blob: 2987fe3039708b729a653f0e85049f4ca470d15d [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 /*
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300122 * Don't call trylock_super as it is a potential
Tim Chend23da152014-06-04 16:10:47 -0700123 * scalability bottleneck. The counts could get updated
124 * between super_cache_count and super_cache_scan anyway.
125 * Call to super_cache_count with shrinker_rwsem held
Vladimir Davydov503c3582015-02-12 14:58:47 -0800126 * ensures the safety of call to list_lru_shrink_count() and
Tim Chend23da152014-06-04 16:10:47 -0700127 * s_op->nr_cached_objects().
128 */
Dave Chinner0a234c62013-08-28 10:17:57 +1000129 if (sb->s_op && sb->s_op->nr_cached_objects)
Vladimir Davydov4101b622015-02-12 14:58:51 -0800130 total_objects = sb->s_op->nr_cached_objects(sb, sc);
Dave Chinner0a234c62013-08-28 10:17:57 +1000131
Vladimir Davydov503c3582015-02-12 14:58:47 -0800132 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
133 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
Dave Chinner0a234c62013-08-28 10:17:57 +1000134
Glauber Costa55f841c2013-08-28 10:17:53 +1000135 total_objects = vfs_pressure_ratio(total_objects);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000136 return total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000137}
138
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200139static void destroy_super_work(struct work_struct *work)
140{
141 struct super_block *s = container_of(work, struct super_block,
142 destroy_work);
143 int i;
144
145 for (i = 0; i < SB_FREEZE_LEVELS; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200146 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200147 kfree(s);
148}
149
150static void destroy_super_rcu(struct rcu_head *head)
151{
152 struct super_block *s = container_of(head, struct super_block, rcu);
153 INIT_WORK(&s->destroy_work, destroy_super_work);
154 schedule_work(&s->destroy_work);
155}
156
Al Viro7eb5e882013-10-01 15:09:58 -0400157/**
158 * destroy_super - frees a superblock
159 * @s: superblock to free
160 *
161 * Frees a superblock.
162 */
163static void destroy_super(struct super_block *s)
Jan Kara5accdf82012-06-12 16:20:34 +0200164{
Al Viro7eb5e882013-10-01 15:09:58 -0400165 list_lru_destroy(&s->s_dentry_lru);
166 list_lru_destroy(&s->s_inode_lru);
Al Viro7eb5e882013-10-01 15:09:58 -0400167 security_sb_free(s);
168 WARN_ON(!list_empty(&s->s_mounts));
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500169 put_user_ns(s->s_user_ns);
Al Viro7eb5e882013-10-01 15:09:58 -0400170 kfree(s->s_subtype);
171 kfree(s->s_options);
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200172 call_rcu(&s->rcu, destroy_super_rcu);
Jan Kara5accdf82012-06-12 16:20:34 +0200173}
174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175/**
176 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700177 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100178 * @flags: the mount flags
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500179 * @user_ns: User namespace for the super_block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 *
181 * Allocates and initializes a new &struct super_block. alloc_super()
182 * returns a pointer new superblock or %NULL if allocation had failed.
183 */
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500184static struct super_block *alloc_super(struct file_system_type *type, int flags,
185 struct user_namespace *user_ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800187 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700188 static const struct super_operations default_op;
Al Viro7eb5e882013-10-01 15:09:58 -0400189 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Al Viro7eb5e882013-10-01 15:09:58 -0400191 if (!s)
192 return NULL;
193
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800194 INIT_LIST_HEAD(&s->s_mounts);
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500195 s->s_user_ns = get_user_ns(user_ns);
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800196
Al Viro7eb5e882013-10-01 15:09:58 -0400197 if (security_sb_alloc(s))
198 goto fail;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200199
Al Viro7eb5e882013-10-01 15:09:58 -0400200 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200201 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
202 sb_writers_name[i],
203 &type->s_writers_key[i]))
Al Viro7eb5e882013-10-01 15:09:58 -0400204 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 }
Al Viro7eb5e882013-10-01 15:09:58 -0400206 init_waitqueue_head(&s->s_writers.wait_unfrozen);
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100207 s->s_bdi = &noop_backing_dev_info;
Al Viro7eb5e882013-10-01 15:09:58 -0400208 s->s_flags = flags;
Eric W. Biedermancc50a072016-06-09 15:44:48 -0500209 if (s->s_user_ns != &init_user_ns)
Eric W. Biederman67690f92016-05-18 13:50:06 -0500210 s->s_iflags |= SB_I_NODEV;
Al Viro7eb5e882013-10-01 15:09:58 -0400211 INIT_HLIST_NODE(&s->s_instances);
212 INIT_HLIST_BL_HEAD(&s->s_anon);
Dave Chinnere97fedb2015-03-04 13:40:00 -0500213 mutex_init(&s->s_sync_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400214 INIT_LIST_HEAD(&s->s_inodes);
Dave Chinner74278da2015-03-04 12:37:22 -0500215 spin_lock_init(&s->s_inode_list_lock);
Dave Chinner6c60d2b2016-07-26 15:21:50 -0700216 INIT_LIST_HEAD(&s->s_inodes_wb);
217 spin_lock_init(&s->s_inode_wblist_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400218
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800219 if (list_lru_init_memcg(&s->s_dentry_lru))
Al Viro7eb5e882013-10-01 15:09:58 -0400220 goto fail;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800221 if (list_lru_init_memcg(&s->s_inode_lru))
Al Viro7eb5e882013-10-01 15:09:58 -0400222 goto fail;
223
Al Viro7eb5e882013-10-01 15:09:58 -0400224 init_rwsem(&s->s_umount);
225 lockdep_set_class(&s->s_umount, &type->s_umount_key);
226 /*
227 * sget() can have s_umount recursion.
