blob: abe2541fb28c2872bcc81411f9978366e9fadc68 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/super.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * super.c contains code to handle: - mount structures
7 * - super-block tables
8 * - filesystem drivers list
9 * - mount system call
10 * - umount system call
11 * - ustat system call
12 *
13 * GK 2/5/95 - Changed to support mounting the root fs via NFS
14 *
15 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
16 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
17 * Added options to /proc/mounts:
Jan Engelhardt96de0e22007-10-19 23:21:04 +020018 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
Linus Torvalds1da177e2005-04-16 15:20:36 -070019 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
20 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
21 */
22
Paul Gortmaker630d9c42011-11-16 23:57:37 -050023#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/mount.h>
27#include <linux/security.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/writeback.h> /* for the emergency remount stuff */
29#include <linux/idr.h>
Ingo Molnar353ab6e2006-03-26 01:37:12 -080030#include <linux/mutex.h>
Jens Axboe5477d0f2010-04-29 20:33:35 +020031#include <linux/backing-dev.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110032#include <linux/rculist_bl.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060033#include <linux/cleancache.h>
Al Viro40401532012-02-13 03:58:52 +000034#include <linux/fsnotify.h>
Jan Kara5accdf82012-06-12 16:20:34 +020035#include <linux/lockdep.h>
Eric W. Biederman6e4eab52016-05-24 09:29:01 -050036#include <linux/user_namespace.h>
Al Viro6d59e7f2008-03-22 15:48:17 -040037#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39
Al Viro15d0f5e2015-02-02 10:07:59 -070040static LIST_HEAD(super_blocks);
41static DEFINE_SPINLOCK(sb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Jan Kara5accdf82012-06-12 16:20:34 +020043static char *sb_writers_name[SB_FREEZE_LEVELS] = {
44 "sb_writers",
45 "sb_pagefaults",
46 "sb_internal",
47};
48
Dave Chinnerb0d40c92011-07-08 14:14:42 +100049/*
50 * One thing we have to be careful of with a per-sb shrinker is that we don't
51 * drop the last active reference to the superblock from within the shrinker.
52 * If that happens we could trigger unregistering the shrinker from within the
53 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
54 * take a passive reference to the superblock to avoid this from occurring.
55 */
Dave Chinner0a234c62013-08-28 10:17:57 +100056static unsigned long super_cache_scan(struct shrinker *shrink,
57 struct shrink_control *sc)
Dave Chinnerb0d40c92011-07-08 14:14:42 +100058{
59 struct super_block *sb;
Dave Chinner0a234c62013-08-28 10:17:57 +100060 long fs_objects = 0;
61 long total_objects;
62 long freed = 0;
63 long dentries;
64 long inodes;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100065
66 sb = container_of(shrink, struct super_block, s_shrink);
67
68 /*
69 * Deadlock avoidance. We may hold various FS locks, and we don't want
70 * to recurse into the FS that called us in clear_inode() and friends..
71 */
Dave Chinner0a234c62013-08-28 10:17:57 +100072 if (!(sc->gfp_mask & __GFP_FS))
73 return SHRINK_STOP;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100074
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +030075 if (!trylock_super(sb))
Dave Chinner0a234c62013-08-28 10:17:57 +100076 return SHRINK_STOP;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100077
Al Virod0407902013-07-19 21:12:31 +040078 if (sb->s_op->nr_cached_objects)
Vladimir Davydov4101b622015-02-12 14:58:51 -080079 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100080
Vladimir Davydov503c3582015-02-12 14:58:47 -080081 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
82 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
Dave Chinnerf6041562013-08-28 10:18:00 +100083 total_objects = dentries + inodes + fs_objects + 1;
Tetsuo Handa475d0db2014-05-17 20:56:38 +090084 if (!total_objects)
85 total_objects = 1;
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100086
Dave Chinner0a234c62013-08-28 10:17:57 +100087 /* proportion the scan between the caches */
Dave Chinnerf6041562013-08-28 10:18:00 +100088 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +100089 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
Vladimir Davydov503c3582015-02-12 14:58:47 -080090 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
Dave Chinnerb0d40c92011-07-08 14:14:42 +100091
Dave Chinner0a234c62013-08-28 10:17:57 +100092 /*
93 * prune the dcache first as the icache is pinned by it, then
94 * prune the icache, followed by the filesystem specific caches
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -080095 *
96 * Ensure that we always scan at least one object - memcg kmem
97 * accounting uses this to fully empty the caches.
Dave Chinner0a234c62013-08-28 10:17:57 +100098 */
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -080099 sc->nr_to_scan = dentries + 1;
Vladimir Davydov503c3582015-02-12 14:58:47 -0800100 freed = prune_dcache_sb(sb, sc);
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -0800101 sc->nr_to_scan = inodes + 1;
Vladimir Davydov503c3582015-02-12 14:58:47 -0800102 freed += prune_icache_sb(sb, sc);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000103
Dave Chinner0a234c62013-08-28 10:17:57 +1000104 if (fs_objects) {
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -0800105 sc->nr_to_scan = fs_objects + 1;
Vladimir Davydov4101b622015-02-12 14:58:51 -0800106 freed += sb->s_op->free_cached_objects(sb, sc);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000107 }
108
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300109 up_read(&sb->s_umount);
Dave Chinner0a234c62013-08-28 10:17:57 +1000110 return freed;
111}
112
113static unsigned long super_cache_count(struct shrinker *shrink,
114 struct shrink_control *sc)
115{
116 struct super_block *sb;
117 long total_objects = 0;
118
119 sb = container_of(shrink, struct super_block, s_shrink);
120
Tim Chend23da152014-06-04 16:10:47 -0700121 /*
Dave Chinnerd6f62ec2018-05-11 11:20:57 +1000122 * We don't call trylock_super() here as it is a scalability bottleneck,
123 * so we're exposed to partial setup state. The shrinker rwsem does not
124 * protect filesystem operations backing list_lru_shrink_count() or
125 * s_op->nr_cached_objects(). Counts can change between
126 * super_cache_count and super_cache_scan, so we really don't need locks
127 * here.
