blob: 583a0124bc394d2919b2fbc346948c5ab9684201 [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>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060034#include <linux/cleancache.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;
Vladimir Davydov3cb29d12015-04-14 15:46:48 -0700260 s->cleancache_poolid = CLEANCACHE_NO_POOL;
Al Viro7eb5e882013-10-01 15:09:58 -0400261
262 s->s_shrink.seeks = DEFAULT_SEEKS;
263 s->s_shrink.scan_objects = super_cache_scan;
264 s->s_shrink.count_objects = super_cache_count;
265 s->s_shrink.batch = 1024;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800266 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
Tetsuo Handa8e049442018-04-04 19:53:07 +0900267 if (prealloc_shrinker(&s->s_shrink))
268 goto fail;
Kirill Tkhaic92e8e12018-08-17 15:47:50 -0700269 if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
Kirill Tkhai2b3648a2018-08-17 15:47:45 -0700270 goto fail;
Kirill Tkhaic92e8e12018-08-17 15:47:50 -0700271 if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
Kirill Tkhai2b3648a2018-08-17 15:47:45 -0700272 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000274
Al Viro7eb5e882013-10-01 15:09:58 -0400275fail:
Al Viro02008942017-10-11 13:48:55 -0400276 destroy_unused_super(s);
Al Viro7eb5e882013-10-01 15:09:58 -0400277 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278}
279
280/* Superblock refcounting */
281
282/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400283 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 */
Al Viroc645b932017-10-11 13:48:55 -0400285static void __put_super(struct super_block *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286{
Al Viroc645b932017-10-11 13:48:55 -0400287 if (!--s->s_count) {
288 list_del_init(&s->s_list);
289 WARN_ON(s->s_dentry_lru.node);
290 WARN_ON(s->s_inode_lru.node);
291 WARN_ON(!list_empty(&s->s_mounts));
292 security_sb_free(s);
293 put_user_ns(s->s_user_ns);
294 kfree(s->s_subtype);
295 call_rcu(&s->rcu, destroy_super_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297}
298
299/**
300 * put_super - drop a temporary reference to superblock
301 * @sb: superblock in question
302 *
303 * Drops a temporary reference, frees superblock if there's no
304 * references left.
305 */
Al Virof47ec3f22011-11-21 21:15:42 -0500306static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
308 spin_lock(&sb_lock);
309 __put_super(sb);
310 spin_unlock(&sb_lock);
311}
312
313
314/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400315 * deactivate_locked_super - drop an active reference to superblock
316 * @s: superblock to deactivate
317 *
Masanari Iidabd7ced92016-02-02 22:31:06 +0900318 * Drops an active reference to superblock, converting it into a temporary
Al Viro1712ac82010-03-22 15:22:31 -0400319 * one if there is no other active references left. In that case we
320 * tell fs driver to shut it down and drop the temporary reference we
321 * had just acquired.
322 *
323 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400324 */
325void deactivate_locked_super(struct super_block *s)
326{
327 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400328 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400329 cleancache_invalidate_fs(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000330 unregister_shrinker(&s->s_shrink);
Dave Chinner28f2cd4f2014-06-04 16:10:46 -0700331 fs->kill_sb(s);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000332
Vladimir Davydovc0a5b562015-02-12 14:59:07 -0800333 /*
334 * Since list_lru_destroy() may sleep, we cannot call it from
335 * put_super(), where we hold the sb_lock. Therefore we destroy
336 * the lru lists right now.
337 */
338 list_lru_destroy(&s->s_dentry_lru);
339 list_lru_destroy(&s->s_inode_lru);
340
Al Viro74dbbdd2009-05-06 01:07:50 -0400341 put_filesystem(fs);
342 put_super(s);
343 } else {
344 up_write(&s->s_umount);
345 }
346}
347
348EXPORT_SYMBOL(deactivate_locked_super);
349
350/**
Al Viro1712ac82010-03-22 15:22:31 -0400351 * deactivate_super - drop an active reference to superblock
352 * @s: superblock to deactivate
353 *
354 * Variant of deactivate_locked_super(), except that superblock is *not*
355 * locked by caller. If we are going to drop the final active reference,
356 * lock will be acquired prior to that.
357 */
358void deactivate_super(struct super_block *s)
359{
360 if (!atomic_add_unless(&s->s_active, -1, 1)) {
361 down_write(&s->s_umount);
362 deactivate_locked_super(s);
363 }
364}
365
366EXPORT_SYMBOL(deactivate_super);
367
368/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 * grab_super - acquire an active reference
370 * @s: reference we are trying to make active
371 *
372 * Tries to acquire an active reference. grab_super() is used when we
373 * had just found a superblock in super_blocks or fs_type->fs_supers
374 * and want to turn it into a full-blown active reference. grab_super()
375 * is called with sb_lock held and drops it. Returns 1 in case of
376 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400377 * dying when grab_super() had been called). Note that this is only
378 * called for superblocks not in rundown mode (== ones still on ->fs_supers
379 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700381static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382{
383 s->s_count++;
384 spin_unlock(&sb_lock);
385 down_write(&s->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100386 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
Al Viroacfec9a2013-07-20 03:13:55 +0400387 put_super(s);
388 return 1;
389 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 up_write(&s->s_umount);
391 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 return 0;
393}
394
David Howellscf9a2ae2006-08-29 19:05:54 +0100395/*
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300396 * trylock_super - try to grab ->s_umount shared
Wanpeng Li331cbde2012-06-09 11:10:55 +0800397 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000398 *
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300399 * Try to prevent fs shutdown. This is used in places where we
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000400 * cannot take an active reference but we need to ensure that the
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300401 * filesystem is not shut down while we are working on it. It returns
402 * false if we cannot acquire s_umount or if we lose the race and
403 * filesystem already got into shutdown, and returns true with the s_umount
404 * lock held in read mode in case of success. On successful return,
405 * the caller must drop the s_umount lock when done.
406 *
407 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
408 * The reason why it's safe is that we are OK with doing trylock instead
409 * of down_read(). There's a couple of places that are OK with that, but
410 * it's very much not a general-purpose interface.
