blk-cgroup: fix rcu lockdep warning in blkg_lookup()
[linux-block.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>
22d94f49 34#include <linux/fscrypt.h>
40401530 35#include <linux/fsnotify.h>
5accdf82 36#include <linux/lockdep.h>
6e4eab57 37#include <linux/user_namespace.h>
9bc61ab1 38#include <linux/fs_context.h>
e262e32d 39#include <uapi/linux/mount.h>
6d59e7f5 40#include "internal.h"
1da177e4 41
880b9577 42static int thaw_super_locked(struct super_block *sb, enum freeze_holder who);
1da177e4 43
15d0f5ea
AV
44static LIST_HEAD(super_blocks);
45static DEFINE_SPINLOCK(sb_lock);
1da177e4 46
5accdf82
JK
47static char *sb_writers_name[SB_FREEZE_LEVELS] = {
48 "sb_writers",
49 "sb_pagefaults",
50 "sb_internal",
51};
52
5e874914 53static inline void __super_lock(struct super_block *sb, bool excl)
0ed33598
CB
54{
55 if (excl)
56 down_write(&sb->s_umount);
57 else
58 down_read(&sb->s_umount);
59}
60
61static inline void super_unlock(struct super_block *sb, bool excl)
62{
63 if (excl)
64 up_write(&sb->s_umount);
65 else
66 up_read(&sb->s_umount);
67}
68
5e874914 69static inline void __super_lock_excl(struct super_block *sb)
0ed33598 70{
5e874914 71 __super_lock(sb, true);
0ed33598
CB
72}
73
74static inline void super_unlock_excl(struct super_block *sb)
75{
76 super_unlock(sb, true);
77}
78
79static inline void super_unlock_shared(struct super_block *sb)
80{
81 super_unlock(sb, false);
82}
83
5e874914
CB
84static inline bool wait_born(struct super_block *sb)
85{
86 unsigned int flags;
87
88 /*
89 * Pairs with smp_store_release() in super_wake() and ensures
90 * that we see SB_BORN or SB_DYING after we're woken.
91 */
92 flags = smp_load_acquire(&sb->s_flags);
93 return flags & (SB_BORN | SB_DYING);
94}
95
96/**
97 * super_lock - wait for superblock to become ready and lock it
98 * @sb: superblock to wait for
99 * @excl: whether exclusive access is required
100 *
101 * If the superblock has neither passed through vfs_get_tree() or
102 * generic_shutdown_super() yet wait for it to happen. Either superblock
103 * creation will succeed and SB_BORN is set by vfs_get_tree() or we're
104 * woken and we'll see SB_DYING.
105 *
106 * The caller must have acquired a temporary reference on @sb->s_count.
107 *
108 * Return: This returns true if SB_BORN was set, false if SB_DYING was
109 * set. The function acquires s_umount and returns with it held.
110 */
111static __must_check bool super_lock(struct super_block *sb, bool excl)
112{
113
114 lockdep_assert_not_held(&sb->s_umount);
115
116relock:
117 __super_lock(sb, excl);
118
119 /*
120 * Has gone through generic_shutdown_super() in the meantime.
121 * @sb->s_root is NULL and @sb->s_active is 0. No one needs to
122 * grab a reference to this. Tell them so.
123 */
124 if (sb->s_flags & SB_DYING)
125 return false;
126
127 /* Has called ->get_tree() successfully. */
128 if (sb->s_flags & SB_BORN)
129 return true;
130
131 super_unlock(sb, excl);
132
133 /* wait until the superblock is ready or dying */
134 wait_var_event(&sb->s_flags, wait_born(sb));
135
136 /*
137 * Neither SB_BORN nor SB_DYING are ever unset so we never loop.
138 * Just reacquire @sb->s_umount for the caller.
139 */
140 goto relock;
141}
142
143/* wait and acquire read-side of @sb->s_umount */
144static inline bool super_lock_shared(struct super_block *sb)
145{
146 return super_lock(sb, false);
147}
148
149/* wait and acquire write-side of @sb->s_umount */
150static inline bool super_lock_excl(struct super_block *sb)
151{
152 return super_lock(sb, true);
153}
154
155/* wake waiters */
2c18a63b 156#define SUPER_WAKE_FLAGS (SB_BORN | SB_DYING | SB_DEAD)
5e874914
CB
157static void super_wake(struct super_block *sb, unsigned int flag)
158{
159 WARN_ON_ONCE((flag & ~SUPER_WAKE_FLAGS));
160 WARN_ON_ONCE(hweight32(flag & SUPER_WAKE_FLAGS) > 1);
161
162 /*
163 * Pairs with smp_load_acquire() in super_lock() to make sure
164 * all initializations in the superblock are seen by the user
165 * seeing SB_BORN sent.
166 */
167 smp_store_release(&sb->s_flags, sb->s_flags | flag);
168 /*
169 * Pairs with the barrier in prepare_to_wait_event() to make sure
170 * ___wait_var_event() either sees SB_BORN set or
171 * waitqueue_active() check in wake_up_var() sees the waiter.
172 */
173 smp_mb();
174 wake_up_var(&sb->s_flags);
175}
176
b0d40c92
DC
177/*
178 * One thing we have to be careful of with a per-sb shrinker is that we don't
179 * drop the last active reference to the superblock from within the shrinker.
180 * If that happens we could trigger unregistering the shrinker from within the
8a0e8bb1 181 * shrinker path and that leads to deadlock on the shrinker_mutex. Hence we
b0d40c92
DC
182 * take a passive reference to the superblock to avoid this from occurring.
183 */
0a234c6d
DC
184static unsigned long super_cache_scan(struct shrinker *shrink,
185 struct shrink_control *sc)
b0d40c92
DC
186{
187 struct super_block *sb;
0a234c6d
DC
188 long fs_objects = 0;
189 long total_objects;
190 long freed = 0;
191 long dentries;
192 long inodes;
b0d40c92 193
1720f5dd 194 sb = shrink->private_data;
b0d40c92
DC
195
196 /*
197 * Deadlock avoidance. We may hold various FS locks, and we don't want
198 * to recurse into the FS that called us in clear_inode() and friends..
199 */
0a234c6d
DC
200 if (!(sc->gfp_mask & __GFP_FS))
201 return SHRINK_STOP;
b0d40c92 202
d8ce82ef 203 if (!super_trylock_shared(sb))
0a234c6d 204 return SHRINK_STOP;
b0d40c92 205
d0407903 206 if (sb->s_op->nr_cached_objects)
4101b624 207 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
0e1fdafd 208
503c358c
VD
209 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
210 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
f6041567 211 total_objects = dentries + inodes + fs_objects + 1;
475d0db7
TH
212 if (!total_objects)
213 total_objects = 1;
0e1fdafd 214
0a234c6d 215 /* proportion the scan between the caches */
f6041567 216 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
bc3b14cb 217 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
503c358c 218 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
b0d40c92 219
0a234c6d
DC
220 /*
221 * prune the dcache first as the icache is pinned by it, then
222 * prune the icache, followed by the filesystem specific caches
49e7e7ff
VD
223 *
224 * Ensure that we always scan at least one object - memcg kmem
225 * accounting uses this to fully empty the caches.
0a234c6d 226 */
49e7e7ff 227 sc->nr_to_scan = dentries + 1;
503c358c 228 freed = prune_dcache_sb(sb, sc);
49e7e7ff 229 sc->nr_to_scan = inodes + 1;
503c358c 230 freed += prune_icache_sb(sb, sc);
0a234c6d
DC
231
232 if (fs_objects) {
49e7e7ff 233 sc->nr_to_scan = fs_objects + 1;
4101b624 234 freed += sb->s_op->free_cached_objects(sb, sc);
b0d40c92
DC
235 }
236
0ed33598 237 super_unlock_shared(sb);
0a234c6d
DC
238 return freed;
239}
240
241static unsigned long super_cache_count(struct shrinker *shrink,
242 struct shrink_control *sc)
243{
244 struct super_block *sb;
245 long total_objects = 0;
246
1720f5dd 247 sb = shrink->private_data;
0a234c6d 248
d23da150 249 /*
d8ce82ef
CB
250 * We don't call super_trylock_shared() here as it is a scalability
251 * bottleneck, so we're exposed to partial setup state. The shrinker
252 * rwsem does not protect filesystem operations backing
253 * list_lru_shrink_count() or s_op->nr_cached_objects(). Counts can
254 * change between super_cache_count and super_cache_scan, so we really
255 * don't need locks here.
