blob: c2ff475c1711f38aa0b4b5296c68589e68fa1fc2 [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. Biedermana001e742016-06-06 15:48:04 -0500473 if (!(flags & MS_KERNMOUNT) &&
474 !(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. Biederman6e4eab52016-05-24 09:29:01 -0500503 s = alloc_super(type, flags, 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
544 /* Ensure the requestor has permissions over the target filesystem */
545 if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
546 return ERR_PTR(-EPERM);
547
548 return sget_userns(type, test, set, flags, user_ns, data);
549}
550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551EXPORT_SYMBOL(sget);
552
553void drop_super(struct super_block *sb)
554{
555 up_read(&sb->s_umount);
556 put_super(sb);
557}
558
559EXPORT_SYMBOL(drop_super);
560
Christoph Hellwige5004752009-05-05 16:08:56 +0200561/**
Al Viro01a05b32010-03-23 06:06:58 -0400562 * iterate_supers - call function for all active superblocks
563 * @f: function to call
564 * @arg: argument to pass to it
565 *
566 * Scans the superblock list and calls given function, passing it
567 * locked superblock and given argument.
568 */
569void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
570{
Al Virodca33252010-07-25 02:31:46 +0400571 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400572
573 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400574 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500575 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400576 continue;
577 sb->s_count++;
578 spin_unlock(&sb_lock);
579
580 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500581 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400582 f(sb, arg);
583 up_read(&sb->s_umount);
584
585 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400586 if (p)
587 __put_super(p);
588 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400589 }
Al Virodca33252010-07-25 02:31:46 +0400590 if (p)
591 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400592 spin_unlock(&sb_lock);
593}
594
595/**
Al Viro43e15cd2011-06-03 20:16:57 -0400596 * iterate_supers_type - call function for superblocks of given type
597 * @type: fs type
598 * @f: function to call
599 * @arg: argument to pass to it
600 *
601 * Scans the superblock list and calls given function, passing it
602 * locked superblock and given argument.
603 */
604void iterate_supers_type(struct file_system_type *type,
605 void (*f)(struct super_block *, void *), void *arg)
606{
607 struct super_block *sb, *p = NULL;
608
609 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800610 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400611 sb->s_count++;
612 spin_unlock(&sb_lock);
613
614 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500615 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400616 f(sb, arg);
617 up_read(&sb->s_umount);
618
619 spin_lock(&sb_lock);
620 if (p)
621 __put_super(p);
622 p = sb;
623 }
624 if (p)
625 __put_super(p);
626 spin_unlock(&sb_lock);
627}
628
629EXPORT_SYMBOL(iterate_supers_type);
630
631/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 * get_super - get the superblock of a device
633 * @bdev: device to get the superblock for
634 *
635 * Scans the superblock list and finds the superblock of the file system
636 * mounted on the device given. %NULL is returned if no match is found.
637 */
638
Al Virodf40c012010-03-22 20:23:25 -0400639struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700641 struct super_block *sb;
642
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 if (!bdev)
644 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700645
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700647rescan:
648 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500649 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400650 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700651 if (sb->s_bdev == bdev) {
652 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700654 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400655 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500656 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700657 return sb;
658 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400659 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700660 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400661 __put_super(sb);
662 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 }
664 }
665 spin_unlock(&sb_lock);
666 return NULL;
667}
668
669EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200670
671/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100672 * get_super_thawed - get thawed superblock of a device
673 * @bdev: device to get the superblock for
674 *
675 * Scans the superblock list and finds the superblock of the file system
676 * mounted on the device. The superblock is returned once it is thawed
677 * (or immediately if it was not frozen). %NULL is returned if no match
678 * is found.
