]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - fs/super.c
Merge branch 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[mirror_ubuntu-eoan-kernel.git] / fs / super.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
1da177e4
LT
2/*
3 * linux/fs/super.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * super.c contains code to handle: - mount structures
8 * - super-block tables
9 * - filesystem drivers list
10 * - mount system call
11 * - umount system call
12 * - ustat system call
13 *
14 * GK 2/5/95 - Changed to support mounting the root fs via NFS
15 *
16 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
17 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
18 * Added options to /proc/mounts:
96de0e25 19 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
1da177e4
LT
20 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
21 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
22 */
23
630d9c47 24#include <linux/export.h>
1da177e4 25#include <linux/slab.h>
1da177e4 26#include <linux/blkdev.h>
1da177e4
LT
27#include <linux/mount.h>
28#include <linux/security.h>
1da177e4
LT
29#include <linux/writeback.h> /* for the emergency remount stuff */
30#include <linux/idr.h>
353ab6e9 31#include <linux/mutex.h>
5477d0fa 32#include <linux/backing-dev.h>
ceb5bdc2 33#include <linux/rculist_bl.h>
c515e1fd 34#include <linux/cleancache.h>
40401530 35#include <linux/fsnotify.h>
5accdf82 36#include <linux/lockdep.h>
6e4eab57 37#include <linux/user_namespace.h>
6d59e7f5 38#include "internal.h"
1da177e4 39
08fdc8a0 40static int thaw_super_locked(struct super_block *sb);
1da177e4 41
15d0f5ea
AV
42static LIST_HEAD(super_blocks);
43static DEFINE_SPINLOCK(sb_lock);
1da177e4 44
5accdf82
JK
45static char *sb_writers_name[SB_FREEZE_LEVELS] = {
46 "sb_writers",
47 "sb_pagefaults",
48 "sb_internal",
49};
50
b0d40c92
DC
51/*
52 * One thing we have to be careful of with a per-sb shrinker is that we don't
53 * drop the last active reference to the superblock from within the shrinker.
54 * If that happens we could trigger unregistering the shrinker from within the
55 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
56 * take a passive reference to the superblock to avoid this from occurring.
57 */
0a234c6d
DC
58static unsigned long super_cache_scan(struct shrinker *shrink,
59 struct shrink_control *sc)
b0d40c92
DC
60{
61 struct super_block *sb;
0a234c6d
DC
62 long fs_objects = 0;
63 long total_objects;
64 long freed = 0;
65 long dentries;
66 long inodes;
b0d40c92
DC
67
68 sb = container_of(shrink, struct super_block, s_shrink);
69
70 /*
71 * Deadlock avoidance. We may hold various FS locks, and we don't want
72 * to recurse into the FS that called us in clear_inode() and friends..
73 */
0a234c6d
DC
74 if (!(sc->gfp_mask & __GFP_FS))
75 return SHRINK_STOP;
b0d40c92 76
eb6ef3df 77 if (!trylock_super(sb))
0a234c6d 78 return SHRINK_STOP;
b0d40c92 79
d0407903 80 if (sb->s_op->nr_cached_objects)
4101b624 81 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
0e1fdafd 82
503c358c
VD
83 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
84 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
f6041567 85 total_objects = dentries + inodes + fs_objects + 1;
475d0db7
TH
86 if (!total_objects)
87 total_objects = 1;
0e1fdafd 88
0a234c6d 89 /* proportion the scan between the caches */
f6041567 90 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
bc3b14cb 91 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
503c358c 92 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
b0d40c92 93
0a234c6d
DC
94 /*
95 * prune the dcache first as the icache is pinned by it, then
96 * prune the icache, followed by the filesystem specific caches
49e7e7ff
VD
97 *
98 * Ensure that we always scan at least one object - memcg kmem
99 * accounting uses this to fully empty the caches.
0a234c6d 100 */
49e7e7ff 101 sc->nr_to_scan = dentries + 1;
503c358c 102 freed = prune_dcache_sb(sb, sc);
49e7e7ff 103 sc->nr_to_scan = inodes + 1;
503c358c 104 freed += prune_icache_sb(sb, sc);
0a234c6d
DC
105
106 if (fs_objects) {
49e7e7ff 107 sc->nr_to_scan = fs_objects + 1;
4101b624 108 freed += sb->s_op->free_cached_objects(sb, sc);
b0d40c92
DC
109 }
110
eb6ef3df 111 up_read(&sb->s_umount);
0a234c6d
DC
112 return freed;
113}
114
115static unsigned long super_cache_count(struct shrinker *shrink,
116 struct shrink_control *sc)
117{
118 struct super_block *sb;
119 long total_objects = 0;
120
121 sb = container_of(shrink, struct super_block, s_shrink);
122
d23da150 123 /*
79f546a6
DC
124 * We don't call trylock_super() here as it is a scalability bottleneck,
125 * so we're exposed to partial setup state. The shrinker rwsem does not
126 * protect filesystem operations backing list_lru_shrink_count() or
127 * s_op->nr_cached_objects(). Counts can change between
128 * super_cache_count and super_cache_scan, so we really don't need locks
129 * here.
130 *
131 * However, if we are currently mounting the superblock, the underlying
132 * filesystem might be in a state of partial construction and hence it
133 * is dangerous to access it. trylock_super() uses a SB_BORN check to
134 * avoid this situation, so do the same here. The memory barrier is
135 * matched with the one in mount_fs() as we don't hold locks here.
d23da150 136 */
79f546a6
DC
137 if (!(sb->s_flags & SB_BORN))
138 return 0;
139 smp_rmb();
140
0a234c6d 141 if (sb->s_op && sb->s_op->nr_cached_objects)
4101b624 142 total_objects = sb->s_op->nr_cached_objects(sb, sc);
0a234c6d 143
503c358c
VD
144 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
145 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
0a234c6d 146
9b996468
KT
147 if (!total_objects)
148 return SHRINK_EMPTY;
149
55f841ce 150 total_objects = vfs_pressure_ratio(total_objects);
0e1fdafd 151 return total_objects;
b0d40c92
DC
152}
153
853b39a7
ON
154static void destroy_super_work(struct work_struct *work)
155{
156 struct super_block *s = container_of(work, struct super_block,
157 destroy_work);
158 int i;
159
160 for (i = 0; i < SB_FREEZE_LEVELS; i++)
8129ed29 161 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
853b39a7
ON
162 kfree(s);
163}
164
165static void destroy_super_rcu(struct rcu_head *head)
166{
167 struct super_block *s = container_of(head, struct super_block, rcu);
168 INIT_WORK(&s->destroy_work, destroy_super_work);
169 schedule_work(&s->destroy_work);
170}
171
0200894d
AV
172/* Free a superblock that has never been seen by anyone */
173static void destroy_unused_super(struct super_block *s)
5accdf82 174{
0200894d
AV
175 if (!s)
176 return;
177 up_write(&s->s_umount);
7eb5e882
AV
178 list_lru_destroy(&s->s_dentry_lru);
179 list_lru_destroy(&s->s_inode_lru);
7eb5e882 180 security_sb_free(s);
6e4eab57 181 put_user_ns(s->s_user_ns);
7eb5e882 182 kfree(s->s_subtype);
8e04944f 183 free_prealloced_shrinker(&s->s_shrink);
0200894d
AV
184 /* no delays needed */
185 destroy_super_work(&s->destroy_work);
5accdf82
JK
186}
187
1da177e4
LT
188/**
189 * alloc_super - create new superblock
fe2bbc48 190 * @type: filesystem type superblock should belong to
9249e17f 191 * @flags: the mount flags
6e4eab57 192 * @user_ns: User namespace for the super_block
1da177e4
LT
193 *
194 * Allocates and initializes a new &struct super_block. alloc_super()
195 * returns a pointer new superblock or %NULL if allocation had failed.
