blob: 0225c20f877047abb0a47a856c6099594964b6aa [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
115 if (!grab_super_passive(sb))
116 return 0;
117
118 if (sb->s_op && sb->s_op->nr_cached_objects)
Dave Chinner9b17c622013-08-28 10:18:05 +1000119 total_objects = sb->s_op->nr_cached_objects(sb,
120 sc->nid);
Dave Chinner0a234c62013-08-28 10:17:57 +1000121
Dave Chinner9b17c622013-08-28 10:18:05 +1000122 total_objects += list_lru_count_node(&sb->s_dentry_lru,
123 sc->nid);
124 total_objects += list_lru_count_node(&sb->s_inode_lru,
125 sc->nid);
Dave Chinner0a234c62013-08-28 10:17:57 +1000126
Glauber Costa55f841c2013-08-28 10:17:53 +1000127 total_objects = vfs_pressure_ratio(total_objects);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000128 drop_super(sb);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000129 return total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000130}
131
Jan Kara5accdf82012-06-12 16:20:34 +0200132static int init_sb_writers(struct super_block *s, struct file_system_type *type)
133{
134 int err;
135 int i;
136
137 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
138 err = percpu_counter_init(&s->s_writers.counter[i], 0);
139 if (err < 0)
140 goto err_out;
141 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
142 &type->s_writers_key[i], 0);
143 }
144 init_waitqueue_head(&s->s_writers.wait);
145 init_waitqueue_head(&s->s_writers.wait_unfrozen);
146 return 0;
147err_out:
148 while (--i >= 0)
149 percpu_counter_destroy(&s->s_writers.counter[i]);
150 return err;
151}
152
153static void destroy_sb_writers(struct super_block *s)
154{
155 int i;
156
157 for (i = 0; i < SB_FREEZE_LEVELS; i++)
158 percpu_counter_destroy(&s->s_writers.counter[i]);
159}
160
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161/**
162 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700163 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100164 * @flags: the mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 *
166 * Allocates and initializes a new &struct super_block. alloc_super()
167 * returns a pointer new superblock or %NULL if allocation had failed.
168 */
David Howells9249e172012-06-25 12:55:37 +0100169static struct super_block *alloc_super(struct file_system_type *type, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800171 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700172 static const struct super_operations default_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
174 if (s) {
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200175 if (security_sb_alloc(s))
176 goto out_free_sb;
177
Nick Piggin6416ccb2010-08-18 04:37:38 +1000178#ifdef CONFIG_SMP
179 s->s_files = alloc_percpu(struct list_head);
Jan Kara5accdf82012-06-12 16:20:34 +0200180 if (!s->s_files)
181 goto err_out;
182 else {
Nick Piggin6416ccb2010-08-18 04:37:38 +1000183 int i;
184
185 for_each_possible_cpu(i)
186 INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
187 }
188#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 INIT_LIST_HEAD(&s->s_files);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000190#endif
Jan Kara5accdf82012-06-12 16:20:34 +0200191 if (init_sb_writers(s, type))
192 goto err_out;
David Howells9249e172012-06-25 12:55:37 +0100193 s->s_flags = flags;
Jens Axboe95f28602011-03-17 11:13:12 +0100194 s->s_bdi = &default_backing_dev_info;
Al Viroa5166162011-12-12 22:53:00 -0500195 INIT_HLIST_NODE(&s->s_instances);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100196 INIT_HLIST_BL_HEAD(&s->s_anon);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 INIT_LIST_HEAD(&s->s_inodes);
Glauber Costa5ca302c2013-08-28 10:18:18 +1000198
199 if (list_lru_init(&s->s_dentry_lru))
200 goto err_out;
201 if (list_lru_init(&s->s_inode_lru))
202 goto err_out_dentry_lru;
203
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100204 INIT_LIST_HEAD(&s->s_mounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 init_rwsem(&s->s_umount);
Arjan van de Ven897c6ff2006-07-03 00:25:28 -0700206 lockdep_set_class(&s->s_umount, &type->s_umount_key);
Ingo Molnarcf516242006-07-03 00:25:27 -0700207 /*
Peter Zijlstraada723d2009-02-18 14:48:30 -0800208 * sget() can have s_umount recursion.
209 *
210 * When it cannot find a suitable sb, it allocates a new
211 * one (this one), and tries again to find a suitable old
212 * one.
213 *
214 * In case that succeeds, it will acquire the s_umount
215 * lock of the old one. Since these are clearly distrinct
216 * locks, and this object isn't exposed yet, there's no
217 * risk of deadlocks.
218 *
219 * Annotate this by putting this lock in a different
220 * subclass.