679 */
680struct super_block *get_super_thawed(struct block_device *bdev)
681{
682 while (1) {
683 struct super_block *s = get_super(bdev);
Jan Kara5accdf82012-06-12 16:20:34 +0200684 if (!s || s->s_writers.frozen == SB_UNFROZEN)
Jan Kara6b6dc832012-02-10 11:03:00 +0100685 return s;
686 up_read(&s->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +0200687 wait_event(s->s_writers.wait_unfrozen,
688 s->s_writers.frozen == SB_UNFROZEN);
Jan Kara6b6dc832012-02-10 11:03:00 +0100689 put_super(s);
690 }
691}
692EXPORT_SYMBOL(get_super_thawed);
693
694/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200695 * get_active_super - get an active reference to the superblock of a device
696 * @bdev: device to get the superblock for
697 *
698 * Scans the superblock list and finds the superblock of the file system
699 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400700 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200701 */
702struct super_block *get_active_super(struct block_device *bdev)
703{
704 struct super_block *sb;
705
706 if (!bdev)
707 return NULL;
708
Al Viro14945832010-03-22 20:15:33 -0400709restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200710 spin_lock(&sb_lock);
711 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500712 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400713 continue;
Al Viro14945832010-03-22 20:15:33 -0400714 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400715 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400716 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400717 up_write(&sb->s_umount);
718 return sb;
Al Viro14945832010-03-22 20:15:33 -0400719 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200720 }
721 spin_unlock(&sb_lock);
722 return NULL;
723}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
Al Virodf40c012010-03-22 20:23:25 -0400725struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700727 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700730rescan:
731 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500732 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400733 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700734 if (sb->s_dev == dev) {
735 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700737 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400738 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500739 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700740 return sb;
741 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400742 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700743 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400744 __put_super(sb);
745 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 }
747 }
748 spin_unlock(&sb_lock);
749 return NULL;
750}
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 * do_remount_sb - asks filesystem to change mount options.
754 * @sb: superblock in question
755 * @flags: numeric part of options
756 * @data: the rest of options
757 * @force: whether or not to force the change
758 *
759 * Alters the mount options of a mounted file system.
760 */
761int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
762{
763 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400764 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200765
Jan Kara5accdf82012-06-12 16:20:34 +0200766 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200767 return -EBUSY;
768
David Howells93614012006-09-30 20:45:40 +0200769#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
771 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200772#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200773
Nick Piggind208bbd2009-12-21 16:28:53 -0800774 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800775
Al Viro0aec09d2014-08-07 07:32:06 -0400776 if (remount_ro) {
Al Virofdab6842015-01-11 10:57:27 -0500777 if (!hlist_empty(&sb->s_pins)) {
Al Viro0aec09d2014-08-07 07:32:06 -0400778 up_write(&sb->s_umount);
Al Virofdab6842015-01-11 10:57:27 -0500779 group_pin_kill(&sb->s_pins);
Al Viro0aec09d2014-08-07 07:32:06 -0400780 down_write(&sb->s_umount);
781 if (!sb->s_root)
782 return 0;
783 if (sb->s_writers.frozen != SB_UNFROZEN)
784 return -EBUSY;
785 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
786 }
787 }
788 shrink_dcache_sb(sb);
789
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 /* If we are remounting RDONLY and current sb is read/write,
791 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800792 if (remount_ro) {
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100793 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400794 sb->s_readonly_remount = 1;
795 smp_wmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100796 } else {
797 retval = sb_prepare_remount_readonly(sb);
798 if (retval)
799 return retval;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100800 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 }
802
803 if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 retval = sb->s_op->remount_fs(sb, &flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200805 if (retval) {
806 if (!force)
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100807 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200808 /* If forced remount, go ahead despite any errors */
809 WARN(1, "forced remount of a %s fs returned %i\n",
810 sb->s_type->name, retval);
811 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 }
813 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100814 /* Needs to be ordered wrt mnt_is_readonly() */
815 smp_wmb();
816 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400817
Nick Piggind208bbd2009-12-21 16:28:53 -0800818 /*
819 * Some filesystems modify their metadata via some other path than the
820 * bdev buffer cache (eg. use a private mapping, or directories in
821 * pagecache, etc). Also file data modifications go via their own
822 * mappings. So If we try to mount readonly then copy the filesystem
823 * from bdev, we could get stale data, so invalidate it to give a best
824 * effort at coherency.
825 */
826 if (remount_ro && sb->s_bdev)
827 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 return 0;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100829
830cancel_readonly:
831 sb->s_readonly_remount = 0;
832 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
834
Jens Axboea2a95372009-03-17 09:38:40 +0100835static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836{
Al Virodca33252010-07-25 02:31:46 +0400837 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
839 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400840 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500841 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400842 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 sb->s_count++;
844 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400845 down_write(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500846 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
847 !(sb->s_flags & MS_RDONLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 * What lock protects sb->s_flags??
