blob: 74914b1bae70ff09d451450cb596fb3b5639d969 [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>
Al Viro6d59e7f2008-03-22 15:48:17 -040036#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38
Al Viro15d0f5e2015-02-02 10:07:59 -070039static LIST_HEAD(super_blocks);
40static DEFINE_SPINLOCK(sb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Jan Kara5accdf82012-06-12 16:20:34 +020042static char *sb_writers_name[SB_FREEZE_LEVELS] = {
43 "sb_writers",
44 "sb_pagefaults",
45 "sb_internal",
46};
47
Dave Chinnerb0d40c92011-07-08 14:14:42 +100048/*
49 * One thing we have to be careful of with a per-sb shrinker is that we don't
50 * drop the last active reference to the superblock from within the shrinker.
51 * If that happens we could trigger unregistering the shrinker from within the
52 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
53 * take a passive reference to the superblock to avoid this from occurring.
54 */
Dave Chinner0a234c62013-08-28 10:17:57 +100055static unsigned long super_cache_scan(struct shrinker *shrink,
56 struct shrink_control *sc)
Dave Chinnerb0d40c92011-07-08 14:14:42 +100057{
58 struct super_block *sb;
Dave Chinner0a234c62013-08-28 10:17:57 +100059 long fs_objects = 0;
60 long total_objects;
61 long freed = 0;
62 long dentries;
63 long inodes;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100064
65 sb = container_of(shrink, struct super_block, s_shrink);
66
67 /*
68 * Deadlock avoidance. We may hold various FS locks, and we don't want
69 * to recurse into the FS that called us in clear_inode() and friends..
70 */
Dave Chinner0a234c62013-08-28 10:17:57 +100071 if (!(sc->gfp_mask & __GFP_FS))
72 return SHRINK_STOP;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100073
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +030074 if (!trylock_super(sb))
Dave Chinner0a234c62013-08-28 10:17:57 +100075 return SHRINK_STOP;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100076
Al Virod0407902013-07-19 21:12:31 +040077 if (sb->s_op->nr_cached_objects)
Vladimir Davydov4101b622015-02-12 14:58:51 -080078 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100079
Vladimir Davydov503c3582015-02-12 14:58:47 -080080 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
81 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
Dave Chinnerf6041562013-08-28 10:18:00 +100082 total_objects = dentries + inodes + fs_objects + 1;
Tetsuo Handa475d0db2014-05-17 20:56:38 +090083 if (!total_objects)
84 total_objects = 1;
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100085
Dave Chinner0a234c62013-08-28 10:17:57 +100086 /* proportion the scan between the caches */
Dave Chinnerf6041562013-08-28 10:18:00 +100087 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +100088 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
Vladimir Davydov503c3582015-02-12 14:58:47 -080089 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
Dave Chinnerb0d40c92011-07-08 14:14:42 +100090
Dave Chinner0a234c62013-08-28 10:17:57 +100091 /*
92 * prune the dcache first as the icache is pinned by it, then
93 * prune the icache, followed by the filesystem specific caches
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -080094 *
95 * Ensure that we always scan at least one object - memcg kmem
96 * accounting uses this to fully empty the caches.
Dave Chinner0a234c62013-08-28 10:17:57 +100097 */
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -080098 sc->nr_to_scan = dentries + 1;
Vladimir Davydov503c3582015-02-12 14:58:47 -080099 freed = prune_dcache_sb(sb, sc);
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -0800100 sc->nr_to_scan = inodes + 1;
Vladimir Davydov503c3582015-02-12 14:58:47 -0800101 freed += prune_icache_sb(sb, sc);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000102
Dave Chinner0a234c62013-08-28 10:17:57 +1000103 if (fs_objects) {
Vladimir Davydov49e7e7f2015-02-12 14:59:17 -0800104 sc->nr_to_scan = fs_objects + 1;
Vladimir Davydov4101b622015-02-12 14:58:51 -0800105 freed += sb->s_op->free_cached_objects(sb, sc);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000106 }
107
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300108 up_read(&sb->s_umount);
Dave Chinner0a234c62013-08-28 10:17:57 +1000109 return freed;
110}
111
112static unsigned long super_cache_count(struct shrinker *shrink,
113 struct shrink_control *sc)
114{
115 struct super_block *sb;
116 long total_objects = 0;
117
118 sb = container_of(shrink, struct super_block, s_shrink);
119
Tim Chend23da152014-06-04 16:10:47 -0700120 /*
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300121 * Don't call trylock_super as it is a potential
Tim Chend23da152014-06-04 16:10:47 -0700122 * scalability bottleneck. The counts could get updated
123 * between super_cache_count and super_cache_scan anyway.
124 * Call to super_cache_count with shrinker_rwsem held
Vladimir Davydov503c3582015-02-12 14:58:47 -0800125 * ensures the safety of call to list_lru_shrink_count() and
Tim Chend23da152014-06-04 16:10:47 -0700126 * s_op->nr_cached_objects().