196 */
6e4eab57
EB
197static struct super_block *alloc_super(struct file_system_type *type, int flags,
198 struct user_namespace *user_ns)
1da177e4 199{
11b0b5ab 200 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
b87221de 201 static const struct super_operations default_op;
7eb5e882
AV
202 int i;
203
204 if (!s)
205 return NULL;
1da177e4 206
b5bd856a 207 INIT_LIST_HEAD(&s->s_mounts);
6e4eab57 208 s->s_user_ns = get_user_ns(user_ns);
ca0168e8
AV
209 init_rwsem(&s->s_umount);
210 lockdep_set_class(&s->s_umount, &type->s_umount_key);
211 /*
212 * sget() can have s_umount recursion.
213 *
214 * When it cannot find a suitable sb, it allocates a new
215 * one (this one), and tries again to find a suitable old
216 * one.
217 *
218 * In case that succeeds, it will acquire the s_umount
219 * lock of the old one. Since these are clearly distrinct
220 * locks, and this object isn't exposed yet, there's no
221 * risk of deadlocks.
222 *
223 * Annotate this by putting this lock in a different
224 * subclass.
225 */
226 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
b5bd856a 227
7eb5e882
AV
228 if (security_sb_alloc(s))
229 goto fail;
7b7a8665 230
7eb5e882 231 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
8129ed29
ON
232 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
233 sb_writers_name[i],
234 &type->s_writers_key[i]))
7eb5e882 235 goto fail;
1da177e4 236 }
7eb5e882 237 init_waitqueue_head(&s->s_writers.wait_unfrozen);
df0ce26c 238 s->s_bdi = &noop_backing_dev_info;
7eb5e882 239 s->s_flags = flags;
cc50a07a 240 if (s->s_user_ns != &init_user_ns)
67690f93 241 s->s_iflags |= SB_I_NODEV;
7eb5e882 242 INIT_HLIST_NODE(&s->s_instances);
f1ee6162 243 INIT_HLIST_BL_HEAD(&s->s_roots);
e97fedb9 244 mutex_init(&s->s_sync_lock);
7eb5e882 245 INIT_LIST_HEAD(&s->s_inodes);
74278da9 246 spin_lock_init(&s->s_inode_list_lock);
6c60d2b5
DC
247 INIT_LIST_HEAD(&s->s_inodes_wb);
248 spin_lock_init(&s->s_inode_wblist_lock);
7eb5e882 249
7eb5e882
AV
250 s->s_count = 1;
251 atomic_set(&s->s_active, 1);
252 mutex_init(&s->s_vfs_rename_mutex);
253 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
bc8230ee 254 init_rwsem(&s->s_dquot.dqio_sem);
7eb5e882
AV
255 s->s_maxbytes = MAX_NON_LFS;
256 s->s_op = &default_op;
257 s->s_time_gran = 1000000000;
3cb29d11 258 s->cleancache_poolid = CLEANCACHE_NO_POOL;
7eb5e882
AV
259
260 s->s_shrink.seeks = DEFAULT_SEEKS;
261 s->s_shrink.scan_objects = super_cache_scan;
262 s->s_shrink.count_objects = super_cache_count;
263 s->s_shrink.batch = 1024;
2acb60a0 264 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
8e04944f
TH
265 if (prealloc_shrinker(&s->s_shrink))
266 goto fail;
c92e8e10 267 if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
2b3648a6 268 goto fail;
c92e8e10 269 if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
2b3648a6 270 goto fail;
1da177e4 271 return s;
5ca302c8 272
7eb5e882 273fail:
0200894d 274 destroy_unused_super(s);
7eb5e882 275 return NULL;
1da177e4
LT
276}
277
278/* Superblock refcounting */
279
280/*
35cf7ba0 281 * Drop a superblock's refcount. The caller must hold sb_lock.
1da177e4 282 */
c645b930 283static void __put_super(struct super_block *s)
1da177e4 284{
c645b930
AV
285 if (!--s->s_count) {
286 list_del_init(&s->s_list);
287 WARN_ON(s->s_dentry_lru.node);
288 WARN_ON(s->s_inode_lru.node);
289 WARN_ON(!list_empty(&s->s_mounts));
290 security_sb_free(s);
291 put_user_ns(s->s_user_ns);
292 kfree(s->s_subtype);
293 call_rcu(&s->rcu, destroy_super_rcu);
1da177e4 294 }
1da177e4
LT
295}
296
297/**
298 * put_super - drop a temporary reference to superblock
299 * @sb: superblock in question
300 *
301 * Drops a temporary reference, frees superblock if there's no
302 * references left.
303 */
f47ec3f2 304static void put_super(struct super_block *sb)
1da177e4
LT
305{
306 spin_lock(&sb_lock);
307 __put_super(sb);
308 spin_unlock(&sb_lock);
309}
310
311
312/**
1712ac8f 313 * deactivate_locked_super - drop an active reference to superblock
1da177e4
LT
314 * @s: superblock to deactivate
315 *
bd7ced98 316 * Drops an active reference to superblock, converting it into a temporary
1712ac8f 317 * one if there is no other active references left. In that case we
1da177e4
LT
318 * tell fs driver to shut it down and drop the temporary reference we
319 * had just acquired.
1712ac8f
AV
320 *
321 * Caller holds exclusive lock on superblock; that lock is released.
1da177e4 322 */
1712ac8f 323void deactivate_locked_super(struct super_block *s)
1da177e4
LT
324{
325 struct file_system_type *fs = s->s_type;
b20bd1a5 326 if (atomic_dec_and_test(&s->s_active)) {
3167760f 327 cleancache_invalidate_fs(s);
b0d40c92 328 unregister_shrinker(&s->s_shrink);
28f2cd4f 329 fs->kill_sb(s);
f5e1dd34 330
c0a5b560
VD
331 /*
332 * Since list_lru_destroy() may sleep, we cannot call it from
333 * put_super(), where we hold the sb_lock. Therefore we destroy
334 * the lru lists right now.
335 */
336 list_lru_destroy(&s->s_dentry_lru);
337 list_lru_destroy(&s->s_inode_lru);
338
1da177e4
LT
339 put_filesystem(fs);
340 put_super(s);
1712ac8f
AV
341 } else {
342 up_write(&s->s_umount);
1da177e4
LT
343 }
344}
345
1712ac8f 346EXPORT_SYMBOL(deactivate_locked_super);
1da177e4 347
74dbbdd7 348/**
1712ac8f 349 * deactivate_super - drop an active reference to superblock
74dbbdd7
AV
350 * @s: superblock to deactivate
351 *
1712ac8f
AV
352 * Variant of deactivate_locked_super(), except that superblock is *not*
353 * locked by caller. If we are going to drop the final active reference,
354 * lock will be acquired prior to that.
74dbbdd7 355 */
1712ac8f 356void deactivate_super(struct super_block *s)
74dbbdd7 357{
1712ac8f
AV
358 if (!atomic_add_unless(&s->s_active, -1, 1)) {
359 down_write(&s->s_umount);
360 deactivate_locked_super(s);
74dbbdd7
AV
361 }
362}
363
1712ac8f 364EXPORT_SYMBOL(deactivate_super);
74dbbdd7 365
1da177e4
LT
366/**
367 * grab_super - acquire an active reference
368 * @s: reference we are trying to make active
369 *
370 * Tries to acquire an active reference. grab_super() is used when we
371 * had just found a superblock in super_blocks or fs_type->fs_supers
372 * and want to turn it into a full-blown active reference. grab_super()
373 * is called with sb_lock held and drops it. Returns 1 in case of
374 * success, 0 if we had failed (superblock contents was already dead or
acfec9a5
AV
375 * dying when grab_super() had been called). Note that this is only
376 * called for superblocks not in rundown mode (== ones still on ->fs_supers
377 * of their type), so increment of ->s_count is OK here.
