Szeredide2a4a52018-07-18 15:44:43 +02002102 if (f.file->f_op->flock)
Al Viro2903ff02012-08-28 12:52:22 -04002103 error = f.file->f_op->flock(f.file,
Kuniyuki Iwashimadb4abb42022-07-16 21:35:32 -07002104 (can_sleep) ? F_SETLKW : F_SETLK,
Kuniyuki Iwashima4149be72022-07-16 21:35:31 -07002105 &fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 else
Kuniyuki Iwashima4149be72022-07-16 21:35:31 -07002107 error = locks_lock_file_wait(f.file, &fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
David Howells932c29a2022-08-17 19:41:27 +01002109 locks_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 out_putf:
Al Viro2903ff02012-08-28 12:52:22 -04002111 fdput(f);
Kuniyuki Iwashimadb4abb42022-07-16 21:35:32 -07002112
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 return error;
2114}
2115
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002116/**
2117 * vfs_test_lock - test file byte range lock
2118 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04002119 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002120 *
2121 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2122 * setting conf->fl_type to something other than F_UNLCK.
2123 */
2124int vfs_test_lock(struct file *filp, struct file_lock *fl)
2125{
Jeff Layton7e8e5cc82022-11-11 06:46:52 -05002126 WARN_ON_ONCE(filp != fl->fl_file);
Miklos Szeredide2a4a52018-07-18 15:44:43 +02002127 if (filp->f_op->lock)
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002128 return filp->f_op->lock(filp, F_GETLK, fl);
2129 posix_test_lock(filp, fl);
2130 return 0;
2131}
2132EXPORT_SYMBOL_GPL(vfs_test_lock);
2133
Benjamin Coddington9d5b86a2017-07-16 10:28:22 -04002134/**
2135 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2136 * @fl: The file_lock who's fl_pid should be translated
2137 * @ns: The namespace into which the pid should be translated
2138 *
2139 * Used to tranlate a fl_pid into a namespace virtual pid number
2140 */
2141static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2142{
2143 pid_t vnr;
2144 struct pid *pid;
2145
2146 if (IS_OFDLCK(fl))
2147 return -1;
2148 if (IS_REMOTELCK(fl))
2149 return fl->fl_pid;
Konstantin Khorenko826d7bc2018-06-08 17:27:11 +03002150 /*
2151 * If the flock owner process is dead and its pid has been already
2152 * freed, the translation below won't work, but we still want to show
2153 * flock owner pid number in init pidns.
2154 */
2155 if (ns == &init_pid_ns)
2156 return (pid_t)fl->fl_pid;
Benjamin Coddington9d5b86a2017-07-16 10:28:22 -04002157
2158 rcu_read_lock();
2159 pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2160 vnr = pid_nr_ns(pid, ns);
2161 rcu_read_unlock();
2162 return vnr;
2163}
2164
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05002165static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2166{
Benjamin Coddington9d5b86a2017-07-16 10:28:22 -04002167 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05002168#if BITS_PER_LONG == 32
2169 /*
2170 * Make sure we can represent the posix lock via
2171 * legacy 32bit flock.
2172 */
2173 if (fl->fl_start > OFFT_OFFSET_MAX)
2174 return -EOVERFLOW;
2175 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2176 return -EOVERFLOW;
2177#endif
2178 flock->l_start = fl->fl_start;
2179 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2180 fl->fl_end - fl->fl_start + 1;
2181 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04002182 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05002183 return 0;
2184}
2185
2186#if BITS_PER_LONG == 32
2187static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2188{
Benjamin Coddington9d5b86a2017-07-16 10:28:22 -04002189 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05002190 flock->l_start = fl->fl_start;
2191 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2192 fl->fl_end - fl->fl_start + 1;
2193 flock->l_whence = 0;
2194 flock->l_type = fl->fl_type;
2195}
2196#endif
2197
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198/* Report the first existing lock that would conflict with l.
2199 * This implements the F_GETLK command of fcntl().