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000411 */
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300412bool trylock_super(struct super_block *sb)
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000413{
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000414 if (down_read_trylock(&sb->s_umount)) {
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300415 if (!hlist_unhashed(&sb->s_instances) &&
David Howellse462ec52017-07-17 08:45:35 +0100416 sb->s_root && (sb->s_flags & SB_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000417 return true;
418 up_read(&sb->s_umount);
419 }
420
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000421 return false;
422}
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424/**
425 * generic_shutdown_super - common helper for ->kill_sb()
426 * @sb: superblock to kill
427 *
428 * generic_shutdown_super() does all fs-independent work on superblock
429 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
430 * that need destruction out of superblock, call generic_shutdown_super()
431 * and release aforementioned objects. Note: dentries and inodes _are_
432 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700433 *
434 * Upon calling this function, the filesystem may no longer alter or
435 * rearrange the set of dentries belonging to this super_block, nor may it
436 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 */
438void generic_shutdown_super(struct super_block *sb)
439{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800440 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
David Howellsc636ebd2006-10-11 01:22:19 -0700442 if (sb->s_root) {
443 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200444 sync_filesystem(sb);
David Howellse462ec52017-07-17 08:45:35 +0100445 sb->s_flags &= ~SB_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800446
Amir Goldstein1e6cb722018-09-01 10:41:11 +0300447 fsnotify_sb_delete(sb);
Tejun Heoa1a0e232016-02-29 18:28:53 -0500448 cgroup_writeback_umount();
Al Viro63997e92010-10-25 20:49:35 -0400449
450 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200452 if (sb->s_dio_done_wq) {
453 destroy_workqueue(sb->s_dio_done_wq);
454 sb->s_dio_done_wq = NULL;
455 }
456
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 if (sop->put_super)
458 sop->put_super(sb);
459
Al Viro63997e92010-10-25 20:49:35 -0400460 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800461 printk("VFS: Busy inodes after unmount of %s. "
462 "Self-destruct in 5 seconds. Have a nice day...\n",
463 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 }
466 spin_lock(&sb_lock);
467 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500468 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 spin_unlock(&sb_lock);
470 up_write(&sb->s_umount);
Jan Karac1844d52017-04-12 12:24:47 +0200471 if (sb->s_bdi != &noop_backing_dev_info) {
Jan Karafca39342017-04-12 12:24:28 +0200472 bdi_put(sb->s_bdi);
473 sb->s_bdi = &noop_backing_dev_info;
Jan Karafca39342017-04-12 12:24:28 +0200474 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475}
476
477EXPORT_SYMBOL(generic_shutdown_super);
478
479/**
Al Virocb50b342018-12-23 17:25:47 -0500480 * sget_fc - Find or create a superblock
481 * @fc: Filesystem context.
482 * @test: Comparison callback
483 * @set: Setup callback
484 *
485 * Find or create a superblock using the parameters stored in the filesystem
486 * context and the two callback functions.
487 *
488 * If an extant superblock is matched, then that will be returned with an
489 * elevated reference count that the caller must transfer or discard.
490 *
491 * If no match is made, a new superblock will be allocated and basic
492 * initialisation will be performed (s_type, s_fs_info and s_id will be set and
493 * the set() callback will be invoked), the superblock will be published and it
494 * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
495 * as yet unset.
496 */
497struct super_block *sget_fc(struct fs_context *fc,
498 int (*test)(struct super_block *, struct fs_context *),
499 int (*set)(struct super_block *, struct fs_context *))
500{
501 struct super_block *s = NULL;
502 struct super_block *old;
503 struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
504 int err;
505
506 if (!(fc->sb_flags & SB_KERNMOUNT) &&
507 fc->purpose != FS_CONTEXT_FOR_SUBMOUNT) {
508 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
509 * over the namespace.
510 */
511 if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT)) {
512 if (!capable(CAP_SYS_ADMIN))
513 return ERR_PTR(-EPERM);
514 } else {
515 if (!ns_capable(fc->user_ns, CAP_SYS_ADMIN))
516 return ERR_PTR(-EPERM);
517 }
518 }
519
520retry:
521 spin_lock(&sb_lock);
522 if (test) {
523 hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
524 if (test(old, fc))
525 goto share_extant_sb;
526 }
527 }
528 if (!s) {
529 spin_unlock(&sb_lock);
530 s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
531 if (!s)
532 return ERR_PTR(-ENOMEM);
533 goto retry;
534 }
535
536 s->s_fs_info = fc->s_fs_info;
537 err = set(s, fc);
538 if (err) {
539 s->s_fs_info = NULL;
540 spin_unlock(&sb_lock);
541 destroy_unused_super(s);
542 return ERR_PTR(err);
543 }
544 fc->s_fs_info = NULL;
545 s->s_type = fc->fs_type;
546 strlcpy(s->s_id, s->s_type->name, sizeof(s->s_id));
547 list_add_tail(&s->s_list, &super_blocks);
548 hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
549 spin_unlock(&sb_lock);
550 get_filesystem(s->s_type);
551 register_shrinker_prepared(&s->s_shrink);
552 return s;
553
554share_extant_sb:
555 if (user_ns != old->s_user_ns) {
556 spin_unlock(&sb_lock);
557 destroy_unused_super(s);
558 return ERR_PTR(-EBUSY);
559 }
560 if (!grab_super(old))
561 goto retry;
562 destroy_unused_super(s);
563 return old;
564}
565EXPORT_SYMBOL(sget_fc);
566
567/**
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500568 * sget_userns - find or create a superblock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 * @type: filesystem type superblock should belong to
570 * @test: comparison callback
571 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100572 * @flags: mount flags
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500573 * @user_ns: User namespace for the super_block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 * @data: argument to each of them
575 */
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500576struct super_block *sget_userns(struct file_system_type *type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 int (*test)(struct super_block *,void *),
578 int (*set)(struct super_block *,void *),
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500579 int flags, struct user_namespace *user_ns,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 void *data)
581{
582 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700583 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 int err;
585
David Howellse462ec52017-07-17 08:45:35 +0100586 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) &&
Eric W. Biedermana001e742016-06-06 15:48:04 -0500587 !(type->fs_flags & FS_USERNS_MOUNT) &&
588 !capable(CAP_SYS_ADMIN))
589 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590retry:
591 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700592 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800593 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700594 if (!test(old, data))
595 continue;
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500596 if (user_ns != old->s_user_ns) {
597 spin_unlock(&sb_lock);
Al Viro02008942017-10-11 13:48:55 -0400598 destroy_unused_super(s);
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500599 return ERR_PTR(-EBUSY);
600 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700601 if (!grab_super(old))
602 goto retry;
Al Viro02008942017-10-11 13:48:55 -0400603 destroy_unused_super(s);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700604 return old;
605 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 }
607 if (!s) {
608 spin_unlock(&sb_lock);
David Howellse462ec52017-07-17 08:45:35 +0100609 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 if (!s)
611 return ERR_PTR(-ENOMEM);
612 goto retry;
613 }
David Howellsdd111b32017-07-04 17:25:09 +0100614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 err = set(s, data);
616 if (err) {
617 spin_unlock(&sb_lock);
Al Viro02008942017-10-11 13:48:55 -0400618 destroy_unused_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 return ERR_PTR(err);
620 }
621 s->s_type = type;
622 strlcpy(s->s_id, type->name, sizeof(s->s_id));
623 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500624 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 spin_unlock(&sb_lock);
626 get_filesystem(type);
Tetsuo Handa8e049442018-04-04 19:53:07 +0900627 register_shrinker_prepared(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 return s;
629}
630
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500631EXPORT_SYMBOL(sget_userns);
632
633/**
634 * sget - find or create a superblock
635 * @type: filesystem type superblock should belong to
636 * @test: comparison callback
637 * @set: setup callback
638 * @flags: mount flags
639 * @data: argument to each of them
640 */
641struct super_block *sget(struct file_system_type *type,
642 int (*test)(struct super_block *,void *),
643 int (*set)(struct super_block *,void *),
644 int flags,
645 void *data)
646{
647 struct user_namespace *user_ns = current_user_ns();
648
Eric W. Biederman93faccbb2017-02-01 06:06:16 +1300649 /* We don't yet pass the user namespace of the parent
650 * mount through to here so always use &init_user_ns
651 * until that changes.
