blob: 954aeb80e202be0a40fc42eeca465c033ab1fc9a [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 Davydovc0a5b562015-02-12 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);
Al Viro63997e92010-10-25 20:49:35 -0400418
419 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200421 if (sb->s_dio_done_wq) {
422 destroy_workqueue(sb->s_dio_done_wq);
423 sb->s_dio_done_wq = NULL;
424 }
425
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 if (sop->put_super)
427 sop->put_super(sb);
428
Al Viro63997e92010-10-25 20:49:35 -0400429 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800430 printk("VFS: Busy inodes after unmount of %s. "
431 "Self-destruct in 5 seconds. Have a nice day...\n",
432 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 }
435 spin_lock(&sb_lock);
436 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500437 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 spin_unlock(&sb_lock);
439 up_write(&sb->s_umount);
440}
441
442EXPORT_SYMBOL(generic_shutdown_super);
443
444/**
445 * sget - find or create a superblock
446 * @type: filesystem type superblock should belong to
447 * @test: comparison callback
448 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100449 * @flags: mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 * @data: argument to each of them
451 */
452struct super_block *sget(struct file_system_type *type,
453 int (*test)(struct super_block *,void *),
454 int (*set)(struct super_block *,void *),
David Howells9249e172012-06-25 12:55:37 +0100455 int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 void *data)
457{
458 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700459 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 int err;
461
462retry:
463 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700464 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800465 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700466 if (!test(old, data))
467 continue;
468 if (!grab_super(old))
469 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700470 if (s) {
471 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700472 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400473 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700474 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700475 return old;
476 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 }
478 if (!s) {
479 spin_unlock(&sb_lock);
David Howells9249e172012-06-25 12:55:37 +0100480 s = alloc_super(type, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 if (!s)
482 return ERR_PTR(-ENOMEM);
483 goto retry;
484 }
485
486 err = set(s, data);
487 if (err) {
488 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700489 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 destroy_super(s);
491 return ERR_PTR(err);
492 }
493 s->s_type = type;
494 strlcpy(s->s_id, type->name, sizeof(s->s_id));
495 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500496 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 spin_unlock(&sb_lock);
498 get_filesystem(type);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000499 register_shrinker(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 return s;
501}
502
503EXPORT_SYMBOL(sget);
504
505void drop_super(struct super_block *sb)
506{
507 up_read(&sb->s_umount);
508 put_super(sb);
509}
510
511EXPORT_SYMBOL(drop_super);
512
Christoph Hellwige5004752009-05-05 16:08:56 +0200513/**
Al Viro01a05b32010-03-23 06:06:58 -0400514 * iterate_supers - call function for all active superblocks
515 * @f: function to call
516 * @arg: argument to pass to it
517 *
518 * Scans the superblock list and calls given function, passing it
519 * locked superblock and given argument.
520 */
521void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
522{
Al Virodca33252010-07-25 02:31:46 +0400523 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400524
525 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400526 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500527 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400528 continue;
529 sb->s_count++;
530 spin_unlock(&sb_lock);
531
532 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500533 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400534 f(sb, arg);
535 up_read(&sb->s_umount);
536
537 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400538 if (p)
539 __put_super(p);
540 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400541 }
Al Virodca33252010-07-25 02:31:46 +0400542 if (p)
543 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400544 spin_unlock(&sb_lock);
545}
546
547/**
Al Viro43e15cd2011-06-03 20:16:57 -0400548 * iterate_supers_type - call function for superblocks of given type
549 * @type: fs type
550 * @f: function to call
551 * @arg: argument to pass to it
552 *
553 * Scans the superblock list and calls given function, passing it
554 * locked superblock and given argument.
