blob: f5f96e52e0cdee8eeac0437d1aa5207ae68fbaa0 [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>
Al Viro6d59e7f2008-03-22 15:48:17 -040038#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +020040static int thaw_super_locked(struct super_block *sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Al Viro15d0f5e2015-02-02 10:07:59 -070042static LIST_HEAD(super_blocks);
43static DEFINE_SPINLOCK(sb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Jan Kara5accdf82012-06-12 16:20:34 +020045static char *sb_writers_name[SB_FREEZE_LEVELS] = {
46 "sb_writers",
47 "sb_pagefaults",
48 "sb_internal",
49};
50
Dave Chinnerb0d40c92011-07-08 14:14:42 +100051/*
52 * One thing we have to be careful of with a per-sb shrinker is that we don't
53 * drop the last active reference to the superblock from within the shrinker.
54 * If that happens we could trigger unregistering the shrinker from within the
55 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
56 * take a passive reference to the superblock to avoid this from occurring.
57 */
Dave Chinner0a234c62013-08-28 10:17:57 +100058static unsigned long super_cache_scan(struct shrinker *shrink,
59 struct shrink_control *sc)
Dave Chinnerb0d40c92011-07-08 14:14:42 +100060{
61 struct super_block *sb;
Dave Chinner0a234c62013-08-28 10:17:57 +100062 long fs_objects = 0;
63 long total_objects;
64 long freed = 0;
65 long dentries;
66 long inodes;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100067
68 sb = container_of(shrink, struct super_block, s_shrink);
69
70 /*
71 * Deadlock avoidance. We may hold various FS locks, and we don't want
72 * to recurse into the FS that called us in clear_inode() and friends..
73 */
Dave Chinner0a234c62013-08-28 10:17:57 +100074 if (!(sc->gfp_mask & __GFP_FS))
75 return SHRINK_STOP;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100076
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +030077 if (!trylock_super(sb))
Dave Chinner0a234c62013-08-28 10:17:57 +100078 return SHRINK_STOP;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100079
Al Virod0407902013-07-19 21:12:31 +040080 if (sb->s_op->nr_cached_objects)
Vladimir Davydov4101b622015-02-12 14:58:51 -080081 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100082
Vladimir Davydov503c3582015-02-12 14:58:47 -080083 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
84 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
Dave Chinnerf6041562013-08-28 10:18:00 +100085 total_objects = dentries + inodes + fs_objects + 1;
Tetsuo Handa475d0db2014-05-17 20:56:38 +090086 if (!total_objects)
87 total_objects = 1;
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100088
Dave Chinner0a234c62013-08-28 10:17:57 +100089 /* proportion the scan between the caches */
Dave Chinnerf6041562013-08-28 10:18:00 +100090 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +100091 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
Vladimir Davydov503c3582015-02-12 14:58:47 -080092 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
Dave Chinnerb0d40c92011-07-08 14:14:42 +100093
Dave Chinner0a234c62013-08-28 10:17:57 +100094 /*
95 * prune the dcache first as the icache is pinned by it, then
96 * prune the icache, followed by the filesystem specific caches
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -080097 *
98 * Ensure that we always scan at least one object - memcg kmem
99 * accounting uses this to fully empty the caches.
Dave Chinner0a234c62013-08-28 10:17:57 +1000100 */
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -0800101 sc->nr_to_scan = dentries + 1;
Vladimir Davydov503c3582015-02-12 14:58:47 -0800102 freed = prune_dcache_sb(sb, sc);
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -0800103 sc->nr_to_scan = inodes + 1;
Vladimir Davydov503c3582015-02-12 14:58:47 -0800104 freed += prune_icache_sb(sb, sc);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000105
Dave Chinner0a234c62013-08-28 10:17:57 +1000106 if (fs_objects) {
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -0800107 sc->nr_to_scan = fs_objects + 1;
Vladimir Davydov4101b622015-02-12 14:58:51 -0800108 freed += sb->s_op->free_cached_objects(sb, sc);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000109 }
110
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300111 up_read(&sb->s_umount);
Dave Chinner0a234c62013-08-28 10:17:57 +1000112 return freed;
113}
114
115static unsigned long super_cache_count(struct shrinker *shrink,
116 struct shrink_control *sc)
117{
118 struct super_block *sb;
119 long total_objects = 0;
120
121 sb = container_of(shrink, struct super_block, s_shrink);
122
Tim Chend23da152014-06-04 16:10:47 -0700123 /*
Dave Chinner79f546a2018-05-11 11:20:57 +1000124 * We don't call trylock_super() here as it is a scalability bottleneck,
125 * so we're exposed to partial setup state. The shrinker rwsem does not
126 * protect filesystem operations backing list_lru_shrink_count() or
127 * s_op->nr_cached_objects(). Counts can change between
128 * super_cache_count and super_cache_scan, so we really don't need locks
129 * here.
130 *
131 * However, if we are currently mounting the superblock, the underlying
132 * filesystem might be in a state of partial construction and hence it
133 * is dangerous to access it. trylock_super() uses a SB_BORN check to
134 * avoid this situation, so do the same here. The memory barrier is
135 * matched with the one in mount_fs() as we don't hold locks here.
Tim Chend23da152014-06-04 16:10:47 -0700136 */
Dave Chinner79f546a2018-05-11 11:20:57 +1000137 if (!(sb->s_flags & SB_BORN))
138 return 0;
139 smp_rmb();
140
Dave Chinner0a234c62013-08-28 10:17:57 +1000141 if (sb->s_op && sb->s_op->nr_cached_objects)
Vladimir Davydov4101b622015-02-12 14:58:51 -0800142 total_objects = sb->s_op->nr_cached_objects(sb, sc);
Dave Chinner0a234c62013-08-28 10:17:57 +1000143
Vladimir Davydov503c3582015-02-12 14:58:47 -0800144 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
145 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
Dave Chinner0a234c62013-08-28 10:17:57 +1000146
Glauber Costa55f841c2013-08-28 10:17:53 +1000147 total_objects = vfs_pressure_ratio(total_objects);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000148 return total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000149}
150
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200151static void destroy_super_work(struct work_struct *work)
152{
153 struct super_block *s = container_of(work, struct super_block,
154 destroy_work);
155 int i;
156
157 for (i = 0; i < SB_FREEZE_LEVELS; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200158 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200159 kfree(s);
160}
161
162static void destroy_super_rcu(struct rcu_head *head)
163{
164 struct super_block *s = container_of(head, struct super_block, rcu);
165 INIT_WORK(&s->destroy_work, destroy_super_work);
166 schedule_work(&s->destroy_work);
167}
168
Al Viro02008942017-10-11 13:48:55 -0400169/* Free a superblock that has never been seen by anyone */
170static void destroy_unused_super(struct super_block *s)
Jan Kara5accdf82012-06-12 16:20:34 +0200171{
Al Viro02008942017-10-11 13:48:55 -0400172 if (!s)
173 return;
174 up_write(&s->s_umount);
Al Viro7eb5e882013-10-01 15:09:58 -0400175 list_lru_destroy(&s->s_dentry_lru);
176 list_lru_destroy(&s->s_inode_lru);
Al Viro7eb5e882013-10-01 15:09:58 -0400177 security_sb_free(s);
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500178 put_user_ns(s->s_user_ns);
Al Viro7eb5e882013-10-01 15:09:58 -0400179 kfree(s->s_subtype);
Tetsuo Handa8e049442018-04-04 19:53:07 +0900180 free_prealloced_shrinker(&s->s_shrink);
Al Viro02008942017-10-11 13:48:55 -0400181 /* no delays needed */
182 destroy_super_work(&s->destroy_work);
Jan Kara5accdf82012-06-12 16:20:34 +0200183}
184
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185/**
186 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700187 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100188 * @flags: the mount flags
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500189 * @user_ns: User namespace for the super_block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 *
191 * Allocates and initializes a new &struct super_block. alloc_super()
192 * returns a pointer new superblock or %NULL if allocation had failed.
