blob: 166c4ee0d0ed02a00591c9b00cdb69eb8759ca61 [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);
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200171 call_rcu(&s->rcu, destroy_super_rcu);
Jan Kara5accdf82012-06-12 16:20:34 +0200172}
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174/**
175 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700176 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100177 * @flags: the mount flags
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500178 * @user_ns: User namespace for the super_block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 *
180 * Allocates and initializes a new &struct super_block. alloc_super()
181 * returns a pointer new superblock or %NULL if allocation had failed.
182 */
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500183static struct super_block *alloc_super(struct file_system_type *type, int flags,
184 struct user_namespace *user_ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800186 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700187 static const struct super_operations default_op;
Al Viro7eb5e882013-10-01 15:09:58 -0400188 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
Al Viro7eb5e882013-10-01 15:09:58 -0400190 if (!s)
191 return NULL;
192
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800193 INIT_LIST_HEAD(&s->s_mounts);
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500194 s->s_user_ns = get_user_ns(user_ns);
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800195
Al Viro7eb5e882013-10-01 15:09:58 -0400196 if (security_sb_alloc(s))
197 goto fail;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200198
Al Viro7eb5e882013-10-01 15:09:58 -0400199 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200200 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
201 sb_writers_name[i],
202 &type->s_writers_key[i]))
Al Viro7eb5e882013-10-01 15:09:58 -0400203 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 }
Al Viro7eb5e882013-10-01 15:09:58 -0400205 init_waitqueue_head(&s->s_writers.wait_unfrozen);
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100206 s->s_bdi = &noop_backing_dev_info;
Al Viro7eb5e882013-10-01 15:09:58 -0400207 s->s_flags = flags;
Eric W. Biedermancc50a072016-06-09 15:44:48 -0500208 if (s->s_user_ns != &init_user_ns)
Eric W. Biederman67690f92016-05-18 13:50:06 -0500209 s->s_iflags |= SB_I_NODEV;
Al Viro7eb5e882013-10-01 15:09:58 -0400210 INIT_HLIST_NODE(&s->s_instances);
211 INIT_HLIST_BL_HEAD(&s->s_anon);
Dave Chinnere97fedb2015-03-04 13:40:00 -0500212 mutex_init(&s->s_sync_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400213 INIT_LIST_HEAD(&s->s_inodes);
Dave Chinner74278da92015-03-04 12:37:22 -0500214 spin_lock_init(&s->s_inode_list_lock);
Dave Chinner6c60d2b2016-07-26 15:21:50 -0700215 INIT_LIST_HEAD(&s->s_inodes_wb);
216 spin_lock_init(&s->s_inode_wblist_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400217
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800218 if (list_lru_init_memcg(&s->s_dentry_lru))
Al Viro7eb5e882013-10-01 15:09:58 -0400219 goto fail;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800220 if (list_lru_init_memcg(&s->s_inode_lru))
Al Viro7eb5e882013-10-01 15:09:58 -0400221 goto fail;
222
Al Viro7eb5e882013-10-01 15:09:58 -0400223 init_rwsem(&s->s_umount);
224 lockdep_set_class(&s->s_umount, &type->s_umount_key);
225 /*
226 * sget() can have s_umount recursion.
227 *
228 * When it cannot find a suitable sb, it allocates a new
229 * one (this one), and tries again to find a suitable old
230 * one.
231 *
232 * In case that succeeds, it will acquire the s_umount
233 * lock of the old one. Since these are clearly distrinct
234 * locks, and this object isn't exposed yet, there's no
235 * risk of deadlocks.
236 *
237 * Annotate this by putting this lock in a different
238 * subclass.
239 */
240 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
241 s->s_count = 1;
242 atomic_set(&s->s_active, 1);
243 mutex_init(&s->s_vfs_rename_mutex);
244 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
Jan Karabc8230e2017-06-08 14:39:48 +0200245 init_rwsem(&s->s_dquot.dqio_sem);
Al Viro7eb5e882013-10-01 15:09:58 -0400246 s->s_maxbytes = MAX_NON_LFS;
247 s->s_op = &default_op;
248 s->s_time_gran = 1000000000;
Vladimir Davydov3cb29d12015-04-14 15:46:48 -0700249 s->cleancache_poolid = CLEANCACHE_NO_POOL;
Al Viro7eb5e882013-10-01 15:09:58 -0400250
251 s->s_shrink.seeks = DEFAULT_SEEKS;
252 s->s_shrink.scan_objects = super_cache_scan;
253 s->s_shrink.count_objects = super_cache_count;
254 s->s_shrink.batch = 1024;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800255 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000257
Al Viro7eb5e882013-10-01 15:09:58 -0400258fail:
259 destroy_super(s);
260 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261}
262
263/* Superblock refcounting */
264
265/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400266 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 */
Al Virof47ec3f22011-11-21 21:15:42 -0500268static void __put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 if (!--sb->s_count) {
Al Viro551de6f2010-03-22 19:36:35 -0400271 list_del_init(&sb->s_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 destroy_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274}
275
276/**
277 * put_super - drop a temporary reference to superblock
278 * @sb: superblock in question
279 *
280 * Drops a temporary reference, frees superblock if there's no
281 * references left.
282 */
Al Virof47ec3f22011-11-21 21:15:42 -0500283static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284{
285 spin_lock(&sb_lock);
286 __put_super(sb);
287 spin_unlock(&sb_lock);
288}
289
290
291/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400292 * deactivate_locked_super - drop an active reference to superblock
293 * @s: superblock to deactivate
294 *
Masanari Iidabd7ced92016-02-02 22:31:06 +0900295 * Drops an active reference to superblock, converting it into a temporary
Al Viro1712ac82010-03-22 15:22:31 -0400296 * one if there is no other active references left. In that case we
297 * tell fs driver to shut it down and drop the temporary reference we
298 * had just acquired.
