blob: 32a81f3467e06835abdf8e5a3beb91acc3a2c161 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/module.h>
24#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 Viro6d59e7f2008-03-22 15:48:17 -040035#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37
Linus Torvalds1da177e2005-04-16 15:20:36 -070038LIST_HEAD(super_blocks);
39DEFINE_SPINLOCK(sb_lock);
40
Dave Chinnerb0d40c92011-07-08 14:14:42 +100041/*
42 * One thing we have to be careful of with a per-sb shrinker is that we don't
43 * drop the last active reference to the superblock from within the shrinker.
44 * If that happens we could trigger unregistering the shrinker from within the
45 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
46 * take a passive reference to the superblock to avoid this from occurring.
47 */
48static int prune_super(struct shrinker *shrink, struct shrink_control *sc)
49{
50 struct super_block *sb;
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100051 int fs_objects = 0;
52 int total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100053
54 sb = container_of(shrink, struct super_block, s_shrink);
55
56 /*
57 * Deadlock avoidance. We may hold various FS locks, and we don't want
58 * to recurse into the FS that called us in clear_inode() and friends..
59 */
60 if (sc->nr_to_scan && !(sc->gfp_mask & __GFP_FS))
61 return -1;
62
63 if (!grab_super_passive(sb))
Mikulas Patocka09f363c2011-10-31 17:08:57 -070064 return !sc->nr_to_scan ? 0 : -1;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100065
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100066 if (sb->s_op && sb->s_op->nr_cached_objects)
67 fs_objects = sb->s_op->nr_cached_objects(sb);
68
69 total_objects = sb->s_nr_dentry_unused +
70 sb->s_nr_inodes_unused + fs_objects + 1;
71
Dave Chinnerb0d40c92011-07-08 14:14:42 +100072 if (sc->nr_to_scan) {
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100073 int dentries;
74 int inodes;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100075
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100076 /* proportion the scan between the caches */
77 dentries = (sc->nr_to_scan * sb->s_nr_dentry_unused) /
78 total_objects;
79 inodes = (sc->nr_to_scan * sb->s_nr_inodes_unused) /
80 total_objects;
81 if (fs_objects)
82 fs_objects = (sc->nr_to_scan * fs_objects) /
83 total_objects;
84 /*
85 * prune the dcache first as the icache is pinned by it, then
86 * prune the icache, followed by the filesystem specific caches
87 */
88 prune_dcache_sb(sb, dentries);
89 prune_icache_sb(sb, inodes);
Dave Chinnerb0d40c92011-07-08 14:14:42 +100090
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100091 if (fs_objects && sb->s_op->free_cached_objects) {
92 sb->s_op->free_cached_objects(sb, fs_objects);
93 fs_objects = sb->s_op->nr_cached_objects(sb);
94 }
95 total_objects = sb->s_nr_dentry_unused +
96 sb->s_nr_inodes_unused + fs_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100097 }
98
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100099 total_objects = (total_objects / 100) * sysctl_vfs_cache_pressure;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000100 drop_super(sb);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000101 return total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000102}
103
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104/**
105 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700106 * @type: filesystem type superblock should belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 *
108 * Allocates and initializes a new &struct super_block. alloc_super()
109 * returns a pointer new superblock or %NULL if allocation had failed.
110 */
Ingo Molnarcf516242006-07-03 00:25:27 -0700111static struct super_block *alloc_super(struct file_system_type *type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800113 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700114 static const struct super_operations default_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116 if (s) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 if (security_sb_alloc(s)) {
118 kfree(s);
119 s = NULL;
120 goto out;
121 }
Nick Piggin6416ccb2010-08-18 04:37:38 +1000122#ifdef CONFIG_SMP
123 s->s_files = alloc_percpu(struct list_head);
124 if (!s->s_files) {
125 security_sb_free(s);
126 kfree(s);
127 s = NULL;
128 goto out;
129 } else {
130 int i;
131
132 for_each_possible_cpu(i)
133 INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
134 }
135#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 INIT_LIST_HEAD(&s->s_files);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000137#endif
Jens Axboe95f28602011-03-17 11:13:12 +0100138 s->s_bdi = &default_backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 INIT_LIST_HEAD(&s->s_instances);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100140 INIT_HLIST_BL_HEAD(&s->s_anon);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 INIT_LIST_HEAD(&s->s_inodes);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700142 INIT_LIST_HEAD(&s->s_dentry_lru);
Dave Chinner98b745c2011-07-08 14:14:39 +1000143 INIT_LIST_HEAD(&s->s_inode_lru);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000144 spin_lock_init(&s->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 init_rwsem(&s->s_umount);
Ingo Molnar7892f2f2006-01-09 15:59:25 -0800146 mutex_init(&s->s_lock);
Arjan van de Ven897c6ff2006-07-03 00:25:28 -0700147 lockdep_set_class(&s->s_umount, &type->s_umount_key);
Ingo Molnarcf516242006-07-03 00:25:27 -0700148 /*
149 * The locking rules for s_lock are up to the
150 * filesystem. For example ext3fs has different
151 * lock ordering than usbfs:
152 */
153 lockdep_set_class(&s->s_lock, &type->s_lock_key);
Peter Zijlstraada723d2009-02-18 14:48:30 -0800154 /*
155 * sget() can have s_umount recursion.