228 *
229 * When it cannot find a suitable sb, it allocates a new
230 * one (this one), and tries again to find a suitable old
231 * one.
232 *
233 * In case that succeeds, it will acquire the s_umount
234 * lock of the old one. Since these are clearly distrinct
235 * locks, and this object isn't exposed yet, there's no
236 * risk of deadlocks.
237 *
238 * Annotate this by putting this lock in a different
239 * subclass.
240 */
241 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
242 s->s_count = 1;
243 atomic_set(&s->s_active, 1);
244 mutex_init(&s->s_vfs_rename_mutex);
245 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
246 mutex_init(&s->s_dquot.dqio_mutex);
247 mutex_init(&s->s_dquot.dqonoff_mutex);
Al Viro7eb5e882013-10-01 15:09:58 -0400248 s->s_maxbytes = MAX_NON_LFS;
249 s->s_op = &default_op;
250 s->s_time_gran = 1000000000;
Vladimir Davydov3cb29d12015-04-14 15:46:48 -0700251 s->cleancache_poolid = CLEANCACHE_NO_POOL;
Al Viro7eb5e882013-10-01 15:09:58 -0400252
253 s->s_shrink.seeks = DEFAULT_SEEKS;
254 s->s_shrink.scan_objects = super_cache_scan;
255 s->s_shrink.count_objects = super_cache_count;
256 s->s_shrink.batch = 1024;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800257 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000259
Al Viro7eb5e882013-10-01 15:09:58 -0400260fail:
261 destroy_super(s);
262 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263}
264
265/* Superblock refcounting */
266
267/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400268 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 */
Al Virof47ec3f2011-11-21 21:15:42 -0500270static void __put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 if (!--sb->s_count) {
Al Viro551de6f2010-03-22 19:36:35 -0400273 list_del_init(&sb->s_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 destroy_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277
278/**
279 * put_super - drop a temporary reference to superblock
280 * @sb: superblock in question
281 *
282 * Drops a temporary reference, frees superblock if there's no
283 * references left.
284 */
Al Virof47ec3f2011-11-21 21:15:42 -0500285static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286{
287 spin_lock(&sb_lock);
288 __put_super(sb);
289 spin_unlock(&sb_lock);
290}
291
292
293/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400294 * deactivate_locked_super - drop an active reference to superblock
295 * @s: superblock to deactivate
296 *
Masanari Iidabd7ced92016-02-02 22:31:06 +0900297 * Drops an active reference to superblock, converting it into a temporary
Al Viro1712ac82010-03-22 15:22:31 -0400298 * one if there is no other active references left. In that case we
299 * tell fs driver to shut it down and drop the temporary reference we
300 * had just acquired.
301 *
302 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400303 */
304void deactivate_locked_super(struct super_block *s)
305{
306 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400307 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400308 cleancache_invalidate_fs(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000309 unregister_shrinker(&s->s_shrink);
Dave Chinner28f2cd4f2014-06-04 16:10:46 -0700310 fs->kill_sb(s);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000311
Vladimir Davydovc0a5b562015-02-12 14:59:07 -0800312 /*
313 * Since list_lru_destroy() may sleep, we cannot call it from
314 * put_super(), where we hold the sb_lock. Therefore we destroy
315 * the lru lists right now.
316 */
317 list_lru_destroy(&s->s_dentry_lru);
318 list_lru_destroy(&s->s_inode_lru);
319
Al Viro74dbbdd2009-05-06 01:07:50 -0400320 put_filesystem(fs);
321 put_super(s);
322 } else {
323 up_write(&s->s_umount);
324 }
325}
326
327EXPORT_SYMBOL(deactivate_locked_super);
328
329/**
Al Viro1712ac82010-03-22 15:22:31 -0400330 * deactivate_super - drop an active reference to superblock
331 * @s: superblock to deactivate
332 *
333 * Variant of deactivate_locked_super(), except that superblock is *not*
334 * locked by caller. If we are going to drop the final active reference,
335 * lock will be acquired prior to that.
336 */
337void deactivate_super(struct super_block *s)
338{
339 if (!atomic_add_unless(&s->s_active, -1, 1)) {
340 down_write(&s->s_umount);
341 deactivate_locked_super(s);
342 }
343}
344
345EXPORT_SYMBOL(deactivate_super);
346
347/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 * grab_super - acquire an active reference
349 * @s: reference we are trying to make active
350 *
351 * Tries to acquire an active reference. grab_super() is used when we
352 * had just found a superblock in super_blocks or fs_type->fs_supers
353 * and want to turn it into a full-blown active reference. grab_super()
354 * is called with sb_lock held and drops it. Returns 1 in case of
355 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400356 * dying when grab_super() had been called). Note that this is only
357 * called for superblocks not in rundown mode (== ones still on ->fs_supers
358 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700360static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361{
362 s->s_count++;
363 spin_unlock(&sb_lock);
364 down_write(&s->s_umount);
Al Viroacfec9a2013-07-20 03:13:55 +0400365 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
366 put_super(s);
367 return 1;
368 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 up_write(&s->s_umount);
370 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 return 0;
372}
373
David Howellscf9a2ae2006-08-29 19:05:54 +0100374/*
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300375 * trylock_super - try to grab ->s_umount shared
Wanpeng Li331cbde2012-06-09 11:10:55 +0800376 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000377 *
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300378 * Try to prevent fs shutdown. This is used in places where we
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000379 * cannot take an active reference but we need to ensure that the
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300380 * filesystem is not shut down while we are working on it. It returns
381 * false if we cannot acquire s_umount or if we lose the race and
382 * filesystem already got into shutdown, and returns true with the s_umount
383 * lock held in read mode in case of success. On successful return,
384 * the caller must drop the s_umount lock when done.