127 */
Dave Chinner0a234c62013-08-28 10:17:57 +1000128 if (sb->s_op && sb->s_op->nr_cached_objects)
Vladimir Davydov4101b622015-02-12 14:58:51 -0800129 total_objects = sb->s_op->nr_cached_objects(sb, sc);
Dave Chinner0a234c62013-08-28 10:17:57 +1000130
Vladimir Davydov503c3582015-02-12 14:58:47 -0800131 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
132 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
Dave Chinner0a234c62013-08-28 10:17:57 +1000133
Glauber Costa55f841c2013-08-28 10:17:53 +1000134 total_objects = vfs_pressure_ratio(total_objects);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000135 return total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000136}
137
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200138static void destroy_super_work(struct work_struct *work)
139{
140 struct super_block *s = container_of(work, struct super_block,
141 destroy_work);
142 int i;
143
144 for (i = 0; i < SB_FREEZE_LEVELS; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200145 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200146 kfree(s);
147}
148
149static void destroy_super_rcu(struct rcu_head *head)
150{
151 struct super_block *s = container_of(head, struct super_block, rcu);
152 INIT_WORK(&s->destroy_work, destroy_super_work);
153 schedule_work(&s->destroy_work);
154}
155
Al Viro7eb5e882013-10-01 15:09:58 -0400156/**
157 * destroy_super - frees a superblock
158 * @s: superblock to free
159 *
160 * Frees a superblock.
161 */
162static void destroy_super(struct super_block *s)
Jan Kara5accdf82012-06-12 16:20:34 +0200163{
Al Viro7eb5e882013-10-01 15:09:58 -0400164 list_lru_destroy(&s->s_dentry_lru);
165 list_lru_destroy(&s->s_inode_lru);
Al Viro7eb5e882013-10-01 15:09:58 -0400166 security_sb_free(s);
167 WARN_ON(!list_empty(&s->s_mounts));
168 kfree(s->s_subtype);
169 kfree(s->s_options);
Oleg Nesterov853b39a2015-07-22 20:21:13 +0200170 call_rcu(&s->rcu, destroy_super_rcu);
Jan Kara5accdf82012-06-12 16:20:34 +0200171}
172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173/**
174 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700175 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100176 * @flags: the mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 *
178 * Allocates and initializes a new &struct super_block. alloc_super()
179 * returns a pointer new superblock or %NULL if allocation had failed.
180 */
David Howells9249e172012-06-25 12:55:37 +0100181static struct super_block *alloc_super(struct file_system_type *type, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800183 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700184 static const struct super_operations default_op;
Al Viro7eb5e882013-10-01 15:09:58 -0400185 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Al Viro7eb5e882013-10-01 15:09:58 -0400187 if (!s)
188 return NULL;
189
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800190 INIT_LIST_HEAD(&s->s_mounts);
191
Al Viro7eb5e882013-10-01 15:09:58 -0400192 if (security_sb_alloc(s))
193 goto fail;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200194
Al Viro7eb5e882013-10-01 15:09:58 -0400195 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
Oleg Nesterov8129ed22015-08-11 17:05:04 +0200196 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
197 sb_writers_name[i],
198 &type->s_writers_key[i]))
Al Viro7eb5e882013-10-01 15:09:58 -0400199 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 }
Al Viro7eb5e882013-10-01 15:09:58 -0400201 init_waitqueue_head(&s->s_writers.wait_unfrozen);
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100202 s->s_bdi = &noop_backing_dev_info;
Al Viro7eb5e882013-10-01 15:09:58 -0400203 s->s_flags = flags;
Al Viro7eb5e882013-10-01 15:09:58 -0400204 INIT_HLIST_NODE(&s->s_instances);
205 INIT_HLIST_BL_HEAD(&s->s_anon);
Dave Chinnere97fedb2015-03-04 13:40:00 -0500206 mutex_init(&s->s_sync_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400207 INIT_LIST_HEAD(&s->s_inodes);
Dave Chinner74278da2015-03-04 12:37:22 -0500208 spin_lock_init(&s->s_inode_list_lock);
Al Viro7eb5e882013-10-01 15:09:58 -0400209
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800210 if (list_lru_init_memcg(&s->s_dentry_lru))
Al Viro7eb5e882013-10-01 15:09:58 -0400211 goto fail;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800212 if (list_lru_init_memcg(&s->s_inode_lru))
Al Viro7eb5e882013-10-01 15:09:58 -0400213 goto fail;
214
Al Viro7eb5e882013-10-01 15:09:58 -0400215 init_rwsem(&s->s_umount);
216 lockdep_set_class(&s->s_umount, &type->s_umount_key);
217 /*
218 * sget() can have s_umount recursion.
219 *
220 * When it cannot find a suitable sb, it allocates a new
221 * one (this one), and tries again to find a suitable old
222 * one.
223 *
224 * In case that succeeds, it will acquire the s_umount
225 * lock of the old one. Since these are clearly distrinct
226 * locks, and this object isn't exposed yet, there's no
227 * risk of deadlocks.
228 *
229 * Annotate this by putting this lock in a different
230 * subclass.
231 */
232 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
233 s->s_count = 1;
234 atomic_set(&s->s_active, 1);
235 mutex_init(&s->s_vfs_rename_mutex);
236 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
237 mutex_init(&s->s_dquot.dqio_mutex);
238 mutex_init(&s->s_dquot.dqonoff_mutex);
Al Viro7eb5e882013-10-01 15:09:58 -0400239 s->s_maxbytes = MAX_NON_LFS;
240 s->s_op = &default_op;
241 s->s_time_gran = 1000000000;
Vladimir Davydov3cb29d12015-04-14 15:46:48 -0700242 s->cleancache_poolid = CLEANCACHE_NO_POOL;
Al Viro7eb5e882013-10-01 15:09:58 -0400243
244 s->s_shrink.seeks = DEFAULT_SEEKS;
245 s->s_shrink.scan_objects = super_cache_scan;
246 s->s_shrink.count_objects = super_cache_count;
247 s->s_shrink.batch = 1024;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800248 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000250
Al Viro7eb5e882013-10-01 15:09:58 -0400251fail:
252 destroy_super(s);
253 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254}
255
256/* Superblock refcounting */
257
258/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400259 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 */
Al Virof47ec3f2011-11-21 21:15:42 -0500261static void __put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 if (!--sb->s_count) {
Al Viro551de6f2010-03-22 19:36:35 -0400264 list_del_init(&sb->s_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 destroy_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267}
268
269/**
270 * put_super - drop a temporary reference to superblock
271 * @sb: superblock in question
272 *
273 * Drops a temporary reference, frees superblock if there's no
274 * references left.