2200 */
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002201int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202{
Benjamin Coddington52306e82017-07-16 10:28:21 -04002203 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 int error;
2205
Benjamin Coddington52306e82017-07-16 10:28:21 -04002206 fl = locks_alloc_lock();
2207 if (fl == NULL)
2208 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 error = -EINVAL;
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002210 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 goto out;
2212
Benjamin Coddington52306e82017-07-16 10:28:21 -04002213 error = flock_to_posix_lock(filp, fl, flock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 if (error)
2215 goto out;
2216
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002217 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05002218 error = -EINVAL;
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002219 if (flock->l_pid != 0)
Jeff Layton90478932014-03-04 10:30:23 -05002220 goto out;
2221
Benjamin Coddington52306e82017-07-16 10:28:21 -04002222 fl->fl_flags |= FL_OFDLCK;
2223 fl->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002224 }
2225
Benjamin Coddington52306e82017-07-16 10:28:21 -04002226 error = vfs_test_lock(filp, fl);
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002227 if (error)
2228 goto out;
NeilBrown7bbd1fc2018-11-30 10:04:08 +11002229
Benjamin Coddington52306e82017-07-16 10:28:21 -04002230 flock->l_type = fl->fl_type;
2231 if (fl->fl_type != F_UNLCK) {
2232 error = posix_lock_to_flock(flock, fl);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05002233 if (error)
Benjamin Coddington52306e82017-07-16 10:28:21 -04002234 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236out:
Benjamin Coddington52306e82017-07-16 10:28:21 -04002237 locks_free_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 return error;
2239}
2240
Marc Eshel7723ec92007-01-18 15:08:55 -05002241/**
2242 * vfs_lock_file - file byte range lock
2243 * @filp: The file to apply the lock to
2244 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2245 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05002246 * @conf: Place to return a copy of the conflicting lock, if found.
2247 *
2248 * A caller that doesn't care about the conflicting lock may pass NULL
2249 * as the final argument.
2250 *
2251 * If the filesystem defines a private ->lock() method, then @conf will
2252 * be left unchanged; so a caller that cares should initialize it to
2253 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05002254 *
2255 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2256 * locks, the ->lock() interface may return asynchronously, before the lock has
2257 * been granted or denied by the underlying filesystem, if (and only if)
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002258 * lm_grant is set. Callers expecting ->lock() to return asynchronously
Marc Eshel2beb6612006-12-05 23:31:28 -05002259 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2260 * the request is for a blocking lock. When ->lock() does return asynchronously,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002261 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05002262 * request completes.
2263 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07002264 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2265 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05002266 * nonzero return code and the file system should release the lock. The file
2267 * system is also responsible to keep a corresponding posix lock when it
2268 * grants a lock so the VFS can find out which locks are locally held and do
2269 * the correct lock cleanup when required.
2270 * The underlying filesystem must not drop the kernel lock or call
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002271 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05002272 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05002273 */
Marc Eshel150b3932007-01-18 16:15:35 -05002274int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05002275{
Jeff Layton7e8e5cc82022-11-11 06:46:52 -05002276 WARN_ON_ONCE(filp != fl->fl_file);
Miklos Szeredide2a4a52018-07-18 15:44:43 +02002277 if (filp->f_op->lock)
Marc Eshel7723ec92007-01-18 15:08:55 -05002278 return filp->f_op->lock(filp, cmd, fl);
2279 else
Marc Eshel150b3932007-01-18 16:15:35 -05002280 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05002281}
2282EXPORT_SYMBOL_GPL(vfs_lock_file);
2283
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002284static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2285 struct file_lock *fl)
2286{
2287 int error;
2288
2289 error = security_file_lock(filp, fl->fl_type);
2290 if (error)
2291 return error;
2292
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002293 for (;;) {
2294 error = vfs_lock_file(filp, cmd, fl, NULL);
2295 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002296 break;
Linus Torvaldsdcf23ac2020-03-18 07:52:21 -04002297 error = wait_event_interruptible(fl->fl_wait,
2298 list_empty(&fl->fl_blocked_member));
NeilBrown16306a62018-11-30 10:04:08 +11002299 if (error)
2300 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002301 }
NeilBrown16306a62018-11-30 10:04:08 +11002302 locks_delete_block(fl);
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002303
2304 return error;
2305}
2306
Benjamin Coddington6ca7d912015-10-15 09:07:07 -04002307/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002308static int
2309check_fmode_for_setlk(struct file_lock *fl)
2310{
2311 switch (fl->fl_type) {
2312 case F_RDLCK:
2313 if (!(fl->fl_file->f_mode & FMODE_READ))
2314 return -EBADF;
2315 break;
2316 case F_WRLCK:
2317 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2318 return -EBADF;
2319 }
2320 return 0;
2321}
2322
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323/* Apply the lock described by l to an open file descriptor.