156 *
157 * When it cannot find a suitable sb, it allocates a new
158 * one (this one), and tries again to find a suitable old
159 * one.
160 *
161 * In case that succeeds, it will acquire the s_umount
162 * lock of the old one. Since these are clearly distrinct
163 * locks, and this object isn't exposed yet, there's no
164 * risk of deadlocks.
165 *
166 * Annotate this by putting this lock in a different
167 * subclass.
168 */
169 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
Al Virob20bd1a2010-03-22 08:53:19 -0400170 s->s_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 atomic_set(&s->s_active, 1);
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800172 mutex_init(&s->s_vfs_rename_mutex);
Roland Dreier51ee0492010-04-27 14:23:57 -0700173 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
Ingo Molnard3be9152006-03-23 03:00:29 -0800174 mutex_init(&s->s_dquot.dqio_mutex);
175 mutex_init(&s->s_dquot.dqonoff_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 init_rwsem(&s->s_dquot.dqptr_sem);
177 init_waitqueue_head(&s->s_wait_unfrozen);
178 s->s_maxbytes = MAX_NON_LFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 s->s_op = &default_op;
180 s->s_time_gran = 1000000000;
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600181 s->cleancache_poolid = -1;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000182
183 s->s_shrink.seeks = DEFAULT_SEEKS;
184 s->s_shrink.shrink = prune_super;
Dave Chinner8ab47662011-07-08 14:14:45 +1000185 s->s_shrink.batch = 1024;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 }
187out:
188 return s;
189}
190
191/**
192 * destroy_super - frees a superblock
193 * @s: superblock to free
194 *
195 * Frees a superblock.
196 */
197static inline void destroy_super(struct super_block *s)
198{
Nick Piggin6416ccb2010-08-18 04:37:38 +1000199#ifdef CONFIG_SMP
200 free_percpu(s->s_files);
201#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 security_sb_free(s);
Miklos Szeredi79c0b2d2007-05-08 00:25:43 -0700203 kfree(s->s_subtype);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800204 kfree(s->s_options);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 kfree(s);
206}
207
208/* Superblock refcounting */
209
210/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400211 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 */
Al Viro35cf7ba2010-03-22 21:13:53 -0400213void __put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 if (!--sb->s_count) {
Al Viro551de6f2010-03-22 19:36:35 -0400216 list_del_init(&sb->s_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 destroy_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219}
220
221/**
222 * put_super - drop a temporary reference to superblock
223 * @sb: superblock in question
224 *
225 * Drops a temporary reference, frees superblock if there's no
226 * references left.
227 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200228void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
230 spin_lock(&sb_lock);
231 __put_super(sb);
232 spin_unlock(&sb_lock);
233}
234
235
236/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400237 * deactivate_locked_super - drop an active reference to superblock
238 * @s: superblock to deactivate
239 *
Al Viro1712ac82010-03-22 15:22:31 -0400240 * Drops an active reference to superblock, converting it into a temprory
241 * one if there is no other active references left. In that case we
242 * tell fs driver to shut it down and drop the temporary reference we
243 * had just acquired.
244 *
245 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400246 */
247void deactivate_locked_super(struct super_block *s)
248{
249 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400250 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600251 cleancache_flush_fs(s);
Al Viro74dbbdd2009-05-06 01:07:50 -0400252 fs->kill_sb(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000253
254 /* caches are now gone, we can safely kill the shrinker now */
255 unregister_shrinker(&s->s_shrink);
256
Boaz Harroshd863b502011-02-10 15:01:20 -0800257 /*
258 * We need to call rcu_barrier so all the delayed rcu free
259 * inodes are flushed before we release the fs module.