79f546a6
DC
256 *
257 * However, if we are currently mounting the superblock, the underlying
258 * filesystem might be in a state of partial construction and hence it
d8ce82ef
CB
259 * is dangerous to access it. super_trylock_shared() uses a SB_BORN check
260 * to avoid this situation, so do the same here. The memory barrier is
79f546a6 261 * matched with the one in mount_fs() as we don't hold locks here.
d23da150 262 */
79f546a6
DC
263 if (!(sb->s_flags & SB_BORN))
264 return 0;
265 smp_rmb();
266
0a234c6d 267 if (sb->s_op && sb->s_op->nr_cached_objects)
4101b624 268 total_objects = sb->s_op->nr_cached_objects(sb, sc);
0a234c6d 269
503c358c
VD
270 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
271 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
0a234c6d 272
9b996468
KT
273 if (!total_objects)
274 return SHRINK_EMPTY;
275
55f841ce 276 total_objects = vfs_pressure_ratio(total_objects);
0e1fdafd 277 return total_objects;
b0d40c92
DC
278}
279
853b39a7
ON
280static void destroy_super_work(struct work_struct *work)
281{
282 struct super_block *s = container_of(work, struct super_block,
283 destroy_work);
284 int i;
285
286 for (i = 0; i < SB_FREEZE_LEVELS; i++)
8129ed29 287 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
853b39a7
ON
288 kfree(s);
289}
290
291static void destroy_super_rcu(struct rcu_head *head)
292{
293 struct super_block *s = container_of(head, struct super_block, rcu);
294 INIT_WORK(&s->destroy_work, destroy_super_work);
295 schedule_work(&s->destroy_work);
296}
297
0200894d
AV
298/* Free a superblock that has never been seen by anyone */
299static void destroy_unused_super(struct super_block *s)
5accdf82 300{
0200894d
AV
301 if (!s)
302 return;
0ed33598 303 super_unlock_excl(s);
7eb5e882
AV
304 list_lru_destroy(&s->s_dentry_lru);
305 list_lru_destroy(&s->s_inode_lru);
7eb5e882 306 security_sb_free(s);
6e4eab57 307 put_user_ns(s->s_user_ns);
7eb5e882 308 kfree(s->s_subtype);
1720f5dd 309 shrinker_free(s->s_shrink);
0200894d
AV
310 /* no delays needed */
311 destroy_super_work(&s->destroy_work);
5accdf82
JK
312}
313
1da177e4
LT
314/**
315 * alloc_super - create new superblock
fe2bbc48 316 * @type: filesystem type superblock should belong to
9249e17f 317 * @flags: the mount flags
6e4eab57 318 * @user_ns: User namespace for the super_block
1da177e4
LT
319 *
320 * Allocates and initializes a new &struct super_block. alloc_super()
321 * returns a pointer new superblock or %NULL if allocation had failed.
322 */
6e4eab57
EB
323static struct super_block *alloc_super(struct file_system_type *type, int flags,
324 struct user_namespace *user_ns)
1da177e4 325{
11b0b5ab 326 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
b87221de 327 static const struct super_operations default_op;
7eb5e882
AV
328 int i;
329
330 if (!s)
331 return NULL;
1da177e4 332
b5bd856a 333 INIT_LIST_HEAD(&s->s_mounts);
6e4eab57 334 s->s_user_ns = get_user_ns(user_ns);
ca0168e8
AV
335 init_rwsem(&s->s_umount);
336 lockdep_set_class(&s->s_umount, &type->s_umount_key);
337 /*
338 * sget() can have s_umount recursion.
339 *
340 * When it cannot find a suitable sb, it allocates a new
341 * one (this one), and tries again to find a suitable old
342 * one.
343 *
344 * In case that succeeds, it will acquire the s_umount
345 * lock of the old one. Since these are clearly distrinct
346 * locks, and this object isn't exposed yet, there's no
347 * risk of deadlocks.
348 *
349 * Annotate this by putting this lock in a different
350 * subclass.
351 */
352 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
b5bd856a 353
7eb5e882
AV
354 if (security_sb_alloc(s))
355 goto fail;
7b7a8665 356
7eb5e882 357 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
8129ed29
ON
358 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
359 sb_writers_name[i],
360 &type->s_writers_key[i]))
7eb5e882 361 goto fail;
1da177e4 362 }
df0ce26c 363 s->s_bdi = &noop_backing_dev_info;
7eb5e882 364 s->s_flags = flags;
cc50a07a 365 if (s->s_user_ns != &init_user_ns)
67690f93 366 s->s_iflags |= SB_I_NODEV;
7eb5e882 367 INIT_HLIST_NODE(&s->s_instances);
f1ee6162 368 INIT_HLIST_BL_HEAD(&s->s_roots);
e97fedb9 369 mutex_init(&s->s_sync_lock);
7eb5e882 370 INIT_LIST_HEAD(&s->s_inodes);
74278da9 371 spin_lock_init(&s->s_inode_list_lock);
6c60d2b5
DC
372 INIT_LIST_HEAD(&s->s_inodes_wb);
373 spin_lock_init(&s->s_inode_wblist_lock);
7eb5e882 374
7eb5e882
AV
375 s->s_count = 1;
376 atomic_set(&s->s_active, 1);
377 mutex_init(&s->s_vfs_rename_mutex);
378 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
bc8230ee 379 init_rwsem(&s->s_dquot.dqio_sem);
7eb5e882
AV
380 s->s_maxbytes = MAX_NON_LFS;
381 s->s_op = &default_op;
382 s->s_time_gran = 1000000000;
188d20bc
DD
383 s->s_time_min = TIME64_MIN;
384 s->s_time_max = TIME64_MAX;
7eb5e882 385
1720f5dd
QZ
386 s->s_shrink = shrinker_alloc(SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE,
387 "sb-%s", type->name);
388 if (!s->s_shrink)
8e04944f 389 goto fail;
1720f5dd
QZ
390
391 s->s_shrink->scan_objects = super_cache_scan;
392 s->s_shrink->count_objects = super_cache_count;
393 s->s_shrink->batch = 1024;
394 s->s_shrink->private_data = s;
395
396 if (list_lru_init_memcg(&s->s_dentry_lru, s->s_shrink))
2b3648a6 397 goto fail;
1720f5dd 398 if (list_lru_init_memcg(&s->s_inode_lru, s->s_shrink))
2b3648a6 399 goto fail;
1da177e4 400 return s;
5ca302c8 401
7eb5e882 402fail:
0200894d 403 destroy_unused_super(s);
7eb5e882 404 return NULL;
1da177e4
LT
405}
406
407/* Superblock refcounting */
408
409/*
35cf7ba0 410 * Drop a superblock's refcount. The caller must hold sb_lock.
1da177e4 411 */
c645b930 412static void __put_super(struct super_block *s)
1da177e4 413{
c645b930
AV
414 if (!--s->s_count) {
415 list_del_init(&s->s_list);
416 WARN_ON(s->s_dentry_lru.node);
417 WARN_ON(s->s_inode_lru.node);
418 WARN_ON(!list_empty(&s->s_mounts));
419 security_sb_free(s);
420 put_user_ns(s->s_user_ns);
421 kfree(s->s_subtype);
422 call_rcu(&s->rcu, destroy_super_rcu);
1da177e4 423 }
1da177e4
LT
424}
425
426/**
427 * put_super - drop a temporary reference to superblock
428 * @sb: superblock in question
429 *
430 * Drops a temporary reference, frees superblock if there's no
431 * references left.
432 */
60b49885 433void put_super(struct super_block *sb)
1da177e4
LT
434{
435 spin_lock(&sb_lock);
436 __put_super(sb);
437 spin_unlock(&sb_lock);
438}
439
dc3216b1
CB
440static void kill_super_notify(struct super_block *sb)
441{
442 lockdep_assert_not_held(&sb->s_umount);
443
444 /* already notified earlier */
445 if (sb->s_flags & SB_DEAD)
446 return;
447
448 /*
449 * Remove it from @fs_supers so it isn't found by new
450 * sget{_fc}() walkers anymore. Any concurrent mounter still
451 * managing to grab a temporary reference is guaranteed to
452 * already see SB_DYING and will wait until we notify them about
453 * SB_DEAD.
454 */
455 spin_lock(&sb_lock);
456 hlist_del_init(&sb->s_instances);
457 spin_unlock(&sb_lock);
458
459 /*
460 * Let concurrent mounts know that this thing is really dead.
461 * We don't need @sb->s_umount here as every concurrent caller
462 * will see SB_DYING and either discard the superblock or wait
463 * for SB_DEAD.
464 */
465 super_wake(sb, SB_DEAD);
466}
1da177e4
LT
467
468/**
1712ac8f 469 * deactivate_locked_super - drop an active reference to superblock
1da177e4
LT
470 * @s: superblock to deactivate
471 *
bd7ced98 472 * Drops an active reference to superblock, converting it into a temporary
1712ac8f 473 * one if there is no other active references left. In that case we
1da177e4
LT
474 * tell fs driver to shut it down and drop the temporary reference we
475 * had just acquired.
1712ac8f
AV
476 *
477 * Caller holds exclusive lock on superblock; that lock is released.
1da177e4 478 */
1712ac8f 479void deactivate_locked_super(struct super_block *s)
1da177e4
LT
480{
481 struct file_system_type *fs = s->s_type;
b20bd1a5 482 if (atomic_dec_and_test(&s->s_active)) {
1720f5dd 483 shrinker_free(s->s_shrink);
28f2cd4f 484 fs->kill_sb(s);
f5e1dd34 485
dc3216b1
CB
486 kill_super_notify(s);
487
c0a5b560
VD
488 /*
489 * Since list_lru_destroy() may sleep, we cannot call it from
490 * put_super(), where we hold the sb_lock. Therefore we destroy
491 * the lru lists right now.
