blob: b9a214d2fe98b8b37a7560ebf1a7d1dc7c5d83e0 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070039LIST_HEAD(super_blocks);
40DEFINE_SPINLOCK(sb_lock);
41
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
74 if (!grab_super_passive(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)
Dave Chinner9b17c622013-08-28 10:18:05 +100078 fs_objects = sb->s_op->nr_cached_objects(sb, sc->nid);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100079
Dave Chinner9b17c622013-08-28 10:18:05 +100080 inodes = list_lru_count_node(&sb->s_inode_lru, sc->nid);
81 dentries = list_lru_count_node(&sb->s_dentry_lru, sc->nid);
Dave Chinnerf6041562013-08-28 10:18:00 +100082 total_objects = dentries + inodes + fs_objects + 1;
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100083
Dave Chinner0a234c62013-08-28 10:17:57 +100084 /* proportion the scan between the caches */
Dave Chinnerf6041562013-08-28 10:18:00 +100085 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +100086 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
Dave Chinnerb0d40c92011-07-08 14:14:42 +100087
Dave Chinner0a234c62013-08-28 10:17:57 +100088 /*
89 * prune the dcache first as the icache is pinned by it, then
90 * prune the icache, followed by the filesystem specific caches
91 */
Dave Chinner9b17c622013-08-28 10:18:05 +100092 freed = prune_dcache_sb(sb, dentries, sc->nid);
93 freed += prune_icache_sb(sb, inodes, sc->nid);
Dave Chinnerb0d40c92011-07-08 14:14:42 +100094
Dave Chinner0a234c62013-08-28 10:17:57 +100095 if (fs_objects) {
96 fs_objects = mult_frac(sc->nr_to_scan, fs_objects,
97 total_objects);
Dave Chinner9b17c622013-08-28 10:18:05 +100098 freed += sb->s_op->free_cached_objects(sb, fs_objects,
99 sc->nid);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000100 }
101
Dave Chinner0a234c62013-08-28 10:17:57 +1000102 drop_super(sb);
103 return freed;
104}
105
106static unsigned long super_cache_count(struct shrinker *shrink,
107 struct shrink_control *sc)
108{
109 struct super_block *sb;
110 long total_objects = 0;
111
112 sb = container_of(shrink, struct super_block, s_shrink);
113
Tim Chend23da152014-06-04 16:10:47 -0700114 /*
115 * Don't call grab_super_passive as it is a potential
116 * scalability bottleneck. The counts could get updated
117 * between super_cache_count and super_cache_scan anyway.
118 * Call to super_cache_count with shrinker_rwsem held
119 * ensures the safety of call to list_lru_count_node() and
120 * s_op->nr_cached_objects().
121 */
Dave Chinner0a234c62013-08-28 10:17:57 +1000122 if (sb->s_op && sb->s_op->nr_cached_objects)
Dave Chinner9b17c622013-08-28 10:18:05 +1000123 total_objects = sb->s_op->nr_cached_objects(sb,
124 sc->nid);
Dave Chinner0a234c62013-08-28 10:17:57 +1000125
Dave Chinner9b17c622013-08-28 10:18:05 +1000126 total_objects += list_lru_count_node(&sb->s_dentry_lru,
127 sc->nid);
128 total_objects += list_lru_count_node(&sb->s_inode_lru,
129 sc->nid);
Dave Chinner0a234c62013-08-28 10:17:57 +1000130
Glauber Costa55f841c2013-08-28 10:17:53 +1000131 total_objects = vfs_pressure_ratio(total_objects);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000132 return total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000133}
134
Al Viro7eb5e882013-10-01 15:09:58 -0400135/**
136 * destroy_super - frees a superblock
137 * @s: superblock to free
138 *
139 * Frees a superblock.
140 */
141static void destroy_super(struct super_block *s)
Jan Kara5accdf82012-06-12 16:20:34 +0200142{
143 int i;
Al Viro7eb5e882013-10-01 15:09:58 -0400144 list_lru_destroy(&s->s_dentry_lru);
145 list_lru_destroy(&s->s_inode_lru);
Jan Kara5accdf82012-06-12 16:20:34 +0200146 for (i = 0; i < SB_FREEZE_LEVELS; i++)
147 percpu_counter_destroy(&s->s_writers.counter[i]);
Al Viro7eb5e882013-10-01 15:09:58 -0400148 security_sb_free(s);
149 WARN_ON(!list_empty(&s->s_mounts));
150 kfree(s->s_subtype);
151 kfree(s->s_options);
Al Viroe2fec7c2013-10-04 17:06:56 -0400152 kfree_rcu(s, rcu);
Jan Kara5accdf82012-06-12 16:20:34 +0200153}
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155/**
156 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700157 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100158 * @flags: the mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 *
160 * Allocates and initializes a new &struct super_block. alloc_super()
161 * returns a pointer new superblock or %NULL if allocation had failed.
