blob: d20d5b11dedf80a7d6c2f39872e6a892b21d7206 [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/acct.h>
26#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/security.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/writeback.h> /* for the emergency remount stuff */
30#include <linux/idr.h>
Ingo Molnar353ab6e2006-03-26 01:37:12 -080031#include <linux/mutex.h>
Jens Axboe5477d0f2010-04-29 20:33:35 +020032#include <linux/backing-dev.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110033#include <linux/rculist_bl.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060034#include <linux/cleancache.h>
Al Viro40401532012-02-13 03:58:52 +000035#include <linux/fsnotify.h>
Jan Kara5accdf82012-06-12 16:20:34 +020036#include <linux/lockdep.h>
Al Viro6d59e7f2008-03-22 15:48:17 -040037#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040LIST_HEAD(super_blocks);
41DEFINE_SPINLOCK(sb_lock);
42
Jan Kara5accdf82012-06-12 16:20:34 +020043static char *sb_writers_name[SB_FREEZE_LEVELS] = {
44 "sb_writers",
45 "sb_pagefaults",
46 "sb_internal",
47};
48
Dave Chinnerb0d40c92011-07-08 14:14:42 +100049/*
50 * One thing we have to be careful of with a per-sb shrinker is that we don't
51 * drop the last active reference to the superblock from within the shrinker.
52 * If that happens we could trigger unregistering the shrinker from within the
53 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
54 * take a passive reference to the superblock to avoid this from occurring.
55 */
Dave Chinner0a234c62013-08-28 10:17:57 +100056static unsigned long super_cache_scan(struct shrinker *shrink,
57 struct shrink_control *sc)
Dave Chinnerb0d40c92011-07-08 14:14:42 +100058{
59 struct super_block *sb;
Dave Chinner0a234c62013-08-28 10:17:57 +100060 long fs_objects = 0;
61 long total_objects;
62 long freed = 0;
63 long dentries;
64 long inodes;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100065
66 sb = container_of(shrink, struct super_block, s_shrink);
67
68 /*
69 * Deadlock avoidance. We may hold various FS locks, and we don't want
70 * to recurse into the FS that called us in clear_inode() and friends..
71 */
Dave Chinner0a234c62013-08-28 10:17:57 +100072 if (!(sc->gfp_mask & __GFP_FS))
73 return SHRINK_STOP;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100074
75 if (!grab_super_passive(sb))
Dave Chinner0a234c62013-08-28 10:17:57 +100076 return SHRINK_STOP;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100077
Al Virod0407902013-07-19 21:12:31 +040078 if (sb->s_op->nr_cached_objects)
Dave Chinner9b17c622013-08-28 10:18:05 +100079 fs_objects = sb->s_op->nr_cached_objects(sb, sc->nid);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100080
Dave Chinner9b17c622013-08-28 10:18:05 +100081 inodes = list_lru_count_node(&sb->s_inode_lru, sc->nid);
82 dentries = list_lru_count_node(&sb->s_dentry_lru, sc->nid);
Dave Chinnerf6041562013-08-28 10:18:00 +100083 total_objects = dentries + inodes + fs_objects + 1;
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100084
Dave Chinner0a234c62013-08-28 10:17:57 +100085 /* proportion the scan between the caches */
Dave Chinnerf6041562013-08-28 10:18:00 +100086 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +100087 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
Dave Chinnerb0d40c92011-07-08 14:14:42 +100088
Dave Chinner0a234c62013-08-28 10:17:57 +100089 /*
90 * prune the dcache first as the icache is pinned by it, then
91 * prune the icache, followed by the filesystem specific caches
92 */
Dave Chinner9b17c622013-08-28 10:18:05 +100093 freed = prune_dcache_sb(sb, dentries, sc->nid);
94 freed += prune_icache_sb(sb, inodes, sc->nid);
Dave Chinnerb0d40c92011-07-08 14:14:42 +100095
Dave Chinner0a234c62013-08-28 10:17:57 +100096 if (fs_objects) {
97 fs_objects = mult_frac(sc->nr_to_scan, fs_objects,
98 total_objects);
Dave Chinner9b17c622013-08-28 10:18:05 +100099 freed += sb->s_op->free_cached_objects(sb, fs_objects,
100 sc->nid);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000101 }
102
Dave Chinner0a234c62013-08-28 10:17:57 +1000103 drop_super(sb);
104 return freed;
105}
106
107static unsigned long super_cache_count(struct shrinker *shrink,
108 struct shrink_control *sc)
109{
110 struct super_block *sb;
111 long total_objects = 0;
112
113 sb = container_of(shrink, struct super_block, s_shrink);
114
Tim Chend23da152014-06-04 16:10:47 -0700115 /*
116 * Don't call grab_super_passive as it is a potential
117 * scalability bottleneck. The counts could get updated
118 * between super_cache_count and super_cache_scan anyway.
119 * Call to super_cache_count with shrinker_rwsem held
120 * ensures the safety of call to list_lru_count_node() and
121 * s_op->nr_cached_objects().