555 */
556void iterate_supers_type(struct file_system_type *type,
557 void (*f)(struct super_block *, void *), void *arg)
558{
559 struct super_block *sb, *p = NULL;
560
561 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800562 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400563 sb->s_count++;
564 spin_unlock(&sb_lock);
565
566 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500567 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400568 f(sb, arg);
569 up_read(&sb->s_umount);
570
571 spin_lock(&sb_lock);
572 if (p)
573 __put_super(p);
574 p = sb;
575 }
576 if (p)
577 __put_super(p);
578 spin_unlock(&sb_lock);
579}
580
581EXPORT_SYMBOL(iterate_supers_type);
582
583/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 * get_super - get the superblock of a device
585 * @bdev: device to get the superblock for
586 *
587 * Scans the superblock list and finds the superblock of the file system
588 * mounted on the device given. %NULL is returned if no match is found.
589 */
590
Al Virodf40c012010-03-22 20:23:25 -0400591struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700593 struct super_block *sb;
594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 if (!bdev)
596 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700599rescan:
600 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500601 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400602 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700603 if (sb->s_bdev == bdev) {
604 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700606 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400607 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500608 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700609 return sb;
610 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400611 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700612 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400613 __put_super(sb);
614 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 }
616 }
617 spin_unlock(&sb_lock);
618 return NULL;
619}
620
621EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200622
623/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100624 * get_super_thawed - get thawed superblock of a device
625 * @bdev: device to get the superblock for
626 *
627 * Scans the superblock list and finds the superblock of the file system
628 * mounted on the device. The superblock is returned once it is thawed
629 * (or immediately if it was not frozen). %NULL is returned if no match
630 * is found.
631 */
632struct super_block *get_super_thawed(struct block_device *bdev)
633{
634 while (1) {
635 struct super_block *s = get_super(bdev);
Jan Kara5accdf82012-06-12 16:20:34 +0200636 if (!s || s->s_writers.frozen == SB_UNFROZEN)
Jan Kara6b6dc832012-02-10 11:03:00 +0100637 return s;
638 up_read(&s->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +0200639 wait_event(s->s_writers.wait_unfrozen,
640 s->s_writers.frozen == SB_UNFROZEN);
Jan Kara6b6dc832012-02-10 11:03:00 +0100641 put_super(s);
642 }
643}
644EXPORT_SYMBOL(get_super_thawed);
645
646/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200647 * get_active_super - get an active reference to the superblock of a device
648 * @bdev: device to get the superblock for
649 *
650 * Scans the superblock list and finds the superblock of the file system
651 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400652 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200653 */
654struct super_block *get_active_super(struct block_device *bdev)
655{
656 struct super_block *sb;
657
658 if (!bdev)
659 return NULL;
660
Al Viro14945832010-03-22 20:15:33 -0400661restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200662 spin_lock(&sb_lock);
663 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500664 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400665 continue;
Al Viro14945832010-03-22 20:15:33 -0400666 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400667 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400668 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400669 up_write(&sb->s_umount);
670 return sb;
Al Viro14945832010-03-22 20:15:33 -0400671 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200672 }
673 spin_unlock(&sb_lock);
674 return NULL;
675}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
Al Virodf40c012010-03-22 20:23:25 -0400677struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700679 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700682rescan:
683 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500684 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400685 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700686 if (sb->s_dev == dev) {
687 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700689 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400690 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500691 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700692 return sb;
693 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400694 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700695 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400696 __put_super(sb);
697 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 }
699 }
700 spin_unlock(&sb_lock);
701 return NULL;
702}
703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 * do_remount_sb - asks filesystem to change mount options.
706 * @sb: superblock in question
707 * @flags: numeric part of options
708 * @data: the rest of options
709 * @force: whether or not to force the change
710 *
711 * Alters the mount options of a mounted file system.
