blob: 734ed584a946e185bacf6d0586e7ab8205fe0435 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/fs/super.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * super.c contains code to handle: - mount structures
8 * - super-block tables
9 * - filesystem drivers list
10 * - mount system call
11 * - umount system call
12 * - ustat system call
13 *
14 * GK 2/5/95 - Changed to support mounting the root fs via NFS
15 *
16 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
17 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
18 * Added options to /proc/mounts:
Jan Engelhardt96de0e22007-10-19 23:21:04 +020019 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
Linus Torvalds1da177e2005-04-16 15:20:36 -070020 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
21 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
22 */
23
Paul Gortmaker630d9c42011-11-16 23:57:37 -050024#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/security.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/writeback.h> /* for the emergency remount stuff */
30#include <linux/idr.h>
Ingo Molnar353ab6e2006-03-26 01:37:12 -080031#include <linux/mutex.h>
Jens Axboe5477d0f2010-04-29 20:33:35 +020032#include <linux/backing-dev.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110033#include <linux/rculist_bl.h>
Eric Biggers22d94f42019-08-04 19:35:46 -070034#include <linux/fscrypt.h>
Al Viro40401532012-02-13 03:58:52 +000035#include <linux/fsnotify.h>
Jan Kara5accdf82012-06-12 16:20:34 +020036#include <linux/lockdep.h>
Eric W. Biederman6e4eab52016-05-24 09:29:01 -050037#include <linux/user_namespace.h>
David Howells9bc61ab2018-11-04 03:19:03 -050038#include <linux/fs_context.h>
David Howellse262e32d2018-11-01 23:07:23 +000039#include <uapi/linux/mount.h>
Al Viro6d59e7f2008-03-22 15:48:17 -040040#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +020042static int thaw_super_locked(struct super_block *sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Al Viro15d0f5e2015-02-02 10:07:59 -070044static LIST_HEAD(super_blocks);
45static DEFINE_SPINLOCK(sb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Jan Kara5accdf82012-06-12 16:20:34 +020047static char *sb_writers_name[SB_FREEZE_LEVELS] = {
48 "sb_writers",
49 "sb_pagefaults",
50 "sb_internal",
51};
52
Dave Chinnerb0d40c92011-07-08 14:14:42 +100053/*
54 * One thing we have to be careful of with a per-sb shrinker is that we don't
55 * drop the last active reference to the superblock from within the shrinker.
56 * If that happens we could trigger unregistering the shrinker from within the
57 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
58 * take a passive reference to the superblock to avoid this from occurring.
59 */
Dave Chinner0a234c62013-08-28 10:17:57 +100060static unsigned long super_cache_scan(struct shrinker *shrink,
61 struct shrink_control *sc)
Dave Chinnerb0d40c92011-07-08 14:14:42 +100062{
63 struct super_block *sb;
Dave Chinner0a234c62013-08-28 10:17:57 +100064 long fs_objects = 0;
65 long total_objects;
66 long freed = 0;
67 long dentries;
68 long inodes;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100069
70 sb = container_of(shrink, struct super_block, s_shrink);
71
72 /*
73 * Deadlock avoidance. We may hold various FS locks, and we don't want
74 * to recurse into the FS that called us in clear_inode() and friends..
75 */
Dave Chinner0a234c62013-08-28 10:17:57 +100076 if (!(sc->gfp_mask & __GFP_FS))
77 return SHRINK_STOP;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100078
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +030079 if (!trylock_super(sb))
Dave Chinner0a234c62013-08-28 10:17:57 +100080 return SHRINK_STOP;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100081
Al Virod0407902013-07-19 21:12:31 +040082 if (sb->s_op->nr_cached_objects)
Vladimir Davydov4101b622015-02-12 14:58:51 -080083 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100084
Vladimir Davydov503c3582015-02-12 14:58:47 -080085 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
86 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
Dave Chinnerf6041562013-08-28 10:18:00 +100087 total_objects = dentries + inodes + fs_objects + 1;
Tetsuo Handa475d0db2014-05-17 20:56:38 +090088 if (!total_objects)
89 total_objects = 1;
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100090
Dave Chinner0a234c62013-08-28 10:17:57 +100091 /* proportion the scan between the caches */
Dave Chinnerf6041562013-08-28 10:18:00 +100092 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +100093 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
Vladimir Davydov503c3582015-02-12 14:58:47 -080094 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
Dave Chinnerb0d40c92011-07-08 14:14:42 +100095
Dave Chinner0a234c62013-08-28 10:17:57 +100096 /*
97 * prune the dcache first as the icache is pinned by it, then
98 * prune the icache, followed by the filesystem specific caches
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -080099 *
100 * Ensure that we always scan at least one object - memcg kmem
101 * accounting uses this to fully empty the caches.
Dave Chinner0a234c62013-08-28 10:17:57 +1000102 */
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -0800103 sc->nr_to_scan = dentries + 1;
Vladimir Davydov503c3582015-02-12 14:58:47 -0800104 freed = prune_dcache_sb(sb, sc);
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -0800105 sc->nr_to_scan = inodes + 1;
Vladimir Davydov503c3582015-02-12 14:58:47 -0800106 freed += prune_icache_sb(sb, sc);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000107
Dave Chinner0a234c62013-08-28 10:17:57 +1000108 if (fs_objects) {
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -0800109 sc->nr_to_scan = fs_objects + 1;
Vladimir Davydov4101b622015-02-12 14:58:51 -0800110 freed += sb->s_op->free_cached_objects(sb, sc);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000111 }
112
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300113 up_read(&sb->s_umount);
Dave Chinner0a234c62013-08-28 10:17:57 +1000114 return freed;
115}
116
117static unsigned long super_cache_count(struct shrinker *shrink,
118 struct shrink_control *sc)
119{
120 struct super_block *sb;
121 long total_objects = 0;
122
123 sb = container_of(shrink, struct super_block, s_shrink);
124
Tim Chend23da152014-06-04 16:10:47 -0700125 /*
Dave Chinner79f546a2018-05-11 11:20:57 +1000126 * We don't call trylock_super() here as it is a scalability bottleneck,
127 * so we're exposed to partial setup state. The shrinker rwsem does not
128 * protect filesystem operations backing list_lru_shrink_count() or
129 * s_op->nr_cached_objects(). Counts can change between
130 * super_cache_count and super_cache_scan, so we really don't need locks
131 * here.
132 *
133 * However, if we are currently mounting the superblock, the underlying
134 * filesystem might be in a state of partial construction and hence it
135 * is dangerous to access it. trylock_super() uses a SB_BORN check to
136 * avoid this situation, so do the same here. The memory barrier is
137 * matched with the one in mount_fs() as we don't hold locks here.
Tim Chend23da152014-06-04 16:10:47 -0700138 */
Dave Chinner79f546a2018-05-11 11:20:57 +1000139 if (!(sb->s_flags & SB_BORN))
140 return 0;
141 smp_rmb();
142
Dave Chinner0a234c62013-08-28 10:17:57 +1000143 if (sb->s_op && sb->s_op->nr_cached_objects)
Vladimir Davydov4101b622015-02-12 14:58:51 -0800144 total_objects = sb->s_op->nr_cached_objects(sb, sc);
Dave Chinner0a234c62013-08-28 10:17:57 +1000145
Vladimir Davydov503c3582015-02-12 14:58:47 -0800146 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
147 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
Dave Chinner0a234c62013-08-28 10:17:57 +1000148
Kirill Tkhai9b996462018-08-17 15:48:21 -0700149 if (!total_objects)
150 return SHRINK_EMPTY;
151
Glauber Costa55f841c2013-08-28 10:17:53 +1000152 total_objects = vfs_pressure_ratio(total_objects);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000153 return total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000154}
155
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200156static void destroy_super_work(struct work_struct *work)
157{
158 struct super_block *s = container_of(work, struct super_block,
159 destroy_work);
160 int i;
161
162 for (i = 0; i < SB_FREEZE_LEVELS; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200163 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200164 kfree(s);
165}
166
167static void destroy_super_rcu(struct rcu_head *head)
168{
169 struct super_block *s = container_of(head, struct super_block, rcu);
170 INIT_WORK(&s->destroy_work, destroy_super_work);
171 schedule_work(&s->destroy_work);
172}
173
Al Viro02008942017-10-11 13:48:55 -0400174/* Free a superblock that has never been seen by anyone */
175static void destroy_unused_super(struct super_block *s)
Jan Kara5accdf82012-06-12 16:20:34 +0200176{
Al Viro02008942017-10-11 13:48:55 -0400177 if (!s)
178 return;
179 up_write(&s->s_umount);
Al Viro7eb5e882013-10-01 15:09:58 -0400180 list_lru_destroy(&s->s_dentry_lru);
181 list_lru_destroy(&s->s_inode_lru);
Al Viro7eb5e882013-10-01 15:09:58 -0400182 security_sb_free(s);
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500183 put_user_ns(s->s_user_ns);
Al Viro7eb5e882013-10-01 15:09:58 -0400184 kfree(s->s_subtype);
Tetsuo Handa8e049442018-04-04 19:53:07 +0900185 free_prealloced_shrinker(&s->s_shrink);
Al Viro02008942017-10-11 13:48:55 -0400186 /* no delays needed */
187 destroy_super_work(&s->destroy_work);
Jan Kara5accdf82012-06-12 16:20:34 +0200188}
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190/**
191 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700192 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100193 * @flags: the mount flags
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500194 * @user_ns: User namespace for the super_block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 *
196 * Allocates and initializes a new &struct super_block. alloc_super()
197 * returns a pointer new superblock or %NULL if allocation had failed.
