blob: 68307c029228c51efc0e7967c90c39edd1a85113 [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 */
56static int prune_super(struct shrinker *shrink, struct shrink_control *sc)
57{
58 struct super_block *sb;
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100059 int fs_objects = 0;
60 int total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100061
62 sb = container_of(shrink, struct super_block, s_shrink);
63
64 /*
65 * Deadlock avoidance. We may hold various FS locks, and we don't want
66 * to recurse into the FS that called us in clear_inode() and friends..
67 */
68 if (sc->nr_to_scan && !(sc->gfp_mask & __GFP_FS))
69 return -1;
70
71 if (!grab_super_passive(sb))
Minchan Kim8e125cd2012-07-31 16:43:16 -070072 return -1;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100073
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100074 if (sb->s_op && sb->s_op->nr_cached_objects)
75 fs_objects = sb->s_op->nr_cached_objects(sb);
76
77 total_objects = sb->s_nr_dentry_unused +
78 sb->s_nr_inodes_unused + fs_objects + 1;
79
Dave Chinnerb0d40c92011-07-08 14:14:42 +100080 if (sc->nr_to_scan) {
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100081 int dentries;
82 int inodes;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100083
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100084 /* proportion the scan between the caches */
85 dentries = (sc->nr_to_scan * sb->s_nr_dentry_unused) /
86 total_objects;
87 inodes = (sc->nr_to_scan * sb->s_nr_inodes_unused) /
88 total_objects;
89 if (fs_objects)
90 fs_objects = (sc->nr_to_scan * fs_objects) /
91 total_objects;
92 /*
93 * prune the dcache first as the icache is pinned by it, then
94 * prune the icache, followed by the filesystem specific caches
95 */
96 prune_dcache_sb(sb, dentries);
97 prune_icache_sb(sb, inodes);
Dave Chinnerb0d40c92011-07-08 14:14:42 +100098
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100099 if (fs_objects && sb->s_op->free_cached_objects) {
100 sb->s_op->free_cached_objects(sb, fs_objects);
101 fs_objects = sb->s_op->nr_cached_objects(sb);
102 }
103 total_objects = sb->s_nr_dentry_unused +
104 sb->s_nr_inodes_unused + fs_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000105 }
106
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000107 total_objects = (total_objects / 100) * sysctl_vfs_cache_pressure;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000108 drop_super(sb);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000109 return total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000110}
111
Jan Kara5accdf82012-06-12 16:20:34 +0200112static int init_sb_writers(struct super_block *s, struct file_system_type *type)
113{
114 int err;
115 int i;
116
117 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
118 err = percpu_counter_init(&s->s_writers.counter[i], 0);
119 if (err < 0)
120 goto err_out;
121 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
122 &type->s_writers_key[i], 0);
123 }
124 init_waitqueue_head(&s->s_writers.wait);
125 init_waitqueue_head(&s->s_writers.wait_unfrozen);
126 return 0;
127err_out:
128 while (--i >= 0)
129 percpu_counter_destroy(&s->s_writers.counter[i]);
130 return err;
131}
132
133static void destroy_sb_writers(struct super_block *s)
134{
135 int i;
136
137 for (i = 0; i < SB_FREEZE_LEVELS; i++)
138 percpu_counter_destroy(&s->s_writers.counter[i]);
139}
140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141/**
142 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700143 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100144 * @flags: the mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 *
146 * Allocates and initializes a new &struct super_block. alloc_super()
147 * returns a pointer new superblock or %NULL if allocation had failed.
148 */
David Howells9249e172012-06-25 12:55:37 +0100149static struct super_block *alloc_super(struct file_system_type *type, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800151 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700152 static const struct super_operations default_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
154 if (s) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 if (security_sb_alloc(s)) {
Jan Kara5accdf82012-06-12 16:20:34 +0200156 /*
157 * We cannot call security_sb_free() without
158 * security_sb_alloc() succeeding. So bail out manually
159 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 kfree(s);
161 s = NULL;
162 goto out;
163 }
Nick Piggin6416ccb2010-08-18 04:37:38 +1000164#ifdef CONFIG_SMP
165 s->s_files = alloc_percpu(struct list_head);
Jan Kara5accdf82012-06-12 16:20:34 +0200166 if (!s->s_files)
167 goto err_out;
168 else {
Nick Piggin6416ccb2010-08-18 04:37:38 +1000169 int i;
170
171 for_each_possible_cpu(i)
172 INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
173 }
174#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 INIT_LIST_HEAD(&s->s_files);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000176#endif
Jan Kara5accdf82012-06-12 16:20:34 +0200177 if (init_sb_writers(s, type))
178 goto err_out;
David Howells9249e172012-06-25 12:55:37 +0100179 s->s_flags = flags;
Jens Axboe95f28602011-03-17 11:13:12 +0100180 s->s_bdi = &default_backing_dev_info;
Al Viroa5166162011-12-12 22:53:00 -0500181 INIT_HLIST_NODE(&s->s_instances);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100182 INIT_HLIST_BL_HEAD(&s->s_anon);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 INIT_LIST_HEAD(&s->s_inodes);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700184 INIT_LIST_HEAD(&s->s_dentry_lru);
Dave Chinner98b745c2011-07-08 14:14:39 +1000185 INIT_LIST_HEAD(&s->s_inode_lru);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000186 spin_lock_init(&s->s_inode_lru_lock);
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100187 INIT_LIST_HEAD(&s->s_mounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 init_rwsem(&s->s_umount);
Arjan van de Ven897c6ff2006-07-03 00:25:28 -0700189 lockdep_set_class(&s->s_umount, &type->s_umount_key);
Ingo Molnarcf516242006-07-03 00:25:27 -0700190 /*
Peter Zijlstraada723d2009-02-18 14:48:30 -0800191 * sget() can have s_umount recursion.
