blob: 743bb7118053071381d8e9bc0f09e634d2264061 [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
Al Viro7eb5e882013-10-01 15:09:58 -0400132/**
133 * destroy_super - frees a superblock
134 * @s: superblock to free
135 *
136 * Frees a superblock.
137 */
138static void destroy_super(struct super_block *s)
Jan Kara5accdf82012-06-12 16:20:34 +0200139{
140 int i;
Al Viro7eb5e882013-10-01 15:09:58 -0400141 list_lru_destroy(&s->s_dentry_lru);
142 list_lru_destroy(&s->s_inode_lru);
143#ifdef CONFIG_SMP
144 free_percpu(s->s_files);
145#endif
Jan Kara5accdf82012-06-12 16:20:34 +0200146 for (i = 0; i < SB_FREEZE_LEVELS; i++)
147 percpu_counter_destroy(&s->s_writers.counter[i]);
Al Viro7eb5e882013-10-01 15:09:58 -0400148 security_sb_free(s);
149 WARN_ON(!list_empty(&s->s_mounts));
150 kfree(s->s_subtype);
151 kfree(s->s_options);
Al Viroe2fec7c2013-10-04 17:06:56 -0400152 kfree_rcu(s, rcu);
Jan Kara5accdf82012-06-12 16:20:34 +0200153}
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155/**
156 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700157 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100158 * @flags: the mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 *
160 * Allocates and initializes a new &struct super_block. alloc_super()
161 * returns a pointer new superblock or %NULL if allocation had failed.
162 */
David Howells9249e172012-06-25 12:55:37 +0100163static struct super_block *alloc_super(struct file_system_type *type, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800165 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700166 static const struct super_operations default_op;
Al Viro7eb5e882013-10-01 15:09:58 -0400167 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Al Viro7eb5e882013-10-01 15:09:58 -0400169 if (!s)
170 return NULL;
171
172 if (security_sb_alloc(s))
173 goto fail;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200174
Nick Piggin6416ccb2010-08-18 04:37:38 +1000175#ifdef CONFIG_SMP
Al Viro7eb5e882013-10-01 15:09:58 -0400176 s->s_files = alloc_percpu(struct list_head);
177 if (!s->s_files)
178 goto fail;
179 for_each_possible_cpu(i)
180 INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
Nick Piggin6416ccb2010-08-18 04:37:38 +1000181#else
Al Viro7eb5e882013-10-01 15:09:58 -0400182 INIT_LIST_HEAD(&s->s_files);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000183#endif
Al Viro7eb5e882013-10-01 15:09:58 -0400184 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
185 if (percpu_counter_init(&s->s_writers.counter[i], 0) < 0)
186 goto fail;
187 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
188 &type->s_writers_key[i], 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 }
Al Viro7eb5e882013-10-01 15:09:58 -0400190 init_waitqueue_head(&s->s_writers.wait);
191 init_waitqueue_head(&s->s_writers.wait_unfrozen);
192 s->s_flags = flags;
193 s->s_bdi = &default_backing_dev_info;
194 INIT_HLIST_NODE(&s->s_instances);
195 INIT_HLIST_BL_HEAD(&s->s_anon);
196 INIT_LIST_HEAD(&s->s_inodes);
197
198 if (list_lru_init(&s->s_dentry_lru))
199 goto fail;
200 if (list_lru_init(&s->s_inode_lru))
201 goto fail;
202
203 INIT_LIST_HEAD(&s->s_mounts);
204 init_rwsem(&s->s_umount);
205 lockdep_set_class(&s->s_umount, &type->s_umount_key);
206 /*
207 * sget() can have s_umount recursion.
208 *
209 * When it cannot find a suitable sb, it allocates a new
210 * one (this one), and tries again to find a suitable old
211 * one.
212 *
213 * In case that succeeds, it will acquire the s_umount
214 * lock of the old one. Since these are clearly distrinct
215 * locks, and this object isn't exposed yet, there's no
216 * risk of deadlocks.
217 *
218 * Annotate this by putting this lock in a different
219 * subclass.