712 */
713int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
714{
715 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400716 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200717
Jan Kara5accdf82012-06-12 16:20:34 +0200718 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200719 return -EBUSY;
720
David Howells93614012006-09-30 20:45:40 +0200721#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
723 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200724#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200725
Nick Piggind208bbd2009-12-21 16:28:53 -0800726 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800727
Al Viro0aec09d2014-08-07 07:32:06 -0400728 if (remount_ro) {
Al Virofdab6842015-01-11 10:57:27 -0500729 if (!hlist_empty(&sb->s_pins)) {
Al Viro0aec09d2014-08-07 07:32:06 -0400730 up_write(&sb->s_umount);
Al Virofdab6842015-01-11 10:57:27 -0500731 group_pin_kill(&sb->s_pins);
Al Viro0aec09d2014-08-07 07:32:06 -0400732 down_write(&sb->s_umount);
733 if (!sb->s_root)
734 return 0;
735 if (sb->s_writers.frozen != SB_UNFROZEN)
736 return -EBUSY;
737 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
738 }
739 }
740 shrink_dcache_sb(sb);
741
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 /* If we are remounting RDONLY and current sb is read/write,
743 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800744 if (remount_ro) {
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100745 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400746 sb->s_readonly_remount = 1;
747 smp_wmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100748 } else {
749 retval = sb_prepare_remount_readonly(sb);
750 if (retval)
751 return retval;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100752 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 }
754
755 if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 retval = sb->s_op->remount_fs(sb, &flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200757 if (retval) {
758 if (!force)
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100759 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200760 /* If forced remount, go ahead despite any errors */
761 WARN(1, "forced remount of a %s fs returned %i\n",
762 sb->s_type->name, retval);
763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 }
765 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100766 /* Needs to be ordered wrt mnt_is_readonly() */
767 smp_wmb();
768 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400769
Nick Piggind208bbd2009-12-21 16:28:53 -0800770 /*
771 * Some filesystems modify their metadata via some other path than the
772 * bdev buffer cache (eg. use a private mapping, or directories in
773 * pagecache, etc). Also file data modifications go via their own
774 * mappings. So If we try to mount readonly then copy the filesystem
775 * from bdev, we could get stale data, so invalidate it to give a best
776 * effort at coherency.
777 */
778 if (remount_ro && sb->s_bdev)
779 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 return 0;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100781
782cancel_readonly:
783 sb->s_readonly_remount = 0;
784 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785}
786
Jens Axboea2a95372009-03-17 09:38:40 +0100787static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788{
Al Virodca33252010-07-25 02:31:46 +0400789 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
791 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400792 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500793 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400794 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 sb->s_count++;
796 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400797 down_write(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500798 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
799 !(sb->s_flags & MS_RDONLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 * What lock protects sb->s_flags??
802 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 }
Al Viro443b94b2009-05-05 23:48:50 -0400805 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400807 if (p)
808 __put_super(p);
809 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 }
Al Virodca33252010-07-25 02:31:46 +0400811 if (p)
812 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100814 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 printk("Emergency Remount complete\n");
816}
817
818void emergency_remount(void)
819{
Jens Axboea2a95372009-03-17 09:38:40 +0100820 struct work_struct *work;
821
822 work = kmalloc(sizeof(*work), GFP_ATOMIC);
823 if (work) {
824 INIT_WORK(work, do_emergency_remount);
825 schedule_work(work);
826 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827}
828
829/*
830 * Unnamed block devices are dummy devices used by virtual
831 * filesystems which don't use real block-devices. -- jrs
832 */
833
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400834static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Thomas Bächlera2a4dc42014-04-03 21:55:37 +0200836/* Many userspace utilities consider an FSID of 0 invalid.
837 * Always return at least 1 from get_anon_bdev.
