blob: f3a8c008e16430be1ba08d9f575c1df9faeb0f0a [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
Kirill Tkhai9b996462018-08-17 15:48:21 -0700147 if (!total_objects)
148 return SHRINK_EMPTY;
149
Glauber Costa55f841c2013-08-28 10:17:53 +1000150 total_objects = vfs_pressure_ratio(total_objects);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000151 return total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000152}
153
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200154static void destroy_super_work(struct work_struct *work)
155{
156 struct super_block *s = container_of(work, struct super_block,
157 destroy_work);
158 int i;
159
160 for (i = 0; i < SB_FREEZE_LEVELS; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200161 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200162 kfree(s);
163}
164
165static void destroy_super_rcu(struct rcu_head *head)
166{
167 struct super_block *s = container_of(head, struct super_block, rcu);
168 INIT_WORK(&s->destroy_work, destroy_super_work);
169 schedule_work(&s->destroy_work);
170}
171
Al Viro02008942017-10-11 13:48:55 -0400172/* Free a superblock that has never been seen by anyone */
173static void destroy_unused_super(struct super_block *s)
Jan Kara5accdf82012-06-12 16:20:34 +0200174{
Al Viro02008942017-10-11 13:48:55 -0400175 if (!s)
176 return;
177 up_write(&s->s_umount);
Al Viro7eb5e882013-10-01 15:09:58 -0400178 list_lru_destroy(&s->s_dentry_lru);
179 list_lru_destroy(&s->s_inode_lru);
Al Viro7eb5e882013-10-01 15:09:58 -0400180 security_sb_free(s);
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500181 put_user_ns(s->s_user_ns);
Al Viro7eb5e882013-10-01 15:09:58 -0400182 kfree(s->s_subtype);
Tetsuo Handa8e049442018-04-04 19:53:07 +0900183 free_prealloced_shrinker(&s->s_shrink);
Al Viro02008942017-10-11 13:48:55 -0400184 /* no delays needed */
185 destroy_super_work(&s->destroy_work);
Jan Kara5accdf82012-06-12 16:20:34 +0200186}
187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188/**
189 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700190 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100191 * @flags: the mount flags
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500192 * @user_ns: User namespace for the super_block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 *
194 * Allocates and initializes a new &struct super_block. alloc_super()
195 * returns a pointer new superblock or %NULL if allocation had failed.
196 */
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500197static struct super_block *alloc_super(struct file_system_type *type, int flags,
198 struct user_namespace *user_ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800200 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700201 static const struct super_operations default_op;
Al Viro7eb5e882013-10-01 15:09:58 -0400202 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
Al Viro7eb5e882013-10-01 15:09:58 -0400204 if (!s)
205 return NULL;
206
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800207 INIT_LIST_HEAD(&s->s_mounts);
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500208 s->s_user_ns = get_user_ns(user_ns);
Al Viroca0168e2017-12-05 09:32:25 -0500209 init_rwsem(&s->s_umount);
210 lockdep_set_class(&s->s_umount, &type->s_umount_key);
211 /*
212 * sget() can have s_umount recursion.
213 *
214 * When it cannot find a suitable sb, it allocates a new
215 * one (this one), and tries again to find a suitable old
216 * one.
217 *
218 * In case that succeeds, it will acquire the s_umount
219 * lock of the old one. Since these are clearly distrinct
220 * locks, and this object isn't exposed yet, there's no
221 * risk of deadlocks.
222 *
223 * Annotate this by putting this lock in a different
224 * subclass.
225 */
226 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800227
Al Viro7eb5e882013-10-01 15:09:58 -0400228 if (security_sb_alloc(s))
229 goto fail;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200230
Al Viro7eb5e882013-10-01 15:09:58 -0400231 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200232 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
233 sb_writers_name[i],
234 &type->s_writers_key[i]))
Al Viro7eb5e882013-10-01 15:09:58 -0400235 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 }
Al Viro7eb5e882013-10-01 15:09:58 -0400237 init_waitqueue_head(&s->s_writers.wait_unfrozen);
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100238 s->s_bdi = &noop_backing_dev_info;
Al Viro7eb5e882013-10-01 15:09:58 -0400239 s->s_flags = flags;
Eric W. Biedermancc50a072016-06-09 15:44:48 -0500240 if (s->s_user_ns != &init_user_ns)
Eric W. Biederman67690f92016-05-18 13:50:06 -0500241 s->s_iflags |= SB_I_NODEV;
Al Viro7eb5e882013-10-01 15:09:58 -0400242 INIT_HLIST_NODE(&s->s_instances);
NeilBrownf1ee6162017-12-21 09:45:40 +1100243 INIT_HLIST_BL_HEAD(&s->s_roots);
Dave Chinnere97fedb2015-03-04 13:40:00 -0500244 mutex_init(&s->s_sync_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400245 INIT_LIST_HEAD(&s->s_inodes);
Dave Chinner74278da2015-03-04 12:37:22 -0500246 spin_lock_init(&s->s_inode_list_lock);
Dave Chinner6c60d2b2016-07-26 15:21:50 -0700247 INIT_LIST_HEAD(&s->s_inodes_wb);
248 spin_lock_init(&s->s_inode_wblist_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400249
Al Viro7eb5e882013-10-01 15:09:58 -0400250 s->s_count = 1;
251 atomic_set(&s->s_active, 1);
252 mutex_init(&s->s_vfs_rename_mutex);
253 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
Jan Karabc8230e2017-06-08 14:39:48 +0200254 init_rwsem(&s->s_dquot.dqio_sem);
Al Viro7eb5e882013-10-01 15:09:58 -0400255 s->s_maxbytes = MAX_NON_LFS;
256 s->s_op = &default_op;
257 s->s_time_gran = 1000000000;
Vladimir Davydov3cb29d12015-04-14 15:46:48 -0700258 s->cleancache_poolid = CLEANCACHE_NO_POOL;
Al Viro7eb5e882013-10-01 15:09:58 -0400259
260 s->s_shrink.seeks = DEFAULT_SEEKS;
261 s->s_shrink.scan_objects = super_cache_scan;
262 s->s_shrink.count_objects = super_cache_count;
263 s->s_shrink.batch = 1024;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800264 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
Tetsuo Handa8e049442018-04-04 19:53:07 +0900265 if (prealloc_shrinker(&s->s_shrink))
266 goto fail;
Kirill Tkhaic92e8e12018-08-17 15:47:50 -0700267 if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
Kirill Tkhai2b3648a2018-08-17 15:47:45 -0700268 goto fail;
Kirill Tkhaic92e8e12018-08-17 15:47:50 -0700269 if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
Kirill Tkhai2b3648a2018-08-17 15:47:45 -0700270 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000272
Al Viro7eb5e882013-10-01 15:09:58 -0400273fail:
Al Viro02008942017-10-11 13:48:55 -0400274 destroy_unused_super(s);
Al Viro7eb5e882013-10-01 15:09:58 -0400275 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277
278/* Superblock refcounting */
279
280/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400281 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 */
Al Viroc645b932017-10-11 13:48:55 -0400283static void __put_super(struct super_block *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284{
Al Viroc645b932017-10-11 13:48:55 -0400285 if (!--s->s_count) {
286 list_del_init(&s->s_list);
287 WARN_ON(s->s_dentry_lru.node);
288 WARN_ON(s->s_inode_lru.node);
289 WARN_ON(!list_empty(&s->s_mounts));
290 security_sb_free(s);
291 put_user_ns(s->s_user_ns);
292 kfree(s->s_subtype);
293 call_rcu(&s->rcu, destroy_super_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
296
297/**
298 * put_super - drop a temporary reference to superblock
299 * @sb: superblock in question
300 *
301 * Drops a temporary reference, frees superblock if there's no
302 * references left.
