blob: 2b7dc90ccdbb4ae1ceac7725967cec2a643e21c7 [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
Al Viro7eb5e882013-10-01 15:09:58 -0400138/**
139 * destroy_super - frees a superblock
140 * @s: superblock to free
141 *
142 * Frees a superblock.
143 */
144static void destroy_super(struct super_block *s)
Jan Kara5accdf82012-06-12 16:20:34 +0200145{
146 int i;
Al Viro7eb5e882013-10-01 15:09:58 -0400147 list_lru_destroy(&s->s_dentry_lru);
148 list_lru_destroy(&s->s_inode_lru);
Jan Kara5accdf82012-06-12 16:20:34 +0200149 for (i = 0; i < SB_FREEZE_LEVELS; i++)
150 percpu_counter_destroy(&s->s_writers.counter[i]);
Al Viro7eb5e882013-10-01 15:09:58 -0400151 security_sb_free(s);
152 WARN_ON(!list_empty(&s->s_mounts));
153 kfree(s->s_subtype);
154 kfree(s->s_options);
Al Viroe2fec7c2013-10-04 17:06:56 -0400155 kfree_rcu(s, rcu);
Jan Kara5accdf82012-06-12 16:20:34 +0200156}
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158/**
159 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700160 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100161 * @flags: the mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 *
163 * Allocates and initializes a new &struct super_block. alloc_super()
164 * returns a pointer new superblock or %NULL if allocation had failed.
165 */
David Howells9249e172012-06-25 12:55:37 +0100166static struct super_block *alloc_super(struct file_system_type *type, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800168 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700169 static const struct super_operations default_op;
Al Viro7eb5e882013-10-01 15:09:58 -0400170 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
Al Viro7eb5e882013-10-01 15:09:58 -0400172 if (!s)
173 return NULL;
174
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800175 INIT_LIST_HEAD(&s->s_mounts);
176
Al Viro7eb5e882013-10-01 15:09:58 -0400177 if (security_sb_alloc(s))
178 goto fail;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200179
Al Viro7eb5e882013-10-01 15:09:58 -0400180 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
Tejun Heo908c7f12014-09-08 09:51:29 +0900181 if (percpu_counter_init(&s->s_writers.counter[i], 0,
182 GFP_KERNEL) < 0)
Al Viro7eb5e882013-10-01 15:09:58 -0400183 goto fail;
184 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
185 &type->s_writers_key[i], 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 }
Al Viro7eb5e882013-10-01 15:09:58 -0400187 init_waitqueue_head(&s->s_writers.wait);
188 init_waitqueue_head(&s->s_writers.wait_unfrozen);
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100189 s->s_bdi = &noop_backing_dev_info;
Al Viro7eb5e882013-10-01 15:09:58 -0400190 s->s_flags = flags;
Al Viro7eb5e882013-10-01 15:09:58 -0400191 INIT_HLIST_NODE(&s->s_instances);
192 INIT_HLIST_BL_HEAD(&s->s_anon);
193 INIT_LIST_HEAD(&s->s_inodes);
194
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800195 if (list_lru_init_memcg(&s->s_dentry_lru))
Al Viro7eb5e882013-10-01 15:09:58 -0400196 goto fail;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800197 if (list_lru_init_memcg(&s->s_inode_lru))
Al Viro7eb5e882013-10-01 15:09:58 -0400198 goto fail;
199
Al Viro7eb5e882013-10-01 15:09:58 -0400200 init_rwsem(&s->s_umount);
201 lockdep_set_class(&s->s_umount, &type->s_umount_key);
202 /*
203 * sget() can have s_umount recursion.
204 *
205 * When it cannot find a suitable sb, it allocates a new
206 * one (this one), and tries again to find a suitable old
207 * one.
208 *
209 * In case that succeeds, it will acquire the s_umount
210 * lock of the old one. Since these are clearly distrinct
211 * locks, and this object isn't exposed yet, there's no
212 * risk of deadlocks.
213 *
214 * Annotate this by putting this lock in a different
215 * subclass.