128 *
129 * However, if we are currently mounting the superblock, the underlying
130 * filesystem might be in a state of partial construction and hence it
131 * is dangerous to access it. trylock_super() uses a MS_BORN check to
132 * avoid this situation, so do the same here. The memory barrier is
133 * matched with the one in mount_fs() as we don't hold locks here.
Tim Chend23da152014-06-04 16:10:47 -0700134 */
Dave Chinnerd6f62ec2018-05-11 11:20:57 +1000135 if (!(sb->s_flags & MS_BORN))
136 return 0;
137 smp_rmb();
138
Dave Chinner0a234c62013-08-28 10:17:57 +1000139 if (sb->s_op && sb->s_op->nr_cached_objects)
Vladimir Davydov4101b622015-02-12 14:58:51 -0800140 total_objects = sb->s_op->nr_cached_objects(sb, sc);
Dave Chinner0a234c62013-08-28 10:17:57 +1000141
Vladimir Davydov503c3582015-02-12 14:58:47 -0800142 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
143 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
Dave Chinner0a234c62013-08-28 10:17:57 +1000144
Glauber Costa55f841c2013-08-28 10:17:53 +1000145 total_objects = vfs_pressure_ratio(total_objects);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000146 return total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000147}
148
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200149static void destroy_super_work(struct work_struct *work)
150{
151 struct super_block *s = container_of(work, struct super_block,
152 destroy_work);
153 int i;
154
155 for (i = 0; i < SB_FREEZE_LEVELS; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200156 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200157 kfree(s);
158}
159
160static void destroy_super_rcu(struct rcu_head *head)
161{
162 struct super_block *s = container_of(head, struct super_block, rcu);
163 INIT_WORK(&s->destroy_work, destroy_super_work);
164 schedule_work(&s->destroy_work);
165}
166
Al Viro7eb5e882013-10-01 15:09:58 -0400167/**
168 * destroy_super - frees a superblock
169 * @s: superblock to free
170 *
171 * Frees a superblock.
172 */
173static void destroy_super(struct super_block *s)
Jan Kara5accdf82012-06-12 16:20:34 +0200174{
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);
178 WARN_ON(!list_empty(&s->s_mounts));
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500179 put_user_ns(s->s_user_ns);
Al Viro7eb5e882013-10-01 15:09:58 -0400180 kfree(s->s_subtype);
181 kfree(s->s_options);
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200182 call_rcu(&s->rcu, destroy_super_rcu);
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);
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800206
Al Viro7eb5e882013-10-01 15:09:58 -0400207 if (security_sb_alloc(s))
208 goto fail;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200209
Al Viro7eb5e882013-10-01 15:09:58 -0400210 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200211 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
212 sb_writers_name[i],
213 &type->s_writers_key[i]))
Al Viro7eb5e882013-10-01 15:09:58 -0400214 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 }
Al Viro7eb5e882013-10-01 15:09:58 -0400216 init_waitqueue_head(&s->s_writers.wait_unfrozen);
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100217 s->s_bdi = &noop_backing_dev_info;
Al Viro7eb5e882013-10-01 15:09:58 -0400218 s->s_flags = flags;
Eric W. Biedermancc50a072016-06-09 15:44:48 -0500219 if (s->s_user_ns != &init_user_ns)
Eric W. Biederman67690f92016-05-18 13:50:06 -0500220 s->s_iflags |= SB_I_NODEV;
Al Viro7eb5e882013-10-01 15:09:58 -0400221 INIT_HLIST_NODE(&s->s_instances);
222 INIT_HLIST_BL_HEAD(&s->s_anon);
Dave Chinnere97fedb2015-03-04 13:40:00 -0500223 mutex_init(&s->s_sync_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400224 INIT_LIST_HEAD(&s->s_inodes);
Dave Chinner74278da2015-03-04 12:37:22 -0500225 spin_lock_init(&s->s_inode_list_lock);
Dave Chinner6c60d2b2016-07-26 15:21:50 -0700226 INIT_LIST_HEAD(&s->s_inodes_wb);
227 spin_lock_init(&s->s_inode_wblist_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400228
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800229 if (list_lru_init_memcg(&s->s_dentry_lru))
Al Viro7eb5e882013-10-01 15:09:58 -0400230 goto fail;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800231 if (list_lru_init_memcg(&s->s_inode_lru))
Al Viro7eb5e882013-10-01 15:09:58 -0400232 goto fail;
233
Al Viro7eb5e882013-10-01 15:09:58 -0400234 init_rwsem(&s->s_umount);
235 lockdep_set_class(&s->s_umount, &type->s_umount_key);
236 /*
237 * sget() can have s_umount recursion.
238 *
239 * When it cannot find a suitable sb, it allocates a new
240 * one (this one), and tries again to find a suitable old
241 * one.
242 *
243 * In case that succeeds, it will acquire the s_umount
244 * lock of the old one. Since these are clearly distrinct
245 * locks, and this object isn't exposed yet, there's no
246 * risk of deadlocks.
247 *
248 * Annotate this by putting this lock in a different
249 * subclass.
250 */
251 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
252 s->s_count = 1;
253 atomic_set(&s->s_active, 1);
254 mutex_init(&s->s_vfs_rename_mutex);
255 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
256 mutex_init(&s->s_dquot.dqio_mutex);
257 mutex_init(&s->s_dquot.dqonoff_mutex);
Al Viro7eb5e882013-10-01 15:09:58 -0400258 s->s_maxbytes = MAX_NON_LFS;
259 s->s_op = &default_op;
260 s->s_time_gran = 1000000000;
Vladimir Davydov3cb29d12015-04-14 15:46:48 -0700261 s->cleancache_poolid = CLEANCACHE_NO_POOL;
Al Viro7eb5e882013-10-01 15:09:58 -0400262
263 s->s_shrink.seeks = DEFAULT_SEEKS;
264 s->s_shrink.scan_objects = super_cache_scan;
265 s->s_shrink.count_objects = super_cache_count;
266 s->s_shrink.batch = 1024;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800267 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
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:
271 destroy_super(s);
272 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 Virof47ec3f2011-11-21 21:15:42 -0500280static void __put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 if (!--sb->s_count) {
Al Viro551de6f2010-03-22 19:36:35 -0400283 list_del_init(&sb->s_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 destroy_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286}
287
288/**
289 * put_super - drop a temporary reference to superblock
290 * @sb: superblock in question
291 *
292 * Drops a temporary reference, frees superblock if there's no
293 * references left.
