]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/super.c
Merge tag 'sti-drm-next-2017-01-06' of https://github.com/vinceab/linux into drm...
[mirror_ubuntu-artful-kernel.git] / fs / super.c
CommitLineData
1da177e4
LT
1/*
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:
96de0e25 18 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
1da177e4
LT
19 * 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
630d9c47 23#include <linux/export.h>
1da177e4 24#include <linux/slab.h>
1da177e4 25#include <linux/blkdev.h>
1da177e4
LT
26#include <linux/mount.h>
27#include <linux/security.h>
1da177e4
LT
28#include <linux/writeback.h> /* for the emergency remount stuff */
29#include <linux/idr.h>
353ab6e9 30#include <linux/mutex.h>
5477d0fa 31#include <linux/backing-dev.h>
ceb5bdc2 32#include <linux/rculist_bl.h>
c515e1fd 33#include <linux/cleancache.h>
40401530 34#include <linux/fsnotify.h>
5accdf82 35#include <linux/lockdep.h>
6e4eab57 36#include <linux/user_namespace.h>
6d59e7f5 37#include "internal.h"
1da177e4
LT
38
39
15d0f5ea
AV
40static LIST_HEAD(super_blocks);
41static DEFINE_SPINLOCK(sb_lock);
1da177e4 42
5accdf82
JK
43static char *sb_writers_name[SB_FREEZE_LEVELS] = {
44 "sb_writers",
45 "sb_pagefaults",
46 "sb_internal",
47};
48
b0d40c92
DC
49/*
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 */
0a234c6d
DC
56static unsigned long super_cache_scan(struct shrinker *shrink,
57 struct shrink_control *sc)
b0d40c92
DC
58{
59 struct super_block *sb;
0a234c6d
DC
60 long fs_objects = 0;
61 long total_objects;
62 long freed = 0;
63 long dentries;
64 long inodes;
b0d40c92
DC
65
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 */
0a234c6d
DC
72 if (!(sc->gfp_mask & __GFP_FS))
73 return SHRINK_STOP;
b0d40c92 74
eb6ef3df 75 if (!trylock_super(sb))
0a234c6d 76 return SHRINK_STOP;
b0d40c92 77
d0407903 78 if (sb->s_op->nr_cached_objects)
4101b624 79 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
0e1fdafd 80
503c358c
VD
81 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
82 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
f6041567 83 total_objects = dentries + inodes + fs_objects + 1;
475d0db7
TH
84 if (!total_objects)
85 total_objects = 1;
0e1fdafd 86
0a234c6d 87 /* proportion the scan between the caches */
f6041567 88 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
bc3b14cb 89 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
503c358c 90 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
b0d40c92 91
0a234c6d
DC
92 /*
93 * prune the dcache first as the icache is pinned by it, then
94 * prune the icache, followed by the filesystem specific caches
49e7e7ff
VD
95 *
96 * Ensure that we always scan at least one object - memcg kmem
97 * accounting uses this to fully empty the caches.
0a234c6d 98 */
49e7e7ff 99 sc->nr_to_scan = dentries + 1;
503c358c 100 freed = prune_dcache_sb(sb, sc);
49e7e7ff 101 sc->nr_to_scan = inodes + 1;
503c358c 102 freed += prune_icache_sb(sb, sc);
0a234c6d
DC
103
104 if (fs_objects) {
49e7e7ff 105 sc->nr_to_scan = fs_objects + 1;
4101b624 106 freed += sb->s_op->free_cached_objects(sb, sc);
b0d40c92
DC
107 }
108
eb6ef3df 109 up_read(&sb->s_umount);
0a234c6d
DC
110 return freed;
111}
112
113static unsigned long super_cache_count(struct shrinker *shrink,
114 struct shrink_control *sc)
115{
116 struct super_block *sb;
117 long total_objects = 0;
118
119 sb = container_of(shrink, struct super_block, s_shrink);
120
d23da150 121 /*
eb6ef3df 122 * Don't call trylock_super as it is a potential
d23da150
TC
123 * scalability bottleneck. The counts could get updated
124 * between super_cache_count and super_cache_scan anyway.
125 * Call to super_cache_count with shrinker_rwsem held
503c358c 126 * ensures the safety of call to list_lru_shrink_count() and
d23da150
TC
127 * s_op->nr_cached_objects().
128 */
0a234c6d 129 if (sb->s_op && sb->s_op->nr_cached_objects)
4101b624 130 total_objects = sb->s_op->nr_cached_objects(sb, sc);
0a234c6d 131
503c358c
VD
132 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
133 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
0a234c6d 134
55f841ce 135 total_objects = vfs_pressure_ratio(total_objects);
0e1fdafd 136 return total_objects;
b0d40c92
DC
137}
138
853b39a7
ON
139static void destroy_super_work(struct work_struct *work)
140{
141 struct super_block *s = container_of(work, struct super_block,
142 destroy_work);
143 int i;
144
145 for (i = 0; i < SB_FREEZE_LEVELS; i++)
8129ed29 146 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
853b39a7
ON
147 kfree(s);
148}
149
150static void destroy_super_rcu(struct rcu_head *head)
151{
152 struct super_block *s = container_of(head, struct super_block, rcu);
153 INIT_WORK(&s->destroy_work, destroy_super_work);
154 schedule_work(&s->destroy_work);
155}
156
7eb5e882
AV
157/**
158 * destroy_super - frees a superblock
159 * @s: superblock to free
160 *
161 * Frees a superblock.
162 */
163static void destroy_super(struct super_block *s)
5accdf82 164{
7eb5e882
AV
165 list_lru_destroy(&s->s_dentry_lru);
166 list_lru_destroy(&s->s_inode_lru);
7eb5e882
AV
167 security_sb_free(s);
168 WARN_ON(!list_empty(&s->s_mounts));
6e4eab57 169 put_user_ns(s->s_user_ns);
7eb5e882
AV
170 kfree(s->s_subtype);
171 kfree(s->s_options);
853b39a7 172 call_rcu(&s->rcu, destroy_super_rcu);
5accdf82
JK
173}
174
1da177e4
LT
175/**
176 * alloc_super - create new superblock
fe2bbc48 177 * @type: filesystem type superblock should belong to
9249e17f 178 * @flags: the mount flags
6e4eab57 179 * @user_ns: User namespace for the super_block
1da177e4
LT
180 *
181 * Allocates and initializes a new &struct super_block. alloc_super()
182 * returns a pointer new superblock or %NULL if allocation had failed.