220 */
221 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
222 s->s_count = 1;
223 atomic_set(&s->s_active, 1);
224 mutex_init(&s->s_vfs_rename_mutex);
225 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
226 mutex_init(&s->s_dquot.dqio_mutex);
227 mutex_init(&s->s_dquot.dqonoff_mutex);
228 init_rwsem(&s->s_dquot.dqptr_sem);
229 s->s_maxbytes = MAX_NON_LFS;
230 s->s_op = &default_op;
231 s->s_time_gran = 1000000000;
232 s->cleancache_poolid = -1;
233
234 s->s_shrink.seeks = DEFAULT_SEEKS;
235 s->s_shrink.scan_objects = super_cache_scan;
236 s->s_shrink.count_objects = super_cache_count;
237 s->s_shrink.batch = 1024;
238 s->s_shrink.flags = SHRINKER_NUMA_AWARE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000240
Al Viro7eb5e882013-10-01 15:09:58 -0400241fail:
242 destroy_super(s);
243 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244}
245
246/* Superblock refcounting */
247
248/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400249 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 */
Al Virof47ec3f2011-11-21 21:15:42 -0500251static void __put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 if (!--sb->s_count) {
Al Viro551de6f2010-03-22 19:36:35 -0400254 list_del_init(&sb->s_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 destroy_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
258
259/**
260 * put_super - drop a temporary reference to superblock
261 * @sb: superblock in question
262 *
263 * Drops a temporary reference, frees superblock if there's no
264 * references left.
265 */
Al Virof47ec3f2011-11-21 21:15:42 -0500266static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267{
268 spin_lock(&sb_lock);
269 __put_super(sb);
270 spin_unlock(&sb_lock);
271}
272
273
274/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400275 * deactivate_locked_super - drop an active reference to superblock
276 * @s: superblock to deactivate
277 *
Al Viro1712ac82010-03-22 15:22:31 -0400278 * Drops an active reference to superblock, converting it into a temprory
279 * one if there is no other active references left. In that case we
280 * tell fs driver to shut it down and drop the temporary reference we
281 * had just acquired.
282 *
283 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400284 */
285void deactivate_locked_super(struct super_block *s)
286{
287 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400288 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400289 cleancache_invalidate_fs(s);
Al Viro74dbbdd2009-05-06 01:07:50 -0400290 fs->kill_sb(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000291
292 /* caches are now gone, we can safely kill the shrinker now */
293 unregister_shrinker(&s->s_shrink);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000294
Al Viro74dbbdd2009-05-06 01:07:50 -0400295 put_filesystem(fs);
296 put_super(s);
297 } else {
298 up_write(&s->s_umount);
299 }
300}
301
302EXPORT_SYMBOL(deactivate_locked_super);
303
304/**
Al Viro1712ac82010-03-22 15:22:31 -0400305 * deactivate_super - drop an active reference to superblock
306 * @s: superblock to deactivate
307 *
308 * Variant of deactivate_locked_super(), except that superblock is *not*
309 * locked by caller. If we are going to drop the final active reference,
310 * lock will be acquired prior to that.
311 */
312void deactivate_super(struct super_block *s)
313{
314 if (!atomic_add_unless(&s->s_active, -1, 1)) {
315 down_write(&s->s_umount);
316 deactivate_locked_super(s);
317 }
318}
319
320EXPORT_SYMBOL(deactivate_super);
321
322/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 * grab_super - acquire an active reference
324 * @s: reference we are trying to make active
325 *
326 * Tries to acquire an active reference. grab_super() is used when we
327 * had just found a superblock in super_blocks or fs_type->fs_supers
328 * and want to turn it into a full-blown active reference. grab_super()
329 * is called with sb_lock held and drops it. Returns 1 in case of
330 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400331 * dying when grab_super() had been called). Note that this is only
332 * called for superblocks not in rundown mode (== ones still on ->fs_supers
333 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700335static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336{
337 s->s_count++;
338 spin_unlock(&sb_lock);
339 down_write(&s->s_umount);
Al Viroacfec9a2013-07-20 03:13:55 +0400340 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
341 put_super(s);
342 return 1;
343 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 up_write(&s->s_umount);
345 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 return 0;
347}
348
David Howellscf9a2ae2006-08-29 19:05:54 +0100349/*
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000350 * grab_super_passive - acquire a passive reference
Wanpeng Li331cbde2012-06-09 11:10:55 +0800351 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000352 *
353 * Tries to acquire a passive reference. This is used in places where we
354 * cannot take an active reference but we need to ensure that the
355 * superblock does not go away while we are working on it. It returns
356 * false if a reference was not gained, and returns true with the s_umount
357 * lock held in read mode if a reference is gained. On successful return,
358 * the caller must drop the s_umount lock and the passive reference when
359 * done.