198 */
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500199static struct super_block *alloc_super(struct file_system_type *type, int flags,
200 struct user_namespace *user_ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800202 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700203 static const struct super_operations default_op;
Al Viro7eb5e882013-10-01 15:09:58 -0400204 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
Al Viro7eb5e882013-10-01 15:09:58 -0400206 if (!s)
207 return NULL;
208
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800209 INIT_LIST_HEAD(&s->s_mounts);
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500210 s->s_user_ns = get_user_ns(user_ns);
Al Viroca0168e2017-12-05 09:32:25 -0500211 init_rwsem(&s->s_umount);
212 lockdep_set_class(&s->s_umount, &type->s_umount_key);
213 /*
214 * sget() can have s_umount recursion.
215 *
216 * When it cannot find a suitable sb, it allocates a new
217 * one (this one), and tries again to find a suitable old
218 * one.
219 *
220 * In case that succeeds, it will acquire the s_umount
221 * lock of the old one. Since these are clearly distrinct
222 * locks, and this object isn't exposed yet, there's no
223 * risk of deadlocks.
224 *
225 * Annotate this by putting this lock in a different
226 * subclass.
227 */
228 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800229
Al Viro7eb5e882013-10-01 15:09:58 -0400230 if (security_sb_alloc(s))
231 goto fail;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200232
Al Viro7eb5e882013-10-01 15:09:58 -0400233 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200234 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
235 sb_writers_name[i],
236 &type->s_writers_key[i]))
Al Viro7eb5e882013-10-01 15:09:58 -0400237 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 }
Al Viro7eb5e882013-10-01 15:09:58 -0400239 init_waitqueue_head(&s->s_writers.wait_unfrozen);
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100240 s->s_bdi = &noop_backing_dev_info;
Al Viro7eb5e882013-10-01 15:09:58 -0400241 s->s_flags = flags;
Eric W. Biedermancc50a072016-06-09 15:44:48 -0500242 if (s->s_user_ns != &init_user_ns)
Eric W. Biederman67690f92016-05-18 13:50:06 -0500243 s->s_iflags |= SB_I_NODEV;
Al Viro7eb5e882013-10-01 15:09:58 -0400244 INIT_HLIST_NODE(&s->s_instances);
NeilBrownf1ee6162017-12-21 09:45:40 +1100245 INIT_HLIST_BL_HEAD(&s->s_roots);
Dave Chinnere97fedb2015-03-04 13:40:00 -0500246 mutex_init(&s->s_sync_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400247 INIT_LIST_HEAD(&s->s_inodes);
Dave Chinner74278da92015-03-04 12:37:22 -0500248 spin_lock_init(&s->s_inode_list_lock);
Dave Chinner6c60d2b2016-07-26 15:21:50 -0700249 INIT_LIST_HEAD(&s->s_inodes_wb);
250 spin_lock_init(&s->s_inode_wblist_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400251
Al Viro7eb5e882013-10-01 15:09:58 -0400252 s->s_count = 1;
253 atomic_set(&s->s_active, 1);
254 mutex_init(&s->s_vfs_rename_mutex);
255 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
Jan Karabc8230e2017-06-08 14:39:48 +0200256 init_rwsem(&s->s_dquot.dqio_sem);
Al Viro7eb5e882013-10-01 15:09:58 -0400257 s->s_maxbytes = MAX_NON_LFS;
258 s->s_op = &default_op;
259 s->s_time_gran = 1000000000;
Deepa Dinamani188d20bc2018-01-21 18:04:23 -0800260 s->s_time_min = TIME64_MIN;
261 s->s_time_max = TIME64_MAX;
Al Viro7eb5e882013-10-01 15:09:58 -0400262
263 s->s_shrink.seeks = DEFAULT_SEEKS;
264 s->s_shrink.scan_objects = super_cache_scan;
265 s->s_shrink.count_objects = super_cache_count;
266 s->s_shrink.batch = 1024;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800267 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
Roman Gushchine33c2672022-05-31 20:22:24 -0700268 if (prealloc_shrinker(&s->s_shrink, "sb-%s", type->name))
Tetsuo Handa8e049442018-04-04 19:53:07 +0900269 goto fail;
Kirill Tkhaic92e8e12018-08-17 15:47:50 -0700270 if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
Kirill Tkhai2b3648a2018-08-17 15:47:45 -0700271 goto fail;
Kirill Tkhaic92e8e12018-08-17 15:47:50 -0700272 if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
Kirill Tkhai2b3648a2018-08-17 15:47:45 -0700273 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000275
Al Viro7eb5e882013-10-01 15:09:58 -0400276fail:
Al Viro02008942017-10-11 13:48:55 -0400277 destroy_unused_super(s);
Al Viro7eb5e882013-10-01 15:09:58 -0400278 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279}
280
281/* Superblock refcounting */
282
283/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400284 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 */
Al Viroc645b932017-10-11 13:48:55 -0400286static void __put_super(struct super_block *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287{
Al Viroc645b932017-10-11 13:48:55 -0400288 if (!--s->s_count) {
289 list_del_init(&s->s_list);
290 WARN_ON(s->s_dentry_lru.node);
291 WARN_ON(s->s_inode_lru.node);
292 WARN_ON(!list_empty(&s->s_mounts));
293 security_sb_free(s);
Eric Biggers22d94f42019-08-04 19:35:46 -0700294 fscrypt_sb_free(s);
Al Viroc645b932017-10-11 13:48:55 -0400295 put_user_ns(s->s_user_ns);
296 kfree(s->s_subtype);
297 call_rcu(&s->rcu, destroy_super_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299}
300
301/**
302 * put_super - drop a temporary reference to superblock
303 * @sb: superblock in question
304 *
305 * Drops a temporary reference, frees superblock if there's no
306 * references left.
307 */
Christoph Hellwig60b49882020-11-16 15:21:18 +0100308void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309{
310 spin_lock(&sb_lock);
311 __put_super(sb);
312 spin_unlock(&sb_lock);
313}
314
315
316/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400317 * deactivate_locked_super - drop an active reference to superblock
318 * @s: superblock to deactivate
319 *
Masanari Iidabd7ced92016-02-02 22:31:06 +0900320 * Drops an active reference to superblock, converting it into a temporary
Al Viro1712ac82010-03-22 15:22:31 -0400321 * one if there is no other active references left. In that case we
322 * tell fs driver to shut it down and drop the temporary reference we
323 * had just acquired.
324 *
325 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400326 */
327void deactivate_locked_super(struct super_block *s)
328{
329 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400330 if (atomic_dec_and_test(&s->s_active)) {
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000331 unregister_shrinker(&s->s_shrink);
Dave Chinner28f2cd4f2014-06-04 16:10:46 -0700332 fs->kill_sb(s);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000333
Vladimir Davydovc0a5b562015-02-12 14:59:07 -0800334 /*
335 * Since list_lru_destroy() may sleep, we cannot call it from
336 * put_super(), where we hold the sb_lock. Therefore we destroy
337 * the lru lists right now.
338 */
339 list_lru_destroy(&s->s_dentry_lru);
340 list_lru_destroy(&s->s_inode_lru);
341
Al Viro74dbbdd2009-05-06 01:07:50 -0400342 put_filesystem(fs);
343 put_super(s);
344 } else {
345 up_write(&s->s_umount);
346 }
347}
348
349EXPORT_SYMBOL(deactivate_locked_super);
350
351/**
Al Viro1712ac82010-03-22 15:22:31 -0400352 * deactivate_super - drop an active reference to superblock
353 * @s: superblock to deactivate
354 *
355 * Variant of deactivate_locked_super(), except that superblock is *not*
356 * locked by caller. If we are going to drop the final active reference,
357 * lock will be acquired prior to that.