193 */
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500194static struct super_block *alloc_super(struct file_system_type *type, int flags,
195 struct user_namespace *user_ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800197 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700198 static const struct super_operations default_op;
Al Viro7eb5e882013-10-01 15:09:58 -0400199 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Al Viro7eb5e882013-10-01 15:09:58 -0400201 if (!s)
202 return NULL;
203
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800204 INIT_LIST_HEAD(&s->s_mounts);
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500205 s->s_user_ns = get_user_ns(user_ns);
Al Viroca0168e2017-12-05 09:32:25 -0500206 init_rwsem(&s->s_umount);
207 lockdep_set_class(&s->s_umount, &type->s_umount_key);
208 /*
209 * sget() can have s_umount recursion.
210 *
211 * When it cannot find a suitable sb, it allocates a new
212 * one (this one), and tries again to find a suitable old
213 * one.
214 *
215 * In case that succeeds, it will acquire the s_umount
216 * lock of the old one. Since these are clearly distrinct
217 * locks, and this object isn't exposed yet, there's no
218 * risk of deadlocks.
219 *
220 * Annotate this by putting this lock in a different
221 * subclass.
222 */
223 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800224
Al Viro7eb5e882013-10-01 15:09:58 -0400225 if (security_sb_alloc(s))
226 goto fail;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200227
Al Viro7eb5e882013-10-01 15:09:58 -0400228 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200229 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
230 sb_writers_name[i],
231 &type->s_writers_key[i]))
Al Viro7eb5e882013-10-01 15:09:58 -0400232 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 }
Al Viro7eb5e882013-10-01 15:09:58 -0400234 init_waitqueue_head(&s->s_writers.wait_unfrozen);
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100235 s->s_bdi = &noop_backing_dev_info;
Al Viro7eb5e882013-10-01 15:09:58 -0400236 s->s_flags = flags;
Eric W. Biedermancc50a072016-06-09 15:44:48 -0500237 if (s->s_user_ns != &init_user_ns)
Eric W. Biederman67690f92016-05-18 13:50:06 -0500238 s->s_iflags |= SB_I_NODEV;
Al Viro7eb5e882013-10-01 15:09:58 -0400239 INIT_HLIST_NODE(&s->s_instances);
NeilBrownf1ee6162017-12-21 09:45:40 +1100240 INIT_HLIST_BL_HEAD(&s->s_roots);
Dave Chinnere97fedb2015-03-04 13:40:00 -0500241 mutex_init(&s->s_sync_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400242 INIT_LIST_HEAD(&s->s_inodes);
Dave Chinner74278da2015-03-04 12:37:22 -0500243 spin_lock_init(&s->s_inode_list_lock);
Dave Chinner6c60d2b2016-07-26 15:21:50 -0700244 INIT_LIST_HEAD(&s->s_inodes_wb);
245 spin_lock_init(&s->s_inode_wblist_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400246
Al Viro7eb5e882013-10-01 15:09:58 -0400247 s->s_count = 1;
248 atomic_set(&s->s_active, 1);
249 mutex_init(&s->s_vfs_rename_mutex);
250 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
Jan Karabc8230e2017-06-08 14:39:48 +0200251 init_rwsem(&s->s_dquot.dqio_sem);
Al Viro7eb5e882013-10-01 15:09:58 -0400252 s->s_maxbytes = MAX_NON_LFS;
253 s->s_op = &default_op;
254 s->s_time_gran = 1000000000;
Vladimir Davydov3cb29d12015-04-14 15:46:48 -0700255 s->cleancache_poolid = CLEANCACHE_NO_POOL;
Al Viro7eb5e882013-10-01 15:09:58 -0400256
257 s->s_shrink.seeks = DEFAULT_SEEKS;
258 s->s_shrink.scan_objects = super_cache_scan;
259 s->s_shrink.count_objects = super_cache_count;
260 s->s_shrink.batch = 1024;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800261 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
Tetsuo Handa8e049442018-04-04 19:53:07 +0900262 if (prealloc_shrinker(&s->s_shrink))
263 goto fail;
Kirill Tkhaic92e8e12018-08-17 15:47:50 -0700264 if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
Kirill Tkhai2b3648a2018-08-17 15:47:45 -0700265 goto fail;
Kirill Tkhaic92e8e12018-08-17 15:47:50 -0700266 if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
Kirill Tkhai2b3648a2018-08-17 15:47:45 -0700267 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000269
Al Viro7eb5e882013-10-01 15:09:58 -0400270fail:
Al Viro02008942017-10-11 13:48:55 -0400271 destroy_unused_super(s);
Al Viro7eb5e882013-10-01 15:09:58 -0400272 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273}
274
275/* Superblock refcounting */
276
277/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400278 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 */
Al Viroc645b932017-10-11 13:48:55 -0400280static void __put_super(struct super_block *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
Al Viroc645b932017-10-11 13:48:55 -0400282 if (!--s->s_count) {
283 list_del_init(&s->s_list);
284 WARN_ON(s->s_dentry_lru.node);
285 WARN_ON(s->s_inode_lru.node);
286 WARN_ON(!list_empty(&s->s_mounts));
287 security_sb_free(s);
288 put_user_ns(s->s_user_ns);
289 kfree(s->s_subtype);
290 call_rcu(&s->rcu, destroy_super_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292}
293
294/**
295 * put_super - drop a temporary reference to superblock
296 * @sb: superblock in question
297 *
298 * Drops a temporary reference, frees superblock if there's no
299 * references left.
300 */
Al Virof47ec3f22011-11-21 21:15:42 -0500301static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
303 spin_lock(&sb_lock);
304 __put_super(sb);
305 spin_unlock(&sb_lock);
306}
307
308
309/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400310 * deactivate_locked_super - drop an active reference to superblock
311 * @s: superblock to deactivate
312 *
Masanari Iidabd7ced92016-02-02 22:31:06 +0900313 * Drops an active reference to superblock, converting it into a temporary
Al Viro1712ac82010-03-22 15:22:31 -0400314 * one if there is no other active references left. In that case we
315 * tell fs driver to shut it down and drop the temporary reference we
316 * had just acquired.