192 *
193 * When it cannot find a suitable sb, it allocates a new
194 * one (this one), and tries again to find a suitable old
195 * one.
196 *
197 * In case that succeeds, it will acquire the s_umount
198 * lock of the old one. Since these are clearly distrinct
199 * locks, and this object isn't exposed yet, there's no
200 * risk of deadlocks.
201 *
202 * Annotate this by putting this lock in a different
203 * subclass.
204 */
205 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
Al Virob20bd1a2010-03-22 08:53:19 -0400206 s->s_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 atomic_set(&s->s_active, 1);
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800208 mutex_init(&s->s_vfs_rename_mutex);
Roland Dreier51ee0492010-04-27 14:23:57 -0700209 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
Ingo Molnard3be9152006-03-23 03:00:29 -0800210 mutex_init(&s->s_dquot.dqio_mutex);
211 mutex_init(&s->s_dquot.dqonoff_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 init_rwsem(&s->s_dquot.dqptr_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 s->s_maxbytes = MAX_NON_LFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 s->s_op = &default_op;
215 s->s_time_gran = 1000000000;
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600216 s->cleancache_poolid = -1;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000217
218 s->s_shrink.seeks = DEFAULT_SEEKS;
219 s->s_shrink.shrink = prune_super;
Dave Chinner8ab47662011-07-08 14:14:45 +1000220 s->s_shrink.batch = 1024;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 }
222out:
223 return s;
Jan Kara5accdf82012-06-12 16:20:34 +0200224err_out:
225 security_sb_free(s);
226#ifdef CONFIG_SMP
227 if (s->s_files)
228 free_percpu(s->s_files);
229#endif
230 destroy_sb_writers(s);
231 kfree(s);
232 s = NULL;
233 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234}
235
236/**
237 * destroy_super - frees a superblock
238 * @s: superblock to free
239 *
240 * Frees a superblock.
241 */
242static inline void destroy_super(struct super_block *s)
243{
Nick Piggin6416ccb2010-08-18 04:37:38 +1000244#ifdef CONFIG_SMP
245 free_percpu(s->s_files);
246#endif
Jan Kara5accdf82012-06-12 16:20:34 +0200247 destroy_sb_writers(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 security_sb_free(s);
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +0100249 WARN_ON(!list_empty(&s->s_mounts));
Miklos Szeredi79c0b2d2007-05-08 00:25:43 -0700250 kfree(s->s_subtype);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800251 kfree(s->s_options);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 kfree(s);
253}
254
255/* Superblock refcounting */
256
257/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400258 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 */
Al Virof47ec3f2011-11-21 21:15:42 -0500260static void __put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 if (!--sb->s_count) {
Al Viro551de6f2010-03-22 19:36:35 -0400263 list_del_init(&sb->s_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 destroy_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266}
267
268/**
269 * put_super - drop a temporary reference to superblock
270 * @sb: superblock in question
271 *
272 * Drops a temporary reference, frees superblock if there's no
273 * references left.
274 */
Al Virof47ec3f2011-11-21 21:15:42 -0500275static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
277 spin_lock(&sb_lock);
278 __put_super(sb);
279 spin_unlock(&sb_lock);
280}
281
282
283/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400284 * deactivate_locked_super - drop an active reference to superblock
285 * @s: superblock to deactivate
286 *
Al Viro1712ac82010-03-22 15:22:31 -0400287 * Drops an active reference to superblock, converting it into a temprory
288 * one if there is no other active references left. In that case we
289 * tell fs driver to shut it down and drop the temporary reference we
290 * had just acquired.
291 *
292 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400293 */
294void deactivate_locked_super(struct super_block *s)
295{
296 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400297 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400298 cleancache_invalidate_fs(s);
Al Viro74dbbdd2009-05-06 01:07:50 -0400299 fs->kill_sb(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000300
301 /* caches are now gone, we can safely kill the shrinker now */
302 unregister_shrinker(&s->s_shrink);
Al Viro74dbbdd2009-05-06 01:07:50 -0400303 put_filesystem(fs);
304 put_super(s);
305 } else {
306 up_write(&s->s_umount);
307 }
308}
309
310EXPORT_SYMBOL(deactivate_locked_super);
311
312/**
Al Viro1712ac82010-03-22 15:22:31 -0400313 * deactivate_super - drop an active reference to superblock
314 * @s: superblock to deactivate
315 *
316 * Variant of deactivate_locked_super(), except that superblock is *not*
317 * locked by caller. If we are going to drop the final active reference,
318 * lock will be acquired prior to that.