216 */
217 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
218 s->s_count = 1;
219 atomic_set(&s->s_active, 1);
220 mutex_init(&s->s_vfs_rename_mutex);
221 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
222 mutex_init(&s->s_dquot.dqio_mutex);
223 mutex_init(&s->s_dquot.dqonoff_mutex);
Al Viro7eb5e882013-10-01 15:09:58 -0400224 s->s_maxbytes = MAX_NON_LFS;
225 s->s_op = &default_op;
226 s->s_time_gran = 1000000000;
227 s->cleancache_poolid = -1;
228
229 s->s_shrink.seeks = DEFAULT_SEEKS;
230 s->s_shrink.scan_objects = super_cache_scan;
231 s->s_shrink.count_objects = super_cache_count;
232 s->s_shrink.batch = 1024;
Vladimir Davydov2acb60a2015-02-12 14:59:14 -0800233 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000235
Al Viro7eb5e882013-10-01 15:09:58 -0400236fail:
237 destroy_super(s);
238 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239}
240
241/* Superblock refcounting */
242
243/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400244 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 */
Al Virof47ec3f2011-11-21 21:15:42 -0500246static void __put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 if (!--sb->s_count) {
Al Viro551de6f2010-03-22 19:36:35 -0400249 list_del_init(&sb->s_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 destroy_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252}
253
254/**
255 * put_super - drop a temporary reference to superblock
256 * @sb: superblock in question
257 *
258 * Drops a temporary reference, frees superblock if there's no
259 * references left.
260 */
Al Virof47ec3f2011-11-21 21:15:42 -0500261static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262{
263 spin_lock(&sb_lock);
264 __put_super(sb);
265 spin_unlock(&sb_lock);
266}
267
268
269/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400270 * deactivate_locked_super - drop an active reference to superblock
271 * @s: superblock to deactivate
272 *
Al Viro1712ac82010-03-22 15:22:31 -0400273 * Drops an active reference to superblock, converting it into a temprory
274 * one if there is no other active references left. In that case we
275 * tell fs driver to shut it down and drop the temporary reference we
276 * had just acquired.
277 *
278 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400279 */
280void deactivate_locked_super(struct super_block *s)
281{
282 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400283 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400284 cleancache_invalidate_fs(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000285 unregister_shrinker(&s->s_shrink);
Dave Chinner28f2cd4f2014-06-04 16:10:46 -0700286 fs->kill_sb(s);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000287
Vladimir Davydovc0a5b562015-02-12 14:59:07 -0800288 /*
289 * Since list_lru_destroy() may sleep, we cannot call it from
290 * put_super(), where we hold the sb_lock. Therefore we destroy
291 * the lru lists right now.
292 */
293 list_lru_destroy(&s->s_dentry_lru);
294 list_lru_destroy(&s->s_inode_lru);
295
Al Viro74dbbdd2009-05-06 01:07:50 -0400296 put_filesystem(fs);
297 put_super(s);
298 } else {
299 up_write(&s->s_umount);
300 }
301}
302
303EXPORT_SYMBOL(deactivate_locked_super);
304
305/**
Al Viro1712ac82010-03-22 15:22:31 -0400306 * deactivate_super - drop an active reference to superblock
307 * @s: superblock to deactivate
308 *
309 * Variant of deactivate_locked_super(), except that superblock is *not*
310 * locked by caller. If we are going to drop the final active reference,
311 * lock will be acquired prior to that.
312 */
313void deactivate_super(struct super_block *s)
314{
315 if (!atomic_add_unless(&s->s_active, -1, 1)) {
316 down_write(&s->s_umount);
317 deactivate_locked_super(s);
318 }
319}
320
321EXPORT_SYMBOL(deactivate_super);
322
323/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 * grab_super - acquire an active reference
325 * @s: reference we are trying to make active
326 *
327 * Tries to acquire an active reference. grab_super() is used when we
328 * had just found a superblock in super_blocks or fs_type->fs_supers
329 * and want to turn it into a full-blown active reference. grab_super()
330 * is called with sb_lock held and drops it. Returns 1 in case of
331 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400332 * dying when grab_super() had been called). Note that this is only
333 * called for superblocks not in rundown mode (== ones still on ->fs_supers
334 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700336static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
338 s->s_count++;
339 spin_unlock(&sb_lock);
340 down_write(&s->s_umount);
Al Viroacfec9a2013-07-20 03:13:55 +0400341 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
342 put_super(s);
343 return 1;
344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 up_write(&s->s_umount);
346 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 return 0;
348}
349
David Howellscf9a2ae2006-08-29 19:05:54 +0100350/*
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300351 * trylock_super - try to grab ->s_umount shared
Wanpeng Li331cbde2012-06-09 11:10:55 +0800352 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000353 *
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300354 * Try to prevent fs shutdown. This is used in places where we
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000355 * cannot take an active reference but we need to ensure that the
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300356 * filesystem is not shut down while we are working on it. It returns
357 * false if we cannot acquire s_umount or if we lose the race and
358 * filesystem already got into shutdown, and returns true with the s_umount
359 * lock held in read mode in case of success. On successful return,
360 * the caller must drop the s_umount lock when done.
361 *
362 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
363 * The reason why it's safe is that we are OK with doing trylock instead
364 * of down_read(). There's a couple of places that are OK with that, but
365 * it's very much not a general-purpose interface.