317 *
318 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400319 */
320void deactivate_locked_super(struct super_block *s)
321{
322 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400323 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400324 cleancache_invalidate_fs(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000325 unregister_shrinker(&s->s_shrink);
Dave Chinner28f2cd4f2014-06-04 16:10:46 -0700326 fs->kill_sb(s);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000327
Vladimir Davydovc0a5b562015-02-12 14:59:07 -0800328 /*
329 * Since list_lru_destroy() may sleep, we cannot call it from
330 * put_super(), where we hold the sb_lock. Therefore we destroy
331 * the lru lists right now.
332 */
333 list_lru_destroy(&s->s_dentry_lru);
334 list_lru_destroy(&s->s_inode_lru);
335
Al Viro74dbbdd2009-05-06 01:07:50 -0400336 put_filesystem(fs);
337 put_super(s);
338 } else {
339 up_write(&s->s_umount);
340 }
341}
342
343EXPORT_SYMBOL(deactivate_locked_super);
344
345/**
Al Viro1712ac82010-03-22 15:22:31 -0400346 * deactivate_super - drop an active reference to superblock
347 * @s: superblock to deactivate
348 *
349 * Variant of deactivate_locked_super(), except that superblock is *not*
350 * locked by caller. If we are going to drop the final active reference,
351 * lock will be acquired prior to that.
352 */
353void deactivate_super(struct super_block *s)
354{
355 if (!atomic_add_unless(&s->s_active, -1, 1)) {
356 down_write(&s->s_umount);
357 deactivate_locked_super(s);
358 }
359}
360
361EXPORT_SYMBOL(deactivate_super);
362
363/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 * grab_super - acquire an active reference
365 * @s: reference we are trying to make active
366 *
367 * Tries to acquire an active reference. grab_super() is used when we
368 * had just found a superblock in super_blocks or fs_type->fs_supers
369 * and want to turn it into a full-blown active reference. grab_super()
370 * is called with sb_lock held and drops it. Returns 1 in case of
371 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400372 * dying when grab_super() had been called). Note that this is only
373 * called for superblocks not in rundown mode (== ones still on ->fs_supers
374 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700376static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377{
378 s->s_count++;
379 spin_unlock(&sb_lock);
380 down_write(&s->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100381 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
Al Viroacfec9a2013-07-20 03:13:55 +0400382 put_super(s);
383 return 1;
384 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 up_write(&s->s_umount);
386 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 return 0;
388}
389
David Howellscf9a2ae2006-08-29 19:05:54 +0100390/*
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300391 * trylock_super - try to grab ->s_umount shared
Wanpeng Li331cbde2012-06-09 11:10:55 +0800392 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000393 *
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300394 * Try to prevent fs shutdown. This is used in places where we
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000395 * cannot take an active reference but we need to ensure that the
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300396 * filesystem is not shut down while we are working on it. It returns
397 * false if we cannot acquire s_umount or if we lose the race and
398 * filesystem already got into shutdown, and returns true with the s_umount
399 * lock held in read mode in case of success. On successful return,
400 * the caller must drop the s_umount lock when done.
401 *
402 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
403 * The reason why it's safe is that we are OK with doing trylock instead
404 * of down_read(). There's a couple of places that are OK with that, but
405 * it's very much not a general-purpose interface.
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000406 */
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300407bool trylock_super(struct super_block *sb)
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000408{
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000409 if (down_read_trylock(&sb->s_umount)) {
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300410 if (!hlist_unhashed(&sb->s_instances) &&
David Howellse462ec52017-07-17 08:45:35 +0100411 sb->s_root && (sb->s_flags & SB_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000412 return true;
413 up_read(&sb->s_umount);
414 }
415
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000416 return false;
417}
418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419/**
420 * generic_shutdown_super - common helper for ->kill_sb()
421 * @sb: superblock to kill
422 *
423 * generic_shutdown_super() does all fs-independent work on superblock
424 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
425 * that need destruction out of superblock, call generic_shutdown_super()
426 * and release aforementioned objects. Note: dentries and inodes _are_
427 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700428 *
429 * Upon calling this function, the filesystem may no longer alter or
430 * rearrange the set of dentries belonging to this super_block, nor may it
431 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 */
433void generic_shutdown_super(struct super_block *sb)
434{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800435 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
David Howellsc636ebd2006-10-11 01:22:19 -0700437 if (sb->s_root) {
438 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200439 sync_filesystem(sb);
David Howellse462ec52017-07-17 08:45:35 +0100440 sb->s_flags &= ~SB_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800441
Dave Chinner74278da2015-03-04 12:37:22 -0500442 fsnotify_unmount_inodes(sb);
Tejun Heoa1a0e232016-02-29 18:28:53 -0500443 cgroup_writeback_umount();
Al Viro63997e92010-10-25 20:49:35 -0400444
445 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200447 if (sb->s_dio_done_wq) {
448 destroy_workqueue(sb->s_dio_done_wq);
449 sb->s_dio_done_wq = NULL;
450 }
451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 if (sop->put_super)
453 sop->put_super(sb);
454
Al Viro63997e92010-10-25 20:49:35 -0400455 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800456 printk("VFS: Busy inodes after unmount of %s. "
457 "Self-destruct in 5 seconds. Have a nice day...\n",
458 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 }
461 spin_lock(&sb_lock);
462 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500463 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 spin_unlock(&sb_lock);
465 up_write(&sb->s_umount);
Jan Karac1844d52017-04-12 12:24:47 +0200466 if (sb->s_bdi != &noop_backing_dev_info) {
Jan Karafca39342017-04-12 12:24:28 +0200467 bdi_put(sb->s_bdi);
468 sb->s_bdi = &noop_backing_dev_info;
Jan Karafca39342017-04-12 12:24:28 +0200469 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470}
471
472EXPORT_SYMBOL(generic_shutdown_super);
473
474/**
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500475 * sget_userns - find or create a superblock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 * @type: filesystem type superblock should belong to
477 * @test: comparison callback
478 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100479 * @flags: mount flags
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500480 * @user_ns: User namespace for the super_block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 * @data: argument to each of them
482 */
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500483struct super_block *sget_userns(struct file_system_type *type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 int (*test)(struct super_block *,void *),
485 int (*set)(struct super_block *,void *),
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500486 int flags, struct user_namespace *user_ns,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 void *data)
488{
489 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700490 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 int err;
492
David Howellse462ec52017-07-17 08:45:35 +0100493 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) &&
Eric W. Biedermana001e742016-06-06 15:48:04 -0500494 !(type->fs_flags & FS_USERNS_MOUNT) &&
495 !capable(CAP_SYS_ADMIN))
496 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497retry:
498 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700499 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800500 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700501 if (!test(old, data))
502 continue;
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500503 if (user_ns != old->s_user_ns) {
504 spin_unlock(&sb_lock);
Al Viro02008942017-10-11 13:48:55 -0400505 destroy_unused_super(s);
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500506 return ERR_PTR(-EBUSY);
507 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700508 if (!grab_super(old))
509 goto retry;
Al Viro02008942017-10-11 13:48:55 -0400510 destroy_unused_super(s);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700511 return old;
512 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 }
514 if (!s) {
515 spin_unlock(&sb_lock);
David Howellse462ec52017-07-17 08:45:35 +0100516 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 if (!s)
518 return ERR_PTR(-ENOMEM);
519 goto retry;
520 }
David Howellsdd111b32017-07-04 17:25:09 +0100521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 err = set(s, data);
523 if (err) {
524 spin_unlock(&sb_lock);
Al Viro02008942017-10-11 13:48:55 -0400525 destroy_unused_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 return ERR_PTR(err);
527 }
528 s->s_type = type;
529 strlcpy(s->s_id, type->name, sizeof(s->s_id));
530 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500531 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 spin_unlock(&sb_lock);
533 get_filesystem(type);
Tetsuo Handa8e049442018-04-04 19:53:07 +0900534 register_shrinker_prepared(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 return s;
536}
537
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500538EXPORT_SYMBOL(sget_userns);
539
540/**
541 * sget - find or create a superblock
542 * @type: filesystem type superblock should belong to
543 * @test: comparison callback
544 * @set: setup callback
545 * @flags: mount flags
546 * @data: argument to each of them
547 */
548struct super_block *sget(struct file_system_type *type,
549 int (*test)(struct super_block *,void *),
550 int (*set)(struct super_block *,void *),
551 int flags,
552 void *data)
553{
554 struct user_namespace *user_ns = current_user_ns();
555
Eric W. Biederman93faccbb2017-02-01 06:06:16 +1300556 /* We don't yet pass the user namespace of the parent
557 * mount through to here so always use &init_user_ns
558 * until that changes.