275 */
Al Virof47ec3f2011-11-21 21:15:42 -0500276static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277{
278 spin_lock(&sb_lock);
279 __put_super(sb);
280 spin_unlock(&sb_lock);
281}
282
283
284/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400285 * deactivate_locked_super - drop an active reference to superblock
286 * @s: superblock to deactivate
287 *
Al Viro1712ac82010-03-22 15:22:31 -0400288 * Drops an active reference to superblock, converting it into a temprory
289 * one if there is no other active references left. In that case we
290 * tell fs driver to shut it down and drop the temporary reference we
291 * had just acquired.
292 *
293 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400294 */
295void deactivate_locked_super(struct super_block *s)
296{
297 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400298 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400299 cleancache_invalidate_fs(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000300 unregister_shrinker(&s->s_shrink);
Dave Chinner28f2cd4f2014-06-04 16:10:46 -0700301 fs->kill_sb(s);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000302
Vladimir Davydovc0a5b5602015-02-12 14:59:07 -0800303 /*
304 * Since list_lru_destroy() may sleep, we cannot call it from
305 * put_super(), where we hold the sb_lock. Therefore we destroy
306 * the lru lists right now.
307 */
308 list_lru_destroy(&s->s_dentry_lru);
309 list_lru_destroy(&s->s_inode_lru);
310
Al Viro74dbbdd2009-05-06 01:07:50 -0400311 put_filesystem(fs);
312 put_super(s);
313 } else {
314 up_write(&s->s_umount);
315 }
316}
317
318EXPORT_SYMBOL(deactivate_locked_super);
319
320/**
Al Viro1712ac82010-03-22 15:22:31 -0400321 * deactivate_super - drop an active reference to superblock
322 * @s: superblock to deactivate
323 *
324 * Variant of deactivate_locked_super(), except that superblock is *not*
325 * locked by caller. If we are going to drop the final active reference,
326 * lock will be acquired prior to that.
327 */
328void deactivate_super(struct super_block *s)
329{
330 if (!atomic_add_unless(&s->s_active, -1, 1)) {
331 down_write(&s->s_umount);
332 deactivate_locked_super(s);
333 }
334}
335
336EXPORT_SYMBOL(deactivate_super);
337
338/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 * grab_super - acquire an active reference
340 * @s: reference we are trying to make active
341 *
342 * Tries to acquire an active reference. grab_super() is used when we
343 * had just found a superblock in super_blocks or fs_type->fs_supers
344 * and want to turn it into a full-blown active reference. grab_super()
345 * is called with sb_lock held and drops it. Returns 1 in case of
346 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400347 * dying when grab_super() had been called). Note that this is only
348 * called for superblocks not in rundown mode (== ones still on ->fs_supers
349 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700351static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352{
353 s->s_count++;
354 spin_unlock(&sb_lock);
355 down_write(&s->s_umount);
Al Viroacfec9a2013-07-20 03:13:55 +0400356 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
357 put_super(s);
358 return 1;
359 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 up_write(&s->s_umount);
361 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 return 0;
363}
364
David Howellscf9a2ae2006-08-29 19:05:54 +0100365/*
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300366 * trylock_super - try to grab ->s_umount shared
Wanpeng Li331cbde2012-06-09 11:10:55 +0800367 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000368 *
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300369 * Try to prevent fs shutdown. This is used in places where we
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000370 * cannot take an active reference but we need to ensure that the
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300371 * filesystem is not shut down while we are working on it. It returns
372 * false if we cannot acquire s_umount or if we lose the race and
373 * filesystem already got into shutdown, and returns true with the s_umount
374 * lock held in read mode in case of success. On successful return,
375 * the caller must drop the s_umount lock when done.
376 *
377 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
378 * The reason why it's safe is that we are OK with doing trylock instead
379 * of down_read(). There's a couple of places that are OK with that, but
380 * it's very much not a general-purpose interface.