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000366 */
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300367bool trylock_super(struct super_block *sb)
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000368{
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000369 if (down_read_trylock(&sb->s_umount)) {
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300370 if (!hlist_unhashed(&sb->s_instances) &&
371 sb->s_root && (sb->s_flags & MS_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000372 return true;
373 up_read(&sb->s_umount);
374 }
375
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000376 return false;
377}
378
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379/**
380 * generic_shutdown_super - common helper for ->kill_sb()
381 * @sb: superblock to kill
382 *
383 * generic_shutdown_super() does all fs-independent work on superblock
384 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
385 * that need destruction out of superblock, call generic_shutdown_super()
386 * and release aforementioned objects. Note: dentries and inodes _are_
387 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700388 *
389 * Upon calling this function, the filesystem may no longer alter or
390 * rearrange the set of dentries belonging to this super_block, nor may it
391 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 */
393void generic_shutdown_super(struct super_block *sb)
394{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800395 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
David Howellsc636ebd2006-10-11 01:22:19 -0700397 if (sb->s_root) {
398 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200399 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 sb->s_flags &= ~MS_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800401
Al Viro63997e92010-10-25 20:49:35 -0400402 fsnotify_unmount_inodes(&sb->s_inodes);
403
404 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200406 if (sb->s_dio_done_wq) {
407 destroy_workqueue(sb->s_dio_done_wq);
408 sb->s_dio_done_wq = NULL;
409 }
410
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 if (sop->put_super)
412 sop->put_super(sb);
413
Al Viro63997e92010-10-25 20:49:35 -0400414 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800415 printk("VFS: Busy inodes after unmount of %s. "
416 "Self-destruct in 5 seconds. Have a nice day...\n",
417 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 }
420 spin_lock(&sb_lock);
421 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500422 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 spin_unlock(&sb_lock);
424 up_write(&sb->s_umount);
425}
426
427EXPORT_SYMBOL(generic_shutdown_super);
428
429/**
430 * sget - find or create a superblock
431 * @type: filesystem type superblock should belong to
432 * @test: comparison callback
433 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100434 * @flags: mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 * @data: argument to each of them
436 */
437struct super_block *sget(struct file_system_type *type,
438 int (*test)(struct super_block *,void *),
439 int (*set)(struct super_block *,void *),
David Howells9249e172012-06-25 12:55:37 +0100440 int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 void *data)
442{
443 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700444 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 int err;
446
447retry:
448 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700449 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800450 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700451 if (!test(old, data))
452 continue;
453 if (!grab_super(old))
454 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700455 if (s) {
456 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700457 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400458 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700459 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700460 return old;
461 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 }
463 if (!s) {
464 spin_unlock(&sb_lock);
David Howells9249e172012-06-25 12:55:37 +0100465 s = alloc_super(type, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 if (!s)
467 return ERR_PTR(-ENOMEM);
468 goto retry;
469 }
470
471 err = set(s, data);
472 if (err) {
473 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700474 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 destroy_super(s);
476 return ERR_PTR(err);
477 }
478 s->s_type = type;
479 strlcpy(s->s_id, type->name, sizeof(s->s_id));
480 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500481 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 spin_unlock(&sb_lock);
483 get_filesystem(type);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000484 register_shrinker(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 return s;
486}
487
488EXPORT_SYMBOL(sget);
489
490void drop_super(struct super_block *sb)
491{
492 up_read(&sb->s_umount);
493 put_super(sb);
494}
495
496EXPORT_SYMBOL(drop_super);
497
Christoph Hellwige5004752009-05-05 16:08:56 +0200498/**
Al Viro01a05b32010-03-23 06:06:58 -0400499 * iterate_supers - call function for all active superblocks
500 * @f: function to call
501 * @arg: argument to pass to it
502 *
503 * Scans the superblock list and calls given function, passing it
504 * locked superblock and given argument.
505 */
506void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
507{
Al Virodca33252010-07-25 02:31:46 +0400508 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400509
510 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400511 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500512 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400513 continue;
514 sb->s_count++;
515 spin_unlock(&sb_lock);
516
517 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500518 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400519 f(sb, arg);
520 up_read(&sb->s_umount);
521
522 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400523 if (p)
524 __put_super(p);
525 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400526 }
Al Virodca33252010-07-25 02:31:46 +0400527 if (p)
528 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400529 spin_unlock(&sb_lock);
530}
531
532/**
Al Viro43e15cd2011-06-03 20:16:57 -0400533 * iterate_supers_type - call function for superblocks of given type
534 * @type: fs type
535 * @f: function to call
536 * @arg: argument to pass to it
537 *
538 * Scans the superblock list and calls given function, passing it
539 * locked superblock and given argument.