260 */
261 rcu_barrier();
Al Viro74dbbdd2009-05-06 01:07:50 -0400262 put_filesystem(fs);
263 put_super(s);
264 } else {
265 up_write(&s->s_umount);
266 }
267}
268
269EXPORT_SYMBOL(deactivate_locked_super);
270
271/**
Al Viro1712ac82010-03-22 15:22:31 -0400272 * deactivate_super - drop an active reference to superblock
273 * @s: superblock to deactivate
274 *
275 * Variant of deactivate_locked_super(), except that superblock is *not*
276 * locked by caller. If we are going to drop the final active reference,
277 * lock will be acquired prior to that.
278 */
279void deactivate_super(struct super_block *s)
280{
281 if (!atomic_add_unless(&s->s_active, -1, 1)) {
282 down_write(&s->s_umount);
283 deactivate_locked_super(s);
284 }
285}
286
287EXPORT_SYMBOL(deactivate_super);
288
289/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 * grab_super - acquire an active reference
291 * @s: reference we are trying to make active
292 *
293 * Tries to acquire an active reference. grab_super() is used when we
294 * had just found a superblock in super_blocks or fs_type->fs_supers
295 * and want to turn it into a full-blown active reference. grab_super()
296 * is called with sb_lock held and drops it. Returns 1 in case of
297 * success, 0 if we had failed (superblock contents was already dead or
298 * dying when grab_super() had been called).
299 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700300static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
Al Virob20bd1a2010-03-22 08:53:19 -0400302 if (atomic_inc_not_zero(&s->s_active)) {
303 spin_unlock(&sb_lock);
Al Virob20bd1a2010-03-22 08:53:19 -0400304 return 1;
305 }
306 /* it's going away */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 s->s_count++;
308 spin_unlock(&sb_lock);
Al Viro1712ac82010-03-22 15:22:31 -0400309 /* wait for it to die */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 up_write(&s->s_umount);
312 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 return 0;
314}
315
David Howellscf9a2ae2006-08-29 19:05:54 +0100316/*
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000317 * grab_super_passive - acquire a passive reference
318 * @s: reference we are trying to grab
319 *
320 * Tries to acquire a passive reference. This is used in places where we
321 * cannot take an active reference but we need to ensure that the
322 * superblock does not go away while we are working on it. It returns
323 * false if a reference was not gained, and returns true with the s_umount
324 * lock held in read mode if a reference is gained. On successful return,
325 * the caller must drop the s_umount lock and the passive reference when
326 * done.
327 */
328bool grab_super_passive(struct super_block *sb)
329{
330 spin_lock(&sb_lock);
331 if (list_empty(&sb->s_instances)) {
332 spin_unlock(&sb_lock);
333 return false;
334 }
335
336 sb->s_count++;
337 spin_unlock(&sb_lock);
338
339 if (down_read_trylock(&sb->s_umount)) {
340 if (sb->s_root)
341 return true;
342 up_read(&sb->s_umount);
343 }
344
345 put_super(sb);
346 return false;
347}
348
349/*
Al Viro914e2632006-10-18 13:55:46 -0400350 * Superblock locking. We really ought to get rid of these two.
351 */
352void lock_super(struct super_block * sb)
353{
Al Viro914e2632006-10-18 13:55:46 -0400354 mutex_lock(&sb->s_lock);
355}
356
357void unlock_super(struct super_block * sb)
358{
Al Viro914e2632006-10-18 13:55:46 -0400359 mutex_unlock(&sb->s_lock);
360}
361
362EXPORT_SYMBOL(lock_super);
363EXPORT_SYMBOL(unlock_super);
364
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365/**
366 * generic_shutdown_super - common helper for ->kill_sb()
367 * @sb: superblock to kill
368 *
369 * generic_shutdown_super() does all fs-independent work on superblock
370 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
371 * that need destruction out of superblock, call generic_shutdown_super()
372 * and release aforementioned objects. Note: dentries and inodes _are_
373 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700374 *
375 * Upon calling this function, the filesystem may no longer alter or
376 * rearrange the set of dentries belonging to this super_block, nor may it
377 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 */
379void generic_shutdown_super(struct super_block *sb)
380{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800381 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
David Howellsc636ebd2006-10-11 01:22:19 -0700383 if (sb->s_root) {
384 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200385 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 sb->s_flags &= ~MS_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800387
Al Viro63997e92010-10-25 20:49:35 -0400388 fsnotify_unmount_inodes(&sb->s_inodes);
389
390 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 if (sop->put_super)
393 sop->put_super(sb);
394
Al Viro63997e92010-10-25 20:49:35 -0400395 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800396 printk("VFS: Busy inodes after unmount of %s. "