221 */
222 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
Al Virob20bd1a2010-03-22 08:53:19 -0400223 s->s_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 atomic_set(&s->s_active, 1);
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800225 mutex_init(&s->s_vfs_rename_mutex);
Roland Dreier51ee0492010-04-27 14:23:57 -0700226 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
Ingo Molnard3be9152006-03-23 03:00:29 -0800227 mutex_init(&s->s_dquot.dqio_mutex);
228 mutex_init(&s->s_dquot.dqonoff_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 init_rwsem(&s->s_dquot.dqptr_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 s->s_maxbytes = MAX_NON_LFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 s->s_op = &default_op;
232 s->s_time_gran = 1000000000;
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600233 s->cleancache_poolid = -1;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000234
235 s->s_shrink.seeks = DEFAULT_SEEKS;
Dave Chinner0a234c62013-08-28 10:17:57 +1000236 s->s_shrink.scan_objects = super_cache_scan;
237 s->s_shrink.count_objects = super_cache_count;
Dave Chinner8ab47662011-07-08 14:14:45 +1000238 s->s_shrink.batch = 1024;
Dave Chinner9b17c622013-08-28 10:18:05 +1000239 s->s_shrink.flags = SHRINKER_NUMA_AWARE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 }
241out:
242 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000243
244err_out_dentry_lru:
245 list_lru_destroy(&s->s_dentry_lru);
Jan Kara5accdf82012-06-12 16:20:34 +0200246err_out:
247 security_sb_free(s);
248#ifdef CONFIG_SMP
249 if (s->s_files)
250 free_percpu(s->s_files);
251#endif
252 destroy_sb_writers(s);
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200253out_free_sb:
Jan Kara5accdf82012-06-12 16:20:34 +0200254 kfree(s);
255 s = NULL;
256 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
258
259/**
260 * destroy_super - frees a superblock
261 * @s: superblock to free
262 *
263 * Frees a superblock.
264 */
265static inline void destroy_super(struct super_block *s)
266{
Al Viroc2d22ec2013-10-01 13:11:21 -0400267 list_lru_destroy(&s->s_dentry_lru);
268 list_lru_destroy(&s->s_inode_lru);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000269#ifdef CONFIG_SMP
270 free_percpu(s->s_files);
271#endif
Jan Kara5accdf82012-06-12 16:20:34 +0200272 destroy_sb_writers(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 security_sb_free(s);
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100274 WARN_ON(!list_empty(&s->s_mounts));
Miklos Szeredi79c0b2d2007-05-08 00:25:43 -0700275 kfree(s->s_subtype);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800276 kfree(s->s_options);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 kfree(s);
278}
279
280/* Superblock refcounting */
281
282/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400283 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 */
Al Virof47ec3f2011-11-21 21:15:42 -0500285static void __put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 if (!--sb->s_count) {
Al Viro551de6f2010-03-22 19:36:35 -0400288 list_del_init(&sb->s_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 destroy_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291}
292
293/**
294 * put_super - drop a temporary reference to superblock
295 * @sb: superblock in question
296 *
297 * Drops a temporary reference, frees superblock if there's no
298 * references left.
299 */
Al Virof47ec3f2011-11-21 21:15:42 -0500300static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
302 spin_lock(&sb_lock);
303 __put_super(sb);
304 spin_unlock(&sb_lock);
305}
306
307
308/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400309 * deactivate_locked_super - drop an active reference to superblock
310 * @s: superblock to deactivate
311 *
Al Viro1712ac82010-03-22 15:22:31 -0400312 * Drops an active reference to superblock, converting it into a temprory
313 * one if there is no other active references left. In that case we
314 * tell fs driver to shut it down and drop the temporary reference we
315 * had just acquired.
316 *
317 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400318 */
319void deactivate_locked_super(struct super_block *s)
320{
321 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400322 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400323 cleancache_invalidate_fs(s);
Al Viro74dbbdd2009-05-06 01:07:50 -0400324 fs->kill_sb(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000325
326 /* caches are now gone, we can safely kill the shrinker now */
327 unregister_shrinker(&s->s_shrink);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000328
Al Viro74dbbdd2009-05-06 01:07:50 -0400329 put_filesystem(fs);
330 put_super(s);
331 } else {
332 up_write(&s->s_umount);
333 }
334}
335
336EXPORT_SYMBOL(deactivate_locked_super);
337
338/**
Al Viro1712ac82010-03-22 15:22:31 -0400339 * deactivate_super - drop an active reference to superblock
340 * @s: superblock to deactivate
341 *
342 * Variant of deactivate_locked_super(), except that superblock is *not*
343 * locked by caller. If we are going to drop the final active reference,
344 * lock will be acquired prior to that.