540 */
541void iterate_supers_type(struct file_system_type *type,
542 void (*f)(struct super_block *, void *), void *arg)
543{
544 struct super_block *sb, *p = NULL;
545
546 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800547 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400548 sb->s_count++;
549 spin_unlock(&sb_lock);
550
551 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500552 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400553 f(sb, arg);
554 up_read(&sb->s_umount);
555
556 spin_lock(&sb_lock);
557 if (p)
558 __put_super(p);
559 p = sb;
560 }
561 if (p)
562 __put_super(p);
563 spin_unlock(&sb_lock);
564}
565
566EXPORT_SYMBOL(iterate_supers_type);
567
568/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 * get_super - get the superblock of a device
570 * @bdev: device to get the superblock for
571 *
572 * Scans the superblock list and finds the superblock of the file system
573 * mounted on the device given. %NULL is returned if no match is found.
574 */
575
Al Virodf40c012010-03-22 20:23:25 -0400576struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700578 struct super_block *sb;
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 if (!bdev)
581 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700584rescan:
585 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500586 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400587 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700588 if (sb->s_bdev == bdev) {
589 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700591 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400592 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500593 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700594 return sb;
595 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400596 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700597 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400598 __put_super(sb);
599 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 }
601 }
602 spin_unlock(&sb_lock);
603 return NULL;
604}
605
606EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200607
608/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100609 * get_super_thawed - get thawed superblock of a device
610 * @bdev: device to get the superblock for
611 *
612 * Scans the superblock list and finds the superblock of the file system
613 * mounted on the device. The superblock is returned once it is thawed
614 * (or immediately if it was not frozen). %NULL is returned if no match
615 * is found.
616 */
617struct super_block *get_super_thawed(struct block_device *bdev)
618{
619 while (1) {
620 struct super_block *s = get_super(bdev);
Jan Kara5accdf82012-06-12 16:20:34 +0200621 if (!s || s->s_writers.frozen == SB_UNFROZEN)
Jan Kara6b6dc832012-02-10 11:03:00 +0100622 return s;
623 up_read(&s->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +0200624 wait_event(s->s_writers.wait_unfrozen,
625 s->s_writers.frozen == SB_UNFROZEN);
Jan Kara6b6dc832012-02-10 11:03:00 +0100626 put_super(s);
627 }
628}
629EXPORT_SYMBOL(get_super_thawed);
630
631/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200632 * get_active_super - get an active reference to the superblock of a device
633 * @bdev: device to get the superblock for
634 *
635 * Scans the superblock list and finds the superblock of the file system
636 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400637 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200638 */
639struct super_block *get_active_super(struct block_device *bdev)
640{
641 struct super_block *sb;
642
643 if (!bdev)
644 return NULL;
645
Al Viro14945832010-03-22 20:15:33 -0400646restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200647 spin_lock(&sb_lock);
648 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500649 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400650 continue;
Al Viro14945832010-03-22 20:15:33 -0400651 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400652 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400653 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400654 up_write(&sb->s_umount);
655 return sb;
Al Viro14945832010-03-22 20:15:33 -0400656 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200657 }
658 spin_unlock(&sb_lock);
659 return NULL;
660}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
Al Virodf40c012010-03-22 20:23:25 -0400662struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700664 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700667rescan:
668 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500669 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400670 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700671 if (sb->s_dev == dev) {
672 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700674 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400675 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500676 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700677 return sb;
678 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400679 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700680 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400681 __put_super(sb);
682 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 }
684 }
685 spin_unlock(&sb_lock);
686 return NULL;
687}
688
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 * do_remount_sb - asks filesystem to change mount options.
691 * @sb: superblock in question
692 * @flags: numeric part of options
693 * @data: the rest of options
694 * @force: whether or not to force the change
695 *
696 * Alters the mount options of a mounted file system.
697 */
698int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
699{
700 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400701 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200702
Jan Kara5accdf82012-06-12 16:20:34 +0200703 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200704 return -EBUSY;
705
David Howells93614012006-09-30 20:45:40 +0200706#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
708 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200709#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200710
Nick Piggind208bbd2009-12-21 16:28:53 -0800711 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800712
Al Viro0aec09d2014-08-07 07:32:06 -0400713 if (remount_ro) {
Al Virofdab6842015-01-11 10:57:27 -0500714 if (!hlist_empty(&sb->s_pins)) {
Al Viro0aec09d2014-08-07 07:32:06 -0400715 up_write(&sb->s_umount);
Al Virofdab6842015-01-11 10:57:27 -0500716 group_pin_kill(&sb->s_pins);
Al Viro0aec09d2014-08-07 07:32:06 -0400717 down_write(&sb->s_umount);
718 if (!sb->s_root)
719 return 0;
720 if (sb->s_writers.frozen != SB_UNFROZEN)
721 return -EBUSY;
722 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
723 }
724 }
725 shrink_dcache_sb(sb);
726
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 /* If we are remounting RDONLY and current sb is read/write,
728 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800729 if (remount_ro) {
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100730 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400731 sb->s_readonly_remount = 1;
732 smp_wmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100733 } else {
734 retval = sb_prepare_remount_readonly(sb);
735 if (retval)
736 return retval;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100737 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 }
739
740 if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 retval = sb->s_op->remount_fs(sb, &flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200742 if (retval) {
743 if (!force)
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100744 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200745 /* If forced remount, go ahead despite any errors */
746 WARN(1, "forced remount of a %s fs returned %i\n",
747 sb->s_type->name, retval);
748 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 }
750 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100751 /* Needs to be ordered wrt mnt_is_readonly() */
752 smp_wmb();
753 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400754
Nick Piggind208bbd2009-12-21 16:28:53 -0800755 /*
756 * Some filesystems modify their metadata via some other path than the
757 * bdev buffer cache (eg. use a private mapping, or directories in
758 * pagecache, etc). Also file data modifications go via their own
759 * mappings. So If we try to mount readonly then copy the filesystem
760 * from bdev, we could get stale data, so invalidate it to give a best
761 * effort at coherency.