303 */
Al Virof47ec3f2011-11-21 21:15:42 -0500304static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
306 spin_lock(&sb_lock);
307 __put_super(sb);
308 spin_unlock(&sb_lock);
309}
310
311
312/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400313 * deactivate_locked_super - drop an active reference to superblock
314 * @s: superblock to deactivate
315 *
Masanari Iidabd7ced92016-02-02 22:31:06 +0900316 * Drops an active reference to superblock, converting it into a temporary
Al Viro1712ac82010-03-22 15:22:31 -0400317 * one if there is no other active references left. In that case we
318 * tell fs driver to shut it down and drop the temporary reference we
319 * had just acquired.
320 *
321 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400322 */
323void deactivate_locked_super(struct super_block *s)
324{
325 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400326 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400327 cleancache_invalidate_fs(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000328 unregister_shrinker(&s->s_shrink);
Dave Chinner28f2cd4f2014-06-04 16:10:46 -0700329 fs->kill_sb(s);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000330
Vladimir Davydovc0a5b562015-02-12 14:59:07 -0800331 /*
332 * Since list_lru_destroy() may sleep, we cannot call it from
333 * put_super(), where we hold the sb_lock. Therefore we destroy
334 * the lru lists right now.
335 */
336 list_lru_destroy(&s->s_dentry_lru);
337 list_lru_destroy(&s->s_inode_lru);
338
Al Viro74dbbdd2009-05-06 01:07:50 -0400339 put_filesystem(fs);
340 put_super(s);
341 } else {
342 up_write(&s->s_umount);
343 }
344}
345
346EXPORT_SYMBOL(deactivate_locked_super);
347
348/**
Al Viro1712ac82010-03-22 15:22:31 -0400349 * deactivate_super - drop an active reference to superblock
350 * @s: superblock to deactivate
351 *
352 * Variant of deactivate_locked_super(), except that superblock is *not*
353 * locked by caller. If we are going to drop the final active reference,
354 * lock will be acquired prior to that.
355 */
356void deactivate_super(struct super_block *s)
357{
358 if (!atomic_add_unless(&s->s_active, -1, 1)) {
359 down_write(&s->s_umount);
360 deactivate_locked_super(s);
361 }
362}
363
364EXPORT_SYMBOL(deactivate_super);
365
366/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 * grab_super - acquire an active reference
368 * @s: reference we are trying to make active
369 *
370 * Tries to acquire an active reference. grab_super() is used when we
371 * had just found a superblock in super_blocks or fs_type->fs_supers
372 * and want to turn it into a full-blown active reference. grab_super()
373 * is called with sb_lock held and drops it. Returns 1 in case of
374 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400375 * dying when grab_super() had been called). Note that this is only
376 * called for superblocks not in rundown mode (== ones still on ->fs_supers
377 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700379static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380{
381 s->s_count++;
382 spin_unlock(&sb_lock);
383 down_write(&s->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100384 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
Al Viroacfec9a2013-07-20 03:13:55 +0400385 put_super(s);
386 return 1;
387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 up_write(&s->s_umount);
389 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 return 0;
391}
392
David Howellscf9a2ae2006-08-29 19:05:54 +0100393/*
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300394 * trylock_super - try to grab ->s_umount shared
Wanpeng Li331cbde2012-06-09 11:10:55 +0800395 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000396 *
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300397 * Try to prevent fs shutdown. This is used in places where we
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000398 * cannot take an active reference but we need to ensure that the
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300399 * filesystem is not shut down while we are working on it. It returns
400 * false if we cannot acquire s_umount or if we lose the race and
401 * filesystem already got into shutdown, and returns true with the s_umount
402 * lock held in read mode in case of success. On successful return,
403 * the caller must drop the s_umount lock when done.
404 *
405 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
406 * The reason why it's safe is that we are OK with doing trylock instead
407 * of down_read(). There's a couple of places that are OK with that, but
408 * it's very much not a general-purpose interface.