345 */
346void deactivate_super(struct super_block *s)
347{
348 if (!atomic_add_unless(&s->s_active, -1, 1)) {
349 down_write(&s->s_umount);
350 deactivate_locked_super(s);
351 }
352}
353
354EXPORT_SYMBOL(deactivate_super);
355
356/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 * grab_super - acquire an active reference
358 * @s: reference we are trying to make active
359 *
360 * Tries to acquire an active reference. grab_super() is used when we
361 * had just found a superblock in super_blocks or fs_type->fs_supers
362 * and want to turn it into a full-blown active reference. grab_super()
363 * is called with sb_lock held and drops it. Returns 1 in case of
364 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400365 * dying when grab_super() had been called). Note that this is only
366 * called for superblocks not in rundown mode (== ones still on ->fs_supers
367 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700369static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370{
371 s->s_count++;
372 spin_unlock(&sb_lock);
373 down_write(&s->s_umount);
Al Viroacfec9a2013-07-20 03:13:55 +0400374 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
375 put_super(s);
376 return 1;
377 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 up_write(&s->s_umount);
379 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 return 0;
381}
382
David Howellscf9a2ae2006-08-29 19:05:54 +0100383/*
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000384 * grab_super_passive - acquire a passive reference
Wanpeng Li331cbde2012-06-09 11:10:55 +0800385 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000386 *
387 * Tries to acquire a passive reference. This is used in places where we
388 * cannot take an active reference but we need to ensure that the
389 * superblock does not go away while we are working on it. It returns
390 * false if a reference was not gained, and returns true with the s_umount
391 * lock held in read mode if a reference is gained. On successful return,
392 * the caller must drop the s_umount lock and the passive reference when
393 * done.
394 */
395bool grab_super_passive(struct super_block *sb)
396{
397 spin_lock(&sb_lock);
Al Viroa5166162011-12-12 22:53:00 -0500398 if (hlist_unhashed(&sb->s_instances)) {
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000399 spin_unlock(&sb_lock);
400 return false;
401 }
402
403 sb->s_count++;
404 spin_unlock(&sb_lock);
405
406 if (down_read_trylock(&sb->s_umount)) {
Al Virodabe0dc2012-01-03 21:01:29 -0500407 if (sb->s_root && (sb->s_flags & MS_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000408 return true;
409 up_read(&sb->s_umount);
410 }
411
412 put_super(sb);
413 return false;
414}
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416/**
417 * generic_shutdown_super - common helper for ->kill_sb()
418 * @sb: superblock to kill
419 *
420 * generic_shutdown_super() does all fs-independent work on superblock
421 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
422 * that need destruction out of superblock, call generic_shutdown_super()
423 * and release aforementioned objects. Note: dentries and inodes _are_
424 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700425 *
426 * Upon calling this function, the filesystem may no longer alter or
427 * rearrange the set of dentries belonging to this super_block, nor may it
428 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 */
430void generic_shutdown_super(struct super_block *sb)
431{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800432 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
David Howellsc636ebd2006-10-11 01:22:19 -0700434 if (sb->s_root) {
435 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200436 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 sb->s_flags &= ~MS_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800438
Al Viro63997e92010-10-25 20:49:35 -0400439 fsnotify_unmount_inodes(&sb->s_inodes);
440
441 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200443 if (sb->s_dio_done_wq) {
444 destroy_workqueue(sb->s_dio_done_wq);
445 sb->s_dio_done_wq = NULL;
446 }
447
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 if (sop->put_super)
449 sop->put_super(sb);
450
Al Viro63997e92010-10-25 20:49:35 -0400451 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800452 printk("VFS: Busy inodes after unmount of %s. "
453 "Self-destruct in 5 seconds. Have a nice day...\n",
454 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 }
457 spin_lock(&sb_lock);
458 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500459 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 spin_unlock(&sb_lock);
461 up_write(&sb->s_umount);
462}
463
464EXPORT_SYMBOL(generic_shutdown_super);
465
466/**
467 * sget - find or create a superblock
468 * @type: filesystem type superblock should belong to
469 * @test: comparison callback
470 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100471 * @flags: mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 * @data: argument to each of them
473 */
474struct super_block *sget(struct file_system_type *type,
475 int (*test)(struct super_block *,void *),
476 int (*set)(struct super_block *,void *),
David Howells9249e172012-06-25 12:55:37 +0100477 int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 void *data)
479{
480 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700481 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 int err;
483
484retry:
485 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700486 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800487 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700488 if (!test(old, data))
489 continue;
490 if (!grab_super(old))
491 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700492 if (s) {
493 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700494 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400495 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700496 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700497 return old;
498 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 }
500 if (!s) {
501 spin_unlock(&sb_lock);
David Howells9249e172012-06-25 12:55:37 +0100502 s = alloc_super(type, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 if (!s)
504 return ERR_PTR(-ENOMEM);
505 goto retry;
506 }
507
508 err = set(s, data);
509 if (err) {
510 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700511 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 destroy_super(s);
513 return ERR_PTR(err);
514 }
515 s->s_type = type;
516 strlcpy(s->s_id, type->name, sizeof(s->s_id));
517 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500518 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 spin_unlock(&sb_lock);
520 get_filesystem(type);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000521 register_shrinker(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 return s;
523}
524
525EXPORT_SYMBOL(sget);
526
527void drop_super(struct super_block *sb)
528{
529 up_read(&sb->s_umount);
530 put_super(sb);
531}
532
533EXPORT_SYMBOL(drop_super);
534
Christoph Hellwige5004752009-05-05 16:08:56 +0200535/**
Al Viro01a05b32010-03-23 06:06:58 -0400536 * iterate_supers - call function for all active superblocks
537 * @f: function to call
538 * @arg: argument to pass to it
539 *
540 * Scans the superblock list and calls given function, passing it
541 * locked superblock and given argument.