2324 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2325 */
Peter Staubachc2936212005-07-27 11:45:09 -07002326int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002327 struct flock *flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328{
2329 struct file_lock *file_lock = locks_alloc_lock();
Jeff Laytonc65454a2022-11-25 08:48:37 -05002330 struct inode *inode = file_inode(filp);
Al Viro0b2bac22008-05-06 13:58:34 -04002331 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 int error;
2333
2334 if (file_lock == NULL)
2335 return -ENOLCK;
2336
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002337 error = flock_to_posix_lock(filp, file_lock, flock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 if (error)
2339 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002340
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002341 error = check_fmode_for_setlk(file_lock);
2342 if (error)
2343 goto out;
2344
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002345 /*
2346 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002347 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002348 */
2349 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002350 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002351 error = -EINVAL;
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002352 if (flock->l_pid != 0)
Jeff Layton90478932014-03-04 10:30:23 -05002353 goto out;
2354
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002355 cmd = F_SETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002356 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002357 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002358 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002359 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002360 error = -EINVAL;
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002361 if (flock->l_pid != 0)
Jeff Layton90478932014-03-04 10:30:23 -05002362 goto out;
2363
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002364 cmd = F_SETLKW;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002365 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002366 file_lock->fl_owner = filp;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002367 fallthrough;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002368 case F_SETLKW:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 file_lock->fl_flags |= FL_SLEEP;
2370 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002371
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002372 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
Peter Staubachc2936212005-07-27 11:45:09 -07002374 /*
Jeff Layton0752ba802016-01-08 07:30:43 -05002375 * Attempt to detect a close/fcntl race and recover by releasing the
2376 * lock that was just acquired. There is no need to do that when we're
2377 * unlocking though, or for OFD locks.
Peter Staubachc2936212005-07-27 11:45:09 -07002378 */
Jeff Layton0752ba802016-01-08 07:30:43 -05002379 if (!error && file_lock->fl_type != F_UNLCK &&
2380 !(file_lock->fl_flags & FL_OFDLCK)) {
Eric W. Biederman120ce2b2020-11-20 17:14:25 -06002381 struct files_struct *files = current->files;
Jeff Layton7f3697e2016-01-07 16:38:10 -05002382 /*
2383 * We need that spin_lock here - it prevents reordering between
2384 * update of i_flctx->flc_posix and check for it done in
2385 * close(). rcu_read_lock() wouldn't do.
2386 */
Eric W. Biederman120ce2b2020-11-20 17:14:25 -06002387 spin_lock(&files->file_lock);
2388 f = files_lookup_fd_locked(files, fd);
2389 spin_unlock(&files->file_lock);
Jeff Layton7f3697e2016-01-07 16:38:10 -05002390 if (f != filp) {
2391 file_lock->fl_type = F_UNLCK;
2392 error = do_lock_file_wait(filp, cmd, file_lock);
2393 WARN_ON_ONCE(error);
2394 error = -EBADF;
2395 }
Peter Staubachc2936212005-07-27 11:45:09 -07002396 }
Peter Staubachc2936212005-07-27 11:45:09 -07002397out:
Jeff Layton18909102016-01-06 21:26:10 -05002398 trace_fcntl_setlk(inode, file_lock, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 locks_free_lock(file_lock);
2400 return error;
2401}
2402
2403#if BITS_PER_LONG == 32
2404/* Report the first existing lock that would conflict with l.
2405 * This implements the F_GETLK command of fcntl().
2406 */
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002407int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408{
Benjamin Coddington52306e82017-07-16 10:28:21 -04002409 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 int error;
2411
Benjamin Coddington52306e82017-07-16 10:28:21 -04002412 fl = locks_alloc_lock();
2413 if (fl == NULL)
2414 return -ENOMEM;
2415
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 error = -EINVAL;
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002417 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 goto out;
2419
Benjamin Coddington52306e82017-07-16 10:28:21 -04002420 error = flock64_to_posix_lock(filp, fl, flock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 if (error)
2422 goto out;
2423
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002424 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05002425 error = -EINVAL;
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002426 if (flock->l_pid != 0)
Jeff Layton90478932014-03-04 10:30:23 -05002427 goto out;
2428
Benjamin Coddington52306e82017-07-16 10:28:21 -04002429 fl->fl_flags |= FL_OFDLCK;
2430 fl->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002431 }
2432
Benjamin Coddington52306e82017-07-16 10:28:21 -04002433 error = vfs_test_lock(filp, fl);
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002434 if (error)
2435 goto out;
2436
Benjamin Coddington52306e82017-07-16 10:28:21 -04002437 flock->l_type = fl->fl_type;
2438 if (fl->fl_type != F_UNLCK)
2439 posix_lock_to_flock64(flock, fl);
Kinglong Meef3282962014-08-22 10:18:43 -04002440
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441out:
Benjamin Coddington52306e82017-07-16 10:28:21 -04002442 locks_free_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 return error;
2444}
2445
2446/* Apply the lock described by l to an open file descriptor.