358 */
359void deactivate_super(struct super_block *s)
360{
Yufen Yucc234022020-03-05 16:06:39 +0800361 if (!atomic_add_unless(&s->s_active, -1, 1)) {
Al Viro1712ac82010-03-22 15:22:31 -0400362 down_write(&s->s_umount);
363 deactivate_locked_super(s);
364 }
365}
366
367EXPORT_SYMBOL(deactivate_super);
368
369/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 * grab_super - acquire an active reference
371 * @s: reference we are trying to make active
372 *
373 * Tries to acquire an active reference. grab_super() is used when we
374 * had just found a superblock in super_blocks or fs_type->fs_supers
375 * and want to turn it into a full-blown active reference. grab_super()
376 * is called with sb_lock held and drops it. Returns 1 in case of
377 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400378 * dying when grab_super() had been called). Note that this is only
379 * called for superblocks not in rundown mode (== ones still on ->fs_supers
380 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700382static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383{
384 s->s_count++;
385 spin_unlock(&sb_lock);
386 down_write(&s->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100387 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
Al Viroacfec9a2013-07-20 03:13:55 +0400388 put_super(s);
389 return 1;
390 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 up_write(&s->s_umount);
392 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 return 0;
394}
395
David Howellscf9a2ae2006-08-29 19:05:54 +0100396/*
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300397 * trylock_super - try to grab ->s_umount shared
Wanpeng Li331cbde2012-06-09 11:10:55 +0800398 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000399 *
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300400 * Try to prevent fs shutdown. This is used in places where we
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000401 * cannot take an active reference but we need to ensure that the
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300402 * filesystem is not shut down while we are working on it. It returns
403 * false if we cannot acquire s_umount or if we lose the race and
404 * filesystem already got into shutdown, and returns true with the s_umount
405 * lock held in read mode in case of success. On successful return,
406 * the caller must drop the s_umount lock when done.
407 *
408 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
409 * The reason why it's safe is that we are OK with doing trylock instead
410 * of down_read(). There's a couple of places that are OK with that, but
411 * it's very much not a general-purpose interface.
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000412 */
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300413bool trylock_super(struct super_block *sb)
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000414{
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000415 if (down_read_trylock(&sb->s_umount)) {
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300416 if (!hlist_unhashed(&sb->s_instances) &&
David Howellse462ec52017-07-17 08:45:35 +0100417 sb->s_root && (sb->s_flags & SB_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000418 return true;
419 up_read(&sb->s_umount);
420 }
421
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000422 return false;
423}
424
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425/**
Daniil Lunev04b94072022-07-27 16:44:24 +1000426 * retire_super - prevents superblock from being reused
427 * @sb: superblock to retire
428 *
429 * The function marks superblock to be ignored in superblock test, which
430 * prevents it from being reused for any new mounts. If the superblock has
431 * a private bdi, it also unregisters it, but doesn't reduce the refcount
432 * of the superblock to prevent potential races. The refcount is reduced
433 * by generic_shutdown_super(). The function can not be called
434 * concurrently with generic_shutdown_super(). It is safe to call the
435 * function multiple times, subsequent calls have no effect.
436 *
437 * The marker will affect the re-use only for block-device-based
438 * superblocks. Other superblocks will still get marked if this function
439 * is used, but that will not affect their reusability.
440 */
441void retire_super(struct super_block *sb)
442{
443 WARN_ON(!sb->s_bdev);
444 down_write(&sb->s_umount);
445 if (sb->s_iflags & SB_I_PERSB_BDI) {
446 bdi_unregister(sb->s_bdi);
447 sb->s_iflags &= ~SB_I_PERSB_BDI;
448 }
449 sb->s_iflags |= SB_I_RETIRED;
450 up_write(&sb->s_umount);
451}
452EXPORT_SYMBOL(retire_super);
453
454/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 * generic_shutdown_super - common helper for ->kill_sb()
456 * @sb: superblock to kill
457 *
458 * generic_shutdown_super() does all fs-independent work on superblock
459 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
460 * that need destruction out of superblock, call generic_shutdown_super()
461 * and release aforementioned objects. Note: dentries and inodes _are_
462 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700463 *
464 * Upon calling this function, the filesystem may no longer alter or
465 * rearrange the set of dentries belonging to this super_block, nor may it
466 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 */
468void generic_shutdown_super(struct super_block *sb)
469{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800470 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
David Howellsc636ebd2006-10-11 01:22:19 -0700472 if (sb->s_root) {
473 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200474 sync_filesystem(sb);
David Howellse462ec52017-07-17 08:45:35 +0100475 sb->s_flags &= ~SB_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800476
Tejun Heoa1a0e232016-02-29 18:28:53 -0500477 cgroup_writeback_umount();
Al Viro63997e92010-10-25 20:49:35 -0400478
Eric Sandeen1edc8eb22019-12-06 10:55:59 -0600479 /* evict all inodes with zero refcount */
Al Viro63997e92010-10-25 20:49:35 -0400480 evict_inodes(sb);
Eric Sandeen1edc8eb22019-12-06 10:55:59 -0600481 /* only nonzero refcount inodes can have marks */
482 fsnotify_sb_delete(sb);
Mickaël Salaün83e804f2021-04-22 17:41:16 +0200483 security_sb_delete(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200485 if (sb->s_dio_done_wq) {
486 destroy_workqueue(sb->s_dio_done_wq);
487 sb->s_dio_done_wq = NULL;
488 }
489
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 if (sop->put_super)
491 sop->put_super(sb);
492
Al Viro63997e92010-10-25 20:49:35 -0400493 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800494 printk("VFS: Busy inodes after unmount of %s. "
495 "Self-destruct in 5 seconds. Have a nice day...\n",
496 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 }
499 spin_lock(&sb_lock);
500 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500501 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 spin_unlock(&sb_lock);
503 up_write(&sb->s_umount);
Jan Karac1844d52017-04-12 12:24:47 +0200504 if (sb->s_bdi != &noop_backing_dev_info) {
Christoph Hellwig0b3ea092021-11-05 13:36:58 -0700505 if (sb->s_iflags & SB_I_PERSB_BDI)
506 bdi_unregister(sb->s_bdi);
Jan Karafca39342017-04-12 12:24:28 +0200507 bdi_put(sb->s_bdi);
508 sb->s_bdi = &noop_backing_dev_info;
Jan Karafca39342017-04-12 12:24:28 +0200509 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510}
511
512EXPORT_SYMBOL(generic_shutdown_super);
513
Al Viro20284ab2019-05-12 17:31:45 -0400514bool mount_capable(struct fs_context *fc)
Al Viro0ce0cf12019-05-12 15:42:48 -0400515{
Al Viro20284ab2019-05-12 17:31:45 -0400516 if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
Al Viro0ce0cf12019-05-12 15:42:48 -0400517 return capable(CAP_SYS_ADMIN);
518 else
Al Viroc2c44ec2019-07-27 08:29:57 -0400519 return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
Al Viro0ce0cf12019-05-12 15:42:48 -0400520}
521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522/**
Al Virocb50b342018-12-23 17:25:47 -0500523 * sget_fc - Find or create a superblock
524 * @fc: Filesystem context.
525 * @test: Comparison callback
526 * @set: Setup callback
527 *
528 * Find or create a superblock using the parameters stored in the filesystem
529 * context and the two callback functions.
530 *
531 * If an extant superblock is matched, then that will be returned with an
532 * elevated reference count that the caller must transfer or discard.
533 *
534 * If no match is made, a new superblock will be allocated and basic
535 * initialisation will be performed (s_type, s_fs_info and s_id will be set and
536 * the set() callback will be invoked), the superblock will be published and it
537 * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
538 * as yet unset.
539 */
540struct super_block *sget_fc(struct fs_context *fc,
541 int (*test)(struct super_block *, struct fs_context *),
542 int (*set)(struct super_block *, struct fs_context *))
543{
544 struct super_block *s = NULL;
545 struct super_block *old;
546 struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
547 int err;
548
Al Virocb50b342018-12-23 17:25:47 -0500549retry:
550 spin_lock(&sb_lock);
551 if (test) {
552 hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
553 if (test(old, fc))
554 goto share_extant_sb;
555 }
556 }
557 if (!s) {
558 spin_unlock(&sb_lock);
559 s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
560 if (!s)
561 return ERR_PTR(-ENOMEM);
562 goto retry;
563 }
564
565 s->s_fs_info = fc->s_fs_info;
566 err = set(s, fc);
567 if (err) {
568 s->s_fs_info = NULL;
569 spin_unlock(&sb_lock);
570 destroy_unused_super(s);
571 return ERR_PTR(err);
572 }
573 fc->s_fs_info = NULL;
574 s->s_type = fc->fs_type;
David Howellsc80fa7c2019-03-25 16:38:23 +0000575 s->s_iflags |= fc->s_iflags;
Al Virocb50b342018-12-23 17:25:47 -0500576 strlcpy(s->s_id, s->s_type->name, sizeof(s->s_id));
577 list_add_tail(&s->s_list, &super_blocks);
578 hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
579 spin_unlock(&sb_lock);
580 get_filesystem(s->s_type);
581 register_shrinker_prepared(&s->s_shrink);
582 return s;
583
584share_extant_sb:
585 if (user_ns != old->s_user_ns) {
586 spin_unlock(&sb_lock);
587 destroy_unused_super(s);
588 return ERR_PTR(-EBUSY);
589 }
590 if (!grab_super(old))
591 goto retry;
592 destroy_unused_super(s);
593 return old;
594}
595EXPORT_SYMBOL(sget_fc);
596
597/**
David Howells023d0662019-03-25 16:38:28 +0000598 * sget - find or create a superblock
599 * @type: filesystem type superblock should belong to
600 * @test: comparison callback
601 * @set: setup callback
602 * @flags: mount flags
603 * @data: argument to each of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 */
David Howells023d0662019-03-25 16:38:28 +0000605struct super_block *sget(struct file_system_type *type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 int (*test)(struct super_block *,void *),
607 int (*set)(struct super_block *,void *),
David Howells023d0662019-03-25 16:38:28 +0000608 int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 void *data)
610{
David Howells023d0662019-03-25 16:38:28 +0000611 struct user_namespace *user_ns = current_user_ns();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700613 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 int err;
615
David Howells023d0662019-03-25 16:38:28 +0000616 /* We don't yet pass the user namespace of the parent
617 * mount through to here so always use &init_user_ns
618 * until that changes.