397 "Self-destruct in 5 seconds. Have a nice day...\n",
398 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 }
401 spin_lock(&sb_lock);
402 /* should be initialized for __put_super_and_need_restart() */
Al Viro551de6f2010-03-22 19:36:35 -0400403 list_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 spin_unlock(&sb_lock);
405 up_write(&sb->s_umount);
406}
407
408EXPORT_SYMBOL(generic_shutdown_super);
409
410/**
411 * sget - find or create a superblock
412 * @type: filesystem type superblock should belong to
413 * @test: comparison callback
414 * @set: setup callback
415 * @data: argument to each of them
416 */
417struct super_block *sget(struct file_system_type *type,
418 int (*test)(struct super_block *,void *),
419 int (*set)(struct super_block *,void *),
420 void *data)
421{
422 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700423 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 int err;
425
426retry:
427 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700428 if (test) {
429 list_for_each_entry(old, &type->fs_supers, s_instances) {
430 if (!test(old, data))
431 continue;
432 if (!grab_super(old))
433 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700434 if (s) {
435 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700436 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400437 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700438 }
Al Virod3f21472010-03-23 11:11:05 -0400439 down_write(&old->s_umount);
Al Viro7a4dec52010-08-09 12:05:43 -0400440 if (unlikely(!(old->s_flags & MS_BORN))) {
441 deactivate_locked_super(old);
442 goto retry;
443 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700444 return old;
445 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 }
447 if (!s) {
448 spin_unlock(&sb_lock);
Ingo Molnarcf516242006-07-03 00:25:27 -0700449 s = alloc_super(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 if (!s)
451 return ERR_PTR(-ENOMEM);
452 goto retry;
453 }
454
455 err = set(s, data);
456 if (err) {
457 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700458 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 destroy_super(s);
460 return ERR_PTR(err);
461 }
462 s->s_type = type;
463 strlcpy(s->s_id, type->name, sizeof(s->s_id));
464 list_add_tail(&s->s_list, &super_blocks);
465 list_add(&s->s_instances, &type->fs_supers);
466 spin_unlock(&sb_lock);
467 get_filesystem(type);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000468 register_shrinker(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 return s;
470}
471
472EXPORT_SYMBOL(sget);
473
474void drop_super(struct super_block *sb)
475{
476 up_read(&sb->s_umount);
477 put_super(sb);
478}
479
480EXPORT_SYMBOL(drop_super);
481
Christoph Hellwige5004752009-05-05 16:08:56 +0200482/**
483 * sync_supers - helper for periodic superblock writeback
484 *
485 * Call the write_super method if present on all dirty superblocks in
486 * the system. This is for the periodic writeback used by most older
487 * filesystems. For data integrity superblock writeback use
488 * sync_filesystems() instead.
489 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 * Note: check the dirty flag before waiting, so we don't
491 * hold up the sync while mounting a device. (The newly
492 * mounted device won't need syncing.)
493 */
494void sync_supers(void)
495{
Al Virodca33252010-07-25 02:31:46 +0400496 struct super_block *sb, *p = NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700497
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400499 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viro551de6f2010-03-22 19:36:35 -0400500 if (list_empty(&sb->s_instances))
501 continue;
Christoph Hellwige5004752009-05-05 16:08:56 +0200502 if (sb->s_op->write_super && sb->s_dirt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 sb->s_count++;
504 spin_unlock(&sb_lock);
Christoph Hellwige5004752009-05-05 16:08:56 +0200505
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 down_read(&sb->s_umount);
Christoph Hellwige5004752009-05-05 16:08:56 +0200507 if (sb->s_root && sb->s_dirt)
508 sb->s_op->write_super(sb);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700509 up_read(&sb->s_umount);
Christoph Hellwige5004752009-05-05 16:08:56 +0200510
Kirill Korotaev618f0632005-06-23 00:09:54 -0700511 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400512 if (p)
513 __put_super(p);
514 p = sb;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700515 }
516 }
Al Virodca33252010-07-25 02:31:46 +0400517 if (p)
518 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 spin_unlock(&sb_lock);
520}
521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522/**
Al Viro01a05b32010-03-23 06:06:58 -0400523 * iterate_supers - call function for all active superblocks
524 * @f: function to call
525 * @arg: argument to pass to it
526 *
527 * Scans the superblock list and calls given function, passing it
528 * locked superblock and given argument.