122 */
Dave Chinner0a234c62013-08-28 10:17:57 +1000123 if (sb->s_op && sb->s_op->nr_cached_objects)
Dave Chinner9b17c622013-08-28 10:18:05 +1000124 total_objects = sb->s_op->nr_cached_objects(sb,
125 sc->nid);
Dave Chinner0a234c62013-08-28 10:17:57 +1000126
Dave Chinner9b17c622013-08-28 10:18:05 +1000127 total_objects += list_lru_count_node(&sb->s_dentry_lru,
128 sc->nid);
129 total_objects += list_lru_count_node(&sb->s_inode_lru,
130 sc->nid);
Dave Chinner0a234c62013-08-28 10:17:57 +1000131
Glauber Costa55f841c2013-08-28 10:17:53 +1000132 total_objects = vfs_pressure_ratio(total_objects);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000133 return total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000134}
135
Al Viro7eb5e882013-10-01 15:09:58 -0400136/**
137 * destroy_super - frees a superblock
138 * @s: superblock to free
139 *
140 * Frees a superblock.
141 */
142static void destroy_super(struct super_block *s)
Jan Kara5accdf82012-06-12 16:20:34 +0200143{
144 int i;
Al Viro7eb5e882013-10-01 15:09:58 -0400145 list_lru_destroy(&s->s_dentry_lru);
146 list_lru_destroy(&s->s_inode_lru);
Jan Kara5accdf82012-06-12 16:20:34 +0200147 for (i = 0; i < SB_FREEZE_LEVELS; i++)
148 percpu_counter_destroy(&s->s_writers.counter[i]);
Al Viro7eb5e882013-10-01 15:09:58 -0400149 security_sb_free(s);
150 WARN_ON(!list_empty(&s->s_mounts));
151 kfree(s->s_subtype);
152 kfree(s->s_options);
Al Viroe2fec7c2013-10-04 17:06:56 -0400153 kfree_rcu(s, rcu);
Jan Kara5accdf82012-06-12 16:20:34 +0200154}
155
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156/**
157 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700158 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100159 * @flags: the mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 *
161 * Allocates and initializes a new &struct super_block. alloc_super()
162 * returns a pointer new superblock or %NULL if allocation had failed.
163 */
David Howells9249e172012-06-25 12:55:37 +0100164static struct super_block *alloc_super(struct file_system_type *type, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800166 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700167 static const struct super_operations default_op;
Al Viro7eb5e882013-10-01 15:09:58 -0400168 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
Al Viro7eb5e882013-10-01 15:09:58 -0400170 if (!s)
171 return NULL;
172
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800173 INIT_LIST_HEAD(&s->s_mounts);
174
Al Viro7eb5e882013-10-01 15:09:58 -0400175 if (security_sb_alloc(s))
176 goto fail;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200177
Al Viro7eb5e882013-10-01 15:09:58 -0400178 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
179 if (percpu_counter_init(&s->s_writers.counter[i], 0) < 0)
180 goto fail;
181 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
182 &type->s_writers_key[i], 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 }
Al Viro7eb5e882013-10-01 15:09:58 -0400184 init_waitqueue_head(&s->s_writers.wait);
185 init_waitqueue_head(&s->s_writers.wait_unfrozen);
186 s->s_flags = flags;
187 s->s_bdi = &default_backing_dev_info;
188 INIT_HLIST_NODE(&s->s_instances);
189 INIT_HLIST_BL_HEAD(&s->s_anon);
190 INIT_LIST_HEAD(&s->s_inodes);
191
192 if (list_lru_init(&s->s_dentry_lru))
193 goto fail;
194 if (list_lru_init(&s->s_inode_lru))
195 goto fail;
196
Al Viro7eb5e882013-10-01 15:09:58 -0400197 init_rwsem(&s->s_umount);
198 lockdep_set_class(&s->s_umount, &type->s_umount_key);
199 /*
200 * sget() can have s_umount recursion.
201 *
202 * When it cannot find a suitable sb, it allocates a new
203 * one (this one), and tries again to find a suitable old
204 * one.
205 *
206 * In case that succeeds, it will acquire the s_umount
207 * lock of the old one. Since these are clearly distrinct
208 * locks, and this object isn't exposed yet, there's no
209 * risk of deadlocks.
210 *
211 * Annotate this by putting this lock in a different
212 * subclass.
213 */
214 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
215 s->s_count = 1;
216 atomic_set(&s->s_active, 1);
217 mutex_init(&s->s_vfs_rename_mutex);
218 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
219 mutex_init(&s->s_dquot.dqio_mutex);
220 mutex_init(&s->s_dquot.dqonoff_mutex);
221 init_rwsem(&s->s_dquot.dqptr_sem);
222 s->s_maxbytes = MAX_NON_LFS;
223 s->s_op = &default_op;
224 s->s_time_gran = 1000000000;
225 s->cleancache_poolid = -1;
226
227 s->s_shrink.seeks = DEFAULT_SEEKS;
228 s->s_shrink.scan_objects = super_cache_scan;
229 s->s_shrink.count_objects = super_cache_count;
230 s->s_shrink.batch = 1024;
231 s->s_shrink.flags = SHRINKER_NUMA_AWARE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000233
Al Viro7eb5e882013-10-01 15:09:58 -0400234fail:
235 destroy_super(s);
236 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237}
238
239/* Superblock refcounting */
240
241/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400242 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 */
Al Virof47ec3f2011-11-21 21:15:42 -0500244static void __put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 if (!--sb->s_count) {
Al Viro551de6f2010-03-22 19:36:35 -0400247 list_del_init(&sb->s_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 destroy_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250}
251
252/**
253 * put_super - drop a temporary reference to superblock
254 * @sb: superblock in question
255 *
256 * Drops a temporary reference, frees superblock if there's no
257 * references left.