294 */
Al Virof47ec3f2011-11-21 21:15:42 -0500295static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
297 spin_lock(&sb_lock);
298 __put_super(sb);
299 spin_unlock(&sb_lock);
300}
301
302
303/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400304 * deactivate_locked_super - drop an active reference to superblock
305 * @s: superblock to deactivate
306 *
Masanari Iidabd7ced92016-02-02 22:31:06 +0900307 * Drops an active reference to superblock, converting it into a temporary
Al Viro1712ac82010-03-22 15:22:31 -0400308 * one if there is no other active references left. In that case we
309 * tell fs driver to shut it down and drop the temporary reference we
310 * had just acquired.
311 *
312 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400313 */
314void deactivate_locked_super(struct super_block *s)
315{
316 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400317 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400318 cleancache_invalidate_fs(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000319 unregister_shrinker(&s->s_shrink);
Dave Chinner28f2cd4f2014-06-04 16:10:46 -0700320 fs->kill_sb(s);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000321
Vladimir Davydovc0a5b562015-02-12 14:59:07 -0800322 /*
323 * Since list_lru_destroy() may sleep, we cannot call it from
324 * put_super(), where we hold the sb_lock. Therefore we destroy
325 * the lru lists right now.
326 */
327 list_lru_destroy(&s->s_dentry_lru);
328 list_lru_destroy(&s->s_inode_lru);
329
Al Viro74dbbdd2009-05-06 01:07:50 -0400330 put_filesystem(fs);
331 put_super(s);
332 } else {
333 up_write(&s->s_umount);
334 }
335}
336
337EXPORT_SYMBOL(deactivate_locked_super);
338
339/**
Al Viro1712ac82010-03-22 15:22:31 -0400340 * deactivate_super - drop an active reference to superblock
341 * @s: superblock to deactivate
342 *
343 * Variant of deactivate_locked_super(), except that superblock is *not*
344 * locked by caller. If we are going to drop the final active reference,
345 * lock will be acquired prior to that.
346 */
347void deactivate_super(struct super_block *s)
348{
349 if (!atomic_add_unless(&s->s_active, -1, 1)) {
350 down_write(&s->s_umount);
351 deactivate_locked_super(s);
352 }
353}
354
355EXPORT_SYMBOL(deactivate_super);
356
357/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 * grab_super - acquire an active reference
359 * @s: reference we are trying to make active
360 *
361 * Tries to acquire an active reference. grab_super() is used when we
362 * had just found a superblock in super_blocks or fs_type->fs_supers
363 * and want to turn it into a full-blown active reference. grab_super()
364 * is called with sb_lock held and drops it. Returns 1 in case of
365 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400366 * dying when grab_super() had been called). Note that this is only
367 * called for superblocks not in rundown mode (== ones still on ->fs_supers
368 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700370static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
372 s->s_count++;
373 spin_unlock(&sb_lock);
374 down_write(&s->s_umount);
Al Viroacfec9a2013-07-20 03:13:55 +0400375 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
376 put_super(s);
377 return 1;
378 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 up_write(&s->s_umount);
380 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 return 0;
382}
383
David Howellscf9a2ae2006-08-29 19:05:54 +0100384/*
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300385 * trylock_super - try to grab ->s_umount shared
Wanpeng Li331cbde2012-06-09 11:10:55 +0800386 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000387 *
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300388 * Try to prevent fs shutdown. This is used in places where we
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000389 * cannot take an active reference but we need to ensure that the
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300390 * filesystem is not shut down while we are working on it. It returns
391 * false if we cannot acquire s_umount or if we lose the race and
392 * filesystem already got into shutdown, and returns true with the s_umount
393 * lock held in read mode in case of success. On successful return,
394 * the caller must drop the s_umount lock when done.
395 *
396 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
397 * The reason why it's safe is that we are OK with doing trylock instead
398 * of down_read(). There's a couple of places that are OK with that, but
399 * it's very much not a general-purpose interface.