360 */
361bool grab_super_passive(struct super_block *sb)
362{
363 spin_lock(&sb_lock);
Al Viroa5166162011-12-12 22:53:00 -0500364 if (hlist_unhashed(&sb->s_instances)) {
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000365 spin_unlock(&sb_lock);
366 return false;
367 }
368
369 sb->s_count++;
370 spin_unlock(&sb_lock);
371
372 if (down_read_trylock(&sb->s_umount)) {
Al Virodabe0dc2012-01-03 21:01:29 -0500373 if (sb->s_root && (sb->s_flags & MS_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000374 return true;
375 up_read(&sb->s_umount);
376 }
377
378 put_super(sb);
379 return false;
380}
381
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382/**
383 * generic_shutdown_super - common helper for ->kill_sb()
384 * @sb: superblock to kill
385 *
386 * generic_shutdown_super() does all fs-independent work on superblock
387 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
388 * that need destruction out of superblock, call generic_shutdown_super()
389 * and release aforementioned objects. Note: dentries and inodes _are_
390 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700391 *
392 * Upon calling this function, the filesystem may no longer alter or
393 * rearrange the set of dentries belonging to this super_block, nor may it
394 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 */
396void generic_shutdown_super(struct super_block *sb)
397{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800398 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
David Howellsc636ebd2006-10-11 01:22:19 -0700400 if (sb->s_root) {
401 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200402 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 sb->s_flags &= ~MS_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800404
Al Viro63997e92010-10-25 20:49:35 -0400405 fsnotify_unmount_inodes(&sb->s_inodes);
406
407 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200409 if (sb->s_dio_done_wq) {
410 destroy_workqueue(sb->s_dio_done_wq);
411 sb->s_dio_done_wq = NULL;
412 }
413
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 if (sop->put_super)
415 sop->put_super(sb);
416
Al Viro63997e92010-10-25 20:49:35 -0400417 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800418 printk("VFS: Busy inodes after unmount of %s. "
419 "Self-destruct in 5 seconds. Have a nice day...\n",
420 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 }
423 spin_lock(&sb_lock);
424 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500425 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 spin_unlock(&sb_lock);
427 up_write(&sb->s_umount);
428}
429
430EXPORT_SYMBOL(generic_shutdown_super);
431
432/**
433 * sget - find or create a superblock
434 * @type: filesystem type superblock should belong to
435 * @test: comparison callback
436 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100437 * @flags: mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 * @data: argument to each of them
439 */
440struct super_block *sget(struct file_system_type *type,
441 int (*test)(struct super_block *,void *),
442 int (*set)(struct super_block *,void *),
David Howells9249e172012-06-25 12:55:37 +0100443 int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 void *data)
445{
446 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700447 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 int err;
449
450retry:
451 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700452 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800453 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700454 if (!test(old, data))
455 continue;
456 if (!grab_super(old))
457 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700458 if (s) {
459 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700460 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400461 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700462 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700463 return old;
464 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 }
466 if (!s) {
467 spin_unlock(&sb_lock);
David Howells9249e172012-06-25 12:55:37 +0100468 s = alloc_super(type, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 if (!s)
470 return ERR_PTR(-ENOMEM);
471 goto retry;
472 }
473
474 err = set(s, data);
475 if (err) {
476 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700477 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 destroy_super(s);
479 return ERR_PTR(err);
480 }
481 s->s_type = type;
482 strlcpy(s->s_id, type->name, sizeof(s->s_id));
483 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500484 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 spin_unlock(&sb_lock);
486 get_filesystem(type);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000487 register_shrinker(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 return s;
489}
490
491EXPORT_SYMBOL(sget);
492
493void drop_super(struct super_block *sb)
494{
495 up_read(&sb->s_umount);
496 put_super(sb);
497}
498
499EXPORT_SYMBOL(drop_super);
500
Christoph Hellwige5004752009-05-05 16:08:56 +0200501/**
Al Viro01a05b32010-03-23 06:06:58 -0400502 * iterate_supers - call function for all active superblocks
503 * @f: function to call
504 * @arg: argument to pass to it
505 *
506 * Scans the superblock list and calls given function, passing it
507 * locked superblock and given argument.
508 */
509void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
510{
Al Virodca33252010-07-25 02:31:46 +0400511 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400512
513 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400514 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500515 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400516 continue;
517 sb->s_count++;
518 spin_unlock(&sb_lock);
519
520 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500521 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400522 f(sb, arg);
523 up_read(&sb->s_umount);
524
525 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400526 if (p)
527 __put_super(p);
528 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400529 }
Al Virodca33252010-07-25 02:31:46 +0400530 if (p)
531 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400532 spin_unlock(&sb_lock);
533}
534
535/**
Al Viro43e15cd2011-06-03 20:16:57 -0400536 * iterate_supers_type - call function for superblocks of given type
537 * @type: fs type
538 * @f: function to call
539 * @arg: argument to pass to it
540 *
541 * Scans the superblock list and calls given function, passing it
542 * locked superblock and given argument.