258 */
Al Virof47ec3f2011-11-21 21:15:42 -0500259static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
261 spin_lock(&sb_lock);
262 __put_super(sb);
263 spin_unlock(&sb_lock);
264}
265
266
267/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400268 * deactivate_locked_super - drop an active reference to superblock
269 * @s: superblock to deactivate
270 *
Al Viro1712ac82010-03-22 15:22:31 -0400271 * Drops an active reference to superblock, converting it into a temprory
272 * one if there is no other active references left. In that case we
273 * tell fs driver to shut it down and drop the temporary reference we
274 * had just acquired.
275 *
276 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400277 */
278void deactivate_locked_super(struct super_block *s)
279{
280 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400281 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400282 cleancache_invalidate_fs(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000283 unregister_shrinker(&s->s_shrink);
Dave Chinner28f2cd4f2014-06-04 16:10:46 -0700284 fs->kill_sb(s);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000285
Al Viro74dbbdd2009-05-06 01:07:50 -0400286 put_filesystem(fs);
287 put_super(s);
288 } else {
289 up_write(&s->s_umount);
290 }
291}
292
293EXPORT_SYMBOL(deactivate_locked_super);
294
295/**
Al Viro1712ac82010-03-22 15:22:31 -0400296 * deactivate_super - drop an active reference to superblock
297 * @s: superblock to deactivate
298 *
299 * Variant of deactivate_locked_super(), except that superblock is *not*
300 * locked by caller. If we are going to drop the final active reference,
301 * lock will be acquired prior to that.
302 */
303void deactivate_super(struct super_block *s)
304{
305 if (!atomic_add_unless(&s->s_active, -1, 1)) {
306 down_write(&s->s_umount);
307 deactivate_locked_super(s);
308 }
309}
310
311EXPORT_SYMBOL(deactivate_super);
312
313/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 * grab_super - acquire an active reference
315 * @s: reference we are trying to make active
316 *
317 * Tries to acquire an active reference. grab_super() is used when we
318 * had just found a superblock in super_blocks or fs_type->fs_supers
319 * and want to turn it into a full-blown active reference. grab_super()
320 * is called with sb_lock held and drops it. Returns 1 in case of
321 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400322 * dying when grab_super() had been called). Note that this is only
323 * called for superblocks not in rundown mode (== ones still on ->fs_supers
324 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700326static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327{
328 s->s_count++;
329 spin_unlock(&sb_lock);
330 down_write(&s->s_umount);
Al Viroacfec9a2013-07-20 03:13:55 +0400331 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
332 put_super(s);
333 return 1;
334 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 up_write(&s->s_umount);
336 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 return 0;
338}
339
David Howellscf9a2ae2006-08-29 19:05:54 +0100340/*
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000341 * grab_super_passive - acquire a passive reference
Wanpeng Li331cbde2012-06-09 11:10:55 +0800342 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000343 *
344 * Tries to acquire a passive reference. This is used in places where we
345 * cannot take an active reference but we need to ensure that the
346 * superblock does not go away while we are working on it. It returns
347 * false if a reference was not gained, and returns true with the s_umount
348 * lock held in read mode if a reference is gained. On successful return,
349 * the caller must drop the s_umount lock and the passive reference when
350 * done.
351 */
352bool grab_super_passive(struct super_block *sb)
353{
354 spin_lock(&sb_lock);
Al Viroa5166162011-12-12 22:53:00 -0500355 if (hlist_unhashed(&sb->s_instances)) {
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000356 spin_unlock(&sb_lock);
357 return false;
358 }
359
360 sb->s_count++;
361 spin_unlock(&sb_lock);
362
363 if (down_read_trylock(&sb->s_umount)) {
Al Virodabe0dc2012-01-03 21:01:29 -0500364 if (sb->s_root && (sb->s_flags & MS_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000365 return true;
366 up_read(&sb->s_umount);
367 }
368
369 put_super(sb);
370 return false;
371}
372
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373/**
374 * generic_shutdown_super - common helper for ->kill_sb()
375 * @sb: superblock to kill
376 *
377 * generic_shutdown_super() does all fs-independent work on superblock
378 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
379 * that need destruction out of superblock, call generic_shutdown_super()
380 * and release aforementioned objects. Note: dentries and inodes _are_
381 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700382 *
383 * Upon calling this function, the filesystem may no longer alter or
384 * rearrange the set of dentries belonging to this super_block, nor may it
385 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 */
387void generic_shutdown_super(struct super_block *sb)
388{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800389 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
David Howellsc636ebd2006-10-11 01:22:19 -0700391 if (sb->s_root) {
392 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200393 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 sb->s_flags &= ~MS_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800395
Al Viro63997e92010-10-25 20:49:35 -0400396 fsnotify_unmount_inodes(&sb->s_inodes);
397
398 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200400 if (sb->s_dio_done_wq) {
401 destroy_workqueue(sb->s_dio_done_wq);
402 sb->s_dio_done_wq = NULL;
403 }
404
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 if (sop->put_super)
406 sop->put_super(sb);
407
Al Viro63997e92010-10-25 20:49:35 -0400408 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800409 printk("VFS: Busy inodes after unmount of %s. "
410 "Self-destruct in 5 seconds. Have a nice day...\n",
411 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 }
414 spin_lock(&sb_lock);
415 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500416 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 spin_unlock(&sb_lock);
418 up_write(&sb->s_umount);
419}
420
421EXPORT_SYMBOL(generic_shutdown_super);
422
423/**
424 * sget - find or create a superblock