1da177e4 378 */
9c4dbee7 379static int grab_super(struct super_block *s) __releases(sb_lock)
1da177e4
LT
380{
381 s->s_count++;
382 spin_unlock(&sb_lock);
383 down_write(&s->s_umount);
e462ec50 384 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
acfec9a5
AV
385 put_super(s);
386 return 1;
387 }
1da177e4
LT
388 up_write(&s->s_umount);
389 put_super(s);
1da177e4
LT
390 return 0;
391}
392
12ad3ab6 393/*
eb6ef3df 394 * trylock_super - try to grab ->s_umount shared
331cbdee 395 * @sb: reference we are trying to grab
12ad3ab6 396 *
eb6ef3df 397 * Try to prevent fs shutdown. This is used in places where we
12ad3ab6 398 * cannot take an active reference but we need to ensure that the
eb6ef3df
KK
399 * filesystem is not shut down while we are working on it. It returns
400 * false if we cannot acquire s_umount or if we lose the race and
401 * filesystem already got into shutdown, and returns true with the s_umount
402 * lock held in read mode in case of success. On successful return,
403 * the caller must drop the s_umount lock when done.
404 *
405 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
406 * The reason why it's safe is that we are OK with doing trylock instead
407 * of down_read(). There's a couple of places that are OK with that, but
408 * it's very much not a general-purpose interface.
12ad3ab6 409 */
eb6ef3df 410bool trylock_super(struct super_block *sb)
12ad3ab6 411{
12ad3ab6 412 if (down_read_trylock(&sb->s_umount)) {
eb6ef3df 413 if (!hlist_unhashed(&sb->s_instances) &&
e462ec50 414 sb->s_root && (sb->s_flags & SB_BORN))
12ad3ab6
DC
415 return true;
416 up_read(&sb->s_umount);
417 }
418
12ad3ab6
DC
419 return false;
420}
421
1da177e4
LT
422/**
423 * generic_shutdown_super - common helper for ->kill_sb()
424 * @sb: superblock to kill
425 *
426 * generic_shutdown_super() does all fs-independent work on superblock
427 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
428 * that need destruction out of superblock, call generic_shutdown_super()
429 * and release aforementioned objects. Note: dentries and inodes _are_
430 * taken care of and do not need specific handling.
c636ebdb
DH
431 *
432 * Upon calling this function, the filesystem may no longer alter or
433 * rearrange the set of dentries belonging to this super_block, nor may it
434 * change the attachments of dentries to inodes.
1da177e4
LT
435 */
436void generic_shutdown_super(struct super_block *sb)
437{
ee9b6d61 438 const struct super_operations *sop = sb->s_op;
1da177e4 439
c636ebdb
DH
440 if (sb->s_root) {
441 shrink_dcache_for_umount(sb);
60b0680f 442 sync_filesystem(sb);
e462ec50 443 sb->s_flags &= ~SB_ACTIVE;
efaee192 444
74278da9 445 fsnotify_unmount_inodes(sb);
a1a0e23e 446 cgroup_writeback_umount();
63997e98
AV
447
448 evict_inodes(sb);
1da177e4 449
7b7a8665
CH
450 if (sb->s_dio_done_wq) {
451 destroy_workqueue(sb->s_dio_done_wq);
452 sb->s_dio_done_wq = NULL;
453 }
454
1da177e4
LT
455 if (sop->put_super)
456 sop->put_super(sb);
457
63997e98 458 if (!list_empty(&sb->s_inodes)) {
7b4fe29e
DJ
459 printk("VFS: Busy inodes after unmount of %s. "
460 "Self-destruct in 5 seconds. Have a nice day...\n",
461 sb->s_id);
1da177e4 462 }
1da177e4
LT
463 }
464 spin_lock(&sb_lock);
465 /* should be initialized for __put_super_and_need_restart() */
a5166169 466 hlist_del_init(&sb->s_instances);
1da177e4
LT
467 spin_unlock(&sb_lock);
468 up_write(&sb->s_umount);
c1844d53 469 if (sb->s_bdi != &noop_backing_dev_info) {
fca39346
JK
470 bdi_put(sb->s_bdi);
471 sb->s_bdi = &noop_backing_dev_info;
fca39346 472 }
1da177e4
LT
473}
474
475EXPORT_SYMBOL(generic_shutdown_super);
476
477/**
6e4eab57 478 * sget_userns - find or create a superblock
1da177e4
LT
479 * @type: filesystem type superblock should belong to
480 * @test: comparison callback
481 * @set: setup callback
9249e17f 482 * @flags: mount flags
6e4eab57 483 * @user_ns: User namespace for the super_block
1da177e4
LT
484 * @data: argument to each of them
485 */
6e4eab57 486struct super_block *sget_userns(struct file_system_type *type,
1da177e4
LT
487 int (*test)(struct super_block *,void *),
488 int (*set)(struct super_block *,void *),
6e4eab57 489 int flags, struct user_namespace *user_ns,
1da177e4
LT
490 void *data)
491{
492 struct super_block *s = NULL;
d4730127 493 struct super_block *old;
1da177e4
LT
494 int err;
495
e462ec50 496 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) &&
a001e74c
EB
497 !(type->fs_flags & FS_USERNS_MOUNT) &&
498 !capable(CAP_SYS_ADMIN))
499 return ERR_PTR(-EPERM);
1da177e4
LT
500retry:
501 spin_lock(&sb_lock);
d4730127 502 if (test) {
b67bfe0d 503 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
d4730127
MK
504 if (!test(old, data))
505 continue;
6e4eab57
EB
506 if (user_ns != old->s_user_ns) {
507 spin_unlock(&sb_lock);
0200894d 508 destroy_unused_super(s);
6e4eab57
EB
509 return ERR_PTR(-EBUSY);
510 }
d4730127
MK
511 if (!grab_super(old))
512 goto retry;
0200894d 513 destroy_unused_super(s);
d4730127
MK
514 return old;
515 }
1da177e4
LT
516 }
517 if (!s) {
518 spin_unlock(&sb_lock);
e462ec50 519 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
1da177e4
LT
520 if (!s)
521 return ERR_PTR(-ENOMEM);
522 goto retry;
523 }
dd111b31 524
1da177e4
LT
525 err = set(s, data);
526 if (err) {
527 spin_unlock(&sb_lock);
0200894d 528 destroy_unused_super(s);
1da177e4
LT
529 return ERR_PTR(err);
530 }
531 s->s_type = type;
532 strlcpy(s->s_id, type->name, sizeof(s->s_id));
533 list_add_tail(&s->s_list, &super_blocks);
a5166169 534 hlist_add_head(&s->s_instances, &type->fs_supers);
1da177e4
LT
535 spin_unlock(&sb_lock);
536 get_filesystem(type);
8e04944f 537 register_shrinker_prepared(&s->s_shrink);
1da177e4
LT
538 return s;
539}
540
6e4eab57
EB
541EXPORT_SYMBOL(sget_userns);
542
543/**
544 * sget - find or create a superblock
545 * @type: filesystem type superblock should belong to
546 * @test: comparison callback
547 * @set: setup callback
548 * @flags: mount flags
549 * @data: argument to each of them
550 */
551struct super_block *sget(struct file_system_type *type,
552 int (*test)(struct super_block *,void *),
553 int (*set)(struct super_block *,void *),
554 int flags,
555 void *data)
556{
557 struct user_namespace *user_ns = current_user_ns();
558
93faccbb
EB
559 /* We don't yet pass the user namespace of the parent
560 * mount through to here so always use &init_user_ns
561 * until that changes.