543 */
544void iterate_supers_type(struct file_system_type *type,
545 void (*f)(struct super_block *, void *), void *arg)
546{
547 struct super_block *sb, *p = NULL;
548
549 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800550 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400551 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 Viro43e15cd2011-06-03 20:16:57 -0400556 f(sb, arg);
557 up_read(&sb->s_umount);
558
559 spin_lock(&sb_lock);
560 if (p)
561 __put_super(p);
562 p = sb;
563 }
564 if (p)
565 __put_super(p);
566 spin_unlock(&sb_lock);
567}
568
569EXPORT_SYMBOL(iterate_supers_type);
570
571/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 * get_super - get the superblock of a device
573 * @bdev: device to get the superblock for
574 *
575 * Scans the superblock list and finds the superblock of the file system
576 * mounted on the device given. %NULL is returned if no match is found.
577 */
578
Al Virodf40c012010-03-22 20:23:25 -0400579struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700581 struct super_block *sb;
582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 if (!bdev)
584 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700587rescan:
588 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500589 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400590 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700591 if (sb->s_bdev == bdev) {
592 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700594 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400595 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500596 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700597 return sb;
598 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400599 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700600 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400601 __put_super(sb);
602 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 }
604 }
605 spin_unlock(&sb_lock);
606 return NULL;
607}
608
609EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200610
611/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100612 * get_super_thawed - get thawed superblock of a device
613 * @bdev: device to get the superblock for
614 *
615 * Scans the superblock list and finds the superblock of the file system
616 * mounted on the device. The superblock is returned once it is thawed
617 * (or immediately if it was not frozen). %NULL is returned if no match
618 * is found.
619 */
620struct super_block *get_super_thawed(struct block_device *bdev)
621{
622 while (1) {
623 struct super_block *s = get_super(bdev);
Jan Kara5accdf82012-06-12 16:20:34 +0200624 if (!s || s->s_writers.frozen == SB_UNFROZEN)
Jan Kara6b6dc832012-02-10 11:03:00 +0100625 return s;
626 up_read(&s->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +0200627 wait_event(s->s_writers.wait_unfrozen,
628 s->s_writers.frozen == SB_UNFROZEN);
Jan Kara6b6dc832012-02-10 11:03:00 +0100629 put_super(s);
630 }
631}
632EXPORT_SYMBOL(get_super_thawed);
633
634/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200635 * get_active_super - get an active reference to the superblock of a device
636 * @bdev: device to get the superblock for
637 *
638 * Scans the superblock list and finds the superblock of the file system
639 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400640 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200641 */
642struct super_block *get_active_super(struct block_device *bdev)
643{
644 struct super_block *sb;
645
646 if (!bdev)
647 return NULL;
648
Al Viro14945832010-03-22 20:15:33 -0400649restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200650 spin_lock(&sb_lock);
651 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500652 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400653 continue;
Al Viro14945832010-03-22 20:15:33 -0400654 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400655 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400656 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400657 up_write(&sb->s_umount);
658 return sb;
Al Viro14945832010-03-22 20:15:33 -0400659 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200660 }
661 spin_unlock(&sb_lock);
662 return NULL;
663}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Al Virodf40c012010-03-22 20:23:25 -0400665struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700667 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700670rescan:
671 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500672 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400673 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700674 if (sb->s_dev == dev) {
675 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700677 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400678 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500679 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700680 return sb;
681 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400682 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700683 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400684 __put_super(sb);
685 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 }
687 }
688 spin_unlock(&sb_lock);
689 return NULL;
690}
691
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 * do_remount_sb - asks filesystem to change mount options.
694 * @sb: superblock in question
695 * @flags: numeric part of options
696 * @data: the rest of options
697 * @force: whether or not to force the change
698 *
699 * Alters the mount options of a mounted file system.
700 */
701int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
702{
703 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400704 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200705
Jan Kara5accdf82012-06-12 16:20:34 +0200706 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200707 return -EBUSY;
708
David Howells93614012006-09-30 20:45:40 +0200709#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
711 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200712#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200713
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 if (flags & MS_RDONLY)
715 acct_auto_close(sb);
716 shrink_dcache_sb(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200717 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Nick Piggind208bbd2009-12-21 16:28:53 -0800719 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800720
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 /* If we are remounting RDONLY and current sb is read/write,
722 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800723 if (remount_ro) {
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100724 if (force) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 mark_files_ro(sb);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100726 } else {
727 retval = sb_prepare_remount_readonly(sb);
728 if (retval)
729 return retval;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100730 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 }
732
733 if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 retval = sb->s_op->remount_fs(sb, &flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200735 if (retval) {
736 if (!force)
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100737 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200738 /* If forced remount, go ahead despite any errors */
739 WARN(1, "forced remount of a %s fs returned %i\n",
740 sb->s_type->name, retval);
741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 }
743 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100744 /* Needs to be ordered wrt mnt_is_readonly() */
745 smp_wmb();
746 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400747
Nick Piggind208bbd2009-12-21 16:28:53 -0800748 /*
749 * Some filesystems modify their metadata via some other path than the
750 * bdev buffer cache (eg. use a private mapping, or directories in
751 * pagecache, etc). Also file data modifications go via their own
752 * mappings. So If we try to mount readonly then copy the filesystem
753 * from bdev, we could get stale data, so invalidate it to give a best
754 * effort at coherency.