559 */
David Howellse462ec52017-07-17 08:45:35 +0100560 if (flags & SB_SUBMOUNT)
Eric W. Biederman93faccbb2017-02-01 06:06:16 +1300561 user_ns = &init_user_ns;
562
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500563 /* Ensure the requestor has permissions over the target filesystem */
David Howellse462ec52017-07-17 08:45:35 +0100564 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500565 return ERR_PTR(-EPERM);
566
567 return sget_userns(type, test, set, flags, user_ns, data);
568}
569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570EXPORT_SYMBOL(sget);
571
572void drop_super(struct super_block *sb)
573{
574 up_read(&sb->s_umount);
575 put_super(sb);
576}
577
578EXPORT_SYMBOL(drop_super);
579
Jan Karaba6379f2016-11-23 12:53:00 +0100580void drop_super_exclusive(struct super_block *sb)
581{
582 up_write(&sb->s_umount);
583 put_super(sb);
584}
585EXPORT_SYMBOL(drop_super_exclusive);
586
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +0200587static void __iterate_supers(void (*f)(struct super_block *))
588{
589 struct super_block *sb, *p = NULL;
590
591 spin_lock(&sb_lock);
592 list_for_each_entry(sb, &super_blocks, s_list) {
593 if (hlist_unhashed(&sb->s_instances))
594 continue;
595 sb->s_count++;
596 spin_unlock(&sb_lock);
597
598 f(sb);
599
600 spin_lock(&sb_lock);
601 if (p)
602 __put_super(p);
603 p = sb;
604 }
605 if (p)
606 __put_super(p);
607 spin_unlock(&sb_lock);
608}
Christoph Hellwige5004752009-05-05 16:08:56 +0200609/**
Al Viro01a05b32010-03-23 06:06:58 -0400610 * iterate_supers - call function for all active superblocks
611 * @f: function to call
612 * @arg: argument to pass to it
613 *
614 * Scans the superblock list and calls given function, passing it
615 * locked superblock and given argument.
616 */
617void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
618{
Al Virodca33252010-07-25 02:31:46 +0400619 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400620
621 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400622 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500623 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400624 continue;
625 sb->s_count++;
626 spin_unlock(&sb_lock);
627
628 down_read(&sb->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100629 if (sb->s_root && (sb->s_flags & SB_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400630 f(sb, arg);
631 up_read(&sb->s_umount);
632
633 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400634 if (p)
635 __put_super(p);
636 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400637 }
Al Virodca33252010-07-25 02:31:46 +0400638 if (p)
639 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400640 spin_unlock(&sb_lock);
641}
642
643/**
Al Viro43e15cd2011-06-03 20:16:57 -0400644 * iterate_supers_type - call function for superblocks of given type
645 * @type: fs type
646 * @f: function to call
647 * @arg: argument to pass to it
648 *
649 * Scans the superblock list and calls given function, passing it
650 * locked superblock and given argument.
651 */
652void iterate_supers_type(struct file_system_type *type,
653 void (*f)(struct super_block *, void *), void *arg)
654{
655 struct super_block *sb, *p = NULL;
656
657 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800658 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400659 sb->s_count++;
660 spin_unlock(&sb_lock);
661
662 down_read(&sb->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100663 if (sb->s_root && (sb->s_flags & SB_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400664 f(sb, arg);
665 up_read(&sb->s_umount);
666
667 spin_lock(&sb_lock);
668 if (p)
669 __put_super(p);
670 p = sb;
671 }
672 if (p)
673 __put_super(p);
674 spin_unlock(&sb_lock);
675}
676
677EXPORT_SYMBOL(iterate_supers_type);
678
Jan Karaba6379f2016-11-23 12:53:00 +0100679static struct super_block *__get_super(struct block_device *bdev, bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700681 struct super_block *sb;
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 if (!bdev)
684 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700687rescan:
688 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500689 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400690 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700691 if (sb->s_bdev == bdev) {
692 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 spin_unlock(&sb_lock);
Jan Karaba6379f2016-11-23 12:53:00 +0100694 if (!excl)
695 down_read(&sb->s_umount);
696 else
697 down_write(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400698 /* still alive? */
David Howellse462ec52017-07-17 08:45:35 +0100699 if (sb->s_root && (sb->s_flags & SB_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700700 return sb;
Jan Karaba6379f2016-11-23 12:53:00 +0100701 if (!excl)
702 up_read(&sb->s_umount);
703 else
704 up_write(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400705 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700706 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400707 __put_super(sb);
708 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 }
710 }
711 spin_unlock(&sb_lock);
712 return NULL;
713}
714
Jan Karaba6379f2016-11-23 12:53:00 +0100715/**
716 * get_super - get the superblock of a device
717 * @bdev: device to get the superblock for
718 *
719 * Scans the superblock list and finds the superblock of the file system
720 * mounted on the device given. %NULL is returned if no match is found.
721 */
722struct super_block *get_super(struct block_device *bdev)
723{
724 return __get_super(bdev, false);
725}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200727
Jan Karaba6379f2016-11-23 12:53:00 +0100728static struct super_block *__get_super_thawed(struct block_device *bdev,
729 bool excl)
730{
731 while (1) {
732 struct super_block *s = __get_super(bdev, excl);
733 if (!s || s->s_writers.frozen == SB_UNFROZEN)
734 return s;
735 if (!excl)
736 up_read(&s->s_umount);
737 else
738 up_write(&s->s_umount);
739 wait_event(s->s_writers.wait_unfrozen,
740 s->s_writers.frozen == SB_UNFROZEN);
741 put_super(s);
742 }
743}
744
Christoph Hellwig45042302009-08-03 23:28:35 +0200745/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100746 * get_super_thawed - get thawed superblock of a device
747 * @bdev: device to get the superblock for
748 *
749 * Scans the superblock list and finds the superblock of the file system
750 * mounted on the device. The superblock is returned once it is thawed
751 * (or immediately if it was not frozen). %NULL is returned if no match
752 * is found.