183 */
6e4eab57
EB
184static struct super_block *alloc_super(struct file_system_type *type, int flags,
185 struct user_namespace *user_ns)
1da177e4 186{
11b0b5ab 187 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
b87221de 188 static const struct super_operations default_op;
7eb5e882
AV
189 int i;
190
191 if (!s)
192 return NULL;
1da177e4 193
b5bd856a 194 INIT_LIST_HEAD(&s->s_mounts);
6e4eab57 195 s->s_user_ns = get_user_ns(user_ns);
b5bd856a 196
7eb5e882
AV
197 if (security_sb_alloc(s))
198 goto fail;
7b7a8665 199
7eb5e882 200 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
8129ed29
ON
201 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
202 sb_writers_name[i],
203 &type->s_writers_key[i]))
7eb5e882 204 goto fail;
1da177e4 205 }
7eb5e882 206 init_waitqueue_head(&s->s_writers.wait_unfrozen);
df0ce26c 207 s->s_bdi = &noop_backing_dev_info;
7eb5e882 208 s->s_flags = flags;
cc50a07a 209 if (s->s_user_ns != &init_user_ns)
67690f93 210 s->s_iflags |= SB_I_NODEV;
7eb5e882
AV
211 INIT_HLIST_NODE(&s->s_instances);
212 INIT_HLIST_BL_HEAD(&s->s_anon);
e97fedb9 213 mutex_init(&s->s_sync_lock);
7eb5e882 214 INIT_LIST_HEAD(&s->s_inodes);
74278da9 215 spin_lock_init(&s->s_inode_list_lock);
6c60d2b5
DC
216 INIT_LIST_HEAD(&s->s_inodes_wb);
217 spin_lock_init(&s->s_inode_wblist_lock);
7eb5e882 218
2acb60a0 219 if (list_lru_init_memcg(&s->s_dentry_lru))
7eb5e882 220 goto fail;
2acb60a0 221 if (list_lru_init_memcg(&s->s_inode_lru))
7eb5e882
AV
222 goto fail;
223
7eb5e882
AV
224 init_rwsem(&s->s_umount);
225 lockdep_set_class(&s->s_umount, &type->s_umount_key);
226 /*
227 * sget() can have s_umount recursion.
228 *
229 * When it cannot find a suitable sb, it allocates a new
230 * one (this one), and tries again to find a suitable old
231 * one.
232 *
233 * In case that succeeds, it will acquire the s_umount
234 * lock of the old one. Since these are clearly distrinct
235 * locks, and this object isn't exposed yet, there's no
236 * risk of deadlocks.
237 *
238 * Annotate this by putting this lock in a different
239 * subclass.
240 */
241 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
242 s->s_count = 1;
243 atomic_set(&s->s_active, 1);
244 mutex_init(&s->s_vfs_rename_mutex);
245 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
246 mutex_init(&s->s_dquot.dqio_mutex);
7eb5e882
AV
247 s->s_maxbytes = MAX_NON_LFS;
248 s->s_op = &default_op;
249 s->s_time_gran = 1000000000;
3cb29d11 250 s->cleancache_poolid = CLEANCACHE_NO_POOL;
7eb5e882
AV
251
252 s->s_shrink.seeks = DEFAULT_SEEKS;
253 s->s_shrink.scan_objects = super_cache_scan;
254 s->s_shrink.count_objects = super_cache_count;
255 s->s_shrink.batch = 1024;
2acb60a0 256 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
1da177e4 257 return s;
5ca302c8 258
7eb5e882
AV
259fail:
260 destroy_super(s);
261 return NULL;
1da177e4
LT
262}
263
264/* Superblock refcounting */
265
266/*
35cf7ba0 267 * Drop a superblock's refcount. The caller must hold sb_lock.
1da177e4 268 */
f47ec3f2 269static void __put_super(struct super_block *sb)
1da177e4 270{
1da177e4 271 if (!--sb->s_count) {
551de6f3 272 list_del_init(&sb->s_list);
1da177e4 273 destroy_super(sb);
1da177e4 274 }
1da177e4
LT
275}
276
277/**
278 * put_super - drop a temporary reference to superblock
279 * @sb: superblock in question
280 *
281 * Drops a temporary reference, frees superblock if there's no
282 * references left.
283 */
f47ec3f2 284static void put_super(struct super_block *sb)
1da177e4
LT
285{
286 spin_lock(&sb_lock);
287 __put_super(sb);
288 spin_unlock(&sb_lock);
289}
290
291
292/**
1712ac8f 293 * deactivate_locked_super - drop an active reference to superblock
1da177e4
LT
294 * @s: superblock to deactivate
295 *
bd7ced98 296 * Drops an active reference to superblock, converting it into a temporary
1712ac8f 297 * one if there is no other active references left. In that case we
1da177e4
LT
298 * tell fs driver to shut it down and drop the temporary reference we
299 * had just acquired.
1712ac8f
AV
300 *
301 * Caller holds exclusive lock on superblock; that lock is released.
1da177e4 302 */
1712ac8f 303void deactivate_locked_super(struct super_block *s)
1da177e4
LT
304{
305 struct file_system_type *fs = s->s_type;
b20bd1a5 306 if (atomic_dec_and_test(&s->s_active)) {
3167760f 307 cleancache_invalidate_fs(s);
b0d40c92 308 unregister_shrinker(&s->s_shrink);
28f2cd4f 309 fs->kill_sb(s);
f5e1dd34 310
c0a5b560
VD
311 /*
312 * Since list_lru_destroy() may sleep, we cannot call it from
313 * put_super(), where we hold the sb_lock. Therefore we destroy
314 * the lru lists right now.
315 */
316 list_lru_destroy(&s->s_dentry_lru);
317 list_lru_destroy(&s->s_inode_lru);
318
1da177e4
LT
319 put_filesystem(fs);
320 put_super(s);
1712ac8f
AV
321 } else {
322 up_write(&s->s_umount);
1da177e4
LT
323 }
324}
325
1712ac8f 326EXPORT_SYMBOL(deactivate_locked_super);
1da177e4 327
74dbbdd7 328/**
1712ac8f 329 * deactivate_super - drop an active reference to superblock
74dbbdd7
AV
330 * @s: superblock to deactivate
331 *
1712ac8f
AV
332 * Variant of deactivate_locked_super(), except that superblock is *not*
333 * locked by caller. If we are going to drop the final active reference,
334 * lock will be acquired prior to that.
74dbbdd7 335 */
1712ac8f 336void deactivate_super(struct super_block *s)
74dbbdd7 337{
1712ac8f
AV
338 if (!atomic_add_unless(&s->s_active, -1, 1)) {
339 down_write(&s->s_umount);
340 deactivate_locked_super(s);
74dbbdd7
AV
341 }
342}
343
1712ac8f 344EXPORT_SYMBOL(deactivate_super);
74dbbdd7 345
1da177e4
LT
346/**
347 * grab_super - acquire an active reference
348 * @s: reference we are trying to make active
349 *
350 * Tries to acquire an active reference. grab_super() is used when we
351 * had just found a superblock in super_blocks or fs_type->fs_supers
352 * and want to turn it into a full-blown active reference. grab_super()
353 * is called with sb_lock held and drops it. Returns 1 in case of
354 * success, 0 if we had failed (superblock contents was already dead or
acfec9a5
AV
355 * dying when grab_super() had been called). Note that this is only
356 * called for superblocks not in rundown mode (== ones still on ->fs_supers
357 * of their type), so increment of ->s_count is OK here.