162 */
David Howells9249e172012-06-25 12:55:37 +0100163static struct super_block *alloc_super(struct file_system_type *type, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800165 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700166 static const struct super_operations default_op;
Al Viro7eb5e882013-10-01 15:09:58 -0400167 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Al Viro7eb5e882013-10-01 15:09:58 -0400169 if (!s)
170 return NULL;
171
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800172 INIT_LIST_HEAD(&s->s_mounts);
173
Al Viro7eb5e882013-10-01 15:09:58 -0400174 if (security_sb_alloc(s))
175 goto fail;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200176
Al Viro7eb5e882013-10-01 15:09:58 -0400177 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
178 if (percpu_counter_init(&s->s_writers.counter[i], 0) < 0)
179 goto fail;
180 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
181 &type->s_writers_key[i], 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 }
Al Viro7eb5e882013-10-01 15:09:58 -0400183 init_waitqueue_head(&s->s_writers.wait);
184 init_waitqueue_head(&s->s_writers.wait_unfrozen);
185 s->s_flags = flags;
186 s->s_bdi = &default_backing_dev_info;
187 INIT_HLIST_NODE(&s->s_instances);
188 INIT_HLIST_BL_HEAD(&s->s_anon);
189 INIT_LIST_HEAD(&s->s_inodes);
190
191 if (list_lru_init(&s->s_dentry_lru))
192 goto fail;
193 if (list_lru_init(&s->s_inode_lru))
194 goto fail;
195
Al Viro7eb5e882013-10-01 15:09:58 -0400196 init_rwsem(&s->s_umount);
197 lockdep_set_class(&s->s_umount, &type->s_umount_key);
198 /*
199 * sget() can have s_umount recursion.
200 *
201 * When it cannot find a suitable sb, it allocates a new
202 * one (this one), and tries again to find a suitable old
203 * one.
204 *
205 * In case that succeeds, it will acquire the s_umount
206 * lock of the old one. Since these are clearly distrinct
207 * locks, and this object isn't exposed yet, there's no
208 * risk of deadlocks.
209 *
210 * Annotate this by putting this lock in a different
211 * subclass.
212 */
213 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
214 s->s_count = 1;
215 atomic_set(&s->s_active, 1);
216 mutex_init(&s->s_vfs_rename_mutex);
217 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
218 mutex_init(&s->s_dquot.dqio_mutex);
219 mutex_init(&s->s_dquot.dqonoff_mutex);
Al Viro7eb5e882013-10-01 15:09:58 -0400220 s->s_maxbytes = MAX_NON_LFS;
221 s->s_op = &default_op;
222 s->s_time_gran = 1000000000;
223 s->cleancache_poolid = -1;
224
225 s->s_shrink.seeks = DEFAULT_SEEKS;
226 s->s_shrink.scan_objects = super_cache_scan;
227 s->s_shrink.count_objects = super_cache_count;
228 s->s_shrink.batch = 1024;
229 s->s_shrink.flags = SHRINKER_NUMA_AWARE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000231
Al Viro7eb5e882013-10-01 15:09:58 -0400232fail:
233 destroy_super(s);
234 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235}
236
237/* Superblock refcounting */
238
239/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400240 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 */
Al Virof47ec3f2011-11-21 21:15:42 -0500242static void __put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 if (!--sb->s_count) {
Al Viro551de6f2010-03-22 19:36:35 -0400245 list_del_init(&sb->s_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 destroy_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248}
249
250/**
251 * put_super - drop a temporary reference to superblock
252 * @sb: superblock in question
253 *
254 * Drops a temporary reference, frees superblock if there's no
255 * references left.
256 */
Al Virof47ec3f2011-11-21 21:15:42 -0500257static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
259 spin_lock(&sb_lock);
260 __put_super(sb);
261 spin_unlock(&sb_lock);
262}
263
264
265/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400266 * deactivate_locked_super - drop an active reference to superblock
267 * @s: superblock to deactivate
268 *
Al Viro1712ac82010-03-22 15:22:31 -0400269 * Drops an active reference to superblock, converting it into a temprory
270 * one if there is no other active references left. In that case we
271 * tell fs driver to shut it down and drop the temporary reference we
272 * had just acquired.
273 *
274 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400275 */
276void deactivate_locked_super(struct super_block *s)
277{
278 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400279 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400280 cleancache_invalidate_fs(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000281 unregister_shrinker(&s->s_shrink);
Dave Chinner28f2cd4f2014-06-04 16:10:46 -0700282 fs->kill_sb(s);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000283
Al Viro74dbbdd2009-05-06 01:07:50 -0400284 put_filesystem(fs);
285 put_super(s);
286 } else {
287 up_write(&s->s_umount);
288 }
289}
290
291EXPORT_SYMBOL(deactivate_locked_super);
292
293/**
Al Viro1712ac82010-03-22 15:22:31 -0400294 * deactivate_super - drop an active reference to superblock
295 * @s: superblock to deactivate
296 *
297 * Variant of deactivate_locked_super(), except that superblock is *not*
298 * locked by caller. If we are going to drop the final active reference,
299 * lock will be acquired prior to that.