755 */
756 if (remount_ro && sb->s_bdev)
757 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 return 0;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100759
760cancel_readonly:
761 sb->s_readonly_remount = 0;
762 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763}
764
Jens Axboea2a95372009-03-17 09:38:40 +0100765static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766{
Al Virodca33252010-07-25 02:31:46 +0400767 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
769 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400770 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500771 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400772 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 sb->s_count++;
774 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400775 down_write(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500776 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
777 !(sb->s_flags & MS_RDONLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 * What lock protects sb->s_flags??
780 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 }
Al Viro443b94b2009-05-05 23:48:50 -0400783 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400785 if (p)
786 __put_super(p);
787 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
Al Virodca33252010-07-25 02:31:46 +0400789 if (p)
790 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100792 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 printk("Emergency Remount complete\n");
794}
795
796void emergency_remount(void)
797{
Jens Axboea2a95372009-03-17 09:38:40 +0100798 struct work_struct *work;
799
800 work = kmalloc(sizeof(*work), GFP_ATOMIC);
801 if (work) {
802 INIT_WORK(work, do_emergency_remount);
803 schedule_work(work);
804 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805}
806
807/*
808 * Unnamed block devices are dummy devices used by virtual
809 * filesystems which don't use real block-devices. -- jrs
810 */
811
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400812static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Al Viroc63e09e2009-06-24 02:05:18 -0400814static int unnamed_dev_start = 0; /* don't bother trying below it */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
Al Viro0ee5dc62011-07-07 15:44:25 -0400816int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817{
818 int dev;
819 int error;
820
821 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400822 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 return -ENOMEM;
824 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400825 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400826 if (!error)
827 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 spin_unlock(&unnamed_dev_lock);
829 if (error == -EAGAIN)
830 /* We raced and lost with another CPU. */
831 goto retry;
832 else if (error)
833 return -EAGAIN;
834
Tejun Heoe8c8d1b2013-02-27 17:05:04 -0800835 if (dev == (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400837 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400838 if (unnamed_dev_start > dev)
839 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 spin_unlock(&unnamed_dev_lock);
841 return -EMFILE;
842 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400843 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 return 0;
845}
Al Viro0ee5dc62011-07-07 15:44:25 -0400846EXPORT_SYMBOL(get_anon_bdev);
847
848void free_anon_bdev(dev_t dev)
849{
850 int slot = MINOR(dev);
851 spin_lock(&unnamed_dev_lock);
852 ida_remove(&unnamed_dev_ida, slot);
853 if (slot < unnamed_dev_start)
854 unnamed_dev_start = slot;
855 spin_unlock(&unnamed_dev_lock);
856}
857EXPORT_SYMBOL(free_anon_bdev);
858
859int set_anon_super(struct super_block *s, void *data)
860{
861 int error = get_anon_bdev(&s->s_dev);
862 if (!error)
863 s->s_bdi = &noop_backing_dev_info;
864 return error;
865}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866
867EXPORT_SYMBOL(set_anon_super);
868
869void kill_anon_super(struct super_block *sb)
870{
Al Viro0ee5dc62011-07-07 15:44:25 -0400871 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400873 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874}
875
876EXPORT_SYMBOL(kill_anon_super);
877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878void kill_litter_super(struct super_block *sb)
879{
880 if (sb->s_root)
881 d_genocide(sb->s_root);
882 kill_anon_super(sb);
883}
884
885EXPORT_SYMBOL(kill_litter_super);
886
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700887static int ns_test_super(struct super_block *sb, void *data)
888{
889 return sb->s_fs_info == data;
890}
891
892static int ns_set_super(struct super_block *sb, void *data)
893{
894 sb->s_fs_info = data;
895 return set_anon_super(sb, NULL);
896}
897
Al Viroceefda62010-07-26 13:16:50 +0400898struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
899 void *data, int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700900{
901 struct super_block *sb;
902
David Howells9249e172012-06-25 12:55:37 +0100903 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700904 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +0400905 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700906
907 if (!sb->s_root) {
908 int err;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700909 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
910 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400911 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +0400912 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700913 }
914
915 sb->s_flags |= MS_ACTIVE;
916 }
917
Al Viroceefda62010-07-26 13:16:50 +0400918 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700919}
920
Al Viroceefda62010-07-26 13:16:50 +0400921EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700922
David Howells93614012006-09-30 20:45:40 +0200923#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924static int set_bdev_super(struct super_block *s, void *data)
925{
926 s->s_bdev = data;
927 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +0200928
929 /*
930 * We set the bdi here to the queue backing, file systems can
931 * overwrite this in ->fill_super()
932 */
933 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 return 0;
935}
936
937static int test_bdev_super(struct super_block *s, void *data)
938{
939 return (void *)s->s_bdev == data;
940}
941
Al Viro152a0832010-07-25 00:46:55 +0400942struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +0400944 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945{
946 struct block_device *bdev;
947 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +0100948 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 int error = 0;
950
Al Viro30c40d22008-02-22 19:50:45 -0500951 if (!(flags & MS_RDONLY))
952 mode |= FMODE_WRITE;
953
Tejun Heod4d77622010-11-13 11:55:18 +0100954 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +0400956 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
958 /*
959 * once the super is inserted into the list by sget, s_umount
960 * will protect the lockfs code from trying to start a snapshot
961 * while we are mounting