753 */
754struct super_block *get_super_thawed(struct block_device *bdev)
755{
Jan Karaba6379f2016-11-23 12:53:00 +0100756 return __get_super_thawed(bdev, false);
Jan Kara6b6dc832012-02-10 11:03:00 +0100757}
758EXPORT_SYMBOL(get_super_thawed);
759
760/**
Jan Karaba6379f2016-11-23 12:53:00 +0100761 * get_super_exclusive_thawed - get thawed superblock of a device
762 * @bdev: device to get the superblock for
763 *
764 * Scans the superblock list and finds the superblock of the file system
765 * mounted on the device. The superblock is returned once it is thawed
766 * (or immediately if it was not frozen) and s_umount semaphore is held
767 * in exclusive mode. %NULL is returned if no match is found.
768 */
769struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
770{
771 return __get_super_thawed(bdev, true);
772}
773EXPORT_SYMBOL(get_super_exclusive_thawed);
774
775/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200776 * get_active_super - get an active reference to the superblock of a device
777 * @bdev: device to get the superblock for
778 *
779 * Scans the superblock list and finds the superblock of the file system
780 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400781 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200782 */
783struct super_block *get_active_super(struct block_device *bdev)
784{
785 struct super_block *sb;
786
787 if (!bdev)
788 return NULL;
789
Al Viro14945832010-03-22 20:15:33 -0400790restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200791 spin_lock(&sb_lock);
792 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500793 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400794 continue;
Al Viro14945832010-03-22 20:15:33 -0400795 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400796 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400797 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400798 up_write(&sb->s_umount);
799 return sb;
Al Viro14945832010-03-22 20:15:33 -0400800 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200801 }
802 spin_unlock(&sb_lock);
803 return NULL;
804}
David Howellsdd111b32017-07-04 17:25:09 +0100805
Al Virodf40c012010-03-22 20:23:25 -0400806struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700808 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700811rescan:
812 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500813 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400814 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700815 if (sb->s_dev == dev) {
816 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700818 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400819 /* still alive? */
David Howellse462ec52017-07-17 08:45:35 +0100820 if (sb->s_root && (sb->s_flags & SB_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700821 return sb;
822 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400823 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700824 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400825 __put_super(sb);
826 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 }
828 }
829 spin_unlock(&sb_lock);
830 return NULL;
831}
832
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 * do_remount_sb - asks filesystem to change mount options.
835 * @sb: superblock in question
David Howellse462ec52017-07-17 08:45:35 +0100836 * @sb_flags: revised superblock flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 * @data: the rest of options
838 * @force: whether or not to force the change
839 *
840 * Alters the mount options of a mounted file system.
841 */
David Howellse462ec52017-07-17 08:45:35 +0100842int do_remount_sb(struct super_block *sb, int sb_flags, void *data, int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843{
844 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400845 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200846
Jan Kara5accdf82012-06-12 16:20:34 +0200847 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200848 return -EBUSY;
849
David Howells93614012006-09-30 20:45:40 +0200850#ifdef CONFIG_BLOCK
David Howellse462ec52017-07-17 08:45:35 +0100851 if (!(sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200853#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200854
David Howellse462ec52017-07-17 08:45:35 +0100855 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
Nick Piggind208bbd2009-12-21 16:28:53 -0800856
Al Viro0aec09d2014-08-07 07:32:06 -0400857 if (remount_ro) {
Al Virofdab6842015-01-11 10:57:27 -0500858 if (!hlist_empty(&sb->s_pins)) {
Al Viro0aec09d2014-08-07 07:32:06 -0400859 up_write(&sb->s_umount);
Al Virofdab6842015-01-11 10:57:27 -0500860 group_pin_kill(&sb->s_pins);
Al Viro0aec09d2014-08-07 07:32:06 -0400861 down_write(&sb->s_umount);
862 if (!sb->s_root)
863 return 0;
864 if (sb->s_writers.frozen != SB_UNFROZEN)
865 return -EBUSY;
David Howellse462ec52017-07-17 08:45:35 +0100866 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
Al Viro0aec09d2014-08-07 07:32:06 -0400867 }
868 }
869 shrink_dcache_sb(sb);
870
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 /* If we are remounting RDONLY and current sb is read/write,
872 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800873 if (remount_ro) {
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100874 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400875 sb->s_readonly_remount = 1;
876 smp_wmb();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100877 } else {
878 retval = sb_prepare_remount_readonly(sb);
879 if (retval)
880 return retval;
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 }
883
884 if (sb->s_op->remount_fs) {
David Howellse462ec52017-07-17 08:45:35 +0100885 retval = sb->s_op->remount_fs(sb, &sb_flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200886 if (retval) {
887 if (!force)
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100888 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200889 /* If forced remount, go ahead despite any errors */
890 WARN(1, "forced remount of a %s fs returned %i\n",
891 sb->s_type->name, retval);
892 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 }
David Howellse462ec52017-07-17 08:45:35 +0100894 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (sb_flags & MS_RMT_MASK);
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100895 /* Needs to be ordered wrt mnt_is_readonly() */
896 smp_wmb();
897 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400898
Nick Piggind208bbd2009-12-21 16:28:53 -0800899 /*
900 * Some filesystems modify their metadata via some other path than the
901 * bdev buffer cache (eg. use a private mapping, or directories in
902 * pagecache, etc). Also file data modifications go via their own
903 * mappings. So If we try to mount readonly then copy the filesystem
904 * from bdev, we could get stale data, so invalidate it to give a best
905 * effort at coherency.
906 */
907 if (remount_ro && sb->s_bdev)
908 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 return 0;
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100910
911cancel_readonly:
912 sb->s_readonly_remount = 0;
913 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914}
915
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +0200916static void do_emergency_remount_callback(struct super_block *sb)
917{
918 down_write(&sb->s_umount);
919 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
920 !sb_rdonly(sb)) {
921 /*
922 * What lock protects sb->s_flags??