300 */
301void deactivate_super(struct super_block *s)
302{
303 if (!atomic_add_unless(&s->s_active, -1, 1)) {
304 down_write(&s->s_umount);
305 deactivate_locked_super(s);
306 }
307}
308
309EXPORT_SYMBOL(deactivate_super);
310
311/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 * grab_super - acquire an active reference
313 * @s: reference we are trying to make active
314 *
315 * Tries to acquire an active reference. grab_super() is used when we
316 * had just found a superblock in super_blocks or fs_type->fs_supers
317 * and want to turn it into a full-blown active reference. grab_super()
318 * is called with sb_lock held and drops it. Returns 1 in case of
319 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400320 * dying when grab_super() had been called). Note that this is only
321 * called for superblocks not in rundown mode (== ones still on ->fs_supers
322 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700324static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 s->s_count++;
327 spin_unlock(&sb_lock);
328 down_write(&s->s_umount);
Al Viroacfec9a2013-07-20 03:13:55 +0400329 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
330 put_super(s);
331 return 1;
332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 up_write(&s->s_umount);
334 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 return 0;
336}
337
David Howellscf9a2ae2006-08-29 19:05:54 +0100338/*
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000339 * grab_super_passive - acquire a passive reference
Wanpeng Li331cbde2012-06-09 11:10:55 +0800340 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000341 *
342 * Tries to acquire a passive reference. This is used in places where we
343 * cannot take an active reference but we need to ensure that the
344 * superblock does not go away while we are working on it. It returns
345 * false if a reference was not gained, and returns true with the s_umount
346 * lock held in read mode if a reference is gained. On successful return,
347 * the caller must drop the s_umount lock and the passive reference when
348 * done.
349 */
350bool grab_super_passive(struct super_block *sb)
351{
352 spin_lock(&sb_lock);
Al Viroa5166162011-12-12 22:53:00 -0500353 if (hlist_unhashed(&sb->s_instances)) {
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000354 spin_unlock(&sb_lock);
355 return false;
356 }
357
358 sb->s_count++;
359 spin_unlock(&sb_lock);
360
361 if (down_read_trylock(&sb->s_umount)) {
Al Virodabe0dc2012-01-03 21:01:29 -0500362 if (sb->s_root && (sb->s_flags & MS_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000363 return true;
364 up_read(&sb->s_umount);
365 }
366
367 put_super(sb);
368 return false;
369}
370
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371/**
372 * generic_shutdown_super - common helper for ->kill_sb()
373 * @sb: superblock to kill
374 *
375 * generic_shutdown_super() does all fs-independent work on superblock
376 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
377 * that need destruction out of superblock, call generic_shutdown_super()
378 * and release aforementioned objects. Note: dentries and inodes _are_
379 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700380 *
381 * Upon calling this function, the filesystem may no longer alter or
382 * rearrange the set of dentries belonging to this super_block, nor may it
383 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 */
385void generic_shutdown_super(struct super_block *sb)
386{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800387 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
David Howellsc636ebd2006-10-11 01:22:19 -0700389 if (sb->s_root) {
390 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200391 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 sb->s_flags &= ~MS_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800393
Al Viro63997e92010-10-25 20:49:35 -0400394 fsnotify_unmount_inodes(&sb->s_inodes);
395
396 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200398 if (sb->s_dio_done_wq) {
399 destroy_workqueue(sb->s_dio_done_wq);
400 sb->s_dio_done_wq = NULL;
401 }
402
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 if (sop->put_super)
404 sop->put_super(sb);
405
Al Viro63997e92010-10-25 20:49:35 -0400406 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800407 printk("VFS: Busy inodes after unmount of %s. "
408 "Self-destruct in 5 seconds. Have a nice day...\n",
409 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 }
412 spin_lock(&sb_lock);
413 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500414 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 spin_unlock(&sb_lock);
416 up_write(&sb->s_umount);
417}
418
419EXPORT_SYMBOL(generic_shutdown_super);
420
421/**
422 * sget - find or create a superblock
423 * @type: filesystem type superblock should belong to
424 * @test: comparison callback
425 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100426 * @flags: mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 * @data: argument to each of them
428 */
429struct super_block *sget(struct file_system_type *type,
430 int (*test)(struct super_block *,void *),
431 int (*set)(struct super_block *,void *),
David Howells9249e172012-06-25 12:55:37 +0100432 int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 void *data)
434{
435 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700436 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 int err;
438
439retry:
440 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700441 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800442 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700443 if (!test(old, data))
444 continue;
445 if (!grab_super(old))
446 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700447 if (s) {
448 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700449 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400450 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700451 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700452 return old;
453 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 }
455 if (!s) {
456 spin_unlock(&sb_lock);
David Howells9249e172012-06-25 12:55:37 +0100457 s = alloc_super(type, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 if (!s)
459 return ERR_PTR(-ENOMEM);
460 goto retry;
461 }
462
463 err = set(s, data);
464 if (err) {
465 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700466 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 destroy_super(s);
468 return ERR_PTR(err);
469 }
470 s->s_type = type;
471 strlcpy(s->s_id, type->name, sizeof(s->s_id));
472 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500473 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 spin_unlock(&sb_lock);
475 get_filesystem(type);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000476 register_shrinker(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 return s;
478}
479
480EXPORT_SYMBOL(sget);
481
482void drop_super(struct super_block *sb)
483{
484 up_read(&sb->s_umount);
485 put_super(sb);
486}
487
488EXPORT_SYMBOL(drop_super);
489
Christoph Hellwige5004752009-05-05 16:08:56 +0200490/**
Al Viro01a05b32010-03-23 06:06:58 -0400491 * iterate_supers - call function for all active superblocks
492 * @f: function to call
493 * @arg: argument to pass to it
494 *
495 * Scans the superblock list and calls given function, passing it
496 * locked superblock and given argument.