850 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 }
Al Viro443b94b2009-05-05 23:48:50 -0400853 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400855 if (p)
856 __put_super(p);
857 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 }
Al Virodca33252010-07-25 02:31:46 +0400859 if (p)
860 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100862 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 printk("Emergency Remount complete\n");
864}
865
866void emergency_remount(void)
867{
Jens Axboea2a95372009-03-17 09:38:40 +0100868 struct work_struct *work;
869
870 work = kmalloc(sizeof(*work), GFP_ATOMIC);
871 if (work) {
872 INIT_WORK(work, do_emergency_remount);
873 schedule_work(work);
874 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
877/*
878 * Unnamed block devices are dummy devices used by virtual
879 * filesystems which don't use real block-devices. -- jrs
880 */
881
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400882static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Thomas Bächlera2a4dc42014-04-03 21:55:37 +0200884/* Many userspace utilities consider an FSID of 0 invalid.
885 * Always return at least 1 from get_anon_bdev.
886 */
887static int unnamed_dev_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
Al Viro0ee5dc62011-07-07 15:44:25 -0400889int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890{
891 int dev;
892 int error;
893
894 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400895 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 return -ENOMEM;
897 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400898 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400899 if (!error)
900 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 spin_unlock(&unnamed_dev_lock);
902 if (error == -EAGAIN)
903 /* We raced and lost with another CPU. */
904 goto retry;
905 else if (error)
906 return -EAGAIN;
907
Wang YanQing1af95de2015-06-23 18:54:45 +0800908 if (dev >= (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400910 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400911 if (unnamed_dev_start > dev)
912 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 spin_unlock(&unnamed_dev_lock);
914 return -EMFILE;
915 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400916 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 return 0;
918}
Al Viro0ee5dc62011-07-07 15:44:25 -0400919EXPORT_SYMBOL(get_anon_bdev);
920
921void free_anon_bdev(dev_t dev)
922{
923 int slot = MINOR(dev);
924 spin_lock(&unnamed_dev_lock);
925 ida_remove(&unnamed_dev_ida, slot);
926 if (slot < unnamed_dev_start)
927 unnamed_dev_start = slot;
928 spin_unlock(&unnamed_dev_lock);
929}
930EXPORT_SYMBOL(free_anon_bdev);
931
932int set_anon_super(struct super_block *s, void *data)
933{
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100934 return get_anon_bdev(&s->s_dev);
Al Viro0ee5dc62011-07-07 15:44:25 -0400935}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
937EXPORT_SYMBOL(set_anon_super);
938
939void kill_anon_super(struct super_block *sb)
940{
Al Viro0ee5dc62011-07-07 15:44:25 -0400941 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400943 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944}
945
946EXPORT_SYMBOL(kill_anon_super);
947
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948void kill_litter_super(struct super_block *sb)
949{
950 if (sb->s_root)
951 d_genocide(sb->s_root);
952 kill_anon_super(sb);
953}
954
955EXPORT_SYMBOL(kill_litter_super);
956
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700957static int ns_test_super(struct super_block *sb, void *data)
958{
959 return sb->s_fs_info == data;
960}
961
962static int ns_set_super(struct super_block *sb, void *data)
963{
964 sb->s_fs_info = data;
965 return set_anon_super(sb, NULL);
966}
967
Eric W. Biedermand91ee872016-05-23 14:51:59 -0500968struct dentry *mount_ns(struct file_system_type *fs_type,
969 int flags, void *data, void *ns, struct user_namespace *user_ns,
970 int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700971{
972 struct super_block *sb;
973
Eric W. Biedermand91ee872016-05-23 14:51:59 -0500974 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
975 * over the namespace.