299 *
300 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400301 */
302void deactivate_locked_super(struct super_block *s)
303{
304 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400305 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400306 cleancache_invalidate_fs(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000307 unregister_shrinker(&s->s_shrink);
Dave Chinner28f2cd4f2014-06-04 16:10:46 -0700308 fs->kill_sb(s);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000309
Vladimir Davydovc0a5b562015-02-12 14:59:07 -0800310 /*
311 * Since list_lru_destroy() may sleep, we cannot call it from
312 * put_super(), where we hold the sb_lock. Therefore we destroy
313 * the lru lists right now.
314 */
315 list_lru_destroy(&s->s_dentry_lru);
316 list_lru_destroy(&s->s_inode_lru);
317
Al Viro74dbbdd2009-05-06 01:07:50 -0400318 put_filesystem(fs);
319 put_super(s);
320 } else {
321 up_write(&s->s_umount);
322 }
323}
324
325EXPORT_SYMBOL(deactivate_locked_super);
326
327/**
Al Viro1712ac82010-03-22 15:22:31 -0400328 * deactivate_super - drop an active reference to superblock
329 * @s: superblock to deactivate
330 *
331 * Variant of deactivate_locked_super(), except that superblock is *not*
332 * locked by caller. If we are going to drop the final active reference,
333 * lock will be acquired prior to that.
334 */
335void deactivate_super(struct super_block *s)
336{
337 if (!atomic_add_unless(&s->s_active, -1, 1)) {
338 down_write(&s->s_umount);
339 deactivate_locked_super(s);
340 }
341}
342
343EXPORT_SYMBOL(deactivate_super);
344
345/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 * grab_super - acquire an active reference
347 * @s: reference we are trying to make active
348 *
349 * Tries to acquire an active reference. grab_super() is used when we
350 * had just found a superblock in super_blocks or fs_type->fs_supers
351 * and want to turn it into a full-blown active reference. grab_super()
352 * is called with sb_lock held and drops it. Returns 1 in case of
353 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400354 * dying when grab_super() had been called). Note that this is only
355 * called for superblocks not in rundown mode (== ones still on ->fs_supers
356 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700358static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359{
360 s->s_count++;
361 spin_unlock(&sb_lock);
362 down_write(&s->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100363 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
Al Viroacfec9a2013-07-20 03:13:55 +0400364 put_super(s);
365 return 1;
366 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 up_write(&s->s_umount);
368 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 return 0;
370}
371
David Howellscf9a2ae2006-08-29 19:05:54 +0100372/*
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300373 * trylock_super - try to grab ->s_umount shared
Wanpeng Li331cbde2012-06-09 11:10:55 +0800374 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000375 *
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300376 * Try to prevent fs shutdown. This is used in places where we
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000377 * cannot take an active reference but we need to ensure that the
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300378 * filesystem is not shut down while we are working on it. It returns
379 * false if we cannot acquire s_umount or if we lose the race and
380 * filesystem already got into shutdown, and returns true with the s_umount
381 * lock held in read mode in case of success. On successful return,
382 * the caller must drop the s_umount lock when done.
383 *
384 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
385 * The reason why it's safe is that we are OK with doing trylock instead
386 * of down_read(). There's a couple of places that are OK with that, but
387 * it's very much not a general-purpose interface.
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000388 */
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300389bool trylock_super(struct super_block *sb)
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000390{
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000391 if (down_read_trylock(&sb->s_umount)) {
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300392 if (!hlist_unhashed(&sb->s_instances) &&
David Howellse462ec52017-07-17 08:45:35 +0100393 sb->s_root && (sb->s_flags & SB_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000394 return true;
395 up_read(&sb->s_umount);
396 }
397
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000398 return false;
399}
400
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401/**
402 * generic_shutdown_super - common helper for ->kill_sb()
403 * @sb: superblock to kill
404 *
405 * generic_shutdown_super() does all fs-independent work on superblock
406 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
407 * that need destruction out of superblock, call generic_shutdown_super()
408 * and release aforementioned objects. Note: dentries and inodes _are_
409 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700410 *
411 * Upon calling this function, the filesystem may no longer alter or
412 * rearrange the set of dentries belonging to this super_block, nor may it
413 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 */
415void generic_shutdown_super(struct super_block *sb)
416{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800417 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
David Howellsc636ebd2006-10-11 01:22:19 -0700419 if (sb->s_root) {
420 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200421 sync_filesystem(sb);
David Howellse462ec52017-07-17 08:45:35 +0100422 sb->s_flags &= ~SB_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800423
Dave Chinner74278da92015-03-04 12:37:22 -0500424 fsnotify_unmount_inodes(sb);
Tejun Heoa1a0e232016-02-29 18:28:53 -0500425 cgroup_writeback_umount();
Al Viro63997e92010-10-25 20:49:35 -0400426
427 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200429 if (sb->s_dio_done_wq) {
430 destroy_workqueue(sb->s_dio_done_wq);
431 sb->s_dio_done_wq = NULL;
432 }
433
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 if (sop->put_super)
435 sop->put_super(sb);
436
Al Viro63997e92010-10-25 20:49:35 -0400437 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800438 printk("VFS: Busy inodes after unmount of %s. "
439 "Self-destruct in 5 seconds. Have a nice day...\n",
440 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 }
443 spin_lock(&sb_lock);
444 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500445 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 spin_unlock(&sb_lock);
447 up_write(&sb->s_umount);
Jan Karac1844d52017-04-12 12:24:47 +0200448 if (sb->s_bdi != &noop_backing_dev_info) {