619 */
620 if (flags & SB_SUBMOUNT)
621 user_ns = &init_user_ns;
622
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623retry:
624 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700625 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800626 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700627 if (!test(old, data))
628 continue;
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500629 if (user_ns != old->s_user_ns) {
630 spin_unlock(&sb_lock);
Al Viro02008942017-10-11 13:48:55 -0400631 destroy_unused_super(s);
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500632 return ERR_PTR(-EBUSY);
633 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700634 if (!grab_super(old))
635 goto retry;
Al Viro02008942017-10-11 13:48:55 -0400636 destroy_unused_super(s);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700637 return old;
638 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 }
640 if (!s) {
641 spin_unlock(&sb_lock);
David Howellse462ec52017-07-17 08:45:35 +0100642 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 if (!s)
644 return ERR_PTR(-ENOMEM);
645 goto retry;
646 }
David Howellsdd111b32017-07-04 17:25:09 +0100647
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 err = set(s, data);
649 if (err) {
650 spin_unlock(&sb_lock);
Al Viro02008942017-10-11 13:48:55 -0400651 destroy_unused_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 return ERR_PTR(err);
653 }
654 s->s_type = type;
655 strlcpy(s->s_id, type->name, sizeof(s->s_id));
656 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500657 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 spin_unlock(&sb_lock);
659 get_filesystem(type);
Tetsuo Handa8e049442018-04-04 19:53:07 +0900660 register_shrinker_prepared(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 return s;
662}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663EXPORT_SYMBOL(sget);
664
665void drop_super(struct super_block *sb)
666{
667 up_read(&sb->s_umount);
668 put_super(sb);
669}
670
671EXPORT_SYMBOL(drop_super);
672
Jan Karaba6379f2016-11-23 12:53:00 +0100673void drop_super_exclusive(struct super_block *sb)
674{
675 up_write(&sb->s_umount);
676 put_super(sb);
677}
678EXPORT_SYMBOL(drop_super_exclusive);
679
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +0200680static void __iterate_supers(void (*f)(struct super_block *))
681{
682 struct super_block *sb, *p = NULL;
683
684 spin_lock(&sb_lock);
685 list_for_each_entry(sb, &super_blocks, s_list) {
686 if (hlist_unhashed(&sb->s_instances))
687 continue;
688 sb->s_count++;
689 spin_unlock(&sb_lock);
690
691 f(sb);
692
693 spin_lock(&sb_lock);
694 if (p)
695 __put_super(p);
696 p = sb;
697 }
698 if (p)
699 __put_super(p);
700 spin_unlock(&sb_lock);
701}
Christoph Hellwige5004752009-05-05 16:08:56 +0200702/**
Al Viro01a05b32010-03-23 06:06:58 -0400703 * iterate_supers - call function for all active superblocks
704 * @f: function to call
705 * @arg: argument to pass to it
706 *
707 * Scans the superblock list and calls given function, passing it
708 * locked superblock and given argument.
709 */
710void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
711{
Al Virodca33252010-07-25 02:31:46 +0400712 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400713
714 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400715 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500716 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400717 continue;
718 sb->s_count++;
719 spin_unlock(&sb_lock);
720
721 down_read(&sb->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100722 if (sb->s_root && (sb->s_flags & SB_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400723 f(sb, arg);
724 up_read(&sb->s_umount);
725
726 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400727 if (p)
728 __put_super(p);
729 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400730 }
Al Virodca33252010-07-25 02:31:46 +0400731 if (p)
732 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400733 spin_unlock(&sb_lock);
734}
735
736/**
Al Viro43e15cd2011-06-03 20:16:57 -0400737 * iterate_supers_type - call function for superblocks of given type
738 * @type: fs type
739 * @f: function to call
740 * @arg: argument to pass to it
741 *
742 * Scans the superblock list and calls given function, passing it
743 * locked superblock and given argument.
744 */
745void iterate_supers_type(struct file_system_type *type,
746 void (*f)(struct super_block *, void *), void *arg)
747{
748 struct super_block *sb, *p = NULL;
749
750 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800751 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400752 sb->s_count++;
753 spin_unlock(&sb_lock);
754
755 down_read(&sb->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100756 if (sb->s_root && (sb->s_flags & SB_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400757 f(sb, arg);
758 up_read(&sb->s_umount);
759
760 spin_lock(&sb_lock);
761 if (p)
762 __put_super(p);
763 p = sb;
764 }
765 if (p)
766 __put_super(p);
767 spin_unlock(&sb_lock);
768}
769
770EXPORT_SYMBOL(iterate_supers_type);
771
Christoph Hellwig4e7b5672020-11-23 13:38:40 +0100772/**
773 * get_super - get the superblock of a device
774 * @bdev: device to get the superblock for
775 *
776 * Scans the superblock list and finds the superblock of the file system
777 * mounted on the device given. %NULL is returned if no match is found.
778 */
779struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700781 struct super_block *sb;
782
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 if (!bdev)
784 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700785
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700787rescan:
788 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500789 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400790 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700791 if (sb->s_bdev == bdev) {
792 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 spin_unlock(&sb_lock);
Christoph Hellwig4e7b5672020-11-23 13:38:40 +0100794 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400795 /* still alive? */
David Howellse462ec52017-07-17 08:45:35 +0100796 if (sb->s_root && (sb->s_flags & SB_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700797 return sb;
Christoph Hellwig4e7b5672020-11-23 13:38:40 +0100798 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400799 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700800 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400801 __put_super(sb);
802 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 }
804 }
805 spin_unlock(&sb_lock);
806 return NULL;
807}
808
Jan Karaba6379f2016-11-23 12:53:00 +0100809/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200810 * get_active_super - get an active reference to the superblock of a device
811 * @bdev: device to get the superblock for
812 *
813 * Scans the superblock list and finds the superblock of the file system
814 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400815 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200816 */
817struct super_block *get_active_super(struct block_device *bdev)
818{
819 struct super_block *sb;
820
821 if (!bdev)
822 return NULL;
823
Al Viro14945832010-03-22 20:15:33 -0400824restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200825 spin_lock(&sb_lock);
826 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500827 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400828 continue;
Al Viro14945832010-03-22 20:15:33 -0400829 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400830 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400831 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400832 up_write(&sb->s_umount);
833 return sb;
Al Viro14945832010-03-22 20:15:33 -0400834 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200835 }
836 spin_unlock(&sb_lock);
837 return NULL;
838}
David Howellsdd111b32017-07-04 17:25:09 +0100839
Christoph Hellwig4e7b5672020-11-23 13:38:40 +0100840struct super_block *user_get_super(dev_t dev, bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700842 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700845rescan:
846 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500847 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400848 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700849 if (sb->s_dev == dev) {
850 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 spin_unlock(&sb_lock);
Christoph Hellwig4e7b5672020-11-23 13:38:40 +0100852 if (excl)
853 down_write(&sb->s_umount);
854 else
855 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400856 /* still alive? */
David Howellse462ec52017-07-17 08:45:35 +0100857 if (sb->s_root && (sb->s_flags & SB_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700858 return sb;
Christoph Hellwig4e7b5672020-11-23 13:38:40 +0100859 if (excl)
860 up_write(&sb->s_umount);
861 else
862 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400863 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700864 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400865 __put_super(sb);
866 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 }
868 }
869 spin_unlock(&sb_lock);
870 return NULL;
871}
872
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873/**
David Howells8d0347f2018-11-04 09:28:36 -0500874 * reconfigure_super - asks filesystem to change superblock parameters
875 * @fc: The superblock and configuration
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 *
David Howells8d0347f2018-11-04 09:28:36 -0500877 * Alters the configuration parameters of a live superblock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 */
David Howells8d0347f2018-11-04 09:28:36 -0500879int reconfigure_super(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880{
David Howells8d0347f2018-11-04 09:28:36 -0500881 struct super_block *sb = fc->root->d_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 int retval;
David Howells8d0347f2018-11-04 09:28:36 -0500883 bool remount_ro = false;
884 bool force = fc->sb_flags & SB_FORCE;
Christoph Hellwig45042302009-08-03 23:28:35 +0200885
David Howells8d0347f2018-11-04 09:28:36 -0500886 if (fc->sb_flags_mask & ~MS_RMT_MASK)
887 return -EINVAL;
Jan Kara5accdf82012-06-12 16:20:34 +0200888 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200889 return -EBUSY;
890
David Howells8d0347f2018-11-04 09:28:36 -0500891 retval = security_sb_remount(sb, fc->security);
892 if (retval)
893 return retval;
894
895 if (fc->sb_flags_mask & SB_RDONLY) {
David Howells93614012006-09-30 20:45:40 +0200896#ifdef CONFIG_BLOCK
Christoph Hellwig6f0d9682021-01-09 11:42:50 +0100897 if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
898 bdev_read_only(sb->s_bdev))
David Howells8d0347f2018-11-04 09:28:36 -0500899 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200900#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200901
David Howells8d0347f2018-11-04 09:28:36 -0500902 remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
903 }
Nick Piggind208bbd2009-12-21 16:28:53 -0800904
Al Viro0aec09d2014-08-07 07:32:06 -0400905 if (remount_ro) {
Al Virofdab6842015-01-11 10:57:27 -0500906 if (!hlist_empty(&sb->s_pins)) {
Al Viro0aec09d2014-08-07 07:32:06 -0400907 up_write(&sb->s_umount);
Al Virofdab6842015-01-11 10:57:27 -0500908 group_pin_kill(&sb->s_pins);
Al Viro0aec09d2014-08-07 07:32:06 -0400909 down_write(&sb->s_umount);
910 if (!sb->s_root)
911 return 0;
912 if (sb->s_writers.frozen != SB_UNFROZEN)
913 return -EBUSY;
David Howells8d0347f2018-11-04 09:28:36 -0500914 remount_ro = !sb_rdonly(sb);
Al Viro0aec09d2014-08-07 07:32:06 -0400915 }
916 }
917 shrink_dcache_sb(sb);
918
David Howells8d0347f2018-11-04 09:28:36 -0500919 /* If we are reconfiguring to RDONLY and current sb is read/write,
920 * make sure there are no files open for writing.