762 */
763 if (remount_ro && sb->s_bdev)
764 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 return 0;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100766
767cancel_readonly:
768 sb->s_readonly_remount = 0;
769 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770}
771
Jens Axboea2a95372009-03-17 09:38:40 +0100772static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773{
Al Virodca33252010-07-25 02:31:46 +0400774 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
776 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400777 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500778 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400779 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 sb->s_count++;
781 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400782 down_write(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500783 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
784 !(sb->s_flags & MS_RDONLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 * What lock protects sb->s_flags??
787 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 }
Al Viro443b94b2009-05-05 23:48:50 -0400790 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400792 if (p)
793 __put_super(p);
794 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 }
Al Virodca33252010-07-25 02:31:46 +0400796 if (p)
797 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100799 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 printk("Emergency Remount complete\n");
801}
802
803void emergency_remount(void)
804{
Jens Axboea2a95372009-03-17 09:38:40 +0100805 struct work_struct *work;
806
807 work = kmalloc(sizeof(*work), GFP_ATOMIC);
808 if (work) {
809 INIT_WORK(work, do_emergency_remount);
810 schedule_work(work);
811 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812}
813
814/*
815 * Unnamed block devices are dummy devices used by virtual
816 * filesystems which don't use real block-devices. -- jrs
817 */
818
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400819static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Thomas Bächlera2a4dc42014-04-03 21:55:37 +0200821/* Many userspace utilities consider an FSID of 0 invalid.
822 * Always return at least 1 from get_anon_bdev.
823 */
824static int unnamed_dev_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
Al Viro0ee5dc62011-07-07 15:44:25 -0400826int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827{
828 int dev;
829 int error;
830
831 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400832 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 return -ENOMEM;
834 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400835 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400836 if (!error)
837 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 spin_unlock(&unnamed_dev_lock);
839 if (error == -EAGAIN)
840 /* We raced and lost with another CPU. */
841 goto retry;
842 else if (error)
843 return -EAGAIN;
844
Tejun Heoe8c8d1b2013-02-27 17:05:04 -0800845 if (dev == (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400847 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400848 if (unnamed_dev_start > dev)
849 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 spin_unlock(&unnamed_dev_lock);
851 return -EMFILE;
852 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400853 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 return 0;
855}
Al Viro0ee5dc62011-07-07 15:44:25 -0400856EXPORT_SYMBOL(get_anon_bdev);
857
858void free_anon_bdev(dev_t dev)
859{
860 int slot = MINOR(dev);
861 spin_lock(&unnamed_dev_lock);
862 ida_remove(&unnamed_dev_ida, slot);
863 if (slot < unnamed_dev_start)
864 unnamed_dev_start = slot;
865 spin_unlock(&unnamed_dev_lock);
866}
867EXPORT_SYMBOL(free_anon_bdev);
868
869int set_anon_super(struct super_block *s, void *data)
870{
Christoph Hellwigdf0ce262015-01-14 10:42:41 +0100871 return get_anon_bdev(&s->s_dev);
Al Viro0ee5dc62011-07-07 15:44:25 -0400872}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
874EXPORT_SYMBOL(set_anon_super);
875
876void kill_anon_super(struct super_block *sb)
877{
Al Viro0ee5dc62011-07-07 15:44:25 -0400878 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400880 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881}
882
883EXPORT_SYMBOL(kill_anon_super);
884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885void kill_litter_super(struct super_block *sb)
886{
887 if (sb->s_root)
888 d_genocide(sb->s_root);
889 kill_anon_super(sb);
890}
891
892EXPORT_SYMBOL(kill_litter_super);
893
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700894static int ns_test_super(struct super_block *sb, void *data)
895{
896 return sb->s_fs_info == data;
897}
898
899static int ns_set_super(struct super_block *sb, void *data)
900{
901 sb->s_fs_info = data;
902 return set_anon_super(sb, NULL);
903}
904
Al Viroceefda62010-07-26 13:16:50 +0400905struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
906 void *data, int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700907{
908 struct super_block *sb;
909
David Howells9249e172012-06-25 12:55:37 +0100910 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700911 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +0400912 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700913
914 if (!sb->s_root) {
915 int err;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700916 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
917 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400918 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +0400919 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700920 }
921
922 sb->s_flags |= MS_ACTIVE;
923 }
924
Al Viroceefda62010-07-26 13:16:50 +0400925 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700926}
927
Al Viroceefda62010-07-26 13:16:50 +0400928EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700929
David Howells93614012006-09-30 20:45:40 +0200930#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931static int set_bdev_super(struct super_block *s, void *data)
932{
933 s->s_bdev = data;
934 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +0200935
936 /*
937 * We set the bdi here to the queue backing, file systems can
938 * overwrite this in ->fill_super()
939 */
940 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 return 0;
942}
943
944static int test_bdev_super(struct super_block *s, void *data)
945{
946 return (void *)s->s_bdev == data;
947}
948
Al Viro152a0832010-07-25 00:46:55 +0400949struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +0400951 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952{
953 struct block_device *bdev;
954 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +0100955 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 int error = 0;
957
Al Viro30c40d22008-02-22 19:50:45 -0500958 if (!(flags & MS_RDONLY))
959 mode |= FMODE_WRITE;
960
Tejun Heod4d77622010-11-13 11:55:18 +0100961 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +0400963 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
965 /*
966 * once the super is inserted into the list by sget, s_umount