385 *
386 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
387 * The reason why it's safe is that we are OK with doing trylock instead
388 * of down_read(). There's a couple of places that are OK with that, but
389 * it's very much not a general-purpose interface.
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000390 */
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300391bool trylock_super(struct super_block *sb)
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000392{
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000393 if (down_read_trylock(&sb->s_umount)) {
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300394 if (!hlist_unhashed(&sb->s_instances) &&
395 sb->s_root && (sb->s_flags & MS_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000396 return true;
397 up_read(&sb->s_umount);
398 }
399
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000400 return false;
401}
402
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403/**
404 * generic_shutdown_super - common helper for ->kill_sb()
405 * @sb: superblock to kill
406 *
407 * generic_shutdown_super() does all fs-independent work on superblock
408 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
409 * that need destruction out of superblock, call generic_shutdown_super()
410 * and release aforementioned objects. Note: dentries and inodes _are_
411 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700412 *
413 * Upon calling this function, the filesystem may no longer alter or
414 * rearrange the set of dentries belonging to this super_block, nor may it
415 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 */
417void generic_shutdown_super(struct super_block *sb)
418{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800419 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
David Howellsc636ebd2006-10-11 01:22:19 -0700421 if (sb->s_root) {
422 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200423 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 sb->s_flags &= ~MS_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800425
Dave Chinner74278da2015-03-04 12:37:22 -0500426 fsnotify_unmount_inodes(sb);
Tejun Heoa1a0e232016-02-29 18:28:53 -0500427 cgroup_writeback_umount();
Al Viro63997e92010-10-25 20:49:35 -0400428
429 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200431 if (sb->s_dio_done_wq) {
432 destroy_workqueue(sb->s_dio_done_wq);
433 sb->s_dio_done_wq = NULL;
434 }
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 if (sop->put_super)
437 sop->put_super(sb);
438
Al Viro63997e92010-10-25 20:49:35 -0400439 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800440 printk("VFS: Busy inodes after unmount of %s. "
441 "Self-destruct in 5 seconds. Have a nice day...\n",
442 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 }
445 spin_lock(&sb_lock);
446 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500447 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 spin_unlock(&sb_lock);
449 up_write(&sb->s_umount);
450}
451
452EXPORT_SYMBOL(generic_shutdown_super);
453
454/**
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500455 * sget_userns - find or create a superblock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 * @type: filesystem type superblock should belong to
457 * @test: comparison callback
458 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100459 * @flags: mount flags
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500460 * @user_ns: User namespace for the super_block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 * @data: argument to each of them
462 */
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500463struct super_block *sget_userns(struct file_system_type *type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 int (*test)(struct super_block *,void *),
465 int (*set)(struct super_block *,void *),
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500466 int flags, struct user_namespace *user_ns,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 void *data)
468{
469 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700470 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 int err;
472
Eric W. Biedermand3381fa2017-02-01 06:06:16 +1300473 if (!(flags & (MS_KERNMOUNT|MS_SUBMOUNT)) &&
Eric W. Biedermana001e742016-06-06 15:48:04 -0500474 !(type->fs_flags & FS_USERNS_MOUNT) &&
475 !capable(CAP_SYS_ADMIN))
476 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477retry:
478 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700479 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800480 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700481 if (!test(old, data))
482 continue;
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500483 if (user_ns != old->s_user_ns) {
484 spin_unlock(&sb_lock);
485 if (s) {
486 up_write(&s->s_umount);
487 destroy_super(s);
488 }
489 return ERR_PTR(-EBUSY);
490 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700491 if (!grab_super(old))
492 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700493 if (s) {
494 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700495 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400496 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700497 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700498 return old;
499 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 }
501 if (!s) {
502 spin_unlock(&sb_lock);
Eric W. Biedermand3381fa2017-02-01 06:06:16 +1300503 s = alloc_super(type, (flags & ~MS_SUBMOUNT), user_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 if (!s)
505 return ERR_PTR(-ENOMEM);
506 goto retry;
507 }
508
509 err = set(s, data);
510 if (err) {
511 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700512 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 destroy_super(s);
514 return ERR_PTR(err);
515 }
516 s->s_type = type;
517 strlcpy(s->s_id, type->name, sizeof(s->s_id));
518 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500519 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 spin_unlock(&sb_lock);
521 get_filesystem(type);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000522 register_shrinker(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 return s;
524}
525
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500526EXPORT_SYMBOL(sget_userns);
527
528/**
529 * sget - find or create a superblock
530 * @type: filesystem type superblock should belong to
531 * @test: comparison callback
532 * @set: setup callback
533 * @flags: mount flags
534 * @data: argument to each of them
535 */
536struct super_block *sget(struct file_system_type *type,
537 int (*test)(struct super_block *,void *),
538 int (*set)(struct super_block *,void *),
539 int flags,
540 void *data)
541{
542 struct user_namespace *user_ns = current_user_ns();
543
Eric W. Biedermand3381fa2017-02-01 06:06:16 +1300544 /* We don't yet pass the user namespace of the parent
545 * mount through to here so always use &init_user_ns
546 * until that changes.