319 */
320void deactivate_super(struct super_block *s)
321{
322 if (!atomic_add_unless(&s->s_active, -1, 1)) {
323 down_write(&s->s_umount);
324 deactivate_locked_super(s);
325 }
326}
327
328EXPORT_SYMBOL(deactivate_super);
329
330/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 * grab_super - acquire an active reference
332 * @s: reference we are trying to make active
333 *
334 * Tries to acquire an active reference. grab_super() is used when we
335 * had just found a superblock in super_blocks or fs_type->fs_supers
336 * and want to turn it into a full-blown active reference. grab_super()
337 * is called with sb_lock held and drops it. Returns 1 in case of
338 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400339 * dying when grab_super() had been called). Note that this is only
340 * called for superblocks not in rundown mode (== ones still on ->fs_supers
341 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700343static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344{
345 s->s_count++;
346 spin_unlock(&sb_lock);
347 down_write(&s->s_umount);
Al Viroacfec9a2013-07-20 03:13:55 +0400348 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
349 put_super(s);
350 return 1;
351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 up_write(&s->s_umount);
353 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 return 0;
355}
356
David Howellscf9a2ae2006-08-29 19:05:54 +0100357/*
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000358 * grab_super_passive - acquire a passive reference
Wanpeng Li331cbde2012-06-09 11:10:55 +0800359 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000360 *
361 * Tries to acquire a passive reference. This is used in places where we
362 * cannot take an active reference but we need to ensure that the
363 * superblock does not go away while we are working on it. It returns
364 * false if a reference was not gained, and returns true with the s_umount
365 * lock held in read mode if a reference is gained. On successful return,
366 * the caller must drop the s_umount lock and the passive reference when
367 * done.
368 */
369bool grab_super_passive(struct super_block *sb)
370{
371 spin_lock(&sb_lock);
Al Viroa5166162011-12-12 22:53:00 -0500372 if (hlist_unhashed(&sb->s_instances)) {
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000373 spin_unlock(&sb_lock);
374 return false;
375 }
376
377 sb->s_count++;
378 spin_unlock(&sb_lock);
379
380 if (down_read_trylock(&sb->s_umount)) {
Al Virodabe0dc2012-01-03 21:01:29 -0500381 if (sb->s_root && (sb->s_flags & MS_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000382 return true;
383 up_read(&sb->s_umount);
384 }
385
386 put_super(sb);
387 return false;
388}
389
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390/**
391 * generic_shutdown_super - common helper for ->kill_sb()
392 * @sb: superblock to kill
393 *
394 * generic_shutdown_super() does all fs-independent work on superblock
395 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
396 * that need destruction out of superblock, call generic_shutdown_super()
397 * and release aforementioned objects. Note: dentries and inodes _are_
398 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700399 *
400 * Upon calling this function, the filesystem may no longer alter or
401 * rearrange the set of dentries belonging to this super_block, nor may it
402 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 */
404void generic_shutdown_super(struct super_block *sb)
405{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800406 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
David Howellsc636ebd2006-10-11 01:22:19 -0700408 if (sb->s_root) {
409 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200410 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 sb->s_flags &= ~MS_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800412
Al Viro63997e92010-10-25 20:49:35 -0400413 fsnotify_unmount_inodes(&sb->s_inodes);
414
415 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 if (sop->put_super)
418 sop->put_super(sb);
419
Al Viro63997e92010-10-25 20:49:35 -0400420 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800421 printk("VFS: Busy inodes after unmount of %s. "
422 "Self-destruct in 5 seconds. Have a nice day...\n",
423 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 }
426 spin_lock(&sb_lock);
427 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500428 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 spin_unlock(&sb_lock);
430 up_write(&sb->s_umount);
431}
432
433EXPORT_SYMBOL(generic_shutdown_super);
434
435/**
436 * sget - find or create a superblock
437 * @type: filesystem type superblock should belong to
438 * @test: comparison callback
439 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100440 * @flags: mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 * @data: argument to each of them
442 */
443struct super_block *sget(struct file_system_type *type,
444 int (*test)(struct super_block *,void *),
445 int (*set)(struct super_block *,void *),
David Howells9249e172012-06-25 12:55:37 +0100446 int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 void *data)
448{
449 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700450 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 int err;
452
453retry:
454 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700455 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800456 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700457 if (!test(old, data))
458 continue;
459 if (!grab_super(old))
460 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700461 if (s) {
462 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700463 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400464 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700465 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700466 return old;
467 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 }
469 if (!s) {
470 spin_unlock(&sb_lock);
David Howells9249e172012-06-25 12:55:37 +0100471 s = alloc_super(type, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 if (!s)
473 return ERR_PTR(-ENOMEM);
474 goto retry;
475 }
476
477 err = set(s, data);
478 if (err) {
479 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700480 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 destroy_super(s);
482 return ERR_PTR(err);
483 }
484 s->s_type = type;
485 strlcpy(s->s_id, type->name, sizeof(s->s_id));
486 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500487 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 spin_unlock(&sb_lock);
489 get_filesystem(type);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000490 register_shrinker(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 return s;
492}
493
494EXPORT_SYMBOL(sget);
495
496void drop_super(struct super_block *sb)
497{
498 up_read(&sb->s_umount);
499 put_super(sb);
500}
501
502EXPORT_SYMBOL(drop_super);
503
Christoph Hellwige5004752009-05-05 16:08:56 +0200504/**
Al Viro01a05b32010-03-23 06:06:58 -0400505 * iterate_supers - call function for all active superblocks
506 * @f: function to call
507 * @arg: argument to pass to it
508 *
509 * Scans the superblock list and calls given function, passing it
510 * locked superblock and given argument.