497 */
498void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
499{
Al Virodca33252010-07-25 02:31:46 +0400500 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400501
502 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400503 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500504 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400505 continue;
506 sb->s_count++;
507 spin_unlock(&sb_lock);
508
509 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500510 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400511 f(sb, arg);
512 up_read(&sb->s_umount);
513
514 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400515 if (p)
516 __put_super(p);
517 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400518 }
Al Virodca33252010-07-25 02:31:46 +0400519 if (p)
520 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400521 spin_unlock(&sb_lock);
522}
523
524/**
Al Viro43e15cd2011-06-03 20:16:57 -0400525 * iterate_supers_type - call function for superblocks of given type
526 * @type: fs type
527 * @f: function to call
528 * @arg: argument to pass to it
529 *
530 * Scans the superblock list and calls given function, passing it
531 * locked superblock and given argument.
532 */
533void iterate_supers_type(struct file_system_type *type,
534 void (*f)(struct super_block *, void *), void *arg)
535{
536 struct super_block *sb, *p = NULL;
537
538 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800539 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400540 sb->s_count++;
541 spin_unlock(&sb_lock);
542
543 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500544 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400545 f(sb, arg);
546 up_read(&sb->s_umount);
547
548 spin_lock(&sb_lock);
549 if (p)
550 __put_super(p);
551 p = sb;
552 }
553 if (p)
554 __put_super(p);
555 spin_unlock(&sb_lock);
556}
557
558EXPORT_SYMBOL(iterate_supers_type);
559
560/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 * get_super - get the superblock of a device
562 * @bdev: device to get the superblock for
563 *
564 * Scans the superblock list and finds the superblock of the file system
565 * mounted on the device given. %NULL is returned if no match is found.
566 */
567
Al Virodf40c012010-03-22 20:23:25 -0400568struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700570 struct super_block *sb;
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 if (!bdev)
573 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700576rescan:
577 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500578 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400579 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700580 if (sb->s_bdev == bdev) {
581 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700583 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400584 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500585 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700586 return sb;
587 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400588 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700589 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400590 __put_super(sb);
591 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 }
593 }
594 spin_unlock(&sb_lock);
595 return NULL;
596}
597
598EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200599
600/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100601 * get_super_thawed - get thawed superblock of a device
602 * @bdev: device to get the superblock for
603 *
604 * Scans the superblock list and finds the superblock of the file system
605 * mounted on the device. The superblock is returned once it is thawed
606 * (or immediately if it was not frozen). %NULL is returned if no match
607 * is found.
608 */
609struct super_block *get_super_thawed(struct block_device *bdev)
610{
611 while (1) {
612 struct super_block *s = get_super(bdev);
Jan Kara5accdf82012-06-12 16:20:34 +0200613 if (!s || s->s_writers.frozen == SB_UNFROZEN)
Jan Kara6b6dc832012-02-10 11:03:00 +0100614 return s;
615 up_read(&s->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +0200616 wait_event(s->s_writers.wait_unfrozen,
617 s->s_writers.frozen == SB_UNFROZEN);
Jan Kara6b6dc832012-02-10 11:03:00 +0100618 put_super(s);
619 }
620}
621EXPORT_SYMBOL(get_super_thawed);
622
623/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200624 * get_active_super - get an active reference to the superblock of a device
625 * @bdev: device to get the superblock for
626 *
627 * Scans the superblock list and finds the superblock of the file system
628 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400629 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200630 */
631struct super_block *get_active_super(struct block_device *bdev)
632{
633 struct super_block *sb;
634
635 if (!bdev)
636 return NULL;
637
Al Viro14945832010-03-22 20:15:33 -0400638restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200639 spin_lock(&sb_lock);
640 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500641 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400642 continue;
Al Viro14945832010-03-22 20:15:33 -0400643 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400644 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400645 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400646 up_write(&sb->s_umount);
647 return sb;
Al Viro14945832010-03-22 20:15:33 -0400648 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200649 }
650 spin_unlock(&sb_lock);
651 return NULL;
652}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Al Virodf40c012010-03-22 20:23:25 -0400654struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700656 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700659rescan:
660 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500661 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400662 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700663 if (sb->s_dev == dev) {
664 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700666 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400667 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500668 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700669 return sb;
670 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400671 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700672 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400673 __put_super(sb);
674 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 }
676 }
677 spin_unlock(&sb_lock);
678 return NULL;
679}
680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 * do_remount_sb - asks filesystem to change mount options.