838 */
839static int unnamed_dev_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
Al Viro0ee5dc62011-07-07 15:44:25 -0400841int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842{
843 int dev;
844 int error;
845
846 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400847 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 return -ENOMEM;
849 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400850 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400851 if (!error)
852 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 spin_unlock(&unnamed_dev_lock);
854 if (error == -EAGAIN)
855 /* We raced and lost with another CPU. */
856 goto retry;
857 else if (error)
858 return -EAGAIN;
859
Wang YanQing1af95de2015-06-23 18:54:45 +0800860 if (dev >= (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400862 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400863 if (unnamed_dev_start > dev)
864 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 spin_unlock(&unnamed_dev_lock);
866 return -EMFILE;
867 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400868 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 return 0;
870}
Al Viro0ee5dc62011-07-07 15:44:25 -0400871EXPORT_SYMBOL(get_anon_bdev);
872
873void free_anon_bdev(dev_t dev)
874{
875 int slot = MINOR(dev);
876 spin_lock(&unnamed_dev_lock);
877 ida_remove(&unnamed_dev_ida, slot);
878 if (slot < unnamed_dev_start)
879 unnamed_dev_start = slot;
880 spin_unlock(&unnamed_dev_lock);
881}
882EXPORT_SYMBOL(free_anon_bdev);
883
884int set_anon_super(struct super_block *s, void *data)
885{
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100886 return get_anon_bdev(&s->s_dev);
Al Viro0ee5dc62011-07-07 15:44:25 -0400887}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
889EXPORT_SYMBOL(set_anon_super);
890
891void kill_anon_super(struct super_block *sb)
892{
Al Viro0ee5dc62011-07-07 15:44:25 -0400893 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400895 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896}
897
898EXPORT_SYMBOL(kill_anon_super);
899
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900void kill_litter_super(struct super_block *sb)
901{
902 if (sb->s_root)
903 d_genocide(sb->s_root);
904 kill_anon_super(sb);
905}
906
907EXPORT_SYMBOL(kill_litter_super);
908
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700909static int ns_test_super(struct super_block *sb, void *data)
910{
911 return sb->s_fs_info == data;
912}
913
914static int ns_set_super(struct super_block *sb, void *data)
915{
916 sb->s_fs_info = data;
917 return set_anon_super(sb, NULL);
918}
919
Al Viroceefda62010-07-26 13:16:50 +0400920struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
921 void *data, int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700922{
923 struct super_block *sb;
924
David Howells9249e172012-06-25 12:55:37 +0100925 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700926 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +0400927 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700928
929 if (!sb->s_root) {
930 int err;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700931 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
932 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400933 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +0400934 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700935 }
936
937 sb->s_flags |= MS_ACTIVE;
938 }
939
Al Viroceefda62010-07-26 13:16:50 +0400940 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700941}
942
Al Viroceefda62010-07-26 13:16:50 +0400943EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700944
David Howells93614012006-09-30 20:45:40 +0200945#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946static int set_bdev_super(struct super_block *s, void *data)
947{
948 s->s_bdev = data;
949 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +0200950
951 /*
952 * We set the bdi here to the queue backing, file systems can
953 * overwrite this in ->fill_super()
954 */
955 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 return 0;
957}
958
959static int test_bdev_super(struct super_block *s, void *data)
960{
961 return (void *)s->s_bdev == data;
962}
963
Al Viro152a0832010-07-25 00:46:55 +0400964struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +0400966 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967{
968 struct block_device *bdev;
969 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +0100970 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 int error = 0;
972
Al Viro30c40d22008-02-22 19:50:45 -0500973 if (!(flags & MS_RDONLY))
974 mode |= FMODE_WRITE;
975
Tejun Heod4d77622010-11-13 11:55:18 +0100976 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +0400978 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
980 /*
981 * once the super is inserted into the list by sget, s_umount
982 * will protect the lockfs code from trying to start a snapshot
983 * while we are mounting
984 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200985 mutex_lock(&bdev->bd_fsfreeze_mutex);
986 if (bdev->bd_fsfreeze_count > 0) {
987 mutex_unlock(&bdev->bd_fsfreeze_mutex);
988 error = -EBUSY;
989 goto error_bdev;
990 }
David Howells9249e172012-06-25 12:55:37 +0100991 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
992 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200993 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -0700995 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
997 if (s->s_root) {
998 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400999 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001000 error = -EBUSY;
1001 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 }
David Howells454e2392006-06-23 02:02:57 -07001003
Tejun Heo4f331f02010-07-20 15:18:07 -07001004 /*
1005 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +01001006 * __invalidate_device(). blkdev_put() acquires
1007 * bd_mutex and can't be called under s_umount. Drop
1008 * s_umount temporarily. This is safe as we're
1009 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -07001010 */
1011 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +01001012 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -07001013 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 } else {