562 */
e462ec50 563 if (flags & SB_SUBMOUNT)
93faccbb
EB
564 user_ns = &init_user_ns;
565
6e4eab57 566 /* Ensure the requestor has permissions over the target filesystem */
e462ec50 567 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
6e4eab57
EB
568 return ERR_PTR(-EPERM);
569
570 return sget_userns(type, test, set, flags, user_ns, data);
571}
572
1da177e4
LT
573EXPORT_SYMBOL(sget);
574
575void drop_super(struct super_block *sb)
576{
577 up_read(&sb->s_umount);
578 put_super(sb);
579}
580
581EXPORT_SYMBOL(drop_super);
582
ba6379f7
JK
583void drop_super_exclusive(struct super_block *sb)
584{
585 up_write(&sb->s_umount);
586 put_super(sb);
587}
588EXPORT_SYMBOL(drop_super_exclusive);
589
fa7c1d50
MG
590static void __iterate_supers(void (*f)(struct super_block *))
591{
592 struct super_block *sb, *p = NULL;
593
594 spin_lock(&sb_lock);
595 list_for_each_entry(sb, &super_blocks, s_list) {
596 if (hlist_unhashed(&sb->s_instances))
597 continue;
598 sb->s_count++;
599 spin_unlock(&sb_lock);
600
601 f(sb);
602
603 spin_lock(&sb_lock);
604 if (p)
605 __put_super(p);
606 p = sb;
607 }
608 if (p)
609 __put_super(p);
610 spin_unlock(&sb_lock);
611}
01a05b33
AV
612/**
613 * iterate_supers - call function for all active superblocks
614 * @f: function to call
615 * @arg: argument to pass to it
616 *
617 * Scans the superblock list and calls given function, passing it
618 * locked superblock and given argument.
619 */
620void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
621{
dca33252 622 struct super_block *sb, *p = NULL;
01a05b33
AV
623
624 spin_lock(&sb_lock);
dca33252 625 list_for_each_entry(sb, &super_blocks, s_list) {
a5166169 626 if (hlist_unhashed(&sb->s_instances))
01a05b33
AV
627 continue;
628 sb->s_count++;
629 spin_unlock(&sb_lock);
630
631 down_read(&sb->s_umount);
e462ec50 632 if (sb->s_root && (sb->s_flags & SB_BORN))
01a05b33
AV
633 f(sb, arg);
634 up_read(&sb->s_umount);
635
636 spin_lock(&sb_lock);
dca33252
AV
637 if (p)
638 __put_super(p);
639 p = sb;
01a05b33 640 }
dca33252
AV
641 if (p)
642 __put_super(p);
01a05b33
AV
643 spin_unlock(&sb_lock);
644}
645
43e15cdb
AV
646/**
647 * iterate_supers_type - call function for superblocks of given type
648 * @type: fs type
649 * @f: function to call
650 * @arg: argument to pass to it
651 *
652 * Scans the superblock list and calls given function, passing it
653 * locked superblock and given argument.
654 */
655void iterate_supers_type(struct file_system_type *type,
656 void (*f)(struct super_block *, void *), void *arg)
657{
658 struct super_block *sb, *p = NULL;
659
660 spin_lock(&sb_lock);
b67bfe0d 661 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
43e15cdb
AV
662 sb->s_count++;
663 spin_unlock(&sb_lock);
664
665 down_read(&sb->s_umount);
e462ec50 666 if (sb->s_root && (sb->s_flags & SB_BORN))
43e15cdb
AV
667 f(sb, arg);
668 up_read(&sb->s_umount);
669
670 spin_lock(&sb_lock);
671 if (p)
672 __put_super(p);
673 p = sb;
674 }
675 if (p)
676 __put_super(p);
677 spin_unlock(&sb_lock);
678}
679
680EXPORT_SYMBOL(iterate_supers_type);
681
ba6379f7 682static struct super_block *__get_super(struct block_device *bdev, bool excl)
1da177e4 683{
618f0636
KK
684 struct super_block *sb;
685
1da177e4
LT
686 if (!bdev)
687 return NULL;
618f0636 688
1da177e4 689 spin_lock(&sb_lock);
618f0636
KK
690rescan:
691 list_for_each_entry(sb, &super_blocks, s_list) {
a5166169 692 if (hlist_unhashed(&sb->s_instances))
551de6f3 693 continue;
618f0636
KK
694 if (sb->s_bdev == bdev) {
695 sb->s_count++;
1da177e4 696 spin_unlock(&sb_lock);
ba6379f7
JK
697 if (!excl)
698 down_read(&sb->s_umount);
699 else
700 down_write(&sb->s_umount);
df40c01a 701 /* still alive? */
e462ec50 702 if (sb->s_root && (sb->s_flags & SB_BORN))
618f0636 703 return sb;
ba6379f7
JK
704 if (!excl)
705 up_read(&sb->s_umount);
706 else
707 up_write(&sb->s_umount);
df40c01a 708 /* nope, got unmounted */
618f0636 709 spin_lock(&sb_lock);
df40c01a
AV
710 __put_super(sb);
711 goto rescan;
1da177e4
LT
712 }
713 }
714 spin_unlock(&sb_lock);
715 return NULL;
716}
717
6b6dc836 718/**
ba6379f7 719 * get_super - get the superblock of a device
6b6dc836
JK
720 * @bdev: device to get the superblock for
721 *
722 * Scans the superblock list and finds the superblock of the file system
ba6379f7 723 * mounted on the device given. %NULL is returned if no match is found.
6b6dc836 724 */
ba6379f7
JK
725struct super_block *get_super(struct block_device *bdev)
726{
727 return __get_super(bdev, false);
728}
729EXPORT_SYMBOL(get_super);
730
731static struct super_block *__get_super_thawed(struct block_device *bdev,
732 bool excl)
6b6dc836
JK
733{
734 while (1) {
ba6379f7 735 struct super_block *s = __get_super(bdev, excl);
5accdf82 736 if (!s || s->s_writers.frozen == SB_UNFROZEN)
6b6dc836 737 return s;
ba6379f7
JK
738 if (!excl)
739 up_read(&s->s_umount);
740 else
741 up_write(&s->s_umount);
5accdf82
JK
742 wait_event(s->s_writers.wait_unfrozen,
743 s->s_writers.frozen == SB_UNFROZEN);
6b6dc836
JK
744 put_super(s);
745 }
746}
ba6379f7
JK
747
748/**
749 * get_super_thawed - get thawed superblock of a device
750 * @bdev: device to get the superblock for
751 *
752 * Scans the superblock list and finds the superblock of the file system
753 * mounted on the device. The superblock is returned once it is thawed
754 * (or immediately if it was not frozen). %NULL is returned if no match
755 * is found.
756 */
757struct super_block *get_super_thawed(struct block_device *bdev)
758{
759 return __get_super_thawed(bdev, false);
760}
6b6dc836
JK
761EXPORT_SYMBOL(get_super_thawed);
762
ba6379f7
JK
763/**
764 * get_super_exclusive_thawed - get thawed superblock of a device
765 * @bdev: device to get the superblock for
766 *
767 * Scans the superblock list and finds the superblock of the file system
768 * mounted on the device. The superblock is returned once it is thawed
769 * (or immediately if it was not frozen) and s_umount semaphore is held
770 * in exclusive mode. %NULL is returned if no match is found.