1da177e4 358 */
9c4dbee7 359static int grab_super(struct super_block *s) __releases(sb_lock)
1da177e4
LT
360{
361 s->s_count++;
362 spin_unlock(&sb_lock);
363 down_write(&s->s_umount);
acfec9a5
AV
364 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
365 put_super(s);
366 return 1;
367 }
1da177e4
LT
368 up_write(&s->s_umount);
369 put_super(s);
1da177e4
LT
370 return 0;
371}
372
12ad3ab6 373/*
eb6ef3df 374 * trylock_super - try to grab ->s_umount shared
331cbdee 375 * @sb: reference we are trying to grab
12ad3ab6 376 *
eb6ef3df 377 * Try to prevent fs shutdown. This is used in places where we
12ad3ab6 378 * cannot take an active reference but we need to ensure that the
eb6ef3df
KK
379 * filesystem is not shut down while we are working on it. It returns
380 * false if we cannot acquire s_umount or if we lose the race and
381 * filesystem already got into shutdown, and returns true with the s_umount
382 * lock held in read mode in case of success. On successful return,
383 * the caller must drop the s_umount lock when done.
384 *
385 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
386 * The reason why it's safe is that we are OK with doing trylock instead
387 * of down_read(). There's a couple of places that are OK with that, but
388 * it's very much not a general-purpose interface.
12ad3ab6 389 */
eb6ef3df 390bool trylock_super(struct super_block *sb)
12ad3ab6 391{
12ad3ab6 392 if (down_read_trylock(&sb->s_umount)) {
eb6ef3df
KK
393 if (!hlist_unhashed(&sb->s_instances) &&
394 sb->s_root && (sb->s_flags & MS_BORN))
12ad3ab6
DC
395 return true;
396 up_read(&sb->s_umount);
397 }
398
12ad3ab6
DC
399 return false;
400}
401
1da177e4
LT
402/**
403 * generic_shutdown_super - common helper for ->kill_sb()
404 * @sb: superblock to kill
405 *
406 * generic_shutdown_super() does all fs-independent work on superblock
407 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
408 * that need destruction out of superblock, call generic_shutdown_super()
409 * and release aforementioned objects. Note: dentries and inodes _are_
410 * taken care of and do not need specific handling.
c636ebdb
DH
411 *
412 * Upon calling this function, the filesystem may no longer alter or
413 * rearrange the set of dentries belonging to this super_block, nor may it
414 * change the attachments of dentries to inodes.
1da177e4
LT
415 */
416void generic_shutdown_super(struct super_block *sb)
417{
ee9b6d61 418 const struct super_operations *sop = sb->s_op;
1da177e4 419
c636ebdb
DH
420 if (sb->s_root) {
421 shrink_dcache_for_umount(sb);
60b0680f 422 sync_filesystem(sb);
1da177e4 423 sb->s_flags &= ~MS_ACTIVE;
efaee192 424
74278da9 425 fsnotify_unmount_inodes(sb);
a1a0e23e 426 cgroup_writeback_umount();
63997e98
AV
427
428 evict_inodes(sb);
1da177e4 429
7b7a8665
CH
430 if (sb->s_dio_done_wq) {
431 destroy_workqueue(sb->s_dio_done_wq);
432 sb->s_dio_done_wq = NULL;
433 }
434
1da177e4
LT
435 if (sop->put_super)
436 sop->put_super(sb);
437
63997e98 438 if (!list_empty(&sb->s_inodes)) {
7b4fe29e
DJ
439 printk("VFS: Busy inodes after unmount of %s. "
440 "Self-destruct in 5 seconds. Have a nice day...\n",
441 sb->s_id);
1da177e4 442 }
1da177e4
LT
443 }
444 spin_lock(&sb_lock);
445 /* should be initialized for __put_super_and_need_restart() */
a5166169 446 hlist_del_init(&sb->s_instances);
1da177e4
LT
447 spin_unlock(&sb_lock);
448 up_write(&sb->s_umount);
449}
450
451EXPORT_SYMBOL(generic_shutdown_super);
452
453/**
6e4eab57 454 * sget_userns - find or create a superblock
1da177e4
LT
455 * @type: filesystem type superblock should belong to
456 * @test: comparison callback
457 * @set: setup callback
9249e17f 458 * @flags: mount flags
6e4eab57 459 * @user_ns: User namespace for the super_block
1da177e4
LT
460 * @data: argument to each of them
461 */
6e4eab57 462struct super_block *sget_userns(struct file_system_type *type,
1da177e4
LT
463 int (*test)(struct super_block *,void *),
464 int (*set)(struct super_block *,void *),
6e4eab57 465 int flags, struct user_namespace *user_ns,
1da177e4
LT
466 void *data)
467{
468 struct super_block *s = NULL;
d4730127 469 struct super_block *old;
1da177e4
LT
470 int err;
471
a001e74c
EB
472 if (!(flags & MS_KERNMOUNT) &&
473 !(type->fs_flags & FS_USERNS_MOUNT) &&
474 !capable(CAP_SYS_ADMIN))
475 return ERR_PTR(-EPERM);
1da177e4
LT
476retry:
477 spin_lock(&sb_lock);
d4730127 478 if (test) {
b67bfe0d 479 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
d4730127
MK
480 if (!test(old, data))
481 continue;
6e4eab57
EB
482 if (user_ns != old->s_user_ns) {
483 spin_unlock(&sb_lock);
484 if (s) {
485 up_write(&s->s_umount);
486 destroy_super(s);
487 }
488 return ERR_PTR(-EBUSY);
489 }
d4730127
MK
490 if (!grab_super(old))
491 goto retry;
a3cfbb53
LZ
492 if (s) {
493 up_write(&s->s_umount);
d4730127 494 destroy_super(s);
7a4dec53 495 s = NULL;
a3cfbb53 496 }
d4730127
MK
497 return old;
498 }
1da177e4
LT
499 }
500 if (!s) {
501 spin_unlock(&sb_lock);
6e4eab57 502 s = alloc_super(type, flags, user_ns);
1da177e4
LT
503 if (!s)
504 return ERR_PTR(-ENOMEM);
505 goto retry;
506 }
507
508 err = set(s, data);
509 if (err) {
510 spin_unlock(&sb_lock);
a3cfbb53 511 up_write(&s->s_umount);
1da177e4
LT
512 destroy_super(s);
513 return ERR_PTR(err);
514 }
515 s->s_type = type;
516 strlcpy(s->s_id, type->name, sizeof(s->s_id));
517 list_add_tail(&s->s_list, &super_blocks);
a5166169 518 hlist_add_head(&s->s_instances, &type->fs_supers);
1da177e4
LT
519 spin_unlock(&sb_lock);
520 get_filesystem(type);
b0d40c92 521 register_shrinker(&s->s_shrink);
1da177e4
LT
522 return s;
523}
524
6e4eab57
EB
525EXPORT_SYMBOL(sget_userns);
526
527/**
528 * sget - find or create a superblock
529 * @type: filesystem type superblock should belong to
530 * @test: comparison callback
531 * @set: setup callback
532 * @flags: mount flags
533 * @data: argument to each of them
534 */
535struct super_block *sget(struct file_system_type *type,
536 int (*test)(struct super_block *,void *),
537 int (*set)(struct super_block *,void *),
538 int flags,
539 void *data)
540{
541 struct user_namespace *user_ns = current_user_ns();
542
543 /* Ensure the requestor has permissions over the target filesystem */
544 if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
545 return ERR_PTR(-EPERM);
546
547 return sget_userns(type, test, set, flags, user_ns, data);
548}
549
1da177e4
LT
550EXPORT_SYMBOL(sget);
551
552void drop_super(struct super_block *sb)
553{
554 up_read(&sb->s_umount);
555 put_super(sb);
556}
557
558EXPORT_SYMBOL(drop_super);
559
ba6379f7
JK
560void drop_super_exclusive(struct super_block *sb)
561{
562 up_write(&sb->s_umount);
563 put_super(sb);
564}
565EXPORT_SYMBOL(drop_super_exclusive);
566
01a05b33
AV
567/**
568 * iterate_supers - call function for all active superblocks
569 * @f: function to call
570 * @arg: argument to pass to it
571 *
572 * Scans the superblock list and calls given function, passing it
573 * locked superblock and given argument.