921 */
Nick Piggind208bbd2009-12-21 16:28:53 -0800922 if (remount_ro) {
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100923 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400924 sb->s_readonly_remount = 1;
925 smp_wmb();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100926 } else {
927 retval = sb_prepare_remount_readonly(sb);
928 if (retval)
929 return retval;
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100930 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 }
932
Al Virof3a09c92018-12-23 18:55:56 -0500933 if (fc->ops->reconfigure) {
934 retval = fc->ops->reconfigure(fc);
935 if (retval) {
936 if (!force)
937 goto cancel_readonly;
938 /* If forced remount, go ahead despite any errors */
939 WARN(1, "forced remount of a %s fs returned %i\n",
940 sb->s_type->name, retval);
941 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 }
David Howells8d0347f2018-11-04 09:28:36 -0500943
944 WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
945 (fc->sb_flags & fc->sb_flags_mask)));
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100946 /* Needs to be ordered wrt mnt_is_readonly() */
947 smp_wmb();
948 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400949
Nick Piggind208bbd2009-12-21 16:28:53 -0800950 /*
951 * Some filesystems modify their metadata via some other path than the
952 * bdev buffer cache (eg. use a private mapping, or directories in
953 * pagecache, etc). Also file data modifications go via their own
954 * mappings. So If we try to mount readonly then copy the filesystem
955 * from bdev, we could get stale data, so invalidate it to give a best
956 * effort at coherency.
957 */
958 if (remount_ro && sb->s_bdev)
959 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 return 0;
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100961
962cancel_readonly:
963 sb->s_readonly_remount = 0;
964 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965}
966
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +0200967static void do_emergency_remount_callback(struct super_block *sb)
968{
969 down_write(&sb->s_umount);
970 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
971 !sb_rdonly(sb)) {
David Howells8d0347f2018-11-04 09:28:36 -0500972 struct fs_context *fc;
973
974 fc = fs_context_for_reconfigure(sb->s_root,
975 SB_RDONLY | SB_FORCE, SB_RDONLY);
976 if (!IS_ERR(fc)) {
977 if (parse_monolithic_mount_data(fc, NULL) == 0)
978 (void)reconfigure_super(fc);
979 put_fs_context(fc);
980 }
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +0200981 }
982 up_write(&sb->s_umount);
983}
984
Jens Axboea2a95372009-03-17 09:38:40 +0100985static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986{
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +0200987 __iterate_supers(do_emergency_remount_callback);
Jens Axboea2a95372009-03-17 09:38:40 +0100988 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 printk("Emergency Remount complete\n");
990}
991
992void emergency_remount(void)
993{
Jens Axboea2a95372009-03-17 09:38:40 +0100994 struct work_struct *work;
995
996 work = kmalloc(sizeof(*work), GFP_ATOMIC);
997 if (work) {
998 INIT_WORK(work, do_emergency_remount);
999 schedule_work(work);
1000 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001}
1002
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001003static void do_thaw_all_callback(struct super_block *sb)
1004{
1005 down_write(&sb->s_umount);
Al Viro1c18d2a2018-05-10 15:09:41 -04001006 if (sb->s_root && sb->s_flags & SB_BORN) {
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001007 emergency_thaw_bdev(sb);
1008 thaw_super_locked(sb);
1009 } else {
1010 up_write(&sb->s_umount);
1011 }
1012}
1013
1014static void do_thaw_all(struct work_struct *work)
1015{
1016 __iterate_supers(do_thaw_all_callback);
1017 kfree(work);
1018 printk(KERN_WARNING "Emergency Thaw complete\n");
1019}
1020
1021/**
1022 * emergency_thaw_all -- forcibly thaw every frozen filesystem
1023 *
1024 * Used for emergency unfreeze of all filesystems via SysRq
1025 */
1026void emergency_thaw_all(void)
1027{
1028 struct work_struct *work;
1029
1030 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1031 if (work) {
1032 INIT_WORK(work, do_thaw_all);
1033 schedule_work(work);
1034 }
1035}
1036
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +04001037static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
Matthew Wilcox5a668472018-06-11 12:03:31 -04001039/**
1040 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1041 * @p: Pointer to a dev_t.
1042 *
1043 * Filesystems which don't use real block devices can call this function
1044 * to allocate a virtual block device.
1045 *
1046 * Context: Any context. Frequently called while holding sb_lock.
1047 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1048 * or -ENOMEM if memory allocation failed.
1049 */
Al Viro0ee5dc62011-07-07 15:44:25 -04001050int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051{
1052 int dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Matthew Wilcox5a668472018-06-11 12:03:31 -04001054 /*
1055 * Many userspace utilities consider an FSID of 0 invalid.
1056 * Always return at least 1 from get_anon_bdev.
1057 */
1058 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1059 GFP_ATOMIC);
1060 if (dev == -ENOSPC)
1061 dev = -EMFILE;
1062 if (dev < 0)
1063 return dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
Matthew Wilcox5a668472018-06-11 12:03:31 -04001065 *p = MKDEV(0, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 return 0;
1067}
Al Viro0ee5dc62011-07-07 15:44:25 -04001068EXPORT_SYMBOL(get_anon_bdev);
1069
1070void free_anon_bdev(dev_t dev)
1071{
Matthew Wilcox5a668472018-06-11 12:03:31 -04001072 ida_free(&unnamed_dev_ida, MINOR(dev));
Al Viro0ee5dc62011-07-07 15:44:25 -04001073}
1074EXPORT_SYMBOL(free_anon_bdev);
1075
1076int set_anon_super(struct super_block *s, void *data)
1077{
Christoph Hellwigdf0ce262015-01-14 10:42:41 +01001078 return get_anon_bdev(&s->s_dev);
Al Viro0ee5dc62011-07-07 15:44:25 -04001079}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080EXPORT_SYMBOL(set_anon_super);
1081
1082void kill_anon_super(struct super_block *sb)
1083{
Al Viro0ee5dc62011-07-07 15:44:25 -04001084 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -04001086 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088EXPORT_SYMBOL(kill_anon_super);
1089
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090void kill_litter_super(struct super_block *sb)
1091{
1092 if (sb->s_root)
1093 d_genocide(sb->s_root);
1094 kill_anon_super(sb);
1095}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096EXPORT_SYMBOL(kill_litter_super);
1097
Al Virocb50b342018-12-23 17:25:47 -05001098int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1099{
1100 return set_anon_super(sb, NULL);
1101}
1102EXPORT_SYMBOL(set_anon_super_fc);
1103
1104static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1105{
1106 return sb->s_fs_info == fc->s_fs_info;
1107}
1108
1109static int test_single_super(struct super_block *s, struct fs_context *fc)
1110{
1111 return 1;
1112}
1113
1114/**
1115 * vfs_get_super - Get a superblock with a search key set in s_fs_info.
1116 * @fc: The filesystem context holding the parameters
1117 * @keying: How to distinguish superblocks
1118 * @fill_super: Helper to initialise a new superblock
1119 *
1120 * Search for a superblock and create a new one if not found. The search
1121 * criterion is controlled by @keying. If the search fails, a new superblock
1122 * is created and @fill_super() is called to initialise it.
1123 *
1124 * @keying can take one of a number of values:
1125 *
1126 * (1) vfs_get_single_super - Only one superblock of this type may exist on the
1127 * system. This is typically used for special system filesystems.
1128 *
1129 * (2) vfs_get_keyed_super - Multiple superblocks may exist, but they must have
1130 * distinct keys (where the key is in s_fs_info). Searching for the same
1131 * key again will turn up the superblock for that key.