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000381 */
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300382bool trylock_super(struct super_block *sb)
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000383{
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000384 if (down_read_trylock(&sb->s_umount)) {
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300385 if (!hlist_unhashed(&sb->s_instances) &&
386 sb->s_root && (sb->s_flags & MS_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000387 return true;
388 up_read(&sb->s_umount);
389 }
390
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000391 return false;
392}
393
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394/**
395 * generic_shutdown_super - common helper for ->kill_sb()
396 * @sb: superblock to kill
397 *
398 * generic_shutdown_super() does all fs-independent work on superblock
399 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
400 * that need destruction out of superblock, call generic_shutdown_super()
401 * and release aforementioned objects. Note: dentries and inodes _are_
402 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700403 *
404 * Upon calling this function, the filesystem may no longer alter or
405 * rearrange the set of dentries belonging to this super_block, nor may it
406 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 */
408void generic_shutdown_super(struct super_block *sb)
409{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800410 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
David Howellsc636ebd2006-10-11 01:22:19 -0700412 if (sb->s_root) {
413 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200414 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 sb->s_flags &= ~MS_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800416
Dave Chinner74278da2015-03-04 12:37:22 -0500417 fsnotify_unmount_inodes(sb);
Tejun Heoa1a0e232016-02-29 18:28:53 -0500418 cgroup_writeback_umount();
Al Viro63997e92010-10-25 20:49:35 -0400419
420 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200422 if (sb->s_dio_done_wq) {
423 destroy_workqueue(sb->s_dio_done_wq);
424 sb->s_dio_done_wq = NULL;
425 }
426
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 if (sop->put_super)
428 sop->put_super(sb);
429
Al Viro63997e92010-10-25 20:49:35 -0400430 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800431 printk("VFS: Busy inodes after unmount of %s. "
432 "Self-destruct in 5 seconds. Have a nice day...\n",
433 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 }
436 spin_lock(&sb_lock);
437 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500438 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 spin_unlock(&sb_lock);
440 up_write(&sb->s_umount);
441}
442
443EXPORT_SYMBOL(generic_shutdown_super);
444
445/**
446 * sget - find or create a superblock
447 * @type: filesystem type superblock should belong to
448 * @test: comparison callback
449 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100450 * @flags: mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 * @data: argument to each of them
452 */
453struct super_block *sget(struct file_system_type *type,
454 int (*test)(struct super_block *,void *),
455 int (*set)(struct super_block *,void *),
David Howells9249e172012-06-25 12:55:37 +0100456 int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 void *data)
458{
459 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700460 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 int err;
462
463retry:
464 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700465 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800466 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700467 if (!test(old, data))
468 continue;
469 if (!grab_super(old))
470 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700471 if (s) {
472 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700473 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400474 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700475 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700476 return old;
477 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 }
479 if (!s) {
480 spin_unlock(&sb_lock);
David Howells9249e172012-06-25 12:55:37 +0100481 s = alloc_super(type, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 if (!s)
483 return ERR_PTR(-ENOMEM);
484 goto retry;
485 }
486
487 err = set(s, data);
488 if (err) {
489 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700490 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 destroy_super(s);
492 return ERR_PTR(err);
493 }
494 s->s_type = type;
495 strlcpy(s->s_id, type->name, sizeof(s->s_id));
496 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500497 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 spin_unlock(&sb_lock);
499 get_filesystem(type);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000500 register_shrinker(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 return s;
502}
503
504EXPORT_SYMBOL(sget);
505
506void drop_super(struct super_block *sb)
507{
508 up_read(&sb->s_umount);
509 put_super(sb);
510}
511
512EXPORT_SYMBOL(drop_super);
513
Christoph Hellwige5004752009-05-05 16:08:56 +0200514/**
Al Viro01a05b32010-03-23 06:06:58 -0400515 * iterate_supers - call function for all active superblocks
516 * @f: function to call
517 * @arg: argument to pass to it
518 *
519 * Scans the superblock list and calls given function, passing it
520 * locked superblock and given argument.
521 */
522void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
523{
Al Virodca33252010-07-25 02:31:46 +0400524 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400525
526 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400527 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500528 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400529 continue;
530 sb->s_count++;
531 spin_unlock(&sb_lock);
532
533 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500534 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400535 f(sb, arg);
536 up_read(&sb->s_umount);
537
538 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400539 if (p)
540 __put_super(p);
541 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400542 }
Al Virodca33252010-07-25 02:31:46 +0400543 if (p)
544 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400545 spin_unlock(&sb_lock);
546}
547
548/**
Al Viro43e15cd2011-06-03 20:16:57 -0400549 * iterate_supers_type - call function for superblocks of given type
550 * @type: fs type
551 * @f: function to call
552 * @arg: argument to pass to it
553 *
554 * Scans the superblock list and calls given function, passing it
555 * locked superblock and given argument.
556 */
557void iterate_supers_type(struct file_system_type *type,
558 void (*f)(struct super_block *, void *), void *arg)
559{
560 struct super_block *sb, *p = NULL;
561
562 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800563 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400564 sb->s_count++;
565 spin_unlock(&sb_lock);
566
567 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500568 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400569 f(sb, arg);
570 up_read(&sb->s_umount);
571
572 spin_lock(&sb_lock);
573 if (p)
574 __put_super(p);
575 p = sb;
576 }
577 if (p)
578 __put_super(p);
579 spin_unlock(&sb_lock);
580}
581
582EXPORT_SYMBOL(iterate_supers_type);
583
584/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 * get_super - get the superblock of a device
586 * @bdev: device to get the superblock for
587 *
588 * Scans the superblock list and finds the superblock of the file system
589 * mounted on the device given. %NULL is returned if no match is found.
590 */
591
Al Virodf40c012010-03-22 20:23:25 -0400592struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700594 struct super_block *sb;
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 if (!bdev)
597 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700598
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700600rescan:
601 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500602 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400603 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700604 if (sb->s_bdev == bdev) {
605 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700607 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400608 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500609 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700610 return sb;
611 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400612 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700613 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400614 __put_super(sb);
615 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 }
617 }
618 spin_unlock(&sb_lock);
619 return NULL;
620}
621
622EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200623
624/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100625 * get_super_thawed - get thawed superblock of a device
626 * @bdev: device to get the superblock for
627 *
628 * Scans the superblock list and finds the superblock of the file system
629 * mounted on the device. The superblock is returned once it is thawed
630 * (or immediately if it was not frozen). %NULL is returned if no match
631 * is found.