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000400 */
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300401bool trylock_super(struct super_block *sb)
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000402{
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000403 if (down_read_trylock(&sb->s_umount)) {
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300404 if (!hlist_unhashed(&sb->s_instances) &&
405 sb->s_root && (sb->s_flags & MS_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000406 return true;
407 up_read(&sb->s_umount);
408 }
409
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000410 return false;
411}
412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413/**
414 * generic_shutdown_super - common helper for ->kill_sb()
415 * @sb: superblock to kill
416 *
417 * generic_shutdown_super() does all fs-independent work on superblock
418 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
419 * that need destruction out of superblock, call generic_shutdown_super()
420 * and release aforementioned objects. Note: dentries and inodes _are_
421 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700422 *
423 * Upon calling this function, the filesystem may no longer alter or
424 * rearrange the set of dentries belonging to this super_block, nor may it
425 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 */
427void generic_shutdown_super(struct super_block *sb)
428{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800429 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
David Howellsc636ebd2006-10-11 01:22:19 -0700431 if (sb->s_root) {
432 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200433 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 sb->s_flags &= ~MS_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800435
Dave Chinner74278da2015-03-04 12:37:22 -0500436 fsnotify_unmount_inodes(sb);
Tejun Heoa1a0e232016-02-29 18:28:53 -0500437 cgroup_writeback_umount();
Al Viro63997e92010-10-25 20:49:35 -0400438
439 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200441 if (sb->s_dio_done_wq) {
442 destroy_workqueue(sb->s_dio_done_wq);
443 sb->s_dio_done_wq = NULL;
444 }
445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 if (sop->put_super)
447 sop->put_super(sb);
448
Al Viro63997e92010-10-25 20:49:35 -0400449 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800450 printk("VFS: Busy inodes after unmount of %s. "
451 "Self-destruct in 5 seconds. Have a nice day...\n",
452 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 }
455 spin_lock(&sb_lock);
456 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500457 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 spin_unlock(&sb_lock);
459 up_write(&sb->s_umount);
460}
461
462EXPORT_SYMBOL(generic_shutdown_super);
463
464/**
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500465 * sget_userns - find or create a superblock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 * @type: filesystem type superblock should belong to
467 * @test: comparison callback
468 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100469 * @flags: mount flags
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500470 * @user_ns: User namespace for the super_block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 * @data: argument to each of them
472 */
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500473struct super_block *sget_userns(struct file_system_type *type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 int (*test)(struct super_block *,void *),
475 int (*set)(struct super_block *,void *),
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500476 int flags, struct user_namespace *user_ns,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 void *data)
478{
479 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700480 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 int err;
482
Eric W. Biedermand3381fa2017-02-01 06:06:16 +1300483 if (!(flags & (MS_KERNMOUNT|MS_SUBMOUNT)) &&
Eric W. Biedermana001e742016-06-06 15:48:04 -0500484 !(type->fs_flags & FS_USERNS_MOUNT) &&
485 !capable(CAP_SYS_ADMIN))
486 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487retry:
488 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700489 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800490 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700491 if (!test(old, data))
492 continue;
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500493 if (user_ns != old->s_user_ns) {
494 spin_unlock(&sb_lock);
495 if (s) {
496 up_write(&s->s_umount);
497 destroy_super(s);
498 }
499 return ERR_PTR(-EBUSY);
500 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700501 if (!grab_super(old))
502 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700503 if (s) {
504 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700505 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400506 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700507 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700508 return old;
509 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 }
511 if (!s) {
512 spin_unlock(&sb_lock);
Eric W. Biedermand3381fa2017-02-01 06:06:16 +1300513 s = alloc_super(type, (flags & ~MS_SUBMOUNT), user_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 if (!s)
515 return ERR_PTR(-ENOMEM);
516 goto retry;
517 }
518
519 err = set(s, data);
520 if (err) {
521 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700522 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 destroy_super(s);
524 return ERR_PTR(err);
525 }
526 s->s_type = type;
527 strlcpy(s->s_id, type->name, sizeof(s->s_id));
528 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500529 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 spin_unlock(&sb_lock);
531 get_filesystem(type);
Al Virofd7cbb52017-12-18 15:05:07 -0500532 err = register_shrinker(&s->s_shrink);
533 if (err) {
534 deactivate_locked_super(s);
535 s = ERR_PTR(err);
536 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 return s;
538}
539
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500540EXPORT_SYMBOL(sget_userns);
541
542/**
543 * sget - find or create a superblock
544 * @type: filesystem type superblock should belong to
545 * @test: comparison callback
546 * @set: setup callback
547 * @flags: mount flags
548 * @data: argument to each of them
549 */
550struct super_block *sget(struct file_system_type *type,
551 int (*test)(struct super_block *,void *),
552 int (*set)(struct super_block *,void *),
553 int flags,
554 void *data)
555{
556 struct user_namespace *user_ns = current_user_ns();
557
Eric W. Biedermand3381fa2017-02-01 06:06:16 +1300558 /* We don't yet pass the user namespace of the parent
559 * mount through to here so always use &init_user_ns
560 * until that changes.
561 */
562 if (flags & MS_SUBMOUNT)
563 user_ns = &init_user_ns;
564
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500565 /* Ensure the requestor has permissions over the target filesystem */
Eric W. Biedermand3381fa2017-02-01 06:06:16 +1300566 if (!(flags & (MS_KERNMOUNT|MS_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500567 return ERR_PTR(-EPERM);
568
569 return sget_userns(type, test, set, flags, user_ns, data);
570}
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572EXPORT_SYMBOL(sget);
573
574void drop_super(struct super_block *sb)
575{
576 up_read(&sb->s_umount);
577 put_super(sb);
578}
579
580EXPORT_SYMBOL(drop_super);
581
Christoph Hellwige5004752009-05-05 16:08:56 +0200582/**
Al Viro01a05b32010-03-23 06:06:58 -0400583 * iterate_supers - call function for all active superblocks
584 * @f: function to call
585 * @arg: argument to pass to it
586 *
587 * Scans the superblock list and calls given function, passing it
588 * locked superblock and given argument.
589 */
590void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
591{
Al Virodca33252010-07-25 02:31:46 +0400592 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400593
594 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400595 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500596 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400597 continue;
598 sb->s_count++;
599 spin_unlock(&sb_lock);
600
601 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500602 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400603 f(sb, arg);
604 up_read(&sb->s_umount);
605
606 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400607 if (p)
608 __put_super(p);
609 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400610 }
Al Virodca33252010-07-25 02:31:46 +0400611 if (p)
612 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400613 spin_unlock(&sb_lock);
614}
615
616/**
Al Viro43e15cd2011-06-03 20:16:57 -0400617 * iterate_supers_type - call function for superblocks of given type
618 * @type: fs type
619 * @f: function to call
620 * @arg: argument to pass to it
621 *
622 * Scans the superblock list and calls given function, passing it
623 * locked superblock and given argument.
624 */
625void iterate_supers_type(struct file_system_type *type,
626 void (*f)(struct super_block *, void *), void *arg)
627{
628 struct super_block *sb, *p = NULL;
629
630 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800631 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400632 sb->s_count++;
633 spin_unlock(&sb_lock);
634
635 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500636 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400637 f(sb, arg);
638 up_read(&sb->s_umount);
639
640 spin_lock(&sb_lock);
641 if (p)
642 __put_super(p);
643 p = sb;
644 }
645 if (p)
646 __put_super(p);
647 spin_unlock(&sb_lock);
648}
649
650EXPORT_SYMBOL(iterate_supers_type);
651
652/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 * get_super - get the superblock of a device
654 * @bdev: device to get the superblock for
655 *
656 * Scans the superblock list and finds the superblock of the file system
657 * mounted on the device given. %NULL is returned if no match is found.