511 */
512void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
513{
Al Virodca33252010-07-25 02:31:46 +0400514 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400515
516 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400517 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500518 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400519 continue;
520 sb->s_count++;
521 spin_unlock(&sb_lock);
522
523 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500524 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400525 f(sb, arg);
526 up_read(&sb->s_umount);
527
528 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400529 if (p)
530 __put_super(p);
531 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400532 }
Al Virodca33252010-07-25 02:31:46 +0400533 if (p)
534 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400535 spin_unlock(&sb_lock);
536}
537
538/**
Al Viro43e15cd2011-06-03 20:16:57 -0400539 * iterate_supers_type - call function for superblocks of given type
540 * @type: fs type
541 * @f: function to call
542 * @arg: argument to pass to it
543 *
544 * Scans the superblock list and calls given function, passing it
545 * locked superblock and given argument.
546 */
547void iterate_supers_type(struct file_system_type *type,
548 void (*f)(struct super_block *, void *), void *arg)
549{
550 struct super_block *sb, *p = NULL;
551
552 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800553 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400554 sb->s_count++;
555 spin_unlock(&sb_lock);
556
557 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500558 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400559 f(sb, arg);
560 up_read(&sb->s_umount);
561
562 spin_lock(&sb_lock);
563 if (p)
564 __put_super(p);
565 p = sb;
566 }
567 if (p)
568 __put_super(p);
569 spin_unlock(&sb_lock);
570}
571
572EXPORT_SYMBOL(iterate_supers_type);
573
574/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 * get_super - get the superblock of a device
576 * @bdev: device to get the superblock for
577 *
578 * Scans the superblock list and finds the superblock of the file system
579 * mounted on the device given. %NULL is returned if no match is found.
580 */
581
Al Virodf40c012010-03-22 20:23:25 -0400582struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700584 struct super_block *sb;
585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 if (!bdev)
587 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700590rescan:
591 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500592 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400593 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700594 if (sb->s_bdev == bdev) {
595 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700597 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400598 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500599 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700600 return sb;
601 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400602 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700603 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400604 __put_super(sb);
605 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 }
607 }
608 spin_unlock(&sb_lock);
609 return NULL;
610}
611
612EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200613
614/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100615 * get_super_thawed - get thawed superblock of a device
616 * @bdev: device to get the superblock for
617 *
618 * Scans the superblock list and finds the superblock of the file system
619 * mounted on the device. The superblock is returned once it is thawed
620 * (or immediately if it was not frozen). %NULL is returned if no match
621 * is found.
622 */
623struct super_block *get_super_thawed(struct block_device *bdev)
624{
625 while (1) {
626 struct super_block *s = get_super(bdev);
Jan Kara5accdf82012-06-12 16:20:34 +0200627 if (!s || s->s_writers.frozen == SB_UNFROZEN)
Jan Kara6b6dc832012-02-10 11:03:00 +0100628 return s;
629 up_read(&s->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +0200630 wait_event(s->s_writers.wait_unfrozen,
631 s->s_writers.frozen == SB_UNFROZEN);
Jan Kara6b6dc832012-02-10 11:03:00 +0100632 put_super(s);
633 }
634}
635EXPORT_SYMBOL(get_super_thawed);
636
637/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200638 * get_active_super - get an active reference to the superblock of a device
639 * @bdev: device to get the superblock for
640 *
641 * Scans the superblock list and finds the superblock of the file system
642 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400643 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200644 */
645struct super_block *get_active_super(struct block_device *bdev)
646{
647 struct super_block *sb;
648
649 if (!bdev)
650 return NULL;
651
Al Viro14945832010-03-22 20:15:33 -0400652restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200653 spin_lock(&sb_lock);
654 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500655 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400656 continue;
Al Viro14945832010-03-22 20:15:33 -0400657 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400658 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400659 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400660 up_write(&sb->s_umount);
661 return sb;
Al Viro14945832010-03-22 20:15:33 -0400662 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200663 }
664 spin_unlock(&sb_lock);
665 return NULL;
666}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Al Virodf40c012010-03-22 20:23:25 -0400668struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700670 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700673rescan:
674 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500675 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400676 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700677 if (sb->s_dev == dev) {
678 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700680 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400681 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500682 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700683 return sb;
684 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400685 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700686 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400687 __put_super(sb);
688 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 }
690 }
691 spin_unlock(&sb_lock);
692 return NULL;
693}
694
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 * do_remount_sb - asks filesystem to change mount options.