542 */
543void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
544{
Al Virodca33252010-07-25 02:31:46 +0400545 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400546
547 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400548 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500549 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400550 continue;
551 sb->s_count++;
552 spin_unlock(&sb_lock);
553
554 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500555 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400556 f(sb, arg);
557 up_read(&sb->s_umount);
558
559 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400560 if (p)
561 __put_super(p);
562 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400563 }
Al Virodca33252010-07-25 02:31:46 +0400564 if (p)
565 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400566 spin_unlock(&sb_lock);
567}
568
569/**
Al Viro43e15cd2011-06-03 20:16:57 -0400570 * iterate_supers_type - call function for superblocks of given type
571 * @type: fs type
572 * @f: function to call
573 * @arg: argument to pass to it
574 *
575 * Scans the superblock list and calls given function, passing it
576 * locked superblock and given argument.
577 */
578void iterate_supers_type(struct file_system_type *type,
579 void (*f)(struct super_block *, void *), void *arg)
580{
581 struct super_block *sb, *p = NULL;
582
583 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800584 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400585 sb->s_count++;
586 spin_unlock(&sb_lock);
587
588 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500589 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400590 f(sb, arg);
591 up_read(&sb->s_umount);
592
593 spin_lock(&sb_lock);
594 if (p)
595 __put_super(p);
596 p = sb;
597 }
598 if (p)
599 __put_super(p);
600 spin_unlock(&sb_lock);
601}
602
603EXPORT_SYMBOL(iterate_supers_type);
604
605/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 * get_super - get the superblock of a device
607 * @bdev: device to get the superblock for
608 *
609 * Scans the superblock list and finds the superblock of the file system
610 * mounted on the device given. %NULL is returned if no match is found.
611 */
612
Al Virodf40c012010-03-22 20:23:25 -0400613struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700615 struct super_block *sb;
616
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 if (!bdev)
618 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700621rescan:
622 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500623 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400624 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700625 if (sb->s_bdev == bdev) {
626 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700628 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400629 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500630 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700631 return sb;
632 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400633 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700634 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400635 __put_super(sb);
636 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 }
638 }
639 spin_unlock(&sb_lock);
640 return NULL;
641}
642
643EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200644
645/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100646 * get_super_thawed - get thawed superblock of a device
647 * @bdev: device to get the superblock for
648 *
649 * Scans the superblock list and finds the superblock of the file system
650 * mounted on the device. The superblock is returned once it is thawed
651 * (or immediately if it was not frozen). %NULL is returned if no match
652 * is found.
653 */
654struct super_block *get_super_thawed(struct block_device *bdev)
655{
656 while (1) {
657 struct super_block *s = get_super(bdev);
Jan Kara5accdf82012-06-12 16:20:34 +0200658 if (!s || s->s_writers.frozen == SB_UNFROZEN)
Jan Kara6b6dc832012-02-10 11:03:00 +0100659 return s;
660 up_read(&s->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +0200661 wait_event(s->s_writers.wait_unfrozen,
662 s->s_writers.frozen == SB_UNFROZEN);
Jan Kara6b6dc832012-02-10 11:03:00 +0100663 put_super(s);
664 }
665}
666EXPORT_SYMBOL(get_super_thawed);
667
668/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200669 * get_active_super - get an active reference to the superblock of a device
670 * @bdev: device to get the superblock for
671 *
672 * Scans the superblock list and finds the superblock of the file system
673 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400674 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200675 */
676struct super_block *get_active_super(struct block_device *bdev)
677{
678 struct super_block *sb;
679
680 if (!bdev)
681 return NULL;
682
Al Viro14945832010-03-22 20:15:33 -0400683restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200684 spin_lock(&sb_lock);
685 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500686 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400687 continue;
Al Viro14945832010-03-22 20:15:33 -0400688 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400689 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400690 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400691 up_write(&sb->s_umount);
692 return sb;
Al Viro14945832010-03-22 20:15:33 -0400693 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200694 }
695 spin_unlock(&sb_lock);
696 return NULL;
697}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Al Virodf40c012010-03-22 20:23:25 -0400699struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700701 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700704rescan:
705 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500706 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400707 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700708 if (sb->s_dev == dev) {
709 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700711 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400712 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500713 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700714 return sb;
715 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400716 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700717 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400718 __put_super(sb);
719 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 }
721 }
722 spin_unlock(&sb_lock);
723 return NULL;
724}
725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 * do_remount_sb - asks filesystem to change mount options.