658 */
659
Al Virodf40c012010-03-22 20:23:25 -0400660struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700662 struct super_block *sb;
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 if (!bdev)
665 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700668rescan:
669 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500670 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400671 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700672 if (sb->s_bdev == bdev) {
673 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700675 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400676 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500677 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700678 return sb;
679 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400680 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700681 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400682 __put_super(sb);
683 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 }
685 }
686 spin_unlock(&sb_lock);
687 return NULL;
688}
689
690EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200691
692/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100693 * get_super_thawed - get thawed superblock of a device
694 * @bdev: device to get the superblock for
695 *
696 * Scans the superblock list and finds the superblock of the file system
697 * mounted on the device. The superblock is returned once it is thawed
698 * (or immediately if it was not frozen). %NULL is returned if no match
699 * is found.
700 */
701struct super_block *get_super_thawed(struct block_device *bdev)
702{
703 while (1) {
704 struct super_block *s = get_super(bdev);
Jan Kara5accdf82012-06-12 16:20:34 +0200705 if (!s || s->s_writers.frozen == SB_UNFROZEN)
Jan Kara6b6dc832012-02-10 11:03:00 +0100706 return s;
707 up_read(&s->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +0200708 wait_event(s->s_writers.wait_unfrozen,
709 s->s_writers.frozen == SB_UNFROZEN);
Jan Kara6b6dc832012-02-10 11:03:00 +0100710 put_super(s);
711 }
712}
713EXPORT_SYMBOL(get_super_thawed);
714
715/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200716 * get_active_super - get an active reference to 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. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400721 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200722 */
723struct super_block *get_active_super(struct block_device *bdev)
724{
725 struct super_block *sb;
726
727 if (!bdev)
728 return NULL;
729
Al Viro14945832010-03-22 20:15:33 -0400730restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200731 spin_lock(&sb_lock);
732 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500733 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400734 continue;
Al Viro14945832010-03-22 20:15:33 -0400735 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400736 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400737 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400738 up_write(&sb->s_umount);
739 return sb;
Al Viro14945832010-03-22 20:15:33 -0400740 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200741 }
742 spin_unlock(&sb_lock);
743 return NULL;
744}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
Al Virodf40c012010-03-22 20:23:25 -0400746struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700748 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700751rescan:
752 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500753 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400754 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700755 if (sb->s_dev == dev) {
756 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700758 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400759 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500760 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700761 return sb;
762 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400763 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700764 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400765 __put_super(sb);
766 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 }
768 }
769 spin_unlock(&sb_lock);
770 return NULL;
771}
772
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 * do_remount_sb - asks filesystem to change mount options.
775 * @sb: superblock in question
776 * @flags: numeric part of options
777 * @data: the rest of options
778 * @force: whether or not to force the change
779 *
780 * Alters the mount options of a mounted file system.
781 */
782int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
783{
784 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400785 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200786
Jan Kara5accdf82012-06-12 16:20:34 +0200787 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200788 return -EBUSY;
789
David Howells93614012006-09-30 20:45:40 +0200790#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
792 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200793#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200794
Nick Piggind208bbd2009-12-21 16:28:53 -0800795 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800796
Al Viro0aec09d2014-08-07 07:32:06 -0400797 if (remount_ro) {
Al Virofdab6842015-01-11 10:57:27 -0500798 if (!hlist_empty(&sb->s_pins)) {
Al Viro0aec09d2014-08-07 07:32:06 -0400799 up_write(&sb->s_umount);
Al Virofdab6842015-01-11 10:57:27 -0500800 group_pin_kill(&sb->s_pins);
Al Viro0aec09d2014-08-07 07:32:06 -0400801 down_write(&sb->s_umount);
802 if (!sb->s_root)
803 return 0;
804 if (sb->s_writers.frozen != SB_UNFROZEN)
805 return -EBUSY;
806 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
807 }
808 }
809 shrink_dcache_sb(sb);
810
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 /* If we are remounting RDONLY and current sb is read/write,
812 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800813 if (remount_ro) {
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100814 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400815 sb->s_readonly_remount = 1;
816 smp_wmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100817 } else {
818 retval = sb_prepare_remount_readonly(sb);
819 if (retval)
820 return retval;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100821 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 }
823
824 if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 retval = sb->s_op->remount_fs(sb, &flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200826 if (retval) {
827 if (!force)
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100828 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200829 /* If forced remount, go ahead despite any errors */
830 WARN(1, "forced remount of a %s fs returned %i\n",
831 sb->s_type->name, retval);
832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 }
834 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100835 /* Needs to be ordered wrt mnt_is_readonly() */
836 smp_wmb();
837 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400838
Nick Piggind208bbd2009-12-21 16:28:53 -0800839 /*
840 * Some filesystems modify their metadata via some other path than the
841 * bdev buffer cache (eg. use a private mapping, or directories in
842 * pagecache, etc). Also file data modifications go via their own
843 * mappings. So If we try to mount readonly then copy the filesystem
844 * from bdev, we could get stale data, so invalidate it to give a best
845 * effort at coherency.
846 */
847 if (remount_ro && sb->s_bdev)
848 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 return 0;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100850
851cancel_readonly:
852 sb->s_readonly_remount = 0;
853 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854}
855
Jens Axboea2a95372009-03-17 09:38:40 +0100856static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857{
Al Virodca33252010-07-25 02:31:46 +0400858 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859
860 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400861 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500862 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400863 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 sb->s_count++;
865 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400866 down_write(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500867 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
868 !(sb->s_flags & MS_RDONLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 * What lock protects sb->s_flags??