771 */
772struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
773{
774 return __get_super_thawed(bdev, true);
775}
776EXPORT_SYMBOL(get_super_exclusive_thawed);
777
4504230a
CH
778/**
779 * get_active_super - get an active reference to the superblock of a device
780 * @bdev: device to get the superblock for
781 *
782 * Scans the superblock list and finds the superblock of the file system
783 * mounted on the device given. Returns the superblock with an active
d3f21473 784 * reference or %NULL if none was found.
4504230a
CH
785 */
786struct super_block *get_active_super(struct block_device *bdev)
787{
788 struct super_block *sb;
789
790 if (!bdev)
791 return NULL;
792
1494583d 793restart:
4504230a
CH
794 spin_lock(&sb_lock);
795 list_for_each_entry(sb, &super_blocks, s_list) {
a5166169 796 if (hlist_unhashed(&sb->s_instances))
551de6f3 797 continue;
1494583d 798 if (sb->s_bdev == bdev) {
acfec9a5 799 if (!grab_super(sb))
1494583d 800 goto restart;
acfec9a5
AV
801 up_write(&sb->s_umount);
802 return sb;
1494583d 803 }
4504230a
CH
804 }
805 spin_unlock(&sb_lock);
806 return NULL;
807}
dd111b31 808
df40c01a 809struct super_block *user_get_super(dev_t dev)
1da177e4 810{
618f0636 811 struct super_block *sb;
1da177e4 812
1da177e4 813 spin_lock(&sb_lock);
618f0636
KK
814rescan:
815 list_for_each_entry(sb, &super_blocks, s_list) {
a5166169 816 if (hlist_unhashed(&sb->s_instances))
551de6f3 817 continue;
618f0636
KK
818 if (sb->s_dev == dev) {
819 sb->s_count++;
1da177e4 820 spin_unlock(&sb_lock);
618f0636 821 down_read(&sb->s_umount);
df40c01a 822 /* still alive? */
e462ec50 823 if (sb->s_root && (sb->s_flags & SB_BORN))
618f0636
KK
824 return sb;
825 up_read(&sb->s_umount);
df40c01a 826 /* nope, got unmounted */
618f0636 827 spin_lock(&sb_lock);
df40c01a
AV
828 __put_super(sb);
829 goto rescan;
1da177e4
LT
830 }
831 }
832 spin_unlock(&sb_lock);
833 return NULL;
834}
835
1da177e4
LT
836/**
837 * do_remount_sb - asks filesystem to change mount options.
838 * @sb: superblock in question
e462ec50 839 * @sb_flags: revised superblock flags
1da177e4
LT
840 * @data: the rest of options
841 * @force: whether or not to force the change
842 *
843 * Alters the mount options of a mounted file system.
844 */
e462ec50 845int do_remount_sb(struct super_block *sb, int sb_flags, void *data, int force)
1da177e4
LT
846{
847 int retval;
c79d967d 848 int remount_ro;
4504230a 849
5accdf82 850 if (sb->s_writers.frozen != SB_UNFROZEN)
4504230a
CH
851 return -EBUSY;
852
9361401e 853#ifdef CONFIG_BLOCK
e462ec50 854 if (!(sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
1da177e4 855 return -EACCES;
9361401e 856#endif
4504230a 857
e462ec50 858 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
d208bbdd 859
0aec09d0 860 if (remount_ro) {
fdab684d 861 if (!hlist_empty(&sb->s_pins)) {
0aec09d0 862 up_write(&sb->s_umount);
fdab684d 863 group_pin_kill(&sb->s_pins);
0aec09d0
AV
864 down_write(&sb->s_umount);
865 if (!sb->s_root)
866 return 0;
867 if (sb->s_writers.frozen != SB_UNFROZEN)
868 return -EBUSY;
e462ec50 869 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
0aec09d0
AV
870 }
871 }
872 shrink_dcache_sb(sb);
873
1da177e4
LT
874 /* If we are remounting RDONLY and current sb is read/write,
875 make sure there are no rw files opened */
d208bbdd 876 if (remount_ro) {
4ed5e82f 877 if (force) {
eee5cc27
AV
878 sb->s_readonly_remount = 1;
879 smp_wmb();
4ed5e82f
MS
880 } else {
881 retval = sb_prepare_remount_readonly(sb);
882 if (retval)
883 return retval;
4ed5e82f 884 }
1da177e4
LT
885 }
886
887 if (sb->s_op->remount_fs) {
e462ec50 888 retval = sb->s_op->remount_fs(sb, &sb_flags, data);
2833eb2b
MS
889 if (retval) {
890 if (!force)
4ed5e82f 891 goto cancel_readonly;
2833eb2b
MS
892 /* If forced remount, go ahead despite any errors */
893 WARN(1, "forced remount of a %s fs returned %i\n",
894 sb->s_type->name, retval);
895 }
1da177e4 896 }
e462ec50 897 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (sb_flags & MS_RMT_MASK);
4ed5e82f
MS
898 /* Needs to be ordered wrt mnt_is_readonly() */
899 smp_wmb();
900 sb->s_readonly_remount = 0;
c79d967d 901
d208bbdd
NP
902 /*
903 * Some filesystems modify their metadata via some other path than the
904 * bdev buffer cache (eg. use a private mapping, or directories in
905 * pagecache, etc). Also file data modifications go via their own
906 * mappings. So If we try to mount readonly then copy the filesystem
907 * from bdev, we could get stale data, so invalidate it to give a best
908 * effort at coherency.
909 */
910 if (remount_ro && sb->s_bdev)
911 invalidate_bdev(sb->s_bdev);
1da177e4 912 return 0;
4ed5e82f
MS
913
914cancel_readonly:
915 sb->s_readonly_remount = 0;
916 return retval;
1da177e4
LT
917}
918
fa7c1d50 919static void do_emergency_remount_callback(struct super_block *sb)
1da177e4 920{
fa7c1d50
MG
921 down_write(&sb->s_umount);
922 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
923 !sb_rdonly(sb)) {
924 /*
925 * What lock protects sb->s_flags??
926 */
927 do_remount_sb(sb, SB_RDONLY, NULL, 1);
1da177e4 928 }
fa7c1d50
MG
929 up_write(&sb->s_umount);
930}
931
932static void do_emergency_remount(struct work_struct *work)
933{
934 __iterate_supers(do_emergency_remount_callback);
a2a9537a 935 kfree(work);
1da177e4
LT
936 printk("Emergency Remount complete\n");
937}
938
939void emergency_remount(void)
940{
a2a9537a
JA
941 struct work_struct *work;
942
943 work = kmalloc(sizeof(*work), GFP_ATOMIC);
944 if (work) {
945 INIT_WORK(work, do_emergency_remount);
946 schedule_work(work);
947 }
1da177e4
LT
948}
949
08fdc8a0
MG
950static void do_thaw_all_callback(struct super_block *sb)
951{
952 down_write(&sb->s_umount);
1c18d2a1 953 if (sb->s_root && sb->s_flags & SB_BORN) {
08fdc8a0
MG
954 emergency_thaw_bdev(sb);
955 thaw_super_locked(sb);
956 } else {
957 up_write(&sb->s_umount);
958 }
959}
960
961static void do_thaw_all(struct work_struct *work)
962{
963 __iterate_supers(do_thaw_all_callback);
964 kfree(work);
965 printk(KERN_WARNING "Emergency Thaw complete\n");
966}
967
968/**
969 * emergency_thaw_all -- forcibly thaw every frozen filesystem
970 *
971 * Used for emergency unfreeze of all filesystems via SysRq
972 */
973void emergency_thaw_all(void)
974{
975 struct work_struct *work;
976
977 work = kmalloc(sizeof(*work), GFP_ATOMIC);
978 if (work) {
979 INIT_WORK(work, do_thaw_all);
980 schedule_work(work);
981 }
982}
983
ad76cbc6 984static DEFINE_IDA(unnamed_dev_ida);
1da177e4 985
5a66847e
MW
986/**
987 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
988 * @p: Pointer to a dev_t.