2447 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2448 */
Peter Staubachc2936212005-07-27 11:45:09 -07002449int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002450 struct flock64 *flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451{
2452 struct file_lock *file_lock = locks_alloc_lock();
Al Viro0b2bac22008-05-06 13:58:34 -04002453 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 int error;
2455
2456 if (file_lock == NULL)
2457 return -ENOLCK;
2458
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002459 error = flock64_to_posix_lock(filp, file_lock, flock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 if (error)
2461 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002462
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002463 error = check_fmode_for_setlk(file_lock);
2464 if (error)
2465 goto out;
2466
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002467 /*
2468 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002469 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002470 */
2471 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002472 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002473 error = -EINVAL;
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002474 if (flock->l_pid != 0)
Jeff Layton90478932014-03-04 10:30:23 -05002475 goto out;
2476
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002477 cmd = F_SETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002478 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002479 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002480 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002481 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002482 error = -EINVAL;
Christoph Hellwiga75d30c2017-05-27 06:07:19 -04002483 if (flock->l_pid != 0)
Jeff Layton90478932014-03-04 10:30:23 -05002484 goto out;
2485
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002486 cmd = F_SETLKW64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002487 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002488 file_lock->fl_owner = filp;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002489 fallthrough;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002490 case F_SETLKW64:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 file_lock->fl_flags |= FL_SLEEP;
2492 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002493
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002494 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495
Peter Staubachc2936212005-07-27 11:45:09 -07002496 /*
Jeff Layton0752ba802016-01-08 07:30:43 -05002497 * Attempt to detect a close/fcntl race and recover by releasing the
2498 * lock that was just acquired. There is no need to do that when we're
2499 * unlocking though, or for OFD locks.
Peter Staubachc2936212005-07-27 11:45:09 -07002500 */
Jeff Layton0752ba802016-01-08 07:30:43 -05002501 if (!error && file_lock->fl_type != F_UNLCK &&
2502 !(file_lock->fl_flags & FL_OFDLCK)) {
Eric W. Biederman120ce2b2020-11-20 17:14:25 -06002503 struct files_struct *files = current->files;
Jeff Layton7f3697e2016-01-07 16:38:10 -05002504 /*
2505 * We need that spin_lock here - it prevents reordering between
2506 * update of i_flctx->flc_posix and check for it done in
2507 * close(). rcu_read_lock() wouldn't do.
2508 */
Eric W. Biederman120ce2b2020-11-20 17:14:25 -06002509 spin_lock(&files->file_lock);
2510 f = files_lookup_fd_locked(files, fd);
2511 spin_unlock(&files->file_lock);
Jeff Layton7f3697e2016-01-07 16:38:10 -05002512 if (f != filp) {
2513 file_lock->fl_type = F_UNLCK;
2514 error = do_lock_file_wait(filp, cmd, file_lock);
2515 WARN_ON_ONCE(error);
2516 error = -EBADF;
2517 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519out:
2520 locks_free_lock(file_lock);
2521 return error;
2522}
2523#endif /* BITS_PER_LONG == 32 */
2524
2525/*
2526 * This function is called when the file is being removed
2527 * from the task's fd array. POSIX locks belonging to this task
2528 * are deleted at this time.
2529 */
2530void locks_remove_posix(struct file *filp, fl_owner_t owner)
2531{
Jeff Layton18909102016-01-06 21:26:10 -05002532 int error;
Jeff Laytonc65454a2022-11-25 08:48:37 -05002533 struct inode *inode = file_inode(filp);
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002534 struct file_lock lock;
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002535 struct file_lock_context *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536
2537 /*
2538 * If there are no locks held on this file, we don't need to call
2539 * posix_lock_file(). Another process could be setting a lock on this
2540 * file at the same time, but we wouldn't remove that lock anyway.