574 */
575void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
576{
dca33252 577 struct super_block *sb, *p = NULL;
01a05b33
AV
578
579 spin_lock(&sb_lock);
dca33252 580 list_for_each_entry(sb, &super_blocks, s_list) {
a5166169 581 if (hlist_unhashed(&sb->s_instances))
01a05b33
AV
582 continue;
583 sb->s_count++;
584 spin_unlock(&sb_lock);
585
586 down_read(&sb->s_umount);
dabe0dc1 587 if (sb->s_root && (sb->s_flags & MS_BORN))
01a05b33
AV
588 f(sb, arg);
589 up_read(&sb->s_umount);
590
591 spin_lock(&sb_lock);
dca33252
AV
592 if (p)
593 __put_super(p);
594 p = sb;
01a05b33 595 }
dca33252
AV
596 if (p)
597 __put_super(p);
01a05b33
AV
598 spin_unlock(&sb_lock);
599}
600
43e15cdb
AV
601/**
602 * iterate_supers_type - call function for superblocks of given type
603 * @type: fs type
604 * @f: function to call
605 * @arg: argument to pass to it
606 *
607 * Scans the superblock list and calls given function, passing it
608 * locked superblock and given argument.
609 */
610void iterate_supers_type(struct file_system_type *type,
611 void (*f)(struct super_block *, void *), void *arg)
612{
613 struct super_block *sb, *p = NULL;
614
615 spin_lock(&sb_lock);
b67bfe0d 616 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
43e15cdb
AV
617 sb->s_count++;
618 spin_unlock(&sb_lock);
619
620 down_read(&sb->s_umount);
dabe0dc1 621 if (sb->s_root && (sb->s_flags & MS_BORN))
43e15cdb
AV
622 f(sb, arg);
623 up_read(&sb->s_umount);
624
625 spin_lock(&sb_lock);
626 if (p)
627 __put_super(p);
628 p = sb;
629 }
630 if (p)
631 __put_super(p);
632 spin_unlock(&sb_lock);
633}
634
635EXPORT_SYMBOL(iterate_supers_type);
636
ba6379f7 637static struct super_block *__get_super(struct block_device *bdev, bool excl)
1da177e4 638{
618f0636
KK
639 struct super_block *sb;
640
1da177e4
LT
641 if (!bdev)
642 return NULL;
618f0636 643
1da177e4 644 spin_lock(&sb_lock);
618f0636
KK
645rescan:
646 list_for_each_entry(sb, &super_blocks, s_list) {
a5166169 647 if (hlist_unhashed(&sb->s_instances))
551de6f3 648 continue;
618f0636
KK
649 if (sb->s_bdev == bdev) {
650 sb->s_count++;
1da177e4 651 spin_unlock(&sb_lock);
ba6379f7
JK
652 if (!excl)
653 down_read(&sb->s_umount);
654 else
655 down_write(&sb->s_umount);
df40c01a 656 /* still alive? */
dabe0dc1 657 if (sb->s_root && (sb->s_flags & MS_BORN))
618f0636 658 return sb;
ba6379f7
JK
659 if (!excl)
660 up_read(&sb->s_umount);
661 else
662 up_write(&sb->s_umount);
df40c01a 663 /* nope, got unmounted */
618f0636 664 spin_lock(&sb_lock);
df40c01a
AV
665 __put_super(sb);
666 goto rescan;
1da177e4
LT
667 }
668 }
669 spin_unlock(&sb_lock);
670 return NULL;
671}
672
6b6dc836 673/**
ba6379f7 674 * get_super - get the superblock of a device
6b6dc836
JK
675 * @bdev: device to get the superblock for
676 *
677 * Scans the superblock list and finds the superblock of the file system
ba6379f7 678 * mounted on the device given. %NULL is returned if no match is found.
6b6dc836 679 */
ba6379f7
JK
680struct super_block *get_super(struct block_device *bdev)
681{
682 return __get_super(bdev, false);
683}
684EXPORT_SYMBOL(get_super);
685
686static struct super_block *__get_super_thawed(struct block_device *bdev,
687 bool excl)
6b6dc836
JK
688{
689 while (1) {
ba6379f7 690 struct super_block *s = __get_super(bdev, excl);
5accdf82 691 if (!s || s->s_writers.frozen == SB_UNFROZEN)
6b6dc836 692 return s;
ba6379f7
JK
693 if (!excl)
694 up_read(&s->s_umount);
695 else
696 up_write(&s->s_umount);
5accdf82
JK
697 wait_event(s->s_writers.wait_unfrozen,
698 s->s_writers.frozen == SB_UNFROZEN);
6b6dc836
JK
699 put_super(s);
700 }
701}
ba6379f7
JK
702
703/**
704 * get_super_thawed - get thawed superblock of a device
705 * @bdev: device to get the superblock for
706 *
707 * Scans the superblock list and finds the superblock of the file system
708 * mounted on the device. The superblock is returned once it is thawed
709 * (or immediately if it was not frozen). %NULL is returned if no match
710 * is found.
711 */
712struct super_block *get_super_thawed(struct block_device *bdev)
713{
714 return __get_super_thawed(bdev, false);
715}
6b6dc836
JK
716EXPORT_SYMBOL(get_super_thawed);
717
ba6379f7
JK
718/**
719 * get_super_exclusive_thawed - get thawed superblock of a device
720 * @bdev: device to get the superblock for
721 *
722 * Scans the superblock list and finds the superblock of the file system
723 * mounted on the device. The superblock is returned once it is thawed
724 * (or immediately if it was not frozen) and s_umount semaphore is held
725 * in exclusive mode. %NULL is returned if no match is found.
726 */
727struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
728{
729 return __get_super_thawed(bdev, true);
730}
731EXPORT_SYMBOL(get_super_exclusive_thawed);
732
4504230a
CH
733/**
734 * get_active_super - get an active reference to the superblock of a device
735 * @bdev: device to get the superblock for
736 *
737 * Scans the superblock list and finds the superblock of the file system
738 * mounted on the device given. Returns the superblock with an active
d3f21473 739 * reference or %NULL if none was found.