697 * @sb: superblock in question
698 * @flags: numeric part of options
699 * @data: the rest of options
700 * @force: whether or not to force the change
701 *
702 * Alters the mount options of a mounted file system.
703 */
704int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
705{
706 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400707 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200708
Jan Kara5accdf82012-06-12 16:20:34 +0200709 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200710 return -EBUSY;
711
David Howells93614012006-09-30 20:45:40 +0200712#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
714 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200715#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200716
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 if (flags & MS_RDONLY)
718 acct_auto_close(sb);
719 shrink_dcache_sb(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200720 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Nick Piggind208bbd2009-12-21 16:28:53 -0800722 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800723
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 /* If we are remounting RDONLY and current sb is read/write,
725 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800726 if (remount_ro) {
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100727 if (force) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 mark_files_ro(sb);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100729 } else {
730 retval = sb_prepare_remount_readonly(sb);
731 if (retval)
732 return retval;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100733 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 }
735
736 if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 retval = sb->s_op->remount_fs(sb, &flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200738 if (retval) {
739 if (!force)
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100740 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200741 /* If forced remount, go ahead despite any errors */
742 WARN(1, "forced remount of a %s fs returned %i\n",
743 sb->s_type->name, retval);
744 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 }
746 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100747 /* Needs to be ordered wrt mnt_is_readonly() */
748 smp_wmb();
749 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400750
Nick Piggind208bbd2009-12-21 16:28:53 -0800751 /*
752 * Some filesystems modify their metadata via some other path than the
753 * bdev buffer cache (eg. use a private mapping, or directories in
754 * pagecache, etc). Also file data modifications go via their own
755 * mappings. So If we try to mount readonly then copy the filesystem
756 * from bdev, we could get stale data, so invalidate it to give a best
757 * effort at coherency.
758 */
759 if (remount_ro && sb->s_bdev)
760 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 return 0;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100762
763cancel_readonly:
764 sb->s_readonly_remount = 0;
765 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766}
767
Jens Axboea2a95372009-03-17 09:38:40 +0100768static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
Al Virodca33252010-07-25 02:31:46 +0400770 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
772 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400773 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500774 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400775 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 sb->s_count++;
777 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400778 down_write(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500779 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
780 !(sb->s_flags & MS_RDONLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 * What lock protects sb->s_flags??
783 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 }
Al Viro443b94b2009-05-05 23:48:50 -0400786 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400788 if (p)
789 __put_super(p);
790 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 }
Al Virodca33252010-07-25 02:31:46 +0400792 if (p)
793 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100795 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 printk("Emergency Remount complete\n");
797}
798
799void emergency_remount(void)
800{
Jens Axboea2a95372009-03-17 09:38:40 +0100801 struct work_struct *work;
802
803 work = kmalloc(sizeof(*work), GFP_ATOMIC);
804 if (work) {
805 INIT_WORK(work, do_emergency_remount);
806 schedule_work(work);
807 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808}
809
810/*
811 * Unnamed block devices are dummy devices used by virtual
812 * filesystems which don't use real block-devices. -- jrs
813 */
814
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400815static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Al Viroc63e09e2009-06-24 02:05:18 -0400817static int unnamed_dev_start = 0; /* don't bother trying below it */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Al Viro0ee5dc62011-07-07 15:44:25 -0400819int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820{
821 int dev;
822 int error;
823
824 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400825 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 return -ENOMEM;
827 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400828 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400829 if (!error)
830 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 spin_unlock(&unnamed_dev_lock);
832 if (error == -EAGAIN)
833 /* We raced and lost with another CPU. */
834 goto retry;
835 else if (error)
836 return -EAGAIN;
837
Tejun Heoe8c8d1b2013-02-27 17:05:04 -0800838 if (dev == (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400840 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400841 if (unnamed_dev_start > dev)
842 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 spin_unlock(&unnamed_dev_lock);
844 return -EMFILE;
845 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400846 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 return 0;
848}
Al Viro0ee5dc62011-07-07 15:44:25 -0400849EXPORT_SYMBOL(get_anon_bdev);
850
851void free_anon_bdev(dev_t dev)
852{
853 int slot = MINOR(dev);
854 spin_lock(&unnamed_dev_lock);
855 ida_remove(&unnamed_dev_ida, slot);
856 if (slot < unnamed_dev_start)
857 unnamed_dev_start = slot;
858 spin_unlock(&unnamed_dev_lock);
859}
860EXPORT_SYMBOL(free_anon_bdev);
861
862int set_anon_super(struct super_block *s, void *data)
863{
864 int error = get_anon_bdev(&s->s_dev);