923 */
924 do_remount_sb(sb, SB_RDONLY, NULL, 1);
925 }
926 up_write(&sb->s_umount);
927}
928
Jens Axboea2a95372009-03-17 09:38:40 +0100929static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930{
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +0200931 __iterate_supers(do_emergency_remount_callback);
Jens Axboea2a95372009-03-17 09:38:40 +0100932 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 printk("Emergency Remount complete\n");
934}
935
936void emergency_remount(void)
937{
Jens Axboea2a95372009-03-17 09:38:40 +0100938 struct work_struct *work;
939
940 work = kmalloc(sizeof(*work), GFP_ATOMIC);
941 if (work) {
942 INIT_WORK(work, do_emergency_remount);
943 schedule_work(work);
944 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945}
946
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +0200947static void do_thaw_all_callback(struct super_block *sb)
948{
949 down_write(&sb->s_umount);
Al Viro1c18d2a2018-05-10 15:09:41 -0400950 if (sb->s_root && sb->s_flags & SB_BORN) {
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +0200951 emergency_thaw_bdev(sb);
952 thaw_super_locked(sb);
953 } else {
954 up_write(&sb->s_umount);
955 }
956}
957
958static void do_thaw_all(struct work_struct *work)
959{
960 __iterate_supers(do_thaw_all_callback);
961 kfree(work);
962 printk(KERN_WARNING "Emergency Thaw complete\n");
963}
964
965/**
966 * emergency_thaw_all -- forcibly thaw every frozen filesystem
967 *
968 * Used for emergency unfreeze of all filesystems via SysRq
969 */
970void emergency_thaw_all(void)
971{
972 struct work_struct *work;
973
974 work = kmalloc(sizeof(*work), GFP_ATOMIC);
975 if (work) {
976 INIT_WORK(work, do_thaw_all);
977 schedule_work(work);
978 }
979}
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981/*
982 * Unnamed block devices are dummy devices used by virtual
983 * filesystems which don't use real block-devices. -- jrs
984 */
985
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400986static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Thomas Bächlera2a4dc42014-04-03 21:55:37 +0200988/* Many userspace utilities consider an FSID of 0 invalid.
989 * Always return at least 1 from get_anon_bdev.
990 */
991static int unnamed_dev_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
Al Viro0ee5dc62011-07-07 15:44:25 -0400993int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
995 int dev;
996 int error;
997
998 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400999 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 return -ENOMEM;
1001 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -04001002 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -04001003 if (!error)
1004 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 spin_unlock(&unnamed_dev_lock);
1006 if (error == -EAGAIN)
1007 /* We raced and lost with another CPU. */
1008 goto retry;
1009 else if (error)
1010 return -EAGAIN;
1011
Wang YanQing1af95de2015-06-23 18:54:45 +08001012 if (dev >= (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +04001014 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -04001015 if (unnamed_dev_start > dev)
1016 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 spin_unlock(&unnamed_dev_lock);
1018 return -EMFILE;
1019 }
Al Viro0ee5dc62011-07-07 15:44:25 -04001020 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 return 0;
1022}
Al Viro0ee5dc62011-07-07 15:44:25 -04001023EXPORT_SYMBOL(get_anon_bdev);
1024
1025void free_anon_bdev(dev_t dev)
1026{
1027 int slot = MINOR(dev);
1028 spin_lock(&unnamed_dev_lock);
1029 ida_remove(&unnamed_dev_ida, slot);
1030 if (slot < unnamed_dev_start)
1031 unnamed_dev_start = slot;
1032 spin_unlock(&unnamed_dev_lock);
1033}
1034EXPORT_SYMBOL(free_anon_bdev);
1035
1036int set_anon_super(struct super_block *s, void *data)
1037{
Christoph Hellwigdf0ce262015-01-14 10:42:41 +01001038 return get_anon_bdev(&s->s_dev);
Al Viro0ee5dc62011-07-07 15:44:25 -04001039}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041EXPORT_SYMBOL(set_anon_super);
1042
1043void kill_anon_super(struct super_block *sb)
1044{
Al Viro0ee5dc62011-07-07 15:44:25 -04001045 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -04001047 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048}
1049
1050EXPORT_SYMBOL(kill_anon_super);
1051
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052void kill_litter_super(struct super_block *sb)
1053{
1054 if (sb->s_root)
1055 d_genocide(sb->s_root);
1056 kill_anon_super(sb);
1057}
1058
1059EXPORT_SYMBOL(kill_litter_super);
1060
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001061static int ns_test_super(struct super_block *sb, void *data)
1062{
1063 return sb->s_fs_info == data;
1064}
1065
1066static int ns_set_super(struct super_block *sb, void *data)
1067{
1068 sb->s_fs_info = data;
1069 return set_anon_super(sb, NULL);
1070}
1071
Eric W. Biedermand91ee872016-05-23 14:51:59 -05001072struct dentry *mount_ns(struct file_system_type *fs_type,
1073 int flags, void *data, void *ns, struct user_namespace *user_ns,
1074 int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001075{
1076 struct super_block *sb;
1077
Eric W. Biedermand91ee872016-05-23 14:51:59 -05001078 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1079 * over the namespace.
1080 */
David Howellse462ec52017-07-17 08:45:35 +01001081 if (!(flags & SB_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
Eric W. Biedermand91ee872016-05-23 14:51:59 -05001082 return ERR_PTR(-EPERM);
1083
Eric W. Biederman6e4eab52016-05-24 09:29:01 -05001084 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1085 user_ns, ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001086 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +04001087 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001088
1089 if (!sb->s_root) {
1090 int err;
David Howellse462ec52017-07-17 08:45:35 +01001091 err = fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001092 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001093 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +04001094 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001095 }
1096
David Howellse462ec52017-07-17 08:45:35 +01001097 sb->s_flags |= SB_ACTIVE;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001098 }
1099
Al Viroceefda62010-07-26 13:16:50 +04001100 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001101}
1102
Al Viroceefda62010-07-26 13:16:50 +04001103EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001104
David Howells93614012006-09-30 20:45:40 +02001105#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106static int set_bdev_super(struct super_block *s, void *data)
1107{
1108 s->s_bdev = data;
1109 s->s_dev = s->s_bdev->bd_dev;
Jan Kara13eec232017-04-12 12:24:29 +02001110 s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
Jens Axboe32a88aa2009-09-16 15:02:33 +02001111
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 return 0;
1113}
1114
1115static int test_bdev_super(struct super_block *s, void *data)
1116{
1117 return (void *)s->s_bdev == data;
1118}
1119
Al Viro152a0832010-07-25 00:46:55 +04001120struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +04001122 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123{
1124 struct block_device *bdev;
1125 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +01001126 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 int error = 0;
1128
David Howellse462ec52017-07-17 08:45:35 +01001129 if (!(flags & SB_RDONLY))
Al Viro30c40d22008-02-22 19:50:45 -05001130 mode |= FMODE_WRITE;
1131
Tejun Heod4d77622010-11-13 11:55:18 +01001132 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +04001134 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
1136 /*
1137 * once the super is inserted into the list by sget, s_umount
1138 * will protect the lockfs code from trying to start a snapshot
1139 * while we are mounting
1140 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001141 mutex_lock(&bdev->bd_fsfreeze_mutex);
1142 if (bdev->bd_fsfreeze_count > 0) {
1143 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1144 error = -EBUSY;
1145 goto error_bdev;
1146 }
David Howellse462ec52017-07-17 08:45:35 +01001147 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
David Howells9249e172012-06-25 12:55:37 +01001148 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001149 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -07001151 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
1153 if (s->s_root) {
David Howellse462ec52017-07-17 08:45:35 +01001154 if ((flags ^ s->s_flags) & SB_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001155 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001156 error = -EBUSY;
1157 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 }
David Howells454e2392006-06-23 02:02:57 -07001159
Tejun Heo4f331f02010-07-20 15:18:07 -07001160 /*
1161 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +01001162 * __invalidate_device(). blkdev_put() acquires
1163 * bd_mutex and can't be called under s_umount. Drop
1164 * s_umount temporarily. This is safe as we're
1165 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -07001166 */