683 * @sb: superblock in question
684 * @flags: numeric part of options
685 * @data: the rest of options
686 * @force: whether or not to force the change
687 *
688 * Alters the mount options of a mounted file system.
689 */
690int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
691{
692 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400693 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200694
Jan Kara5accdf82012-06-12 16:20:34 +0200695 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200696 return -EBUSY;
697
David Howells93614012006-09-30 20:45:40 +0200698#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
700 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200701#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200702
Nick Piggind208bbd2009-12-21 16:28:53 -0800703 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800704
Al Viro0aec09d2014-08-07 07:32:06 -0400705 if (remount_ro) {
706 if (sb->s_pins.first) {
707 up_write(&sb->s_umount);
Al Viro8fa1f1c2014-05-21 18:22:52 -0400708 sb_pin_kill(sb);
Al Viro0aec09d2014-08-07 07:32:06 -0400709 down_write(&sb->s_umount);
710 if (!sb->s_root)
711 return 0;
712 if (sb->s_writers.frozen != SB_UNFROZEN)
713 return -EBUSY;
714 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
715 }
716 }
717 shrink_dcache_sb(sb);
718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 /* If we are remounting RDONLY and current sb is read/write,
720 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800721 if (remount_ro) {
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100722 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400723 sb->s_readonly_remount = 1;
724 smp_wmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100725 } else {
726 retval = sb_prepare_remount_readonly(sb);
727 if (retval)
728 return retval;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100729 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 }
731
732 if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 retval = sb->s_op->remount_fs(sb, &flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200734 if (retval) {
735 if (!force)
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100736 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200737 /* If forced remount, go ahead despite any errors */
738 WARN(1, "forced remount of a %s fs returned %i\n",
739 sb->s_type->name, retval);
740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 }
742 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100743 /* Needs to be ordered wrt mnt_is_readonly() */
744 smp_wmb();
745 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400746
Nick Piggind208bbd2009-12-21 16:28:53 -0800747 /*
748 * Some filesystems modify their metadata via some other path than the
749 * bdev buffer cache (eg. use a private mapping, or directories in
750 * pagecache, etc). Also file data modifications go via their own
751 * mappings. So If we try to mount readonly then copy the filesystem
752 * from bdev, we could get stale data, so invalidate it to give a best
753 * effort at coherency.
754 */
755 if (remount_ro && sb->s_bdev)
756 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 return 0;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100758
759cancel_readonly:
760 sb->s_readonly_remount = 0;
761 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
763
Jens Axboea2a95372009-03-17 09:38:40 +0100764static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765{
Al Virodca33252010-07-25 02:31:46 +0400766 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
768 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400769 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500770 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400771 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 sb->s_count++;
773 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400774 down_write(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500775 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
776 !(sb->s_flags & MS_RDONLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 * What lock protects sb->s_flags??
779 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 }
Al Viro443b94b2009-05-05 23:48:50 -0400782 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400784 if (p)
785 __put_super(p);
786 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 }
Al Virodca33252010-07-25 02:31:46 +0400788 if (p)
789 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100791 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 printk("Emergency Remount complete\n");
793}
794
795void emergency_remount(void)
796{
Jens Axboea2a95372009-03-17 09:38:40 +0100797 struct work_struct *work;
798
799 work = kmalloc(sizeof(*work), GFP_ATOMIC);
800 if (work) {
801 INIT_WORK(work, do_emergency_remount);
802 schedule_work(work);
803 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805
806/*
807 * Unnamed block devices are dummy devices used by virtual
808 * filesystems which don't use real block-devices. -- jrs
809 */
810
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400811static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Thomas Bächlera2a4dc42014-04-03 21:55:37 +0200813/* Many userspace utilities consider an FSID of 0 invalid.
814 * Always return at least 1 from get_anon_bdev.