492 */
493 list_lru_destroy(&s->s_dentry_lru);
494 list_lru_destroy(&s->s_inode_lru);
495
1da177e4
LT
496 put_filesystem(fs);
497 put_super(s);
1712ac8f 498 } else {
0ed33598 499 super_unlock_excl(s);
1da177e4
LT
500 }
501}
502
1712ac8f 503EXPORT_SYMBOL(deactivate_locked_super);
1da177e4 504
74dbbdd7 505/**
1712ac8f 506 * deactivate_super - drop an active reference to superblock
74dbbdd7
AV
507 * @s: superblock to deactivate
508 *
1712ac8f
AV
509 * Variant of deactivate_locked_super(), except that superblock is *not*
510 * locked by caller. If we are going to drop the final active reference,
511 * lock will be acquired prior to that.
74dbbdd7 512 */
1712ac8f 513void deactivate_super(struct super_block *s)
74dbbdd7 514{
cc23402c 515 if (!atomic_add_unless(&s->s_active, -1, 1)) {
5e874914 516 __super_lock_excl(s);
1712ac8f 517 deactivate_locked_super(s);
74dbbdd7
AV
518 }
519}
520
1712ac8f 521EXPORT_SYMBOL(deactivate_super);
74dbbdd7 522
1da177e4
LT
523/**
524 * grab_super - acquire an active reference
525 * @s: reference we are trying to make active
526 *
527 * Tries to acquire an active reference. grab_super() is used when we
528 * had just found a superblock in super_blocks or fs_type->fs_supers
529 * and want to turn it into a full-blown active reference. grab_super()
530 * is called with sb_lock held and drops it. Returns 1 in case of
531 * success, 0 if we had failed (superblock contents was already dead or
acfec9a5
AV
532 * dying when grab_super() had been called). Note that this is only
533 * called for superblocks not in rundown mode (== ones still on ->fs_supers
534 * of their type), so increment of ->s_count is OK here.
1da177e4 535 */
9c4dbee7 536static int grab_super(struct super_block *s) __releases(sb_lock)
1da177e4 537{
5e874914
CB
538 bool born;
539
1da177e4
LT
540 s->s_count++;
541 spin_unlock(&sb_lock);
5e874914
CB
542 born = super_lock_excl(s);
543 if (born && atomic_inc_not_zero(&s->s_active)) {
acfec9a5
AV
544 put_super(s);
545 return 1;
546 }
0ed33598 547 super_unlock_excl(s);
1da177e4 548 put_super(s);
1da177e4
LT
549 return 0;
550}
551
2c18a63b
CB
552static inline bool wait_dead(struct super_block *sb)
553{
554 unsigned int flags;
555
556 /*
557 * Pairs with memory barrier in super_wake() and ensures
558 * that we see SB_DEAD after we're woken.
559 */
560 flags = smp_load_acquire(&sb->s_flags);
561 return flags & SB_DEAD;
562}
563
564/**
565 * grab_super_dead - acquire an active reference to a superblock
566 * @sb: superblock to acquire
567 *
568 * Acquire a temporary reference on a superblock and try to trade it for
569 * an active reference. This is used in sget{_fc}() to wait for a
570 * superblock to either become SB_BORN or for it to pass through
571 * sb->kill() and be marked as SB_DEAD.
572 *
573 * Return: This returns true if an active reference could be acquired,
574 * false if not.
575 */
576static bool grab_super_dead(struct super_block *sb)
577{
578
579 sb->s_count++;
580 if (grab_super(sb)) {
581 put_super(sb);
582 lockdep_assert_held(&sb->s_umount);
583 return true;
584 }
585 wait_var_event(&sb->s_flags, wait_dead(sb));
2c18a63b 586 lockdep_assert_not_held(&sb->s_umount);
345a5c4a 587 put_super(sb);
2c18a63b
CB
588 return false;
589}
590
12ad3ab6 591/*
d8ce82ef 592 * super_trylock_shared - try to grab ->s_umount shared
331cbdee 593 * @sb: reference we are trying to grab
12ad3ab6 594 *
eb6ef3df 595 * Try to prevent fs shutdown. This is used in places where we
12ad3ab6 596 * cannot take an active reference but we need to ensure that the
eb6ef3df
KK
597 * filesystem is not shut down while we are working on it. It returns
598 * false if we cannot acquire s_umount or if we lose the race and
599 * filesystem already got into shutdown, and returns true with the s_umount
600 * lock held in read mode in case of success. On successful return,
601 * the caller must drop the s_umount lock when done.
602 *
603 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
604 * The reason why it's safe is that we are OK with doing trylock instead
605 * of down_read(). There's a couple of places that are OK with that, but
606 * it's very much not a general-purpose interface.
12ad3ab6 607 */
d8ce82ef 608bool super_trylock_shared(struct super_block *sb)
12ad3ab6 609{
12ad3ab6 610 if (down_read_trylock(&sb->s_umount)) {
5e874914
CB
611 if (!(sb->s_flags & SB_DYING) && sb->s_root &&
612 (sb->s_flags & SB_BORN))
12ad3ab6 613 return true;
0ed33598 614 super_unlock_shared(sb);
12ad3ab6
DC
615 }
616
12ad3ab6
DC
617 return false;
618}
619
04b94071
DL
620/**
621 * retire_super - prevents superblock from being reused
622 * @sb: superblock to retire
623 *
624 * The function marks superblock to be ignored in superblock test, which
625 * prevents it from being reused for any new mounts. If the superblock has
626 * a private bdi, it also unregisters it, but doesn't reduce the refcount
627 * of the superblock to prevent potential races. The refcount is reduced
628 * by generic_shutdown_super(). The function can not be called
629 * concurrently with generic_shutdown_super(). It is safe to call the
630 * function multiple times, subsequent calls have no effect.
631 *
632 * The marker will affect the re-use only for block-device-based
633 * superblocks. Other superblocks will still get marked if this function
634 * is used, but that will not affect their reusability.
635 */
636void retire_super(struct super_block *sb)
637{
638 WARN_ON(!sb->s_bdev);
5e874914 639 __super_lock_excl(sb);
04b94071
DL
640 if (sb->s_iflags & SB_I_PERSB_BDI) {
641 bdi_unregister(sb->s_bdi);
642 sb->s_iflags &= ~SB_I_PERSB_BDI;
643 }
644 sb->s_iflags |= SB_I_RETIRED;
0ed33598 645 super_unlock_excl(sb);
04b94071
DL
646}
647EXPORT_SYMBOL(retire_super);
648
1da177e4
LT
649/**
650 * generic_shutdown_super - common helper for ->kill_sb()
651 * @sb: superblock to kill
652 *
653 * generic_shutdown_super() does all fs-independent work on superblock
654 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
655 * that need destruction out of superblock, call generic_shutdown_super()
656 * and release aforementioned objects. Note: dentries and inodes _are_
657 * taken care of and do not need specific handling.
c636ebdb
DH
658 *
659 * Upon calling this function, the filesystem may no longer alter or
660 * rearrange the set of dentries belonging to this super_block, nor may it
661 * change the attachments of dentries to inodes.
1da177e4
LT
662 */
663void generic_shutdown_super(struct super_block *sb)
664{
ee9b6d61 665 const struct super_operations *sop = sb->s_op;
1da177e4 666
c636ebdb
DH
667 if (sb->s_root) {
668 shrink_dcache_for_umount(sb);
60b0680f 669 sync_filesystem(sb);
e462ec50 670 sb->s_flags &= ~SB_ACTIVE;
efaee192 671
a1a0e23e 672 cgroup_writeback_umount();
63997e98 673
ccb820dc 674 /* Evict all inodes with zero refcount. */
63997e98 675 evict_inodes(sb);
ccb820dc
EB
676
677 /*
678 * Clean up and evict any inodes that still have references due
679 * to fsnotify or the security policy.
680 */
1edc8eb2 681 fsnotify_sb_delete(sb);
83e804f0 682 security_sb_delete(sb);
1da177e4 683
ccb820dc
EB
684 /*
685 * Now that all potentially-encrypted inodes have been evicted,
686 * the fscrypt keyring can be destroyed.
687 */
688 fscrypt_destroy_keyring(sb);
689
7b7a8665
CH
690 if (sb->s_dio_done_wq) {
691 destroy_workqueue(sb->s_dio_done_wq);
692 sb->s_dio_done_wq = NULL;
693 }
694
1da177e4
LT
695 if (sop->put_super)
696 sop->put_super(sb);
697
47d58691
JH
698 if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes),
699 "VFS: Busy inodes after unmount of %s (%s)",
700 sb->s_id, sb->s_type->name)) {
701 /*
702 * Adding a proper bailout path here would be hard, but
703 * we can at least make it more likely that a later
704 * iput_final() or such crashes cleanly.
705 */
706 struct inode *inode;
707
708 spin_lock(&sb->s_inode_list_lock);
709 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
710 inode->i_op = VFS_PTR_POISON;
711 inode->i_sb = VFS_PTR_POISON;
712 inode->i_mapping = VFS_PTR_POISON;
713 }
714 spin_unlock(&sb->s_inode_list_lock);
1da177e4 715 }
1da177e4 716 }
5e874914
CB
717 /*
718 * Broadcast to everyone that grabbed a temporary reference to this
719 * superblock before we removed it from @fs_supers that the superblock
720 * is dying. Every walker of @fs_supers outside of sget{_fc}() will now
721 * discard this superblock and treat it as dead.