2541 */
Jeff Layton401a8b82022-11-16 09:02:30 -05002542 ctx = locks_inode_context(inode);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05002543 if (!ctx || list_empty(&ctx->flc_posix))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 return;
2545
NeilBrownd6367d62018-11-30 10:04:08 +11002546 locks_init_lock(&lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002548 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 lock.fl_start = 0;
2550 lock.fl_end = OFFSET_MAX;
2551 lock.fl_owner = owner;
2552 lock.fl_pid = current->tgid;
2553 lock.fl_file = filp;
2554 lock.fl_ops = NULL;
2555 lock.fl_lmops = NULL;
2556
Jeff Layton18909102016-01-06 21:26:10 -05002557 error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2560 lock.fl_ops->fl_release_private(&lock);
Miklos Szeredic568d682016-09-16 12:44:20 +02002561 trace_locks_remove_posix(inode, &lock, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563EXPORT_SYMBOL(locks_remove_posix);
2564
Jeff Layton3d8e5602015-01-16 15:05:58 -05002565/* The i_flctx must be valid when calling into here */
Jeff Laytondd459bb2015-01-16 15:05:54 -05002566static void
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002567locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
Jeff Laytondd459bb2015-01-16 15:05:54 -05002568{
NeilBrownd6367d62018-11-30 10:04:08 +11002569 struct file_lock fl;
Jeff Laytonc65454a2022-11-25 08:48:37 -05002570 struct inode *inode = file_inode(filp);
Jeff Laytondd459bb2015-01-16 15:05:54 -05002571
Jeff Layton3d8e5602015-01-16 15:05:58 -05002572 if (list_empty(&flctx->flc_flock))
Jeff Laytondd459bb2015-01-16 15:05:54 -05002573 return;
2574
Kuniyuki Iwashima4149be72022-07-16 21:35:31 -07002575 flock_make_lock(filp, &fl, F_UNLCK);
NeilBrownd6367d62018-11-30 10:04:08 +11002576 fl.fl_flags |= FL_CLOSE;
2577
Miklos Szeredide2a4a52018-07-18 15:44:43 +02002578 if (filp->f_op->flock)
Jeff Laytondd459bb2015-01-16 15:05:54 -05002579 filp->f_op->flock(filp, F_SETLKW, &fl);
2580 else
Jeff Laytonbcd7f78d2015-07-11 06:43:02 -04002581 flock_lock_inode(inode, &fl);
Jeff Laytondd459bb2015-01-16 15:05:54 -05002582
2583 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2584 fl.fl_ops->fl_release_private(&fl);
2585}
2586
Jeff Layton3d8e5602015-01-16 15:05:58 -05002587/* The i_flctx must be valid when calling into here */
Jeff Layton8634b512015-01-16 15:05:55 -05002588static void
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002589locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
Jeff Layton8634b512015-01-16 15:05:55 -05002590{
Jeff Layton8634b512015-01-16 15:05:55 -05002591 struct file_lock *fl, *tmp;
2592 LIST_HEAD(dispose);
2593
Jeff Layton3d8e5602015-01-16 15:05:58 -05002594 if (list_empty(&ctx->flc_lease))
Jeff Layton8634b512015-01-16 15:05:55 -05002595 return;
2596
Peter Zijlstra02e525b22019-02-21 15:38:40 +01002597 percpu_down_read(&file_rwsem);
Jeff Layton6109c852015-01-16 15:05:57 -05002598 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05002599 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
Jeff Laytonc4e136c2015-02-16 19:37:42 -05002600 if (filp == fl->fl_file)
2601 lease_modify(fl, F_UNLCK, &dispose);
Jeff Layton6109c852015-01-16 15:05:57 -05002602 spin_unlock(&ctx->flc_lock);
Peter Zijlstra02e525b22019-02-21 15:38:40 +01002603 percpu_up_read(&file_rwsem);
Peter Zijlstra5f430862016-10-08 10:12:28 +02002604
Jeff Layton8634b512015-01-16 15:05:55 -05002605 locks_dispose_list(&dispose);
2606}
2607
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608/*
2609 * This function is called on the last close of an open file.