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000409 */
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300410bool trylock_super(struct super_block *sb)
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000411{
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000412 if (down_read_trylock(&sb->s_umount)) {
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300413 if (!hlist_unhashed(&sb->s_instances) &&
David Howellse462ec52017-07-17 08:45:35 +0100414 sb->s_root && (sb->s_flags & SB_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000415 return true;
416 up_read(&sb->s_umount);
417 }
418
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000419 return false;
420}
421
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422/**
423 * generic_shutdown_super - common helper for ->kill_sb()
424 * @sb: superblock to kill
425 *
426 * generic_shutdown_super() does all fs-independent work on superblock
427 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
428 * that need destruction out of superblock, call generic_shutdown_super()
429 * and release aforementioned objects. Note: dentries and inodes _are_
430 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700431 *
432 * Upon calling this function, the filesystem may no longer alter or
433 * rearrange the set of dentries belonging to this super_block, nor may it
434 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 */
436void generic_shutdown_super(struct super_block *sb)
437{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800438 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
David Howellsc636ebd2006-10-11 01:22:19 -0700440 if (sb->s_root) {
441 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200442 sync_filesystem(sb);
David Howellse462ec52017-07-17 08:45:35 +0100443 sb->s_flags &= ~SB_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800444
Dave Chinner74278da2015-03-04 12:37:22 -0500445 fsnotify_unmount_inodes(sb);
Tejun Heoa1a0e232016-02-29 18:28:53 -0500446 cgroup_writeback_umount();
Al Viro63997e92010-10-25 20:49:35 -0400447
448 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200450 if (sb->s_dio_done_wq) {
451 destroy_workqueue(sb->s_dio_done_wq);
452 sb->s_dio_done_wq = NULL;
453 }
454
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 if (sop->put_super)
456 sop->put_super(sb);
457
Al Viro63997e92010-10-25 20:49:35 -0400458 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800459 printk("VFS: Busy inodes after unmount of %s. "
460 "Self-destruct in 5 seconds. Have a nice day...\n",
461 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 }
464 spin_lock(&sb_lock);
465 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500466 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 spin_unlock(&sb_lock);
468 up_write(&sb->s_umount);
Jan Karac1844d52017-04-12 12:24:47 +0200469 if (sb->s_bdi != &noop_backing_dev_info) {
Jan Karafca39342017-04-12 12:24:28 +0200470 bdi_put(sb->s_bdi);
471 sb->s_bdi = &noop_backing_dev_info;
Jan Karafca39342017-04-12 12:24:28 +0200472 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473}
474
475EXPORT_SYMBOL(generic_shutdown_super);
476
477/**
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500478 * sget_userns - find or create a superblock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 * @type: filesystem type superblock should belong to
480 * @test: comparison callback
481 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100482 * @flags: mount flags
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500483 * @user_ns: User namespace for the super_block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 * @data: argument to each of them
485 */
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500486struct super_block *sget_userns(struct file_system_type *type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 int (*test)(struct super_block *,void *),
488 int (*set)(struct super_block *,void *),
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500489 int flags, struct user_namespace *user_ns,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 void *data)
491{
492 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700493 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 int err;
495
David Howellse462ec52017-07-17 08:45:35 +0100496 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) &&
Eric W. Biedermana001e742016-06-06 15:48:04 -0500497 !(type->fs_flags & FS_USERNS_MOUNT) &&
498 !capable(CAP_SYS_ADMIN))
499 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500retry:
501 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700502 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800503 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700504 if (!test(old, data))
505 continue;
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500506 if (user_ns != old->s_user_ns) {
507 spin_unlock(&sb_lock);
Al Viro02008942017-10-11 13:48:55 -0400508 destroy_unused_super(s);
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500509 return ERR_PTR(-EBUSY);
510 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700511 if (!grab_super(old))
512 goto retry;
Al Viro02008942017-10-11 13:48:55 -0400513 destroy_unused_super(s);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700514 return old;
515 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 }
517 if (!s) {
518 spin_unlock(&sb_lock);
David Howellse462ec52017-07-17 08:45:35 +0100519 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 if (!s)
521 return ERR_PTR(-ENOMEM);
522 goto retry;
523 }
David Howellsdd111b32017-07-04 17:25:09 +0100524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 err = set(s, data);
526 if (err) {
527 spin_unlock(&sb_lock);
Al Viro02008942017-10-11 13:48:55 -0400528 destroy_unused_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 return ERR_PTR(err);
530 }
531 s->s_type = type;
532 strlcpy(s->s_id, type->name, sizeof(s->s_id));
533 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500534 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 spin_unlock(&sb_lock);
536 get_filesystem(type);
Tetsuo Handa8e049442018-04-04 19:53:07 +0900537 register_shrinker_prepared(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 return s;
539}
540
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500541EXPORT_SYMBOL(sget_userns);
542
543/**
544 * sget - find or create a superblock
545 * @type: filesystem type superblock should belong to
546 * @test: comparison callback
547 * @set: setup callback
548 * @flags: mount flags
549 * @data: argument to each of them
550 */
551struct super_block *sget(struct file_system_type *type,
552 int (*test)(struct super_block *,void *),
553 int (*set)(struct super_block *,void *),
554 int flags,
555 void *data)
556{
557 struct user_namespace *user_ns = current_user_ns();
558
Eric W. Biederman93faccbb2017-02-01 06:06:16 +1300559 /* We don't yet pass the user namespace of the parent
560 * mount through to here so always use &init_user_ns
561 * until that changes.
562 */
David Howellse462ec52017-07-17 08:45:35 +0100563 if (flags & SB_SUBMOUNT)
Eric W. Biederman93faccbb2017-02-01 06:06:16 +1300564 user_ns = &init_user_ns;
565
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500566 /* Ensure the requestor has permissions over the target filesystem */
David Howellse462ec52017-07-17 08:45:35 +0100567 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500568 return ERR_PTR(-EPERM);
569
570 return sget_userns(type, test, set, flags, user_ns, data);
571}
572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573EXPORT_SYMBOL(sget);
574
575void drop_super(struct super_block *sb)
576{
577 up_read(&sb->s_umount);
578 put_super(sb);
579}
580
581EXPORT_SYMBOL(drop_super);
582
Jan Karaba6379f2016-11-23 12:53:00 +0100583void drop_super_exclusive(struct super_block *sb)
584{
585 up_write(&sb->s_umount);
586 put_super(sb);
587}
588EXPORT_SYMBOL(drop_super_exclusive);
589
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +0200590static void __iterate_supers(void (*f)(struct super_block *))
591{
592 struct super_block *sb, *p = NULL;
593
594 spin_lock(&sb_lock);
595 list_for_each_entry(sb, &super_blocks, s_list) {
596 if (hlist_unhashed(&sb->s_instances))
597 continue;
598 sb->s_count++;
599 spin_unlock(&sb_lock);
600
601 f(sb);
602
603 spin_lock(&sb_lock);
604 if (p)
605 __put_super(p);
606 p = sb;
607 }
608 if (p)
609 __put_super(p);
610 spin_unlock(&sb_lock);
611}
Christoph Hellwige5004752009-05-05 16:08:56 +0200612/**
Al Viro01a05b32010-03-23 06:06:58 -0400613 * iterate_supers - call function for all active superblocks
614 * @f: function to call
615 * @arg: argument to pass to it
616 *
617 * Scans the superblock list and calls given function, passing it
618 * locked superblock and given argument.