871 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 }
Al Viro443b94b2009-05-05 23:48:50 -0400874 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400876 if (p)
877 __put_super(p);
878 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 }
Al Virodca33252010-07-25 02:31:46 +0400880 if (p)
881 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100883 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 printk("Emergency Remount complete\n");
885}
886
887void emergency_remount(void)
888{
Jens Axboea2a95372009-03-17 09:38:40 +0100889 struct work_struct *work;
890
891 work = kmalloc(sizeof(*work), GFP_ATOMIC);
892 if (work) {
893 INIT_WORK(work, do_emergency_remount);
894 schedule_work(work);
895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896}
897
898/*
899 * Unnamed block devices are dummy devices used by virtual
900 * filesystems which don't use real block-devices. -- jrs
901 */
902
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400903static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Thomas Bächlera2a4dc42014-04-03 21:55:37 +0200905/* Many userspace utilities consider an FSID of 0 invalid.
906 * Always return at least 1 from get_anon_bdev.
907 */
908static int unnamed_dev_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Al Viro0ee5dc62011-07-07 15:44:25 -0400910int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911{
912 int dev;
913 int error;
914
915 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400916 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 return -ENOMEM;
918 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400919 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400920 if (!error)
921 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 spin_unlock(&unnamed_dev_lock);
923 if (error == -EAGAIN)
924 /* We raced and lost with another CPU. */
925 goto retry;
926 else if (error)
927 return -EAGAIN;
928
Wang YanQing1af95de2015-06-23 18:54:45 +0800929 if (dev >= (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400931 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400932 if (unnamed_dev_start > dev)
933 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 spin_unlock(&unnamed_dev_lock);
935 return -EMFILE;
936 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400937 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 return 0;
939}
Al Viro0ee5dc62011-07-07 15:44:25 -0400940EXPORT_SYMBOL(get_anon_bdev);
941
942void free_anon_bdev(dev_t dev)
943{
944 int slot = MINOR(dev);
945 spin_lock(&unnamed_dev_lock);
946 ida_remove(&unnamed_dev_ida, slot);
947 if (slot < unnamed_dev_start)
948 unnamed_dev_start = slot;
949 spin_unlock(&unnamed_dev_lock);
950}
951EXPORT_SYMBOL(free_anon_bdev);
952
953int set_anon_super(struct super_block *s, void *data)
954{
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100955 return get_anon_bdev(&s->s_dev);
Al Viro0ee5dc62011-07-07 15:44:25 -0400956}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
958EXPORT_SYMBOL(set_anon_super);
959
960void kill_anon_super(struct super_block *sb)
961{
Al Viro0ee5dc62011-07-07 15:44:25 -0400962 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400964 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965}
966
967EXPORT_SYMBOL(kill_anon_super);
968
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969void kill_litter_super(struct super_block *sb)
970{
971 if (sb->s_root)
972 d_genocide(sb->s_root);
973 kill_anon_super(sb);
974}
975
976EXPORT_SYMBOL(kill_litter_super);
977
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700978static int ns_test_super(struct super_block *sb, void *data)
979{
980 return sb->s_fs_info == data;
981}
982
983static int ns_set_super(struct super_block *sb, void *data)
984{
985 sb->s_fs_info = data;
986 return set_anon_super(sb, NULL);
987}
988
Eric W. Biedermand91ee872016-05-23 14:51:59 -0500989struct dentry *mount_ns(struct file_system_type *fs_type,
990 int flags, void *data, void *ns, struct user_namespace *user_ns,
991 int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700992{
993 struct super_block *sb;
994
Eric W. Biedermand91ee872016-05-23 14:51:59 -0500995 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
996 * over the namespace.
997 */
998 if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
999 return ERR_PTR(-EPERM);
1000
Eric W. Biederman6e4eab52016-05-24 09:29:01 -05001001 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1002 user_ns, ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001003 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +04001004 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001005
1006 if (!sb->s_root) {
1007 int err;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001008 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
1009 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001010 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +04001011 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001012 }
1013
1014 sb->s_flags |= MS_ACTIVE;
1015 }
1016
Al Viroceefda62010-07-26 13:16:50 +04001017 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001018}
1019
Al Viroceefda62010-07-26 13:16:50 +04001020EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001021
David Howells93614012006-09-30 20:45:40 +02001022#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023static int set_bdev_super(struct super_block *s, void *data)
1024{
1025 s->s_bdev = data;
1026 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +02001027
1028 /*
1029 * We set the bdi here to the queue backing, file systems can
1030 * overwrite this in ->fill_super()
1031 */
1032 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 return 0;
1034}
1035
1036static int test_bdev_super(struct super_block *s, void *data)
1037{
1038 return (void *)s->s_bdev == data;
1039}
1040
Al Viro152a0832010-07-25 00:46:55 +04001041struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +04001043 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044{
1045 struct block_device *bdev;
1046 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +01001047 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 int error = 0;
1049
Al Viro30c40d22008-02-22 19:50:45 -05001050 if (!(flags & MS_RDONLY))
1051 mode |= FMODE_WRITE;
1052
Tejun Heod4d77622010-11-13 11:55:18 +01001053 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +04001055 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
1057 /*
1058 * once the super is inserted into the list by sget, s_umount
1059 * will protect the lockfs code from trying to start a snapshot
1060 * while we are mounting
1061 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001062 mutex_lock(&bdev->bd_fsfreeze_mutex);
1063 if (bdev->bd_fsfreeze_count > 0) {
1064 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1065 error = -EBUSY;
1066 goto error_bdev;
1067 }
David Howells9249e172012-06-25 12:55:37 +01001068 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1069 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001070 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -07001072 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
1074 if (s->s_root) {
1075 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001076 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001077 error = -EBUSY;
1078 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 }
David Howells454e2392006-06-23 02:02:57 -07001080
Tejun Heo4f331f02010-07-20 15:18:07 -07001081 /*
1082 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +01001083 * __invalidate_device(). blkdev_put() acquires
1084 * bd_mutex and can't be called under s_umount. Drop
1085 * s_umount temporarily. This is safe as we're
1086 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -07001087 */
1088 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +01001089 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -07001090 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 } else {
Al Viro30c40d22008-02-22 19:50:45 -05001092 s->s_mode = mode;
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001093 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