425 * @type: filesystem type superblock should belong to
426 * @test: comparison callback
427 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100428 * @flags: mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 * @data: argument to each of them
430 */
431struct super_block *sget(struct file_system_type *type,
432 int (*test)(struct super_block *,void *),
433 int (*set)(struct super_block *,void *),
David Howells9249e172012-06-25 12:55:37 +0100434 int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 void *data)
436{
437 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700438 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 int err;
440
441retry:
442 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700443 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800444 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700445 if (!test(old, data))
446 continue;
447 if (!grab_super(old))
448 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700449 if (s) {
450 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700451 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400452 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700453 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700454 return old;
455 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 }
457 if (!s) {
458 spin_unlock(&sb_lock);
David Howells9249e172012-06-25 12:55:37 +0100459 s = alloc_super(type, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 if (!s)
461 return ERR_PTR(-ENOMEM);
462 goto retry;
463 }
464
465 err = set(s, data);
466 if (err) {
467 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700468 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 destroy_super(s);
470 return ERR_PTR(err);
471 }
472 s->s_type = type;
473 strlcpy(s->s_id, type->name, sizeof(s->s_id));
474 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500475 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 spin_unlock(&sb_lock);
477 get_filesystem(type);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000478 register_shrinker(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 return s;
480}
481
482EXPORT_SYMBOL(sget);
483
484void drop_super(struct super_block *sb)
485{
486 up_read(&sb->s_umount);
487 put_super(sb);
488}
489
490EXPORT_SYMBOL(drop_super);
491
Christoph Hellwige5004752009-05-05 16:08:56 +0200492/**
Al Viro01a05b32010-03-23 06:06:58 -0400493 * iterate_supers - call function for all active superblocks
494 * @f: function to call
495 * @arg: argument to pass to it
496 *
497 * Scans the superblock list and calls given function, passing it
498 * locked superblock and given argument.
499 */
500void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
501{
Al Virodca33252010-07-25 02:31:46 +0400502 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400503
504 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400505 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500506 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400507 continue;
508 sb->s_count++;
509 spin_unlock(&sb_lock);
510
511 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500512 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400513 f(sb, arg);
514 up_read(&sb->s_umount);
515
516 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400517 if (p)
518 __put_super(p);
519 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400520 }
Al Virodca33252010-07-25 02:31:46 +0400521 if (p)
522 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400523 spin_unlock(&sb_lock);
524}
525
526/**
Al Viro43e15cd2011-06-03 20:16:57 -0400527 * iterate_supers_type - call function for superblocks of given type
528 * @type: fs type
529 * @f: function to call
530 * @arg: argument to pass to it
531 *
532 * Scans the superblock list and calls given function, passing it
533 * locked superblock and given argument.
534 */
535void iterate_supers_type(struct file_system_type *type,
536 void (*f)(struct super_block *, void *), void *arg)
537{
538 struct super_block *sb, *p = NULL;
539
540 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800541 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400542 sb->s_count++;
543 spin_unlock(&sb_lock);
544
545 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500546 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400547 f(sb, arg);
548 up_read(&sb->s_umount);
549
550 spin_lock(&sb_lock);
551 if (p)
552 __put_super(p);
553 p = sb;
554 }
555 if (p)
556 __put_super(p);
557 spin_unlock(&sb_lock);
558}
559
560EXPORT_SYMBOL(iterate_supers_type);
561
562/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 * get_super - get the superblock of a device
564 * @bdev: device to get the superblock for
565 *
566 * Scans the superblock list and finds the superblock of the file system
567 * mounted on the device given. %NULL is returned if no match is found.
568 */
569
Al Virodf40c012010-03-22 20:23:25 -0400570struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700572 struct super_block *sb;
573
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 if (!bdev)
575 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700576
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700578rescan:
579 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500580 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400581 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700582 if (sb->s_bdev == bdev) {
583 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700585 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400586 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500587 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700588 return sb;
589 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400590 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700591 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400592 __put_super(sb);
593 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
595 }
596 spin_unlock(&sb_lock);
597 return NULL;
598}
599
600EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200601
602/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100603 * get_super_thawed - get thawed superblock of a device
604 * @bdev: device to get the superblock for
605 *
606 * Scans the superblock list and finds the superblock of the file system
607 * mounted on the device. The superblock is returned once it is thawed
608 * (or immediately if it was not frozen). %NULL is returned if no match
609 * is found.