619 */
620void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
621{
Al Virodca33252010-07-25 02:31:46 +0400622 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400623
624 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400625 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500626 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400627 continue;
628 sb->s_count++;
629 spin_unlock(&sb_lock);
630
631 down_read(&sb->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100632 if (sb->s_root && (sb->s_flags & SB_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400633 f(sb, arg);
634 up_read(&sb->s_umount);
635
636 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400637 if (p)
638 __put_super(p);
639 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400640 }
Al Virodca33252010-07-25 02:31:46 +0400641 if (p)
642 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400643 spin_unlock(&sb_lock);
644}
645
646/**
Al Viro43e15cd2011-06-03 20:16:57 -0400647 * iterate_supers_type - call function for superblocks of given type
648 * @type: fs type
649 * @f: function to call
650 * @arg: argument to pass to it
651 *
652 * Scans the superblock list and calls given function, passing it
653 * locked superblock and given argument.
654 */
655void iterate_supers_type(struct file_system_type *type,
656 void (*f)(struct super_block *, void *), void *arg)
657{
658 struct super_block *sb, *p = NULL;
659
660 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800661 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400662 sb->s_count++;
663 spin_unlock(&sb_lock);
664
665 down_read(&sb->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100666 if (sb->s_root && (sb->s_flags & SB_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400667 f(sb, arg);
668 up_read(&sb->s_umount);
669
670 spin_lock(&sb_lock);
671 if (p)
672 __put_super(p);
673 p = sb;
674 }
675 if (p)
676 __put_super(p);
677 spin_unlock(&sb_lock);
678}
679
680EXPORT_SYMBOL(iterate_supers_type);
681
Jan Karaba6379f2016-11-23 12:53:00 +0100682static struct super_block *__get_super(struct block_device *bdev, bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700684 struct super_block *sb;
685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 if (!bdev)
687 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700688
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700690rescan:
691 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500692 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400693 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700694 if (sb->s_bdev == bdev) {
695 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 spin_unlock(&sb_lock);
Jan Karaba6379f2016-11-23 12:53:00 +0100697 if (!excl)
698 down_read(&sb->s_umount);
699 else
700 down_write(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400701 /* still alive? */
David Howellse462ec52017-07-17 08:45:35 +0100702 if (sb->s_root && (sb->s_flags & SB_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700703 return sb;
Jan Karaba6379f2016-11-23 12:53:00 +0100704 if (!excl)
705 up_read(&sb->s_umount);
706 else
707 up_write(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400708 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700709 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400710 __put_super(sb);
711 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 }
713 }
714 spin_unlock(&sb_lock);
715 return NULL;
716}
717
Jan Karaba6379f2016-11-23 12:53:00 +0100718/**
719 * get_super - get the superblock of a device
720 * @bdev: device to get the superblock for
721 *
722 * Scans the superblock list and finds the superblock of the file system
723 * mounted on the device given. %NULL is returned if no match is found.
724 */
725struct super_block *get_super(struct block_device *bdev)
726{
727 return __get_super(bdev, false);
728}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200730
Jan Karaba6379f2016-11-23 12:53:00 +0100731static struct super_block *__get_super_thawed(struct block_device *bdev,
732 bool excl)
733{
734 while (1) {
735 struct super_block *s = __get_super(bdev, excl);
736 if (!s || s->s_writers.frozen == SB_UNFROZEN)
737 return s;
738 if (!excl)
739 up_read(&s->s_umount);
740 else
741 up_write(&s->s_umount);
742 wait_event(s->s_writers.wait_unfrozen,
743 s->s_writers.frozen == SB_UNFROZEN);
744 put_super(s);
745 }
746}
747
Christoph Hellwig45042302009-08-03 23:28:35 +0200748/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100749 * get_super_thawed - get thawed superblock of a device
750 * @bdev: device to get the superblock for
751 *
752 * Scans the superblock list and finds the superblock of the file system
753 * mounted on the device. The superblock is returned once it is thawed
754 * (or immediately if it was not frozen). %NULL is returned if no match
755 * is found.
756 */
757struct super_block *get_super_thawed(struct block_device *bdev)
758{
Jan Karaba6379f2016-11-23 12:53:00 +0100759 return __get_super_thawed(bdev, false);
Jan Kara6b6dc832012-02-10 11:03:00 +0100760}
761EXPORT_SYMBOL(get_super_thawed);
762
763/**
Jan Karaba6379f2016-11-23 12:53:00 +0100764 * get_super_exclusive_thawed - get thawed superblock of a device
765 * @bdev: device to get the superblock for
766 *
767 * Scans the superblock list and finds the superblock of the file system
768 * mounted on the device. The superblock is returned once it is thawed
769 * (or immediately if it was not frozen) and s_umount semaphore is held
770 * in exclusive mode. %NULL is returned if no match is found.
771 */
772struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
773{
774 return __get_super_thawed(bdev, true);
775}
776EXPORT_SYMBOL(get_super_exclusive_thawed);
777
778/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200779 * get_active_super - get an active reference to the superblock of a device
780 * @bdev: device to get the superblock for
781 *
782 * Scans the superblock list and finds the superblock of the file system
783 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400784 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200785 */
786struct super_block *get_active_super(struct block_device *bdev)
787{
788 struct super_block *sb;
789
790 if (!bdev)
791 return NULL;
792
Al Viro14945832010-03-22 20:15:33 -0400793restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200794 spin_lock(&sb_lock);
795 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500796 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400797 continue;
Al Viro14945832010-03-22 20:15:33 -0400798 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400799 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400800 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400801 up_write(&sb->s_umount);
802 return sb;
Al Viro14945832010-03-22 20:15:33 -0400803 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200804 }
805 spin_unlock(&sb_lock);
806 return NULL;
807}
David Howellsdd111b32017-07-04 17:25:09 +0100808
Al Virodf40c012010-03-22 20:23:25 -0400809struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700811 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700814rescan:
815 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500816 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400817 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700818 if (sb->s_dev == dev) {
819 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700821 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400822 /* still alive? */
David Howellse462ec52017-07-17 08:45:35 +0100823 if (sb->s_root && (sb->s_flags & SB_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700824 return sb;
825 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400826 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700827 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400828 __put_super(sb);
829 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 }
831 }
832 spin_unlock(&sb_lock);
833 return NULL;
834}
835
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 * do_remount_sb - asks filesystem to change mount options.