728 * @sb: superblock in question
729 * @flags: numeric part of options
730 * @data: the rest of options
731 * @force: whether or not to force the change
732 *
733 * Alters the mount options of a mounted file system.
734 */
735int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
736{
737 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400738 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200739
Jan Kara5accdf82012-06-12 16:20:34 +0200740 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200741 return -EBUSY;
742
David Howells93614012006-09-30 20:45:40 +0200743#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
745 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200746#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200747
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 if (flags & MS_RDONLY)
749 acct_auto_close(sb);
750 shrink_dcache_sb(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200751 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
Nick Piggind208bbd2009-12-21 16:28:53 -0800753 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800754
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 /* If we are remounting RDONLY and current sb is read/write,
756 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800757 if (remount_ro) {
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100758 if (force) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 mark_files_ro(sb);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100760 } else {
761 retval = sb_prepare_remount_readonly(sb);
762 if (retval)
763 return retval;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 }
766
767 if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 retval = sb->s_op->remount_fs(sb, &flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200769 if (retval) {
770 if (!force)
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100771 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200772 /* If forced remount, go ahead despite any errors */
773 WARN(1, "forced remount of a %s fs returned %i\n",
774 sb->s_type->name, retval);
775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 }
777 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100778 /* Needs to be ordered wrt mnt_is_readonly() */
779 smp_wmb();
780 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400781
Nick Piggind208bbd2009-12-21 16:28:53 -0800782 /*
783 * Some filesystems modify their metadata via some other path than the
784 * bdev buffer cache (eg. use a private mapping, or directories in
785 * pagecache, etc). Also file data modifications go via their own
786 * mappings. So If we try to mount readonly then copy the filesystem
787 * from bdev, we could get stale data, so invalidate it to give a best
788 * effort at coherency.
789 */
790 if (remount_ro && sb->s_bdev)
791 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 return 0;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100793
794cancel_readonly:
795 sb->s_readonly_remount = 0;
796 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797}
798
Jens Axboea2a95372009-03-17 09:38:40 +0100799static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800{
Al Virodca33252010-07-25 02:31:46 +0400801 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802
803 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400804 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500805 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400806 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 sb->s_count++;
808 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400809 down_write(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500810 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
811 !(sb->s_flags & MS_RDONLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 * What lock protects sb->s_flags??
814 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 }
Al Viro443b94b2009-05-05 23:48:50 -0400817 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400819 if (p)
820 __put_super(p);
821 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 }
Al Virodca33252010-07-25 02:31:46 +0400823 if (p)
824 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100826 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 printk("Emergency Remount complete\n");
828}
829
830void emergency_remount(void)
831{
Jens Axboea2a95372009-03-17 09:38:40 +0100832 struct work_struct *work;
833
834 work = kmalloc(sizeof(*work), GFP_ATOMIC);
835 if (work) {
836 INIT_WORK(work, do_emergency_remount);
837 schedule_work(work);
838 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839}
840
841/*
842 * Unnamed block devices are dummy devices used by virtual
843 * filesystems which don't use real block-devices. -- jrs
844 */
845
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400846static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Al Viroc63e09e2009-06-24 02:05:18 -0400848static int unnamed_dev_start = 0; /* don't bother trying below it */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
Al Viro0ee5dc62011-07-07 15:44:25 -0400850int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851{
852 int dev;
853 int error;
854
855 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400856 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 return -ENOMEM;
858 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400859 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400860 if (!error)
861 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 spin_unlock(&unnamed_dev_lock);
863 if (error == -EAGAIN)
864 /* We raced and lost with another CPU. */
865 goto retry;
866 else if (error)
867 return -EAGAIN;
868
Tejun Heoe8c8d1b2013-02-27 17:05:04 -0800869 if (dev == (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400871 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400872 if (unnamed_dev_start > dev)
873 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 spin_unlock(&unnamed_dev_lock);
875 return -EMFILE;
876 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400877 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 return 0;
879}
Al Viro0ee5dc62011-07-07 15:44:25 -0400880EXPORT_SYMBOL(get_anon_bdev);
881
882void free_anon_bdev(dev_t dev)
883{
884 int slot = MINOR(dev);
885 spin_lock(&unnamed_dev_lock);
886 ida_remove(&unnamed_dev_ida, slot);
887 if (slot < unnamed_dev_start)
888 unnamed_dev_start = slot;
889 spin_unlock(&unnamed_dev_lock);
890}
891EXPORT_SYMBOL(free_anon_bdev);
892
893int set_anon_super(struct super_block *s, void *data)
894{
895 int error = get_anon_bdev(&s->s_dev);
896 if (!error)
897 s->s_bdi = &noop_backing_dev_info;
898 return error;
899}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
901EXPORT_SYMBOL(set_anon_super);
902
903void kill_anon_super(struct super_block *sb)
904{
Al Viro0ee5dc62011-07-07 15:44:25 -0400905 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400907 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908}
909
910EXPORT_SYMBOL(kill_anon_super);
911
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912void kill_litter_super(struct super_block *sb)
913{
914 if (sb->s_root)