547 */
548 if (flags & MS_SUBMOUNT)
549 user_ns = &init_user_ns;
550
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500551 /* Ensure the requestor has permissions over the target filesystem */
Eric W. Biedermand3381fa2017-02-01 06:06:16 +1300552 if (!(flags & (MS_KERNMOUNT|MS_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500553 return ERR_PTR(-EPERM);
554
555 return sget_userns(type, test, set, flags, user_ns, data);
556}
557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558EXPORT_SYMBOL(sget);
559
560void drop_super(struct super_block *sb)
561{
562 up_read(&sb->s_umount);
563 put_super(sb);
564}
565
566EXPORT_SYMBOL(drop_super);
567
Christoph Hellwige5004752009-05-05 16:08:56 +0200568/**
Al Viro01a05b32010-03-23 06:06:58 -0400569 * iterate_supers - call function for all active superblocks
570 * @f: function to call
571 * @arg: argument to pass to it
572 *
573 * Scans the superblock list and calls given function, passing it
574 * locked superblock and given argument.
575 */
576void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
577{
Al Virodca33252010-07-25 02:31:46 +0400578 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400579
580 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400581 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500582 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400583 continue;
584 sb->s_count++;
585 spin_unlock(&sb_lock);
586
587 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500588 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400589 f(sb, arg);
590 up_read(&sb->s_umount);
591
592 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400593 if (p)
594 __put_super(p);
595 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400596 }
Al Virodca33252010-07-25 02:31:46 +0400597 if (p)
598 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400599 spin_unlock(&sb_lock);
600}
601
602/**
Al Viro43e15cd2011-06-03 20:16:57 -0400603 * iterate_supers_type - call function for superblocks of given type
604 * @type: fs type
605 * @f: function to call
606 * @arg: argument to pass to it
607 *
608 * Scans the superblock list and calls given function, passing it
609 * locked superblock and given argument.
610 */
611void iterate_supers_type(struct file_system_type *type,
612 void (*f)(struct super_block *, void *), void *arg)
613{
614 struct super_block *sb, *p = NULL;
615
616 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800617 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400618 sb->s_count++;
619 spin_unlock(&sb_lock);
620
621 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500622 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400623 f(sb, arg);
624 up_read(&sb->s_umount);
625
626 spin_lock(&sb_lock);
627 if (p)
628 __put_super(p);
629 p = sb;
630 }
631 if (p)
632 __put_super(p);
633 spin_unlock(&sb_lock);
634}
635
636EXPORT_SYMBOL(iterate_supers_type);
637
638/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 * get_super - get the superblock of a device
640 * @bdev: device to get the superblock for
641 *
642 * Scans the superblock list and finds the superblock of the file system
643 * mounted on the device given. %NULL is returned if no match is found.
644 */
645
Al Virodf40c012010-03-22 20:23:25 -0400646struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700648 struct super_block *sb;
649
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 if (!bdev)
651 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700652
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700654rescan:
655 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500656 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400657 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700658 if (sb->s_bdev == bdev) {
659 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700661 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400662 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500663 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700664 return sb;
665 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400666 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700667 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400668 __put_super(sb);
669 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 }
671 }
672 spin_unlock(&sb_lock);
673 return NULL;
674}
675
676EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200677
678/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100679 * get_super_thawed - get thawed superblock of a device
680 * @bdev: device to get the superblock for
681 *
682 * Scans the superblock list and finds the superblock of the file system
683 * mounted on the device. The superblock is returned once it is thawed
684 * (or immediately if it was not frozen). %NULL is returned if no match
685 * is found.
686 */
687struct super_block *get_super_thawed(struct block_device *bdev)
688{
689 while (1) {
690 struct super_block *s = get_super(bdev);
Jan Kara5accdf82012-06-12 16:20:34 +0200691 if (!s || s->s_writers.frozen == SB_UNFROZEN)
Jan Kara6b6dc832012-02-10 11:03:00 +0100692 return s;
693 up_read(&s->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +0200694 wait_event(s->s_writers.wait_unfrozen,
695 s->s_writers.frozen == SB_UNFROZEN);
Jan Kara6b6dc832012-02-10 11:03:00 +0100696 put_super(s);
697 }
698}
699EXPORT_SYMBOL(get_super_thawed);
700
701/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200702 * get_active_super - get an active reference to the superblock of a device
703 * @bdev: device to get the superblock for
704 *
705 * Scans the superblock list and finds the superblock of the file system
706 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400707 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200708 */
709struct super_block *get_active_super(struct block_device *bdev)
710{
711 struct super_block *sb;
712
713 if (!bdev)
714 return NULL;
715
Al Viro14945832010-03-22 20:15:33 -0400716restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200717 spin_lock(&sb_lock);
718 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500719 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400720 continue;
Al Viro14945832010-03-22 20:15:33 -0400721 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400722 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400723 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400724 up_write(&sb->s_umount);
725 return sb;
Al Viro14945832010-03-22 20:15:33 -0400726 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200727 }
728 spin_unlock(&sb_lock);
729 return NULL;
730}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
Al Virodf40c012010-03-22 20:23:25 -0400732struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700734 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700737rescan:
738 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500739 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400740 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700741 if (sb->s_dev == dev) {
742 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700744 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400745 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500746 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700747 return sb;
748 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400749 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700750 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400751 __put_super(sb);
752 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 }
754 }
755 spin_unlock(&sb_lock);
756 return NULL;
757}
758
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759/**
Daniel Rosenberg29f888e2016-10-26 15:58:22 -0700760 * do_remount_sb2 - asks filesystem to change mount options.