838 * @sb: superblock in question
David Howellse462ec52017-07-17 08:45:35 +0100839 * @sb_flags: revised superblock flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 * @data: the rest of options
841 * @force: whether or not to force the change
842 *
843 * Alters the mount options of a mounted file system.
844 */
David Howellse462ec52017-07-17 08:45:35 +0100845int do_remount_sb(struct super_block *sb, int sb_flags, void *data, int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846{
847 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400848 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200849
Jan Kara5accdf82012-06-12 16:20:34 +0200850 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200851 return -EBUSY;
852
David Howells93614012006-09-30 20:45:40 +0200853#ifdef CONFIG_BLOCK
David Howellse462ec52017-07-17 08:45:35 +0100854 if (!(sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200856#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200857
David Howellse462ec52017-07-17 08:45:35 +0100858 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
Nick Piggind208bbd2009-12-21 16:28:53 -0800859
Al Viro0aec09d2014-08-07 07:32:06 -0400860 if (remount_ro) {
Al Virofdab6842015-01-11 10:57:27 -0500861 if (!hlist_empty(&sb->s_pins)) {
Al Viro0aec09d2014-08-07 07:32:06 -0400862 up_write(&sb->s_umount);
Al Virofdab6842015-01-11 10:57:27 -0500863 group_pin_kill(&sb->s_pins);
Al Viro0aec09d2014-08-07 07:32:06 -0400864 down_write(&sb->s_umount);
865 if (!sb->s_root)
866 return 0;
867 if (sb->s_writers.frozen != SB_UNFROZEN)
868 return -EBUSY;
David Howellse462ec52017-07-17 08:45:35 +0100869 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
Al Viro0aec09d2014-08-07 07:32:06 -0400870 }
871 }
872 shrink_dcache_sb(sb);
873
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 /* If we are remounting RDONLY and current sb is read/write,
875 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800876 if (remount_ro) {
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100877 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400878 sb->s_readonly_remount = 1;
879 smp_wmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100880 } else {
881 retval = sb_prepare_remount_readonly(sb);
882 if (retval)
883 return retval;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 }
886
887 if (sb->s_op->remount_fs) {
David Howellse462ec52017-07-17 08:45:35 +0100888 retval = sb->s_op->remount_fs(sb, &sb_flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200889 if (retval) {
890 if (!force)
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100891 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200892 /* If forced remount, go ahead despite any errors */
893 WARN(1, "forced remount of a %s fs returned %i\n",
894 sb->s_type->name, retval);
895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 }
David Howellse462ec52017-07-17 08:45:35 +0100897 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (sb_flags & MS_RMT_MASK);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100898 /* Needs to be ordered wrt mnt_is_readonly() */
899 smp_wmb();
900 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400901
Nick Piggind208bbd2009-12-21 16:28:53 -0800902 /*
903 * Some filesystems modify their metadata via some other path than the
904 * bdev buffer cache (eg. use a private mapping, or directories in
905 * pagecache, etc). Also file data modifications go via their own
906 * mappings. So If we try to mount readonly then copy the filesystem
907 * from bdev, we could get stale data, so invalidate it to give a best
908 * effort at coherency.
909 */
910 if (remount_ro && sb->s_bdev)
911 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 return 0;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100913
914cancel_readonly:
915 sb->s_readonly_remount = 0;
916 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917}
918
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +0200919static void do_emergency_remount_callback(struct super_block *sb)
920{
921 down_write(&sb->s_umount);
922 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
923 !sb_rdonly(sb)) {
924 /*
925 * What lock protects sb->s_flags??
926 */
927 do_remount_sb(sb, SB_RDONLY, NULL, 1);
928 }
929 up_write(&sb->s_umount);
930}
931
Jens Axboea2a95372009-03-17 09:38:40 +0100932static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933{
Mateusz Guzikfa7c1d52017-10-03 18:17:40 +0200934 __iterate_supers(do_emergency_remount_callback);
Jens Axboea2a95372009-03-17 09:38:40 +0100935 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 printk("Emergency Remount complete\n");
937}
938
939void emergency_remount(void)
940{
Jens Axboea2a95372009-03-17 09:38:40 +0100941 struct work_struct *work;
942
943 work = kmalloc(sizeof(*work), GFP_ATOMIC);
944 if (work) {
945 INIT_WORK(work, do_emergency_remount);
946 schedule_work(work);
947 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948}
949
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +0200950static void do_thaw_all_callback(struct super_block *sb)
951{
952 down_write(&sb->s_umount);
Al Viro1c18d2a2018-05-10 15:09:41 -0400953 if (sb->s_root && sb->s_flags & SB_BORN) {
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +0200954 emergency_thaw_bdev(sb);
955 thaw_super_locked(sb);
956 } else {
957 up_write(&sb->s_umount);
958 }
959}
960
961static void do_thaw_all(struct work_struct *work)
962{
963 __iterate_supers(do_thaw_all_callback);
964 kfree(work);
965 printk(KERN_WARNING "Emergency Thaw complete\n");
966}
967
968/**
969 * emergency_thaw_all -- forcibly thaw every frozen filesystem
970 *
971 * Used for emergency unfreeze of all filesystems via SysRq
972 */
973void emergency_thaw_all(void)
974{
975 struct work_struct *work;
976
977 work = kmalloc(sizeof(*work), GFP_ATOMIC);
978 if (work) {
979 INIT_WORK(work, do_thaw_all);
980 schedule_work(work);
981 }
982}
983
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400984static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
Matthew Wilcox5a668472018-06-11 12:03:31 -0400986/**
987 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
988 * @p: Pointer to a dev_t.
989 *
990 * Filesystems which don't use real block devices can call this function
991 * to allocate a virtual block device.
992 *
993 * Context: Any context. Frequently called while holding sb_lock.
994 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
995 * or -ENOMEM if memory allocation failed.
996 */
Al Viro0ee5dc62011-07-07 15:44:25 -0400997int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998{
999 int dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Matthew Wilcox5a668472018-06-11 12:03:31 -04001001 /*
1002 * Many userspace utilities consider an FSID of 0 invalid.
1003 * Always return at least 1 from get_anon_bdev.