4504230a
CH
740 */
741struct super_block *get_active_super(struct block_device *bdev)
742{
743 struct super_block *sb;
744
745 if (!bdev)
746 return NULL;
747
1494583d 748restart:
4504230a
CH
749 spin_lock(&sb_lock);
750 list_for_each_entry(sb, &super_blocks, s_list) {
a5166169 751 if (hlist_unhashed(&sb->s_instances))
551de6f3 752 continue;
1494583d 753 if (sb->s_bdev == bdev) {
acfec9a5 754 if (!grab_super(sb))
1494583d 755 goto restart;
acfec9a5
AV
756 up_write(&sb->s_umount);
757 return sb;
1494583d 758 }
4504230a
CH
759 }
760 spin_unlock(&sb_lock);
761 return NULL;
762}
1da177e4 763
df40c01a 764struct super_block *user_get_super(dev_t dev)
1da177e4 765{
618f0636 766 struct super_block *sb;
1da177e4 767
1da177e4 768 spin_lock(&sb_lock);
618f0636
KK
769rescan:
770 list_for_each_entry(sb, &super_blocks, s_list) {
a5166169 771 if (hlist_unhashed(&sb->s_instances))
551de6f3 772 continue;
618f0636
KK
773 if (sb->s_dev == dev) {
774 sb->s_count++;
1da177e4 775 spin_unlock(&sb_lock);
618f0636 776 down_read(&sb->s_umount);
df40c01a 777 /* still alive? */
dabe0dc1 778 if (sb->s_root && (sb->s_flags & MS_BORN))
618f0636
KK
779 return sb;
780 up_read(&sb->s_umount);
df40c01a 781 /* nope, got unmounted */
618f0636 782 spin_lock(&sb_lock);
df40c01a
AV
783 __put_super(sb);
784 goto rescan;
1da177e4
LT
785 }
786 }
787 spin_unlock(&sb_lock);
788 return NULL;
789}
790
1da177e4
LT
791/**
792 * do_remount_sb - asks filesystem to change mount options.
793 * @sb: superblock in question
794 * @flags: numeric part of options
795 * @data: the rest of options
796 * @force: whether or not to force the change
797 *
798 * Alters the mount options of a mounted file system.
799 */
800int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
801{
802 int retval;
c79d967d 803 int remount_ro;
4504230a 804
5accdf82 805 if (sb->s_writers.frozen != SB_UNFROZEN)
4504230a
CH
806 return -EBUSY;
807
9361401e 808#ifdef CONFIG_BLOCK
1da177e4
LT
809 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
810 return -EACCES;
9361401e 811#endif
4504230a 812
d208bbdd 813 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
d208bbdd 814
0aec09d0 815 if (remount_ro) {
fdab684d 816 if (!hlist_empty(&sb->s_pins)) {
0aec09d0 817 up_write(&sb->s_umount);
fdab684d 818 group_pin_kill(&sb->s_pins);
0aec09d0
AV
819 down_write(&sb->s_umount);
820 if (!sb->s_root)
821 return 0;
822 if (sb->s_writers.frozen != SB_UNFROZEN)
823 return -EBUSY;
824 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
825 }
826 }
827 shrink_dcache_sb(sb);
828
1da177e4
LT
829 /* If we are remounting RDONLY and current sb is read/write,
830 make sure there are no rw files opened */
d208bbdd 831 if (remount_ro) {
4ed5e82f 832 if (force) {
eee5cc27
AV
833 sb->s_readonly_remount = 1;
834 smp_wmb();
4ed5e82f
MS
835 } else {
836 retval = sb_prepare_remount_readonly(sb);
837 if (retval)
838 return retval;
4ed5e82f 839 }
1da177e4
LT
840 }
841
842 if (sb->s_op->remount_fs) {
1da177e4 843 retval = sb->s_op->remount_fs(sb, &flags, data);
2833eb2b
MS
844 if (retval) {
845 if (!force)
4ed5e82f 846 goto cancel_readonly;
2833eb2b
MS
847 /* If forced remount, go ahead despite any errors */
848 WARN(1, "forced remount of a %s fs returned %i\n",
849 sb->s_type->name, retval);
850 }
1da177e4
LT
851 }
852 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
4ed5e82f
MS
853 /* Needs to be ordered wrt mnt_is_readonly() */
854 smp_wmb();
855 sb->s_readonly_remount = 0;
c79d967d 856
d208bbdd
NP
857 /*
858 * Some filesystems modify their metadata via some other path than the
859 * bdev buffer cache (eg. use a private mapping, or directories in
860 * pagecache, etc). Also file data modifications go via their own
861 * mappings. So If we try to mount readonly then copy the filesystem
862 * from bdev, we could get stale data, so invalidate it to give a best
863 * effort at coherency.
864 */
865 if (remount_ro && sb->s_bdev)
866 invalidate_bdev(sb->s_bdev);
1da177e4 867 return 0;
4ed5e82f
MS
868
869cancel_readonly:
870 sb->s_readonly_remount = 0;
871 return retval;
1da177e4
LT
872}
873
a2a9537a 874static void do_emergency_remount(struct work_struct *work)
1da177e4 875{
dca33252 876 struct super_block *sb, *p = NULL;
1da177e4
LT
877
878 spin_lock(&sb_lock);
dca33252 879 list_for_each_entry(sb, &super_blocks, s_list) {
a5166169 880 if (hlist_unhashed(&sb->s_instances))
551de6f3 881 continue;
1da177e4
LT
882 sb->s_count++;
883 spin_unlock(&sb_lock);
443b94ba 884 down_write(&sb->s_umount);
dabe0dc1
AV
885 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
886 !(sb->s_flags & MS_RDONLY)) {
1da177e4 887 /*
1da177e4
LT
888 * What lock protects sb->s_flags??
889 */
1da177e4 890 do_remount_sb(sb, MS_RDONLY, NULL, 1);
1da177e4 891 }
443b94ba 892 up_write(&sb->s_umount);
1da177e4 893 spin_lock(&sb_lock);
dca33252
AV
894 if (p)
895 __put_super(p);
896 p = sb;
1da177e4 897 }
dca33252
AV
898 if (p)
899 __put_super(p);
1da177e4 900 spin_unlock(&sb_lock);
a2a9537a 901 kfree(work);
1da177e4
LT
902 printk("Emergency Remount complete\n");
903}
904
905void emergency_remount(void)
906{
a2a9537a
JA
907 struct work_struct *work;
908
909 work = kmalloc(sizeof(*work), GFP_ATOMIC);
910 if (work) {
911 INIT_WORK(work, do_emergency_remount);
912 schedule_work(work);
913 }
1da177e4
LT
914}
915
916/*
917 * Unnamed block devices are dummy devices used by virtual
918 * filesystems which don't use real block-devices. -- jrs
919 */
920
ad76cbc6 921static DEFINE_IDA(unnamed_dev_ida);
1da177e4 922static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
a2a4dc49
TB
923/* Many userspace utilities consider an FSID of 0 invalid.
924 * Always return at least 1 from get_anon_bdev.