2c18a63b
CB
722 *
723 * We leave the superblock on @fs_supers so it can be found by
724 * sget{_fc}() until we passed sb->kill_sb().
5e874914
CB
725 */
726 super_wake(sb, SB_DYING);
0ed33598 727 super_unlock_excl(sb);
c1844d53 728 if (sb->s_bdi != &noop_backing_dev_info) {
0b3ea092
CH
729 if (sb->s_iflags & SB_I_PERSB_BDI)
730 bdi_unregister(sb->s_bdi);
fca39346
JK
731 bdi_put(sb->s_bdi);
732 sb->s_bdi = &noop_backing_dev_info;
fca39346 733 }
1da177e4
LT
734}
735
736EXPORT_SYMBOL(generic_shutdown_super);
737
20284ab7 738bool mount_capable(struct fs_context *fc)
0ce0cf12 739{
20284ab7 740 if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
0ce0cf12
AV
741 return capable(CAP_SYS_ADMIN);
742 else
c2c44ec2 743 return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
0ce0cf12
AV
744}
745
cb50b348
AV
746/**
747 * sget_fc - Find or create a superblock
748 * @fc: Filesystem context.
749 * @test: Comparison callback
750 * @set: Setup callback
751 *
22ed7ecd 752 * Create a new superblock or find an existing one.
cb50b348 753 *
22ed7ecd
CB
754 * The @test callback is used to find a matching existing superblock.
755 * Whether or not the requested parameters in @fc are taken into account
756 * is specific to the @test callback that is used. They may even be
757 * completely ignored.
758 *
759 * If an extant superblock is matched, it will be returned unless:
760 *
761 * (1) the namespace the filesystem context @fc and the extant
762 * superblock's namespace differ
763 *
764 * (2) the filesystem context @fc has requested that reusing an extant
765 * superblock is not allowed
766 *
767 * In both cases EBUSY will be returned.
cb50b348
AV
768 *
769 * If no match is made, a new superblock will be allocated and basic
22ed7ecd
CB
770 * initialisation will be performed (s_type, s_fs_info and s_id will be
771 * set and the @set callback will be invoked), the superblock will be
772 * published and it will be returned in a partially constructed state
773 * with SB_BORN and SB_ACTIVE as yet unset.
774 *
775 * Return: On success, an extant or newly created superblock is
776 * returned. On failure an error pointer is returned.
cb50b348
AV
777 */
778struct super_block *sget_fc(struct fs_context *fc,
779 int (*test)(struct super_block *, struct fs_context *),
780 int (*set)(struct super_block *, struct fs_context *))
781{
782 struct super_block *s = NULL;
783 struct super_block *old;
784 struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
785 int err;
786
cb50b348
AV
787retry:
788 spin_lock(&sb_lock);
789 if (test) {
790 hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
791 if (test(old, fc))
792 goto share_extant_sb;
793 }
794 }
795 if (!s) {
796 spin_unlock(&sb_lock);
797 s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
798 if (!s)
799 return ERR_PTR(-ENOMEM);
800 goto retry;
801 }
802
803 s->s_fs_info = fc->s_fs_info;
804 err = set(s, fc);
805 if (err) {
806 s->s_fs_info = NULL;
807 spin_unlock(&sb_lock);
808 destroy_unused_super(s);
809 return ERR_PTR(err);
810 }
811 fc->s_fs_info = NULL;
812 s->s_type = fc->fs_type;
c80fa7c8 813 s->s_iflags |= fc->s_iflags;
c642256b 814 strscpy(s->s_id, s->s_type->name, sizeof(s->s_id));
5e874914
CB
815 /*
816 * Make the superblock visible on @super_blocks and @fs_supers.
817 * It's in a nascent state and users should wait on SB_BORN or
818 * SB_DYING to be set.
819 */
cb50b348
AV
820 list_add_tail(&s->s_list, &super_blocks);
821 hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
822 spin_unlock(&sb_lock);
823 get_filesystem(s->s_type);
1720f5dd 824 shrinker_register(s->s_shrink);
cb50b348
AV
825 return s;
826
827share_extant_sb:
22ed7ecd 828 if (user_ns != old->s_user_ns || fc->exclusive) {
cb50b348
AV
829 spin_unlock(&sb_lock);
830 destroy_unused_super(s);
22ed7ecd
CB
831 if (fc->exclusive)
832 warnfc(fc, "reusing existing filesystem not allowed");
833 else
834 warnfc(fc, "reusing existing filesystem in another namespace not allowed");
cb50b348
AV
835 return ERR_PTR(-EBUSY);
836 }
2c18a63b 837 if (!grab_super_dead(old))
cb50b348
AV
838 goto retry;
839 destroy_unused_super(s);
840 return old;
841}
842EXPORT_SYMBOL(sget_fc);
843
1da177e4 844/**
023d066a
DH
845 * sget - find or create a superblock
846 * @type: filesystem type superblock should belong to
847 * @test: comparison callback
848 * @set: setup callback
849 * @flags: mount flags
850 * @data: argument to each of them
1da177e4 851 */
023d066a 852struct super_block *sget(struct file_system_type *type,
1da177e4
LT
853 int (*test)(struct super_block *,void *),
854 int (*set)(struct super_block *,void *),
023d066a 855 int flags,
1da177e4
LT
856 void *data)
857{
023d066a 858 struct user_namespace *user_ns = current_user_ns();
1da177e4 859 struct super_block *s = NULL;
d4730127 860 struct super_block *old;
1da177e4
LT
861 int err;
862
023d066a
DH
863 /* We don't yet pass the user namespace of the parent
864 * mount through to here so always use &init_user_ns
865 * until that changes.
866 */
867 if (flags & SB_SUBMOUNT)
868 user_ns = &init_user_ns;
869
1da177e4
LT
870retry:
871 spin_lock(&sb_lock);
d4730127 872 if (test) {
b67bfe0d 873 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
d4730127
MK
874 if (!test(old, data))
875 continue;
6e4eab57
EB
876 if (user_ns != old->s_user_ns) {
877 spin_unlock(&sb_lock);
0200894d 878 destroy_unused_super(s);
6e4eab57
EB
879 return ERR_PTR(-EBUSY);
880 }
2c18a63b 881 if (!grab_super_dead(old))
d4730127 882 goto retry;
0200894d 883 destroy_unused_super(s);
d4730127
MK
884 return old;
885 }
1da177e4
LT
886 }
887 if (!s) {
888 spin_unlock(&sb_lock);
e462ec50 889 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
1da177e4
LT
890 if (!s)
891 return ERR_PTR(-ENOMEM);
892 goto retry;
893 }
dd111b31 894
1da177e4
LT
895 err = set(s, data);
896 if (err) {
897 spin_unlock(&sb_lock);
0200894d 898 destroy_unused_super(s);
1da177e4
LT
899 return ERR_PTR(err);
900 }
901 s->s_type = type;
c642256b 902 strscpy(s->s_id, type->name, sizeof(s->s_id));
1da177e4 903 list_add_tail(&s->s_list, &super_blocks);
a5166169 904 hlist_add_head(&s->s_instances, &type->fs_supers);
1da177e4
LT
905 spin_unlock(&sb_lock);
906 get_filesystem(type);
1720f5dd 907 shrinker_register(s->s_shrink);
1da177e4
LT
908 return s;
909}
1da177e4
LT
910EXPORT_SYMBOL(sget);
911
912void drop_super(struct super_block *sb)
913{
0ed33598 914 super_unlock_shared(sb);
1da177e4
LT
915 put_super(sb);
916}
917
918EXPORT_SYMBOL(drop_super);
919
ba6379f7
JK
920void drop_super_exclusive(struct super_block *sb)
921{
0ed33598 922 super_unlock_excl(sb);
ba6379f7
JK
923 put_super(sb);
924}
925EXPORT_SYMBOL(drop_super_exclusive);
926
fa7c1d50
MG
927static void __iterate_supers(void (*f)(struct super_block *))
928{
929 struct super_block *sb, *p = NULL;
930
931 spin_lock(&sb_lock);
932 list_for_each_entry(sb, &super_blocks, s_list) {
5e874914
CB
933 /* Pairs with memory marrier in super_wake(). */
934 if (smp_load_acquire(&sb->s_flags) & SB_DYING)
fa7c1d50
MG
935 continue;
936 sb->s_count++;
937 spin_unlock(&sb_lock);
938
939 f(sb);
940
941 spin_lock(&sb_lock);
942 if (p)
943 __put_super(p);
944 p = sb;
945 }
946 if (p)
947 __put_super(p);
948 spin_unlock(&sb_lock);
949}
01a05b33
AV
950/**
951 * iterate_supers - call function for all active superblocks
952 * @f: function to call
953 * @arg: argument to pass to it
954 *
955 * Scans the superblock list and calls given function, passing it
956 * locked superblock and given argument.