967 * will protect the lockfs code from trying to start a snapshot
968 * while we are mounting
969 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200970 mutex_lock(&bdev->bd_fsfreeze_mutex);
971 if (bdev->bd_fsfreeze_count > 0) {
972 mutex_unlock(&bdev->bd_fsfreeze_mutex);
973 error = -EBUSY;
974 goto error_bdev;
975 }
David Howells9249e172012-06-25 12:55:37 +0100976 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
977 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200978 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -0700980 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
982 if (s->s_root) {
983 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400984 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -0700985 error = -EBUSY;
986 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 }
David Howells454e2392006-06-23 02:02:57 -0700988
Tejun Heo4f331f02010-07-20 15:18:07 -0700989 /*
990 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +0100991 * __invalidate_device(). blkdev_put() acquires
992 * bd_mutex and can't be called under s_umount. Drop
993 * s_umount temporarily. This is safe as we're
994 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -0700995 */
996 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +0100997 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -0700998 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 } else {
1000 char b[BDEVNAME_SIZE];
1001
Al Viro30c40d22008-02-22 19:50:45 -05001002 s->s_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
Pekka Enberge78c9a02006-01-08 01:03:39 -08001004 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001005 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001007 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001008 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001009 }
David Howells454e2392006-06-23 02:02:57 -07001010
1011 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001012 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 }
1014
Al Viro152a0832010-07-25 00:46:55 +04001015 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
David Howells454e2392006-06-23 02:02:57 -07001017error_s:
1018 error = PTR_ERR(s);
1019error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001020 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001021error:
Al Viro152a0832010-07-25 00:46:55 +04001022 return ERR_PTR(error);
1023}
1024EXPORT_SYMBOL(mount_bdev);
1025
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026void kill_block_super(struct super_block *sb)
1027{
1028 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001029 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001031 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 generic_shutdown_super(sb);
1033 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001034 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001035 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036}
1037
1038EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001039#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
Al Viro3c26ff62010-07-25 11:46:36 +04001041struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001043 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044{
1045 int error;
David Howells9249e172012-06-25 12:55:37 +01001046 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
1048 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001049 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001051 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001053 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001054 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 }
1056 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001057 return dget(s->s_root);
1058}
1059EXPORT_SYMBOL(mount_nodev);
1060
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061static int compare_single(struct super_block *s, void *p)
1062{
1063 return 1;
1064}
1065
Al Virofc14f2f2010-07-25 01:48:30 +04001066struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001068 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069{
1070 struct super_block *s;
1071 int error;
1072
David Howells9249e172012-06-25 12:55:37 +01001073 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001075 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 if (!s->s_root) {
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001077 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001079 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001080 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 }
1082 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001083 } else {
1084 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
Al Virofc14f2f2010-07-25 01:48:30 +04001086 return dget(s->s_root);
1087}
1088EXPORT_SYMBOL(mount_single);
1089
Al Viro9d412a42011-03-17 22:08:28 -04001090struct dentry *
1091mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092{
Al Viroc96e41e2010-07-25 00:17:56 +04001093 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001094 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001096 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001097
Eric Parise0007522008-03-05 10:31:54 -05001098 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001100 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001101 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
Eric Parise0007522008-03-05 10:31:54 -05001103 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001104 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 }
1107
Al Viro1a102ff2011-03-16 09:07:58 -04001108 root = type->mount(type, flags, name, data);
1109 if (IS_ERR(root)) {
1110 error = PTR_ERR(root);
1111 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001112 }
Al Viro9d412a42011-03-17 22:08:28 -04001113 sb = root->d_sb;
1114 BUG_ON(!sb);
1115 WARN_ON(!sb->s_bdi);
1116 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001117
Al Viro9d412a42011-03-17 22:08:28 -04001118 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001119 if (error)
1120 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001121
Jeff Layton42cb56a2009-09-18 13:05:53 -07001122 /*
1123 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1124 * but s_maxbytes was an unsigned long long for many releases. Throw
1125 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001126 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001127 */
Al Viro9d412a42011-03-17 22:08:28 -04001128 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1129 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001130
Al Viro9d412a42011-03-17 22:08:28 -04001131 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001132 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001133 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001135 dput(root);
1136 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137out_free_secdata:
1138 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139out:
David Howells454e2392006-06-23 02:02:57 -07001140 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141}
1142
Jan Kara5accdf82012-06-12 16:20:34 +02001143/*
1144 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1145 * instead.