761 * @mnt: mount we are looking at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 * @sb: superblock in question
763 * @flags: numeric part of options
764 * @data: the rest of options
765 * @force: whether or not to force the change
766 *
767 * Alters the mount options of a mounted file system.
768 */
Daniel Rosenberg29f888e2016-10-26 15:58:22 -0700769int do_remount_sb2(struct vfsmount *mnt, struct super_block *sb, int flags, void *data, int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
771 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400772 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200773
Jan Kara5accdf82012-06-12 16:20:34 +0200774 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200775 return -EBUSY;
776
David Howells93614012006-09-30 20:45:40 +0200777#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
779 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200780#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200781
Nick Piggind208bbd2009-12-21 16:28:53 -0800782 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800783
Al Viro0aec09d2014-08-07 07:32:06 -0400784 if (remount_ro) {
Al Virofdab6842015-01-11 10:57:27 -0500785 if (!hlist_empty(&sb->s_pins)) {
Al Viro0aec09d2014-08-07 07:32:06 -0400786 up_write(&sb->s_umount);
Al Virofdab6842015-01-11 10:57:27 -0500787 group_pin_kill(&sb->s_pins);
Al Viro0aec09d2014-08-07 07:32:06 -0400788 down_write(&sb->s_umount);
789 if (!sb->s_root)
790 return 0;
791 if (sb->s_writers.frozen != SB_UNFROZEN)
792 return -EBUSY;
793 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
794 }
795 }
796 shrink_dcache_sb(sb);
797
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 /* If we are remounting RDONLY and current sb is read/write,
799 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800800 if (remount_ro) {
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100801 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400802 sb->s_readonly_remount = 1;
803 smp_wmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100804 } else {
805 retval = sb_prepare_remount_readonly(sb);
806 if (retval)
807 return retval;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100808 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 }
810
Daniel Rosenberg29f888e2016-10-26 15:58:22 -0700811 if (mnt && sb->s_op->remount_fs2) {
812 retval = sb->s_op->remount_fs2(mnt, sb, &flags, data);
813 if (retval) {
814 if (!force)
815 goto cancel_readonly;
816 /* If forced remount, go ahead despite any errors */
817 WARN(1, "forced remount of a %s fs returned %i\n",
818 sb->s_type->name, retval);
819 }
820 } else if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 retval = sb->s_op->remount_fs(sb, &flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200822 if (retval) {
823 if (!force)
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100824 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200825 /* If forced remount, go ahead despite any errors */
826 WARN(1, "forced remount of a %s fs returned %i\n",
827 sb->s_type->name, retval);
828 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 }
830 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100831 /* Needs to be ordered wrt mnt_is_readonly() */
832 smp_wmb();
833 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400834
Nick Piggind208bbd2009-12-21 16:28:53 -0800835 /*
836 * Some filesystems modify their metadata via some other path than the
837 * bdev buffer cache (eg. use a private mapping, or directories in
838 * pagecache, etc). Also file data modifications go via their own
839 * mappings. So If we try to mount readonly then copy the filesystem
840 * from bdev, we could get stale data, so invalidate it to give a best
841 * effort at coherency.
842 */
843 if (remount_ro && sb->s_bdev)
844 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 return 0;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100846
847cancel_readonly:
848 sb->s_readonly_remount = 0;
849 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850}
851
Daniel Rosenberg29f888e2016-10-26 15:58:22 -0700852int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
853{
854 return do_remount_sb2(NULL, sb, flags, data, force);
855}
856
Jens Axboea2a95372009-03-17 09:38:40 +0100857static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858{
Al Virodca33252010-07-25 02:31:46 +0400859 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
861 spin_lock(&sb_lock);
Christian Poetzschebab8682015-07-24 16:42:58 +0100862 list_for_each_entry_reverse(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500863 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400864 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 sb->s_count++;
866 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400867 down_write(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500868 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
869 !(sb->s_flags & MS_RDONLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 * What lock protects sb->s_flags??
872 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 }
Al Viro443b94b2009-05-05 23:48:50 -0400875 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400877 if (p)
878 __put_super(p);
879 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 }
Al Virodca33252010-07-25 02:31:46 +0400881 if (p)
882 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100884 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 printk("Emergency Remount complete\n");
886}
887
888void emergency_remount(void)
889{
Jens Axboea2a95372009-03-17 09:38:40 +0100890 struct work_struct *work;
891
892 work = kmalloc(sizeof(*work), GFP_ATOMIC);
893 if (work) {
894 INIT_WORK(work, do_emergency_remount);
895 schedule_work(work);
896 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
899/*
900 * Unnamed block devices are dummy devices used by virtual
901 * filesystems which don't use real block-devices. -- jrs
902 */
903
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400904static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Thomas Bächlera2a4dc42014-04-03 21:55:37 +0200906/* Many userspace utilities consider an FSID of 0 invalid.
907 * Always return at least 1 from get_anon_bdev.