1132 *
1133 * (3) vfs_get_independent_super - Multiple superblocks may exist and are
1134 * unkeyed. Each call will get a new superblock.
1135 *
1136 * A permissions check is made by sget_fc() unless we're getting a superblock
1137 * for a kernel-internal mount or a submount.
1138 */
1139int vfs_get_super(struct fs_context *fc,
1140 enum vfs_get_super_keying keying,
1141 int (*fill_super)(struct super_block *sb,
1142 struct fs_context *fc))
1143{
1144 int (*test)(struct super_block *, struct fs_context *);
1145 struct super_block *sb;
David Howells43ce4c12019-03-21 09:22:36 +00001146 int err;
Al Virocb50b342018-12-23 17:25:47 -05001147
1148 switch (keying) {
1149 case vfs_get_single_super:
David Howells43ce4c12019-03-21 09:22:36 +00001150 case vfs_get_single_reconf_super:
Al Virocb50b342018-12-23 17:25:47 -05001151 test = test_single_super;
1152 break;
1153 case vfs_get_keyed_super:
1154 test = test_keyed_super;
1155 break;
1156 case vfs_get_independent_super:
1157 test = NULL;
1158 break;
1159 default:
1160 BUG();
1161 }
1162
1163 sb = sget_fc(fc, test, set_anon_super_fc);
1164 if (IS_ERR(sb))
1165 return PTR_ERR(sb);
1166
1167 if (!sb->s_root) {
David Howells43ce4c12019-03-21 09:22:36 +00001168 err = fill_super(sb, fc);
1169 if (err)
1170 goto error;
Al Virocb50b342018-12-23 17:25:47 -05001171
1172 sb->s_flags |= SB_ACTIVE;
David Howells43ce4c12019-03-21 09:22:36 +00001173 fc->root = dget(sb->s_root);
1174 } else {
1175 fc->root = dget(sb->s_root);
1176 if (keying == vfs_get_single_reconf_super) {
1177 err = reconfigure_super(fc);
1178 if (err < 0) {
1179 dput(fc->root);
1180 fc->root = NULL;
1181 goto error;
1182 }
1183 }
Al Virocb50b342018-12-23 17:25:47 -05001184 }
1185
Al Virocb50b342018-12-23 17:25:47 -05001186 return 0;
David Howells43ce4c12019-03-21 09:22:36 +00001187
1188error:
1189 deactivate_locked_super(sb);
1190 return err;
Al Virocb50b342018-12-23 17:25:47 -05001191}
1192EXPORT_SYMBOL(vfs_get_super);
1193
Al Viro2ac295d2019-06-01 20:48:55 -04001194int get_tree_nodev(struct fs_context *fc,
1195 int (*fill_super)(struct super_block *sb,
1196 struct fs_context *fc))
1197{
1198 return vfs_get_super(fc, vfs_get_independent_super, fill_super);
1199}
1200EXPORT_SYMBOL(get_tree_nodev);
1201
Al Viroc23a0bb2019-05-22 21:23:39 -04001202int get_tree_single(struct fs_context *fc,
1203 int (*fill_super)(struct super_block *sb,
1204 struct fs_context *fc))
1205{
1206 return vfs_get_super(fc, vfs_get_single_super, fill_super);
1207}
1208EXPORT_SYMBOL(get_tree_single);
1209
David Howells43ce4c12019-03-21 09:22:36 +00001210int get_tree_single_reconf(struct fs_context *fc,
1211 int (*fill_super)(struct super_block *sb,
1212 struct fs_context *fc))
1213{
1214 return vfs_get_super(fc, vfs_get_single_reconf_super, fill_super);
1215}
1216EXPORT_SYMBOL(get_tree_single_reconf);
1217
Al Viro533770c2019-09-03 19:05:48 -04001218int get_tree_keyed(struct fs_context *fc,
1219 int (*fill_super)(struct super_block *sb,
1220 struct fs_context *fc),
1221 void *key)
1222{
1223 fc->s_fs_info = key;
1224 return vfs_get_super(fc, vfs_get_keyed_super, fill_super);
1225}
1226EXPORT_SYMBOL(get_tree_keyed);
1227
David Howells93614012006-09-30 20:45:40 +02001228#ifdef CONFIG_BLOCK
David Howellsfe62c3a2019-03-27 14:15:16 +00001229
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230static int set_bdev_super(struct super_block *s, void *data)
1231{
1232 s->s_bdev = data;
1233 s->s_dev = s->s_bdev->bd_dev;
Christoph Hellwiga11d7fc2021-08-09 16:17:44 +02001234 s->s_bdi = bdi_get(s->s_bdev->bd_disk->bdi);
Jens Axboe32a88aa2009-09-16 15:02:33 +02001235
Christoph Hellwig36d25482022-04-15 06:52:45 +02001236 if (bdev_stable_writes(s->s_bdev))
Christoph Hellwig1cb039f2020-09-24 08:51:38 +02001237 s->s_iflags |= SB_I_STABLE_WRITES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 return 0;
1239}
1240
David Howellsfe62c3a2019-03-27 14:15:16 +00001241static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1242{
1243 return set_bdev_super(s, fc->sget_key);
1244}
1245
1246static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1247{
Daniil Lunev04b94072022-07-27 16:44:24 +10001248 return !(s->s_iflags & SB_I_RETIRED) && s->s_bdev == fc->sget_key;
David Howellsfe62c3a2019-03-27 14:15:16 +00001249}
1250
1251/**
1252 * get_tree_bdev - Get a superblock based on a single block device
1253 * @fc: The filesystem context holding the parameters
1254 * @fill_super: Helper to initialise a new superblock
1255 */
1256int get_tree_bdev(struct fs_context *fc,
1257 int (*fill_super)(struct super_block *,
1258 struct fs_context *))
1259{
1260 struct block_device *bdev;
1261 struct super_block *s;
1262 fmode_t mode = FMODE_READ | FMODE_EXCL;
1263 int error = 0;
1264
1265 if (!(fc->sb_flags & SB_RDONLY))
1266 mode |= FMODE_WRITE;
1267
1268 if (!fc->source)
1269 return invalf(fc, "No source specified");
1270
1271 bdev = blkdev_get_by_path(fc->source, mode, fc->fs_type);
1272 if (IS_ERR(bdev)) {
1273 errorf(fc, "%s: Can't open blockdev", fc->source);
1274 return PTR_ERR(bdev);
1275 }
1276
1277 /* Once the superblock is inserted into the list by sget_fc(), s_umount
1278 * will protect the lockfs code from trying to start a snapshot while
1279 * we are mounting
1280 */
1281 mutex_lock(&bdev->bd_fsfreeze_mutex);
1282 if (bdev->bd_fsfreeze_count > 0) {
1283 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1284 warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
David Howellsdd7bc812020-04-28 21:27:48 +01001285 blkdev_put(bdev, mode);
David Howellsfe62c3a2019-03-27 14:15:16 +00001286 return -EBUSY;
1287 }
1288
1289 fc->sb_flags |= SB_NOSEC;
1290 fc->sget_key = bdev;
1291 s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc);
1292 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Ian Kent6fcf0c72019-10-02 17:56:33 +08001293 if (IS_ERR(s)) {
1294 blkdev_put(bdev, mode);
David Howellsfe62c3a2019-03-27 14:15:16 +00001295 return PTR_ERR(s);
Ian Kent6fcf0c72019-10-02 17:56:33 +08001296 }
David Howellsfe62c3a2019-03-27 14:15:16 +00001297
1298 if (s->s_root) {
1299 /* Don't summarily change the RO/RW state. */
1300 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1301 warnf(fc, "%pg: Can't mount, would change RO state", bdev);
1302 deactivate_locked_super(s);
1303 blkdev_put(bdev, mode);
1304 return -EBUSY;
1305 }
1306
1307 /*
Christoph Hellwiga8698702021-05-25 08:12:56 +02001308 * s_umount nests inside open_mutex during
David Howellsfe62c3a2019-03-27 14:15:16 +00001309 * __invalidate_device(). blkdev_put() acquires
Christoph Hellwiga8698702021-05-25 08:12:56 +02001310 * open_mutex and can't be called under s_umount. Drop
David Howellsfe62c3a2019-03-27 14:15:16 +00001311 * s_umount temporarily. This is safe as we're
1312 * holding an active reference.