Jan Karafca39342017-04-12 12:24:28 +0200449 bdi_put(sb->s_bdi);
450 sb->s_bdi = &noop_backing_dev_info;
Jan Karafca39342017-04-12 12:24:28 +0200451 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452}
453
454EXPORT_SYMBOL(generic_shutdown_super);
455
456/**
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500457 * sget_userns - find or create a superblock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 * @type: filesystem type superblock should belong to
459 * @test: comparison callback
460 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100461 * @flags: mount flags
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500462 * @user_ns: User namespace for the super_block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 * @data: argument to each of them
464 */
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500465struct super_block *sget_userns(struct file_system_type *type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 int (*test)(struct super_block *,void *),
467 int (*set)(struct super_block *,void *),
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500468 int flags, struct user_namespace *user_ns,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 void *data)
470{
471 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700472 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 int err;
474
David Howellse462ec52017-07-17 08:45:35 +0100475 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) &&
Eric W. Biedermana001e742016-06-06 15:48:04 -0500476 !(type->fs_flags & FS_USERNS_MOUNT) &&
477 !capable(CAP_SYS_ADMIN))
478 return ERR_PTR(-EPERM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479retry:
480 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700481 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800482 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700483 if (!test(old, data))
484 continue;
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500485 if (user_ns != old->s_user_ns) {
486 spin_unlock(&sb_lock);
487 if (s) {
488 up_write(&s->s_umount);
489 destroy_super(s);
490 }
491 return ERR_PTR(-EBUSY);
492 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700493 if (!grab_super(old))
494 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700495 if (s) {
496 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700497 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400498 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700499 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700500 return old;
501 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 }
503 if (!s) {
504 spin_unlock(&sb_lock);
David Howellse462ec52017-07-17 08:45:35 +0100505 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 if (!s)
507 return ERR_PTR(-ENOMEM);
508 goto retry;
509 }
David Howellsdd111b32017-07-04 17:25:09 +0100510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 err = set(s, data);
512 if (err) {
513 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700514 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 destroy_super(s);
516 return ERR_PTR(err);
517 }
518 s->s_type = type;
519 strlcpy(s->s_id, type->name, sizeof(s->s_id));
520 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500521 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 spin_unlock(&sb_lock);
523 get_filesystem(type);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000524 register_shrinker(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 return s;
526}
527
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500528EXPORT_SYMBOL(sget_userns);
529
530/**
531 * sget - find or create a superblock
532 * @type: filesystem type superblock should belong to
533 * @test: comparison callback
534 * @set: setup callback
535 * @flags: mount flags
536 * @data: argument to each of them
537 */
538struct super_block *sget(struct file_system_type *type,
539 int (*test)(struct super_block *,void *),
540 int (*set)(struct super_block *,void *),
541 int flags,
542 void *data)
543{
544 struct user_namespace *user_ns = current_user_ns();
545
Eric W. Biederman93faccbb2017-02-01 06:06:16 +1300546 /* We don't yet pass the user namespace of the parent
547 * mount through to here so always use &init_user_ns
548 * until that changes.
549 */
David Howellse462ec52017-07-17 08:45:35 +0100550 if (flags & SB_SUBMOUNT)
Eric W. Biederman93faccbb2017-02-01 06:06:16 +1300551 user_ns = &init_user_ns;
552
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500553 /* Ensure the requestor has permissions over the target filesystem */
David Howellse462ec52017-07-17 08:45:35 +0100554 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500555 return ERR_PTR(-EPERM);
556
557 return sget_userns(type, test, set, flags, user_ns, data);
558}
559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560EXPORT_SYMBOL(sget);
561
562void drop_super(struct super_block *sb)
563{
564 up_read(&sb->s_umount);
565 put_super(sb);
566}
567
568EXPORT_SYMBOL(drop_super);
569
Jan Karaba6379f2016-11-23 12:53:00 +0100570void drop_super_exclusive(struct super_block *sb)
571{
572 up_write(&sb->s_umount);
573 put_super(sb);
574}
575EXPORT_SYMBOL(drop_super_exclusive);
576
Christoph Hellwige5004752009-05-05 16:08:56 +0200577/**
Al Viro01a05b32010-03-23 06:06:58 -0400578 * iterate_supers - call function for all active superblocks
579 * @f: function to call
580 * @arg: argument to pass to it
581 *
582 * Scans the superblock list and calls given function, passing it
583 * locked superblock and given argument.
584 */
585void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
586{
Al Virodca33252010-07-25 02:31:46 +0400587 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400588
589 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400590 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500591 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400592 continue;
593 sb->s_count++;
594 spin_unlock(&sb_lock);
595
596 down_read(&sb->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100597 if (sb->s_root && (sb->s_flags & SB_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400598 f(sb, arg);
599 up_read(&sb->s_umount);
600
601 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400602 if (p)
603 __put_super(p);
604 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400605 }
Al Virodca33252010-07-25 02:31:46 +0400606 if (p)
607 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400608 spin_unlock(&sb_lock);
609}
610
611/**
Al Viro43e15cd2011-06-03 20:16:57 -0400612 * iterate_supers_type - call function for superblocks of given type
613 * @type: fs type
614 * @f: function to call
615 * @arg: argument to pass to it
616 *
617 * Scans the superblock list and calls given function, passing it
618 * locked superblock and given argument.