865 if (!error)
866 s->s_bdi = &noop_backing_dev_info;
867 return error;
868}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
870EXPORT_SYMBOL(set_anon_super);
871
872void kill_anon_super(struct super_block *sb)
873{
Al Viro0ee5dc62011-07-07 15:44:25 -0400874 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400876 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877}
878
879EXPORT_SYMBOL(kill_anon_super);
880
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881void kill_litter_super(struct super_block *sb)
882{
883 if (sb->s_root)
884 d_genocide(sb->s_root);
885 kill_anon_super(sb);
886}
887
888EXPORT_SYMBOL(kill_litter_super);
889
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700890static int ns_test_super(struct super_block *sb, void *data)
891{
892 return sb->s_fs_info == data;
893}
894
895static int ns_set_super(struct super_block *sb, void *data)
896{
897 sb->s_fs_info = data;
898 return set_anon_super(sb, NULL);
899}
900
Al Viroceefda62010-07-26 13:16:50 +0400901struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
902 void *data, int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700903{
904 struct super_block *sb;
905
David Howells9249e172012-06-25 12:55:37 +0100906 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700907 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +0400908 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700909
910 if (!sb->s_root) {
911 int err;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700912 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
913 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400914 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +0400915 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700916 }
917
918 sb->s_flags |= MS_ACTIVE;
919 }
920
Al Viroceefda62010-07-26 13:16:50 +0400921 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700922}
923
Al Viroceefda62010-07-26 13:16:50 +0400924EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700925
David Howells93614012006-09-30 20:45:40 +0200926#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927static int set_bdev_super(struct super_block *s, void *data)
928{
929 s->s_bdev = data;
930 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +0200931
932 /*
933 * We set the bdi here to the queue backing, file systems can
934 * overwrite this in ->fill_super()
935 */
936 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 return 0;
938}
939
940static int test_bdev_super(struct super_block *s, void *data)
941{
942 return (void *)s->s_bdev == data;
943}
944
Al Viro152a0832010-07-25 00:46:55 +0400945struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +0400947 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948{
949 struct block_device *bdev;
950 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +0100951 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 int error = 0;
953
Al Viro30c40d22008-02-22 19:50:45 -0500954 if (!(flags & MS_RDONLY))
955 mode |= FMODE_WRITE;
956
Tejun Heod4d77622010-11-13 11:55:18 +0100957 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +0400959 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
961 /*
962 * once the super is inserted into the list by sget, s_umount
963 * will protect the lockfs code from trying to start a snapshot
964 * while we are mounting
965 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200966 mutex_lock(&bdev->bd_fsfreeze_mutex);
967 if (bdev->bd_fsfreeze_count > 0) {
968 mutex_unlock(&bdev->bd_fsfreeze_mutex);
969 error = -EBUSY;
970 goto error_bdev;
971 }
David Howells9249e172012-06-25 12:55:37 +0100972 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
973 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200974 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -0700976 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
978 if (s->s_root) {
979 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400980 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -0700981 error = -EBUSY;
982 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 }
David Howells454e2392006-06-23 02:02:57 -0700984
Tejun Heo4f331f02010-07-20 15:18:07 -0700985 /*
986 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +0100987 * __invalidate_device(). blkdev_put() acquires
988 * bd_mutex and can't be called under s_umount. Drop
989 * s_umount temporarily. This is safe as we're
990 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -0700991 */
992 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +0100993 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -0700994 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 } else {
996 char b[BDEVNAME_SIZE];
997
Al Viro30c40d22008-02-22 19:50:45 -0500998 s->s_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
Pekka Enberge78c9a02006-01-08 01:03:39 -08001000 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001001 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001003 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001004 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001005 }
David Howells454e2392006-06-23 02:02:57 -07001006
1007 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001008 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 }
1010
Al Viro152a0832010-07-25 00:46:55 +04001011 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012
David Howells454e2392006-06-23 02:02:57 -07001013error_s:
1014 error = PTR_ERR(s);
1015error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001016 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001017error:
Al Viro152a0832010-07-25 00:46:55 +04001018 return ERR_PTR(error);
1019}
1020EXPORT_SYMBOL(mount_bdev);
1021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022void kill_block_super(struct super_block *sb)
1023{
1024 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001025 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001027 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 generic_shutdown_super(sb);
1029 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001030 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001031 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032}
1033
1034EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001035#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
Al Viro3c26ff62010-07-25 11:46:36 +04001037struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001039 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040{
1041 int error;
David Howells9249e172012-06-25 12:55:37 +01001042 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