610 */
611struct super_block *get_super_thawed(struct block_device *bdev)
612{
613 while (1) {
614 struct super_block *s = get_super(bdev);
Jan Kara5accdf82012-06-12 16:20:34 +0200615 if (!s || s->s_writers.frozen == SB_UNFROZEN)
Jan Kara6b6dc832012-02-10 11:03:00 +0100616 return s;
617 up_read(&s->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +0200618 wait_event(s->s_writers.wait_unfrozen,
619 s->s_writers.frozen == SB_UNFROZEN);
Jan Kara6b6dc832012-02-10 11:03:00 +0100620 put_super(s);
621 }
622}
623EXPORT_SYMBOL(get_super_thawed);
624
625/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200626 * get_active_super - get an active reference to the superblock of a device
627 * @bdev: device to get the superblock for
628 *
629 * Scans the superblock list and finds the superblock of the file system
630 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400631 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200632 */
633struct super_block *get_active_super(struct block_device *bdev)
634{
635 struct super_block *sb;
636
637 if (!bdev)
638 return NULL;
639
Al Viro14945832010-03-22 20:15:33 -0400640restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200641 spin_lock(&sb_lock);
642 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500643 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400644 continue;
Al Viro14945832010-03-22 20:15:33 -0400645 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400646 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400647 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400648 up_write(&sb->s_umount);
649 return sb;
Al Viro14945832010-03-22 20:15:33 -0400650 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200651 }
652 spin_unlock(&sb_lock);
653 return NULL;
654}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Al Virodf40c012010-03-22 20:23:25 -0400656struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700658 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700661rescan:
662 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500663 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400664 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700665 if (sb->s_dev == dev) {
666 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700668 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400669 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500670 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700671 return sb;
672 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400673 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700674 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400675 __put_super(sb);
676 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 }
678 }
679 spin_unlock(&sb_lock);
680 return NULL;
681}
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 * do_remount_sb - asks filesystem to change mount options.
685 * @sb: superblock in question
686 * @flags: numeric part of options
687 * @data: the rest of options
688 * @force: whether or not to force the change
689 *
690 * Alters the mount options of a mounted file system.
691 */
692int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
693{
694 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400695 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200696
Jan Kara5accdf82012-06-12 16:20:34 +0200697 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200698 return -EBUSY;
699
David Howells93614012006-09-30 20:45:40 +0200700#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
702 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200703#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200704
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 if (flags & MS_RDONLY)
706 acct_auto_close(sb);
707 shrink_dcache_sb(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
Nick Piggind208bbd2009-12-21 16:28:53 -0800709 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800710
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 /* If we are remounting RDONLY and current sb is read/write,
712 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800713 if (remount_ro) {
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100714 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400715 sb->s_readonly_remount = 1;
716 smp_wmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100717 } else {
718 retval = sb_prepare_remount_readonly(sb);
719 if (retval)
720 return retval;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100721 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 }
723
724 if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 retval = sb->s_op->remount_fs(sb, &flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200726 if (retval) {
727 if (!force)
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100728 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200729 /* If forced remount, go ahead despite any errors */
730 WARN(1, "forced remount of a %s fs returned %i\n",
731 sb->s_type->name, retval);
732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 }
734 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100735 /* Needs to be ordered wrt mnt_is_readonly() */
736 smp_wmb();
737 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400738
Nick Piggind208bbd2009-12-21 16:28:53 -0800739 /*
740 * Some filesystems modify their metadata via some other path than the
741 * bdev buffer cache (eg. use a private mapping, or directories in
742 * pagecache, etc). Also file data modifications go via their own
743 * mappings. So If we try to mount readonly then copy the filesystem
744 * from bdev, we could get stale data, so invalidate it to give a best
745 * effort at coherency.
746 */
747 if (remount_ro && sb->s_bdev)
748 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 return 0;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100750
751cancel_readonly:
752 sb->s_readonly_remount = 0;
753 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754}
755
Jens Axboea2a95372009-03-17 09:38:40 +0100756static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757{
Al Virodca33252010-07-25 02:31:46 +0400758 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
760 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400761 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500762 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400763 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 sb->s_count++;
765 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400766 down_write(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500767 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
768 !(sb->s_flags & MS_RDONLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 * What lock protects sb->s_flags??
771 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 }
Al Viro443b94b2009-05-05 23:48:50 -0400774 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400776 if (p)
777 __put_super(p);
778 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 }
Al Virodca33252010-07-25 02:31:46 +0400780 if (p)
781 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100783 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 printk("Emergency Remount complete\n");
785}
786
787void emergency_remount(void)
788{
Jens Axboea2a95372009-03-17 09:38:40 +0100789 struct work_struct *work;
790
791 work = kmalloc(sizeof(*work), GFP_ATOMIC);
792 if (work) {
793 INIT_WORK(work, do_emergency_remount);
794 schedule_work(work);
795 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796}
797
798/*
799 * Unnamed block devices are dummy devices used by virtual
800 * filesystems which don't use real block-devices. -- jrs
801 */
802
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400803static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Thomas Bächlera2a4dc42014-04-03 21:55:37 +0200805/* Many userspace utilities consider an FSID of 0 invalid.
806 * Always return at least 1 from get_anon_bdev.