1146 */
1147void __sb_end_write(struct super_block *sb, int level)
1148{
1149 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1150 /*
1151 * Make sure s_writers are updated before we wake up waiters in
1152 * freeze_super().
1153 */
1154 smp_mb();
1155 if (waitqueue_active(&sb->s_writers.wait))
1156 wake_up(&sb->s_writers.wait);
1157 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1158}
1159EXPORT_SYMBOL(__sb_end_write);
1160
1161#ifdef CONFIG_LOCKDEP
1162/*
1163 * We want lockdep to tell us about possible deadlocks with freezing but
1164 * it's it bit tricky to properly instrument it. Getting a freeze protection
1165 * works as getting a read lock but there are subtle problems. XFS for example
1166 * gets freeze protection on internal level twice in some cases, which is OK
1167 * only because we already hold a freeze protection also on higher level. Due
1168 * to these cases we have to tell lockdep we are doing trylock when we
1169 * already hold a freeze protection for a higher freeze level.
1170 */
1171static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1172 unsigned long ip)
1173{
1174 int i;
1175
1176 if (!trylock) {
1177 for (i = 0; i < level - 1; i++)
1178 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1179 trylock = true;
1180 break;
1181 }
1182 }
1183 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1184}
1185#endif
1186
1187/*
1188 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1189 * instead.
1190 */
1191int __sb_start_write(struct super_block *sb, int level, bool wait)
1192{
1193retry:
1194 if (unlikely(sb->s_writers.frozen >= level)) {
1195 if (!wait)
1196 return 0;
1197 wait_event(sb->s_writers.wait_unfrozen,
1198 sb->s_writers.frozen < level);
1199 }
1200
1201#ifdef CONFIG_LOCKDEP
1202 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1203#endif
1204 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1205 /*
1206 * Make sure counter is updated before we check for frozen.
1207 * freeze_super() first sets frozen and then checks the counter.
1208 */
1209 smp_mb();
1210 if (unlikely(sb->s_writers.frozen >= level)) {
1211 __sb_end_write(sb, level);
1212 goto retry;
1213 }
1214 return 1;
1215}
1216EXPORT_SYMBOL(__sb_start_write);
1217
1218/**
1219 * sb_wait_write - wait until all writers to given file system finish
1220 * @sb: the super for which we wait
1221 * @level: type of writers we wait for (normal vs page fault)
1222 *
1223 * This function waits until there are no writers of given type to given file
1224 * system. Caller of this function should make sure there can be no new writers
1225 * of type @level before calling this function. Otherwise this function can
1226 * livelock.
1227 */
1228static void sb_wait_write(struct super_block *sb, int level)
1229{
1230 s64 writers;
1231
1232 /*
1233 * We just cycle-through lockdep here so that it does not complain
1234 * about returning with lock to userspace
1235 */
1236 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1237 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1238
1239 do {
1240 DEFINE_WAIT(wait);
1241
1242 /*
1243 * We use a barrier in prepare_to_wait() to separate setting
1244 * of frozen and checking of the counter
1245 */
1246 prepare_to_wait(&sb->s_writers.wait, &wait,
1247 TASK_UNINTERRUPTIBLE);
1248
1249 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1250 if (writers)
1251 schedule();
1252
1253 finish_wait(&sb->s_writers.wait, &wait);
1254 } while (writers);
1255}
1256
Josef Bacik18e9e512010-03-23 10:34:56 -04001257/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001258 * freeze_super - lock the filesystem and force it into a consistent state
1259 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001260 *
1261 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1262 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1263 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001264 *
1265 * During this function, sb->s_writers.frozen goes through these values:
1266 *
1267 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1268 *
1269 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1270 * writes should be blocked, though page faults are still allowed. We wait for
1271 * all writes to complete and then proceed to the next stage.