1167 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +01001168 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -07001169 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 } else {
Al Viro30c40d22008-02-22 19:50:45 -05001171 s->s_mode = mode;
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001172 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
Pekka Enberge78c9a02006-01-08 01:03:39 -08001173 sb_set_blocksize(s, block_size(bdev));
David Howellse462ec52017-07-17 08:45:35 +01001174 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001176 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001177 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001178 }
David Howells454e2392006-06-23 02:02:57 -07001179
David Howellse462ec52017-07-17 08:45:35 +01001180 s->s_flags |= SB_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001181 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 }
1183
Al Viro152a0832010-07-25 00:46:55 +04001184 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
David Howells454e2392006-06-23 02:02:57 -07001186error_s:
1187 error = PTR_ERR(s);
1188error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001189 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001190error:
Al Viro152a0832010-07-25 00:46:55 +04001191 return ERR_PTR(error);
1192}
1193EXPORT_SYMBOL(mount_bdev);
1194
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195void kill_block_super(struct super_block *sb)
1196{
1197 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001198 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001200 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 generic_shutdown_super(sb);
1202 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001203 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001204 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205}
1206
1207EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001208#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209
Al Viro3c26ff62010-07-25 11:46:36 +04001210struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001212 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213{
1214 int error;
David Howells9249e172012-06-25 12:55:37 +01001215 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216
1217 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001218 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
David Howellse462ec52017-07-17 08:45:35 +01001220 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001222 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001223 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 }
David Howellse462ec52017-07-17 08:45:35 +01001225 s->s_flags |= SB_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001226 return dget(s->s_root);
1227}
1228EXPORT_SYMBOL(mount_nodev);
1229
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230static int compare_single(struct super_block *s, void *p)
1231{
1232 return 1;
1233}
1234
Al Virofc14f2f2010-07-25 01:48:30 +04001235struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001237 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238{
1239 struct super_block *s;
1240 int error;
1241
David Howells9249e172012-06-25 12:55:37 +01001242 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001244 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 if (!s->s_root) {
David Howellse462ec52017-07-17 08:45:35 +01001246 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001248 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001249 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 }
David Howellse462ec52017-07-17 08:45:35 +01001251 s->s_flags |= SB_ACTIVE;
Kay Sievers9329d1be2009-12-18 21:18:15 +01001252 } else {
1253 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 }
Al Virofc14f2f2010-07-25 01:48:30 +04001255 return dget(s->s_root);
1256}
1257EXPORT_SYMBOL(mount_single);
1258
Al Viro9d412a42011-03-17 22:08:28 -04001259struct dentry *
1260mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261{
Al Viroc96e41e2010-07-25 00:17:56 +04001262 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001263 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001265 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001266
Eric Parise0007522008-03-05 10:31:54 -05001267 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001269 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001270 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
Eric Parise0007522008-03-05 10:31:54 -05001272 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001273 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 }
1276
Al Viro1a102ff2011-03-16 09:07:58 -04001277 root = type->mount(type, flags, name, data);
1278 if (IS_ERR(root)) {
1279 error = PTR_ERR(root);
1280 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001281 }
Al Viro9d412a42011-03-17 22:08:28 -04001282 sb = root->d_sb;
1283 BUG_ON(!sb);
1284 WARN_ON(!sb->s_bdi);
Dave Chinner79f546a2018-05-11 11:20:57 +10001285
1286 /*
1287 * Write barrier is for super_cache_count(). We place it before setting
1288 * SB_BORN as the data dependency between the two functions is the
1289 * superblock structure contents that we just set up, not the SB_BORN
1290 * flag.
1291 */
1292 smp_wmb();
David Howellse462ec52017-07-17 08:45:35 +01001293 sb->s_flags |= SB_BORN;
David Howells454e2392006-06-23 02:02:57 -07001294
Al Viro9d412a42011-03-17 22:08:28 -04001295 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001296 if (error)
1297 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001298
Jeff Layton42cb56a2009-09-18 13:05:53 -07001299 /*
1300 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1301 * but s_maxbytes was an unsigned long long for many releases. Throw
1302 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001303 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001304 */
Al Viro9d412a42011-03-17 22:08:28 -04001305 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1306 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001307
Al Viro9d412a42011-03-17 22:08:28 -04001308 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001309 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001310 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001312 dput(root);
1313 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314out_free_secdata:
1315 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316out:
David Howells454e2392006-06-23 02:02:57 -07001317 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318}
1319
Jan Kara5accdf82012-06-12 16:20:34 +02001320/*
Jan Karafca39342017-04-12 12:24:28 +02001321 * Setup private BDI for given superblock. It gets automatically cleaned up
1322 * in generic_shutdown_super().
1323 */
1324int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1325{
1326 struct backing_dev_info *bdi;
1327 int err;
1328 va_list args;
1329
1330 bdi = bdi_alloc(GFP_KERNEL);
1331 if (!bdi)
1332 return -ENOMEM;
1333
1334 bdi->name = sb->s_type->name;
1335
1336 va_start(args, fmt);
Jan Kara7c4cc302017-04-12 12:24:49 +02001337 err = bdi_register_va(bdi, fmt, args);
Jan Karafca39342017-04-12 12:24:28 +02001338 va_end(args);
1339 if (err) {
1340 bdi_put(bdi);
1341 return err;
1342 }
1343 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1344 sb->s_bdi = bdi;
Jan Karafca39342017-04-12 12:24:28 +02001345
1346 return 0;
1347}
1348EXPORT_SYMBOL(super_setup_bdi_name);
1349
1350/*
1351 * Setup private BDI for given superblock. I gets automatically cleaned up
1352 * in generic_shutdown_super().
1353 */
1354int super_setup_bdi(struct super_block *sb)
1355{
1356 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1357
1358 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1359 atomic_long_inc_return(&bdi_seq));
1360}
1361EXPORT_SYMBOL(super_setup_bdi);
1362
1363/*
Jan Kara5accdf82012-06-12 16:20:34 +02001364 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1365 * instead.
1366 */
1367void __sb_end_write(struct super_block *sb, int level)
1368{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001369 percpu_up_read(sb->s_writers.rw_sem + level-1);
Jan Kara5accdf82012-06-12 16:20:34 +02001370}
1371EXPORT_SYMBOL(__sb_end_write);
1372
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001373/*
1374 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1375 * instead.
1376 */
1377int __sb_start_write(struct super_block *sb, int level, bool wait)
1378{
1379 bool force_trylock = false;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001380 int ret = 1;
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001381
1382#ifdef CONFIG_LOCKDEP
1383 /*
1384 * We want lockdep to tell us about possible deadlocks with freezing
1385 * but it's it bit tricky to properly instrument it. Getting a freeze
1386 * protection works as getting a read lock but there are subtle
1387 * problems. XFS for example gets freeze protection on internal level
1388 * twice in some cases, which is OK only because we already hold a
1389 * freeze protection also on higher level. Due to these cases we have
1390 * to use wait == F (trylock mode) which must not fail.