807 */
808static int unnamed_dev_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
Al Viro0ee5dc62011-07-07 15:44:25 -0400810int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811{
812 int dev;
813 int error;
814
815 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400816 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 return -ENOMEM;
818 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400819 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400820 if (!error)
821 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 spin_unlock(&unnamed_dev_lock);
823 if (error == -EAGAIN)
824 /* We raced and lost with another CPU. */
825 goto retry;
826 else if (error)
827 return -EAGAIN;
828
Tejun Heoe8c8d1b2013-02-27 17:05:04 -0800829 if (dev == (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400831 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400832 if (unnamed_dev_start > dev)
833 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 spin_unlock(&unnamed_dev_lock);
835 return -EMFILE;
836 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400837 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 return 0;
839}
Al Viro0ee5dc62011-07-07 15:44:25 -0400840EXPORT_SYMBOL(get_anon_bdev);
841
842void free_anon_bdev(dev_t dev)
843{
844 int slot = MINOR(dev);
845 spin_lock(&unnamed_dev_lock);
846 ida_remove(&unnamed_dev_ida, slot);
847 if (slot < unnamed_dev_start)
848 unnamed_dev_start = slot;
849 spin_unlock(&unnamed_dev_lock);
850}
851EXPORT_SYMBOL(free_anon_bdev);
852
853int set_anon_super(struct super_block *s, void *data)
854{
855 int error = get_anon_bdev(&s->s_dev);
856 if (!error)
857 s->s_bdi = &noop_backing_dev_info;
858 return error;
859}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
861EXPORT_SYMBOL(set_anon_super);
862
863void kill_anon_super(struct super_block *sb)
864{
Al Viro0ee5dc62011-07-07 15:44:25 -0400865 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400867 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868}
869
870EXPORT_SYMBOL(kill_anon_super);
871
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872void kill_litter_super(struct super_block *sb)
873{
874 if (sb->s_root)
875 d_genocide(sb->s_root);
876 kill_anon_super(sb);
877}
878
879EXPORT_SYMBOL(kill_litter_super);
880
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700881static int ns_test_super(struct super_block *sb, void *data)
882{
883 return sb->s_fs_info == data;
884}
885
886static int ns_set_super(struct super_block *sb, void *data)
887{
888 sb->s_fs_info = data;
889 return set_anon_super(sb, NULL);
890}
891
Al Viroceefda62010-07-26 13:16:50 +0400892struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
893 void *data, int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700894{
895 struct super_block *sb;
896
David Howells9249e172012-06-25 12:55:37 +0100897 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700898 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +0400899 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700900
901 if (!sb->s_root) {
902 int err;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700903 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
904 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400905 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +0400906 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700907 }
908
909 sb->s_flags |= MS_ACTIVE;
910 }
911
Al Viroceefda62010-07-26 13:16:50 +0400912 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700913}
914
Al Viroceefda62010-07-26 13:16:50 +0400915EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700916
David Howells93614012006-09-30 20:45:40 +0200917#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918static int set_bdev_super(struct super_block *s, void *data)
919{
920 s->s_bdev = data;
921 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +0200922
923 /*
924 * We set the bdi here to the queue backing, file systems can
925 * overwrite this in ->fill_super()
926 */
927 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 return 0;
929}
930
931static int test_bdev_super(struct super_block *s, void *data)
932{
933 return (void *)s->s_bdev == data;
934}
935
Al Viro152a0832010-07-25 00:46:55 +0400936struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +0400938 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939{
940 struct block_device *bdev;
941 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +0100942 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 int error = 0;
944
Al Viro30c40d22008-02-22 19:50:45 -0500945 if (!(flags & MS_RDONLY))
946 mode |= FMODE_WRITE;
947
Tejun Heod4d77622010-11-13 11:55:18 +0100948 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +0400950 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
952 /*
953 * once the super is inserted into the list by sget, s_umount
954 * will protect the lockfs code from trying to start a snapshot
955 * while we are mounting
956 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200957 mutex_lock(&bdev->bd_fsfreeze_mutex);
958 if (bdev->bd_fsfreeze_count > 0) {
959 mutex_unlock(&bdev->bd_fsfreeze_mutex);
960 error = -EBUSY;
961 goto error_bdev;
962 }
David Howells9249e172012-06-25 12:55:37 +0100963 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
964 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200965 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -0700967 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
969 if (s->s_root) {
970 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400971 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -0700972 error = -EBUSY;
973 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 }
David Howells454e2392006-06-23 02:02:57 -0700975
Tejun Heo4f331f02010-07-20 15:18:07 -0700976 /*
977 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +0100978 * __invalidate_device(). blkdev_put() acquires
979 * bd_mutex and can't be called under s_umount. Drop
980 * s_umount temporarily. This is safe as we're
981 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -0700982 */
983 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +0100984 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -0700985 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 } else {
987 char b[BDEVNAME_SIZE];
988
Al Viro30c40d22008-02-22 19:50:45 -0500989 s->s_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
Pekka Enberge78c9a02006-01-08 01:03:39 -0800991 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -0800992 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400994 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -0700995 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -0800996 }
David Howells454e2392006-06-23 02:02:57 -0700997
998 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -0500999 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 }
1001
Al Viro152a0832010-07-25 00:46:55 +04001002 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
David Howells454e2392006-06-23 02:02:57 -07001004error_s:
1005 error = PTR_ERR(s);
1006error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001007 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001008error:
Al Viro152a0832010-07-25 00:46:55 +04001009 return ERR_PTR(error);
1010}
1011EXPORT_SYMBOL(mount_bdev);
1012
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013void kill_block_super(struct super_block *sb)
1014{