925 */
926static int unnamed_dev_start = 1;
1da177e4 927
0ee5dc67 928int get_anon_bdev(dev_t *p)
1da177e4
LT
929{
930 int dev;
931 int error;
932
933 retry:
ad76cbc6 934 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
1da177e4
LT
935 return -ENOMEM;
936 spin_lock(&unnamed_dev_lock);
c63e09ec 937 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
f21f6220
AV
938 if (!error)
939 unnamed_dev_start = dev + 1;
1da177e4
LT
940 spin_unlock(&unnamed_dev_lock);
941 if (error == -EAGAIN)
942 /* We raced and lost with another CPU. */
943 goto retry;
944 else if (error)
945 return -EAGAIN;
946
1af95de6 947 if (dev >= (1 << MINORBITS)) {
1da177e4 948 spin_lock(&unnamed_dev_lock);
ad76cbc6 949 ida_remove(&unnamed_dev_ida, dev);
f21f6220
AV
950 if (unnamed_dev_start > dev)
951 unnamed_dev_start = dev;
1da177e4
LT
952 spin_unlock(&unnamed_dev_lock);
953 return -EMFILE;
954 }
0ee5dc67 955 *p = MKDEV(0, dev & MINORMASK);
1da177e4
LT
956 return 0;
957}
0ee5dc67 958EXPORT_SYMBOL(get_anon_bdev);
1da177e4 959
0ee5dc67 960void free_anon_bdev(dev_t dev)
1da177e4 961{
0ee5dc67 962 int slot = MINOR(dev);
1da177e4 963 spin_lock(&unnamed_dev_lock);
ad76cbc6 964 ida_remove(&unnamed_dev_ida, slot);
c63e09ec
AV
965 if (slot < unnamed_dev_start)
966 unnamed_dev_start = slot;
1da177e4
LT
967 spin_unlock(&unnamed_dev_lock);
968}
0ee5dc67
AV
969EXPORT_SYMBOL(free_anon_bdev);
970
971int set_anon_super(struct super_block *s, void *data)
972{
df0ce26c 973 return get_anon_bdev(&s->s_dev);
0ee5dc67
AV
974}
975
976EXPORT_SYMBOL(set_anon_super);
977
978void kill_anon_super(struct super_block *sb)
979{
980 dev_t dev = sb->s_dev;
981 generic_shutdown_super(sb);
982 free_anon_bdev(dev);
983}
1da177e4
LT
984
985EXPORT_SYMBOL(kill_anon_super);
986
1da177e4
LT
987void kill_litter_super(struct super_block *sb)
988{
989 if (sb->s_root)
990 d_genocide(sb->s_root);
991 kill_anon_super(sb);
992}
993
994EXPORT_SYMBOL(kill_litter_super);
995
909e6d94
SH
996static int ns_test_super(struct super_block *sb, void *data)
997{
998 return sb->s_fs_info == data;
999}
1000
1001static int ns_set_super(struct super_block *sb, void *data)
1002{
1003 sb->s_fs_info = data;
1004 return set_anon_super(sb, NULL);
1005}
1006
d91ee87d
EB
1007struct dentry *mount_ns(struct file_system_type *fs_type,
1008 int flags, void *data, void *ns, struct user_namespace *user_ns,
1009 int (*fill_super)(struct super_block *, void *, int))
909e6d94
SH
1010{
1011 struct super_block *sb;
1012
d91ee87d
EB
1013 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1014 * over the namespace.
1015 */
1016 if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
1017 return ERR_PTR(-EPERM);
1018
6e4eab57
EB
1019 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1020 user_ns, ns);
909e6d94 1021 if (IS_ERR(sb))
ceefda69 1022 return ERR_CAST(sb);
909e6d94
SH
1023
1024 if (!sb->s_root) {
1025 int err;
909e6d94
SH
1026 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
1027 if (err) {
74dbbdd7 1028 deactivate_locked_super(sb);
ceefda69 1029 return ERR_PTR(err);
909e6d94
SH
1030 }
1031
1032 sb->s_flags |= MS_ACTIVE;
1033 }
1034
ceefda69 1035 return dget(sb->s_root);
909e6d94
SH
1036}
1037
ceefda69 1038EXPORT_SYMBOL(mount_ns);
909e6d94 1039
9361401e 1040#ifdef CONFIG_BLOCK
1da177e4
LT
1041static int set_bdev_super(struct super_block *s, void *data)
1042{
1043 s->s_bdev = data;
1044 s->s_dev = s->s_bdev->bd_dev;
32a88aa1
JA
1045
1046 /*
1047 * We set the bdi here to the queue backing, file systems can
1048 * overwrite this in ->fill_super()
1049 */
1050 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1da177e4
LT
1051 return 0;
1052}
1053
1054static int test_bdev_super(struct super_block *s, void *data)
1055{
1056 return (void *)s->s_bdev == data;
1057}
1058
152a0836 1059struct dentry *mount_bdev(struct file_system_type *fs_type,
1da177e4 1060 int flags, const char *dev_name, void *data,
152a0836 1061 int (*fill_super)(struct super_block *, void *, int))
1da177e4
LT
1062{
1063 struct block_device *bdev;
1064 struct super_block *s;
d4d77629 1065 fmode_t mode = FMODE_READ | FMODE_EXCL;
1da177e4
LT
1066 int error = 0;
1067
30c40d2c
AV
1068 if (!(flags & MS_RDONLY))
1069 mode |= FMODE_WRITE;
1070
d4d77629 1071 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1da177e4 1072 if (IS_ERR(bdev))
152a0836 1073 return ERR_CAST(bdev);
1da177e4
LT
1074
1075 /*
1076 * once the super is inserted into the list by sget, s_umount
1077 * will protect the lockfs code from trying to start a snapshot
1078 * while we are mounting
1079 */
4fadd7bb
CH
1080 mutex_lock(&bdev->bd_fsfreeze_mutex);
1081 if (bdev->bd_fsfreeze_count > 0) {
1082 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1083 error = -EBUSY;
1084 goto error_bdev;
1085 }
9249e17f
DH
1086 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1087 bdev);
4fadd7bb 1088 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1da177e4 1089 if (IS_ERR(s))
454e2398 1090 goto error_s;
1da177e4
LT
1091
1092 if (s->s_root) {
1093 if ((flags ^ s->s_flags) & MS_RDONLY) {
74dbbdd7 1094 deactivate_locked_super(s);
454e2398
DH
1095 error = -EBUSY;
1096 goto error_bdev;
1da177e4 1097 }
454e2398 1098
4f331f01
TH
1099 /*
1100 * s_umount nests inside bd_mutex during
e525fd89
TH
1101 * __invalidate_device(). blkdev_put() acquires
1102 * bd_mutex and can't be called under s_umount. Drop
1103 * s_umount temporarily. This is safe as we're
1104 * holding an active reference.