652 */
David Howellse462ec52017-07-17 08:45:35 +0100653 if (flags & SB_SUBMOUNT)
Eric W. Biederman93faccbb2017-02-01 06:06:16 +1300654 user_ns = &init_user_ns;
655
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500656 /* Ensure the requestor has permissions over the target filesystem */
David Howellse462ec52017-07-17 08:45:35 +0100657 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500658 return ERR_PTR(-EPERM);
659
660 return sget_userns(type, test, set, flags, user_ns, data);
661}
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
Jan Karaba6379f2016-11-23 12:53:00 +0100772static struct super_block *__get_super(struct block_device *bdev, bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700774 struct super_block *sb;
775
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 if (!bdev)
777 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700780rescan:
781 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500782 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400783 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700784 if (sb->s_bdev == bdev) {
785 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 spin_unlock(&sb_lock);
Jan Karaba6379f2016-11-23 12:53:00 +0100787 if (!excl)
788 down_read(&sb->s_umount);
789 else
790 down_write(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400791 /* still alive? */
David Howellse462ec52017-07-17 08:45:35 +0100792 if (sb->s_root && (sb->s_flags & SB_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700793 return sb;
Jan Karaba6379f2016-11-23 12:53:00 +0100794 if (!excl)
795 up_read(&sb->s_umount);
796 else
797 up_write(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400798 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700799 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400800 __put_super(sb);
801 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 }
803 }
804 spin_unlock(&sb_lock);
805 return NULL;
806}
807
Jan Karaba6379f2016-11-23 12:53:00 +0100808/**
809 * get_super - get the superblock of a device
810 * @bdev: device to get the superblock for
811 *
812 * Scans the superblock list and finds the superblock of the file system
813 * mounted on the device given. %NULL is returned if no match is found.
814 */
815struct super_block *get_super(struct block_device *bdev)
816{
817 return __get_super(bdev, false);
818}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200820
Jan Karaba6379f2016-11-23 12:53:00 +0100821static struct super_block *__get_super_thawed(struct block_device *bdev,
822 bool excl)
823{
824 while (1) {
825 struct super_block *s = __get_super(bdev, excl);
826 if (!s || s->s_writers.frozen == SB_UNFROZEN)
827 return s;
828 if (!excl)
829 up_read(&s->s_umount);
830 else
831 up_write(&s->s_umount);
832 wait_event(s->s_writers.wait_unfrozen,
833 s->s_writers.frozen == SB_UNFROZEN);
834 put_super(s);
835 }
836}
837
Christoph Hellwig45042302009-08-03 23:28:35 +0200838/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100839 * get_super_thawed - get thawed superblock of a device
840 * @bdev: device to get the superblock for
841 *
842 * Scans the superblock list and finds the superblock of the file system
843 * mounted on the device. The superblock is returned once it is thawed
844 * (or immediately if it was not frozen). %NULL is returned if no match
845 * is found.
846 */
847struct super_block *get_super_thawed(struct block_device *bdev)
848{
Jan Karaba6379f2016-11-23 12:53:00 +0100849 return __get_super_thawed(bdev, false);
Jan Kara6b6dc832012-02-10 11:03:00 +0100850}
851EXPORT_SYMBOL(get_super_thawed);
852
853/**
Jan Karaba6379f2016-11-23 12:53:00 +0100854 * get_super_exclusive_thawed - get thawed superblock of a device
855 * @bdev: device to get the superblock for
856 *
857 * Scans the superblock list and finds the superblock of the file system
858 * mounted on the device. The superblock is returned once it is thawed
859 * (or immediately if it was not frozen) and s_umount semaphore is held
860 * in exclusive mode. %NULL is returned if no match is found.