619 */
620void iterate_supers_type(struct file_system_type *type,
621 void (*f)(struct super_block *, void *), void *arg)
622{
623 struct super_block *sb, *p = NULL;
624
625 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800626 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400627 sb->s_count++;
628 spin_unlock(&sb_lock);
629
630 down_read(&sb->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100631 if (sb->s_root && (sb->s_flags & SB_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400632 f(sb, arg);
633 up_read(&sb->s_umount);
634
635 spin_lock(&sb_lock);
636 if (p)
637 __put_super(p);
638 p = sb;
639 }
640 if (p)
641 __put_super(p);
642 spin_unlock(&sb_lock);
643}
644
645EXPORT_SYMBOL(iterate_supers_type);
646
Jan Karaba6379f2016-11-23 12:53:00 +0100647static struct super_block *__get_super(struct block_device *bdev, bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700649 struct super_block *sb;
650
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 if (!bdev)
652 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700653
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700655rescan:
656 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500657 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400658 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700659 if (sb->s_bdev == bdev) {
660 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 spin_unlock(&sb_lock);
Jan Karaba6379f2016-11-23 12:53:00 +0100662 if (!excl)
663 down_read(&sb->s_umount);
664 else
665 down_write(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400666 /* still alive? */
David Howellse462ec52017-07-17 08:45:35 +0100667 if (sb->s_root && (sb->s_flags & SB_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700668 return sb;
Jan Karaba6379f2016-11-23 12:53:00 +0100669 if (!excl)
670 up_read(&sb->s_umount);
671 else
672 up_write(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400673 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700674 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400675 __put_super(sb);
676 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 }
678 }
679 spin_unlock(&sb_lock);
680 return NULL;
681}
682
Jan Karaba6379f2016-11-23 12:53:00 +0100683/**
684 * get_super - get the superblock of a device
685 * @bdev: device to get the superblock for
686 *
687 * Scans the superblock list and finds the superblock of the file system
688 * mounted on the device given. %NULL is returned if no match is found.
689 */
690struct super_block *get_super(struct block_device *bdev)
691{
692 return __get_super(bdev, false);
693}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200695
Jan Karaba6379f2016-11-23 12:53:00 +0100696static struct super_block *__get_super_thawed(struct block_device *bdev,
697 bool excl)
698{
699 while (1) {
700 struct super_block *s = __get_super(bdev, excl);
701 if (!s || s->s_writers.frozen == SB_UNFROZEN)
702 return s;
703 if (!excl)
704 up_read(&s->s_umount);
705 else
706 up_write(&s->s_umount);
707 wait_event(s->s_writers.wait_unfrozen,
708 s->s_writers.frozen == SB_UNFROZEN);
709 put_super(s);
710 }
711}
712
Christoph Hellwig45042302009-08-03 23:28:35 +0200713/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100714 * get_super_thawed - get thawed superblock of a device
715 * @bdev: device to get the superblock for
716 *
717 * Scans the superblock list and finds the superblock of the file system
718 * mounted on the device. The superblock is returned once it is thawed
719 * (or immediately if it was not frozen). %NULL is returned if no match
720 * is found.
721 */
722struct super_block *get_super_thawed(struct block_device *bdev)
723{
Jan Karaba6379f2016-11-23 12:53:00 +0100724 return __get_super_thawed(bdev, false);
Jan Kara6b6dc832012-02-10 11:03:00 +0100725}
726EXPORT_SYMBOL(get_super_thawed);
727
728/**
Jan Karaba6379f2016-11-23 12:53:00 +0100729 * get_super_exclusive_thawed - get thawed superblock of a device
730 * @bdev: device to get the superblock for
731 *
732 * Scans the superblock list and finds the superblock of the file system
733 * mounted on the device. The superblock is returned once it is thawed
734 * (or immediately if it was not frozen) and s_umount semaphore is held
735 * in exclusive mode. %NULL is returned if no match is found.
736 */
737struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
738{
739 return __get_super_thawed(bdev, true);
740}
741EXPORT_SYMBOL(get_super_exclusive_thawed);
742
743/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200744 * get_active_super - get an active reference to the superblock of a device
745 * @bdev: device to get the superblock for
746 *
747 * Scans the superblock list and finds the superblock of the file system
748 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400749 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200750 */
751struct super_block *get_active_super(struct block_device *bdev)
752{
753 struct super_block *sb;
754
755 if (!bdev)
756 return NULL;
757
Al Viro14945832010-03-22 20:15:33 -0400758restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200759 spin_lock(&sb_lock);
760 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500761 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400762 continue;
Al Viro14945832010-03-22 20:15:33 -0400763 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400764 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400765 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400766 up_write(&sb->s_umount);
767 return sb;
Al Viro14945832010-03-22 20:15:33 -0400768 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200769 }
770 spin_unlock(&sb_lock);
771 return NULL;
772}
David Howellsdd111b32017-07-04 17:25:09 +0100773
Al Virodf40c012010-03-22 20:23:25 -0400774struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700776 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700779rescan:
780 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500781 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400782 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700783 if (sb->s_dev == dev) {
784 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700786 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400787 /* still alive? */
David Howellse462ec52017-07-17 08:45:35 +0100788 if (sb->s_root && (sb->s_flags & SB_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700789 return sb;
790 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400791 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700792 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400793 __put_super(sb);
794 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 }
796 }
797 spin_unlock(&sb_lock);
798 return NULL;
799}
800
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 * do_remount_sb - asks filesystem to change mount options.
803 * @sb: superblock in question
David Howellse462ec52017-07-17 08:45:35 +0100804 * @sb_flags: revised superblock flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 * @data: the rest of options
806 * @force: whether or not to force the change
807 *
808 * Alters the mount options of a mounted file system.