815 */
816static int unnamed_dev_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
Al Viro0ee5dc62011-07-07 15:44:25 -0400818int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819{
820 int dev;
821 int error;
822
823 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400824 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 return -ENOMEM;
826 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400827 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400828 if (!error)
829 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 spin_unlock(&unnamed_dev_lock);
831 if (error == -EAGAIN)
832 /* We raced and lost with another CPU. */
833 goto retry;
834 else if (error)
835 return -EAGAIN;
836
Tejun Heoe8c8d1b2013-02-27 17:05:04 -0800837 if (dev == (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400839 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400840 if (unnamed_dev_start > dev)
841 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 spin_unlock(&unnamed_dev_lock);
843 return -EMFILE;
844 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400845 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 return 0;
847}
Al Viro0ee5dc62011-07-07 15:44:25 -0400848EXPORT_SYMBOL(get_anon_bdev);
849
850void free_anon_bdev(dev_t dev)
851{
852 int slot = MINOR(dev);
853 spin_lock(&unnamed_dev_lock);
854 ida_remove(&unnamed_dev_ida, slot);
855 if (slot < unnamed_dev_start)
856 unnamed_dev_start = slot;
857 spin_unlock(&unnamed_dev_lock);
858}
859EXPORT_SYMBOL(free_anon_bdev);
860
861int set_anon_super(struct super_block *s, void *data)
862{
863 int error = get_anon_bdev(&s->s_dev);
864 if (!error)
865 s->s_bdi = &noop_backing_dev_info;
866 return error;
867}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
869EXPORT_SYMBOL(set_anon_super);
870
871void kill_anon_super(struct super_block *sb)
872{
Al Viro0ee5dc62011-07-07 15:44:25 -0400873 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400875 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876}
877
878EXPORT_SYMBOL(kill_anon_super);
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880void kill_litter_super(struct super_block *sb)
881{
882 if (sb->s_root)
883 d_genocide(sb->s_root);
884 kill_anon_super(sb);
885}
886
887EXPORT_SYMBOL(kill_litter_super);
888
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700889static int ns_test_super(struct super_block *sb, void *data)
890{
891 return sb->s_fs_info == data;
892}
893
894static int ns_set_super(struct super_block *sb, void *data)
895{
896 sb->s_fs_info = data;
897 return set_anon_super(sb, NULL);
898}
899
Al Viroceefda62010-07-26 13:16:50 +0400900struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
901 void *data, int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700902{
903 struct super_block *sb;
904
David Howells9249e172012-06-25 12:55:37 +0100905 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700906 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +0400907 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700908
909 if (!sb->s_root) {
910 int err;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700911 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
912 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400913 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +0400914 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700915 }
916
917 sb->s_flags |= MS_ACTIVE;
918 }
919
Al Viroceefda62010-07-26 13:16:50 +0400920 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700921}
922
Al Viroceefda62010-07-26 13:16:50 +0400923EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700924
David Howells93614012006-09-30 20:45:40 +0200925#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926static int set_bdev_super(struct super_block *s, void *data)
927{
928 s->s_bdev = data;
929 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +0200930
931 /*
932 * We set the bdi here to the queue backing, file systems can
933 * overwrite this in ->fill_super()
934 */
935 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 return 0;
937}
938
939static int test_bdev_super(struct super_block *s, void *data)
940{
941 return (void *)s->s_bdev == data;
942}
943
Al Viro152a0832010-07-25 00:46:55 +0400944struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +0400946 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947{
948 struct block_device *bdev;
949 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +0100950 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 int error = 0;
952
Al Viro30c40d22008-02-22 19:50:45 -0500953 if (!(flags & MS_RDONLY))
954 mode |= FMODE_WRITE;
955
Tejun Heod4d77622010-11-13 11:55:18 +0100956 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +0400958 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
960 /*
961 * once the super is inserted into the list by sget, s_umount
962 * will protect the lockfs code from trying to start a snapshot
963 * while we are mounting
964 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200965 mutex_lock(&bdev->bd_fsfreeze_mutex);
966 if (bdev->bd_fsfreeze_count > 0) {
967 mutex_unlock(&bdev->bd_fsfreeze_mutex);
968 error = -EBUSY;
969 goto error_bdev;
970 }
David Howells9249e172012-06-25 12:55:37 +0100971 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
972 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200973 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -0700975 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
977 if (s->s_root) {
978 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400979 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -0700980 error = -EBUSY;
981 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 }
David Howells454e2392006-06-23 02:02:57 -0700983
Tejun Heo4f331f02010-07-20 15:18:07 -0700984 /*
985 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +0100986 * __invalidate_device(). blkdev_put() acquires
987 * bd_mutex and can't be called under s_umount. Drop
988 * s_umount temporarily. This is safe as we're
989 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -0700990 */
991 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +0100992 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -0700993 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 } else {
995 char b[BDEVNAME_SIZE];
996
Al Viro30c40d22008-02-22 19:50:45 -0500997 s->s_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
Pekka Enberge78c9a02006-01-08 01:03:39 -0800999 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001000 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001002 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001003 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001004 }
David Howells454e2392006-06-23 02:02:57 -07001005
1006 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001007 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 }
1009
Al Viro152a0832010-07-25 00:46:55 +04001010 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
David Howells454e2392006-06-23 02:02:57 -07001012error_s:
1013 error = PTR_ERR(s);
1014error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001015 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001016error:
Al Viro152a0832010-07-25 00:46:55 +04001017 return ERR_PTR(error);
1018}
1019EXPORT_SYMBOL(mount_bdev);
1020
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021void kill_block_super(struct super_block *sb)
1022{