861 */
862struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
863{
864 return __get_super_thawed(bdev, true);
865}
866EXPORT_SYMBOL(get_super_exclusive_thawed);
867
868/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200869 * get_active_super - get an active reference to the superblock of a device
870 * @bdev: device to get the superblock for
871 *
872 * Scans the superblock list and finds the superblock of the file system
873 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400874 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200875 */
876struct super_block *get_active_super(struct block_device *bdev)
877{
878 struct super_block *sb;
879
880 if (!bdev)
881 return NULL;
882
Al Viro14945832010-03-22 20:15:33 -0400883restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200884 spin_lock(&sb_lock);
885 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500886 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400887 continue;
Al Viro14945832010-03-22 20:15:33 -0400888 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400889 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400890 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400891 up_write(&sb->s_umount);
892 return sb;
Al Viro14945832010-03-22 20:15:33 -0400893 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200894 }
895 spin_unlock(&sb_lock);
896 return NULL;
897}
David Howellsdd111b32017-07-04 17:25:09 +0100898
Al Virodf40c012010-03-22 20:23:25 -0400899struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700901 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700904rescan:
905 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500906 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400907 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700908 if (sb->s_dev == dev) {
909 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700911 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400912 /* still alive? */
David Howellse462ec52017-07-17 08:45:35 +0100913 if (sb->s_root && (sb->s_flags & SB_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700914 return sb;
915 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400916 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700917 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400918 __put_super(sb);
919 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 }
921 }
922 spin_unlock(&sb_lock);
923 return NULL;
924}
925
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926/**
David Howells8d0347f2018-11-04 09:28:36 -0500927 * reconfigure_super - asks filesystem to change superblock parameters
928 * @fc: The superblock and configuration
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 *
David Howells8d0347f2018-11-04 09:28:36 -0500930 * Alters the configuration parameters of a live superblock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 */
David Howells8d0347f2018-11-04 09:28:36 -0500932int reconfigure_super(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933{
David Howells8d0347f2018-11-04 09:28:36 -0500934 struct super_block *sb = fc->root->d_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 int retval;
David Howells8d0347f2018-11-04 09:28:36 -0500936 bool remount_ro = false;
937 bool force = fc->sb_flags & SB_FORCE;
Christoph Hellwig45042302009-08-03 23:28:35 +0200938
David Howells8d0347f2018-11-04 09:28:36 -0500939 if (fc->sb_flags_mask & ~MS_RMT_MASK)
940 return -EINVAL;
Jan Kara5accdf82012-06-12 16:20:34 +0200941 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200942 return -EBUSY;
943
David Howells8d0347f2018-11-04 09:28:36 -0500944 retval = security_sb_remount(sb, fc->security);
945 if (retval)
946 return retval;
947
948 if (fc->sb_flags_mask & SB_RDONLY) {
David Howells93614012006-09-30 20:45:40 +0200949#ifdef CONFIG_BLOCK
David Howells8d0347f2018-11-04 09:28:36 -0500950 if (!(fc->sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
951 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200952#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200953
David Howells8d0347f2018-11-04 09:28:36 -0500954 remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
955 }
Nick Piggind208bbd2009-12-21 16:28:53 -0800956
Al Viro0aec09d2014-08-07 07:32:06 -0400957 if (remount_ro) {
Al Virofdab6842015-01-11 10:57:27 -0500958 if (!hlist_empty(&sb->s_pins)) {
Al Viro0aec09d2014-08-07 07:32:06 -0400959 up_write(&sb->s_umount);
Al Virofdab6842015-01-11 10:57:27 -0500960 group_pin_kill(&sb->s_pins);
Al Viro0aec09d2014-08-07 07:32:06 -0400961 down_write(&sb->s_umount);
962 if (!sb->s_root)
963 return 0;
964 if (sb->s_writers.frozen != SB_UNFROZEN)
965 return -EBUSY;
David Howells8d0347f2018-11-04 09:28:36 -0500966 remount_ro = !sb_rdonly(sb);
Al Viro0aec09d2014-08-07 07:32:06 -0400967 }
968 }
969 shrink_dcache_sb(sb);
970
David Howells8d0347f2018-11-04 09:28:36 -0500971 /* If we are reconfiguring to RDONLY and current sb is read/write,
972 * make sure there are no files open for writing.
973 */
Nick Piggind208bbd2009-12-21 16:28:53 -0800974 if (remount_ro) {
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100975 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400976 sb->s_readonly_remount = 1;
977 smp_wmb();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100978 } else {
979 retval = sb_prepare_remount_readonly(sb);
980 if (retval)
981 return retval;
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100982 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 }
984
Al Virof3a09c92018-12-23 18:55:56 -0500985 if (fc->ops->reconfigure) {
986 retval = fc->ops->reconfigure(fc);
987 if (retval) {
988 if (!force)
989 goto cancel_readonly;
990 /* If forced remount, go ahead despite any errors */
991 WARN(1, "forced remount of a %s fs returned %i\n",
992 sb->s_type->name, retval);
993 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 }
David Howells8d0347f2018-11-04 09:28:36 -0500995
996 WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
997 (fc->sb_flags & fc->sb_flags_mask)));
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100998 /* Needs to be ordered wrt mnt_is_readonly() */
999 smp_wmb();
1000 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -04001001
Nick Piggind208bbd2009-12-21 16:28:53 -08001002 /*
1003 * Some filesystems modify their metadata via some other path than the
1004 * bdev buffer cache (eg. use a private mapping, or directories in
1005 * pagecache, etc). Also file data modifications go via their own
1006 * mappings. So If we try to mount readonly then copy the filesystem
1007 * from bdev, we could get stale data, so invalidate it to give a best
1008 * effort at coherency.
1009 */
1010 if (remount_ro && sb->s_bdev)
1011 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 return 0;
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +01001013
1014cancel_readonly:
1015 sb->s_readonly_remount = 0;
1016 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017}
1018
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +02001019static void do_emergency_remount_callback(struct super_block *sb)
1020{
1021 down_write(&sb->s_umount);
1022 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
1023 !sb_rdonly(sb)) {
David Howells8d0347f2018-11-04 09:28:36 -05001024 struct fs_context *fc;
1025
1026 fc = fs_context_for_reconfigure(sb->s_root,
1027 SB_RDONLY | SB_FORCE, SB_RDONLY);
1028 if (!IS_ERR(fc)) {
1029 if (parse_monolithic_mount_data(fc, NULL) == 0)
1030 (void)reconfigure_super(fc);
1031 put_fs_context(fc);
1032 }
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +02001033 }
1034 up_write(&sb->s_umount);
1035}
1036
Jens Axboea2a95372009-03-17 09:38:40 +01001037static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038{
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +02001039 __iterate_supers(do_emergency_remount_callback);
Jens Axboea2a95372009-03-17 09:38:40 +01001040 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 printk("Emergency Remount complete\n");
1042}
1043
1044void emergency_remount(void)
1045{
Jens Axboea2a95372009-03-17 09:38:40 +01001046 struct work_struct *work;
1047
1048 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1049 if (work) {
1050 INIT_WORK(work, do_emergency_remount);
1051 schedule_work(work);
1052 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053}
1054
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001055static void do_thaw_all_callback(struct super_block *sb)
1056{
1057 down_write(&sb->s_umount);
Al Viro1c18d2a2018-05-10 15:09:41 -04001058 if (sb->s_root && sb->s_flags & SB_BORN) {
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001059 emergency_thaw_bdev(sb);
1060 thaw_super_locked(sb);
1061 } else {
1062 up_write(&sb->s_umount);
1063 }
1064}
1065
1066static void do_thaw_all(struct work_struct *work)
1067{
1068 __iterate_supers(do_thaw_all_callback);
1069 kfree(work);
1070 printk(KERN_WARNING "Emergency Thaw complete\n");
1071}
1072
1073/**
1074 * emergency_thaw_all -- forcibly thaw every frozen filesystem
1075 *
1076 * Used for emergency unfreeze of all filesystems via SysRq
1077 */
1078void emergency_thaw_all(void)
1079{
1080 struct work_struct *work;
1081
1082 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1083 if (work) {
1084 INIT_WORK(work, do_thaw_all);
1085 schedule_work(work);
1086 }
1087}
1088
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +04001089static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
Matthew Wilcox5a668472018-06-11 12:03:31 -04001091/**
1092 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1093 * @p: Pointer to a dev_t.