957 */
958void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
959{
dca33252 960 struct super_block *sb, *p = NULL;
01a05b33
AV
961
962 spin_lock(&sb_lock);
dca33252 963 list_for_each_entry(sb, &super_blocks, s_list) {
5e874914
CB
964 bool born;
965
01a05b33
AV
966 sb->s_count++;
967 spin_unlock(&sb_lock);
968
5e874914
CB
969 born = super_lock_shared(sb);
970 if (born && sb->s_root)
01a05b33 971 f(sb, arg);
0ed33598 972 super_unlock_shared(sb);
01a05b33
AV
973
974 spin_lock(&sb_lock);
dca33252
AV
975 if (p)
976 __put_super(p);
977 p = sb;
01a05b33 978 }
dca33252
AV
979 if (p)
980 __put_super(p);
01a05b33
AV
981 spin_unlock(&sb_lock);
982}
983
43e15cdb
AV
984/**
985 * iterate_supers_type - call function for superblocks of given type
986 * @type: fs type
987 * @f: function to call
988 * @arg: argument to pass to it
989 *
990 * Scans the superblock list and calls given function, passing it
991 * locked superblock and given argument.
992 */
993void iterate_supers_type(struct file_system_type *type,
994 void (*f)(struct super_block *, void *), void *arg)
995{
996 struct super_block *sb, *p = NULL;
997
998 spin_lock(&sb_lock);
b67bfe0d 999 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
5e874914
CB
1000 bool born;
1001
43e15cdb
AV
1002 sb->s_count++;
1003 spin_unlock(&sb_lock);
1004
5e874914
CB
1005 born = super_lock_shared(sb);
1006 if (born && sb->s_root)
43e15cdb 1007 f(sb, arg);
0ed33598 1008 super_unlock_shared(sb);
43e15cdb
AV
1009
1010 spin_lock(&sb_lock);
1011 if (p)
1012 __put_super(p);
1013 p = sb;
1014 }
1015 if (p)
1016 __put_super(p);
1017 spin_unlock(&sb_lock);
1018}
1019
1020EXPORT_SYMBOL(iterate_supers_type);
1021
4504230a
CH
1022/**
1023 * get_active_super - get an active reference to the superblock of a device
1024 * @bdev: device to get the superblock for
1025 *
1026 * Scans the superblock list and finds the superblock of the file system
1027 * mounted on the device given. Returns the superblock with an active
d3f21473 1028 * reference or %NULL if none was found.
4504230a
CH
1029 */
1030struct super_block *get_active_super(struct block_device *bdev)
1031{
1032 struct super_block *sb;
1033
1034 if (!bdev)
1035 return NULL;
1036
1037 spin_lock(&sb_lock);
1038 list_for_each_entry(sb, &super_blocks, s_list) {
1494583d 1039 if (sb->s_bdev == bdev) {
acfec9a5 1040 if (!grab_super(sb))
5e874914 1041 return NULL;
0ed33598 1042 super_unlock_excl(sb);
acfec9a5 1043 return sb;
1494583d 1044 }
4504230a
CH
1045 }
1046 spin_unlock(&sb_lock);
1047 return NULL;
1048}
dd111b31 1049
4e7b5671 1050struct super_block *user_get_super(dev_t dev, bool excl)
1da177e4 1051{
618f0636 1052 struct super_block *sb;
1da177e4 1053
1da177e4 1054 spin_lock(&sb_lock);
618f0636
KK
1055 list_for_each_entry(sb, &super_blocks, s_list) {
1056 if (sb->s_dev == dev) {
5e874914
CB
1057 bool born;
1058
618f0636 1059 sb->s_count++;
1da177e4 1060 spin_unlock(&sb_lock);
df40c01a 1061 /* still alive? */
5e874914
CB
1062 born = super_lock(sb, excl);
1063 if (born && sb->s_root)
618f0636 1064 return sb;
0ed33598 1065 super_unlock(sb, excl);
df40c01a 1066 /* nope, got unmounted */
618f0636 1067 spin_lock(&sb_lock);
df40c01a 1068 __put_super(sb);
5e874914 1069 break;
1da177e4
LT
1070 }
1071 }
1072 spin_unlock(&sb_lock);
1073 return NULL;
1074}
1075
1da177e4 1076/**
8d0347f6
DH
1077 * reconfigure_super - asks filesystem to change superblock parameters
1078 * @fc: The superblock and configuration
1da177e4 1079 *
8d0347f6 1080 * Alters the configuration parameters of a live superblock.
1da177e4 1081 */
8d0347f6 1082int reconfigure_super(struct fs_context *fc)
1da177e4 1083{
8d0347f6 1084 struct super_block *sb = fc->root->d_sb;
1da177e4 1085 int retval;
8d0347f6 1086 bool remount_ro = false;
c541dce8 1087 bool remount_rw = false;
8d0347f6 1088 bool force = fc->sb_flags & SB_FORCE;
4504230a 1089
8d0347f6
DH
1090 if (fc->sb_flags_mask & ~MS_RMT_MASK)
1091 return -EINVAL;
5accdf82 1092 if (sb->s_writers.frozen != SB_UNFROZEN)
4504230a
CH
1093 return -EBUSY;
1094
8d0347f6
DH
1095 retval = security_sb_remount(sb, fc->security);
1096 if (retval)
1097 return retval;
1098
1099 if (fc->sb_flags_mask & SB_RDONLY) {
9361401e 1100#ifdef CONFIG_BLOCK
6f0d9689
CH
1101 if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
1102 bdev_read_only(sb->s_bdev))
8d0347f6 1103 return -EACCES;
9361401e 1104#endif
c541dce8 1105 remount_rw = !(fc->sb_flags & SB_RDONLY) && sb_rdonly(sb);
8d0347f6
DH
1106 remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
1107 }
d208bbdd 1108
0aec09d0 1109 if (remount_ro) {
fdab684d 1110 if (!hlist_empty(&sb->s_pins)) {
0ed33598 1111 super_unlock_excl(sb);
fdab684d 1112 group_pin_kill(&sb->s_pins);
5e874914 1113 __super_lock_excl(sb);
0aec09d0
AV
1114 if (!sb->s_root)
1115 return 0;
1116 if (sb->s_writers.frozen != SB_UNFROZEN)
1117 return -EBUSY;
8d0347f6 1118 remount_ro = !sb_rdonly(sb);
0aec09d0
AV
1119 }
1120 }
1121 shrink_dcache_sb(sb);
1122
8d0347f6
DH
1123 /* If we are reconfiguring to RDONLY and current sb is read/write,
1124 * make sure there are no files open for writing.
1125 */
d208bbdd 1126 if (remount_ro) {
4ed5e82f 1127 if (force) {
d7439fb1 1128 sb_start_ro_state_change(sb);
4ed5e82f
MS
1129 } else {
1130 retval = sb_prepare_remount_readonly(sb);
1131 if (retval)
1132 return retval;
4ed5e82f 1133 }
c541dce8
JK
1134 } else if (remount_rw) {
1135 /*
d7439fb1
JK
1136 * Protect filesystem's reconfigure code from writes from
1137 * userspace until reconfigure finishes.
c541dce8 1138 */
d7439fb1 1139 sb_start_ro_state_change(sb);
1da177e4
LT
1140 }
1141
f3a09c92
AV
1142 if (fc->ops->reconfigure) {
1143 retval = fc->ops->reconfigure(fc);
1144 if (retval) {
1145 if (!force)
1146 goto cancel_readonly;
1147 /* If forced remount, go ahead despite any errors */
1148 WARN(1, "forced remount of a %s fs returned %i\n",
1149 sb->s_type->name, retval);
1150 }
1da177e4 1151 }
8d0347f6
DH
1152
1153 WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
1154 (fc->sb_flags & fc->sb_flags_mask)));
d7439fb1 1155 sb_end_ro_state_change(sb);
c79d967d 1156
d208bbdd
NP
1157 /*
1158 * Some filesystems modify their metadata via some other path than the
1159 * bdev buffer cache (eg. use a private mapping, or directories in
1160 * pagecache, etc). Also file data modifications go via their own
1161 * mappings. So If we try to mount readonly then copy the filesystem
1162 * from bdev, we could get stale data, so invalidate it to give a best
1163 * effort at coherency.