976 */
977 if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
978 return ERR_PTR(-EPERM);
979
Eric W. Biederman6e4eab52016-05-24 09:29:01 -0500980 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
981 user_ns, ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700982 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +0400983 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700984
985 if (!sb->s_root) {
986 int err;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700987 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
988 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400989 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +0400990 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700991 }
992
993 sb->s_flags |= MS_ACTIVE;
994 }
995
Al Viroceefda62010-07-26 13:16:50 +0400996 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700997}
998
Al Viroceefda62010-07-26 13:16:50 +0400999EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -07001000
David Howells93614012006-09-30 20:45:40 +02001001#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002static int set_bdev_super(struct super_block *s, void *data)
1003{
1004 s->s_bdev = data;
1005 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +02001006
1007 /*
1008 * We set the bdi here to the queue backing, file systems can
1009 * overwrite this in ->fill_super()
1010 */
1011 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 return 0;
1013}
1014
1015static int test_bdev_super(struct super_block *s, void *data)
1016{
1017 return (void *)s->s_bdev == data;
1018}
1019
Al Viro152a0832010-07-25 00:46:55 +04001020struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +04001022 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
1024 struct block_device *bdev;
1025 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +01001026 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 int error = 0;
1028
Al Viro30c40d22008-02-22 19:50:45 -05001029 if (!(flags & MS_RDONLY))
1030 mode |= FMODE_WRITE;
1031
Tejun Heod4d77622010-11-13 11:55:18 +01001032 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +04001034 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
1036 /*
1037 * once the super is inserted into the list by sget, s_umount
1038 * will protect the lockfs code from trying to start a snapshot
1039 * while we are mounting
1040 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001041 mutex_lock(&bdev->bd_fsfreeze_mutex);
1042 if (bdev->bd_fsfreeze_count > 0) {
1043 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1044 error = -EBUSY;
1045 goto error_bdev;
1046 }
David Howells9249e172012-06-25 12:55:37 +01001047 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1048 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001049 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -07001051 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
1053 if (s->s_root) {
1054 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001055 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001056 error = -EBUSY;
1057 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 }
David Howells454e2392006-06-23 02:02:57 -07001059
Tejun Heo4f331f02010-07-20 15:18:07 -07001060 /*
1061 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +01001062 * __invalidate_device(). blkdev_put() acquires
1063 * bd_mutex and can't be called under s_umount. Drop
1064 * s_umount temporarily. This is safe as we're
1065 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -07001066 */
1067 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +01001068 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -07001069 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 } else {
Al Viro30c40d22008-02-22 19:50:45 -05001071 s->s_mode = mode;
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001072 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
Pekka Enberge78c9a02006-01-08 01:03:39 -08001073 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001074 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001076 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001077 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001078 }
David Howells454e2392006-06-23 02:02:57 -07001079
1080 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001081 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 }
1083
Al Viro152a0832010-07-25 00:46:55 +04001084 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
David Howells454e2392006-06-23 02:02:57 -07001086error_s:
1087 error = PTR_ERR(s);
1088error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001089 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001090error:
Al Viro152a0832010-07-25 00:46:55 +04001091 return ERR_PTR(error);
1092}
1093EXPORT_SYMBOL(mount_bdev);
1094
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095void kill_block_super(struct super_block *sb)
1096{
1097 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001098 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001100 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 generic_shutdown_super(sb);
1102 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001103 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001104 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105}
1106
1107EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001108#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
Al Viro3c26ff62010-07-25 11:46:36 +04001110struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001112 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113{
1114 int error;
David Howells9249e172012-06-25 12:55:37 +01001115 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
1117 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001118 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001120 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001122 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001123 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 }
1125 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001126 return dget(s->s_root);
1127}
1128EXPORT_SYMBOL(mount_nodev);
1129
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130static int compare_single(struct super_block *s, void *p)
1131{
1132 return 1;
1133}
1134
Al Virofc14f2f2010-07-25 01:48:30 +04001135struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001137 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138{
1139 struct super_block *s;
1140 int error;
1141
David Howells9249e172012-06-25 12:55:37 +01001142 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001144 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 if (!s->s_root) {
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001146 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001148 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001149 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 }
1151 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001152 } else {
1153 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 }
Al Virofc14f2f2010-07-25 01:48:30 +04001155 return dget(s->s_root);
1156}
1157EXPORT_SYMBOL(mount_single);
1158
Al Viro9d412a42011-03-17 22:08:28 -04001159struct dentry *
1160mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161{
Al Viroc96e41e2010-07-25 00:17:56 +04001162 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001163 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001165 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001166
Eric Parise0007522008-03-05 10:31:54 -05001167 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001169 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001170 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
Eric Parise0007522008-03-05 10:31:54 -05001172 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001173 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 }
1176
Al Viro1a102ff2011-03-16 09:07:58 -04001177 root = type->mount(type, flags, name, data);
1178 if (IS_ERR(root)) {
1179 error = PTR_ERR(root);
1180 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001181 }
Al Viro9d412a42011-03-17 22:08:28 -04001182 sb = root->d_sb;
1183 BUG_ON(!sb);
1184 WARN_ON(!sb->s_bdi);
1185 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001186
Al Viro9d412a42011-03-17 22:08:28 -04001187 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001188 if (error)
1189 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001190
Jeff Layton42cb56a2009-09-18 13:05:53 -07001191 /*
1192 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1193 * but s_maxbytes was an unsigned long long for many releases. Throw
1194 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001195 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001196 */
Al Viro9d412a42011-03-17 22:08:28 -04001197 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1198 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001199
Al Viro9d412a42011-03-17 22:08:28 -04001200 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001201 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001202 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001204 dput(root);
1205 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206out_free_secdata:
1207 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208out:
David Howells454e2392006-06-23 02:02:57 -07001209 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210}
1211
Jan Kara5accdf82012-06-12 16:20:34 +02001212/*
1213 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1214 * instead.