1094 *
1095 * Filesystems which don't use real block devices can call this function
1096 * to allocate a virtual block device.
1097 *
1098 * Context: Any context. Frequently called while holding sb_lock.
1099 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1100 * or -ENOMEM if memory allocation failed.
1101 */
Al Viro0ee5dc62011-07-07 15:44:25 -04001102int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103{
1104 int dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Matthew Wilcox5a668472018-06-11 12:03:31 -04001106 /*
1107 * Many userspace utilities consider an FSID of 0 invalid.
1108 * Always return at least 1 from get_anon_bdev.
1109 */
1110 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1111 GFP_ATOMIC);
1112 if (dev == -ENOSPC)
1113 dev = -EMFILE;
1114 if (dev < 0)
1115 return dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
Matthew Wilcox5a668472018-06-11 12:03:31 -04001117 *p = MKDEV(0, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 return 0;
1119}
Al Viro0ee5dc62011-07-07 15:44:25 -04001120EXPORT_SYMBOL(get_anon_bdev);
1121
1122void free_anon_bdev(dev_t dev)
1123{
Matthew Wilcox5a668472018-06-11 12:03:31 -04001124 ida_free(&unnamed_dev_ida, MINOR(dev));
Al Viro0ee5dc62011-07-07 15:44:25 -04001125}
1126EXPORT_SYMBOL(free_anon_bdev);
1127
1128int set_anon_super(struct super_block *s, void *data)
1129{
Christoph Hellwigdf0ce262015-01-14 10:42:41 +01001130 return get_anon_bdev(&s->s_dev);
Al Viro0ee5dc62011-07-07 15:44:25 -04001131}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132EXPORT_SYMBOL(set_anon_super);
1133
1134void kill_anon_super(struct super_block *sb)
1135{
Al Viro0ee5dc62011-07-07 15:44:25 -04001136 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -04001138 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140EXPORT_SYMBOL(kill_anon_super);
1141
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142void kill_litter_super(struct super_block *sb)
1143{
1144 if (sb->s_root)
1145 d_genocide(sb->s_root);
1146 kill_anon_super(sb);
1147}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148EXPORT_SYMBOL(kill_litter_super);
1149
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001150static int ns_test_super(struct super_block *sb, void *data)
1151{
1152 return sb->s_fs_info == data;
1153}
1154
1155static int ns_set_super(struct super_block *sb, void *data)
1156{
1157 sb->s_fs_info = data;
1158 return set_anon_super(sb, NULL);
1159}
1160
Eric W. Biedermand91ee872016-05-23 14:51:59 -05001161struct dentry *mount_ns(struct file_system_type *fs_type,
1162 int flags, void *data, void *ns, struct user_namespace *user_ns,
1163 int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001164{
1165 struct super_block *sb;
1166
Eric W. Biedermand91ee872016-05-23 14:51:59 -05001167 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1168 * over the namespace.
1169 */
David Howellse462ec52017-07-17 08:45:35 +01001170 if (!(flags & SB_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
Eric W. Biedermand91ee872016-05-23 14:51:59 -05001171 return ERR_PTR(-EPERM);
1172
Eric W. Biederman6e4eab52016-05-24 09:29:01 -05001173 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1174 user_ns, ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001175 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +04001176 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001177
1178 if (!sb->s_root) {
1179 int err;
David Howellse462ec52017-07-17 08:45:35 +01001180 err = fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001181 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001182 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +04001183 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001184 }
1185
David Howellse462ec52017-07-17 08:45:35 +01001186 sb->s_flags |= SB_ACTIVE;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001187 }
1188
Al Viroceefda62010-07-26 13:16:50 +04001189 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001190}
1191
Al Viroceefda62010-07-26 13:16:50 +04001192EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001193
Al Virocb50b342018-12-23 17:25:47 -05001194int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1195{
1196 return set_anon_super(sb, NULL);
1197}
1198EXPORT_SYMBOL(set_anon_super_fc);
1199
1200static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1201{
1202 return sb->s_fs_info == fc->s_fs_info;
1203}
1204
1205static int test_single_super(struct super_block *s, struct fs_context *fc)
1206{
1207 return 1;
1208}
1209
1210/**
1211 * vfs_get_super - Get a superblock with a search key set in s_fs_info.
1212 * @fc: The filesystem context holding the parameters
1213 * @keying: How to distinguish superblocks
1214 * @fill_super: Helper to initialise a new superblock
1215 *
1216 * Search for a superblock and create a new one if not found. The search
1217 * criterion is controlled by @keying. If the search fails, a new superblock
1218 * is created and @fill_super() is called to initialise it.
1219 *
1220 * @keying can take one of a number of values:
1221 *
1222 * (1) vfs_get_single_super - Only one superblock of this type may exist on the
1223 * system. This is typically used for special system filesystems.
1224 *
1225 * (2) vfs_get_keyed_super - Multiple superblocks may exist, but they must have
1226 * distinct keys (where the key is in s_fs_info). Searching for the same
1227 * key again will turn up the superblock for that key.
1228 *
1229 * (3) vfs_get_independent_super - Multiple superblocks may exist and are
1230 * unkeyed. Each call will get a new superblock.
1231 *
1232 * A permissions check is made by sget_fc() unless we're getting a superblock
1233 * for a kernel-internal mount or a submount.