1004 */
1005 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1006 GFP_ATOMIC);
1007 if (dev == -ENOSPC)
1008 dev = -EMFILE;
1009 if (dev < 0)
1010 return dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
Matthew Wilcox5a668472018-06-11 12:03:31 -04001012 *p = MKDEV(0, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 return 0;
1014}
Al Viro0ee5dc62011-07-07 15:44:25 -04001015EXPORT_SYMBOL(get_anon_bdev);
1016
1017void free_anon_bdev(dev_t dev)
1018{
Matthew Wilcox5a668472018-06-11 12:03:31 -04001019 ida_free(&unnamed_dev_ida, MINOR(dev));
Al Viro0ee5dc62011-07-07 15:44:25 -04001020}
1021EXPORT_SYMBOL(free_anon_bdev);
1022
1023int set_anon_super(struct super_block *s, void *data)
1024{
Christoph Hellwigdf0ce262015-01-14 10:42:41 +01001025 return get_anon_bdev(&s->s_dev);
Al Viro0ee5dc62011-07-07 15:44:25 -04001026}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027EXPORT_SYMBOL(set_anon_super);
1028
1029void kill_anon_super(struct super_block *sb)
1030{
Al Viro0ee5dc62011-07-07 15:44:25 -04001031 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -04001033 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035EXPORT_SYMBOL(kill_anon_super);
1036
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037void kill_litter_super(struct super_block *sb)
1038{
1039 if (sb->s_root)
1040 d_genocide(sb->s_root);
1041 kill_anon_super(sb);
1042}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043EXPORT_SYMBOL(kill_litter_super);
1044
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001045static int ns_test_super(struct super_block *sb, void *data)
1046{
1047 return sb->s_fs_info == data;
1048}
1049
1050static int ns_set_super(struct super_block *sb, void *data)
1051{
1052 sb->s_fs_info = data;
1053 return set_anon_super(sb, NULL);
1054}
1055
Eric W. Biedermand91ee872016-05-23 14:51:59 -05001056struct dentry *mount_ns(struct file_system_type *fs_type,
1057 int flags, void *data, void *ns, struct user_namespace *user_ns,
1058 int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001059{
1060 struct super_block *sb;
1061
Eric W. Biedermand91ee872016-05-23 14:51:59 -05001062 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1063 * over the namespace.
1064 */
David Howellse462ec52017-07-17 08:45:35 +01001065 if (!(flags & SB_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
Eric W. Biedermand91ee872016-05-23 14:51:59 -05001066 return ERR_PTR(-EPERM);
1067
Eric W. Biederman6e4eab52016-05-24 09:29:01 -05001068 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1069 user_ns, ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001070 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +04001071 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001072
1073 if (!sb->s_root) {
1074 int err;
David Howellse462ec52017-07-17 08:45:35 +01001075 err = fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001076 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001077 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +04001078 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001079 }
1080
David Howellse462ec52017-07-17 08:45:35 +01001081 sb->s_flags |= SB_ACTIVE;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001082 }
1083
Al Viroceefda62010-07-26 13:16:50 +04001084 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001085}
1086
Al Viroceefda62010-07-26 13:16:50 +04001087EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001088
David Howells93614012006-09-30 20:45:40 +02001089#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090static int set_bdev_super(struct super_block *s, void *data)
1091{
1092 s->s_bdev = data;
1093 s->s_dev = s->s_bdev->bd_dev;
Jan Kara13eec232017-04-12 12:24:29 +02001094 s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
Jens Axboe32a88aa2009-09-16 15:02:33 +02001095
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 return 0;
1097}
1098
1099static int test_bdev_super(struct super_block *s, void *data)
1100{
1101 return (void *)s->s_bdev == data;
1102}
1103
Al Viro152a0832010-07-25 00:46:55 +04001104struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +04001106 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107{
1108 struct block_device *bdev;
1109 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +01001110 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 int error = 0;
1112
David Howellse462ec52017-07-17 08:45:35 +01001113 if (!(flags & SB_RDONLY))
Al Viro30c40d22008-02-22 19:50:45 -05001114 mode |= FMODE_WRITE;
1115
Tejun Heod4d77622010-11-13 11:55:18 +01001116 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +04001118 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
1120 /*
1121 * once the super is inserted into the list by sget, s_umount
1122 * will protect the lockfs code from trying to start a snapshot
1123 * while we are mounting
1124 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001125 mutex_lock(&bdev->bd_fsfreeze_mutex);
1126 if (bdev->bd_fsfreeze_count > 0) {
1127 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1128 error = -EBUSY;
1129 goto error_bdev;
1130 }
David Howellse462ec52017-07-17 08:45:35 +01001131 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
David Howells9249e172012-06-25 12:55:37 +01001132 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001133 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -07001135 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
1137 if (s->s_root) {
David Howellse462ec52017-07-17 08:45:35 +01001138 if ((flags ^ s->s_flags) & SB_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001139 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001140 error = -EBUSY;
1141 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 }
David Howells454e2392006-06-23 02:02:57 -07001143
Tejun Heo4f331f02010-07-20 15:18:07 -07001144 /*
1145 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +01001146 * __invalidate_device(). blkdev_put() acquires
1147 * bd_mutex and can't be called under s_umount. Drop
1148 * s_umount temporarily. This is safe as we're
1149 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -07001150 */
1151 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +01001152 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -07001153 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 } else {
Al Viro30c40d22008-02-22 19:50:45 -05001155 s->s_mode = mode;
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001156 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
Pekka Enberge78c9a02006-01-08 01:03:39 -08001157 sb_set_blocksize(s, block_size(bdev));
David Howellse462ec52017-07-17 08:45:35 +01001158 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001160 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001161 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001162 }
David Howells454e2392006-06-23 02:02:57 -07001163
David Howellse462ec52017-07-17 08:45:35 +01001164 s->s_flags |= SB_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001165 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 }
1167
Al Viro152a0832010-07-25 00:46:55 +04001168 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
David Howells454e2392006-06-23 02:02:57 -07001170error_s:
1171 error = PTR_ERR(s);
1172error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001173 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001174error:
Al Viro152a0832010-07-25 00:46:55 +04001175 return ERR_PTR(error);
1176}
1177EXPORT_SYMBOL(mount_bdev);
1178
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179void kill_block_super(struct super_block *sb)
1180{
1181 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001182 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001184 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 generic_shutdown_super(sb);
1186 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001187 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001188 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189}
1190
1191EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001192#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