809 */
David Howellse462ec52017-07-17 08:45:35 +0100810int do_remount_sb(struct super_block *sb, int sb_flags, void *data, int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811{
812 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400813 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200814
Jan Kara5accdf82012-06-12 16:20:34 +0200815 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200816 return -EBUSY;
817
David Howells93614012006-09-30 20:45:40 +0200818#ifdef CONFIG_BLOCK
David Howellse462ec52017-07-17 08:45:35 +0100819 if (!(sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200821#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200822
David Howellse462ec52017-07-17 08:45:35 +0100823 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
Nick Piggind208bbd2009-12-21 16:28:53 -0800824
Al Viro0aec09d2014-08-07 07:32:06 -0400825 if (remount_ro) {
Al Virofdab6842015-01-11 10:57:27 -0500826 if (!hlist_empty(&sb->s_pins)) {
Al Viro0aec09d2014-08-07 07:32:06 -0400827 up_write(&sb->s_umount);
Al Virofdab6842015-01-11 10:57:27 -0500828 group_pin_kill(&sb->s_pins);
Al Viro0aec09d2014-08-07 07:32:06 -0400829 down_write(&sb->s_umount);
830 if (!sb->s_root)
831 return 0;
832 if (sb->s_writers.frozen != SB_UNFROZEN)
833 return -EBUSY;
David Howellse462ec52017-07-17 08:45:35 +0100834 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
Al Viro0aec09d2014-08-07 07:32:06 -0400835 }
836 }
837 shrink_dcache_sb(sb);
838
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 /* If we are remounting RDONLY and current sb is read/write,
840 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800841 if (remount_ro) {
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100842 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400843 sb->s_readonly_remount = 1;
844 smp_wmb();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100845 } else {
846 retval = sb_prepare_remount_readonly(sb);
847 if (retval)
848 return retval;
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100849 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 }
851
852 if (sb->s_op->remount_fs) {
David Howellse462ec52017-07-17 08:45:35 +0100853 retval = sb->s_op->remount_fs(sb, &sb_flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200854 if (retval) {
855 if (!force)
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100856 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200857 /* If forced remount, go ahead despite any errors */
858 WARN(1, "forced remount of a %s fs returned %i\n",
859 sb->s_type->name, retval);
860 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 }
David Howellse462ec52017-07-17 08:45:35 +0100862 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (sb_flags & MS_RMT_MASK);
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100863 /* Needs to be ordered wrt mnt_is_readonly() */
864 smp_wmb();
865 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400866
Nick Piggind208bbd2009-12-21 16:28:53 -0800867 /*
868 * Some filesystems modify their metadata via some other path than the
869 * bdev buffer cache (eg. use a private mapping, or directories in
870 * pagecache, etc). Also file data modifications go via their own
871 * mappings. So If we try to mount readonly then copy the filesystem
872 * from bdev, we could get stale data, so invalidate it to give a best
873 * effort at coherency.
874 */
875 if (remount_ro && sb->s_bdev)
876 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 return 0;
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100878
879cancel_readonly:
880 sb->s_readonly_remount = 0;
881 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882}
883
Jens Axboea2a95372009-03-17 09:38:40 +0100884static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885{
Al Virodca33252010-07-25 02:31:46 +0400886 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
888 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400889 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500890 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400891 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 sb->s_count++;
893 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400894 down_write(&sb->s_umount);
David Howellse462ec52017-07-17 08:45:35 +0100895 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
David Howellsbc98a422017-07-17 08:45:34 +0100896 !sb_rdonly(sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 * What lock protects sb->s_flags??
899 */
David Howellse462ec52017-07-17 08:45:35 +0100900 do_remount_sb(sb, SB_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 }
Al Viro443b94b2009-05-05 23:48:50 -0400902 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400904 if (p)
905 __put_super(p);
906 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 }
Al Virodca33252010-07-25 02:31:46 +0400908 if (p)
909 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100911 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 printk("Emergency Remount complete\n");
913}
914
915void emergency_remount(void)
916{
Jens Axboea2a95372009-03-17 09:38:40 +0100917 struct work_struct *work;
918
919 work = kmalloc(sizeof(*work), GFP_ATOMIC);
920 if (work) {
921 INIT_WORK(work, do_emergency_remount);
922 schedule_work(work);
923 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924}
925
926/*
927 * Unnamed block devices are dummy devices used by virtual
928 * filesystems which don't use real block-devices. -- jrs
929 */
930
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400931static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Thomas Bächlera2a4dc42014-04-03 21:55:37 +0200933/* Many userspace utilities consider an FSID of 0 invalid.
934 * Always return at least 1 from get_anon_bdev.
935 */
936static int unnamed_dev_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937
Al Viro0ee5dc62011-07-07 15:44:25 -0400938int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939{
940 int dev;
941 int error;
942
943 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400944 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 return -ENOMEM;
946 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400947 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400948 if (!error)
949 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 spin_unlock(&unnamed_dev_lock);
951 if (error == -EAGAIN)
952 /* We raced and lost with another CPU. */
953 goto retry;
954 else if (error)
955 return -EAGAIN;
956
Wang YanQing1af95de2015-06-23 18:54:45 +0800957 if (dev >= (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400959 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400960 if (unnamed_dev_start > dev)
961 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 spin_unlock(&unnamed_dev_lock);
963 return -EMFILE;
964 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400965 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 return 0;
967}
Al Viro0ee5dc62011-07-07 15:44:25 -0400968EXPORT_SYMBOL(get_anon_bdev);
969
970void free_anon_bdev(dev_t dev)
971{
972 int slot = MINOR(dev);
973 spin_lock(&unnamed_dev_lock);
974 ida_remove(&unnamed_dev_ida, slot);
975 if (slot < unnamed_dev_start)
976 unnamed_dev_start = slot;
977 spin_unlock(&unnamed_dev_lock);
978}
979EXPORT_SYMBOL(free_anon_bdev);
980
981int set_anon_super(struct super_block *s, void *data)
982{
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100983 return get_anon_bdev(&s->s_dev);
Al Viro0ee5dc62011-07-07 15:44:25 -0400984}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
986EXPORT_SYMBOL(set_anon_super);
987
988void kill_anon_super(struct super_block *sb)
989{
Al Viro0ee5dc62011-07-07 15:44:25 -0400990 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400992 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993}
994
995EXPORT_SYMBOL(kill_anon_super);
996
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997void kill_litter_super(struct super_block *sb)
998{
999 if (sb->s_root)
1000 d_genocide(sb->s_root);
1001 kill_anon_super(sb);
1002}
1003
1004EXPORT_SYMBOL(kill_litter_super);
1005
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001006static int ns_test_super(struct super_block *sb, void *data)
1007{
1008 return sb->s_fs_info == data;
1009}
1010
1011static int ns_set_super(struct super_block *sb, void *data)
1012{
1013 sb->s_fs_info = data;
1014 return set_anon_super(sb, NULL);
1015}
1016
Eric W. Biedermand91ee872016-05-23 14:51:59 -05001017struct dentry *mount_ns(struct file_system_type *fs_type,
1018 int flags, void *data, void *ns, struct user_namespace *user_ns,
1019 int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001020{
1021 struct super_block *sb;
1022
Eric W. Biedermand91ee872016-05-23 14:51:59 -05001023 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1024 * over the namespace.