1015 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001016 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001018 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 generic_shutdown_super(sb);
1020 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001021 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001022 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023}
1024
1025EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001026#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
Al Viro3c26ff62010-07-25 11:46:36 +04001028struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001030 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031{
1032 int error;
David Howells9249e172012-06-25 12:55:37 +01001033 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
1035 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001036 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001038 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001040 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001041 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 }
1043 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001044 return dget(s->s_root);
1045}
1046EXPORT_SYMBOL(mount_nodev);
1047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048static int compare_single(struct super_block *s, void *p)
1049{
1050 return 1;
1051}
1052
Al Virofc14f2f2010-07-25 01:48:30 +04001053struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001055 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056{
1057 struct super_block *s;
1058 int error;
1059
David Howells9249e172012-06-25 12:55:37 +01001060 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001062 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 if (!s->s_root) {
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001064 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001066 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001067 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 }
1069 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001070 } else {
1071 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
Al Virofc14f2f2010-07-25 01:48:30 +04001073 return dget(s->s_root);
1074}
1075EXPORT_SYMBOL(mount_single);
1076
Al Viro9d412a42011-03-17 22:08:28 -04001077struct dentry *
1078mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079{
Al Viroc96e41e2010-07-25 00:17:56 +04001080 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001081 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001083 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001084
Eric Parise0007522008-03-05 10:31:54 -05001085 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001087 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001088 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
Eric Parise0007522008-03-05 10:31:54 -05001090 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001091 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 }
1094
Al Viro1a102ff2011-03-16 09:07:58 -04001095 root = type->mount(type, flags, name, data);
1096 if (IS_ERR(root)) {
1097 error = PTR_ERR(root);
1098 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001099 }
Al Viro9d412a42011-03-17 22:08:28 -04001100 sb = root->d_sb;
1101 BUG_ON(!sb);
1102 WARN_ON(!sb->s_bdi);
Linus Torvalds6c510382011-03-24 10:16:26 -07001103 WARN_ON(sb->s_bdi == &default_backing_dev_info);
Al Viro9d412a42011-03-17 22:08:28 -04001104 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001105
Al Viro9d412a42011-03-17 22:08:28 -04001106 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001107 if (error)
1108 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001109
Jeff Layton42cb56a2009-09-18 13:05:53 -07001110 /*
1111 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1112 * but s_maxbytes was an unsigned long long for many releases. Throw
1113 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001114 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001115 */
Al Viro9d412a42011-03-17 22:08:28 -04001116 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1117 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001118
Al Viro9d412a42011-03-17 22:08:28 -04001119 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001120 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001121 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001123 dput(root);
1124 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125out_free_secdata:
1126 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127out:
David Howells454e2392006-06-23 02:02:57 -07001128 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129}
1130
Jan Kara5accdf82012-06-12 16:20:34 +02001131/*
1132 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1133 * instead.
1134 */
1135void __sb_end_write(struct super_block *sb, int level)
1136{
1137 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1138 /*
1139 * Make sure s_writers are updated before we wake up waiters in
1140 * freeze_super().
1141 */
1142 smp_mb();
1143 if (waitqueue_active(&sb->s_writers.wait))
1144 wake_up(&sb->s_writers.wait);
1145 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1146}
1147EXPORT_SYMBOL(__sb_end_write);
1148
1149#ifdef CONFIG_LOCKDEP
1150/*
1151 * We want lockdep to tell us about possible deadlocks with freezing but
1152 * it's it bit tricky to properly instrument it. Getting a freeze protection
1153 * works as getting a read lock but there are subtle problems. XFS for example
1154 * gets freeze protection on internal level twice in some cases, which is OK
1155 * only because we already hold a freeze protection also on higher level. Due
1156 * to these cases we have to tell lockdep we are doing trylock when we
1157 * already hold a freeze protection for a higher freeze level.
1158 */
1159static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1160 unsigned long ip)
1161{
1162 int i;
1163
1164 if (!trylock) {
1165 for (i = 0; i < level - 1; i++)
1166 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1167 trylock = true;
1168 break;
1169 }
1170 }
1171 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1172}
1173#endif
1174
1175/*
1176 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1177 * instead.
1178 */
1179int __sb_start_write(struct super_block *sb, int level, bool wait)
1180{
1181retry:
1182 if (unlikely(sb->s_writers.frozen >= level)) {
1183 if (!wait)
1184 return 0;
1185 wait_event(sb->s_writers.wait_unfrozen,
1186 sb->s_writers.frozen < level);
1187 }
1188
1189#ifdef CONFIG_LOCKDEP
1190 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1191#endif
1192 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1193 /*
1194 * Make sure counter is updated before we check for frozen.
1195 * freeze_super() first sets frozen and then checks the counter.
1196 */
1197 smp_mb();
1198 if (unlikely(sb->s_writers.frozen >= level)) {
1199 __sb_end_write(sb, level);
1200 goto retry;
1201 }
1202 return 1;
1203}
1204EXPORT_SYMBOL(__sb_start_write);
1205
1206/**
1207 * sb_wait_write - wait until all writers to given file system finish
1208 * @sb: the super for which we wait
1209 * @level: type of writers we wait for (normal vs page fault)
1210 *
1211 * This function waits until there are no writers of given type to given file
1212 * system. Caller of this function should make sure there can be no new writers
1213 * of type @level before calling this function. Otherwise this function can
1214 * livelock.