915 d_genocide(sb->s_root);
916 kill_anon_super(sb);
917}
918
919EXPORT_SYMBOL(kill_litter_super);
920
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700921static int ns_test_super(struct super_block *sb, void *data)
922{
923 return sb->s_fs_info == data;
924}
925
926static int ns_set_super(struct super_block *sb, void *data)
927{
928 sb->s_fs_info = data;
929 return set_anon_super(sb, NULL);
930}
931
Al Viroceefda62010-07-26 13:16:50 +0400932struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
933 void *data, int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700934{
935 struct super_block *sb;
936
David Howells9249e172012-06-25 12:55:37 +0100937 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700938 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +0400939 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700940
941 if (!sb->s_root) {
942 int err;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700943 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
944 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400945 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +0400946 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700947 }
948
949 sb->s_flags |= MS_ACTIVE;
950 }
951
Al Viroceefda62010-07-26 13:16:50 +0400952 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700953}
954
Al Viroceefda62010-07-26 13:16:50 +0400955EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700956
David Howells93614012006-09-30 20:45:40 +0200957#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958static int set_bdev_super(struct super_block *s, void *data)
959{
960 s->s_bdev = data;
961 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +0200962
963 /*
964 * We set the bdi here to the queue backing, file systems can
965 * overwrite this in ->fill_super()
966 */
967 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 return 0;
969}
970
971static int test_bdev_super(struct super_block *s, void *data)
972{
973 return (void *)s->s_bdev == data;
974}
975
Al Viro152a0832010-07-25 00:46:55 +0400976struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +0400978 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979{
980 struct block_device *bdev;
981 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +0100982 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 int error = 0;
984
Al Viro30c40d22008-02-22 19:50:45 -0500985 if (!(flags & MS_RDONLY))
986 mode |= FMODE_WRITE;
987
Tejun Heod4d77622010-11-13 11:55:18 +0100988 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +0400990 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
992 /*
993 * once the super is inserted into the list by sget, s_umount
994 * will protect the lockfs code from trying to start a snapshot
995 * while we are mounting
996 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200997 mutex_lock(&bdev->bd_fsfreeze_mutex);
998 if (bdev->bd_fsfreeze_count > 0) {
999 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1000 error = -EBUSY;
1001 goto error_bdev;
1002 }
David Howells9249e172012-06-25 12:55:37 +01001003 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1004 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +02001005 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -07001007 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009 if (s->s_root) {
1010 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001011 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001012 error = -EBUSY;
1013 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 }
David Howells454e2392006-06-23 02:02:57 -07001015
Tejun Heo4f331f02010-07-20 15:18:07 -07001016 /*
1017 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +01001018 * __invalidate_device(). blkdev_put() acquires
1019 * bd_mutex and can't be called under s_umount. Drop
1020 * s_umount temporarily. This is safe as we're
1021 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -07001022 */
1023 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +01001024 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -07001025 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 } else {
1027 char b[BDEVNAME_SIZE];
1028
Al Viro30c40d22008-02-22 19:50:45 -05001029 s->s_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
Pekka Enberge78c9a02006-01-08 01:03:39 -08001031 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001032 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001034 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001035 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001036 }
David Howells454e2392006-06-23 02:02:57 -07001037
1038 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001039 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 }
1041
Al Viro152a0832010-07-25 00:46:55 +04001042 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
David Howells454e2392006-06-23 02:02:57 -07001044error_s:
1045 error = PTR_ERR(s);
1046error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001047 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001048error:
Al Viro152a0832010-07-25 00:46:55 +04001049 return ERR_PTR(error);
1050}
1051EXPORT_SYMBOL(mount_bdev);
1052
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053void kill_block_super(struct super_block *sb)
1054{
1055 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001056 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001058 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 generic_shutdown_super(sb);
1060 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001061 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001062 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063}
1064
1065EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001066#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
Al Viro3c26ff62010-07-25 11:46:36 +04001068struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001070 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
1072 int error;
David Howells9249e172012-06-25 12:55:37 +01001073 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
1075 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001076 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001078 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001080 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001081 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 }
1083 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001084 return dget(s->s_root);
1085}
1086EXPORT_SYMBOL(mount_nodev);
1087
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088static int compare_single(struct super_block *s, void *p)
1089{
1090 return 1;
1091}
1092
Al Virofc14f2f2010-07-25 01:48:30 +04001093struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001095 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096{
1097 struct super_block *s;
1098 int error;
1099
David Howells9249e172012-06-25 12:55:37 +01001100 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001102 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 if (!s->s_root) {