1023 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001024 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001026 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 generic_shutdown_super(sb);
1028 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001029 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001030 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031}
1032
1033EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001034#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Al Viro3c26ff62010-07-25 11:46:36 +04001036struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001038 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039{
1040 int error;
David Howells9249e172012-06-25 12:55:37 +01001041 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
1043 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001044 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001046 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001048 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001049 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 }
1051 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001052 return dget(s->s_root);
1053}
1054EXPORT_SYMBOL(mount_nodev);
1055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056static int compare_single(struct super_block *s, void *p)
1057{
1058 return 1;
1059}
1060
Al Virofc14f2f2010-07-25 01:48:30 +04001061struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001063 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064{
1065 struct super_block *s;
1066 int error;
1067
David Howells9249e172012-06-25 12:55:37 +01001068 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001070 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 if (!s->s_root) {
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001072 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001074 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001075 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 }
1077 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001078 } else {
1079 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 }
Al Virofc14f2f2010-07-25 01:48:30 +04001081 return dget(s->s_root);
1082}
1083EXPORT_SYMBOL(mount_single);
1084
Al Viro9d412a42011-03-17 22:08:28 -04001085struct dentry *
1086mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087{
Al Viroc96e41e2010-07-25 00:17:56 +04001088 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001089 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001091 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001092
Eric Parise0007522008-03-05 10:31:54 -05001093 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001095 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001096 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
Eric Parise0007522008-03-05 10:31:54 -05001098 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001099 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 }
1102
Al Viro1a102ff2011-03-16 09:07:58 -04001103 root = type->mount(type, flags, name, data);
1104 if (IS_ERR(root)) {
1105 error = PTR_ERR(root);
1106 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001107 }
Al Viro9d412a42011-03-17 22:08:28 -04001108 sb = root->d_sb;
1109 BUG_ON(!sb);
1110 WARN_ON(!sb->s_bdi);
Linus Torvalds6c510382011-03-24 10:16:26 -07001111 WARN_ON(sb->s_bdi == &default_backing_dev_info);
Al Viro9d412a42011-03-17 22:08:28 -04001112 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001113
Al Viro9d412a42011-03-17 22:08:28 -04001114 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001115 if (error)
1116 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001117
Jeff Layton42cb56a2009-09-18 13:05:53 -07001118 /*
1119 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1120 * but s_maxbytes was an unsigned long long for many releases. Throw
1121 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001122 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001123 */
Al Viro9d412a42011-03-17 22:08:28 -04001124 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1125 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001126
Al Viro9d412a42011-03-17 22:08:28 -04001127 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001128 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001129 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001131 dput(root);
1132 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133out_free_secdata:
1134 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135out:
David Howells454e2392006-06-23 02:02:57 -07001136 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137}
1138
Jan Kara5accdf82012-06-12 16:20:34 +02001139/*
1140 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1141 * instead.
1142 */
1143void __sb_end_write(struct super_block *sb, int level)
1144{
1145 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1146 /*
1147 * Make sure s_writers are updated before we wake up waiters in
1148 * freeze_super().
1149 */
1150 smp_mb();
1151 if (waitqueue_active(&sb->s_writers.wait))
1152 wake_up(&sb->s_writers.wait);
1153 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1154}
1155EXPORT_SYMBOL(__sb_end_write);
1156
1157#ifdef CONFIG_LOCKDEP
1158/*
1159 * We want lockdep to tell us about possible deadlocks with freezing but
1160 * it's it bit tricky to properly instrument it. Getting a freeze protection
1161 * works as getting a read lock but there are subtle problems. XFS for example
1162 * gets freeze protection on internal level twice in some cases, which is OK
1163 * only because we already hold a freeze protection also on higher level. Due
1164 * to these cases we have to tell lockdep we are doing trylock when we
1165 * already hold a freeze protection for a higher freeze level.
1166 */
1167static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1168 unsigned long ip)
1169{
1170 int i;
1171
1172 if (!trylock) {
1173 for (i = 0; i < level - 1; i++)
1174 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1175 trylock = true;
1176 break;
1177 }
1178 }
1179 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1180}
1181#endif
1182
1183/*
1184 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1185 * instead.
1186 */
1187int __sb_start_write(struct super_block *sb, int level, bool wait)
1188{
1189retry:
1190 if (unlikely(sb->s_writers.frozen >= level)) {
1191 if (!wait)
1192 return 0;
1193 wait_event(sb->s_writers.wait_unfrozen,
1194 sb->s_writers.frozen < level);
1195 }
1196
1197#ifdef CONFIG_LOCKDEP
1198 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1199#endif
1200 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1201 /*
1202 * Make sure counter is updated before we check for frozen.
1203 * freeze_super() first sets frozen and then checks the counter.
1204 */
1205 smp_mb();
1206 if (unlikely(sb->s_writers.frozen >= level)) {
1207 __sb_end_write(sb, level);
1208 goto retry;
1209 }
1210 return 1;
1211}
1212EXPORT_SYMBOL(__sb_start_write);
1213
1214/**
1215 * sb_wait_write - wait until all writers to given file system finish
1216 * @sb: the super for which we wait
1217 * @level: type of writers we wait for (normal vs page fault)
1218 *
1219 * This function waits until there are no writers of given type to given file
1220 * system. Caller of this function should make sure there can be no new writers
1221 * of type @level before calling this function. Otherwise this function can
1222 * livelock.