1025 */
David Howellse462ec52017-07-17 08:45:35 +01001026 if (!(flags & SB_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
Eric W. Biedermand91ee872016-05-23 14:51:59 -05001027 return ERR_PTR(-EPERM);
1028
Eric W. Biederman6e4eab52016-05-24 09:29:01 -05001029 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1030 user_ns, ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001031 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +04001032 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001033
1034 if (!sb->s_root) {
1035 int err;
David Howellse462ec52017-07-17 08:45:35 +01001036 err = fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001037 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001038 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +04001039 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001040 }
1041
David Howellse462ec52017-07-17 08:45:35 +01001042 sb->s_flags |= SB_ACTIVE;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001043 }
1044
Al Viroceefda62010-07-26 13:16:50 +04001045 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001046}
1047
Al Viroceefda62010-07-26 13:16:50 +04001048EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001049
David Howells93614012006-09-30 20:45:40 +02001050#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051static int set_bdev_super(struct super_block *s, void *data)
1052{
1053 s->s_bdev = data;
1054 s->s_dev = s->s_bdev->bd_dev;
Jan Kara13eec232017-04-12 12:24:29 +02001055 s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
Jens Axboe32a88aa2009-09-16 15:02:33 +02001056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 return 0;
1058}
1059
1060static int test_bdev_super(struct super_block *s, void *data)
1061{
1062 return (void *)s->s_bdev == data;
1063}
1064
Al Viro152a0832010-07-25 00:46:55 +04001065struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +04001067 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068{
1069 struct block_device *bdev;
1070 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +01001071 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 int error = 0;
1073
David Howellse462ec52017-07-17 08:45:35 +01001074 if (!(flags & SB_RDONLY))
Al Viro30c40d22008-02-22 19:50:45 -05001075 mode |= FMODE_WRITE;
1076
Tejun Heod4d77622010-11-13 11:55:18 +01001077 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +04001079 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081 /*
1082 * once the super is inserted into the list by sget, s_umount
1083 * will protect the lockfs code from trying to start a snapshot
1084 * while we are mounting
1085 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001086 mutex_lock(&bdev->bd_fsfreeze_mutex);
1087 if (bdev->bd_fsfreeze_count > 0) {
1088 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1089 error = -EBUSY;
1090 goto error_bdev;
1091 }
David Howellse462ec52017-07-17 08:45:35 +01001092 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
David Howells9249e172012-06-25 12:55:37 +01001093 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001094 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -07001096 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
1098 if (s->s_root) {
David Howellse462ec52017-07-17 08:45:35 +01001099 if ((flags ^ s->s_flags) & SB_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001100 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001101 error = -EBUSY;
1102 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 }
David Howells454e2392006-06-23 02:02:57 -07001104
Tejun Heo4f331f02010-07-20 15:18:07 -07001105 /*
1106 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +01001107 * __invalidate_device(). blkdev_put() acquires
1108 * bd_mutex and can't be called under s_umount. Drop
1109 * s_umount temporarily. This is safe as we're
1110 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -07001111 */
1112 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +01001113 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -07001114 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 } else {
Al Viro30c40d22008-02-22 19:50:45 -05001116 s->s_mode = mode;
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001117 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
Pekka Enberge78c9a02006-01-08 01:03:39 -08001118 sb_set_blocksize(s, block_size(bdev));
David Howellse462ec52017-07-17 08:45:35 +01001119 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001121 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001122 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001123 }
David Howells454e2392006-06-23 02:02:57 -07001124
David Howellse462ec52017-07-17 08:45:35 +01001125 s->s_flags |= SB_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001126 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
1128
Al Viro152a0832010-07-25 00:46:55 +04001129 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
David Howells454e2392006-06-23 02:02:57 -07001131error_s:
1132 error = PTR_ERR(s);
1133error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001134 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001135error:
Al Viro152a0832010-07-25 00:46:55 +04001136 return ERR_PTR(error);
1137}
1138EXPORT_SYMBOL(mount_bdev);
1139
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140void kill_block_super(struct super_block *sb)
1141{
1142 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001143 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001145 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 generic_shutdown_super(sb);
1147 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001148 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001149 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150}
1151
1152EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001153#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154
Al Viro3c26ff62010-07-25 11:46:36 +04001155struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001157 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158{
1159 int error;
David Howells9249e172012-06-25 12:55:37 +01001160 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
1162 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001163 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
David Howellse462ec52017-07-17 08:45:35 +01001165 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001167 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001168 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 }
David Howellse462ec52017-07-17 08:45:35 +01001170 s->s_flags |= SB_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001171 return dget(s->s_root);
1172}
1173EXPORT_SYMBOL(mount_nodev);
1174
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175static int compare_single(struct super_block *s, void *p)
1176{
1177 return 1;
1178}
1179
Al Virofc14f2f2010-07-25 01:48:30 +04001180struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001182 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183{
1184 struct super_block *s;
1185 int error;
1186