Pekka Enberge78c9a02006-01-08 01:03:39 -08001094 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001095 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001097 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001098 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001099 }
David Howells454e2392006-06-23 02:02:57 -07001100
1101 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001102 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 }
1104
Al Viro152a0832010-07-25 00:46:55 +04001105 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106
David Howells454e2392006-06-23 02:02:57 -07001107error_s:
1108 error = PTR_ERR(s);
1109error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001110 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001111error:
Al Viro152a0832010-07-25 00:46:55 +04001112 return ERR_PTR(error);
1113}
1114EXPORT_SYMBOL(mount_bdev);
1115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116void kill_block_super(struct super_block *sb)
1117{
1118 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001119 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001121 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 generic_shutdown_super(sb);
1123 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001124 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001125 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126}
1127
1128EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001129#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
Al Viro3c26ff62010-07-25 11:46:36 +04001131struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001133 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134{
1135 int error;
David Howells9249e172012-06-25 12:55:37 +01001136 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
1138 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001139 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001141 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001143 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001144 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 }
1146 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001147 return dget(s->s_root);
1148}
1149EXPORT_SYMBOL(mount_nodev);
1150
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151static int compare_single(struct super_block *s, void *p)
1152{
1153 return 1;
1154}
1155
Al Virofc14f2f2010-07-25 01:48:30 +04001156struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001158 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159{
1160 struct super_block *s;
1161 int error;
1162
David Howells9249e172012-06-25 12:55:37 +01001163 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001165 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 if (!s->s_root) {
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001167 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001169 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001170 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 }
1172 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001173 } else {
1174 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 }
Al Virofc14f2f2010-07-25 01:48:30 +04001176 return dget(s->s_root);
1177}
1178EXPORT_SYMBOL(mount_single);
1179
Al Viro9d412a42011-03-17 22:08:28 -04001180struct dentry *
1181mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182{
Al Viroc96e41e2010-07-25 00:17:56 +04001183 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001184 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001186 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001187
Eric Parise0007522008-03-05 10:31:54 -05001188 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001190 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001191 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
Eric Parise0007522008-03-05 10:31:54 -05001193 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001194 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 }
1197
Al Viro1a102ff2011-03-16 09:07:58 -04001198 root = type->mount(type, flags, name, data);
1199 if (IS_ERR(root)) {
1200 error = PTR_ERR(root);
1201 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001202 }
Al Viro9d412a42011-03-17 22:08:28 -04001203 sb = root->d_sb;
1204 BUG_ON(!sb);
1205 WARN_ON(!sb->s_bdi);
Dave Chinnerd6f62ec2018-05-11 11:20:57 +10001206
1207 /*
1208 * Write barrier is for super_cache_count(). We place it before setting
1209 * MS_BORN as the data dependency between the two functions is the
1210 * superblock structure contents that we just set up, not the MS_BORN
1211 * flag.
1212 */
1213 smp_wmb();
Al Viro9d412a42011-03-17 22:08:28 -04001214 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001215
Al Viro9d412a42011-03-17 22:08:28 -04001216 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001217 if (error)
1218 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001219
Jeff Layton42cb56a2009-09-18 13:05:53 -07001220 /*
1221 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1222 * but s_maxbytes was an unsigned long long for many releases. Throw
1223 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001224 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001225 */
Al Viro9d412a42011-03-17 22:08:28 -04001226 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1227 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001228
Al Viro9d412a42011-03-17 22:08:28 -04001229 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001230 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001231 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001233 dput(root);
1234 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235out_free_secdata:
1236 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237out:
David Howells454e2392006-06-23 02:02:57 -07001238 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239}
1240
Jan Kara5accdf82012-06-12 16:20:34 +02001241/*
1242 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1243 * instead.
1244 */
1245void __sb_end_write(struct super_block *sb, int level)
1246{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001247 percpu_up_read(sb->s_writers.rw_sem + level-1);
Jan Kara5accdf82012-06-12 16:20:34 +02001248}
1249EXPORT_SYMBOL(__sb_end_write);
1250
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001251/*
1252 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1253 * instead.
1254 */
1255int __sb_start_write(struct super_block *sb, int level, bool wait)
1256{
1257 bool force_trylock = false;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001258 int ret = 1;
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001259
1260#ifdef CONFIG_LOCKDEP
1261 /*
1262 * We want lockdep to tell us about possible deadlocks with freezing
1263 * but it's it bit tricky to properly instrument it. Getting a freeze
1264 * protection works as getting a read lock but there are subtle
1265 * problems. XFS for example gets freeze protection on internal level
1266 * twice in some cases, which is OK only because we already hold a
1267 * freeze protection also on higher level. Due to these cases we have
1268 * to use wait == F (trylock mode) which must not fail.
1269 */
1270 if (wait) {
1271 int i;
1272
1273 for (i = 0; i < level - 1; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001274 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001275 force_trylock = true;
1276 break;
1277 }
1278 }
1279#endif
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001280 if (wait && !force_trylock)
1281 percpu_down_read(sb->s_writers.rw_sem + level-1);
1282 else
1283 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1284
Vincent Stehlé22224a12015-09-21 17:18:34 +02001285 WARN_ON(force_trylock && !ret);
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001286 return ret;
1287}
Jan Kara5accdf82012-06-12 16:20:34 +02001288EXPORT_SYMBOL(__sb_start_write);
1289
1290/**
1291 * sb_wait_write - wait until all writers to given file system finish
1292 * @sb: the super for which we wait
1293 * @level: type of writers we wait for (normal vs page fault)
1294 *
1295 * This function waits until there are no writers of given type to given file
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001296 * system.