1215 */
1216static void sb_wait_write(struct super_block *sb, int level)
1217{
1218 s64 writers;
1219
1220 /*
1221 * We just cycle-through lockdep here so that it does not complain
1222 * about returning with lock to userspace
1223 */
1224 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1225 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1226
1227 do {
1228 DEFINE_WAIT(wait);
1229
1230 /*
1231 * We use a barrier in prepare_to_wait() to separate setting
1232 * of frozen and checking of the counter
1233 */
1234 prepare_to_wait(&sb->s_writers.wait, &wait,
1235 TASK_UNINTERRUPTIBLE);
1236
1237 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1238 if (writers)
1239 schedule();
1240
1241 finish_wait(&sb->s_writers.wait, &wait);
1242 } while (writers);
1243}
1244
Josef Bacik18e9e512010-03-23 10:34:56 -04001245/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001246 * freeze_super - lock the filesystem and force it into a consistent state
1247 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001248 *
1249 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1250 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1251 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001252 *
1253 * During this function, sb->s_writers.frozen goes through these values:
1254 *
1255 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1256 *
1257 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1258 * writes should be blocked, though page faults are still allowed. We wait for
1259 * all writes to complete and then proceed to the next stage.
1260 *
1261 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1262 * but internal fs threads can still modify the filesystem (although they
1263 * should not dirty new pages or inodes), writeback can run etc. After waiting
1264 * for all running page faults we sync the filesystem which will clean all
1265 * dirty pages and inodes (no new dirty pages or inodes can be created when
1266 * sync is running).
1267 *
1268 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1269 * modification are blocked (e.g. XFS preallocation truncation on inode
1270 * reclaim). This is usually implemented by blocking new transactions for
1271 * filesystems that have them and need this additional guard. After all
1272 * internal writers are finished we call ->freeze_fs() to finish filesystem
1273 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1274 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1275 *
1276 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001277 */
1278int freeze_super(struct super_block *sb)
1279{
1280 int ret;
1281
1282 atomic_inc(&sb->s_active);
1283 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001284 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001285 deactivate_locked_super(sb);
1286 return -EBUSY;
1287 }
1288
Al Virodabe0dc2012-01-03 21:01:29 -05001289 if (!(sb->s_flags & MS_BORN)) {
1290 up_write(&sb->s_umount);
1291 return 0; /* sic - it's "nothing to do" */
1292 }
1293
Josef Bacik18e9e512010-03-23 10:34:56 -04001294 if (sb->s_flags & MS_RDONLY) {
Jan Kara5accdf82012-06-12 16:20:34 +02001295 /* Nothing to do really... */
1296 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001297 up_write(&sb->s_umount);
1298 return 0;
1299 }
1300
Jan Kara5accdf82012-06-12 16:20:34 +02001301 /* From now on, no new normal writers can start */
1302 sb->s_writers.frozen = SB_FREEZE_WRITE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001303 smp_wmb();
1304
Jan Kara5accdf82012-06-12 16:20:34 +02001305 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1306 up_write(&sb->s_umount);
1307
1308 sb_wait_write(sb, SB_FREEZE_WRITE);
1309
1310 /* Now we go and block page faults... */
1311 down_write(&sb->s_umount);
1312 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1313 smp_wmb();
1314
1315 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1316
1317 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001318 sync_filesystem(sb);
1319
Jan Kara5accdf82012-06-12 16:20:34 +02001320 /* Now wait for internal filesystem counter */
1321 sb->s_writers.frozen = SB_FREEZE_FS;
Josef Bacik18e9e512010-03-23 10:34:56 -04001322 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001323 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001324
Josef Bacik18e9e512010-03-23 10:34:56 -04001325 if (sb->s_op->freeze_fs) {
1326 ret = sb->s_op->freeze_fs(sb);
1327 if (ret) {
1328 printk(KERN_ERR
1329 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001330 sb->s_writers.frozen = SB_UNFROZEN;
Kazuya Mioe1616302011-12-01 16:51:07 +09001331 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001332 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001333 deactivate_locked_super(sb);
1334 return ret;
1335 }
1336 }
Jan Kara5accdf82012-06-12 16:20:34 +02001337 /*
1338 * This is just for debugging purposes so that fs can warn if it
1339 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1340 */
1341 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001342 up_write(&sb->s_umount);
1343 return 0;
1344}
1345EXPORT_SYMBOL(freeze_super);
1346
1347/**
1348 * thaw_super -- unlock filesystem
1349 * @sb: the super to thaw
1350 *
1351 * Unlocks the filesystem and marks it writeable again after freeze_super().
1352 */
1353int thaw_super(struct super_block *sb)
1354{
1355 int error;
1356
1357 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001358 if (sb->s_writers.frozen == SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001359 up_write(&sb->s_umount);
1360 return -EINVAL;
1361 }
1362
1363 if (sb->s_flags & MS_RDONLY)
1364 goto out;
1365
1366 if (sb->s_op->unfreeze_fs) {
1367 error = sb->s_op->unfreeze_fs(sb);
1368 if (error) {
1369 printk(KERN_ERR
1370 "VFS:Filesystem thaw failed\n");
Josef Bacik18e9e512010-03-23 10:34:56 -04001371 up_write(&sb->s_umount);
1372 return error;
1373 }
1374 }
1375
1376out:
Jan Kara5accdf82012-06-12 16:20:34 +02001377 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001378 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001379 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001380 deactivate_locked_super(sb);
1381
1382 return 0;
1383}
1384EXPORT_SYMBOL(thaw_super);