4f331f01
TH
1105 */
1106 up_write(&s->s_umount);
d4d77629 1107 blkdev_put(bdev, mode);
4f331f01 1108 down_write(&s->s_umount);
1da177e4 1109 } else {
30c40d2c 1110 s->s_mode = mode;
a1c6f057 1111 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
e78c9a00 1112 sb_set_blocksize(s, block_size(bdev));
9b04c997 1113 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1da177e4 1114 if (error) {
74dbbdd7 1115 deactivate_locked_super(s);
454e2398 1116 goto error;
fa675765 1117 }
454e2398
DH
1118
1119 s->s_flags |= MS_ACTIVE;
87d8fe1e 1120 bdev->bd_super = s;
1da177e4
LT
1121 }
1122
152a0836 1123 return dget(s->s_root);
1da177e4 1124
454e2398
DH
1125error_s:
1126 error = PTR_ERR(s);
1127error_bdev:
d4d77629 1128 blkdev_put(bdev, mode);
454e2398 1129error:
152a0836
AV
1130 return ERR_PTR(error);
1131}
1132EXPORT_SYMBOL(mount_bdev);
1133
1da177e4
LT
1134void kill_block_super(struct super_block *sb)
1135{
1136 struct block_device *bdev = sb->s_bdev;
30c40d2c 1137 fmode_t mode = sb->s_mode;
1da177e4 1138
ddbaaf30 1139 bdev->bd_super = NULL;
1da177e4
LT
1140 generic_shutdown_super(sb);
1141 sync_blockdev(bdev);
d4d77629 1142 WARN_ON_ONCE(!(mode & FMODE_EXCL));
e525fd89 1143 blkdev_put(bdev, mode | FMODE_EXCL);
1da177e4
LT
1144}
1145
1146EXPORT_SYMBOL(kill_block_super);
9361401e 1147#endif
1da177e4 1148
3c26ff6e 1149struct dentry *mount_nodev(struct file_system_type *fs_type,
1da177e4 1150 int flags, void *data,
3c26ff6e 1151 int (*fill_super)(struct super_block *, void *, int))
1da177e4
LT
1152{
1153 int error;
9249e17f 1154 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1da177e4
LT
1155
1156 if (IS_ERR(s))
3c26ff6e 1157 return ERR_CAST(s);
1da177e4 1158
9b04c997 1159 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1da177e4 1160 if (error) {
74dbbdd7 1161 deactivate_locked_super(s);
3c26ff6e 1162 return ERR_PTR(error);
1da177e4
LT
1163 }
1164 s->s_flags |= MS_ACTIVE;
3c26ff6e 1165 return dget(s->s_root);
1da177e4 1166}
3c26ff6e
AV
1167EXPORT_SYMBOL(mount_nodev);
1168
1da177e4
LT
1169static int compare_single(struct super_block *s, void *p)
1170{
1171 return 1;
1172}
1173
fc14f2fe 1174struct dentry *mount_single(struct file_system_type *fs_type,
1da177e4 1175 int flags, void *data,
fc14f2fe 1176 int (*fill_super)(struct super_block *, void *, int))
1da177e4
LT
1177{
1178 struct super_block *s;
1179 int error;
1180
9249e17f 1181 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1da177e4 1182 if (IS_ERR(s))
fc14f2fe 1183 return ERR_CAST(s);
1da177e4 1184 if (!s->s_root) {
9b04c997 1185 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1da177e4 1186 if (error) {
74dbbdd7 1187 deactivate_locked_super(s);
fc14f2fe 1188 return ERR_PTR(error);
1da177e4
LT
1189 }
1190 s->s_flags |= MS_ACTIVE;
9329d1be
KS
1191 } else {
1192 do_remount_sb(s, flags, data, 0);
1da177e4 1193 }
fc14f2fe
AV
1194 return dget(s->s_root);
1195}
1196EXPORT_SYMBOL(mount_single);
1197
9d412a43
AV
1198struct dentry *
1199mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1da177e4 1200{
c96e41e9 1201 struct dentry *root;
9d412a43 1202 struct super_block *sb;
1da177e4 1203 char *secdata = NULL;
9d412a43 1204 int error = -ENOMEM;
8089352a 1205
e0007529 1206 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1da177e4 1207 secdata = alloc_secdata();
454e2398 1208 if (!secdata)
9d412a43 1209 goto out;
1da177e4 1210
e0007529 1211 error = security_sb_copy_data(data, secdata);
454e2398 1212 if (error)
1da177e4 1213 goto out_free_secdata;
1da177e4
LT
1214 }
1215
1a102ff9
AV
1216 root = type->mount(type, flags, name, data);
1217 if (IS_ERR(root)) {
1218 error = PTR_ERR(root);
1219 goto out_free_secdata;
c96e41e9 1220 }
9d412a43
AV
1221 sb = root->d_sb;
1222 BUG_ON(!sb);
1223 WARN_ON(!sb->s_bdi);
1224 sb->s_flags |= MS_BORN;
454e2398 1225
9d412a43 1226 error = security_sb_kern_mount(sb, flags, secdata);
5129a469
JE
1227 if (error)
1228 goto out_sb;
454e2398 1229
42cb56ae
JL
1230 /*
1231 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1232 * but s_maxbytes was an unsigned long long for many releases. Throw
1233 * this warning for a little while to try and catch filesystems that
4358b567 1234 * violate this rule.
42cb56ae 1235 */
9d412a43
AV
1236 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1237 "negative value (%lld)\n", type->name, sb->s_maxbytes);
42cb56ae 1238
9d412a43 1239 up_write(&sb->s_umount);
8680e22f 1240 free_secdata(secdata);
9d412a43 1241 return root;
1da177e4 1242out_sb:
9d412a43
AV
1243 dput(root);
1244 deactivate_locked_super(sb);
1da177e4
LT
1245out_free_secdata:
1246 free_secdata(secdata);
1da177e4 1247out:
454e2398 1248 return ERR_PTR(error);
1da177e4
LT
1249}
1250
5accdf82
JK
1251/*
1252 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1253 * instead.
1254 */
1255void __sb_end_write(struct super_block *sb, int level)
1256{
8129ed29 1257 percpu_up_read(sb->s_writers.rw_sem + level-1);
5accdf82
JK
1258}
1259EXPORT_SYMBOL(__sb_end_write);
1260
f4b554af
ON
1261/*
1262 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1263 * instead.
1264 */
1265int __sb_start_write(struct super_block *sb, int level, bool wait)
1266{
1267 bool force_trylock = false;
8129ed29 1268 int ret = 1;
f4b554af
ON
1269
1270#ifdef CONFIG_LOCKDEP
1271 /*
1272 * We want lockdep to tell us about possible deadlocks with freezing
1273 * but it's it bit tricky to properly instrument it. Getting a freeze
1274 * protection works as getting a read lock but there are subtle
1275 * problems. XFS for example gets freeze protection on internal level
1276 * twice in some cases, which is OK only because we already hold a
1277 * freeze protection also on higher level. Due to these cases we have
1278 * to use wait == F (trylock mode) which must not fail.