962 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200963 mutex_lock(&bdev->bd_fsfreeze_mutex);
964 if (bdev->bd_fsfreeze_count > 0) {
965 mutex_unlock(&bdev->bd_fsfreeze_mutex);
966 error = -EBUSY;
967 goto error_bdev;
968 }
David Howells9249e172012-06-25 12:55:37 +0100969 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
970 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200971 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -0700973 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
975 if (s->s_root) {
976 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400977 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -0700978 error = -EBUSY;
979 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 }
David Howells454e2392006-06-23 02:02:57 -0700981
Tejun Heo4f331f02010-07-20 15:18:07 -0700982 /*
983 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +0100984 * __invalidate_device(). blkdev_put() acquires
985 * bd_mutex and can't be called under s_umount. Drop
986 * s_umount temporarily. This is safe as we're
987 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -0700988 */
989 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +0100990 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -0700991 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 } else {
993 char b[BDEVNAME_SIZE];
994
Al Viro30c40d22008-02-22 19:50:45 -0500995 s->s_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
Pekka Enberge78c9a02006-01-08 01:03:39 -0800997 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -0800998 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001000 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001001 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001002 }
David Howells454e2392006-06-23 02:02:57 -07001003
1004 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001005 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 }
1007
Al Viro152a0832010-07-25 00:46:55 +04001008 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
David Howells454e2392006-06-23 02:02:57 -07001010error_s:
1011 error = PTR_ERR(s);
1012error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001013 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001014error:
Al Viro152a0832010-07-25 00:46:55 +04001015 return ERR_PTR(error);
1016}
1017EXPORT_SYMBOL(mount_bdev);
1018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019void kill_block_super(struct super_block *sb)
1020{
1021 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001022 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001024 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 generic_shutdown_super(sb);
1026 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001027 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001028 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029}
1030
1031EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001032#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
Al Viro3c26ff62010-07-25 11:46:36 +04001034struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001036 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037{
1038 int error;
David Howells9249e172012-06-25 12:55:37 +01001039 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001042 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001044 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001046 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001047 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 }
1049 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001050 return dget(s->s_root);
1051}
1052EXPORT_SYMBOL(mount_nodev);
1053
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054static int compare_single(struct super_block *s, void *p)
1055{
1056 return 1;
1057}
1058
Al Virofc14f2f2010-07-25 01:48:30 +04001059struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001061 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062{
1063 struct super_block *s;
1064 int error;
1065
David Howells9249e172012-06-25 12:55:37 +01001066 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001068 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 if (!s->s_root) {
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001070 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001072 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001073 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 }
1075 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001076 } else {
1077 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 }
Al Virofc14f2f2010-07-25 01:48:30 +04001079 return dget(s->s_root);
1080}
1081EXPORT_SYMBOL(mount_single);
1082
Al Viro9d412a42011-03-17 22:08:28 -04001083struct dentry *
1084mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085{
Al Viroc96e41e2010-07-25 00:17:56 +04001086 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001087 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001089 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001090
Eric Parise0007522008-03-05 10:31:54 -05001091 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001093 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001094 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
Eric Parise0007522008-03-05 10:31:54 -05001096 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001097 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 }
1100
Al Viro1a102ff2011-03-16 09:07:58 -04001101 root = type->mount(type, flags, name, data);
1102 if (IS_ERR(root)) {
1103 error = PTR_ERR(root);
1104 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001105 }
Al Viro9d412a42011-03-17 22:08:28 -04001106 sb = root->d_sb;
1107 BUG_ON(!sb);
1108 WARN_ON(!sb->s_bdi);
Linus Torvalds6c510382011-03-24 10:16:26 -07001109 WARN_ON(sb->s_bdi == &default_backing_dev_info);
Al Viro9d412a42011-03-17 22:08:28 -04001110 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001111
Al Viro9d412a42011-03-17 22:08:28 -04001112 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001113 if (error)
1114 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001115
Jeff Layton42cb56a2009-09-18 13:05:53 -07001116 /*
1117 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1118 * but s_maxbytes was an unsigned long long for many releases. Throw
1119 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001120 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001121 */
Al Viro9d412a42011-03-17 22:08:28 -04001122 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1123 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001124
Al Viro9d412a42011-03-17 22:08:28 -04001125 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001126 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001127 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001129 dput(root);
1130 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131out_free_secdata:
1132 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133out:
David Howells454e2392006-06-23 02:02:57 -07001134 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135}
1136
Jan Kara5accdf82012-06-12 16:20:34 +02001137/*
1138 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1139 * instead.