David Howells9249e172012-06-25 12:55:37 +01001187 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001189 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 if (!s->s_root) {
David Howellse462ec52017-07-17 08:45:35 +01001191 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001193 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001194 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 }
David Howellse462ec52017-07-17 08:45:35 +01001196 s->s_flags |= SB_ACTIVE;
Kay Sievers9329d1be2009-12-18 21:18:15 +01001197 } else {
1198 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 }
Al Virofc14f2f2010-07-25 01:48:30 +04001200 return dget(s->s_root);
1201}
1202EXPORT_SYMBOL(mount_single);
1203
Al Viro9d412a42011-03-17 22:08:28 -04001204struct dentry *
1205mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206{
Al Viroc96e41e2010-07-25 00:17:56 +04001207 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001208 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001210 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001211
Eric Parise0007522008-03-05 10:31:54 -05001212 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001214 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001215 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216
Eric Parise0007522008-03-05 10:31:54 -05001217 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001218 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 }
1221
Al Viro1a102ff2011-03-16 09:07:58 -04001222 root = type->mount(type, flags, name, data);
1223 if (IS_ERR(root)) {
1224 error = PTR_ERR(root);
1225 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001226 }
Al Viro9d412a42011-03-17 22:08:28 -04001227 sb = root->d_sb;
1228 BUG_ON(!sb);
1229 WARN_ON(!sb->s_bdi);
David Howellse462ec52017-07-17 08:45:35 +01001230 sb->s_flags |= SB_BORN;
David Howells454e2392006-06-23 02:02:57 -07001231
Al Viro9d412a42011-03-17 22:08:28 -04001232 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001233 if (error)
1234 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001235
Jeff Layton42cb56a2009-09-18 13:05:53 -07001236 /*
1237 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1238 * but s_maxbytes was an unsigned long long for many releases. Throw
1239 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001240 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001241 */
Al Viro9d412a42011-03-17 22:08:28 -04001242 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1243 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001244
Al Viro9d412a42011-03-17 22:08:28 -04001245 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001246 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001247 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001249 dput(root);
1250 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251out_free_secdata:
1252 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253out:
David Howells454e2392006-06-23 02:02:57 -07001254 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255}
1256
Jan Kara5accdf82012-06-12 16:20:34 +02001257/*
Jan Karafca39342017-04-12 12:24:28 +02001258 * Setup private BDI for given superblock. It gets automatically cleaned up
1259 * in generic_shutdown_super().
1260 */
1261int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1262{
1263 struct backing_dev_info *bdi;
1264 int err;
1265 va_list args;
1266
1267 bdi = bdi_alloc(GFP_KERNEL);
1268 if (!bdi)
1269 return -ENOMEM;
1270
1271 bdi->name = sb->s_type->name;
1272
1273 va_start(args, fmt);
Jan Kara7c4cc302017-04-12 12:24:49 +02001274 err = bdi_register_va(bdi, fmt, args);
Jan Karafca39342017-04-12 12:24:28 +02001275 va_end(args);
1276 if (err) {
1277 bdi_put(bdi);
1278 return err;
1279 }
1280 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1281 sb->s_bdi = bdi;
Jan Karafca39342017-04-12 12:24:28 +02001282
1283 return 0;
1284}
1285EXPORT_SYMBOL(super_setup_bdi_name);
1286
1287/*
1288 * Setup private BDI for given superblock. I gets automatically cleaned up
1289 * in generic_shutdown_super().
1290 */
1291int super_setup_bdi(struct super_block *sb)
1292{
1293 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1294
1295 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1296 atomic_long_inc_return(&bdi_seq));
1297}
1298EXPORT_SYMBOL(super_setup_bdi);
1299
1300/*
Jan Kara5accdf82012-06-12 16:20:34 +02001301 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1302 * instead.
1303 */
1304void __sb_end_write(struct super_block *sb, int level)
1305{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001306 percpu_up_read(sb->s_writers.rw_sem + level-1);
Jan Kara5accdf82012-06-12 16:20:34 +02001307}
1308EXPORT_SYMBOL(__sb_end_write);
1309
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001310/*
1311 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1312 * instead.
1313 */
1314int __sb_start_write(struct super_block *sb, int level, bool wait)
1315{
1316 bool force_trylock = false;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001317 int ret = 1;
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001318
1319#ifdef CONFIG_LOCKDEP
1320 /*
1321 * We want lockdep to tell us about possible deadlocks with freezing
1322 * but it's it bit tricky to properly instrument it. Getting a freeze
1323 * protection works as getting a read lock but there are subtle
1324 * problems. XFS for example gets freeze protection on internal level
1325 * twice in some cases, which is OK only because we already hold a
1326 * freeze protection also on higher level. Due to these cases we have
1327 * to use wait == F (trylock mode) which must not fail.
1328 */
1329 if (wait) {
1330 int i;
1331
1332 for (i = 0; i < level - 1; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001333 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001334 force_trylock = true;
1335 break;
1336 }
1337 }
1338#endif
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001339 if (wait && !force_trylock)
1340 percpu_down_read(sb->s_writers.rw_sem + level-1);
1341 else
1342 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1343
Vincent Stehlé22224a12015-09-21 17:18:34 +02001344 WARN_ON(force_trylock && !ret);
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001345 return ret;
1346}
Jan Kara5accdf82012-06-12 16:20:34 +02001347EXPORT_SYMBOL(__sb_start_write);
1348
1349/**
1350 * sb_wait_write - wait until all writers to given file system finish
1351 * @sb: the super for which we wait
1352 * @level: type of writers we wait for (normal vs page fault)
1353 *
1354 * This function waits until there are no writers of given type to given file
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001355 * system.
Jan Kara5accdf82012-06-12 16:20:34 +02001356 */
1357static void sb_wait_write(struct super_block *sb, int level)
1358{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001359 percpu_down_write(sb->s_writers.rw_sem + level-1);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001360}
Jan Kara5accdf82012-06-12 16:20:34 +02001361
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001362/*
1363 * We are going to return to userspace and forget about these locks, the
1364 * ownership goes to the caller of thaw_super() which does unlock().