632 */
633struct super_block *get_super_thawed(struct block_device *bdev)
634{
635 while (1) {
636 struct super_block *s = get_super(bdev);
Jan Kara5accdf82012-06-12 16:20:34 +0200637 if (!s || s->s_writers.frozen == SB_UNFROZEN)
Jan Kara6b6dc832012-02-10 11:03:00 +0100638 return s;
639 up_read(&s->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +0200640 wait_event(s->s_writers.wait_unfrozen,
641 s->s_writers.frozen == SB_UNFROZEN);
Jan Kara6b6dc832012-02-10 11:03:00 +0100642 put_super(s);
643 }
644}
645EXPORT_SYMBOL(get_super_thawed);
646
647/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200648 * get_active_super - get an active reference to the superblock of a device
649 * @bdev: device to get the superblock for
650 *
651 * Scans the superblock list and finds the superblock of the file system
652 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400653 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200654 */
655struct super_block *get_active_super(struct block_device *bdev)
656{
657 struct super_block *sb;
658
659 if (!bdev)
660 return NULL;
661
Al Viro14945832010-03-22 20:15:33 -0400662restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200663 spin_lock(&sb_lock);
664 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500665 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400666 continue;
Al Viro14945832010-03-22 20:15:33 -0400667 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400668 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400669 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400670 up_write(&sb->s_umount);
671 return sb;
Al Viro14945832010-03-22 20:15:33 -0400672 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200673 }
674 spin_unlock(&sb_lock);
675 return NULL;
676}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
Al Virodf40c012010-03-22 20:23:25 -0400678struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700680 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700683rescan:
684 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500685 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400686 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700687 if (sb->s_dev == dev) {
688 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700690 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400691 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500692 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700693 return sb;
694 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400695 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700696 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400697 __put_super(sb);
698 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 }
700 }
701 spin_unlock(&sb_lock);
702 return NULL;
703}
704
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 * do_remount_sb - asks filesystem to change mount options.
707 * @sb: superblock in question
708 * @flags: numeric part of options
709 * @data: the rest of options
710 * @force: whether or not to force the change
711 *
712 * Alters the mount options of a mounted file system.
713 */
714int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
715{
716 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400717 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200718
Jan Kara5accdf82012-06-12 16:20:34 +0200719 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200720 return -EBUSY;
721
David Howells93614012006-09-30 20:45:40 +0200722#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
724 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200725#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200726
Nick Piggind208bbd2009-12-21 16:28:53 -0800727 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800728
Al Viro0aec09d2014-08-07 07:32:06 -0400729 if (remount_ro) {
Al Virofdab6842015-01-11 10:57:27 -0500730 if (!hlist_empty(&sb->s_pins)) {
Al Viro0aec09d2014-08-07 07:32:06 -0400731 up_write(&sb->s_umount);
Al Virofdab6842015-01-11 10:57:27 -0500732 group_pin_kill(&sb->s_pins);
Al Viro0aec09d2014-08-07 07:32:06 -0400733 down_write(&sb->s_umount);
734 if (!sb->s_root)
735 return 0;
736 if (sb->s_writers.frozen != SB_UNFROZEN)
737 return -EBUSY;
738 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
739 }
740 }
741 shrink_dcache_sb(sb);
742
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 /* If we are remounting RDONLY and current sb is read/write,
744 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800745 if (remount_ro) {
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100746 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400747 sb->s_readonly_remount = 1;
748 smp_wmb();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100749 } else {
750 retval = sb_prepare_remount_readonly(sb);
751 if (retval)
752 return retval;
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100753 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 }
755
756 if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 retval = sb->s_op->remount_fs(sb, &flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200758 if (retval) {
759 if (!force)
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100760 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200761 /* If forced remount, go ahead despite any errors */
762 WARN(1, "forced remount of a %s fs returned %i\n",
763 sb->s_type->name, retval);
764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 }
766 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100767 /* Needs to be ordered wrt mnt_is_readonly() */
768 smp_wmb();
769 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400770
Nick Piggind208bbd2009-12-21 16:28:53 -0800771 /*
772 * Some filesystems modify their metadata via some other path than the
773 * bdev buffer cache (eg. use a private mapping, or directories in
774 * pagecache, etc). Also file data modifications go via their own
775 * mappings. So If we try to mount readonly then copy the filesystem
776 * from bdev, we could get stale data, so invalidate it to give a best
777 * effort at coherency.
778 */
779 if (remount_ro && sb->s_bdev)
780 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 return 0;
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100782
783cancel_readonly:
784 sb->s_readonly_remount = 0;
785 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786}
787
Jens Axboea2a95372009-03-17 09:38:40 +0100788static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
Al Virodca33252010-07-25 02:31:46 +0400790 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
792 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400793 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500794 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400795 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 sb->s_count++;
797 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400798 down_write(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500799 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
800 !(sb->s_flags & MS_RDONLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 * What lock protects sb->s_flags??
803 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 }
Al Viro443b94b2009-05-05 23:48:50 -0400806 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400808 if (p)
809 __put_super(p);
810 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 }
Al Virodca33252010-07-25 02:31:46 +0400812 if (p)
813 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100815 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 printk("Emergency Remount complete\n");
817}
818
819void emergency_remount(void)
820{
Jens Axboea2a95372009-03-17 09:38:40 +0100821 struct work_struct *work;
822
823 work = kmalloc(sizeof(*work), GFP_ATOMIC);
824 if (work) {
825 INIT_WORK(work, do_emergency_remount);
826 schedule_work(work);
827 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828}
829
830/*
831 * Unnamed block devices are dummy devices used by virtual
832 * filesystems which don't use real block-devices. -- jrs
833 */
834
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400835static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Thomas Bächlera2a4dc42014-04-03 21:55:37 +0200837/* Many userspace utilities consider an FSID of 0 invalid.