1140 */
1141void __sb_end_write(struct super_block *sb, int level)
1142{
1143 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1144 /*
1145 * Make sure s_writers are updated before we wake up waiters in
1146 * freeze_super().
1147 */
1148 smp_mb();
1149 if (waitqueue_active(&sb->s_writers.wait))
1150 wake_up(&sb->s_writers.wait);
1151 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1152}
1153EXPORT_SYMBOL(__sb_end_write);
1154
1155#ifdef CONFIG_LOCKDEP
1156/*
1157 * We want lockdep to tell us about possible deadlocks with freezing but
1158 * it's it bit tricky to properly instrument it. Getting a freeze protection
1159 * works as getting a read lock but there are subtle problems. XFS for example
1160 * gets freeze protection on internal level twice in some cases, which is OK
1161 * only because we already hold a freeze protection also on higher level. Due
1162 * to these cases we have to tell lockdep we are doing trylock when we
1163 * already hold a freeze protection for a higher freeze level.
1164 */
1165static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1166 unsigned long ip)
1167{
1168 int i;
1169
1170 if (!trylock) {
1171 for (i = 0; i < level - 1; i++)
1172 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1173 trylock = true;
1174 break;
1175 }
1176 }
1177 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1178}
1179#endif
1180
1181/*
1182 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1183 * instead.
1184 */
1185int __sb_start_write(struct super_block *sb, int level, bool wait)
1186{
1187retry:
1188 if (unlikely(sb->s_writers.frozen >= level)) {
1189 if (!wait)
1190 return 0;
1191 wait_event(sb->s_writers.wait_unfrozen,
1192 sb->s_writers.frozen < level);
1193 }
1194
1195#ifdef CONFIG_LOCKDEP
1196 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1197#endif
1198 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1199 /*
1200 * Make sure counter is updated before we check for frozen.
1201 * freeze_super() first sets frozen and then checks the counter.
1202 */
1203 smp_mb();
1204 if (unlikely(sb->s_writers.frozen >= level)) {
1205 __sb_end_write(sb, level);
1206 goto retry;
1207 }
1208 return 1;
1209}
1210EXPORT_SYMBOL(__sb_start_write);
1211
1212/**
1213 * sb_wait_write - wait until all writers to given file system finish
1214 * @sb: the super for which we wait
1215 * @level: type of writers we wait for (normal vs page fault)
1216 *
1217 * This function waits until there are no writers of given type to given file
1218 * system. Caller of this function should make sure there can be no new writers
1219 * of type @level before calling this function. Otherwise this function can
1220 * livelock.
1221 */
1222static void sb_wait_write(struct super_block *sb, int level)
1223{
1224 s64 writers;
1225
1226 /*
1227 * We just cycle-through lockdep here so that it does not complain
1228 * about returning with lock to userspace
1229 */
1230 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1231 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1232
1233 do {
1234 DEFINE_WAIT(wait);
1235
1236 /*
1237 * We use a barrier in prepare_to_wait() to separate setting
1238 * of frozen and checking of the counter
1239 */
1240 prepare_to_wait(&sb->s_writers.wait, &wait,
1241 TASK_UNINTERRUPTIBLE);
1242
1243 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1244 if (writers)
1245 schedule();
1246
1247 finish_wait(&sb->s_writers.wait, &wait);
1248 } while (writers);
1249}
1250
Josef Bacik18e9e512010-03-23 10:34:56 -04001251/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001252 * freeze_super - lock the filesystem and force it into a consistent state
1253 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001254 *
1255 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1256 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1257 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001258 *
1259 * During this function, sb->s_writers.frozen goes through these values:
1260 *
1261 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1262 *
1263 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1264 * writes should be blocked, though page faults are still allowed. We wait for
1265 * all writes to complete and then proceed to the next stage.