2610 */
Jeff Layton78ed8a12014-02-03 12:13:08 -05002611void locks_remove_file(struct file *filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612{
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002613 struct file_lock_context *ctx;
2614
Jeff Laytonc65454a2022-11-25 08:48:37 -05002615 ctx = locks_inode_context(file_inode(filp));
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002616 if (!ctx)
Jeff Layton3d8e5602015-01-16 15:05:58 -05002617 return;
2618
Jeff Laytondd459bb2015-01-16 15:05:54 -05002619 /* remove any OFD locks */
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002620 locks_remove_posix(filp, filp);
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002621
Jeff Laytondd459bb2015-01-16 15:05:54 -05002622 /* remove flock locks */
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002623 locks_remove_flock(filp, ctx);
Jeff Laytondd459bb2015-01-16 15:05:54 -05002624
Jeff Layton8634b512015-01-16 15:05:55 -05002625 /* remove any leases */
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002626 locks_remove_lease(filp, ctx);
Benjamin Coddington39537042017-07-21 13:36:25 -04002627
2628 spin_lock(&ctx->flc_lock);
2629 locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2630 locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2631 locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2632 spin_unlock(&ctx->flc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633}
2634
2635/**
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002636 * vfs_cancel_lock - file byte range unblock lock
2637 * @filp: The file to apply the unblock to
2638 * @fl: The lock to be unblocked
2639 *
2640 * Used by lock managers to cancel blocked requests
2641 */
2642int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2643{
Jeff Layton7e8e5cc82022-11-11 06:46:52 -05002644 WARN_ON_ONCE(filp != fl->fl_file);
Miklos Szeredide2a4a52018-07-18 15:44:43 +02002645 if (filp->f_op->lock)
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002646 return filp->f_op->lock(filp, F_CANCELLK, fl);
2647 return 0;
2648}
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002649EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2650
Jeff Laytonab1ddef2022-11-14 08:33:09 -05002651/**
2652 * vfs_inode_has_locks - are any file locks held on @inode?
2653 * @inode: inode to check for locks
2654 *
2655 * Return true if there are any FL_POSIX or FL_FLOCK locks currently
2656 * set on @inode.
2657 */
2658bool vfs_inode_has_locks(struct inode *inode)
2659{
2660 struct file_lock_context *ctx;
2661 bool ret;
2662
Jeff Layton401a8b82022-11-16 09:02:30 -05002663 ctx = locks_inode_context(inode);
Jeff Laytonab1ddef2022-11-14 08:33:09 -05002664 if (!ctx)
2665 return false;
2666
2667 spin_lock(&ctx->flc_lock);
2668 ret = !list_empty(&ctx->flc_posix) || !list_empty(&ctx->flc_flock);
2669 spin_unlock(&ctx->flc_lock);
2670 return ret;
2671}
2672EXPORT_SYMBOL_GPL(vfs_inode_has_locks);
2673
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002674#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002675#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002676#include <linux/seq_file.h>
2677
Jeff Layton7012b022013-06-21 08:58:22 -04002678struct locks_iterator {
2679 int li_cpu;
2680 loff_t li_pos;
2681};
2682
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002683static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Luo Longjunb8da9b12021-02-25 22:58:29 -05002684 loff_t id, char *pfx, int repeat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685{
2686 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002687 unsigned int fl_pid;
Alexey Gladkov9d78ede2020-05-18 20:07:38 +02002688 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
Jeff Layton90f7d7a2021-09-10 15:36:29 -04002689 int type;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002690
Benjamin Coddington9d5b86a2017-07-16 10:28:22 -04002691 fl_pid = locks_translate_pid(fl, proc_pidns);
2692 /*
Konstantin Khorenko1cf8e5d2018-06-08 17:27:12 +03002693 * If lock owner is dead (and pid is freed) or not visible in current
2694 * pidns, zero is shown as a pid value. Check lock info from
2695 * init_pid_ns to get saved lock pid value.
Benjamin Coddington9d5b86a2017-07-16 10:28:22 -04002696 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697
2698 if (fl->fl_file != NULL)
Jeff Laytonc65454a2022-11-25 08:48:37 -05002699 inode = file_inode(fl->fl_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700
Luo Longjunb8da9b12021-02-25 22:58:29 -05002701 seq_printf(f, "%lld: ", id);
2702
2703 if (repeat)
2704 seq_printf(f, "%*s", repeat - 1 + (int)strlen(pfx), pfx);
2705
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 if (IS_POSIX(fl)) {
Jeff Laytonc918d422014-02-03 12:13:09 -05002707 if (fl->fl_flags & FL_ACCESS)
Fabian Frederick5315c262014-05-09 14:13:05 -04002708 seq_puts(f, "ACCESS");