989 *
990 * Filesystems which don't use real block devices can call this function
991 * to allocate a virtual block device.
992 *
993 * Context: Any context. Frequently called while holding sb_lock.
994 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
995 * or -ENOMEM if memory allocation failed.
996 */
0ee5dc67 997int get_anon_bdev(dev_t *p)
1da177e4
LT
998{
999 int dev;
5a66847e
MW
1000
1001 /*
1002 * Many userspace utilities consider an FSID of 0 invalid.
1003 * Always return at least 1 from get_anon_bdev.
1004 */
1005 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1006 GFP_ATOMIC);
1007 if (dev == -ENOSPC)
1008 dev = -EMFILE;
1009 if (dev < 0)
1010 return dev;
1011
1012 *p = MKDEV(0, dev);
1da177e4
LT
1013 return 0;
1014}
0ee5dc67 1015EXPORT_SYMBOL(get_anon_bdev);
1da177e4 1016
0ee5dc67 1017void free_anon_bdev(dev_t dev)
1da177e4 1018{
5a66847e 1019 ida_free(&unnamed_dev_ida, MINOR(dev));
1da177e4 1020}
0ee5dc67
AV
1021EXPORT_SYMBOL(free_anon_bdev);
1022
1023int set_anon_super(struct super_block *s, void *data)
1024{
df0ce26c 1025 return get_anon_bdev(&s->s_dev);
0ee5dc67 1026}
0ee5dc67
AV
1027EXPORT_SYMBOL(set_anon_super);
1028
1029void kill_anon_super(struct super_block *sb)
1030{
1031 dev_t dev = sb->s_dev;
1032 generic_shutdown_super(sb);
1033 free_anon_bdev(dev);
1034}
1da177e4
LT
1035EXPORT_SYMBOL(kill_anon_super);
1036
1da177e4
LT
1037void kill_litter_super(struct super_block *sb)
1038{
1039 if (sb->s_root)
1040 d_genocide(sb->s_root);
1041 kill_anon_super(sb);
1042}
1da177e4
LT
1043EXPORT_SYMBOL(kill_litter_super);
1044
909e6d94
SH
1045static int ns_test_super(struct super_block *sb, void *data)
1046{
1047 return sb->s_fs_info == data;
1048}
1049
1050static int ns_set_super(struct super_block *sb, void *data)
1051{
1052 sb->s_fs_info = data;
1053 return set_anon_super(sb, NULL);
1054}
1055
d91ee87d
EB
1056struct dentry *mount_ns(struct file_system_type *fs_type,
1057 int flags, void *data, void *ns, struct user_namespace *user_ns,
1058 int (*fill_super)(struct super_block *, void *, int))
909e6d94
SH
1059{
1060 struct super_block *sb;
1061
d91ee87d
EB
1062 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1063 * over the namespace.
1064 */
e462ec50 1065 if (!(flags & SB_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
d91ee87d
EB
1066 return ERR_PTR(-EPERM);
1067
6e4eab57
EB
1068 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1069 user_ns, ns);
909e6d94 1070 if (IS_ERR(sb))
ceefda69 1071 return ERR_CAST(sb);
909e6d94
SH
1072
1073 if (!sb->s_root) {
1074 int err;
e462ec50 1075 err = fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
909e6d94 1076 if (err) {
74dbbdd7 1077 deactivate_locked_super(sb);
ceefda69 1078 return ERR_PTR(err);
909e6d94
SH
1079 }
1080
e462ec50 1081 sb->s_flags |= SB_ACTIVE;
909e6d94
SH
1082 }
1083
ceefda69 1084 return dget(sb->s_root);
909e6d94
SH
1085}
1086
ceefda69 1087EXPORT_SYMBOL(mount_ns);
909e6d94 1088
9361401e 1089#ifdef CONFIG_BLOCK
1da177e4
LT
1090static int set_bdev_super(struct super_block *s, void *data)
1091{
1092 s->s_bdev = data;
1093 s->s_dev = s->s_bdev->bd_dev;
13eec236 1094 s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
32a88aa1 1095
1da177e4
LT
1096 return 0;
1097}
1098
1099static int test_bdev_super(struct super_block *s, void *data)
1100{
1101 return (void *)s->s_bdev == data;
1102}
1103
152a0836 1104struct dentry *mount_bdev(struct file_system_type *fs_type,
1da177e4 1105 int flags, const char *dev_name, void *data,
152a0836 1106 int (*fill_super)(struct super_block *, void *, int))
1da177e4
LT
1107{
1108 struct block_device *bdev;
1109 struct super_block *s;
d4d77629 1110 fmode_t mode = FMODE_READ | FMODE_EXCL;
1da177e4
LT
1111 int error = 0;
1112
e462ec50 1113 if (!(flags & SB_RDONLY))
30c40d2c
AV
1114 mode |= FMODE_WRITE;
1115
d4d77629 1116 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1da177e4 1117 if (IS_ERR(bdev))
152a0836 1118 return ERR_CAST(bdev);
1da177e4
LT
1119
1120 /*
1121 * once the super is inserted into the list by sget, s_umount
1122 * will protect the lockfs code from trying to start a snapshot
1123 * while we are mounting
1124 */
4fadd7bb
CH
1125 mutex_lock(&bdev->bd_fsfreeze_mutex);
1126 if (bdev->bd_fsfreeze_count > 0) {
1127 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1128 error = -EBUSY;
1129 goto error_bdev;
1130 }
e462ec50 1131 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
9249e17f 1132 bdev);
4fadd7bb 1133 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1da177e4 1134 if (IS_ERR(s))
454e2398 1135 goto error_s;
1da177e4
LT
1136
1137 if (s->s_root) {
e462ec50 1138 if ((flags ^ s->s_flags) & SB_RDONLY) {
74dbbdd7 1139 deactivate_locked_super(s);
454e2398
DH
1140 error = -EBUSY;
1141 goto error_bdev;
1da177e4 1142 }
454e2398 1143
4f331f01
TH
1144 /*
1145 * s_umount nests inside bd_mutex during
e525fd89
TH
1146 * __invalidate_device(). blkdev_put() acquires
1147 * bd_mutex and can't be called under s_umount. Drop
1148 * s_umount temporarily. This is safe as we're
1149 * holding an active reference.