1313 */
1314 up_write(&s->s_umount);
1315 blkdev_put(bdev, mode);
1316 down_write(&s->s_umount);
1317 } else {
1318 s->s_mode = mode;
1319 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
Roman Gushchine33c2672022-05-31 20:22:24 -07001320 shrinker_debugfs_rename(&s->s_shrink, "sb-%s:%s",
1321 fc->fs_type->name, s->s_id);
David Howellsfe62c3a2019-03-27 14:15:16 +00001322 sb_set_blocksize(s, block_size(bdev));
1323 error = fill_super(s, fc);
1324 if (error) {
1325 deactivate_locked_super(s);
1326 return error;
1327 }
1328
1329 s->s_flags |= SB_ACTIVE;
1330 bdev->bd_super = s;
1331 }
1332
1333 BUG_ON(fc->root);
1334 fc->root = dget(s->s_root);
1335 return 0;
1336}
1337EXPORT_SYMBOL(get_tree_bdev);
1338
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339static int test_bdev_super(struct super_block *s, void *data)
1340{
Daniil Lunev04b94072022-07-27 16:44:24 +10001341 return !(s->s_iflags & SB_I_RETIRED) && (void *)s->s_bdev == data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342}
1343
Al Viro152a0832010-07-25 00:46:55 +04001344struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +04001346 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
1348 struct block_device *bdev;
1349 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +01001350 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 int error = 0;
1352
David Howellse462ec52017-07-17 08:45:35 +01001353 if (!(flags & SB_RDONLY))
Al Viro30c40d22008-02-22 19:50:45 -05001354 mode |= FMODE_WRITE;
1355
Tejun Heod4d77622010-11-13 11:55:18 +01001356 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +04001358 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
1360 /*
1361 * once the super is inserted into the list by sget, s_umount
1362 * will protect the lockfs code from trying to start a snapshot
1363 * while we are mounting
1364 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001365 mutex_lock(&bdev->bd_fsfreeze_mutex);
1366 if (bdev->bd_fsfreeze_count > 0) {
1367 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1368 error = -EBUSY;
1369 goto error_bdev;
1370 }
David Howellse462ec52017-07-17 08:45:35 +01001371 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
David Howells9249e172012-06-25 12:55:37 +01001372 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001373 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -07001375 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
1377 if (s->s_root) {
David Howellse462ec52017-07-17 08:45:35 +01001378 if ((flags ^ s->s_flags) & SB_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001379 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001380 error = -EBUSY;
1381 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 }
David Howells454e2392006-06-23 02:02:57 -07001383
Tejun Heo4f331f02010-07-20 15:18:07 -07001384 /*
Christoph Hellwiga8698702021-05-25 08:12:56 +02001385 * s_umount nests inside open_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +01001386 * __invalidate_device(). blkdev_put() acquires
Christoph Hellwiga8698702021-05-25 08:12:56 +02001387 * open_mutex and can't be called under s_umount. Drop
Tejun Heoe525fd82010-11-13 11:55:17 +01001388 * s_umount temporarily. This is safe as we're
1389 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -07001390 */
1391 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +01001392 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -07001393 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 } else {
Al Viro30c40d22008-02-22 19:50:45 -05001395 s->s_mode = mode;
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001396 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
Roman Gushchine33c2672022-05-31 20:22:24 -07001397 shrinker_debugfs_rename(&s->s_shrink, "sb-%s:%s",
1398 fs_type->name, s->s_id);
Pekka Enberge78c9a02006-01-08 01:03:39 -08001399 sb_set_blocksize(s, block_size(bdev));
David Howellse462ec52017-07-17 08:45:35 +01001400 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001402 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001403 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001404 }
David Howells454e2392006-06-23 02:02:57 -07001405
David Howellse462ec52017-07-17 08:45:35 +01001406 s->s_flags |= SB_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001407 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 }
1409
Al Viro152a0832010-07-25 00:46:55 +04001410 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
David Howells454e2392006-06-23 02:02:57 -07001412error_s:
1413 error = PTR_ERR(s);
1414error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001415 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001416error:
Al Viro152a0832010-07-25 00:46:55 +04001417 return ERR_PTR(error);
1418}
1419EXPORT_SYMBOL(mount_bdev);
1420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421void kill_block_super(struct super_block *sb)
1422{
1423 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001424 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001426 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 generic_shutdown_super(sb);
1428 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001429 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001430 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431}
1432
1433EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001434#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
Al Viro3c26ff62010-07-25 11:46:36 +04001436struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001438 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439{
1440 int error;
David Howells9249e172012-06-25 12:55:37 +01001441 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
1443 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001444 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
David Howellse462ec52017-07-17 08:45:35 +01001446 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001448 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001449 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 }
David Howellse462ec52017-07-17 08:45:35 +01001451 s->s_flags |= SB_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001452 return dget(s->s_root);
1453}
1454EXPORT_SYMBOL(mount_nodev);
1455
NeilBrowna6097182022-01-17 09:07:26 +11001456int reconfigure_single(struct super_block *s,
1457 int flags, void *data)
David Howells8d0347f2018-11-04 09:28:36 -05001458{
1459 struct fs_context *fc;
1460 int ret;
1461
1462 /* The caller really need to be passing fc down into mount_single(),
1463 * then a chunk of this can be removed. [Bollocks -- AV]
1464 * Better yet, reconfiguration shouldn't happen, but rather the second
1465 * mount should be rejected if the parameters are not compatible.
1466 */
1467 fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1468 if (IS_ERR(fc))
1469 return PTR_ERR(fc);
1470
1471 ret = parse_monolithic_mount_data(fc, data);
1472 if (ret < 0)
1473 goto out;
1474
1475 ret = reconfigure_super(fc);
1476out:
1477 put_fs_context(fc);
1478 return ret;
1479}
1480
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481static int compare_single(struct super_block *s, void *p)
1482{
1483 return 1;
1484}
1485
Al Virofc14f2f2010-07-25 01:48:30 +04001486struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001488 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489{
1490 struct super_block *s;
1491 int error;
1492
David Howells9249e172012-06-25 12:55:37 +01001493 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001495 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 if (!s->s_root) {
David Howellse462ec52017-07-17 08:45:35 +01001497 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
David Howells8d0347f2018-11-04 09:28:36 -05001498 if (!error)
1499 s->s_flags |= SB_ACTIVE;
Kay Sievers9329d1be2009-12-18 21:18:15 +01001500 } else {
David Howells8d0347f2018-11-04 09:28:36 -05001501 error = reconfigure_single(s, flags, data);
1502 }
1503 if (unlikely(error)) {
1504 deactivate_locked_super(s);
1505 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 }
Al Virofc14f2f2010-07-25 01:48:30 +04001507 return dget(s->s_root);
1508}
1509EXPORT_SYMBOL(mount_single);
1510
David Howells9bc61ab2018-11-04 03:19:03 -05001511/**
1512 * vfs_get_tree - Get the mountable root
1513 * @fc: The superblock configuration context.
1514 *
1515 * The filesystem is invoked to get or create a superblock which can then later
1516 * be used for mounting. The filesystem places a pointer to the root to be
1517 * used for mounting in @fc->root.
1518 */
1519int vfs_get_tree(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520{
Al Viro9d412a42011-03-17 22:08:28 -04001521 struct super_block *sb;
David Howells9bc61ab2018-11-04 03:19:03 -05001522 int error;
Al Viro80893522010-02-05 09:30:46 -05001523
Al Virof3a09c92018-12-23 18:55:56 -05001524 if (fc->root)
1525 return -EBUSY;
1526
1527 /* Get the mountable root in fc->root, with a ref on the root and a ref
1528 * on the superblock.
1529 */
1530 error = fc->ops->get_tree(fc);
David Howells9bc61ab2018-11-04 03:19:03 -05001531 if (error < 0)
1532 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
Al Virof3a09c92018-12-23 18:55:56 -05001534 if (!fc->root) {
1535 pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1536 fc->fs_type->name);
1537 /* We don't know what the locking state of the superblock is -
1538 * if there is a superblock.
1539 */
1540 BUG();
1541 }
1542
David Howells9bc61ab2018-11-04 03:19:03 -05001543 sb = fc->root->d_sb;
Al Viro9d412a42011-03-17 22:08:28 -04001544 WARN_ON(!sb->s_bdi);
Dave Chinner79f546a2018-05-11 11:20:57 +10001545
1546 /*
1547 * Write barrier is for super_cache_count(). We place it before setting
1548 * SB_BORN as the data dependency between the two functions is the
1549 * superblock structure contents that we just set up, not the SB_BORN
1550 * flag.
1551 */
1552 smp_wmb();
David Howellse462ec52017-07-17 08:45:35 +01001553 sb->s_flags |= SB_BORN;
David Howells454e2392006-06-23 02:02:57 -07001554
David Howells9bc61ab2018-11-04 03:19:03 -05001555 error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
Al Viroc9ce29e2018-12-20 15:04:50 -05001556 if (unlikely(error)) {
1557 fc_drop_locked(fc);
1558 return error;
Al Viroa10d7c22018-12-05 11:58:35 -05001559 }
1560
Jeff Layton42cb56a2009-09-18 13:05:53 -07001561 /*
1562 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1563 * but s_maxbytes was an unsigned long long for many releases. Throw
1564 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001565 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001566 */
Al Viro9d412a42011-03-17 22:08:28 -04001567 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
David Howells9bc61ab2018-11-04 03:19:03 -05001568 "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001569
David Howells9bc61ab2018-11-04 03:19:03 -05001570 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571}
David Howells9bc61ab2018-11-04 03:19:03 -05001572EXPORT_SYMBOL(vfs_get_tree);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573
Jan Kara5accdf82012-06-12 16:20:34 +02001574/*
Jan Karafca39342017-04-12 12:24:28 +02001575 * Setup private BDI for given superblock. It gets automatically cleaned up
1576 * in generic_shutdown_super().