1391 */
1392 if (wait) {
1393 int i;
1394
1395 for (i = 0; i < level - 1; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001396 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001397 force_trylock = true;
1398 break;
1399 }
1400 }
1401#endif
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001402 if (wait && !force_trylock)
1403 percpu_down_read(sb->s_writers.rw_sem + level-1);
1404 else
1405 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1406
Vincent Stehlé22224a12015-09-21 17:18:34 +02001407 WARN_ON(force_trylock && !ret);
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001408 return ret;
1409}
Jan Kara5accdf82012-06-12 16:20:34 +02001410EXPORT_SYMBOL(__sb_start_write);
1411
1412/**
1413 * sb_wait_write - wait until all writers to given file system finish
1414 * @sb: the super for which we wait
1415 * @level: type of writers we wait for (normal vs page fault)
1416 *
1417 * This function waits until there are no writers of given type to given file
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001418 * system.
Jan Kara5accdf82012-06-12 16:20:34 +02001419 */
1420static void sb_wait_write(struct super_block *sb, int level)
1421{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001422 percpu_down_write(sb->s_writers.rw_sem + level-1);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001423}
Jan Kara5accdf82012-06-12 16:20:34 +02001424
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001425/*
1426 * We are going to return to userspace and forget about these locks, the
1427 * ownership goes to the caller of thaw_super() which does unlock().
1428 */
1429static void lockdep_sb_freeze_release(struct super_block *sb)
1430{
1431 int level;
1432
1433 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1434 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1435}
1436
1437/*
1438 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1439 */
1440static void lockdep_sb_freeze_acquire(struct super_block *sb)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001441{
1442 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001443
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001444 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1445 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001446}
1447
1448static void sb_freeze_unlock(struct super_block *sb)
1449{
1450 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001451
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001452 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1453 percpu_up_write(sb->s_writers.rw_sem + level);
Jan Kara5accdf82012-06-12 16:20:34 +02001454}
1455
Josef Bacik18e9e512010-03-23 10:34:56 -04001456/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001457 * freeze_super - lock the filesystem and force it into a consistent state
1458 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001459 *
1460 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1461 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1462 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001463 *
1464 * During this function, sb->s_writers.frozen goes through these values:
1465 *
1466 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1467 *
1468 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1469 * writes should be blocked, though page faults are still allowed. We wait for
1470 * all writes to complete and then proceed to the next stage.
1471 *
1472 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1473 * but internal fs threads can still modify the filesystem (although they
1474 * should not dirty new pages or inodes), writeback can run etc. After waiting
1475 * for all running page faults we sync the filesystem which will clean all
1476 * dirty pages and inodes (no new dirty pages or inodes can be created when
1477 * sync is running).
1478 *
1479 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1480 * modification are blocked (e.g. XFS preallocation truncation on inode
1481 * reclaim). This is usually implemented by blocking new transactions for
1482 * filesystems that have them and need this additional guard. After all
1483 * internal writers are finished we call ->freeze_fs() to finish filesystem
1484 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1485 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1486 *
1487 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001488 */
1489int freeze_super(struct super_block *sb)
1490{
1491 int ret;
1492
1493 atomic_inc(&sb->s_active);
1494 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001495 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001496 deactivate_locked_super(sb);
1497 return -EBUSY;
1498 }
1499
David Howellse462ec52017-07-17 08:45:35 +01001500 if (!(sb->s_flags & SB_BORN)) {
Al Virodabe0dc2012-01-03 21:01:29 -05001501 up_write(&sb->s_umount);
1502 return 0; /* sic - it's "nothing to do" */
1503 }
1504
David Howellsbc98a422017-07-17 08:45:34 +01001505 if (sb_rdonly(sb)) {
Jan Kara5accdf82012-06-12 16:20:34 +02001506 /* Nothing to do really... */
1507 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001508 up_write(&sb->s_umount);
1509 return 0;
1510 }
1511
Jan Kara5accdf82012-06-12 16:20:34 +02001512 sb->s_writers.frozen = SB_FREEZE_WRITE;
Jan Kara5accdf82012-06-12 16:20:34 +02001513 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1514 up_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001515 sb_wait_write(sb, SB_FREEZE_WRITE);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001516 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001517
1518 /* Now we go and block page faults... */
Jan Kara5accdf82012-06-12 16:20:34 +02001519 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
Jan Kara5accdf82012-06-12 16:20:34 +02001520 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1521
1522 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001523 sync_filesystem(sb);
1524
Jan Kara5accdf82012-06-12 16:20:34 +02001525 /* Now wait for internal filesystem counter */
1526 sb->s_writers.frozen = SB_FREEZE_FS;
Jan Kara5accdf82012-06-12 16:20:34 +02001527 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001528
Josef Bacik18e9e512010-03-23 10:34:56 -04001529 if (sb->s_op->freeze_fs) {
1530 ret = sb->s_op->freeze_fs(sb);
1531 if (ret) {
1532 printk(KERN_ERR
1533 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001534 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001535 sb_freeze_unlock(sb);
Jan Kara5accdf82012-06-12 16:20:34 +02001536 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001537 deactivate_locked_super(sb);
1538 return ret;
1539 }
1540 }
Jan Kara5accdf82012-06-12 16:20:34 +02001541 /*
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001542 * For debugging purposes so that fs can warn if it sees write activity
1543 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
Jan Kara5accdf82012-06-12 16:20:34 +02001544 */
1545 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001546 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001547 up_write(&sb->s_umount);
1548 return 0;
1549}
1550EXPORT_SYMBOL(freeze_super);
1551
1552/**
1553 * thaw_super -- unlock filesystem
1554 * @sb: the super to thaw
1555 *
1556 * Unlocks the filesystem and marks it writeable again after freeze_super().
1557 */
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001558static int thaw_super_locked(struct super_block *sb)
Josef Bacik18e9e512010-03-23 10:34:56 -04001559{
1560 int error;
1561
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001562 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001563 up_write(&sb->s_umount);
1564 return -EINVAL;
1565 }
1566
David Howellsbc98a422017-07-17 08:45:34 +01001567 if (sb_rdonly(sb)) {
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001568 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001569 goto out;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001570 }
Josef Bacik18e9e512010-03-23 10:34:56 -04001571
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001572 lockdep_sb_freeze_acquire(sb);
1573
Josef Bacik18e9e512010-03-23 10:34:56 -04001574 if (sb->s_op->unfreeze_fs) {
1575 error = sb->s_op->unfreeze_fs(sb);
1576 if (error) {
1577 printk(KERN_ERR
1578 "VFS:Filesystem thaw failed\n");
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001579 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001580 up_write(&sb->s_umount);
1581 return error;
1582 }
1583 }
1584
Jan Kara5accdf82012-06-12 16:20:34 +02001585 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001586 sb_freeze_unlock(sb);
1587out:
Jan Kara5accdf82012-06-12 16:20:34 +02001588 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001589 deactivate_locked_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001590 return 0;
1591}
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001592
1593int thaw_super(struct super_block *sb)
1594{
1595 down_write(&sb->s_umount);
1596 return thaw_super_locked(sb);
1597}
Josef Bacik18e9e512010-03-23 10:34:56 -04001598EXPORT_SYMBOL(thaw_super);