838 * Always return at least 1 from get_anon_bdev.
839 */
840static int unnamed_dev_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
Al Viro0ee5dc62011-07-07 15:44:25 -0400842int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843{
844 int dev;
845 int error;
846
847 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400848 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 return -ENOMEM;
850 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400851 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400852 if (!error)
853 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 spin_unlock(&unnamed_dev_lock);
855 if (error == -EAGAIN)
856 /* We raced and lost with another CPU. */
857 goto retry;
858 else if (error)
859 return -EAGAIN;
860
Wang YanQing1af95de2015-06-23 18:54:45 +0800861 if (dev >= (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400863 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400864 if (unnamed_dev_start > dev)
865 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 spin_unlock(&unnamed_dev_lock);
867 return -EMFILE;
868 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400869 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 return 0;
871}
Al Viro0ee5dc62011-07-07 15:44:25 -0400872EXPORT_SYMBOL(get_anon_bdev);
873
874void free_anon_bdev(dev_t dev)
875{
876 int slot = MINOR(dev);
877 spin_lock(&unnamed_dev_lock);
878 ida_remove(&unnamed_dev_ida, slot);
879 if (slot < unnamed_dev_start)
880 unnamed_dev_start = slot;
881 spin_unlock(&unnamed_dev_lock);
882}
883EXPORT_SYMBOL(free_anon_bdev);
884
885int set_anon_super(struct super_block *s, void *data)
886{
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100887 return get_anon_bdev(&s->s_dev);
Al Viro0ee5dc62011-07-07 15:44:25 -0400888}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
890EXPORT_SYMBOL(set_anon_super);
891
892void kill_anon_super(struct super_block *sb)
893{
Al Viro0ee5dc62011-07-07 15:44:25 -0400894 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400896 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
899EXPORT_SYMBOL(kill_anon_super);
900
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901void kill_litter_super(struct super_block *sb)
902{
903 if (sb->s_root)
904 d_genocide(sb->s_root);
905 kill_anon_super(sb);
906}
907
908EXPORT_SYMBOL(kill_litter_super);
909
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700910static int ns_test_super(struct super_block *sb, void *data)
911{
912 return sb->s_fs_info == data;
913}
914
915static int ns_set_super(struct super_block *sb, void *data)
916{
917 sb->s_fs_info = data;
918 return set_anon_super(sb, NULL);
919}
920
Al Viroceefda62010-07-26 13:16:50 +0400921struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
922 void *data, int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700923{
924 struct super_block *sb;
925
David Howells9249e172012-06-25 12:55:37 +0100926 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700927 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +0400928 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700929
930 if (!sb->s_root) {
931 int err;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700932 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
933 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400934 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +0400935 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700936 }
937
938 sb->s_flags |= MS_ACTIVE;
939 }
940
Al Viroceefda62010-07-26 13:16:50 +0400941 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700942}
943
Al Viroceefda62010-07-26 13:16:50 +0400944EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700945
David Howells93614012006-09-30 20:45:40 +0200946#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947static int set_bdev_super(struct super_block *s, void *data)
948{
949 s->s_bdev = data;
950 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +0200951
952 /*
953 * We set the bdi here to the queue backing, file systems can
954 * overwrite this in ->fill_super()
955 */
956 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 return 0;
958}
959
960static int test_bdev_super(struct super_block *s, void *data)
961{
962 return (void *)s->s_bdev == data;
963}
964
Al Viro152a0832010-07-25 00:46:55 +0400965struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +0400967 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968{
969 struct block_device *bdev;
970 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +0100971 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 int error = 0;
973
Al Viro30c40d22008-02-22 19:50:45 -0500974 if (!(flags & MS_RDONLY))
975 mode |= FMODE_WRITE;
976
Tejun Heod4d77622010-11-13 11:55:18 +0100977 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +0400979 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
981 /*
982 * once the super is inserted into the list by sget, s_umount
983 * will protect the lockfs code from trying to start a snapshot
984 * while we are mounting
985 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200986 mutex_lock(&bdev->bd_fsfreeze_mutex);
987 if (bdev->bd_fsfreeze_count > 0) {
988 mutex_unlock(&bdev->bd_fsfreeze_mutex);
989 error = -EBUSY;
990 goto error_bdev;
991 }
David Howells9249e172012-06-25 12:55:37 +0100992 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
993 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200994 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -0700996 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 if (s->s_root) {
999 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001000 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001001 error = -EBUSY;
1002 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 }
David Howells454e2392006-06-23 02:02:57 -07001004
Tejun Heo4f331f02010-07-20 15:18:07 -07001005 /*
1006 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +01001007 * __invalidate_device(). blkdev_put() acquires
1008 * bd_mutex and can't be called under s_umount. Drop
1009 * s_umount temporarily. This is safe as we're
1010 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -07001011 */
1012 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +01001013 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -07001014 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 } else {
Al Viro30c40d22008-02-22 19:50:45 -05001016 s->s_mode = mode;
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001017 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
Pekka Enberge78c9a02006-01-08 01:03:39 -08001018 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001019 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001021 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001022 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001023 }
David Howells454e2392006-06-23 02:02:57 -07001024
1025 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001026 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 }
1028
Al Viro152a0832010-07-25 00:46:55 +04001029 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
David Howells454e2392006-06-23 02:02:57 -07001031error_s:
1032 error = PTR_ERR(s);
1033error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001034 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001035error:
Al Viro152a0832010-07-25 00:46:55 +04001036 return ERR_PTR(error);
1037}
1038EXPORT_SYMBOL(mount_bdev);
1039
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040void kill_block_super(struct super_block *sb)
1041{