1015 char b[BDEVNAME_SIZE];
1016
Al Viro30c40d22008-02-22 19:50:45 -05001017 s->s_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
Pekka Enberge78c9a02006-01-08 01:03:39 -08001019 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001020 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001022 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001023 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001024 }
David Howells454e2392006-06-23 02:02:57 -07001025
1026 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001027 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 }
1029
Al Viro152a0832010-07-25 00:46:55 +04001030 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
David Howells454e2392006-06-23 02:02:57 -07001032error_s:
1033 error = PTR_ERR(s);
1034error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001035 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001036error:
Al Viro152a0832010-07-25 00:46:55 +04001037 return ERR_PTR(error);
1038}
1039EXPORT_SYMBOL(mount_bdev);
1040
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041void kill_block_super(struct super_block *sb)
1042{
1043 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001044 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001046 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 generic_shutdown_super(sb);
1048 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001049 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001050 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051}
1052
1053EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001054#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
Al Viro3c26ff62010-07-25 11:46:36 +04001056struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001058 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
1060 int error;
David Howells9249e172012-06-25 12:55:37 +01001061 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
1063 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001064 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001066 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001068 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001069 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 }
1071 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001072 return dget(s->s_root);
1073}
1074EXPORT_SYMBOL(mount_nodev);
1075
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076static int compare_single(struct super_block *s, void *p)
1077{
1078 return 1;
1079}
1080
Al Virofc14f2f2010-07-25 01:48:30 +04001081struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001083 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084{
1085 struct super_block *s;
1086 int error;
1087
David Howells9249e172012-06-25 12:55:37 +01001088 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001090 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 if (!s->s_root) {
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001092 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001094 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001095 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 }
1097 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001098 } else {
1099 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 }
Al Virofc14f2f2010-07-25 01:48:30 +04001101 return dget(s->s_root);
1102}
1103EXPORT_SYMBOL(mount_single);
1104
Al Viro9d412a42011-03-17 22:08:28 -04001105struct dentry *
1106mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107{
Al Viroc96e41e2010-07-25 00:17:56 +04001108 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001109 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001111 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001112
Eric Parise0007522008-03-05 10:31:54 -05001113 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001115 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001116 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117
Eric Parise0007522008-03-05 10:31:54 -05001118 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001119 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 }
1122
Al Viro1a102ff2011-03-16 09:07:58 -04001123 root = type->mount(type, flags, name, data);
1124 if (IS_ERR(root)) {
1125 error = PTR_ERR(root);
1126 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001127 }
Al Viro9d412a42011-03-17 22:08:28 -04001128 sb = root->d_sb;
1129 BUG_ON(!sb);
1130 WARN_ON(!sb->s_bdi);
1131 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001132
Al Viro9d412a42011-03-17 22:08:28 -04001133 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001134 if (error)
1135 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001136
Jeff Layton42cb56a2009-09-18 13:05:53 -07001137 /*
1138 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1139 * but s_maxbytes was an unsigned long long for many releases. Throw
1140 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001141 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001142 */
Al Viro9d412a42011-03-17 22:08:28 -04001143 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1144 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001145
Al Viro9d412a42011-03-17 22:08:28 -04001146 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001147 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001148 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001150 dput(root);
1151 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152out_free_secdata:
1153 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154out:
David Howells454e2392006-06-23 02:02:57 -07001155 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156}
1157
Jan Kara5accdf82012-06-12 16:20:34 +02001158/*
1159 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1160 * instead.
1161 */
1162void __sb_end_write(struct super_block *sb, int level)
1163{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001164 percpu_up_read(sb->s_writers.rw_sem + level-1);
Jan Kara5accdf82012-06-12 16:20:34 +02001165}
1166EXPORT_SYMBOL(__sb_end_write);
1167
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001168/*
1169 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1170 * instead.