1164 */
1165 if (remount_ro && sb->s_bdev)
1166 invalidate_bdev(sb->s_bdev);
1da177e4 1167 return 0;
4ed5e82f
MS
1168
1169cancel_readonly:
d7439fb1 1170 sb_end_ro_state_change(sb);
4ed5e82f 1171 return retval;
1da177e4
LT
1172}
1173
fa7c1d50 1174static void do_emergency_remount_callback(struct super_block *sb)
1da177e4 1175{
5e874914
CB
1176 bool born = super_lock_excl(sb);
1177
1178 if (born && sb->s_root && sb->s_bdev && !sb_rdonly(sb)) {
8d0347f6
DH
1179 struct fs_context *fc;
1180
1181 fc = fs_context_for_reconfigure(sb->s_root,
1182 SB_RDONLY | SB_FORCE, SB_RDONLY);
1183 if (!IS_ERR(fc)) {
1184 if (parse_monolithic_mount_data(fc, NULL) == 0)
1185 (void)reconfigure_super(fc);
1186 put_fs_context(fc);
1187 }
1da177e4 1188 }
0ed33598 1189 super_unlock_excl(sb);
fa7c1d50
MG
1190}
1191
1192static void do_emergency_remount(struct work_struct *work)
1193{
1194 __iterate_supers(do_emergency_remount_callback);
a2a9537a 1195 kfree(work);
1da177e4
LT
1196 printk("Emergency Remount complete\n");
1197}
1198
1199void emergency_remount(void)
1200{
a2a9537a
JA
1201 struct work_struct *work;
1202
1203 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1204 if (work) {
1205 INIT_WORK(work, do_emergency_remount);
1206 schedule_work(work);
1207 }
1da177e4
LT
1208}
1209
08fdc8a0
MG
1210static void do_thaw_all_callback(struct super_block *sb)
1211{
5e874914
CB
1212 bool born = super_lock_excl(sb);
1213
1214 if (born && sb->s_root) {
4a8b719f
CH
1215 if (IS_ENABLED(CONFIG_BLOCK))
1216 while (sb->s_bdev && !thaw_bdev(sb->s_bdev))
1217 pr_warn("Emergency Thaw on %pg\n", sb->s_bdev);
880b9577 1218 thaw_super_locked(sb, FREEZE_HOLDER_USERSPACE);
08fdc8a0 1219 } else {
0ed33598 1220 super_unlock_excl(sb);
08fdc8a0
MG
1221 }
1222}
1223
1224static void do_thaw_all(struct work_struct *work)
1225{
1226 __iterate_supers(do_thaw_all_callback);
1227 kfree(work);
1228 printk(KERN_WARNING "Emergency Thaw complete\n");
1229}
1230
1231/**
1232 * emergency_thaw_all -- forcibly thaw every frozen filesystem
1233 *
1234 * Used for emergency unfreeze of all filesystems via SysRq
1235 */
1236void emergency_thaw_all(void)
1237{
1238 struct work_struct *work;
1239
1240 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1241 if (work) {
1242 INIT_WORK(work, do_thaw_all);
1243 schedule_work(work);
1244 }
1245}
1246
ad76cbc6 1247static DEFINE_IDA(unnamed_dev_ida);
1da177e4 1248
5a66847e
MW
1249/**
1250 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1251 * @p: Pointer to a dev_t.
1252 *
1253 * Filesystems which don't use real block devices can call this function
1254 * to allocate a virtual block device.
1255 *
1256 * Context: Any context. Frequently called while holding sb_lock.
1257 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1258 * or -ENOMEM if memory allocation failed.
1259 */
0ee5dc67 1260int get_anon_bdev(dev_t *p)
1da177e4
LT
1261{
1262 int dev;
5a66847e
MW
1263
1264 /*
1265 * Many userspace utilities consider an FSID of 0 invalid.
1266 * Always return at least 1 from get_anon_bdev.
1267 */
1268 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1269 GFP_ATOMIC);
1270 if (dev == -ENOSPC)
1271 dev = -EMFILE;
1272 if (dev < 0)
1273 return dev;
1274
1275 *p = MKDEV(0, dev);
1da177e4
LT
1276 return 0;
1277}
0ee5dc67 1278EXPORT_SYMBOL(get_anon_bdev);
1da177e4 1279
0ee5dc67 1280void free_anon_bdev(dev_t dev)
1da177e4 1281{
5a66847e 1282 ida_free(&unnamed_dev_ida, MINOR(dev));
1da177e4 1283}
0ee5dc67
AV
1284EXPORT_SYMBOL(free_anon_bdev);
1285
1286int set_anon_super(struct super_block *s, void *data)
1287{
df0ce26c 1288 return get_anon_bdev(&s->s_dev);
0ee5dc67 1289}
0ee5dc67
AV
1290EXPORT_SYMBOL(set_anon_super);
1291
1292void kill_anon_super(struct super_block *sb)
1293{
1294 dev_t dev = sb->s_dev;
1295 generic_shutdown_super(sb);
dc3216b1 1296 kill_super_notify(sb);
0ee5dc67
AV
1297 free_anon_bdev(dev);
1298}
1da177e4
LT
1299EXPORT_SYMBOL(kill_anon_super);
1300
1da177e4
LT
1301void kill_litter_super(struct super_block *sb)
1302{
1303 if (sb->s_root)
1304 d_genocide(sb->s_root);
1305 kill_anon_super(sb);
1306}
1da177e4
LT
1307EXPORT_SYMBOL(kill_litter_super);
1308
cb50b348
AV
1309int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1310{
1311 return set_anon_super(sb, NULL);
1312}
1313EXPORT_SYMBOL(set_anon_super_fc);
1314
1315static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1316{
1317 return sb->s_fs_info == fc->s_fs_info;
1318}
1319
1320static int test_single_super(struct super_block *s, struct fs_context *fc)
1321{
1322 return 1;
1323}
1324
e062abae 1325static int vfs_get_super(struct fs_context *fc,
cda2ed05
CH
1326 int (*test)(struct super_block *, struct fs_context *),
1327 int (*fill_super)(struct super_block *sb,
1328 struct fs_context *fc))
cb50b348 1329{
cb50b348 1330 struct super_block *sb;
43ce4c1f 1331 int err;
cb50b348 1332
cb50b348
AV
1333 sb = sget_fc(fc, test, set_anon_super_fc);
1334 if (IS_ERR(sb))
1335 return PTR_ERR(sb);
1336
1337 if (!sb->s_root) {
43ce4c1f
DH
1338 err = fill_super(sb, fc);
1339 if (err)
1340 goto error;
cb50b348
AV
1341
1342 sb->s_flags |= SB_ACTIVE;
1343 }
1344
e062abae 1345 fc->root = dget(sb->s_root);
cb50b348 1346 return 0;
43ce4c1f
DH
1347
1348error:
1349 deactivate_locked_super(sb);
1350 return err;
cb50b348 1351}
cb50b348 1352
2ac295d4
AV
1353int get_tree_nodev(struct fs_context *fc,
1354 int (*fill_super)(struct super_block *sb,
1355 struct fs_context *fc))
1356{
e062abae 1357 return vfs_get_super(fc, NULL, fill_super);
2ac295d4
AV
1358}
1359EXPORT_SYMBOL(get_tree_nodev);
1360
c23a0bba
AV
1361int get_tree_single(struct fs_context *fc,
1362 int (*fill_super)(struct super_block *sb,
1363 struct fs_context *fc))
1364{
e062abae 1365 return vfs_get_super(fc, test_single_super, fill_super);
c23a0bba
AV
1366}
1367EXPORT_SYMBOL(get_tree_single);
1368
533770cc
AV
1369int get_tree_keyed(struct fs_context *fc,
1370 int (*fill_super)(struct super_block *sb,
1371 struct fs_context *fc),
1372 void *key)
1373{
1374 fc->s_fs_info = key;
e062abae 1375 return vfs_get_super(fc, test_keyed_super, fill_super);
533770cc
AV
1376}
1377EXPORT_SYMBOL(get_tree_keyed);
1378
69881be3
CB
1379static int set_bdev_super(struct super_block *s, void *data)
1380{
1381 s->s_dev = *(dev_t *)data;
1382 return 0;
1383}
1384
1385static int super_s_dev_set(struct super_block *s, struct fs_context *fc)
1386{
1387 return set_bdev_super(s, fc->sget_key);
1388}
1389
1390static int super_s_dev_test(struct super_block *s, struct fs_context *fc)
1391{
1392 return !(s->s_iflags & SB_I_RETIRED) &&
1393 s->s_dev == *(dev_t *)fc->sget_key;
1394}
1395
1396/**
1397 * sget_dev - Find or create a superblock by device number
1398 * @fc: Filesystem context.
1399 * @dev: device number
1400 *
1401 * Find or create a superblock using the provided device number that
1402 * will be stored in fc->sget_key.
1403 *
1404 * If an extant superblock is matched, then that will be returned with
1405 * an elevated reference count that the caller must transfer or discard.
1406 *
1407 * If no match is made, a new superblock will be allocated and basic
1408 * initialisation will be performed (s_type, s_fs_info, s_id, s_dev will
1409 * be set). The superblock will be published and it will be returned in
1410 * a partially constructed state with SB_BORN and SB_ACTIVE as yet
1411 * unset.
1412 *
1413 * Return: an existing or newly created superblock on success, an error
1414 * pointer on failure.
1415 */
1416struct super_block *sget_dev(struct fs_context *fc, dev_t dev)
1417{
1418 fc->sget_key = &dev;
1419 return sget_fc(fc, super_s_dev_test, super_s_dev_set);
1420}
1421EXPORT_SYMBOL(sget_dev);
1422
9361401e 1423#ifdef CONFIG_BLOCK
9c09a7cf 1424/*
fd146410
JK
1425 * Lock the superblock that is holder of the bdev. Returns the superblock
1426 * pointer if we successfully locked the superblock and it is alive. Otherwise
1427 * we return NULL and just unlock bdev->bd_holder_lock.
9c09a7cf 1428 *
fd146410 1429 * The function must be called with bdev->bd_holder_lock and releases it.