1215 */
1216void __sb_end_write(struct super_block *sb, int level)
1217{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001218 percpu_up_read(sb->s_writers.rw_sem + level-1);
Jan Kara5accdf82012-06-12 16:20:34 +02001219}
1220EXPORT_SYMBOL(__sb_end_write);
1221
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001222/*
1223 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1224 * instead.
1225 */
1226int __sb_start_write(struct super_block *sb, int level, bool wait)
1227{
1228 bool force_trylock = false;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001229 int ret = 1;
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001230
1231#ifdef CONFIG_LOCKDEP
1232 /*
1233 * We want lockdep to tell us about possible deadlocks with freezing
1234 * but it's it bit tricky to properly instrument it. Getting a freeze
1235 * protection works as getting a read lock but there are subtle
1236 * problems. XFS for example gets freeze protection on internal level
1237 * twice in some cases, which is OK only because we already hold a
1238 * freeze protection also on higher level. Due to these cases we have
1239 * to use wait == F (trylock mode) which must not fail.
1240 */
1241 if (wait) {
1242 int i;
1243
1244 for (i = 0; i < level - 1; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001245 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001246 force_trylock = true;
1247 break;
1248 }
1249 }
1250#endif
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001251 if (wait && !force_trylock)
1252 percpu_down_read(sb->s_writers.rw_sem + level-1);
1253 else
1254 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1255
Vincent Stehlé22224a12015-09-21 17:18:34 +02001256 WARN_ON(force_trylock && !ret);
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001257 return ret;
1258}
Jan Kara5accdf82012-06-12 16:20:34 +02001259EXPORT_SYMBOL(__sb_start_write);
1260
1261/**
1262 * sb_wait_write - wait until all writers to given file system finish
1263 * @sb: the super for which we wait
1264 * @level: type of writers we wait for (normal vs page fault)
1265 *
1266 * This function waits until there are no writers of given type to given file
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001267 * system.
Jan Kara5accdf82012-06-12 16:20:34 +02001268 */
1269static void sb_wait_write(struct super_block *sb, int level)
1270{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001271 percpu_down_write(sb->s_writers.rw_sem + level-1);
Oleg Nesterov0e28e012015-08-11 16:28:29 +02001272 /*
1273 * We are going to return to userspace and forget about this lock, the
1274 * ownership goes to the caller of thaw_super() which does unlock.
1275 *
1276 * FIXME: we should do this before return from freeze_super() after we
1277 * called sync_filesystem(sb) and s_op->freeze_fs(sb), and thaw_super()
1278 * should re-acquire these locks before s_op->unfreeze_fs(sb). However
1279 * this leads to lockdep false-positives, so currently we do the early
1280 * release right after acquire.
1281 */
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001282 percpu_rwsem_release(sb->s_writers.rw_sem + level-1, 0, _THIS_IP_);
1283}
Jan Kara5accdf82012-06-12 16:20:34 +02001284
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001285static void sb_freeze_unlock(struct super_block *sb)
1286{
1287 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001288
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001289 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1290 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
Jan Kara5accdf82012-06-12 16:20:34 +02001291
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001292 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1293 percpu_up_write(sb->s_writers.rw_sem + level);
Jan Kara5accdf82012-06-12 16:20:34 +02001294}
1295
Josef Bacik18e9e512010-03-23 10:34:56 -04001296/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001297 * freeze_super - lock the filesystem and force it into a consistent state
1298 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001299 *
1300 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1301 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1302 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001303 *
1304 * During this function, sb->s_writers.frozen goes through these values:
1305 *
1306 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1307 *
1308 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1309 * writes should be blocked, though page faults are still allowed. We wait for
1310 * all writes to complete and then proceed to the next stage.