529 */
530void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
531{
Al Virodca33252010-07-25 02:31:46 +0400532 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400533
534 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400535 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viro01a05b32010-03-23 06:06:58 -0400536 if (list_empty(&sb->s_instances))
537 continue;
538 sb->s_count++;
539 spin_unlock(&sb_lock);
540
541 down_read(&sb->s_umount);
542 if (sb->s_root)
543 f(sb, arg);
544 up_read(&sb->s_umount);
545
546 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400547 if (p)
548 __put_super(p);
549 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400550 }
Al Virodca33252010-07-25 02:31:46 +0400551 if (p)
552 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400553 spin_unlock(&sb_lock);
554}
555
556/**
Al Viro43e15cd2011-06-03 20:16:57 -0400557 * iterate_supers_type - call function for superblocks of given type
558 * @type: fs type
559 * @f: function to call
560 * @arg: argument to pass to it
561 *
562 * Scans the superblock list and calls given function, passing it
563 * locked superblock and given argument.
564 */
565void iterate_supers_type(struct file_system_type *type,
566 void (*f)(struct super_block *, void *), void *arg)
567{
568 struct super_block *sb, *p = NULL;
569
570 spin_lock(&sb_lock);
571 list_for_each_entry(sb, &type->fs_supers, s_instances) {
572 sb->s_count++;
573 spin_unlock(&sb_lock);
574
575 down_read(&sb->s_umount);
576 if (sb->s_root)
577 f(sb, arg);
578 up_read(&sb->s_umount);
579
580 spin_lock(&sb_lock);
581 if (p)
582 __put_super(p);
583 p = sb;
584 }
585 if (p)
586 __put_super(p);
587 spin_unlock(&sb_lock);
588}
589
590EXPORT_SYMBOL(iterate_supers_type);
591
592/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 * get_super - get the superblock of a device
594 * @bdev: device to get the superblock for
595 *
596 * Scans the superblock list and finds the superblock of the file system
597 * mounted on the device given. %NULL is returned if no match is found.
598 */
599
Al Virodf40c012010-03-22 20:23:25 -0400600struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700602 struct super_block *sb;
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 if (!bdev)
605 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700608rescan:
609 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viro551de6f2010-03-22 19:36:35 -0400610 if (list_empty(&sb->s_instances))
611 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700612 if (sb->s_bdev == bdev) {
613 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700615 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400616 /* still alive? */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700617 if (sb->s_root)
618 return sb;
619 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400620 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700621 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400622 __put_super(sb);
623 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 }
625 }
626 spin_unlock(&sb_lock);
627 return NULL;
628}
629
630EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200631
632/**
633 * get_active_super - get an active reference to the superblock of a device
634 * @bdev: device to get the superblock for
635 *
636 * Scans the superblock list and finds the superblock of the file system
637 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400638 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200639 */
640struct super_block *get_active_super(struct block_device *bdev)
641{
642 struct super_block *sb;
643
644 if (!bdev)
645 return NULL;
646
Al Viro14945832010-03-22 20:15:33 -0400647restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200648 spin_lock(&sb_lock);
649 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viro551de6f2010-03-22 19:36:35 -0400650 if (list_empty(&sb->s_instances))
651 continue;
Al Viro14945832010-03-22 20:15:33 -0400652 if (sb->s_bdev == bdev) {
653 if (grab_super(sb)) /* drops sb_lock */
654 return sb;
655 else
656 goto restart;
657 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200658 }
659 spin_unlock(&sb_lock);
660 return NULL;
661}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Al Virodf40c012010-03-22 20:23:25 -0400663struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700665 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700668rescan:
669 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viro551de6f2010-03-22 19:36:35 -0400670 if (list_empty(&sb->s_instances))
671 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700672 if (sb->s_dev == dev) {
673 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700675 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400676 /* still alive? */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700677 if (sb->s_root)
678 return sb;
679 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400680 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700681 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400682 __put_super(sb);
683 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 }
685 }
686 spin_unlock(&sb_lock);
687 return NULL;
688}
689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 * do_remount_sb - asks filesystem to change mount options.
692 * @sb: superblock in question
693 * @flags: numeric part of options
694 * @data: the rest of options
695 * @force: whether or not to force the change
696 *
697 * Alters the mount options of a mounted file system.