9c09a7cf 1430 */
fd146410
JK
1431static struct super_block *bdev_super_lock_shared(struct block_device *bdev)
1432 __releases(&bdev->bd_holder_lock)
87efb390 1433{
fd146410
JK
1434 struct super_block *sb = bdev->bd_holder;
1435 bool born;
1436
1437 lockdep_assert_held(&bdev->bd_holder_lock);
1438 lockdep_assert_not_held(&sb->s_umount);
3b224e1d 1439 lockdep_assert_not_held(&bdev->bd_disk->open_mutex);
fd146410
JK
1440
1441 /* Make sure sb doesn't go away from under us */
1442 spin_lock(&sb_lock);
1443 sb->s_count++;
1444 spin_unlock(&sb_lock);
1445 mutex_unlock(&bdev->bd_holder_lock);
87efb390 1446
fd146410 1447 born = super_lock_shared(sb);
5e874914 1448 if (!born || !sb->s_root || !(sb->s_flags & SB_ACTIVE)) {
0ed33598 1449 super_unlock_shared(sb);
fd146410
JK
1450 put_super(sb);
1451 return NULL;
9c09a7cf 1452 }
fd146410
JK
1453 /*
1454 * The superblock is active and we hold s_umount, we can drop our
1455 * temporary reference now.
1456 */
1457 put_super(sb);
1458 return sb;
9c09a7cf
CH
1459}
1460
d8530de5 1461static void fs_bdev_mark_dead(struct block_device *bdev, bool surprise)
87efb390 1462{
fd146410 1463 struct super_block *sb;
9c09a7cf 1464
fd146410
JK
1465 sb = bdev_super_lock_shared(bdev);
1466 if (!sb)
87efb390
CH
1467 return;
1468
d8530de5
CH
1469 if (!surprise)
1470 sync_filesystem(sb);
1471 shrink_dcache_sb(sb);
e127b9bc 1472 invalidate_inodes(sb);
87efb390
CH
1473 if (sb->s_op->shutdown)
1474 sb->s_op->shutdown(sb);
9c09a7cf 1475
0ed33598 1476 super_unlock_shared(sb);
87efb390
CH
1477}
1478
2142b88c
CH
1479static void fs_bdev_sync(struct block_device *bdev)
1480{
fd146410 1481 struct super_block *sb;
2142b88c 1482
fd146410
JK
1483 sb = bdev_super_lock_shared(bdev);
1484 if (!sb)
2142b88c
CH
1485 return;
1486 sync_filesystem(sb);
0ed33598 1487 super_unlock_shared(sb);
2142b88c
CH
1488}
1489
7ecd0b6f 1490const struct blk_holder_ops fs_holder_ops = {
d8530de5 1491 .mark_dead = fs_bdev_mark_dead,
2142b88c 1492 .sync = fs_bdev_sync,
87efb390 1493};
7ecd0b6f 1494EXPORT_SYMBOL_GPL(fs_holder_ops);
fe62c3a4 1495
cf6da236 1496int setup_bdev_super(struct super_block *sb, int sb_flags,
aca740ce
JK
1497 struct fs_context *fc)
1498{
1499 blk_mode_t mode = sb_open_mode(sb_flags);
f4a48bc3 1500 struct bdev_handle *bdev_handle;
aca740ce
JK
1501 struct block_device *bdev;
1502
f4a48bc3
JK
1503 bdev_handle = bdev_open_by_dev(sb->s_dev, mode, sb, &fs_holder_ops);
1504 if (IS_ERR(bdev_handle)) {
aca740ce
JK
1505 if (fc)
1506 errorf(fc, "%s: Can't open blockdev", fc->source);
f4a48bc3 1507 return PTR_ERR(bdev_handle);
aca740ce 1508 }
f4a48bc3 1509 bdev = bdev_handle->bdev;
aca740ce
JK
1510
1511 /*
1512 * This really should be in blkdev_get_by_dev, but right now can't due
1513 * to legacy issues that require us to allow opening a block device node
1514 * writable from userspace even for a read-only block device.
1515 */
1516 if ((mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
f4a48bc3 1517 bdev_release(bdev_handle);
aca740ce
JK
1518 return -EACCES;
1519 }
1520
1521 /*
1522 * Until SB_BORN flag is set, there can be no active superblock
1523 * references and thus no filesystem freezing. get_active_super() will
1524 * just loop waiting for SB_BORN so even freeze_bdev() cannot proceed.
1525 *
1526 * It is enough to check bdev was not frozen before we set s_bdev.
1527 */
1528 mutex_lock(&bdev->bd_fsfreeze_mutex);
1529 if (bdev->bd_fsfreeze_count > 0) {
1530 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1531 if (fc)
1532 warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
f4a48bc3 1533 bdev_release(bdev_handle);
aca740ce
JK
1534 return -EBUSY;
1535 }
1536 spin_lock(&sb_lock);
f4a48bc3 1537 sb->s_bdev_handle = bdev_handle;
aca740ce
JK
1538 sb->s_bdev = bdev;
1539 sb->s_bdi = bdi_get(bdev->bd_disk->bdi);
1540 if (bdev_stable_writes(bdev))
1541 sb->s_iflags |= SB_I_STABLE_WRITES;
1542 spin_unlock(&sb_lock);
1543 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1544
1545 snprintf(sb->s_id, sizeof(sb->s_id), "%pg", bdev);
1720f5dd 1546 shrinker_debugfs_rename(sb->s_shrink, "sb-%s:%s", sb->s_type->name,
aca740ce
JK
1547 sb->s_id);
1548 sb_set_blocksize(sb, block_size(bdev));
1549 return 0;
fe62c3a4 1550}
cf6da236 1551EXPORT_SYMBOL_GPL(setup_bdev_super);
fe62c3a4
DH
1552
1553/**
1554 * get_tree_bdev - Get a superblock based on a single block device
1555 * @fc: The filesystem context holding the parameters
1556 * @fill_super: Helper to initialise a new superblock
1557 */
1558int get_tree_bdev(struct fs_context *fc,
1559 int (*fill_super)(struct super_block *,
1560 struct fs_context *))
1561{
fe62c3a4 1562 struct super_block *s;
fe62c3a4 1563 int error = 0;
aca740ce 1564 dev_t dev;
fe62c3a4 1565
fe62c3a4
DH
1566 if (!fc->source)
1567 return invalf(fc, "No source specified");
1568
aca740ce
JK
1569 error = lookup_bdev(fc->source, &dev);
1570 if (error) {
1571 errorf(fc, "%s: Can't lookup blockdev", fc->source);
1572 return error;
fe62c3a4
DH
1573 }
1574
1575 fc->sb_flags |= SB_NOSEC;
69881be3 1576 s = sget_dev(fc, dev);
aca740ce 1577 if (IS_ERR(s))
fe62c3a4
DH
1578 return PTR_ERR(s);
1579
1580 if (s->s_root) {
1581 /* Don't summarily change the RO/RW state. */
1582 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
aca740ce 1583 warnf(fc, "%pg: Can't mount, would change RO state", s->s_bdev);
fe62c3a4 1584 deactivate_locked_super(s);
fe62c3a4
DH
1585 return -EBUSY;
1586 }
aca740ce 1587 } else {
fe62c3a4 1588 /*
aca740ce
JK
1589 * We drop s_umount here because we need to open the bdev and
1590 * bdev->open_mutex ranks above s_umount (blkdev_put() ->
560e20e4 1591 * bdev_mark_dead()). It is safe because we have active sb
aca740ce 1592 * reference and SB_BORN is not set yet.
fe62c3a4 1593 */
0ed33598 1594 super_unlock_excl(s);
aca740ce 1595 error = setup_bdev_super(s, fc->sb_flags, fc);
5e874914 1596 __super_lock_excl(s);
aca740ce
JK
1597 if (!error)
1598 error = fill_super(s, fc);
fe62c3a4
DH
1599 if (error) {
1600 deactivate_locked_super(s);
1601 return error;
1602 }
fe62c3a4 1603 s->s_flags |= SB_ACTIVE;
fe62c3a4
DH
1604 }
1605
1606 BUG_ON(fc->root);
1607 fc->root = dget(s->s_root);
1608 return 0;
1609}
1610EXPORT_SYMBOL(get_tree_bdev);
1611
1da177e4
LT
1612static int test_bdev_super(struct super_block *s, void *data)
1613{
aca740ce 1614 return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data;
1da177e4
LT
1615}
1616
152a0836 1617struct dentry *mount_bdev(struct file_system_type *fs_type,
1da177e4 1618 int flags, const char *dev_name, void *data,
152a0836 1619 int (*fill_super)(struct super_block *, void *, int))
1da177e4 1620{
1da177e4 1621 struct super_block *s;
aca740ce
JK
1622 int error;
1623 dev_t dev;
1da177e4 1624
aca740ce
JK
1625 error = lookup_bdev(dev_name, &dev);
1626 if (error)
1627 return ERR_PTR(error);
1da177e4 1628
aca740ce
JK
1629 flags |= SB_NOSEC;
1630 s = sget(fs_type, test_bdev_super, set_bdev_super, flags, &dev);
1da177e4 1631 if (IS_ERR(s))
aca740ce 1632 return ERR_CAST(s);
1da177e4
LT
1633
1634 if (s->s_root) {
e462ec50 1635 if ((flags ^ s->s_flags) & SB_RDONLY) {
74dbbdd7 1636 deactivate_locked_super(s);
aca740ce 1637 return ERR_PTR(-EBUSY);
1da177e4 1638 }
aca740ce 1639 } else {
4f331f01 1640 /*
aca740ce
JK
1641 * We drop s_umount here because we need to open the bdev and
1642 * bdev->open_mutex ranks above s_umount (blkdev_put() ->
560e20e4 1643 * bdev_mark_dead()). It is safe because we have active sb
aca740ce 1644 * reference and SB_BORN is not set yet.