908 */
909static int unnamed_dev_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
Al Viro0ee5dc62011-07-07 15:44:25 -0400911int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912{
913 int dev;
914 int error;
915
916 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400917 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 return -ENOMEM;
919 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400920 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400921 if (!error)
922 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 spin_unlock(&unnamed_dev_lock);
924 if (error == -EAGAIN)
925 /* We raced and lost with another CPU. */
926 goto retry;
927 else if (error)
928 return -EAGAIN;
929
Wang YanQing1af95de2015-06-23 18:54:45 +0800930 if (dev >= (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400932 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400933 if (unnamed_dev_start > dev)
934 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 spin_unlock(&unnamed_dev_lock);
936 return -EMFILE;
937 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400938 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 return 0;
940}
Al Viro0ee5dc62011-07-07 15:44:25 -0400941EXPORT_SYMBOL(get_anon_bdev);
942
943void free_anon_bdev(dev_t dev)
944{
945 int slot = MINOR(dev);
946 spin_lock(&unnamed_dev_lock);
947 ida_remove(&unnamed_dev_ida, slot);
948 if (slot < unnamed_dev_start)
949 unnamed_dev_start = slot;
950 spin_unlock(&unnamed_dev_lock);
951}
952EXPORT_SYMBOL(free_anon_bdev);
953
954int set_anon_super(struct super_block *s, void *data)
955{
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100956 return get_anon_bdev(&s->s_dev);
Al Viro0ee5dc62011-07-07 15:44:25 -0400957}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
959EXPORT_SYMBOL(set_anon_super);
960
961void kill_anon_super(struct super_block *sb)
962{
Al Viro0ee5dc62011-07-07 15:44:25 -0400963 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400965 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966}
967
968EXPORT_SYMBOL(kill_anon_super);
969
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970void kill_litter_super(struct super_block *sb)
971{
972 if (sb->s_root)
973 d_genocide(sb->s_root);
974 kill_anon_super(sb);
975}
976
977EXPORT_SYMBOL(kill_litter_super);
978
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700979static int ns_test_super(struct super_block *sb, void *data)
980{
981 return sb->s_fs_info == data;
982}
983
984static int ns_set_super(struct super_block *sb, void *data)
985{
986 sb->s_fs_info = data;
987 return set_anon_super(sb, NULL);
988}
989
Eric W. Biedermand91ee872016-05-23 14:51:59 -0500990struct dentry *mount_ns(struct file_system_type *fs_type,
991 int flags, void *data, void *ns, struct user_namespace *user_ns,
992 int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700993{
994 struct super_block *sb;
995
Eric W. Biedermand91ee872016-05-23 14:51:59 -0500996 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
997 * over the namespace.
998 */
999 if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
1000 return ERR_PTR(-EPERM);
1001
Eric W. Biederman6e4eab52016-05-24 09:29:01 -05001002 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1003 user_ns, ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001004 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +04001005 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001006
1007 if (!sb->s_root) {
1008 int err;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001009 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
1010 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001011 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +04001012 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001013 }
1014
1015 sb->s_flags |= MS_ACTIVE;
1016 }
1017
Al Viroceefda62010-07-26 13:16:50 +04001018 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001019}
1020
Al Viroceefda62010-07-26 13:16:50 +04001021EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001022
David Howells93614012006-09-30 20:45:40 +02001023#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024static int set_bdev_super(struct super_block *s, void *data)
1025{
1026 s->s_bdev = data;
1027 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +02001028
1029 /*
1030 * We set the bdi here to the queue backing, file systems can
1031 * overwrite this in ->fill_super()
1032 */
1033 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 return 0;
1035}
1036
1037static int test_bdev_super(struct super_block *s, void *data)
1038{
1039 return (void *)s->s_bdev == data;
1040}
1041
Al Viro152a0832010-07-25 00:46:55 +04001042struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +04001044 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045{
1046 struct block_device *bdev;
1047 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +01001048 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 int error = 0;
1050
Al Viro30c40d22008-02-22 19:50:45 -05001051 if (!(flags & MS_RDONLY))
1052 mode |= FMODE_WRITE;
1053
Tejun Heod4d77622010-11-13 11:55:18 +01001054 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +04001056 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
1058 /*
1059 * once the super is inserted into the list by sget, s_umount
1060 * will protect the lockfs code from trying to start a snapshot
1061 * while we are mounting
1062 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001063 mutex_lock(&bdev->bd_fsfreeze_mutex);
1064 if (bdev->bd_fsfreeze_count > 0) {
1065 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1066 error = -EBUSY;
1067 goto error_bdev;
1068 }
David Howells9249e172012-06-25 12:55:37 +01001069 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1070 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001071 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -07001073 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
1075 if (s->s_root) {
1076 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001077 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001078 error = -EBUSY;
1079 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 }
David Howells454e2392006-06-23 02:02:57 -07001081
Tejun Heo4f331f02010-07-20 15:18:07 -07001082 /*
1083 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +01001084 * __invalidate_device(). blkdev_put() acquires
1085 * bd_mutex and can't be called under s_umount. Drop
1086 * s_umount temporarily. This is safe as we're
1087 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -07001088 */
1089 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +01001090 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -07001091 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 } else {
Al Viro30c40d22008-02-22 19:50:45 -05001093 s->s_mode = mode;
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001094 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
Pekka Enberge78c9a02006-01-08 01:03:39 -08001095 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001096 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001098 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001099 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001100 }
David Howells454e2392006-06-23 02:02:57 -07001101
1102 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001103 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 }
1105
Al Viro152a0832010-07-25 00:46:55 +04001106 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107
David Howells454e2392006-06-23 02:02:57 -07001108error_s:
1109 error = PTR_ERR(s);
1110error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001111 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001112error:
Al Viro152a0832010-07-25 00:46:55 +04001113 return ERR_PTR(error);
1114}
1115EXPORT_SYMBOL(mount_bdev);