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002709 else if (IS_OFDLCK(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002710 seq_puts(f, "OFDLCK");
Jeff Laytonc918d422014-02-03 12:13:09 -05002711 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002712 seq_puts(f, "POSIX ");
Jeff Laytonc918d422014-02-03 12:13:09 -05002713
2714 seq_printf(f, " %s ",
Jeff Laytonf7e33bd2021-08-19 14:56:38 -04002715 (inode == NULL) ? "*NOINODE*" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 } else if (IS_FLOCK(fl)) {
Jeff Layton90f7d7a2021-09-10 15:36:29 -04002717 seq_puts(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 } else if (IS_LEASE(fl)) {
Jeff Layton8144f1f2014-08-11 13:36:54 -04002719 if (fl->fl_flags & FL_DELEG)
2720 seq_puts(f, "DELEG ");
2721 else
2722 seq_puts(f, "LEASE ");
2723
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002724 if (lease_breaking(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002725 seq_puts(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 else if (fl->fl_file)
Fabian Frederick5315c262014-05-09 14:13:05 -04002727 seq_puts(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002729 seq_puts(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002731 seq_puts(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 }
Jeff Layton90f7d7a2021-09-10 15:36:29 -04002733 type = IS_LEASE(fl) ? target_leasetype(fl) : fl->fl_type;
Pavel Begunkov43e4cb92019-07-24 20:16:31 +03002734
Jeff Layton90f7d7a2021-09-10 15:36:29 -04002735 seq_printf(f, "%s ", (type == F_WRLCK) ? "WRITE" :
2736 (type == F_RDLCK) ? "READ" : "UNLCK");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 if (inode) {
Jeff Layton36488882015-04-03 09:04:04 -04002738 /* userspace relies on this representation of dev_t */
Amir Goldstein98ca4802019-12-22 20:45:28 +02002739 seq_printf(f, "%d %02x:%02x:%lu ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 MAJOR(inode->i_sb->s_dev),
2741 MINOR(inode->i_sb->s_dev), inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002743 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 }
2745 if (IS_POSIX(fl)) {
2746 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002747 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002749 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002751 seq_puts(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 }
2753}
2754
Luo Longjunb8da9b12021-02-25 22:58:29 -05002755static struct file_lock *get_next_blocked_member(struct file_lock *node)
2756{
2757 struct file_lock *tmp;
2758
2759 /* NULL node or root node */
2760 if (node == NULL || node->fl_blocker == NULL)
2761 return NULL;
2762
2763 /* Next member in the linked list could be itself */
2764 tmp = list_next_entry(node, fl_blocked_member);
2765 if (list_entry_is_head(tmp, &node->fl_blocker->fl_blocked_requests, fl_blocked_member)
2766 || tmp == node) {
2767 return NULL;
2768 }
2769
2770 return tmp;
2771}
2772
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002773static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774{
Jeff Layton7012b022013-06-21 08:58:22 -04002775 struct locks_iterator *iter = f->private;
Luo Longjunb8da9b12021-02-25 22:58:29 -05002776 struct file_lock *cur, *tmp;
Alexey Gladkov9d78ede2020-05-18 20:07:38 +02002777 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
Luo Longjunb8da9b12021-02-25 22:58:29 -05002778 int level = 0;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002779
Luo Longjunb8da9b12021-02-25 22:58:29 -05002780 cur = hlist_entry(v, struct file_lock, fl_link);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002781
Luo Longjunb8da9b12021-02-25 22:58:29 -05002782 if (locks_translate_pid(cur, proc_pidns) == 0)
Nikolay Borisovd67fd442016-08-17 16:18:46 -04002783 return 0;
2784
Luo Longjunb8da9b12021-02-25 22:58:29 -05002785 /* View this crossed linked list as a binary tree, the first member of fl_blocked_requests
2786 * is the left child of current node, the next silibing in fl_blocked_member is the
2787 * right child, we can alse get the parent of current node from fl_blocker, so this
2788 * question becomes traversal of a binary tree
2789 */
2790 while (cur != NULL) {
2791 if (level)
2792 lock_get_status(f, cur, iter->li_pos, "-> ", level);
2793 else
2794 lock_get_status(f, cur, iter->li_pos, "", level);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002795
Luo Longjunb8da9b12021-02-25 22:58:29 -05002796 if (!list_empty(&cur->fl_blocked_requests)) {
2797 /* Turn left */
2798 cur = list_first_entry_or_null(&cur->fl_blocked_requests,
2799 struct file_lock, fl_blocked_member);
2800 level++;
2801 } else {
2802 /* Turn right */
2803 tmp = get_next_blocked_member(cur);
2804 /* Fall back to parent node */
2805 while (tmp == NULL && cur->fl_blocker != NULL) {
2806 cur = cur->fl_blocker;
2807 level--;
2808 tmp = get_next_blocked_member(cur);
2809 }
2810 cur = tmp;
2811 }
2812 }
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002813