4f331f01
TH
1150 */
1151 up_write(&s->s_umount);
d4d77629 1152 blkdev_put(bdev, mode);
4f331f01 1153 down_write(&s->s_umount);
1da177e4 1154 } else {
30c40d2c 1155 s->s_mode = mode;
a1c6f057 1156 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
e78c9a00 1157 sb_set_blocksize(s, block_size(bdev));
e462ec50 1158 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1da177e4 1159 if (error) {
74dbbdd7 1160 deactivate_locked_super(s);
454e2398 1161 goto error;
fa675765 1162 }
454e2398 1163
e462ec50 1164 s->s_flags |= SB_ACTIVE;
87d8fe1e 1165 bdev->bd_super = s;
1da177e4
LT
1166 }
1167
152a0836 1168 return dget(s->s_root);
1da177e4 1169
454e2398
DH
1170error_s:
1171 error = PTR_ERR(s);
1172error_bdev:
d4d77629 1173 blkdev_put(bdev, mode);
454e2398 1174error:
152a0836
AV
1175 return ERR_PTR(error);
1176}
1177EXPORT_SYMBOL(mount_bdev);
1178
1da177e4
LT
1179void kill_block_super(struct super_block *sb)
1180{
1181 struct block_device *bdev = sb->s_bdev;
30c40d2c 1182 fmode_t mode = sb->s_mode;
1da177e4 1183
ddbaaf30 1184 bdev->bd_super = NULL;
1da177e4
LT
1185 generic_shutdown_super(sb);
1186 sync_blockdev(bdev);
d4d77629 1187 WARN_ON_ONCE(!(mode & FMODE_EXCL));
e525fd89 1188 blkdev_put(bdev, mode | FMODE_EXCL);
1da177e4
LT
1189}
1190
1191EXPORT_SYMBOL(kill_block_super);
9361401e 1192#endif
1da177e4 1193
3c26ff6e 1194struct dentry *mount_nodev(struct file_system_type *fs_type,
1da177e4 1195 int flags, void *data,
3c26ff6e 1196 int (*fill_super)(struct super_block *, void *, int))
1da177e4
LT
1197{
1198 int error;
9249e17f 1199 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1da177e4
LT
1200
1201 if (IS_ERR(s))
3c26ff6e 1202 return ERR_CAST(s);
1da177e4 1203
e462ec50 1204 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1da177e4 1205 if (error) {
74dbbdd7 1206 deactivate_locked_super(s);
3c26ff6e 1207 return ERR_PTR(error);
1da177e4 1208 }
e462ec50 1209 s->s_flags |= SB_ACTIVE;
3c26ff6e 1210 return dget(s->s_root);
1da177e4 1211}
3c26ff6e
AV
1212EXPORT_SYMBOL(mount_nodev);
1213
1da177e4
LT
1214static int compare_single(struct super_block *s, void *p)
1215{
1216 return 1;
1217}
1218
fc14f2fe 1219struct dentry *mount_single(struct file_system_type *fs_type,
1da177e4 1220 int flags, void *data,
fc14f2fe 1221 int (*fill_super)(struct super_block *, void *, int))
1da177e4
LT
1222{
1223 struct super_block *s;
1224 int error;
1225
9249e17f 1226 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1da177e4 1227 if (IS_ERR(s))
fc14f2fe 1228 return ERR_CAST(s);
1da177e4 1229 if (!s->s_root) {
e462ec50 1230 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1da177e4 1231 if (error) {
74dbbdd7 1232 deactivate_locked_super(s);
fc14f2fe 1233 return ERR_PTR(error);
1da177e4 1234 }
e462ec50 1235 s->s_flags |= SB_ACTIVE;
9329d1be
KS
1236 } else {
1237 do_remount_sb(s, flags, data, 0);
1da177e4 1238 }
fc14f2fe
AV
1239 return dget(s->s_root);
1240}
1241EXPORT_SYMBOL(mount_single);
1242
9d412a43
AV
1243struct dentry *
1244mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1da177e4 1245{
c96e41e9 1246 struct dentry *root;
9d412a43 1247 struct super_block *sb;
1da177e4 1248 char *secdata = NULL;
9d412a43 1249 int error = -ENOMEM;
8089352a 1250
e0007529 1251 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1da177e4 1252 secdata = alloc_secdata();
454e2398 1253 if (!secdata)
9d412a43 1254 goto out;
1da177e4 1255
e0007529 1256 error = security_sb_copy_data(data, secdata);
454e2398 1257 if (error)
1da177e4 1258 goto out_free_secdata;
1da177e4
LT
1259 }
1260
1a102ff9
AV
1261 root = type->mount(type, flags, name, data);
1262 if (IS_ERR(root)) {
1263 error = PTR_ERR(root);
1264 goto out_free_secdata;
c96e41e9 1265 }
9d412a43
AV
1266 sb = root->d_sb;
1267 BUG_ON(!sb);
1268 WARN_ON(!sb->s_bdi);
79f546a6
DC
1269
1270 /*
1271 * Write barrier is for super_cache_count(). We place it before setting
1272 * SB_BORN as the data dependency between the two functions is the
1273 * superblock structure contents that we just set up, not the SB_BORN
1274 * flag.
1275 */
1276 smp_wmb();
e462ec50 1277 sb->s_flags |= SB_BORN;
454e2398 1278
9d412a43 1279 error = security_sb_kern_mount(sb, flags, secdata);
5129a469
JE
1280 if (error)
1281 goto out_sb;
454e2398 1282
42cb56ae
JL
1283 /*
1284 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1285 * but s_maxbytes was an unsigned long long for many releases. Throw
1286 * this warning for a little while to try and catch filesystems that
4358b567 1287 * violate this rule.
42cb56ae 1288 */
9d412a43
AV
1289 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1290 "negative value (%lld)\n", type->name, sb->s_maxbytes);
42cb56ae 1291
9d412a43 1292 up_write(&sb->s_umount);
8680e22f 1293 free_secdata(secdata);
9d412a43 1294 return root;
1da177e4 1295out_sb:
9d412a43
AV
1296 dput(root);
1297 deactivate_locked_super(sb);
1da177e4
LT
1298out_free_secdata:
1299 free_secdata(secdata);
1da177e4 1300out:
454e2398 1301 return ERR_PTR(error);
1da177e4
LT
1302}
1303
fca39346
JK
1304/*
1305 * Setup private BDI for given superblock. It gets automatically cleaned up
1306 * in generic_shutdown_super().
1307 */
1308int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1309{
1310 struct backing_dev_info *bdi;
1311 int err;
1312 va_list args;
1313
1314 bdi = bdi_alloc(GFP_KERNEL);
1315 if (!bdi)
1316 return -ENOMEM;
1317
1318 bdi->name = sb->s_type->name;
1319
1320 va_start(args, fmt);
7c4cc300 1321 err = bdi_register_va(bdi, fmt, args);
fca39346
JK
1322 va_end(args);
1323 if (err) {
1324 bdi_put(bdi);
1325 return err;
1326 }
1327 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1328 sb->s_bdi = bdi;
fca39346
JK
1329
1330 return 0;
1331}
1332EXPORT_SYMBOL(super_setup_bdi_name);
1333
1334/*
1335 * Setup private BDI for given superblock. I gets automatically cleaned up
1336 * in generic_shutdown_super().
1337 */
1338int super_setup_bdi(struct super_block *sb)
1339{
1340 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1341
1342 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1343 atomic_long_inc_return(&bdi_seq));
1344}
1345EXPORT_SYMBOL(super_setup_bdi);
1346
5accdf82
JK
1347/*
1348 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1349 * instead.
1350 */
1351void __sb_end_write(struct super_block *sb, int level)
1352{
8129ed29 1353 percpu_up_read(sb->s_writers.rw_sem + level-1);
5accdf82
JK
1354}
1355EXPORT_SYMBOL(__sb_end_write);
1356
f4b554af
ON
1357/*
1358 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1359 * instead.
1360 */
1361int __sb_start_write(struct super_block *sb, int level, bool wait)
1362{
1363 bool force_trylock = false;
8129ed29 1364 int ret = 1;
f4b554af
ON
1365
1366#ifdef CONFIG_LOCKDEP
1367 /*
1368 * We want lockdep to tell us about possible deadlocks with freezing
1369 * but it's it bit tricky to properly instrument it. Getting a freeze
1370 * protection works as getting a read lock but there are subtle
1371 * problems. XFS for example gets freeze protection on internal level
1372 * twice in some cases, which is OK only because we already hold a
1373 * freeze protection also on higher level. Due to these cases we have
1374 * to use wait == F (trylock mode) which must not fail.