1042 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001043 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001045 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 generic_shutdown_super(sb);
1047 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001048 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001049 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050}
1051
1052EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001053#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
Al Viro3c26ff62010-07-25 11:46:36 +04001055struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001057 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058{
1059 int error;
David Howells9249e172012-06-25 12:55:37 +01001060 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
1062 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001063 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001065 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001067 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001068 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 }
1070 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001071 return dget(s->s_root);
1072}
1073EXPORT_SYMBOL(mount_nodev);
1074
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075static int compare_single(struct super_block *s, void *p)
1076{
1077 return 1;
1078}
1079
Al Virofc14f2f2010-07-25 01:48:30 +04001080struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001082 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083{
1084 struct super_block *s;
1085 int error;
1086
David Howells9249e172012-06-25 12:55:37 +01001087 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001089 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 if (!s->s_root) {
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001091 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001093 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001094 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 }
1096 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001097 } else {
1098 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 }
Al Virofc14f2f2010-07-25 01:48:30 +04001100 return dget(s->s_root);
1101}
1102EXPORT_SYMBOL(mount_single);
1103
Al Viro9d412a42011-03-17 22:08:28 -04001104struct dentry *
1105mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106{
Al Viroc96e41e2010-07-25 00:17:56 +04001107 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001108 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001110 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001111
Eric Parise0007522008-03-05 10:31:54 -05001112 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001114 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001115 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
Eric Parise0007522008-03-05 10:31:54 -05001117 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001118 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 }
1121
Al Viro1a102ff2011-03-16 09:07:58 -04001122 root = type->mount(type, flags, name, data);
1123 if (IS_ERR(root)) {
1124 error = PTR_ERR(root);
1125 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001126 }
Al Viro9d412a42011-03-17 22:08:28 -04001127 sb = root->d_sb;
1128 BUG_ON(!sb);
1129 WARN_ON(!sb->s_bdi);
1130 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001131
Al Viro9d412a42011-03-17 22:08:28 -04001132 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001133 if (error)
1134 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001135
Jeff Layton42cb56a2009-09-18 13:05:53 -07001136 /*
1137 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1138 * but s_maxbytes was an unsigned long long for many releases. Throw
1139 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001140 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001141 */
Al Viro9d412a42011-03-17 22:08:28 -04001142 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1143 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001144
Al Viro9d412a42011-03-17 22:08:28 -04001145 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001146 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001147 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001149 dput(root);
1150 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151out_free_secdata:
1152 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153out:
David Howells454e2392006-06-23 02:02:57 -07001154 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155}
1156
Jan Kara5accdf82012-06-12 16:20:34 +02001157/*
1158 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1159 * instead.
1160 */
1161void __sb_end_write(struct super_block *sb, int level)
1162{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001163 percpu_up_read(sb->s_writers.rw_sem + level-1);
Jan Kara5accdf82012-06-12 16:20:34 +02001164}
1165EXPORT_SYMBOL(__sb_end_write);
1166
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001167/*
1168 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1169 * instead.
1170 */
1171int __sb_start_write(struct super_block *sb, int level, bool wait)
1172{
1173 bool force_trylock = false;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001174 int ret = 1;
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001175
1176#ifdef CONFIG_LOCKDEP
1177 /*
1178 * We want lockdep to tell us about possible deadlocks with freezing
1179 * but it's it bit tricky to properly instrument it. Getting a freeze
1180 * protection works as getting a read lock but there are subtle
1181 * problems. XFS for example gets freeze protection on internal level
1182 * twice in some cases, which is OK only because we already hold a
1183 * freeze protection also on higher level. Due to these cases we have
1184 * to use wait == F (trylock mode) which must not fail.
1185 */
1186 if (wait) {
1187 int i;
1188
1189 for (i = 0; i < level - 1; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001190 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001191 force_trylock = true;
1192 break;
1193 }
1194 }
1195#endif
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001196 if (wait && !force_trylock)
1197 percpu_down_read(sb->s_writers.rw_sem + level-1);
1198 else
1199 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1200
Vincent Stehlé22224a12015-09-21 17:18:34 +02001201 WARN_ON(force_trylock && !ret);
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001202 return ret;
1203}
Jan Kara5accdf82012-06-12 16:20:34 +02001204EXPORT_SYMBOL(__sb_start_write);
1205
1206/**
1207 * sb_wait_write - wait until all writers to given file system finish
1208 * @sb: the super for which we wait
1209 * @level: type of writers we wait for (normal vs page fault)
1210 *
1211 * This function waits until there are no writers of given type to given file
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001212 * system.
Jan Kara5accdf82012-06-12 16:20:34 +02001213 */
1214static void sb_wait_write(struct super_block *sb, int level)
1215{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001216 percpu_down_write(sb->s_writers.rw_sem + level-1);
Oleg Nesterov0e28e012015-08-11 16:28:29 +02001217 /*
1218 * We are going to return to userspace and forget about this lock, the
1219 * ownership goes to the caller of thaw_super() which does unlock.
1220 *
1221 * FIXME: we should do this before return from freeze_super() after we
1222 * called sync_filesystem(sb) and s_op->freeze_fs(sb), and thaw_super()
1223 * should re-acquire these locks before s_op->unfreeze_fs(sb). However
1224 * this leads to lockdep false-positives, so currently we do the early
1225 * release right after acquire.