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002814 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815}
2816
Andrey Vagin6c8c9032015-04-16 12:49:38 -07002817static void __show_fd_locks(struct seq_file *f,
2818 struct list_head *head, int *id,
2819 struct file *filp, struct files_struct *files)
2820{
2821 struct file_lock *fl;
2822
2823 list_for_each_entry(fl, head, fl_list) {
2824
2825 if (filp != fl->fl_file)
2826 continue;
2827 if (fl->fl_owner != files &&
2828 fl->fl_owner != filp)
2829 continue;
2830
2831 (*id)++;
2832 seq_puts(f, "lock:\t");
Luo Longjunb8da9b12021-02-25 22:58:29 -05002833 lock_get_status(f, fl, *id, "", 0);
Andrey Vagin6c8c9032015-04-16 12:49:38 -07002834 }
2835}
2836
2837void show_fd_locks(struct seq_file *f,
2838 struct file *filp, struct files_struct *files)
2839{
Jeff Laytonc65454a2022-11-25 08:48:37 -05002840 struct inode *inode = file_inode(filp);
Andrey Vagin6c8c9032015-04-16 12:49:38 -07002841 struct file_lock_context *ctx;
2842 int id = 0;
2843
Jeff Layton401a8b82022-11-16 09:02:30 -05002844 ctx = locks_inode_context(inode);
Andrey Vagin6c8c9032015-04-16 12:49:38 -07002845 if (!ctx)
2846 return;
2847
2848 spin_lock(&ctx->flc_lock);
2849 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2850 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2851 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2852 spin_unlock(&ctx->flc_lock);
2853}
2854
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002855static void *locks_start(struct seq_file *f, loff_t *pos)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002856 __acquires(&blocked_lock_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857{
Jeff Layton7012b022013-06-21 08:58:22 -04002858 struct locks_iterator *iter = f->private;
Jerome Marchand99dc8292010-10-26 14:22:33 -07002859
Jeff Layton7012b022013-06-21 08:58:22 -04002860 iter->li_pos = *pos + 1;
Peter Zijlstraaba37662015-06-22 14:16:33 +02002861 percpu_down_write(&file_rwsem);
Jeff Layton7b2296a2013-06-21 08:58:20 -04002862 spin_lock(&blocked_lock_lock);
Peter Zijlstra7c3f6542015-06-22 14:16:34 +02002863 return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864}
2865
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002866static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2867{
Jeff Layton7012b022013-06-21 08:58:22 -04002868 struct locks_iterator *iter = f->private;
2869
2870 ++iter->li_pos;
Peter Zijlstra7c3f6542015-06-22 14:16:34 +02002871 return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002872}
2873
2874static void locks_stop(struct seq_file *f, void *v)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002875 __releases(&blocked_lock_lock)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002876{
Jeff Layton7b2296a2013-06-21 08:58:20 -04002877 spin_unlock(&blocked_lock_lock);
Peter Zijlstraaba37662015-06-22 14:16:33 +02002878 percpu_up_write(&file_rwsem);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002879}
2880
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002881static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002882 .start = locks_start,
2883 .next = locks_next,
2884 .stop = locks_stop,
2885 .show = locks_show,
2886};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002887
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002888static int __init proc_locks_init(void)
2889{
Christoph Hellwig44414d82018-04-24 17:05:17 +02002890 proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
2891 sizeof(struct locks_iterator), NULL);
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002892 return 0;
2893}
Paul Gortmaker91899222015-12-17 14:11:03 -05002894fs_initcall(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002895#endif
2896
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897static int __init filelock_init(void)
2898{
Jeff Layton7012b022013-06-21 08:58:22 -04002899 int i;
2900
Jeff Layton4a075e32015-01-16 15:05:54 -05002901 flctx_cache = kmem_cache_create("file_lock_ctx",
Linus Torvalds37547072021-09-07 11:21:48 -07002902 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
Jeff Layton4a075e32015-01-16 15:05:54 -05002903
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 filelock_cache = kmem_cache_create("file_lock_cache",
Linus Torvalds37547072021-09-07 11:21:48 -07002905 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
Miklos Szerediee19cc42011-07-07 13:06:09 +02002906
Peter Zijlstra7c3f6542015-06-22 14:16:34 +02002907 for_each_possible_cpu(i) {
2908 struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2909
2910 spin_lock_init(&fll->lock);
2911 INIT_HLIST_HEAD(&fll->hlist);
2912 }
Jeff Layton7012b022013-06-21 08:58:22 -04002913
Jeff Layton18f66222019-08-18 14:18:45 -04002914 lease_notifier_chain_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 return 0;
2916}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917core_initcall(filelock_init);