1375 */
1376 if (wait) {
1377 int i;
1378
1379 for (i = 0; i < level - 1; i++)
8129ed29 1380 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
f4b554af
ON
1381 force_trylock = true;
1382 break;
1383 }
1384 }
1385#endif
8129ed29
ON
1386 if (wait && !force_trylock)
1387 percpu_down_read(sb->s_writers.rw_sem + level-1);
1388 else
1389 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1390
22224a17 1391 WARN_ON(force_trylock && !ret);
f4b554af
ON
1392 return ret;
1393}
5accdf82
JK
1394EXPORT_SYMBOL(__sb_start_write);
1395
1396/**
1397 * sb_wait_write - wait until all writers to given file system finish
1398 * @sb: the super for which we wait
1399 * @level: type of writers we wait for (normal vs page fault)
1400 *
1401 * This function waits until there are no writers of given type to given file
8129ed29 1402 * system.
5accdf82
JK
1403 */
1404static void sb_wait_write(struct super_block *sb, int level)
1405{
8129ed29 1406 percpu_down_write(sb->s_writers.rw_sem + level-1);
8129ed29 1407}
5accdf82 1408
f1a96220
ON
1409/*
1410 * We are going to return to userspace and forget about these locks, the
1411 * ownership goes to the caller of thaw_super() which does unlock().
1412 */
1413static void lockdep_sb_freeze_release(struct super_block *sb)
1414{
1415 int level;
1416
1417 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1418 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1419}
1420
1421/*
1422 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1423 */
1424static void lockdep_sb_freeze_acquire(struct super_block *sb)
8129ed29
ON
1425{
1426 int level;
5accdf82 1427
8129ed29
ON
1428 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1429 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
f1a96220
ON
1430}
1431
1432static void sb_freeze_unlock(struct super_block *sb)
1433{
1434 int level;
5accdf82 1435
8129ed29
ON
1436 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1437 percpu_up_write(sb->s_writers.rw_sem + level);
5accdf82
JK
1438}
1439
18e9e510 1440/**
7000d3c4
RD
1441 * freeze_super - lock the filesystem and force it into a consistent state
1442 * @sb: the super to lock
18e9e510
JB
1443 *
1444 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1445 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1446 * -EBUSY.
5accdf82
JK
1447 *
1448 * During this function, sb->s_writers.frozen goes through these values:
1449 *
1450 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1451 *
1452 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1453 * writes should be blocked, though page faults are still allowed. We wait for
1454 * all writes to complete and then proceed to the next stage.
1455 *
1456 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1457 * but internal fs threads can still modify the filesystem (although they
1458 * should not dirty new pages or inodes), writeback can run etc. After waiting
1459 * for all running page faults we sync the filesystem which will clean all
1460 * dirty pages and inodes (no new dirty pages or inodes can be created when
1461 * sync is running).
1462 *
1463 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1464 * modification are blocked (e.g. XFS preallocation truncation on inode
1465 * reclaim). This is usually implemented by blocking new transactions for
1466 * filesystems that have them and need this additional guard. After all
1467 * internal writers are finished we call ->freeze_fs() to finish filesystem
1468 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1469 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1470 *
1471 * sb->s_writers.frozen is protected by sb->s_umount.
18e9e510
JB
1472 */
1473int freeze_super(struct super_block *sb)
1474{
1475 int ret;
1476
1477 atomic_inc(&sb->s_active);
1478 down_write(&sb->s_umount);
5accdf82 1479 if (sb->s_writers.frozen != SB_UNFROZEN) {
18e9e510
JB
1480 deactivate_locked_super(sb);
1481 return -EBUSY;
1482 }
1483
e462ec50 1484 if (!(sb->s_flags & SB_BORN)) {
dabe0dc1
AV
1485 up_write(&sb->s_umount);
1486 return 0; /* sic - it's "nothing to do" */
1487 }
1488
bc98a42c 1489 if (sb_rdonly(sb)) {
5accdf82
JK
1490 /* Nothing to do really... */
1491 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
18e9e510
JB
1492 up_write(&sb->s_umount);
1493 return 0;
1494 }
1495
5accdf82 1496 sb->s_writers.frozen = SB_FREEZE_WRITE;
5accdf82
JK
1497 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1498 up_write(&sb->s_umount);
5accdf82 1499 sb_wait_write(sb, SB_FREEZE_WRITE);
8129ed29 1500 down_write(&sb->s_umount);
5accdf82
JK
1501
1502 /* Now we go and block page faults... */
5accdf82 1503 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
5accdf82
JK
1504 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1505
1506 /* All writers are done so after syncing there won't be dirty data */
18e9e510
JB
1507 sync_filesystem(sb);
1508
5accdf82
JK
1509 /* Now wait for internal filesystem counter */
1510 sb->s_writers.frozen = SB_FREEZE_FS;
5accdf82 1511 sb_wait_write(sb, SB_FREEZE_FS);
18e9e510 1512
18e9e510
JB
1513 if (sb->s_op->freeze_fs) {
1514 ret = sb->s_op->freeze_fs(sb);
1515 if (ret) {
1516 printk(KERN_ERR
1517 "VFS:Filesystem freeze failed\n");
5accdf82 1518 sb->s_writers.frozen = SB_UNFROZEN;
8129ed29 1519 sb_freeze_unlock(sb);
5accdf82 1520 wake_up(&sb->s_writers.wait_unfrozen);
18e9e510
JB
1521 deactivate_locked_super(sb);
1522 return ret;
1523 }
1524 }
5accdf82 1525 /*
89f39af1
ON
1526 * For debugging purposes so that fs can warn if it sees write activity
1527 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
5accdf82
JK
1528 */
1529 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
f1a96220 1530 lockdep_sb_freeze_release(sb);
18e9e510
JB
1531 up_write(&sb->s_umount);
1532 return 0;
1533}
1534EXPORT_SYMBOL(freeze_super);
1535
1536/**
1537 * thaw_super -- unlock filesystem
1538 * @sb: the super to thaw
1539 *
1540 * Unlocks the filesystem and marks it writeable again after freeze_super().
1541 */
08fdc8a0 1542static int thaw_super_locked(struct super_block *sb)
18e9e510
JB
1543{
1544 int error;
1545
89f39af1 1546 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
18e9e510
JB
1547 up_write(&sb->s_umount);
1548 return -EINVAL;
1549 }
1550
bc98a42c 1551 if (sb_rdonly(sb)) {
8129ed29 1552 sb->s_writers.frozen = SB_UNFROZEN;
18e9e510 1553 goto out;
8129ed29 1554 }
18e9e510 1555
f1a96220
ON
1556 lockdep_sb_freeze_acquire(sb);
1557
18e9e510
JB
1558 if (sb->s_op->unfreeze_fs) {
1559 error = sb->s_op->unfreeze_fs(sb);
1560 if (error) {
1561 printk(KERN_ERR
1562 "VFS:Filesystem thaw failed\n");
f1a96220 1563 lockdep_sb_freeze_release(sb);
18e9e510
JB
1564 up_write(&sb->s_umount);
1565 return error;
1566 }
1567 }
1568
5accdf82 1569 sb->s_writers.frozen = SB_UNFROZEN;
8129ed29
ON
1570 sb_freeze_unlock(sb);
1571out:
5accdf82 1572 wake_up(&sb->s_writers.wait_unfrozen);
18e9e510 1573 deactivate_locked_super(sb);
18e9e510
JB
1574 return 0;
1575}
08fdc8a0
MG
1576
1577int thaw_super(struct super_block *sb)
1578{
1579 down_write(&sb->s_umount);
1580 return thaw_super_locked(sb);
1581}
18e9e510 1582EXPORT_SYMBOL(thaw_super);