1266 *
1267 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1268 * but internal fs threads can still modify the filesystem (although they
1269 * should not dirty new pages or inodes), writeback can run etc. After waiting
1270 * for all running page faults we sync the filesystem which will clean all
1271 * dirty pages and inodes (no new dirty pages or inodes can be created when
1272 * sync is running).
1273 *
1274 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1275 * modification are blocked (e.g. XFS preallocation truncation on inode
1276 * reclaim). This is usually implemented by blocking new transactions for
1277 * filesystems that have them and need this additional guard. After all
1278 * internal writers are finished we call ->freeze_fs() to finish filesystem
1279 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1280 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1281 *
1282 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001283 */
1284int freeze_super(struct super_block *sb)
1285{
1286 int ret;
1287
1288 atomic_inc(&sb->s_active);
1289 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001290 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001291 deactivate_locked_super(sb);
1292 return -EBUSY;
1293 }
1294
Al Virodabe0dc2012-01-03 21:01:29 -05001295 if (!(sb->s_flags & MS_BORN)) {
1296 up_write(&sb->s_umount);
1297 return 0; /* sic - it's "nothing to do" */
1298 }
1299
Josef Bacik18e9e512010-03-23 10:34:56 -04001300 if (sb->s_flags & MS_RDONLY) {
Jan Kara5accdf82012-06-12 16:20:34 +02001301 /* Nothing to do really... */
1302 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001303 up_write(&sb->s_umount);
1304 return 0;
1305 }
1306
Jan Kara5accdf82012-06-12 16:20:34 +02001307 /* From now on, no new normal writers can start */
1308 sb->s_writers.frozen = SB_FREEZE_WRITE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001309 smp_wmb();
1310
Jan Kara5accdf82012-06-12 16:20:34 +02001311 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1312 up_write(&sb->s_umount);
1313
1314 sb_wait_write(sb, SB_FREEZE_WRITE);
1315
1316 /* Now we go and block page faults... */
1317 down_write(&sb->s_umount);
1318 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1319 smp_wmb();
1320
1321 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1322
1323 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001324 sync_filesystem(sb);
1325
Jan Kara5accdf82012-06-12 16:20:34 +02001326 /* Now wait for internal filesystem counter */
1327 sb->s_writers.frozen = SB_FREEZE_FS;
Josef Bacik18e9e512010-03-23 10:34:56 -04001328 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001329 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001330
Josef Bacik18e9e512010-03-23 10:34:56 -04001331 if (sb->s_op->freeze_fs) {
1332 ret = sb->s_op->freeze_fs(sb);
1333 if (ret) {
1334 printk(KERN_ERR
1335 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001336 sb->s_writers.frozen = SB_UNFROZEN;
Kazuya Mioe1616302011-12-01 16:51:07 +09001337 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001338 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001339 deactivate_locked_super(sb);
1340 return ret;
1341 }
1342 }
Jan Kara5accdf82012-06-12 16:20:34 +02001343 /*
1344 * This is just for debugging purposes so that fs can warn if it
1345 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1346 */
1347 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001348 up_write(&sb->s_umount);
1349 return 0;
1350}
1351EXPORT_SYMBOL(freeze_super);
1352
1353/**
1354 * thaw_super -- unlock filesystem
1355 * @sb: the super to thaw
1356 *
1357 * Unlocks the filesystem and marks it writeable again after freeze_super().
1358 */
1359int thaw_super(struct super_block *sb)
1360{
1361 int error;
1362
1363 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001364 if (sb->s_writers.frozen == SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001365 up_write(&sb->s_umount);
1366 return -EINVAL;
1367 }
1368
1369 if (sb->s_flags & MS_RDONLY)
1370 goto out;
1371
1372 if (sb->s_op->unfreeze_fs) {
1373 error = sb->s_op->unfreeze_fs(sb);
1374 if (error) {
1375 printk(KERN_ERR
1376 "VFS:Filesystem thaw failed\n");
Josef Bacik18e9e512010-03-23 10:34:56 -04001377 up_write(&sb->s_umount);
1378 return error;
1379 }
1380 }
1381
1382out:
Jan Kara5accdf82012-06-12 16:20:34 +02001383 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001384 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001385 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001386 deactivate_locked_super(sb);
1387
1388 return 0;
1389}
1390EXPORT_SYMBOL(thaw_super);