1044 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001045 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001047 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001049 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001050 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 }
1052 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001053 return dget(s->s_root);
1054}
1055EXPORT_SYMBOL(mount_nodev);
1056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057static int compare_single(struct super_block *s, void *p)
1058{
1059 return 1;
1060}
1061
Al Virofc14f2f2010-07-25 01:48:30 +04001062struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001064 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065{
1066 struct super_block *s;
1067 int error;
1068
David Howells9249e172012-06-25 12:55:37 +01001069 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001071 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 if (!s->s_root) {
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001073 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001075 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001076 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 }
1078 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001079 } else {
1080 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 }
Al Virofc14f2f2010-07-25 01:48:30 +04001082 return dget(s->s_root);
1083}
1084EXPORT_SYMBOL(mount_single);
1085
Al Viro9d412a42011-03-17 22:08:28 -04001086struct dentry *
1087mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088{
Al Viroc96e41e2010-07-25 00:17:56 +04001089 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001090 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001092 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001093
Eric Parise0007522008-03-05 10:31:54 -05001094 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001096 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001097 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098
Eric Parise0007522008-03-05 10:31:54 -05001099 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001100 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 }
1103
Al Viro1a102ff2011-03-16 09:07:58 -04001104 root = type->mount(type, flags, name, data);
1105 if (IS_ERR(root)) {
1106 error = PTR_ERR(root);
1107 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001108 }
Al Viro9d412a42011-03-17 22:08:28 -04001109 sb = root->d_sb;
1110 BUG_ON(!sb);
1111 WARN_ON(!sb->s_bdi);
Linus Torvalds6c510382011-03-24 10:16:26 -07001112 WARN_ON(sb->s_bdi == &default_backing_dev_info);
Al Viro9d412a42011-03-17 22:08:28 -04001113 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001114
Al Viro9d412a42011-03-17 22:08:28 -04001115 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001116 if (error)
1117 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001118
Jeff Layton42cb56a2009-09-18 13:05:53 -07001119 /*
1120 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1121 * but s_maxbytes was an unsigned long long for many releases. Throw
1122 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001123 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001124 */
Al Viro9d412a42011-03-17 22:08:28 -04001125 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1126 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001127
Al Viro9d412a42011-03-17 22:08:28 -04001128 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001129 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001130 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001132 dput(root);
1133 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134out_free_secdata:
1135 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136out:
David Howells454e2392006-06-23 02:02:57 -07001137 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138}
1139
Jan Kara5accdf82012-06-12 16:20:34 +02001140/*
1141 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1142 * instead.
1143 */
1144void __sb_end_write(struct super_block *sb, int level)
1145{
1146 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1147 /*
1148 * Make sure s_writers are updated before we wake up waiters in
1149 * freeze_super().
1150 */
1151 smp_mb();
1152 if (waitqueue_active(&sb->s_writers.wait))
1153 wake_up(&sb->s_writers.wait);
1154 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1155}
1156EXPORT_SYMBOL(__sb_end_write);
1157
1158#ifdef CONFIG_LOCKDEP
1159/*
1160 * We want lockdep to tell us about possible deadlocks with freezing but
1161 * it's it bit tricky to properly instrument it. Getting a freeze protection
1162 * works as getting a read lock but there are subtle problems. XFS for example
1163 * gets freeze protection on internal level twice in some cases, which is OK
1164 * only because we already hold a freeze protection also on higher level. Due
1165 * to these cases we have to tell lockdep we are doing trylock when we
1166 * already hold a freeze protection for a higher freeze level.
1167 */
1168static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1169 unsigned long ip)
1170{
1171 int i;
1172
1173 if (!trylock) {
1174 for (i = 0; i < level - 1; i++)
1175 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1176 trylock = true;
1177 break;
1178 }
1179 }
1180 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1181}
1182#endif
1183
1184/*
1185 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1186 * instead.
1187 */
1188int __sb_start_write(struct super_block *sb, int level, bool wait)
1189{
1190retry:
1191 if (unlikely(sb->s_writers.frozen >= level)) {
1192 if (!wait)
1193 return 0;
1194 wait_event(sb->s_writers.wait_unfrozen,
1195 sb->s_writers.frozen < level);
1196 }
1197
1198#ifdef CONFIG_LOCKDEP
1199 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1200#endif
1201 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1202 /*
1203 * Make sure counter is updated before we check for frozen.
1204 * freeze_super() first sets frozen and then checks the counter.
1205 */
1206 smp_mb();
1207 if (unlikely(sb->s_writers.frozen >= level)) {
1208 __sb_end_write(sb, level);
1209 goto retry;
1210 }
1211 return 1;
1212}
1213EXPORT_SYMBOL(__sb_start_write);
1214
1215/**
1216 * sb_wait_write - wait until all writers to given file system finish
1217 * @sb: the super for which we wait
1218 * @level: type of writers we wait for (normal vs page fault)
1219 *
1220 * This function waits until there are no writers of given type to given file
1221 * system. Caller of this function should make sure there can be no new writers
1222 * of type @level before calling this function. Otherwise this function can
1223 * livelock.