1311 *
1312 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1313 * but internal fs threads can still modify the filesystem (although they
1314 * should not dirty new pages or inodes), writeback can run etc. After waiting
1315 * for all running page faults we sync the filesystem which will clean all
1316 * dirty pages and inodes (no new dirty pages or inodes can be created when
1317 * sync is running).
1318 *
1319 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1320 * modification are blocked (e.g. XFS preallocation truncation on inode
1321 * reclaim). This is usually implemented by blocking new transactions for
1322 * filesystems that have them and need this additional guard. After all
1323 * internal writers are finished we call ->freeze_fs() to finish filesystem
1324 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1325 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1326 *
1327 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001328 */
1329int freeze_super(struct super_block *sb)
1330{
1331 int ret;
1332
1333 atomic_inc(&sb->s_active);
1334 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001335 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001336 deactivate_locked_super(sb);
1337 return -EBUSY;
1338 }
1339
Al Virodabe0dc2012-01-03 21:01:29 -05001340 if (!(sb->s_flags & MS_BORN)) {
1341 up_write(&sb->s_umount);
1342 return 0; /* sic - it's "nothing to do" */
1343 }
1344
Josef Bacik18e9e512010-03-23 10:34:56 -04001345 if (sb->s_flags & MS_RDONLY) {
Jan Kara5accdf82012-06-12 16:20:34 +02001346 /* Nothing to do really... */
1347 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001348 up_write(&sb->s_umount);
1349 return 0;
1350 }
1351
Jan Kara5accdf82012-06-12 16:20:34 +02001352 sb->s_writers.frozen = SB_FREEZE_WRITE;
Jan Kara5accdf82012-06-12 16:20:34 +02001353 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1354 up_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001355 sb_wait_write(sb, SB_FREEZE_WRITE);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001356 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001357
1358 /* Now we go and block page faults... */
Jan Kara5accdf82012-06-12 16:20:34 +02001359 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
Jan Kara5accdf82012-06-12 16:20:34 +02001360 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1361
1362 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001363 sync_filesystem(sb);
1364
Jan Kara5accdf82012-06-12 16:20:34 +02001365 /* Now wait for internal filesystem counter */
1366 sb->s_writers.frozen = SB_FREEZE_FS;
Jan Kara5accdf82012-06-12 16:20:34 +02001367 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001368
Josef Bacik18e9e512010-03-23 10:34:56 -04001369 if (sb->s_op->freeze_fs) {
1370 ret = sb->s_op->freeze_fs(sb);
1371 if (ret) {
1372 printk(KERN_ERR
1373 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001374 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001375 sb_freeze_unlock(sb);
Jan Kara5accdf82012-06-12 16:20:34 +02001376 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001377 deactivate_locked_super(sb);
1378 return ret;
1379 }
1380 }
Jan Kara5accdf82012-06-12 16:20:34 +02001381 /*
1382 * This is just for debugging purposes so that fs can warn if it
1383 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1384 */
1385 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001386 up_write(&sb->s_umount);
1387 return 0;
1388}
1389EXPORT_SYMBOL(freeze_super);
1390
1391/**
1392 * thaw_super -- unlock filesystem
1393 * @sb: the super to thaw
1394 *
1395 * Unlocks the filesystem and marks it writeable again after freeze_super().
1396 */
1397int thaw_super(struct super_block *sb)
1398{
1399 int error;
1400
1401 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001402 if (sb->s_writers.frozen == SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001403 up_write(&sb->s_umount);
1404 return -EINVAL;
1405 }
1406
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001407 if (sb->s_flags & MS_RDONLY) {
1408 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001409 goto out;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001410 }
Josef Bacik18e9e512010-03-23 10:34:56 -04001411
1412 if (sb->s_op->unfreeze_fs) {
1413 error = sb->s_op->unfreeze_fs(sb);
1414 if (error) {
1415 printk(KERN_ERR
1416 "VFS:Filesystem thaw failed\n");
Josef Bacik18e9e512010-03-23 10:34:56 -04001417 up_write(&sb->s_umount);
1418 return error;
1419 }
1420 }
1421
Jan Kara5accdf82012-06-12 16:20:34 +02001422 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001423 sb_freeze_unlock(sb);
1424out:
Jan Kara5accdf82012-06-12 16:20:34 +02001425 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001426 deactivate_locked_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001427 return 0;
1428}
1429EXPORT_SYMBOL(thaw_super);