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117void kill_block_super(struct super_block *sb)
1118{
1119 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001120 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001122 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 generic_shutdown_super(sb);
1124 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001125 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001126 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127}
1128
1129EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001130#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
Al Viro3c26ff62010-07-25 11:46:36 +04001132struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001134 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135{
1136 int error;
David Howells9249e172012-06-25 12:55:37 +01001137 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
1139 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001140 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001142 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001144 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001145 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 }
1147 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001148 return dget(s->s_root);
1149}
1150EXPORT_SYMBOL(mount_nodev);
1151
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152static int compare_single(struct super_block *s, void *p)
1153{
1154 return 1;
1155}
1156
Al Virofc14f2f2010-07-25 01:48:30 +04001157struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001159 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160{
1161 struct super_block *s;
1162 int error;
1163
David Howells9249e172012-06-25 12:55:37 +01001164 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001166 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 if (!s->s_root) {
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001168 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001170 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001171 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 }
1173 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001174 } else {
1175 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 }
Al Virofc14f2f2010-07-25 01:48:30 +04001177 return dget(s->s_root);
1178}
1179EXPORT_SYMBOL(mount_single);
1180
Al Viro9d412a42011-03-17 22:08:28 -04001181struct dentry *
Daniel Rosenberg29f888e2016-10-26 15:58:22 -07001182mount_fs(struct file_system_type *type, int flags, const char *name, struct vfsmount *mnt, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183{
Al Viroc96e41e2010-07-25 00:17:56 +04001184 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001185 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001187 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001188
Eric Parise0007522008-03-05 10:31:54 -05001189 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001191 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001192 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
Eric Parise0007522008-03-05 10:31:54 -05001194 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001195 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 }
1198
Daniel Rosenberg29f888e2016-10-26 15:58:22 -07001199 if (type->mount2)
1200 root = type->mount2(mnt, type, flags, name, data);
1201 else
1202 root = type->mount(type, flags, name, data);
Al Viro1a102ff2011-03-16 09:07:58 -04001203 if (IS_ERR(root)) {
1204 error = PTR_ERR(root);
1205 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001206 }
Al Viro9d412a42011-03-17 22:08:28 -04001207 sb = root->d_sb;
1208 BUG_ON(!sb);
1209 WARN_ON(!sb->s_bdi);
1210 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001211
Al Viro9d412a42011-03-17 22:08:28 -04001212 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001213 if (error)
1214 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001215
Jeff Layton42cb56a2009-09-18 13:05:53 -07001216 /*
1217 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1218 * but s_maxbytes was an unsigned long long for many releases. Throw
1219 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001220 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001221 */
Al Viro9d412a42011-03-17 22:08:28 -04001222 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1223 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001224
Al Viro9d412a42011-03-17 22:08:28 -04001225 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001226 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001227 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001229 dput(root);
1230 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231out_free_secdata:
1232 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233out:
David Howells454e2392006-06-23 02:02:57 -07001234 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235}
1236
Jan Kara5accdf82012-06-12 16:20:34 +02001237/*
1238 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1239 * instead.
1240 */
1241void __sb_end_write(struct super_block *sb, int level)
1242{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001243 percpu_up_read(sb->s_writers.rw_sem + level-1);
Jan Kara5accdf82012-06-12 16:20:34 +02001244}
1245EXPORT_SYMBOL(__sb_end_write);
1246
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001247/*
1248 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1249 * instead.
1250 */
1251int __sb_start_write(struct super_block *sb, int level, bool wait)
1252{
1253 bool force_trylock = false;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001254 int ret = 1;
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001255
1256#ifdef CONFIG_LOCKDEP
1257 /*
1258 * We want lockdep to tell us about possible deadlocks with freezing
1259 * but it's it bit tricky to properly instrument it. Getting a freeze
1260 * protection works as getting a read lock but there are subtle
1261 * problems. XFS for example gets freeze protection on internal level
1262 * twice in some cases, which is OK only because we already hold a
1263 * freeze protection also on higher level. Due to these cases we have
1264 * to use wait == F (trylock mode) which must not fail.
1265 */
1266 if (wait) {
1267 int i;
1268
1269 for (i = 0; i < level - 1; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001270 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001271 force_trylock = true;
1272 break;
1273 }
1274 }
1275#endif
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001276 if (wait && !force_trylock)
1277 percpu_down_read(sb->s_writers.rw_sem + level-1);
1278 else
1279 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1280
Vincent Stehlé22224a12015-09-21 17:18:34 +02001281 WARN_ON(force_trylock && !ret);
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001282 return ret;
1283}
Jan Kara5accdf82012-06-12 16:20:34 +02001284EXPORT_SYMBOL(__sb_start_write);
1285
1286/**
1287 * sb_wait_write - wait until all writers to given file system finish
1288 * @sb: the super for which we wait
1289 * @level: type of writers we wait for (normal vs page fault)
1290 *
1291 * This function waits until there are no writers of given type to given file
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001292 * system.
Jan Kara5accdf82012-06-12 16:20:34 +02001293 */
1294static void sb_wait_write(struct super_block *sb, int level)
1295{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001296 percpu_down_write(sb->s_writers.rw_sem + level-1);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001297}
Jan Kara5accdf82012-06-12 16:20:34 +02001298
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001299/*
1300 * We are going to return to userspace and forget about these locks, the
1301 * ownership goes to the caller of thaw_super() which does unlock().