Al Viro3c26ff62010-07-25 11:46:36 +04001194struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001196 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197{
1198 int error;
David Howells9249e172012-06-25 12:55:37 +01001199 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
1201 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001202 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
David Howellse462ec52017-07-17 08:45:35 +01001204 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001206 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001207 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 }
David Howellse462ec52017-07-17 08:45:35 +01001209 s->s_flags |= SB_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001210 return dget(s->s_root);
1211}
1212EXPORT_SYMBOL(mount_nodev);
1213
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214static int compare_single(struct super_block *s, void *p)
1215{
1216 return 1;
1217}
1218
Al Virofc14f2f2010-07-25 01:48:30 +04001219struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001221 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222{
1223 struct super_block *s;
1224 int error;
1225
David Howells9249e172012-06-25 12:55:37 +01001226 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001228 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 if (!s->s_root) {
David Howellse462ec52017-07-17 08:45:35 +01001230 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001232 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001233 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 }
David Howellse462ec52017-07-17 08:45:35 +01001235 s->s_flags |= SB_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001236 } else {
1237 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 }
Al Virofc14f2f2010-07-25 01:48:30 +04001239 return dget(s->s_root);
1240}
1241EXPORT_SYMBOL(mount_single);
1242
Al Viro9d412a42011-03-17 22:08:28 -04001243struct dentry *
1244mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245{
Al Viroc96e41e2010-07-25 00:17:56 +04001246 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001247 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001249 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001250
Eric Parise0007522008-03-05 10:31:54 -05001251 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001253 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001254 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
Eric Parise0007522008-03-05 10:31:54 -05001256 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001257 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 }
1260
Al Viro1a102ff2011-03-16 09:07:58 -04001261 root = type->mount(type, flags, name, data);
1262 if (IS_ERR(root)) {
1263 error = PTR_ERR(root);
1264 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001265 }
Al Viro9d412a42011-03-17 22:08:28 -04001266 sb = root->d_sb;
1267 BUG_ON(!sb);
1268 WARN_ON(!sb->s_bdi);
Dave Chinner79f546a2018-05-11 11:20:57 +10001269
1270 /*
1271 * Write barrier is for super_cache_count(). We place it before setting
1272 * SB_BORN as the data dependency between the two functions is the
1273 * superblock structure contents that we just set up, not the SB_BORN
1274 * flag.
1275 */
1276 smp_wmb();
David Howellse462ec52017-07-17 08:45:35 +01001277 sb->s_flags |= SB_BORN;
David Howells454e2392006-06-23 02:02:57 -07001278
Al Viro9d412a42011-03-17 22:08:28 -04001279 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001280 if (error)
1281 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001282
Jeff Layton42cb56a2009-09-18 13:05:53 -07001283 /*
1284 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1285 * but s_maxbytes was an unsigned long long for many releases. Throw
1286 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001287 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001288 */
Al Viro9d412a42011-03-17 22:08:28 -04001289 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1290 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001291
Al Viro9d412a42011-03-17 22:08:28 -04001292 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001293 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001294 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001296 dput(root);
1297 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298out_free_secdata:
1299 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300out:
David Howells454e2392006-06-23 02:02:57 -07001301 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302}
1303
Jan Kara5accdf82012-06-12 16:20:34 +02001304/*
Jan Karafca39342017-04-12 12:24:28 +02001305 * Setup private BDI for given superblock. It gets automatically cleaned up
1306 * in generic_shutdown_super().
1307 */
1308int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1309{
1310 struct backing_dev_info *bdi;
1311 int err;
1312 va_list args;
1313
1314 bdi = bdi_alloc(GFP_KERNEL);
1315 if (!bdi)
1316 return -ENOMEM;
1317
1318 bdi->name = sb->s_type->name;
1319
1320 va_start(args, fmt);
Jan Kara7c4cc302017-04-12 12:24:49 +02001321 err = bdi_register_va(bdi, fmt, args);
Jan Karafca39342017-04-12 12:24:28 +02001322 va_end(args);
1323 if (err) {
1324 bdi_put(bdi);
1325 return err;
1326 }
1327 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1328 sb->s_bdi = bdi;
Jan Karafca39342017-04-12 12:24:28 +02001329
1330 return 0;
1331}
1332EXPORT_SYMBOL(super_setup_bdi_name);
1333
1334/*
1335 * Setup private BDI for given superblock. I gets automatically cleaned up
1336 * in generic_shutdown_super().
1337 */
1338int super_setup_bdi(struct super_block *sb)
1339{
1340 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1341
1342 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1343 atomic_long_inc_return(&bdi_seq));
1344}
1345EXPORT_SYMBOL(super_setup_bdi);
1346
1347/*
Jan Kara5accdf82012-06-12 16:20:34 +02001348 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1349 * instead.
1350 */
1351void __sb_end_write(struct super_block *sb, int level)
1352{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001353 percpu_up_read(sb->s_writers.rw_sem + level-1);
Jan Kara5accdf82012-06-12 16:20:34 +02001354}
1355EXPORT_SYMBOL(__sb_end_write);
1356
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001357/*
1358 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1359 * instead.
1360 */
1361int __sb_start_write(struct super_block *sb, int level, bool wait)
1362{
1363 bool force_trylock = false;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001364 int ret = 1;
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001365
1366#ifdef CONFIG_LOCKDEP
1367 /*
1368 * We want lockdep to tell us about possible deadlocks with freezing
1369 * but it's it bit tricky to properly instrument it. Getting a freeze
1370 * protection works as getting a read lock but there are subtle
1371 * problems. XFS for example gets freeze protection on internal level
1372 * twice in some cases, which is OK only because we already hold a
1373 * freeze protection also on higher level. Due to these cases we have
1374 * to use wait == F (trylock mode) which must not fail.
1375 */
1376 if (wait) {
1377 int i;
1378
1379 for (i = 0; i < level - 1; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001380 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001381 force_trylock = true;
1382 break;
1383 }
1384 }
1385#endif
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001386 if (wait && !force_trylock)
1387 percpu_down_read(sb->s_writers.rw_sem + level-1);
1388 else
1389 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1390
Vincent Stehlé22224a12015-09-21 17:18:34 +02001391 WARN_ON(force_trylock && !ret);
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001392 return ret;
1393}
Jan Kara5accdf82012-06-12 16:20:34 +02001394EXPORT_SYMBOL(__sb_start_write);
1395
1396/**
1397 * sb_wait_write - wait until all writers to given file system finish
1398 * @sb: the super for which we wait
1399 * @level: type of writers we wait for (normal vs page fault)
1400 *
1401 * This function waits until there are no writers of given type to given file
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001402 * system.