1226 */
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001227 percpu_rwsem_release(sb->s_writers.rw_sem + level-1, 0, _THIS_IP_);
1228}
Jan Kara5accdf82012-06-12 16:20:34 +02001229
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001230static void sb_freeze_unlock(struct super_block *sb)
1231{
1232 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001233
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001234 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1235 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
Jan Kara5accdf82012-06-12 16:20:34 +02001236
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001237 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1238 percpu_up_write(sb->s_writers.rw_sem + level);
Jan Kara5accdf82012-06-12 16:20:34 +02001239}
1240
Josef Bacik18e9e512010-03-23 10:34:56 -04001241/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001242 * freeze_super - lock the filesystem and force it into a consistent state
1243 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001244 *
1245 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1246 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1247 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001248 *
1249 * During this function, sb->s_writers.frozen goes through these values:
1250 *
1251 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1252 *
1253 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1254 * writes should be blocked, though page faults are still allowed. We wait for
1255 * all writes to complete and then proceed to the next stage.
1256 *
1257 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1258 * but internal fs threads can still modify the filesystem (although they
1259 * should not dirty new pages or inodes), writeback can run etc. After waiting
1260 * for all running page faults we sync the filesystem which will clean all
1261 * dirty pages and inodes (no new dirty pages or inodes can be created when
1262 * sync is running).
1263 *
1264 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1265 * modification are blocked (e.g. XFS preallocation truncation on inode
1266 * reclaim). This is usually implemented by blocking new transactions for
1267 * filesystems that have them and need this additional guard. After all
1268 * internal writers are finished we call ->freeze_fs() to finish filesystem
1269 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1270 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1271 *
1272 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001273 */
1274int freeze_super(struct super_block *sb)
1275{
1276 int ret;
1277
1278 atomic_inc(&sb->s_active);
1279 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001280 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001281 deactivate_locked_super(sb);
1282 return -EBUSY;
1283 }
1284
Al Virodabe0dc2012-01-03 21:01:29 -05001285 if (!(sb->s_flags & MS_BORN)) {
1286 up_write(&sb->s_umount);
1287 return 0; /* sic - it's "nothing to do" */
1288 }
1289
Josef Bacik18e9e512010-03-23 10:34:56 -04001290 if (sb->s_flags & MS_RDONLY) {
Jan Kara5accdf82012-06-12 16:20:34 +02001291 /* Nothing to do really... */
1292 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001293 up_write(&sb->s_umount);
1294 return 0;
1295 }
1296
Jan Kara5accdf82012-06-12 16:20:34 +02001297 sb->s_writers.frozen = SB_FREEZE_WRITE;
Jan Kara5accdf82012-06-12 16:20:34 +02001298 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1299 up_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001300 sb_wait_write(sb, SB_FREEZE_WRITE);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001301 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001302
1303 /* Now we go and block page faults... */
Jan Kara5accdf82012-06-12 16:20:34 +02001304 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
Jan Kara5accdf82012-06-12 16:20:34 +02001305 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1306
1307 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001308 sync_filesystem(sb);
1309
Jan Kara5accdf82012-06-12 16:20:34 +02001310 /* Now wait for internal filesystem counter */
1311 sb->s_writers.frozen = SB_FREEZE_FS;
Jan Kara5accdf82012-06-12 16:20:34 +02001312 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001313
Josef Bacik18e9e512010-03-23 10:34:56 -04001314 if (sb->s_op->freeze_fs) {
1315 ret = sb->s_op->freeze_fs(sb);
1316 if (ret) {
1317 printk(KERN_ERR
1318 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001319 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001320 sb_freeze_unlock(sb);
Jan Kara5accdf82012-06-12 16:20:34 +02001321 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001322 deactivate_locked_super(sb);
1323 return ret;
1324 }
1325 }
Jan Kara5accdf82012-06-12 16:20:34 +02001326 /*
1327 * This is just for debugging purposes so that fs can warn if it
1328 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1329 */
1330 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001331 up_write(&sb->s_umount);
1332 return 0;
1333}
1334EXPORT_SYMBOL(freeze_super);
1335
1336/**
1337 * thaw_super -- unlock filesystem
1338 * @sb: the super to thaw
1339 *
1340 * Unlocks the filesystem and marks it writeable again after freeze_super().
1341 */
1342int thaw_super(struct super_block *sb)
1343{
1344 int error;
1345
1346 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001347 if (sb->s_writers.frozen == SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001348 up_write(&sb->s_umount);
1349 return -EINVAL;
1350 }
1351
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001352 if (sb->s_flags & MS_RDONLY) {
1353 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001354 goto out;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001355 }
Josef Bacik18e9e512010-03-23 10:34:56 -04001356
1357 if (sb->s_op->unfreeze_fs) {
1358 error = sb->s_op->unfreeze_fs(sb);
1359 if (error) {
1360 printk(KERN_ERR
1361 "VFS:Filesystem thaw failed\n");
Josef Bacik18e9e512010-03-23 10:34:56 -04001362 up_write(&sb->s_umount);
1363 return error;
1364 }
1365 }
1366
Jan Kara5accdf82012-06-12 16:20:34 +02001367 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001368 sb_freeze_unlock(sb);
1369out:
Jan Kara5accdf82012-06-12 16:20:34 +02001370 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001371 deactivate_locked_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001372 return 0;
1373}
1374EXPORT_SYMBOL(thaw_super);