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001104 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001106 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001107 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 }
1109 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001110 } else {
1111 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 }
Al Virofc14f2f2010-07-25 01:48:30 +04001113 return dget(s->s_root);
1114}
1115EXPORT_SYMBOL(mount_single);
1116
Al Viro9d412a42011-03-17 22:08:28 -04001117struct dentry *
1118mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119{
Al Viroc96e41e2010-07-25 00:17:56 +04001120 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001121 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001123 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001124
Eric Parise0007522008-03-05 10:31:54 -05001125 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001127 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001128 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129
Eric Parise0007522008-03-05 10:31:54 -05001130 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001131 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 }
1134
Al Viro1a102ff2011-03-16 09:07:58 -04001135 root = type->mount(type, flags, name, data);
1136 if (IS_ERR(root)) {
1137 error = PTR_ERR(root);
1138 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001139 }
Al Viro9d412a42011-03-17 22:08:28 -04001140 sb = root->d_sb;
1141 BUG_ON(!sb);
1142 WARN_ON(!sb->s_bdi);
Linus Torvalds6c510382011-03-24 10:16:26 -07001143 WARN_ON(sb->s_bdi == &default_backing_dev_info);
Al Viro9d412a42011-03-17 22:08:28 -04001144 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001145
Al Viro9d412a42011-03-17 22:08:28 -04001146 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001147 if (error)
1148 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001149
Jeff Layton42cb56a2009-09-18 13:05:53 -07001150 /*
1151 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1152 * but s_maxbytes was an unsigned long long for many releases. Throw
1153 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001154 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001155 */
Al Viro9d412a42011-03-17 22:08:28 -04001156 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1157 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001158
Al Viro9d412a42011-03-17 22:08:28 -04001159 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001160 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001161 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001163 dput(root);
1164 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165out_free_secdata:
1166 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167out:
David Howells454e2392006-06-23 02:02:57 -07001168 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169}
1170
Jan Kara5accdf82012-06-12 16:20:34 +02001171/*
1172 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1173 * instead.
1174 */
1175void __sb_end_write(struct super_block *sb, int level)
1176{
1177 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1178 /*
1179 * Make sure s_writers are updated before we wake up waiters in
1180 * freeze_super().
1181 */
1182 smp_mb();
1183 if (waitqueue_active(&sb->s_writers.wait))
1184 wake_up(&sb->s_writers.wait);
1185 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1186}
1187EXPORT_SYMBOL(__sb_end_write);
1188
1189#ifdef CONFIG_LOCKDEP
1190/*
1191 * We want lockdep to tell us about possible deadlocks with freezing but
1192 * it's it bit tricky to properly instrument it. Getting a freeze protection
1193 * works as getting a read lock but there are subtle problems. XFS for example
1194 * gets freeze protection on internal level twice in some cases, which is OK
1195 * only because we already hold a freeze protection also on higher level. Due
1196 * to these cases we have to tell lockdep we are doing trylock when we
1197 * already hold a freeze protection for a higher freeze level.
1198 */
1199static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1200 unsigned long ip)
1201{
1202 int i;
1203
1204 if (!trylock) {
1205 for (i = 0; i < level - 1; i++)
1206 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1207 trylock = true;
1208 break;
1209 }
1210 }
1211 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1212}
1213#endif
1214
1215/*
1216 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1217 * instead.
1218 */
1219int __sb_start_write(struct super_block *sb, int level, bool wait)
1220{
1221retry:
1222 if (unlikely(sb->s_writers.frozen >= level)) {
1223 if (!wait)
1224 return 0;
1225 wait_event(sb->s_writers.wait_unfrozen,
1226 sb->s_writers.frozen < level);
1227 }
1228
1229#ifdef CONFIG_LOCKDEP
1230 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1231#endif
1232 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1233 /*
1234 * Make sure counter is updated before we check for frozen.
1235 * freeze_super() first sets frozen and then checks the counter.
1236 */
1237 smp_mb();
1238 if (unlikely(sb->s_writers.frozen >= level)) {
1239 __sb_end_write(sb, level);
1240 goto retry;
1241 }
1242 return 1;
1243}
1244EXPORT_SYMBOL(__sb_start_write);
1245
1246/**
1247 * sb_wait_write - wait until all writers to given file system finish
1248 * @sb: the super for which we wait
1249 * @level: type of writers we wait for (normal vs page fault)
1250 *
1251 * This function waits until there are no writers of given type to given file
1252 * system. Caller of this function should make sure there can be no new writers
1253 * of type @level before calling this function. Otherwise this function can
1254 * livelock.
1255 */
1256static void sb_wait_write(struct super_block *sb, int level)
1257{
1258 s64 writers;
1259
1260 /*
1261 * We just cycle-through lockdep here so that it does not complain
1262 * about returning with lock to userspace
1263 */
1264 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1265 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1266
1267 do {
1268 DEFINE_WAIT(wait);
1269
1270 /*
1271 * We use a barrier in prepare_to_wait() to separate setting
1272 * of frozen and checking of the counter
1273 */
1274 prepare_to_wait(&sb->s_writers.wait, &wait,
1275 TASK_UNINTERRUPTIBLE);
1276
1277 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1278 if (writers)
1279 schedule();
1280
1281 finish_wait(&sb->s_writers.wait, &wait);
1282 } while (writers);
1283}
1284
Josef Bacik18e9e512010-03-23 10:34:56 -04001285/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001286 * freeze_super - lock the filesystem and force it into a consistent state
1287 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001288 *
1289 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1290 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1291 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001292 *
1293 * During this function, sb->s_writers.frozen goes through these values:
1294 *
1295 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1296 *
1297 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1298 * writes should be blocked, though page faults are still allowed. We wait for
1299 * all writes to complete and then proceed to the next stage.