Jan Kara5accdf82012-06-12 16:20:34 +02001403 */
1404static void sb_wait_write(struct super_block *sb, int level)
1405{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001406 percpu_down_write(sb->s_writers.rw_sem + level-1);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001407}
Jan Kara5accdf82012-06-12 16:20:34 +02001408
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001409/*
1410 * We are going to return to userspace and forget about these locks, the
1411 * ownership goes to the caller of thaw_super() which does unlock().
1412 */
1413static void lockdep_sb_freeze_release(struct super_block *sb)
1414{
1415 int level;
1416
1417 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1418 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1419}
1420
1421/*
1422 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1423 */
1424static void lockdep_sb_freeze_acquire(struct super_block *sb)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001425{
1426 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001427
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001428 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1429 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001430}
1431
1432static void sb_freeze_unlock(struct super_block *sb)
1433{
1434 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001435
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001436 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1437 percpu_up_write(sb->s_writers.rw_sem + level);
Jan Kara5accdf82012-06-12 16:20:34 +02001438}
1439
Josef Bacik18e9e512010-03-23 10:34:56 -04001440/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001441 * freeze_super - lock the filesystem and force it into a consistent state
1442 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001443 *
1444 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1445 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1446 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001447 *
1448 * During this function, sb->s_writers.frozen goes through these values:
1449 *
1450 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1451 *
1452 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1453 * writes should be blocked, though page faults are still allowed. We wait for
1454 * all writes to complete and then proceed to the next stage.
1455 *
1456 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1457 * but internal fs threads can still modify the filesystem (although they
1458 * should not dirty new pages or inodes), writeback can run etc. After waiting
1459 * for all running page faults we sync the filesystem which will clean all
1460 * dirty pages and inodes (no new dirty pages or inodes can be created when
1461 * sync is running).
1462 *
1463 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1464 * modification are blocked (e.g. XFS preallocation truncation on inode
1465 * reclaim). This is usually implemented by blocking new transactions for
1466 * filesystems that have them and need this additional guard. After all
1467 * internal writers are finished we call ->freeze_fs() to finish filesystem
1468 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1469 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1470 *
1471 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001472 */
1473int freeze_super(struct super_block *sb)
1474{
1475 int ret;
1476
1477 atomic_inc(&sb->s_active);
1478 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001479 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001480 deactivate_locked_super(sb);
1481 return -EBUSY;
1482 }
1483
David Howellse462ec52017-07-17 08:45:35 +01001484 if (!(sb->s_flags & SB_BORN)) {
Al Virodabe0dc2012-01-03 21:01:29 -05001485 up_write(&sb->s_umount);
1486 return 0; /* sic - it's "nothing to do" */
1487 }
1488
David Howellsbc98a422017-07-17 08:45:34 +01001489 if (sb_rdonly(sb)) {
Jan Kara5accdf82012-06-12 16:20:34 +02001490 /* Nothing to do really... */
1491 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001492 up_write(&sb->s_umount);
1493 return 0;
1494 }
1495
Jan Kara5accdf82012-06-12 16:20:34 +02001496 sb->s_writers.frozen = SB_FREEZE_WRITE;
Jan Kara5accdf82012-06-12 16:20:34 +02001497 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1498 up_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001499 sb_wait_write(sb, SB_FREEZE_WRITE);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001500 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001501
1502 /* Now we go and block page faults... */
Jan Kara5accdf82012-06-12 16:20:34 +02001503 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
Jan Kara5accdf82012-06-12 16:20:34 +02001504 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1505
1506 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001507 sync_filesystem(sb);
1508
Jan Kara5accdf82012-06-12 16:20:34 +02001509 /* Now wait for internal filesystem counter */
1510 sb->s_writers.frozen = SB_FREEZE_FS;
Jan Kara5accdf82012-06-12 16:20:34 +02001511 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001512
Josef Bacik18e9e512010-03-23 10:34:56 -04001513 if (sb->s_op->freeze_fs) {
1514 ret = sb->s_op->freeze_fs(sb);
1515 if (ret) {
1516 printk(KERN_ERR
1517 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001518 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001519 sb_freeze_unlock(sb);
Jan Kara5accdf82012-06-12 16:20:34 +02001520 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001521 deactivate_locked_super(sb);
1522 return ret;
1523 }
1524 }
Jan Kara5accdf82012-06-12 16:20:34 +02001525 /*
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001526 * For debugging purposes so that fs can warn if it sees write activity
1527 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
Jan Kara5accdf82012-06-12 16:20:34 +02001528 */
1529 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001530 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001531 up_write(&sb->s_umount);
1532 return 0;
1533}
1534EXPORT_SYMBOL(freeze_super);
1535
1536/**
1537 * thaw_super -- unlock filesystem
1538 * @sb: the super to thaw
1539 *
1540 * Unlocks the filesystem and marks it writeable again after freeze_super().
1541 */
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001542static int thaw_super_locked(struct super_block *sb)
Josef Bacik18e9e512010-03-23 10:34:56 -04001543{
1544 int error;
1545
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001546 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001547 up_write(&sb->s_umount);
1548 return -EINVAL;
1549 }
1550
David Howellsbc98a422017-07-17 08:45:34 +01001551 if (sb_rdonly(sb)) {
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001552 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001553 goto out;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001554 }
Josef Bacik18e9e512010-03-23 10:34:56 -04001555
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001556 lockdep_sb_freeze_acquire(sb);
1557
Josef Bacik18e9e512010-03-23 10:34:56 -04001558 if (sb->s_op->unfreeze_fs) {
1559 error = sb->s_op->unfreeze_fs(sb);
1560 if (error) {
1561 printk(KERN_ERR
1562 "VFS:Filesystem thaw failed\n");
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001563 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001564 up_write(&sb->s_umount);
1565 return error;
1566 }
1567 }
1568
Jan Kara5accdf82012-06-12 16:20:34 +02001569 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001570 sb_freeze_unlock(sb);
1571out:
Jan Kara5accdf82012-06-12 16:20:34 +02001572 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001573 deactivate_locked_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001574 return 0;
1575}
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +02001576
1577int thaw_super(struct super_block *sb)
1578{
1579 down_write(&sb->s_umount);
1580 return thaw_super_locked(sb);
1581}
Josef Bacik18e9e512010-03-23 10:34:56 -04001582EXPORT_SYMBOL(thaw_super);