Jan Kara5accdf82012-06-12 16:20:34 +02001297 */
1298static void sb_wait_write(struct super_block *sb, int level)
1299{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001300 percpu_down_write(sb->s_writers.rw_sem + level-1);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001301}
Jan Kara5accdf82012-06-12 16:20:34 +02001302
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001303/*
1304 * We are going to return to userspace and forget about these locks, the
1305 * ownership goes to the caller of thaw_super() which does unlock().
1306 */
1307static void lockdep_sb_freeze_release(struct super_block *sb)
1308{
1309 int level;
1310
1311 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1312 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1313}
1314
1315/*
1316 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1317 */
1318static void lockdep_sb_freeze_acquire(struct super_block *sb)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001319{
1320 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001321
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001322 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1323 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001324}
1325
1326static void sb_freeze_unlock(struct super_block *sb)
1327{
1328 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001329
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001330 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1331 percpu_up_write(sb->s_writers.rw_sem + level);
Jan Kara5accdf82012-06-12 16:20:34 +02001332}
1333
Josef Bacik18e9e512010-03-23 10:34:56 -04001334/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001335 * freeze_super - lock the filesystem and force it into a consistent state
1336 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001337 *
1338 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1339 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1340 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001341 *
1342 * During this function, sb->s_writers.frozen goes through these values:
1343 *
1344 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1345 *
1346 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1347 * writes should be blocked, though page faults are still allowed. We wait for
1348 * all writes to complete and then proceed to the next stage.
1349 *
1350 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1351 * but internal fs threads can still modify the filesystem (although they
1352 * should not dirty new pages or inodes), writeback can run etc. After waiting
1353 * for all running page faults we sync the filesystem which will clean all
1354 * dirty pages and inodes (no new dirty pages or inodes can be created when
1355 * sync is running).
1356 *
1357 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1358 * modification are blocked (e.g. XFS preallocation truncation on inode
1359 * reclaim). This is usually implemented by blocking new transactions for
1360 * filesystems that have them and need this additional guard. After all
1361 * internal writers are finished we call ->freeze_fs() to finish filesystem
1362 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1363 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1364 *
1365 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001366 */
1367int freeze_super(struct super_block *sb)
1368{
1369 int ret;
1370
1371 atomic_inc(&sb->s_active);
1372 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001373 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001374 deactivate_locked_super(sb);
1375 return -EBUSY;
1376 }
1377
Al Virodabe0dc2012-01-03 21:01:29 -05001378 if (!(sb->s_flags & MS_BORN)) {
1379 up_write(&sb->s_umount);
1380 return 0; /* sic - it's "nothing to do" */
1381 }
1382
Josef Bacik18e9e512010-03-23 10:34:56 -04001383 if (sb->s_flags & MS_RDONLY) {
Jan Kara5accdf82012-06-12 16:20:34 +02001384 /* Nothing to do really... */
1385 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001386 up_write(&sb->s_umount);
1387 return 0;
1388 }
1389
Jan Kara5accdf82012-06-12 16:20:34 +02001390 sb->s_writers.frozen = SB_FREEZE_WRITE;
Jan Kara5accdf82012-06-12 16:20:34 +02001391 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1392 up_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001393 sb_wait_write(sb, SB_FREEZE_WRITE);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001394 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001395
1396 /* Now we go and block page faults... */
Jan Kara5accdf82012-06-12 16:20:34 +02001397 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
Jan Kara5accdf82012-06-12 16:20:34 +02001398 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1399
1400 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001401 sync_filesystem(sb);
1402
Jan Kara5accdf82012-06-12 16:20:34 +02001403 /* Now wait for internal filesystem counter */
1404 sb->s_writers.frozen = SB_FREEZE_FS;
Jan Kara5accdf82012-06-12 16:20:34 +02001405 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001406
Josef Bacik18e9e512010-03-23 10:34:56 -04001407 if (sb->s_op->freeze_fs) {
1408 ret = sb->s_op->freeze_fs(sb);
1409 if (ret) {
1410 printk(KERN_ERR
1411 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001412 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001413 sb_freeze_unlock(sb);
Jan Kara5accdf82012-06-12 16:20:34 +02001414 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001415 deactivate_locked_super(sb);
1416 return ret;
1417 }
1418 }
Jan Kara5accdf82012-06-12 16:20:34 +02001419 /*
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001420 * For debugging purposes so that fs can warn if it sees write activity
1421 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
Jan Kara5accdf82012-06-12 16:20:34 +02001422 */
1423 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001424 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001425 up_write(&sb->s_umount);
1426 return 0;
1427}
1428EXPORT_SYMBOL(freeze_super);
1429
1430/**
1431 * thaw_super -- unlock filesystem
1432 * @sb: the super to thaw
1433 *
1434 * Unlocks the filesystem and marks it writeable again after freeze_super().
1435 */
1436int thaw_super(struct super_block *sb)
1437{
1438 int error;
1439
1440 down_write(&sb->s_umount);
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001441 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001442 up_write(&sb->s_umount);
1443 return -EINVAL;
1444 }
1445
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001446 if (sb->s_flags & MS_RDONLY) {
1447 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001448 goto out;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001449 }
Josef Bacik18e9e512010-03-23 10:34:56 -04001450
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001451 lockdep_sb_freeze_acquire(sb);
1452
Josef Bacik18e9e512010-03-23 10:34:56 -04001453 if (sb->s_op->unfreeze_fs) {
1454 error = sb->s_op->unfreeze_fs(sb);
1455 if (error) {
1456 printk(KERN_ERR
1457 "VFS:Filesystem thaw failed\n");
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001458 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001459 up_write(&sb->s_umount);
1460 return error;
1461 }
1462 }
1463
Jan Kara5accdf82012-06-12 16:20:34 +02001464 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001465 sb_freeze_unlock(sb);
1466out:
Jan Kara5accdf82012-06-12 16:20:34 +02001467 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001468 deactivate_locked_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001469 return 0;
1470}
1471EXPORT_SYMBOL(thaw_super);