1224 */
1225static void sb_wait_write(struct super_block *sb, int level)
1226{
1227 s64 writers;
1228
1229 /*
1230 * We just cycle-through lockdep here so that it does not complain
1231 * about returning with lock to userspace
1232 */
1233 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1234 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1235
1236 do {
1237 DEFINE_WAIT(wait);
1238
1239 /*
1240 * We use a barrier in prepare_to_wait() to separate setting
1241 * of frozen and checking of the counter
1242 */
1243 prepare_to_wait(&sb->s_writers.wait, &wait,
1244 TASK_UNINTERRUPTIBLE);
1245
1246 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1247 if (writers)
1248 schedule();
1249
1250 finish_wait(&sb->s_writers.wait, &wait);
1251 } while (writers);
1252}
1253
Josef Bacik18e9e512010-03-23 10:34:56 -04001254/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001255 * freeze_super - lock the filesystem and force it into a consistent state
1256 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001257 *
1258 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1259 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1260 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001261 *
1262 * During this function, sb->s_writers.frozen goes through these values:
1263 *
1264 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1265 *
1266 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1267 * writes should be blocked, though page faults are still allowed. We wait for
1268 * all writes to complete and then proceed to the next stage.
1269 *
1270 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1271 * but internal fs threads can still modify the filesystem (although they
1272 * should not dirty new pages or inodes), writeback can run etc. After waiting
1273 * for all running page faults we sync the filesystem which will clean all
1274 * dirty pages and inodes (no new dirty pages or inodes can be created when
1275 * sync is running).
1276 *
1277 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1278 * modification are blocked (e.g. XFS preallocation truncation on inode
1279 * reclaim). This is usually implemented by blocking new transactions for
1280 * filesystems that have them and need this additional guard. After all
1281 * internal writers are finished we call ->freeze_fs() to finish filesystem
1282 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1283 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1284 *
1285 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001286 */
1287int freeze_super(struct super_block *sb)
1288{
1289 int ret;
1290
1291 atomic_inc(&sb->s_active);
1292 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001293 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001294 deactivate_locked_super(sb);
1295 return -EBUSY;
1296 }
1297
Al Virodabe0dc2012-01-03 21:01:29 -05001298 if (!(sb->s_flags & MS_BORN)) {
1299 up_write(&sb->s_umount);
1300 return 0; /* sic - it's "nothing to do" */
1301 }
1302
Josef Bacik18e9e512010-03-23 10:34:56 -04001303 if (sb->s_flags & MS_RDONLY) {
Jan Kara5accdf82012-06-12 16:20:34 +02001304 /* Nothing to do really... */
1305 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001306 up_write(&sb->s_umount);
1307 return 0;
1308 }
1309
Jan Kara5accdf82012-06-12 16:20:34 +02001310 /* From now on, no new normal writers can start */
1311 sb->s_writers.frozen = SB_FREEZE_WRITE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001312 smp_wmb();
1313
Jan Kara5accdf82012-06-12 16:20:34 +02001314 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1315 up_write(&sb->s_umount);
1316
1317 sb_wait_write(sb, SB_FREEZE_WRITE);
1318
1319 /* Now we go and block page faults... */
1320 down_write(&sb->s_umount);
1321 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1322 smp_wmb();
1323
1324 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1325
1326 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001327 sync_filesystem(sb);
1328
Jan Kara5accdf82012-06-12 16:20:34 +02001329 /* Now wait for internal filesystem counter */
1330 sb->s_writers.frozen = SB_FREEZE_FS;
Josef Bacik18e9e512010-03-23 10:34:56 -04001331 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001332 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001333
Josef Bacik18e9e512010-03-23 10:34:56 -04001334 if (sb->s_op->freeze_fs) {
1335 ret = sb->s_op->freeze_fs(sb);
1336 if (ret) {
1337 printk(KERN_ERR
1338 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001339 sb->s_writers.frozen = SB_UNFROZEN;
Kazuya Mioe1616302011-12-01 16:51:07 +09001340 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001341 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001342 deactivate_locked_super(sb);
1343 return ret;
1344 }
1345 }
Jan Kara5accdf82012-06-12 16:20:34 +02001346 /*
1347 * This is just for debugging purposes so that fs can warn if it
1348 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1349 */
1350 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001351 up_write(&sb->s_umount);
1352 return 0;
1353}
1354EXPORT_SYMBOL(freeze_super);
1355
1356/**
1357 * thaw_super -- unlock filesystem
1358 * @sb: the super to thaw
1359 *
1360 * Unlocks the filesystem and marks it writeable again after freeze_super().
1361 */
1362int thaw_super(struct super_block *sb)
1363{
1364 int error;
1365
1366 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001367 if (sb->s_writers.frozen == SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001368 up_write(&sb->s_umount);
1369 return -EINVAL;
1370 }
1371
1372 if (sb->s_flags & MS_RDONLY)
1373 goto out;
1374
1375 if (sb->s_op->unfreeze_fs) {
1376 error = sb->s_op->unfreeze_fs(sb);
1377 if (error) {
1378 printk(KERN_ERR
1379 "VFS:Filesystem thaw failed\n");
Josef Bacik18e9e512010-03-23 10:34:56 -04001380 up_write(&sb->s_umount);
1381 return error;
1382 }
1383 }
1384
1385out:
Jan Kara5accdf82012-06-12 16:20:34 +02001386 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001387 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001388 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001389 deactivate_locked_super(sb);
1390
1391 return 0;
1392}
1393EXPORT_SYMBOL(thaw_super);