1171 */
1172int __sb_start_write(struct super_block *sb, int level, bool wait)
1173{
1174 bool force_trylock = false;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001175 int ret = 1;
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001176
1177#ifdef CONFIG_LOCKDEP
1178 /*
1179 * We want lockdep to tell us about possible deadlocks with freezing
1180 * but it's it bit tricky to properly instrument it. Getting a freeze
1181 * protection works as getting a read lock but there are subtle
1182 * problems. XFS for example gets freeze protection on internal level
1183 * twice in some cases, which is OK only because we already hold a
1184 * freeze protection also on higher level. Due to these cases we have
1185 * to use wait == F (trylock mode) which must not fail.
1186 */
1187 if (wait) {
1188 int i;
1189
1190 for (i = 0; i < level - 1; i++)
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001191 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001192 force_trylock = true;
1193 break;
1194 }
1195 }
1196#endif
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001197 if (wait && !force_trylock)
1198 percpu_down_read(sb->s_writers.rw_sem + level-1);
1199 else
1200 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1201
Oleg Nesterovf4b554a2015-07-20 00:50:55 +02001202 WARN_ON(force_trylock & !ret);
1203 return ret;
1204}
Jan Kara5accdf82012-06-12 16:20:34 +02001205EXPORT_SYMBOL(__sb_start_write);
1206
1207/**
1208 * sb_wait_write - wait until all writers to given file system finish
1209 * @sb: the super for which we wait
1210 * @level: type of writers we wait for (normal vs page fault)
1211 *
1212 * This function waits until there are no writers of given type to given file
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001213 * system.
Jan Kara5accdf82012-06-12 16:20:34 +02001214 */
1215static void sb_wait_write(struct super_block *sb, int level)
1216{
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001217 percpu_down_write(sb->s_writers.rw_sem + level-1);
Oleg Nesterov0e28e012015-08-11 16:28:29 +02001218 /*
1219 * We are going to return to userspace and forget about this lock, the
1220 * ownership goes to the caller of thaw_super() which does unlock.
1221 *
1222 * FIXME: we should do this before return from freeze_super() after we
1223 * called sync_filesystem(sb) and s_op->freeze_fs(sb), and thaw_super()
1224 * should re-acquire these locks before s_op->unfreeze_fs(sb). However
1225 * this leads to lockdep false-positives, so currently we do the early
1226 * release right after acquire.
1227 */
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001228 percpu_rwsem_release(sb->s_writers.rw_sem + level-1, 0, _THIS_IP_);
1229}
Jan Kara5accdf82012-06-12 16:20:34 +02001230
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001231static void sb_freeze_unlock(struct super_block *sb)
1232{
1233 int level;
Jan Kara5accdf82012-06-12 16:20:34 +02001234
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001235 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1236 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
Jan Kara5accdf82012-06-12 16:20:34 +02001237
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001238 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1239 percpu_up_write(sb->s_writers.rw_sem + level);
Jan Kara5accdf82012-06-12 16:20:34 +02001240}
1241
Josef Bacik18e9e512010-03-23 10:34:56 -04001242/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001243 * freeze_super - lock the filesystem and force it into a consistent state
1244 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001245 *
1246 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1247 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1248 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001249 *
1250 * During this function, sb->s_writers.frozen goes through these values:
1251 *
1252 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1253 *
1254 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1255 * writes should be blocked, though page faults are still allowed. We wait for
1256 * all writes to complete and then proceed to the next stage.
1257 *
1258 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1259 * but internal fs threads can still modify the filesystem (although they
1260 * should not dirty new pages or inodes), writeback can run etc. After waiting
1261 * for all running page faults we sync the filesystem which will clean all
1262 * dirty pages and inodes (no new dirty pages or inodes can be created when
1263 * sync is running).