1234 */
1235int vfs_get_super(struct fs_context *fc,
1236 enum vfs_get_super_keying keying,
1237 int (*fill_super)(struct super_block *sb,
1238 struct fs_context *fc))
1239{
1240 int (*test)(struct super_block *, struct fs_context *);
1241 struct super_block *sb;
1242
1243 switch (keying) {
1244 case vfs_get_single_super:
1245 test = test_single_super;
1246 break;
1247 case vfs_get_keyed_super:
1248 test = test_keyed_super;
1249 break;
1250 case vfs_get_independent_super:
1251 test = NULL;
1252 break;
1253 default:
1254 BUG();
1255 }
1256
1257 sb = sget_fc(fc, test, set_anon_super_fc);
1258 if (IS_ERR(sb))
1259 return PTR_ERR(sb);
1260
1261 if (!sb->s_root) {
1262 int err = fill_super(sb, fc);
1263 if (err) {
1264 deactivate_locked_super(sb);
1265 return err;
1266 }
1267
1268 sb->s_flags |= SB_ACTIVE;
1269 }
1270
1271 BUG_ON(fc->root);
1272 fc->root = dget(sb->s_root);
1273 return 0;
1274}
1275EXPORT_SYMBOL(vfs_get_super);
1276
David Howells93614012006-09-30 20:45:40 +02001277#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278static int set_bdev_super(struct super_block *s, void *data)
1279{
1280 s->s_bdev = data;
1281 s->s_dev = s->s_bdev->bd_dev;
Jan Kara13eec232017-04-12 12:24:29 +02001282 s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
Jens Axboe32a88aa2009-09-16 15:02:33 +02001283
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 return 0;
1285}
1286
1287static int test_bdev_super(struct super_block *s, void *data)
1288{
1289 return (void *)s->s_bdev == data;
1290}
1291
Al Viro152a0832010-07-25 00:46:55 +04001292struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +04001294 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295{
1296 struct block_device *bdev;
1297 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +01001298 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 int error = 0;
1300
David Howellse462ec52017-07-17 08:45:35 +01001301 if (!(flags & SB_RDONLY))
Al Viro30c40d22008-02-22 19:50:45 -05001302 mode |= FMODE_WRITE;
1303
Tejun Heod4d77622010-11-13 11:55:18 +01001304 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +04001306 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
1308 /*
1309 * once the super is inserted into the list by sget, s_umount
1310 * will protect the lockfs code from trying to start a snapshot
1311 * while we are mounting
1312 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001313 mutex_lock(&bdev->bd_fsfreeze_mutex);
1314 if (bdev->bd_fsfreeze_count > 0) {
1315 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1316 error = -EBUSY;
1317 goto error_bdev;
1318 }
David Howellse462ec52017-07-17 08:45:35 +01001319 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
David Howells9249e172012-06-25 12:55:37 +01001320 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001321 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -07001323 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325 if (s->s_root) {
David Howellse462ec52017-07-17 08:45:35 +01001326 if ((flags ^ s->s_flags) & SB_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001327 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001328 error = -EBUSY;
1329 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 }
David Howells454e2392006-06-23 02:02:57 -07001331
Tejun Heo4f331f02010-07-20 15:18:07 -07001332 /*
1333 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +01001334 * __invalidate_device(). blkdev_put() acquires
1335 * bd_mutex and can't be called under s_umount. Drop
1336 * s_umount temporarily. This is safe as we're
1337 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -07001338 */
1339 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +01001340 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -07001341 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 } else {
Al Viro30c40d22008-02-22 19:50:45 -05001343 s->s_mode = mode;
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001344 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
Pekka Enberge78c9a02006-01-08 01:03:39 -08001345 sb_set_blocksize(s, block_size(bdev));
David Howellse462ec52017-07-17 08:45:35 +01001346 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001348 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001349 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001350 }
David Howells454e2392006-06-23 02:02:57 -07001351
David Howellse462ec52017-07-17 08:45:35 +01001352 s->s_flags |= SB_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001353 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 }
1355
Al Viro152a0832010-07-25 00:46:55 +04001356 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
David Howells454e2392006-06-23 02:02:57 -07001358error_s:
1359 error = PTR_ERR(s);
1360error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001361 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001362error:
Al Viro152a0832010-07-25 00:46:55 +04001363 return ERR_PTR(error);
1364}
1365EXPORT_SYMBOL(mount_bdev);
1366
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367void kill_block_super(struct super_block *sb)
1368{
1369 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001370 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001372 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 generic_shutdown_super(sb);
1374 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001375 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001376 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377}
1378
1379EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001380#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
Al Viro3c26ff62010-07-25 11:46:36 +04001382struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001384 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385{
1386 int error;
David Howells9249e172012-06-25 12:55:37 +01001387 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
1389 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001390 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391
David Howellse462ec52017-07-17 08:45:35 +01001392 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001394 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001395 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 }
David Howellse462ec52017-07-17 08:45:35 +01001397 s->s_flags |= SB_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001398 return dget(s->s_root);
1399}
1400EXPORT_SYMBOL(mount_nodev);
1401
David Howells8d0347f2018-11-04 09:28:36 -05001402static int reconfigure_single(struct super_block *s,
1403 int flags, void *data)
1404{
1405 struct fs_context *fc;
1406 int ret;
1407
1408 /* The caller really need to be passing fc down into mount_single(),
1409 * then a chunk of this can be removed. [Bollocks -- AV]
1410 * Better yet, reconfiguration shouldn't happen, but rather the second
1411 * mount should be rejected if the parameters are not compatible.
1412 */
1413 fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1414 if (IS_ERR(fc))
1415 return PTR_ERR(fc);
1416
1417 ret = parse_monolithic_mount_data(fc, data);
1418 if (ret < 0)
1419 goto out;
1420
1421 ret = reconfigure_super(fc);
1422out:
1423 put_fs_context(fc);
1424 return ret;
1425}
1426
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427static int compare_single(struct super_block *s, void *p)
1428{
1429 return 1;
1430}
1431
Al Virofc14f2f2010-07-25 01:48:30 +04001432struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001434 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435{
1436 struct super_block *s;
1437 int error;
1438
David Howells9249e172012-06-25 12:55:37 +01001439 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001441 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 if (!s->s_root) {
David Howellse462ec52017-07-17 08:45:35 +01001443 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
David Howells8d0347f2018-11-04 09:28:36 -05001444 if (!error)
1445 s->s_flags |= SB_ACTIVE;
Kay Sievers9329d1be2009-12-18 21:18:15 +01001446 } else {
David Howells8d0347f2018-11-04 09:28:36 -05001447 error = reconfigure_single(s, flags, data);
1448 }
1449 if (unlikely(error)) {
1450 deactivate_locked_super(s);
1451 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 }
Al Virofc14f2f2010-07-25 01:48:30 +04001453 return dget(s->s_root);
1454}
1455EXPORT_SYMBOL(mount_single);
1456
David Howells9bc61ab2018-11-04 03:19:03 -05001457/**
1458 * vfs_get_tree - Get the mountable root
1459 * @fc: The superblock configuration context.
1460 *
1461 * The filesystem is invoked to get or create a superblock which can then later
1462 * be used for mounting. The filesystem places a pointer to the root to be
1463 * used for mounting in @fc->root.