698 */
699int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
700{
701 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400702 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200703
704 if (sb->s_frozen != SB_UNFROZEN)
705 return -EBUSY;
706
David Howells93614012006-09-30 20:45:40 +0200707#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
709 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200710#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 if (flags & MS_RDONLY)
713 acct_auto_close(sb);
714 shrink_dcache_sb(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200715 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
Nick Piggind208bbd2009-12-21 16:28:53 -0800717 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 /* If we are remounting RDONLY and current sb is read/write,
720 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800721 if (remount_ro) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 if (force)
723 mark_files_ro(sb);
J. R. Okajimab0895512009-06-17 01:16:50 +0900724 else if (!fs_may_remount_ro(sb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 return -EBUSY;
726 }
727
728 if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 retval = sb->s_op->remount_fs(sb, &flags, data);
J. R. Okajimab0895512009-06-17 01:16:50 +0900730 if (retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 return retval;
732 }
733 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400734
Nick Piggind208bbd2009-12-21 16:28:53 -0800735 /*
736 * Some filesystems modify their metadata via some other path than the
737 * bdev buffer cache (eg. use a private mapping, or directories in
738 * pagecache, etc). Also file data modifications go via their own
739 * mappings. So If we try to mount readonly then copy the filesystem
740 * from bdev, we could get stale data, so invalidate it to give a best
741 * effort at coherency.
742 */
743 if (remount_ro && sb->s_bdev)
744 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 return 0;
746}
747
Jens Axboea2a95372009-03-17 09:38:40 +0100748static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
Al Virodca33252010-07-25 02:31:46 +0400750 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
752 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400753 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viro551de6f2010-03-22 19:36:35 -0400754 if (list_empty(&sb->s_instances))
755 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 sb->s_count++;
757 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400758 down_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
760 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 * What lock protects sb->s_flags??
762 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 }
Al Viro443b94b2009-05-05 23:48:50 -0400765 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400767 if (p)
768 __put_super(p);
769 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 }
Al Virodca33252010-07-25 02:31:46 +0400771 if (p)
772 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100774 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 printk("Emergency Remount complete\n");
776}
777
778void emergency_remount(void)
779{
Jens Axboea2a95372009-03-17 09:38:40 +0100780 struct work_struct *work;
781
782 work = kmalloc(sizeof(*work), GFP_ATOMIC);
783 if (work) {
784 INIT_WORK(work, do_emergency_remount);
785 schedule_work(work);
786 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787}
788
789/*
790 * Unnamed block devices are dummy devices used by virtual
791 * filesystems which don't use real block-devices. -- jrs
792 */
793
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400794static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Al Viroc63e09e2009-06-24 02:05:18 -0400796static int unnamed_dev_start = 0; /* don't bother trying below it */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
Al Viro0ee5dc62011-07-07 15:44:25 -0400798int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799{
800 int dev;
801 int error;
802
803 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400804 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 return -ENOMEM;
806 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400807 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400808 if (!error)
809 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 spin_unlock(&unnamed_dev_lock);
811 if (error == -EAGAIN)
812 /* We raced and lost with another CPU. */
813 goto retry;
814 else if (error)
815 return -EAGAIN;
816
817 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
818 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400819 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400820 if (unnamed_dev_start > dev)
821 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 spin_unlock(&unnamed_dev_lock);
823 return -EMFILE;
824 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400825 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 return 0;
827}
Al Viro0ee5dc62011-07-07 15:44:25 -0400828EXPORT_SYMBOL(get_anon_bdev);
829
830void free_anon_bdev(dev_t dev)
831{
832 int slot = MINOR(dev);
833 spin_lock(&unnamed_dev_lock);
834 ida_remove(&unnamed_dev_ida, slot);
835 if (slot < unnamed_dev_start)
836 unnamed_dev_start = slot;
837 spin_unlock(&unnamed_dev_lock);
838}
839EXPORT_SYMBOL(free_anon_bdev);
840
841int set_anon_super(struct super_block *s, void *data)
842{
843 int error = get_anon_bdev(&s->s_dev);
844 if (!error)
845 s->s_bdi = &noop_backing_dev_info;
846 return error;
847}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
849EXPORT_SYMBOL(set_anon_super);
850
851void kill_anon_super(struct super_block *sb)
852{
Al Viro0ee5dc62011-07-07 15:44:25 -0400853 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400855 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856}
857
858EXPORT_SYMBOL(kill_anon_super);
859
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860void kill_litter_super(struct super_block *sb)
861{
862 if (sb->s_root)