1264 *
1265 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1266 * modification are blocked (e.g. XFS preallocation truncation on inode
1267 * reclaim). This is usually implemented by blocking new transactions for
1268 * filesystems that have them and need this additional guard. After all
1269 * internal writers are finished we call ->freeze_fs() to finish filesystem
1270 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1271 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1272 *
1273 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001274 */
1275int freeze_super(struct super_block *sb)
1276{
1277 int ret;
1278
1279 atomic_inc(&sb->s_active);
1280 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001281 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001282 deactivate_locked_super(sb);
1283 return -EBUSY;
1284 }
1285
Al Virodabe0dc2012-01-03 21:01:29 -05001286 if (!(sb->s_flags & MS_BORN)) {
1287 up_write(&sb->s_umount);
1288 return 0; /* sic - it's "nothing to do" */
1289 }
1290
Josef Bacik18e9e512010-03-23 10:34:56 -04001291 if (sb->s_flags & MS_RDONLY) {
Jan Kara5accdf82012-06-12 16:20:34 +02001292 /* Nothing to do really... */
1293 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001294 up_write(&sb->s_umount);
1295 return 0;
1296 }
1297
Jan Kara5accdf82012-06-12 16:20:34 +02001298 sb->s_writers.frozen = SB_FREEZE_WRITE;
Jan Kara5accdf82012-06-12 16:20:34 +02001299 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1300 up_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001301 sb_wait_write(sb, SB_FREEZE_WRITE);
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001302 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001303
1304 /* Now we go and block page faults... */
Jan Kara5accdf82012-06-12 16:20:34 +02001305 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
Jan Kara5accdf82012-06-12 16:20:34 +02001306 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1307
1308 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001309 sync_filesystem(sb);
1310
Jan Kara5accdf82012-06-12 16:20:34 +02001311 /* Now wait for internal filesystem counter */
1312 sb->s_writers.frozen = SB_FREEZE_FS;
Jan Kara5accdf82012-06-12 16:20:34 +02001313 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001314
Josef Bacik18e9e512010-03-23 10:34:56 -04001315 if (sb->s_op->freeze_fs) {
1316 ret = sb->s_op->freeze_fs(sb);
1317 if (ret) {
1318 printk(KERN_ERR
1319 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001320 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001321 sb_freeze_unlock(sb);
Jan Kara5accdf82012-06-12 16:20:34 +02001322 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001323 deactivate_locked_super(sb);
1324 return ret;
1325 }
1326 }
Jan Kara5accdf82012-06-12 16:20:34 +02001327 /*
1328 * This is just for debugging purposes so that fs can warn if it
1329 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1330 */
1331 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001332 up_write(&sb->s_umount);
1333 return 0;
1334}
1335EXPORT_SYMBOL(freeze_super);
1336
1337/**
1338 * thaw_super -- unlock filesystem
1339 * @sb: the super to thaw
1340 *
1341 * Unlocks the filesystem and marks it writeable again after freeze_super().
1342 */
1343int thaw_super(struct super_block *sb)
1344{
1345 int error;
1346
1347 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001348 if (sb->s_writers.frozen == SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001349 up_write(&sb->s_umount);
1350 return -EINVAL;
1351 }
1352
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001353 if (sb->s_flags & MS_RDONLY) {
1354 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001355 goto out;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001356 }
Josef Bacik18e9e512010-03-23 10:34:56 -04001357
1358 if (sb->s_op->unfreeze_fs) {
1359 error = sb->s_op->unfreeze_fs(sb);
1360 if (error) {
1361 printk(KERN_ERR
1362 "VFS:Filesystem thaw failed\n");
Josef Bacik18e9e512010-03-23 10:34:56 -04001363 up_write(&sb->s_umount);
1364 return error;
1365 }
1366 }
1367
Jan Kara5accdf82012-06-12 16:20:34 +02001368 sb->s_writers.frozen = SB_UNFROZEN;
Oleg Nesterov8129ed22015-08-11 17:05:04 +02001369 sb_freeze_unlock(sb);
1370out:
Jan Kara5accdf82012-06-12 16:20:34 +02001371 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001372 deactivate_locked_super(sb);
Josef Bacik18e9e512010-03-23 10:34:56 -04001373 return 0;
1374}
1375EXPORT_SYMBOL(thaw_super);