1464 */
1465int vfs_get_tree(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466{
Al Viro9d412a42011-03-17 22:08:28 -04001467 struct super_block *sb;
David Howells9bc61ab2018-11-04 03:19:03 -05001468 int error;
Al Viro80893522010-02-05 09:30:46 -05001469
David Howells06a2ae52018-11-01 23:07:26 +00001470 if (fc->fs_type->fs_flags & FS_REQUIRES_DEV && !fc->source) {
1471 errorf(fc, "Filesystem requires source device");
Al Virof3a09c92018-12-23 18:55:56 -05001472 return -ENOENT;
David Howells06a2ae52018-11-01 23:07:26 +00001473 }
Al Virof3a09c92018-12-23 18:55:56 -05001474
1475 if (fc->root)
1476 return -EBUSY;
1477
1478 /* Get the mountable root in fc->root, with a ref on the root and a ref
1479 * on the superblock.
1480 */
1481 error = fc->ops->get_tree(fc);
David Howells9bc61ab2018-11-04 03:19:03 -05001482 if (error < 0)
1483 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Al Virof3a09c92018-12-23 18:55:56 -05001485 if (!fc->root) {
1486 pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1487 fc->fs_type->name);
1488 /* We don't know what the locking state of the superblock is -
1489 * if there is a superblock.
1490 */
1491 BUG();
1492 }
1493
David Howells9bc61ab2018-11-04 03:19:03 -05001494 sb = fc->root->d_sb;
Al Viro9d412a42011-03-17 22:08:28 -04001495 WARN_ON(!sb->s_bdi);
Dave Chinner79f546a2018-05-11 11:20:57 +10001496
Al Viroa0c9a8b2018-11-04 07:18:51 -05001497 if (fc->subtype && !sb->s_subtype) {
1498 sb->s_subtype = fc->subtype;
1499 fc->subtype = NULL;
1500 }
1501
Dave Chinner79f546a2018-05-11 11:20:57 +10001502 /*
1503 * Write barrier is for super_cache_count(). We place it before setting
1504 * SB_BORN as the data dependency between the two functions is the
1505 * superblock structure contents that we just set up, not the SB_BORN
1506 * flag.
1507 */
1508 smp_wmb();
David Howellse462ec52017-07-17 08:45:35 +01001509 sb->s_flags |= SB_BORN;
David Howells454e2392006-06-23 02:02:57 -07001510
David Howells9bc61ab2018-11-04 03:19:03 -05001511 error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
Al Viroc9ce29e2018-12-20 15:04:50 -05001512 if (unlikely(error)) {
1513 fc_drop_locked(fc);
1514 return error;
Al Viroa10d7c22018-12-05 11:58:35 -05001515 }
1516
Jeff Layton42cb56a2009-09-18 13:05:53 -07001517 /*
1518 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1519 * but s_maxbytes was an unsigned long long for many releases. Throw
1520 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001521 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001522 */
Al Viro9d412a42011-03-17 22:08:28 -04001523 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
David Howells9bc61ab2018-11-04 03:19:03 -05001524 "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001525
David Howells9bc61ab2018-11-04 03:19:03 -05001526 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527}
David Howells9bc61ab2018-11-04 03:19:03 -05001528EXPORT_SYMBOL(vfs_get_tree);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
Jan Kara5accdf82012-06-12 16:20:34 +02001530/*
Jan Karafca39342017-04-12 12:24:28 +02001531 * Setup private BDI for given superblock. It gets automatically cleaned up
1532 * in generic_shutdown_super().
1533 */
1534int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1535{
1536 struct backing_dev_info *bdi;
1537 int err;
1538 va_list args;
1539
1540 bdi = bdi_alloc(GFP_KERNEL);
1541 if (!bdi)
1542 return -ENOMEM;
1543
1544 bdi->name = sb->s_type->name;
1545
1546 va_start(args, fmt);
Jan Kara7c4cc302017-04-12 12:24:49 +02001547 err = bdi_register_va(bdi, fmt, args);
Jan Karafca39342017-04-12 12:24:28 +02001548 va_end(args);
1549 if (err) {
1550 bdi_put(bdi);
1551 return err;
1552 }
1553 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1554 sb->s_bdi = bdi;
Jan Karafca39342017-04-12 12:24:28 +02001555
1556 return 0;
1557}
1558EXPORT_SYMBOL(super_setup_bdi_name);
1559
1560/*
1561 * Setup private BDI for given superblock. I gets automatically cleaned up
1562 * in generic_shutdown_super().
1563 */
1564int super_setup_bdi(struct super_block *sb)
1565{
1566 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1567
1568 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1569 atomic_long_inc_return(&bdi_seq));
1570}
1571EXPORT_SYMBOL(super_setup_bdi);
1572
1573/*
Jan Kara5accdf82012-06-12 16:20:34 +02001574 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1575 * instead.
1576 */
1577void __sb_end_write(struct super_block *sb, int level)
1578{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001579 percpu_up_read(sb->s_writers.rw_sem + level-1);
Jan Kara5accdf82012-06-12 16:20:34 +02001580}
1581EXPORT_SYMBOL(__sb_end_write);
1582
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001583/*
1584 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1585 * instead.
1586 */
1587int __sb_start_write(struct super_block *sb, int level, bool wait)
1588{
1589 bool force_trylock = false;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001590 int ret = 1;
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001591
1592#ifdef CONFIG_LOCKDEP
1593 /*
1594 * We want lockdep to tell us about possible deadlocks with freezing
1595 * but it's it bit tricky to properly instrument it. Getting a freeze
1596 * protection works as getting a read lock but there are subtle
1597 * problems. XFS for example gets freeze protection on internal level
1598 * twice in some cases, which is OK only because we already hold a
1599 * freeze protection also on higher level. Due to these cases we have
1600 * to use wait == F (trylock mode) which must not fail.
1601 */
1602 if (wait) {
1603 int i;
1604
1605 for (i = 0; i < level - 1; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001606 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001607 force_trylock = true;
1608 break;
1609 }
1610 }
1611#endif
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001612 if (wait && !force_trylock)
1613 percpu_down_read(sb->s_writers.rw_sem + level-1);
1614 else
1615 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1616
Vincent Stehlé22224a12015-09-21 17:18:34 +02001617 WARN_ON(force_trylock && !ret);
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001618 return ret;
1619}
Jan Kara5accdf82012-06-12 16:20:34 +02001620EXPORT_SYMBOL(__sb_start_write);
1621
1622/**
1623 * sb_wait_write - wait until all writers to given file system finish
1624 * @sb: the super for which we wait
1625 * @level: type of writers we wait for (normal vs page fault)
1626 *
1627 * This function waits until there are no writers of given type to given file
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001628 * system.