1223 */
1224static void sb_wait_write(struct super_block *sb, int level)
1225{
1226 s64 writers;
1227
1228 /*
1229 * We just cycle-through lockdep here so that it does not complain
1230 * about returning with lock to userspace
1231 */
1232 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1233 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1234
1235 do {
1236 DEFINE_WAIT(wait);
1237
1238 /*
1239 * We use a barrier in prepare_to_wait() to separate setting
1240 * of frozen and checking of the counter
1241 */
1242 prepare_to_wait(&sb->s_writers.wait, &wait,
1243 TASK_UNINTERRUPTIBLE);
1244
1245 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1246 if (writers)
1247 schedule();
1248
1249 finish_wait(&sb->s_writers.wait, &wait);
1250 } while (writers);
1251}
1252
Josef Bacik18e9e512010-03-23 10:34:56 -04001253/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001254 * freeze_super - lock the filesystem and force it into a consistent state
1255 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001256 *
1257 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1258 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1259 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001260 *
1261 * During this function, sb->s_writers.frozen goes through these values:
1262 *
1263 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1264 *
1265 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1266 * writes should be blocked, though page faults are still allowed. We wait for
1267 * all writes to complete and then proceed to the next stage.
1268 *
1269 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1270 * but internal fs threads can still modify the filesystem (although they
1271 * should not dirty new pages or inodes), writeback can run etc. After waiting
1272 * for all running page faults we sync the filesystem which will clean all
1273 * dirty pages and inodes (no new dirty pages or inodes can be created when
1274 * sync is running).
1275 *
1276 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1277 * modification are blocked (e.g. XFS preallocation truncation on inode
1278 * reclaim). This is usually implemented by blocking new transactions for
1279 * filesystems that have them and need this additional guard. After all
1280 * internal writers are finished we call ->freeze_fs() to finish filesystem
1281 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1282 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1283 *
1284 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001285 */
1286int freeze_super(struct super_block *sb)
1287{
1288 int ret;
1289
1290 atomic_inc(&sb->s_active);
1291 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001292 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001293 deactivate_locked_super(sb);
1294 return -EBUSY;
1295 }
1296
Al Virodabe0dc2012-01-03 21:01:29 -05001297 if (!(sb->s_flags & MS_BORN)) {
1298 up_write(&sb->s_umount);
1299 return 0; /* sic - it's "nothing to do" */
1300 }
1301
Josef Bacik18e9e512010-03-23 10:34:56 -04001302 if (sb->s_flags & MS_RDONLY) {
Jan Kara5accdf82012-06-12 16:20:34 +02001303 /* Nothing to do really... */
1304 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001305 up_write(&sb->s_umount);
1306 return 0;
1307 }
1308
Jan Kara5accdf82012-06-12 16:20:34 +02001309 /* From now on, no new normal writers can start */
1310 sb->s_writers.frozen = SB_FREEZE_WRITE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001311 smp_wmb();
1312
Jan Kara5accdf82012-06-12 16:20:34 +02001313 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1314 up_write(&sb->s_umount);
1315
1316 sb_wait_write(sb, SB_FREEZE_WRITE);
1317
1318 /* Now we go and block page faults... */
1319 down_write(&sb->s_umount);
1320 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1321 smp_wmb();
1322
1323 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1324
1325 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001326 sync_filesystem(sb);
1327
Jan Kara5accdf82012-06-12 16:20:34 +02001328 /* Now wait for internal filesystem counter */
1329 sb->s_writers.frozen = SB_FREEZE_FS;
Josef Bacik18e9e512010-03-23 10:34:56 -04001330 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001331 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001332
Josef Bacik18e9e512010-03-23 10:34:56 -04001333 if (sb->s_op->freeze_fs) {
1334 ret = sb->s_op->freeze_fs(sb);
1335 if (ret) {
1336 printk(KERN_ERR
1337 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001338 sb->s_writers.frozen = SB_UNFROZEN;
Kazuya Mioe1616302011-12-01 16:51:07 +09001339 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001340 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001341 deactivate_locked_super(sb);
1342 return ret;
1343 }
1344 }
Jan Kara5accdf82012-06-12 16:20:34 +02001345 /*
1346 * This is just for debugging purposes so that fs can warn if it
1347 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1348 */
1349 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001350 up_write(&sb->s_umount);
1351 return 0;
1352}
1353EXPORT_SYMBOL(freeze_super);
1354
1355/**
1356 * thaw_super -- unlock filesystem
1357 * @sb: the super to thaw
1358 *
1359 * Unlocks the filesystem and marks it writeable again after freeze_super().
1360 */
1361int thaw_super(struct super_block *sb)
1362{
1363 int error;
1364
1365 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001366 if (sb->s_writers.frozen == SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001367 up_write(&sb->s_umount);
1368 return -EINVAL;
1369 }
1370
1371 if (sb->s_flags & MS_RDONLY)
1372 goto out;
1373
1374 if (sb->s_op->unfreeze_fs) {
1375 error = sb->s_op->unfreeze_fs(sb);
1376 if (error) {
1377 printk(KERN_ERR
1378 "VFS:Filesystem thaw failed\n");
Josef Bacik18e9e512010-03-23 10:34:56 -04001379 up_write(&sb->s_umount);
1380 return error;
1381 }
1382 }
1383
1384out:
Jan Kara5accdf82012-06-12 16:20:34 +02001385 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001386 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001387 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001388 deactivate_locked_super(sb);
1389
1390 return 0;
1391}
1392EXPORT_SYMBOL(thaw_super);