1302 */
1303static void lockdep_sb_freeze_release(struct super_block *sb)
1304{
1305 int level;
1306
1307 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1308 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1309}
1310
1311/*
1312 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1313 */
1314static void lockdep_sb_freeze_acquire(struct super_block *sb)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001315{
1316 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001317
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001318 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1319 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001320}
1321
1322static void sb_freeze_unlock(struct super_block *sb)
1323{
1324 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001325
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001326 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1327 percpu_up_write(sb->s_writers.rw_sem + level);
Jan Kara5accdf82012-06-12 16:20:34 +02001328}
1329
Josef Bacik18e9e512010-03-23 10:34:56 -04001330/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001331 * freeze_super - lock the filesystem and force it into a consistent state
1332 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001333 *
1334 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1335 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1336 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001337 *
1338 * During this function, sb->s_writers.frozen goes through these values:
1339 *
1340 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1341 *
1342 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1343 * writes should be blocked, though page faults are still allowed. We wait for
1344 * all writes to complete and then proceed to the next stage.
1345 *
1346 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1347 * but internal fs threads can still modify the filesystem (although they
1348 * should not dirty new pages or inodes), writeback can run etc. After waiting
1349 * for all running page faults we sync the filesystem which will clean all
1350 * dirty pages and inodes (no new dirty pages or inodes can be created when
1351 * sync is running).
1352 *
1353 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1354 * modification are blocked (e.g. XFS preallocation truncation on inode
1355 * reclaim). This is usually implemented by blocking new transactions for
1356 * filesystems that have them and need this additional guard. After all
1357 * internal writers are finished we call ->freeze_fs() to finish filesystem
1358 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1359 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1360 *
1361 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001362 */
1363int freeze_super(struct super_block *sb)
1364{
1365 int ret;
1366
1367 atomic_inc(&sb->s_active);
1368 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001369 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001370 deactivate_locked_super(sb);
1371 return -EBUSY;
1372 }
1373
Al Virodabe0dc2012-01-03 21:01:29 -05001374 if (!(sb->s_flags & MS_BORN)) {
1375 up_write(&sb->s_umount);
1376 return 0; /* sic - it's "nothing to do" */
1377 }
1378
Josef Bacik18e9e512010-03-23 10:34:56 -04001379 if (sb->s_flags & MS_RDONLY) {
Jan Kara5accdf82012-06-12 16:20:34 +02001380 /* Nothing to do really... */
1381 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001382 up_write(&sb->s_umount);
1383 return 0;
1384 }
1385
Jan Kara5accdf82012-06-12 16:20:34 +02001386 sb->s_writers.frozen = SB_FREEZE_WRITE;
Jan Kara5accdf82012-06-12 16:20:34 +02001387 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1388 up_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001389 sb_wait_write(sb, SB_FREEZE_WRITE);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001390 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001391
1392 /* Now we go and block page faults... */
Jan Kara5accdf82012-06-12 16:20:34 +02001393 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
Jan Kara5accdf82012-06-12 16:20:34 +02001394 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1395
1396 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001397 sync_filesystem(sb);
1398
Jan Kara5accdf82012-06-12 16:20:34 +02001399 /* Now wait for internal filesystem counter */
1400 sb->s_writers.frozen = SB_FREEZE_FS;
Jan Kara5accdf82012-06-12 16:20:34 +02001401 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001402
Josef Bacik18e9e512010-03-23 10:34:56 -04001403 if (sb->s_op->freeze_fs) {
1404 ret = sb->s_op->freeze_fs(sb);
1405 if (ret) {
1406 printk(KERN_ERR
1407 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001408 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001409 sb_freeze_unlock(sb);
Jan Kara5accdf82012-06-12 16:20:34 +02001410 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001411 deactivate_locked_super(sb);
1412 return ret;
1413 }
1414 }
Jan Kara5accdf82012-06-12 16:20:34 +02001415 /*
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001416 * For debugging purposes so that fs can warn if it sees write activity
1417 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
Jan Kara5accdf82012-06-12 16:20:34 +02001418 */
1419 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001420 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001421 up_write(&sb->s_umount);
1422 return 0;
1423}
1424EXPORT_SYMBOL(freeze_super);
1425
1426/**
1427 * thaw_super -- unlock filesystem
1428 * @sb: the super to thaw
1429 *
1430 * Unlocks the filesystem and marks it writeable again after freeze_super().
1431 */
1432int thaw_super(struct super_block *sb)
1433{
1434 int error;
1435
1436 down_write(&sb->s_umount);
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001437 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001438 up_write(&sb->s_umount);
1439 return -EINVAL;
1440 }
1441
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001442 if (sb->s_flags & MS_RDONLY) {
1443 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001444 goto out;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001445 }
Josef Bacik18e9e512010-03-23 10:34:56 -04001446
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001447 lockdep_sb_freeze_acquire(sb);
1448
Josef Bacik18e9e512010-03-23 10:34:56 -04001449 if (sb->s_op->unfreeze_fs) {
1450 error = sb->s_op->unfreeze_fs(sb);
1451 if (error) {
1452 printk(KERN_ERR
1453 "VFS:Filesystem thaw failed\n");
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001454 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001455 up_write(&sb->s_umount);
1456 return error;
1457 }
1458 }
1459
Jan Kara5accdf82012-06-12 16:20:34 +02001460 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001461 sb_freeze_unlock(sb);
1462out:
Jan Kara5accdf82012-06-12 16:20:34 +02001463 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001464 deactivate_locked_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001465 return 0;
1466}
1467EXPORT_SYMBOL(thaw_super);