863 d_genocide(sb->s_root);
864 kill_anon_super(sb);
865}
866
867EXPORT_SYMBOL(kill_litter_super);
868
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700869static int ns_test_super(struct super_block *sb, void *data)
870{
871 return sb->s_fs_info == data;
872}
873
874static int ns_set_super(struct super_block *sb, void *data)
875{
876 sb->s_fs_info = data;
877 return set_anon_super(sb, NULL);
878}
879
Al Viroceefda62010-07-26 13:16:50 +0400880struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
881 void *data, int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700882{
883 struct super_block *sb;
884
885 sb = sget(fs_type, ns_test_super, ns_set_super, data);
886 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +0400887 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700888
889 if (!sb->s_root) {
890 int err;
891 sb->s_flags = flags;
892 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
893 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400894 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +0400895 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700896 }
897
898 sb->s_flags |= MS_ACTIVE;
899 }
900
Al Viroceefda62010-07-26 13:16:50 +0400901 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700902}
903
Al Viroceefda62010-07-26 13:16:50 +0400904EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700905
David Howells93614012006-09-30 20:45:40 +0200906#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907static int set_bdev_super(struct super_block *s, void *data)
908{
909 s->s_bdev = data;
910 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +0200911
912 /*
913 * We set the bdi here to the queue backing, file systems can
914 * overwrite this in ->fill_super()
915 */
916 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 return 0;
918}
919
920static int test_bdev_super(struct super_block *s, void *data)
921{
922 return (void *)s->s_bdev == data;
923}
924
Al Viro152a0832010-07-25 00:46:55 +0400925struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +0400927 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928{
929 struct block_device *bdev;
930 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +0100931 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 int error = 0;
933
Al Viro30c40d22008-02-22 19:50:45 -0500934 if (!(flags & MS_RDONLY))
935 mode |= FMODE_WRITE;
936
Tejun Heod4d77622010-11-13 11:55:18 +0100937 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +0400939 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
941 /*
942 * once the super is inserted into the list by sget, s_umount
943 * will protect the lockfs code from trying to start a snapshot
944 * while we are mounting
945 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200946 mutex_lock(&bdev->bd_fsfreeze_mutex);
947 if (bdev->bd_fsfreeze_count > 0) {
948 mutex_unlock(&bdev->bd_fsfreeze_mutex);
949 error = -EBUSY;
950 goto error_bdev;
951 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200953 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -0700955 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
957 if (s->s_root) {
958 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400959 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -0700960 error = -EBUSY;
961 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 }
David Howells454e2392006-06-23 02:02:57 -0700963
Tejun Heo4f331f02010-07-20 15:18:07 -0700964 /*
965 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +0100966 * __invalidate_device(). blkdev_put() acquires
967 * bd_mutex and can't be called under s_umount. Drop
968 * s_umount temporarily. This is safe as we're
969 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -0700970 */
971 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +0100972 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -0700973 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 } else {
975 char b[BDEVNAME_SIZE];
976
Al Viro9e1f1de2011-06-03 18:24:58 -0400977 s->s_flags = flags | MS_NOSEC;
Al Viro30c40d22008-02-22 19:50:45 -0500978 s->s_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
Pekka Enberge78c9a02006-01-08 01:03:39 -0800980 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -0800981 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400983 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -0700984 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -0800985 }
David Howells454e2392006-06-23 02:02:57 -0700986
987 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -0500988 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 }
990
Al Viro152a0832010-07-25 00:46:55 +0400991 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
David Howells454e2392006-06-23 02:02:57 -0700993error_s:
994 error = PTR_ERR(s);
995error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +0100996 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -0700997error:
Al Viro152a0832010-07-25 00:46:55 +0400998 return ERR_PTR(error);
999}
1000EXPORT_SYMBOL(mount_bdev);
1001
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002void kill_block_super(struct super_block *sb)
1003{
1004 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001005 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001007 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 generic_shutdown_super(sb);
1009 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001010 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001011 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012}
1013
1014EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001015#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Al Viro3c26ff62010-07-25 11:46:36 +04001017struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001019 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020{
1021 int error;
1022 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
1023
1024 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001025 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