1272 *
1273 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1274 * but internal fs threads can still modify the filesystem (although they
1275 * should not dirty new pages or inodes), writeback can run etc. After waiting
1276 * for all running page faults we sync the filesystem which will clean all
1277 * dirty pages and inodes (no new dirty pages or inodes can be created when
1278 * sync is running).
1279 *
1280 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1281 * modification are blocked (e.g. XFS preallocation truncation on inode
1282 * reclaim). This is usually implemented by blocking new transactions for
1283 * filesystems that have them and need this additional guard. After all
1284 * internal writers are finished we call ->freeze_fs() to finish filesystem
1285 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1286 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1287 *
1288 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001289 */
1290int freeze_super(struct super_block *sb)
1291{
1292 int ret;
1293
1294 atomic_inc(&sb->s_active);
1295 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001296 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001297 deactivate_locked_super(sb);
1298 return -EBUSY;
1299 }
1300
Al Virodabe0dc2012-01-03 21:01:29 -05001301 if (!(sb->s_flags & MS_BORN)) {
1302 up_write(&sb->s_umount);
1303 return 0; /* sic - it's "nothing to do" */
1304 }
1305
Josef Bacik18e9e512010-03-23 10:34:56 -04001306 if (sb->s_flags & MS_RDONLY) {
Jan Kara5accdf82012-06-12 16:20:34 +02001307 /* Nothing to do really... */
1308 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001309 up_write(&sb->s_umount);
1310 return 0;
1311 }
1312
Jan Kara5accdf82012-06-12 16:20:34 +02001313 /* From now on, no new normal writers can start */
1314 sb->s_writers.frozen = SB_FREEZE_WRITE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001315 smp_wmb();
1316
Jan Kara5accdf82012-06-12 16:20:34 +02001317 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1318 up_write(&sb->s_umount);
1319
1320 sb_wait_write(sb, SB_FREEZE_WRITE);
1321
1322 /* Now we go and block page faults... */
1323 down_write(&sb->s_umount);
1324 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1325 smp_wmb();
1326
1327 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1328
1329 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001330 sync_filesystem(sb);
1331
Jan Kara5accdf82012-06-12 16:20:34 +02001332 /* Now wait for internal filesystem counter */
1333 sb->s_writers.frozen = SB_FREEZE_FS;
Josef Bacik18e9e512010-03-23 10:34:56 -04001334 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001335 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001336
Josef Bacik18e9e512010-03-23 10:34:56 -04001337 if (sb->s_op->freeze_fs) {
1338 ret = sb->s_op->freeze_fs(sb);
1339 if (ret) {
1340 printk(KERN_ERR
1341 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001342 sb->s_writers.frozen = SB_UNFROZEN;
Kazuya Mioe1616302011-12-01 16:51:07 +09001343 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001344 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001345 deactivate_locked_super(sb);
1346 return ret;
1347 }
1348 }
Jan Kara5accdf82012-06-12 16:20:34 +02001349 /*
1350 * This is just for debugging purposes so that fs can warn if it
1351 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1352 */
1353 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001354 up_write(&sb->s_umount);
1355 return 0;
1356}
1357EXPORT_SYMBOL(freeze_super);
1358
1359/**
1360 * thaw_super -- unlock filesystem
1361 * @sb: the super to thaw
1362 *
1363 * Unlocks the filesystem and marks it writeable again after freeze_super().
1364 */
1365int thaw_super(struct super_block *sb)
1366{
1367 int error;
1368
1369 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001370 if (sb->s_writers.frozen == SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001371 up_write(&sb->s_umount);
1372 return -EINVAL;
1373 }
1374
1375 if (sb->s_flags & MS_RDONLY)
1376 goto out;
1377
1378 if (sb->s_op->unfreeze_fs) {
1379 error = sb->s_op->unfreeze_fs(sb);
1380 if (error) {
1381 printk(KERN_ERR
1382 "VFS:Filesystem thaw failed\n");
Josef Bacik18e9e512010-03-23 10:34:56 -04001383 up_write(&sb->s_umount);
1384 return error;
1385 }
1386 }
1387
1388out:
Jan Kara5accdf82012-06-12 16:20:34 +02001389 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001390 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001391 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001392 deactivate_locked_super(sb);
1393
1394 return 0;
1395}
1396EXPORT_SYMBOL(thaw_super);