Jan Kara5accdf82012-06-12 16:20:34 +02001629 */
1630static void sb_wait_write(struct super_block *sb, int level)
1631{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001632 percpu_down_write(sb->s_writers.rw_sem + level-1);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001633}
Jan Kara5accdf82012-06-12 16:20:34 +02001634
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001635/*
1636 * We are going to return to userspace and forget about these locks, the
1637 * ownership goes to the caller of thaw_super() which does unlock().
1638 */
1639static void lockdep_sb_freeze_release(struct super_block *sb)
1640{
1641 int level;
1642
1643 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1644 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1645}
1646
1647/*
1648 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1649 */
1650static void lockdep_sb_freeze_acquire(struct super_block *sb)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001651{
1652 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001653
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001654 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1655 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001656}
1657
1658static void sb_freeze_unlock(struct super_block *sb)
1659{
1660 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001661
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001662 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1663 percpu_up_write(sb->s_writers.rw_sem + level);
Jan Kara5accdf82012-06-12 16:20:34 +02001664}
1665
Josef Bacik18e9e512010-03-23 10:34:56 -04001666/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001667 * freeze_super - lock the filesystem and force it into a consistent state
1668 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001669 *
1670 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1671 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1672 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001673 *
1674 * During this function, sb->s_writers.frozen goes through these values:
1675 *
1676 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1677 *
1678 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1679 * writes should be blocked, though page faults are still allowed. We wait for
1680 * all writes to complete and then proceed to the next stage.
1681 *
1682 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1683 * but internal fs threads can still modify the filesystem (although they
1684 * should not dirty new pages or inodes), writeback can run etc. After waiting
1685 * for all running page faults we sync the filesystem which will clean all
1686 * dirty pages and inodes (no new dirty pages or inodes can be created when
1687 * sync is running).
1688 *
1689 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1690 * modification are blocked (e.g. XFS preallocation truncation on inode
1691 * reclaim). This is usually implemented by blocking new transactions for
1692 * filesystems that have them and need this additional guard. After all
1693 * internal writers are finished we call ->freeze_fs() to finish filesystem
1694 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1695 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1696 *
1697 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001698 */
1699int freeze_super(struct super_block *sb)
1700{
1701 int ret;
1702
1703 atomic_inc(&sb->s_active);
1704 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001705 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001706 deactivate_locked_super(sb);
1707 return -EBUSY;
1708 }
1709
David Howellse462ec52017-07-17 08:45:35 +01001710 if (!(sb->s_flags & SB_BORN)) {
Al Virodabe0dc2012-01-03 21:01:29 -05001711 up_write(&sb->s_umount);
1712 return 0; /* sic - it's "nothing to do" */
1713 }
1714
David Howellsbc98a422017-07-17 08:45:34 +01001715 if (sb_rdonly(sb)) {
Jan Kara5accdf82012-06-12 16:20:34 +02001716 /* Nothing to do really... */
1717 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001718 up_write(&sb->s_umount);
1719 return 0;
1720 }
1721
Jan Kara5accdf82012-06-12 16:20:34 +02001722 sb->s_writers.frozen = SB_FREEZE_WRITE;
Jan Kara5accdf82012-06-12 16:20:34 +02001723 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1724 up_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001725 sb_wait_write(sb, SB_FREEZE_WRITE);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001726 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001727
1728 /* Now we go and block page faults... */
Jan Kara5accdf82012-06-12 16:20:34 +02001729 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
Jan Kara5accdf82012-06-12 16:20:34 +02001730 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1731
1732 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001733 sync_filesystem(sb);
1734
Jan Kara5accdf82012-06-12 16:20:34 +02001735 /* Now wait for internal filesystem counter */
1736 sb->s_writers.frozen = SB_FREEZE_FS;
Jan Kara5accdf82012-06-12 16:20:34 +02001737 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001738
Josef Bacik18e9e512010-03-23 10:34:56 -04001739 if (sb->s_op->freeze_fs) {
1740 ret = sb->s_op->freeze_fs(sb);
1741 if (ret) {
1742 printk(KERN_ERR
1743 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001744 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001745 sb_freeze_unlock(sb);
Jan Kara5accdf82012-06-12 16:20:34 +02001746 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001747 deactivate_locked_super(sb);
1748 return ret;
1749 }
1750 }
Jan Kara5accdf82012-06-12 16:20:34 +02001751 /*
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001752 * For debugging purposes so that fs can warn if it sees write activity
1753 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
Jan Kara5accdf82012-06-12 16:20:34 +02001754 */
1755 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001756 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001757 up_write(&sb->s_umount);
1758 return 0;
1759}
1760EXPORT_SYMBOL(freeze_super);
1761
1762/**
1763 * thaw_super -- unlock filesystem
1764 * @sb: the super to thaw
1765 *
1766 * Unlocks the filesystem and marks it writeable again after freeze_super().
1767 */
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001768static int thaw_super_locked(struct super_block *sb)
Josef Bacik18e9e512010-03-23 10:34:56 -04001769{
1770 int error;
1771
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001772 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001773 up_write(&sb->s_umount);
1774 return -EINVAL;
1775 }
1776
David Howellsbc98a422017-07-17 08:45:34 +01001777 if (sb_rdonly(sb)) {
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001778 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001779 goto out;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001780 }
Josef Bacik18e9e512010-03-23 10:34:56 -04001781
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001782 lockdep_sb_freeze_acquire(sb);
1783
Josef Bacik18e9e512010-03-23 10:34:56 -04001784 if (sb->s_op->unfreeze_fs) {
1785 error = sb->s_op->unfreeze_fs(sb);
1786 if (error) {
1787 printk(KERN_ERR
1788 "VFS:Filesystem thaw failed\n");
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001789 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001790 up_write(&sb->s_umount);
1791 return error;
1792 }
1793 }
1794
Jan Kara5accdf82012-06-12 16:20:34 +02001795 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001796 sb_freeze_unlock(sb);
1797out:
Jan Kara5accdf82012-06-12 16:20:34 +02001798 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001799 deactivate_locked_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001800 return 0;
1801}
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001802
1803int thaw_super(struct super_block *sb)
1804{
1805 down_write(&sb->s_umount);
1806 return thaw_super_locked(sb);
1807}
Josef Bacik18e9e512010-03-23 10:34:56 -04001808EXPORT_SYMBOL(thaw_super);