1300 *
1301 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1302 * but internal fs threads can still modify the filesystem (although they
1303 * should not dirty new pages or inodes), writeback can run etc. After waiting
1304 * for all running page faults we sync the filesystem which will clean all
1305 * dirty pages and inodes (no new dirty pages or inodes can be created when
1306 * sync is running).
1307 *
1308 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1309 * modification are blocked (e.g. XFS preallocation truncation on inode
1310 * reclaim). This is usually implemented by blocking new transactions for
1311 * filesystems that have them and need this additional guard. After all
1312 * internal writers are finished we call ->freeze_fs() to finish filesystem
1313 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1314 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1315 *
1316 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001317 */
1318int freeze_super(struct super_block *sb)
1319{
1320 int ret;
1321
1322 atomic_inc(&sb->s_active);
1323 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001324 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001325 deactivate_locked_super(sb);
1326 return -EBUSY;
1327 }
1328
Al Virodabe0dc2012-01-03 21:01:29 -05001329 if (!(sb->s_flags & MS_BORN)) {
1330 up_write(&sb->s_umount);
1331 return 0; /* sic - it's "nothing to do" */
1332 }
1333
Josef Bacik18e9e512010-03-23 10:34:56 -04001334 if (sb->s_flags & MS_RDONLY) {
Jan Kara5accdf82012-06-12 16:20:34 +02001335 /* Nothing to do really... */
1336 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001337 up_write(&sb->s_umount);
1338 return 0;
1339 }
1340
Jan Kara5accdf82012-06-12 16:20:34 +02001341 /* From now on, no new normal writers can start */
1342 sb->s_writers.frozen = SB_FREEZE_WRITE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001343 smp_wmb();
1344
Jan Kara5accdf82012-06-12 16:20:34 +02001345 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1346 up_write(&sb->s_umount);
1347
1348 sb_wait_write(sb, SB_FREEZE_WRITE);
1349
1350 /* Now we go and block page faults... */
1351 down_write(&sb->s_umount);
1352 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1353 smp_wmb();
1354
1355 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1356
1357 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001358 sync_filesystem(sb);
1359
Jan Kara5accdf82012-06-12 16:20:34 +02001360 /* Now wait for internal filesystem counter */
1361 sb->s_writers.frozen = SB_FREEZE_FS;
Josef Bacik18e9e512010-03-23 10:34:56 -04001362 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001363 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001364
Josef Bacik18e9e512010-03-23 10:34:56 -04001365 if (sb->s_op->freeze_fs) {
1366 ret = sb->s_op->freeze_fs(sb);
1367 if (ret) {
1368 printk(KERN_ERR
1369 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001370 sb->s_writers.frozen = SB_UNFROZEN;
Kazuya Mioe1616302011-12-01 16:51:07 +09001371 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001372 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001373 deactivate_locked_super(sb);
1374 return ret;
1375 }
1376 }
Jan Kara5accdf82012-06-12 16:20:34 +02001377 /*
1378 * This is just for debugging purposes so that fs can warn if it
1379 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1380 */
1381 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001382 up_write(&sb->s_umount);
1383 return 0;
1384}
1385EXPORT_SYMBOL(freeze_super);
1386
1387/**
1388 * thaw_super -- unlock filesystem
1389 * @sb: the super to thaw
1390 *
1391 * Unlocks the filesystem and marks it writeable again after freeze_super().
1392 */
1393int thaw_super(struct super_block *sb)
1394{
1395 int error;
1396
1397 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001398 if (sb->s_writers.frozen == SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001399 up_write(&sb->s_umount);
1400 return -EINVAL;
1401 }
1402
1403 if (sb->s_flags & MS_RDONLY)
1404 goto out;
1405
1406 if (sb->s_op->unfreeze_fs) {
1407 error = sb->s_op->unfreeze_fs(sb);
1408 if (error) {
1409 printk(KERN_ERR
1410 "VFS:Filesystem thaw failed\n");
Josef Bacik18e9e512010-03-23 10:34:56 -04001411 up_write(&sb->s_umount);
1412 return error;
1413 }
1414 }
1415
1416out:
Jan Kara5accdf82012-06-12 16:20:34 +02001417 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001418 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001419 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001420 deactivate_locked_super(sb);
1421
1422 return 0;
1423}
1424EXPORT_SYMBOL(thaw_super);