1279 */
1280 if (wait) {
1281 int i;
1282
1283 for (i = 0; i < level - 1; i++)
8129ed29 1284 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
f4b554af
ON
1285 force_trylock = true;
1286 break;
1287 }
1288 }
1289#endif
8129ed29
ON
1290 if (wait && !force_trylock)
1291 percpu_down_read(sb->s_writers.rw_sem + level-1);
1292 else
1293 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1294
22224a17 1295 WARN_ON(force_trylock && !ret);
f4b554af
ON
1296 return ret;
1297}
5accdf82
JK
1298EXPORT_SYMBOL(__sb_start_write);
1299
1300/**
1301 * sb_wait_write - wait until all writers to given file system finish
1302 * @sb: the super for which we wait
1303 * @level: type of writers we wait for (normal vs page fault)
1304 *
1305 * This function waits until there are no writers of given type to given file
8129ed29 1306 * system.
5accdf82
JK
1307 */
1308static void sb_wait_write(struct super_block *sb, int level)
1309{
8129ed29 1310 percpu_down_write(sb->s_writers.rw_sem + level-1);
8129ed29 1311}
5accdf82 1312
f1a96220
ON
1313/*
1314 * We are going to return to userspace and forget about these locks, the
1315 * ownership goes to the caller of thaw_super() which does unlock().
1316 */
1317static void lockdep_sb_freeze_release(struct super_block *sb)
1318{
1319 int level;
1320
1321 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1322 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1323}
1324
1325/*
1326 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1327 */
1328static void lockdep_sb_freeze_acquire(struct super_block *sb)
8129ed29
ON
1329{
1330 int level;
5accdf82 1331
8129ed29
ON
1332 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1333 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
f1a96220
ON
1334}
1335
1336static void sb_freeze_unlock(struct super_block *sb)
1337{
1338 int level;
5accdf82 1339
8129ed29
ON
1340 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1341 percpu_up_write(sb->s_writers.rw_sem + level);
5accdf82
JK
1342}
1343
18e9e510 1344/**
7000d3c4
RD
1345 * freeze_super - lock the filesystem and force it into a consistent state
1346 * @sb: the super to lock
18e9e510
JB
1347 *
1348 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1349 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1350 * -EBUSY.
5accdf82
JK
1351 *
1352 * During this function, sb->s_writers.frozen goes through these values:
1353 *
1354 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1355 *
1356 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1357 * writes should be blocked, though page faults are still allowed. We wait for
1358 * all writes to complete and then proceed to the next stage.
1359 *
1360 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1361 * but internal fs threads can still modify the filesystem (although they
1362 * should not dirty new pages or inodes), writeback can run etc. After waiting
1363 * for all running page faults we sync the filesystem which will clean all
1364 * dirty pages and inodes (no new dirty pages or inodes can be created when
1365 * sync is running).
1366 *
1367 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1368 * modification are blocked (e.g. XFS preallocation truncation on inode
1369 * reclaim). This is usually implemented by blocking new transactions for
1370 * filesystems that have them and need this additional guard. After all
1371 * internal writers are finished we call ->freeze_fs() to finish filesystem
1372 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1373 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1374 *
1375 * sb->s_writers.frozen is protected by sb->s_umount.
18e9e510
JB
1376 */
1377int freeze_super(struct super_block *sb)
1378{
1379 int ret;
1380
1381 atomic_inc(&sb->s_active);
1382 down_write(&sb->s_umount);
5accdf82 1383 if (sb->s_writers.frozen != SB_UNFROZEN) {
18e9e510
JB
1384 deactivate_locked_super(sb);
1385 return -EBUSY;
1386 }
1387
dabe0dc1
AV
1388 if (!(sb->s_flags & MS_BORN)) {
1389 up_write(&sb->s_umount);
1390 return 0; /* sic - it's "nothing to do" */
1391 }
1392
18e9e510 1393 if (sb->s_flags & MS_RDONLY) {
5accdf82
JK
1394 /* Nothing to do really... */
1395 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
18e9e510
JB
1396 up_write(&sb->s_umount);
1397 return 0;
1398 }
1399
5accdf82 1400 sb->s_writers.frozen = SB_FREEZE_WRITE;
5accdf82
JK
1401 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1402 up_write(&sb->s_umount);
5accdf82 1403 sb_wait_write(sb, SB_FREEZE_WRITE);
8129ed29 1404 down_write(&sb->s_umount);
5accdf82
JK
1405
1406 /* Now we go and block page faults... */
5accdf82 1407 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
5accdf82
JK
1408 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1409
1410 /* All writers are done so after syncing there won't be dirty data */
18e9e510
JB
1411 sync_filesystem(sb);
1412
5accdf82
JK
1413 /* Now wait for internal filesystem counter */
1414 sb->s_writers.frozen = SB_FREEZE_FS;
5accdf82 1415 sb_wait_write(sb, SB_FREEZE_FS);
18e9e510 1416
18e9e510
JB
1417 if (sb->s_op->freeze_fs) {
1418 ret = sb->s_op->freeze_fs(sb);
1419 if (ret) {
1420 printk(KERN_ERR
1421 "VFS:Filesystem freeze failed\n");
5accdf82 1422 sb->s_writers.frozen = SB_UNFROZEN;
8129ed29 1423 sb_freeze_unlock(sb);
5accdf82 1424 wake_up(&sb->s_writers.wait_unfrozen);
18e9e510
JB
1425 deactivate_locked_super(sb);
1426 return ret;
1427 }
1428 }
5accdf82 1429 /*
89f39af1
ON
1430 * For debugging purposes so that fs can warn if it sees write activity
1431 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
5accdf82
JK
1432 */
1433 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
f1a96220 1434 lockdep_sb_freeze_release(sb);
18e9e510
JB
1435 up_write(&sb->s_umount);
1436 return 0;
1437}
1438EXPORT_SYMBOL(freeze_super);
1439
1440/**
1441 * thaw_super -- unlock filesystem
1442 * @sb: the super to thaw
1443 *
1444 * Unlocks the filesystem and marks it writeable again after freeze_super().
1445 */
1446int thaw_super(struct super_block *sb)
1447{
1448 int error;
1449
1450 down_write(&sb->s_umount);
89f39af1 1451 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
18e9e510
JB
1452 up_write(&sb->s_umount);
1453 return -EINVAL;
1454 }
1455
8129ed29
ON
1456 if (sb->s_flags & MS_RDONLY) {
1457 sb->s_writers.frozen = SB_UNFROZEN;
18e9e510 1458 goto out;
8129ed29 1459 }
18e9e510 1460
f1a96220
ON
1461 lockdep_sb_freeze_acquire(sb);
1462
18e9e510
JB
1463 if (sb->s_op->unfreeze_fs) {
1464 error = sb->s_op->unfreeze_fs(sb);
1465 if (error) {
1466 printk(KERN_ERR
1467 "VFS:Filesystem thaw failed\n");
f1a96220 1468 lockdep_sb_freeze_release(sb);
18e9e510
JB
1469 up_write(&sb->s_umount);
1470 return error;
1471 }
1472 }
1473
5accdf82 1474 sb->s_writers.frozen = SB_UNFROZEN;
8129ed29
ON
1475 sb_freeze_unlock(sb);
1476out:
5accdf82 1477 wake_up(&sb->s_writers.wait_unfrozen);
18e9e510 1478 deactivate_locked_super(sb);
18e9e510
JB
1479 return 0;
1480}
1481EXPORT_SYMBOL(thaw_super);