1577 */
1578int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1579{
1580 struct backing_dev_info *bdi;
1581 int err;
1582 va_list args;
1583
Christoph Hellwigaef33c22020-05-04 14:48:00 +02001584 bdi = bdi_alloc(NUMA_NO_NODE);
Jan Karafca39342017-04-12 12:24:28 +02001585 if (!bdi)
1586 return -ENOMEM;
1587
Jan Karafca39342017-04-12 12:24:28 +02001588 va_start(args, fmt);
Jan Kara7c4cc302017-04-12 12:24:49 +02001589 err = bdi_register_va(bdi, fmt, args);
Jan Karafca39342017-04-12 12:24:28 +02001590 va_end(args);
1591 if (err) {
1592 bdi_put(bdi);
1593 return err;
1594 }
1595 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1596 sb->s_bdi = bdi;
Christoph Hellwig0b3ea092021-11-05 13:36:58 -07001597 sb->s_iflags |= SB_I_PERSB_BDI;
Jan Karafca39342017-04-12 12:24:28 +02001598
1599 return 0;
1600}
1601EXPORT_SYMBOL(super_setup_bdi_name);
1602
1603/*
1604 * Setup private BDI for given superblock. I gets automatically cleaned up
1605 * in generic_shutdown_super().
1606 */
1607int super_setup_bdi(struct super_block *sb)
1608{
1609 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1610
1611 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1612 atomic_long_inc_return(&bdi_seq));
1613}
1614EXPORT_SYMBOL(super_setup_bdi);
1615
Jan Kara5accdf82012-06-12 16:20:34 +02001616/**
1617 * sb_wait_write - wait until all writers to given file system finish
1618 * @sb: the super for which we wait
1619 * @level: type of writers we wait for (normal vs page fault)
1620 *
1621 * This function waits until there are no writers of given type to given file
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001622 * system.
Jan Kara5accdf82012-06-12 16:20:34 +02001623 */
1624static void sb_wait_write(struct super_block *sb, int level)
1625{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001626 percpu_down_write(sb->s_writers.rw_sem + level-1);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001627}
Jan Kara5accdf82012-06-12 16:20:34 +02001628
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001629/*
1630 * We are going to return to userspace and forget about these locks, the
1631 * ownership goes to the caller of thaw_super() which does unlock().
1632 */
1633static void lockdep_sb_freeze_release(struct super_block *sb)
1634{
1635 int level;
1636
1637 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1638 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1639}
1640
1641/*
1642 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1643 */
1644static void lockdep_sb_freeze_acquire(struct super_block *sb)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001645{
1646 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001647
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001648 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1649 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001650}
1651
Darrick J. Wong2719c7162022-01-30 08:53:16 -08001652static void sb_freeze_unlock(struct super_block *sb, int level)
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001653{
Darrick J. Wong2719c7162022-01-30 08:53:16 -08001654 for (level--; level >= 0; level--)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001655 percpu_up_write(sb->s_writers.rw_sem + level);
Jan Kara5accdf82012-06-12 16:20:34 +02001656}
1657
Josef Bacik18e9e512010-03-23 10:34:56 -04001658/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001659 * freeze_super - lock the filesystem and force it into a consistent state
1660 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001661 *
1662 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1663 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1664 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001665 *
1666 * During this function, sb->s_writers.frozen goes through these values:
1667 *
1668 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1669 *
1670 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1671 * writes should be blocked, though page faults are still allowed. We wait for
1672 * all writes to complete and then proceed to the next stage.
1673 *
1674 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1675 * but internal fs threads can still modify the filesystem (although they
1676 * should not dirty new pages or inodes), writeback can run etc. After waiting
1677 * for all running page faults we sync the filesystem which will clean all
1678 * dirty pages and inodes (no new dirty pages or inodes can be created when
1679 * sync is running).
1680 *
1681 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1682 * modification are blocked (e.g. XFS preallocation truncation on inode
1683 * reclaim). This is usually implemented by blocking new transactions for
1684 * filesystems that have them and need this additional guard. After all
1685 * internal writers are finished we call ->freeze_fs() to finish filesystem
1686 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1687 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1688 *
1689 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001690 */
1691int freeze_super(struct super_block *sb)
1692{
1693 int ret;
1694
1695 atomic_inc(&sb->s_active);
1696 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001697 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001698 deactivate_locked_super(sb);
1699 return -EBUSY;
1700 }
1701
David Howellse462ec52017-07-17 08:45:35 +01001702 if (!(sb->s_flags & SB_BORN)) {
Al Virodabe0dc2012-01-03 21:01:29 -05001703 up_write(&sb->s_umount);
1704 return 0; /* sic - it's "nothing to do" */
1705 }
1706
David Howellsbc98a422017-07-17 08:45:34 +01001707 if (sb_rdonly(sb)) {
Jan Kara5accdf82012-06-12 16:20:34 +02001708 /* Nothing to do really... */
1709 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001710 up_write(&sb->s_umount);
1711 return 0;
1712 }
1713
Jan Kara5accdf82012-06-12 16:20:34 +02001714 sb->s_writers.frozen = SB_FREEZE_WRITE;
Jan Kara5accdf82012-06-12 16:20:34 +02001715 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1716 up_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001717 sb_wait_write(sb, SB_FREEZE_WRITE);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001718 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001719
1720 /* Now we go and block page faults... */
Jan Kara5accdf82012-06-12 16:20:34 +02001721 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
Jan Kara5accdf82012-06-12 16:20:34 +02001722 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1723
1724 /* All writers are done so after syncing there won't be dirty data */
Darrick J. Wong2719c7162022-01-30 08:53:16 -08001725 ret = sync_filesystem(sb);
1726 if (ret) {
1727 sb->s_writers.frozen = SB_UNFROZEN;
1728 sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
1729 wake_up(&sb->s_writers.wait_unfrozen);
1730 deactivate_locked_super(sb);
1731 return ret;
1732 }
Josef Bacik18e9e512010-03-23 10:34:56 -04001733
Jan Kara5accdf82012-06-12 16:20:34 +02001734 /* Now wait for internal filesystem counter */
1735 sb->s_writers.frozen = SB_FREEZE_FS;
Jan Kara5accdf82012-06-12 16:20:34 +02001736 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001737
Josef Bacik18e9e512010-03-23 10:34:56 -04001738 if (sb->s_op->freeze_fs) {
1739 ret = sb->s_op->freeze_fs(sb);
1740 if (ret) {
1741 printk(KERN_ERR
1742 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001743 sb->s_writers.frozen = SB_UNFROZEN;
Darrick J. Wong2719c7162022-01-30 08:53:16 -08001744 sb_freeze_unlock(sb, SB_FREEZE_FS);
Jan Kara5accdf82012-06-12 16:20:34 +02001745 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001746 deactivate_locked_super(sb);
1747 return ret;
1748 }
1749 }
Jan Kara5accdf82012-06-12 16:20:34 +02001750 /*
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001751 * For debugging purposes so that fs can warn if it sees write activity
1752 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
Jan Kara5accdf82012-06-12 16:20:34 +02001753 */
1754 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001755 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001756 up_write(&sb->s_umount);
1757 return 0;
1758}
1759EXPORT_SYMBOL(freeze_super);
1760
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001761static int thaw_super_locked(struct super_block *sb)
Josef Bacik18e9e512010-03-23 10:34:56 -04001762{
1763 int error;
1764
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001765 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001766 up_write(&sb->s_umount);
1767 return -EINVAL;
1768 }
1769
David Howellsbc98a422017-07-17 08:45:34 +01001770 if (sb_rdonly(sb)) {
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001771 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001772 goto out;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001773 }
Josef Bacik18e9e512010-03-23 10:34:56 -04001774
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001775 lockdep_sb_freeze_acquire(sb);
1776
Josef Bacik18e9e512010-03-23 10:34:56 -04001777 if (sb->s_op->unfreeze_fs) {
1778 error = sb->s_op->unfreeze_fs(sb);
1779 if (error) {
1780 printk(KERN_ERR
1781 "VFS:Filesystem thaw failed\n");
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001782 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001783 up_write(&sb->s_umount);
1784 return error;
1785 }
1786 }
1787
Jan Kara5accdf82012-06-12 16:20:34 +02001788 sb->s_writers.frozen = SB_UNFROZEN;
Darrick J. Wong2719c7162022-01-30 08:53:16 -08001789 sb_freeze_unlock(sb, SB_FREEZE_FS);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001790out:
Jan Kara5accdf82012-06-12 16:20:34 +02001791 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001792 deactivate_locked_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001793 return 0;
1794}
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001795
Mauro Carvalho Chehab961f3c82021-01-14 09:04:39 +01001796/**
1797 * thaw_super -- unlock filesystem
1798 * @sb: the super to thaw
1799 *
1800 * Unlocks the filesystem and marks it writeable again after freeze_super().
1801 */
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001802int thaw_super(struct super_block *sb)
1803{
1804 down_write(&sb->s_umount);
1805 return thaw_super_locked(sb);
1806}
Josef Bacik18e9e512010-03-23 10:34:56 -04001807EXPORT_SYMBOL(thaw_super);