1027 s->s_flags = flags;
1028
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001029 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001031 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001032 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 }
1034 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001035 return dget(s->s_root);
1036}
1037EXPORT_SYMBOL(mount_nodev);
1038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039static int compare_single(struct super_block *s, void *p)
1040{
1041 return 1;
1042}
1043
Al Virofc14f2f2010-07-25 01:48:30 +04001044struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001046 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047{
1048 struct super_block *s;
1049 int error;
1050
1051 s = sget(fs_type, compare_single, set_anon_super, NULL);
1052 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001053 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 if (!s->s_root) {
1055 s->s_flags = flags;
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001056 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001058 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001059 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 }
1061 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001062 } else {
1063 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 }
Al Virofc14f2f2010-07-25 01:48:30 +04001065 return dget(s->s_root);
1066}
1067EXPORT_SYMBOL(mount_single);
1068
Al Viro9d412a42011-03-17 22:08:28 -04001069struct dentry *
1070mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
Al Viroc96e41e2010-07-25 00:17:56 +04001072 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001073 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001075 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001076
Eric Parise0007522008-03-05 10:31:54 -05001077 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001079 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001080 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081
Eric Parise0007522008-03-05 10:31:54 -05001082 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001083 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
1086
Al Viro1a102ff2011-03-16 09:07:58 -04001087 root = type->mount(type, flags, name, data);
1088 if (IS_ERR(root)) {
1089 error = PTR_ERR(root);
1090 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001091 }
Al Viro9d412a42011-03-17 22:08:28 -04001092 sb = root->d_sb;
1093 BUG_ON(!sb);
1094 WARN_ON(!sb->s_bdi);
Linus Torvalds6c510382011-03-24 10:16:26 -07001095 WARN_ON(sb->s_bdi == &default_backing_dev_info);
Al Viro9d412a42011-03-17 22:08:28 -04001096 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001097
Al Viro9d412a42011-03-17 22:08:28 -04001098 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001099 if (error)
1100 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001101
Jeff Layton42cb56a2009-09-18 13:05:53 -07001102 /*
1103 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1104 * but s_maxbytes was an unsigned long long for many releases. Throw
1105 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001106 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001107 */
Al Viro9d412a42011-03-17 22:08:28 -04001108 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1109 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001110
Al Viro9d412a42011-03-17 22:08:28 -04001111 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001112 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001113 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001115 dput(root);
1116 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117out_free_secdata:
1118 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119out:
David Howells454e2392006-06-23 02:02:57 -07001120 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121}
1122
Josef Bacik18e9e512010-03-23 10:34:56 -04001123/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001124 * freeze_super - lock the filesystem and force it into a consistent state
1125 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001126 *
1127 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1128 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1129 * -EBUSY.
1130 */
1131int freeze_super(struct super_block *sb)
1132{
1133 int ret;
1134
1135 atomic_inc(&sb->s_active);
1136 down_write(&sb->s_umount);
1137 if (sb->s_frozen) {
1138 deactivate_locked_super(sb);
1139 return -EBUSY;
1140 }
1141
1142 if (sb->s_flags & MS_RDONLY) {
1143 sb->s_frozen = SB_FREEZE_TRANS;
1144 smp_wmb();
1145 up_write(&sb->s_umount);
1146 return 0;
1147 }
1148
1149 sb->s_frozen = SB_FREEZE_WRITE;
1150 smp_wmb();
1151
1152 sync_filesystem(sb);
1153
1154 sb->s_frozen = SB_FREEZE_TRANS;
1155 smp_wmb();
1156
1157 sync_blockdev(sb->s_bdev);
1158 if (sb->s_op->freeze_fs) {
1159 ret = sb->s_op->freeze_fs(sb);
1160 if (ret) {
1161 printk(KERN_ERR
1162 "VFS:Filesystem freeze failed\n");
1163 sb->s_frozen = SB_UNFROZEN;
1164 deactivate_locked_super(sb);
1165 return ret;
1166 }
1167 }
1168 up_write(&sb->s_umount);
1169 return 0;
1170}
1171EXPORT_SYMBOL(freeze_super);
1172
1173/**
1174 * thaw_super -- unlock filesystem
1175 * @sb: the super to thaw
1176 *
1177 * Unlocks the filesystem and marks it writeable again after freeze_super().
1178 */
1179int thaw_super(struct super_block *sb)
1180{
1181 int error;
1182
1183 down_write(&sb->s_umount);
1184 if (sb->s_frozen == SB_UNFROZEN) {
1185 up_write(&sb->s_umount);
1186 return -EINVAL;
1187 }
1188
1189 if (sb->s_flags & MS_RDONLY)
1190 goto out;
1191
1192 if (sb->s_op->unfreeze_fs) {
1193 error = sb->s_op->unfreeze_fs(sb);
1194 if (error) {
1195 printk(KERN_ERR
1196 "VFS:Filesystem thaw failed\n");
1197 sb->s_frozen = SB_FREEZE_TRANS;
1198 up_write(&sb->s_umount);
1199 return error;
1200 }
1201 }
1202
1203out:
1204 sb->s_frozen = SB_UNFROZEN;
1205 smp_wmb();
1206 wake_up(&sb->s_wait_unfrozen);
1207 deactivate_locked_super(sb);
1208
1209 return 0;
1210}
1211EXPORT_SYMBOL(thaw_super);