1365 */
1366static void lockdep_sb_freeze_release(struct super_block *sb)
1367{
1368 int level;
1369
1370 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1371 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1372}
1373
1374/*
1375 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1376 */
1377static void lockdep_sb_freeze_acquire(struct super_block *sb)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001378{
1379 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001380
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001381 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1382 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001383}
1384
1385static void sb_freeze_unlock(struct super_block *sb)
1386{
1387 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001388
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001389 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1390 percpu_up_write(sb->s_writers.rw_sem + level);
Jan Kara5accdf82012-06-12 16:20:34 +02001391}
1392
Josef Bacik18e9e512010-03-23 10:34:56 -04001393/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001394 * freeze_super - lock the filesystem and force it into a consistent state
1395 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001396 *
1397 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1398 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1399 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001400 *
1401 * During this function, sb->s_writers.frozen goes through these values:
1402 *
1403 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1404 *
1405 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1406 * writes should be blocked, though page faults are still allowed. We wait for
1407 * all writes to complete and then proceed to the next stage.
1408 *
1409 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1410 * but internal fs threads can still modify the filesystem (although they
1411 * should not dirty new pages or inodes), writeback can run etc. After waiting
1412 * for all running page faults we sync the filesystem which will clean all
1413 * dirty pages and inodes (no new dirty pages or inodes can be created when
1414 * sync is running).
1415 *
1416 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1417 * modification are blocked (e.g. XFS preallocation truncation on inode
1418 * reclaim). This is usually implemented by blocking new transactions for
1419 * filesystems that have them and need this additional guard. After all
1420 * internal writers are finished we call ->freeze_fs() to finish filesystem
1421 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1422 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1423 *
1424 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001425 */
1426int freeze_super(struct super_block *sb)
1427{
1428 int ret;
1429
1430 atomic_inc(&sb->s_active);
1431 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001432 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001433 deactivate_locked_super(sb);
1434 return -EBUSY;
1435 }
1436
David Howellse462ec52017-07-17 08:45:35 +01001437 if (!(sb->s_flags & SB_BORN)) {
Al Virodabe0dc2012-01-03 21:01:29 -05001438 up_write(&sb->s_umount);
1439 return 0; /* sic - it's "nothing to do" */
1440 }
1441
David Howellsbc98a422017-07-17 08:45:34 +01001442 if (sb_rdonly(sb)) {
Jan Kara5accdf82012-06-12 16:20:34 +02001443 /* Nothing to do really... */
1444 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001445 up_write(&sb->s_umount);
1446 return 0;
1447 }
1448
Jan Kara5accdf82012-06-12 16:20:34 +02001449 sb->s_writers.frozen = SB_FREEZE_WRITE;
Jan Kara5accdf82012-06-12 16:20:34 +02001450 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1451 up_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001452 sb_wait_write(sb, SB_FREEZE_WRITE);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001453 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001454
1455 /* Now we go and block page faults... */
Jan Kara5accdf82012-06-12 16:20:34 +02001456 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
Jan Kara5accdf82012-06-12 16:20:34 +02001457 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1458
1459 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001460 sync_filesystem(sb);
1461
Jan Kara5accdf82012-06-12 16:20:34 +02001462 /* Now wait for internal filesystem counter */
1463 sb->s_writers.frozen = SB_FREEZE_FS;
Jan Kara5accdf82012-06-12 16:20:34 +02001464 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001465
Josef Bacik18e9e512010-03-23 10:34:56 -04001466 if (sb->s_op->freeze_fs) {
1467 ret = sb->s_op->freeze_fs(sb);
1468 if (ret) {
1469 printk(KERN_ERR
1470 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001471 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001472 sb_freeze_unlock(sb);
Jan Kara5accdf82012-06-12 16:20:34 +02001473 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001474 deactivate_locked_super(sb);
1475 return ret;
1476 }
1477 }
Jan Kara5accdf82012-06-12 16:20:34 +02001478 /*
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001479 * For debugging purposes so that fs can warn if it sees write activity
1480 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
Jan Kara5accdf82012-06-12 16:20:34 +02001481 */
1482 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001483 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001484 up_write(&sb->s_umount);
1485 return 0;
1486}
1487EXPORT_SYMBOL(freeze_super);
1488
1489/**
1490 * thaw_super -- unlock filesystem
1491 * @sb: the super to thaw
1492 *
1493 * Unlocks the filesystem and marks it writeable again after freeze_super().
1494 */
1495int thaw_super(struct super_block *sb)
1496{
1497 int error;
1498
1499 down_write(&sb->s_umount);
Oleg Nesterov89f39af2016-09-26 18:07:48 +02001500 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001501 up_write(&sb->s_umount);
1502 return -EINVAL;
1503 }
1504
David Howellsbc98a422017-07-17 08:45:34 +01001505 if (sb_rdonly(sb)) {
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001506 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001507 goto out;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001508 }
Josef Bacik18e9e512010-03-23 10:34:56 -04001509
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001510 lockdep_sb_freeze_acquire(sb);
1511
Josef Bacik18e9e512010-03-23 10:34:56 -04001512 if (sb->s_op->unfreeze_fs) {
1513 error = sb->s_op->unfreeze_fs(sb);
1514 if (error) {
1515 printk(KERN_ERR
1516 "VFS:Filesystem thaw failed\n");
Oleg Nesterovf1a96222016-09-26 18:55:25 +02001517 lockdep_sb_freeze_release(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001518 up_write(&sb->s_umount);
1519 return error;
1520 }
1521 }
1522
Jan Kara5accdf82012-06-12 16:20:34 +02001523 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001524 sb_freeze_unlock(sb);
1525out:
Jan Kara5accdf82012-06-12 16:20:34 +02001526 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001527 deactivate_locked_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001528 return 0;
1529}
1530EXPORT_SYMBOL(thaw_super);