4f331f01 1645 */
0ed33598 1646 super_unlock_excl(s);
aca740ce 1647 error = setup_bdev_super(s, flags, NULL);
5e874914 1648 __super_lock_excl(s);
aca740ce
JK
1649 if (!error)
1650 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1da177e4 1651 if (error) {
74dbbdd7 1652 deactivate_locked_super(s);
aca740ce 1653 return ERR_PTR(error);
fa675765 1654 }
454e2398 1655
e462ec50 1656 s->s_flags |= SB_ACTIVE;
1da177e4
LT
1657 }
1658
152a0836 1659 return dget(s->s_root);
152a0836
AV
1660}
1661EXPORT_SYMBOL(mount_bdev);
1662
1da177e4
LT
1663void kill_block_super(struct super_block *sb)
1664{
1665 struct block_device *bdev = sb->s_bdev;
1666
1da177e4 1667 generic_shutdown_super(sb);
aca740ce
JK
1668 if (bdev) {
1669 sync_blockdev(bdev);
f4a48bc3 1670 bdev_release(sb->s_bdev_handle);
aca740ce 1671 }
1da177e4
LT
1672}
1673
1674EXPORT_SYMBOL(kill_block_super);
9361401e 1675#endif
1da177e4 1676
3c26ff6e 1677struct dentry *mount_nodev(struct file_system_type *fs_type,
1da177e4 1678 int flags, void *data,
3c26ff6e 1679 int (*fill_super)(struct super_block *, void *, int))
1da177e4
LT
1680{
1681 int error;
9249e17f 1682 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1da177e4
LT
1683
1684 if (IS_ERR(s))
3c26ff6e 1685 return ERR_CAST(s);
1da177e4 1686
e462ec50 1687 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1da177e4 1688 if (error) {
74dbbdd7 1689 deactivate_locked_super(s);
3c26ff6e 1690 return ERR_PTR(error);
1da177e4 1691 }
e462ec50 1692 s->s_flags |= SB_ACTIVE;
3c26ff6e 1693 return dget(s->s_root);
1da177e4 1694}
3c26ff6e
AV
1695EXPORT_SYMBOL(mount_nodev);
1696
a6097180
N
1697int reconfigure_single(struct super_block *s,
1698 int flags, void *data)
8d0347f6
DH
1699{
1700 struct fs_context *fc;
1701 int ret;
1702
1703 /* The caller really need to be passing fc down into mount_single(),
1704 * then a chunk of this can be removed. [Bollocks -- AV]
1705 * Better yet, reconfiguration shouldn't happen, but rather the second
1706 * mount should be rejected if the parameters are not compatible.
1707 */
1708 fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1709 if (IS_ERR(fc))
1710 return PTR_ERR(fc);
1711
1712 ret = parse_monolithic_mount_data(fc, data);
1713 if (ret < 0)
1714 goto out;
1715
1716 ret = reconfigure_super(fc);
1717out:
1718 put_fs_context(fc);
1719 return ret;
1720}
1721
1da177e4
LT
1722static int compare_single(struct super_block *s, void *p)
1723{
1724 return 1;
1725}
1726
fc14f2fe 1727struct dentry *mount_single(struct file_system_type *fs_type,
1da177e4 1728 int flags, void *data,
fc14f2fe 1729 int (*fill_super)(struct super_block *, void *, int))
1da177e4
LT
1730{
1731 struct super_block *s;
1732 int error;
1733
9249e17f 1734 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1da177e4 1735 if (IS_ERR(s))
fc14f2fe 1736 return ERR_CAST(s);
1da177e4 1737 if (!s->s_root) {
e462ec50 1738 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
8d0347f6
DH
1739 if (!error)
1740 s->s_flags |= SB_ACTIVE;
9329d1be 1741 } else {
8d0347f6
DH
1742 error = reconfigure_single(s, flags, data);
1743 }
1744 if (unlikely(error)) {
1745 deactivate_locked_super(s);
1746 return ERR_PTR(error);
1da177e4 1747 }
fc14f2fe
AV
1748 return dget(s->s_root);
1749}
1750EXPORT_SYMBOL(mount_single);
1751
9bc61ab1
DH
1752/**
1753 * vfs_get_tree - Get the mountable root
1754 * @fc: The superblock configuration context.
1755 *
1756 * The filesystem is invoked to get or create a superblock which can then later
1757 * be used for mounting. The filesystem places a pointer to the root to be
1758 * used for mounting in @fc->root.
1759 */
1760int vfs_get_tree(struct fs_context *fc)
1da177e4 1761{
9d412a43 1762 struct super_block *sb;
9bc61ab1 1763 int error;
8089352a 1764
f3a09c92
AV
1765 if (fc->root)
1766 return -EBUSY;
1767
1768 /* Get the mountable root in fc->root, with a ref on the root and a ref
1769 * on the superblock.
1770 */
1771 error = fc->ops->get_tree(fc);
9bc61ab1
DH
1772 if (error < 0)
1773 return error;
1da177e4 1774
f3a09c92
AV
1775 if (!fc->root) {
1776 pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1777 fc->fs_type->name);
1778 /* We don't know what the locking state of the superblock is -
1779 * if there is a superblock.
1780 */
1781 BUG();
1782 }
1783
9bc61ab1 1784 sb = fc->root->d_sb;
9d412a43 1785 WARN_ON(!sb->s_bdi);
79f546a6
DC
1786
1787 /*
5e874914
CB
1788 * super_wake() contains a memory barrier which also care of
1789 * ordering for super_cache_count(). We place it before setting
1790 * SB_BORN as the data dependency between the two functions is
1791 * the superblock structure contents that we just set up, not
1792 * the SB_BORN flag.
79f546a6 1793 */
5e874914 1794 super_wake(sb, SB_BORN);
454e2398 1795
9bc61ab1 1796 error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
c9ce29ed
AV
1797 if (unlikely(error)) {
1798 fc_drop_locked(fc);
1799 return error;
a10d7c22
AV
1800 }
1801
42cb56ae
JL
1802 /*
1803 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1804 * but s_maxbytes was an unsigned long long for many releases. Throw
1805 * this warning for a little while to try and catch filesystems that
4358b567 1806 * violate this rule.
42cb56ae 1807 */
9d412a43 1808 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
9bc61ab1 1809 "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
42cb56ae 1810
9bc61ab1 1811 return 0;
1da177e4 1812}
9bc61ab1 1813EXPORT_SYMBOL(vfs_get_tree);
1da177e4 1814
fca39346
JK
1815/*
1816 * Setup private BDI for given superblock. It gets automatically cleaned up
1817 * in generic_shutdown_super().
1818 */
1819int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1820{
1821 struct backing_dev_info *bdi;
1822 int err;
1823 va_list args;
1824
aef33c2f 1825 bdi = bdi_alloc(NUMA_NO_NODE);
fca39346
JK
1826 if (!bdi)
1827 return -ENOMEM;
1828
fca39346 1829 va_start(args, fmt);
7c4cc300 1830 err = bdi_register_va(bdi, fmt, args);
fca39346
JK
1831 va_end(args);
1832 if (err) {
1833 bdi_put(bdi);
1834 return err;
1835 }
1836 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1837 sb->s_bdi = bdi;
0b3ea092 1838 sb->s_iflags |= SB_I_PERSB_BDI;
fca39346
JK
1839
1840 return 0;
1841}
1842EXPORT_SYMBOL(super_setup_bdi_name);
1843
1844/*
1845 * Setup private BDI for given superblock. I gets automatically cleaned up
1846 * in generic_shutdown_super().
1847 */
1848int super_setup_bdi(struct super_block *sb)
1849{
1850 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1851
1852 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1853 atomic_long_inc_return(&bdi_seq));
1854}
1855EXPORT_SYMBOL(super_setup_bdi);
1856
5accdf82
JK
1857/**
1858 * sb_wait_write - wait until all writers to given file system finish
1859 * @sb: the super for which we wait
1860 * @level: type of writers we wait for (normal vs page fault)
1861 *
1862 * This function waits until there are no writers of given type to given file
8129ed29 1863 * system.
5accdf82
JK
1864 */
1865static void sb_wait_write(struct super_block *sb, int level)
1866{
8129ed29 1867 percpu_down_write(sb->s_writers.rw_sem + level-1);
8129ed29 1868}
5accdf82 1869
f1a96220
ON
1870/*
1871 * We are going to return to userspace and forget about these locks, the
1872 * ownership goes to the caller of thaw_super() which does unlock().
1873 */
1874static void lockdep_sb_freeze_release(struct super_block *sb)
1875{
1876 int level;
1877
1878 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1879 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1880}
1881
1882/*
1883 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1884 */
1885static void lockdep_sb_freeze_acquire(struct super_block *sb)
8129ed29
ON
1886{
1887 int level;
5accdf82 1888
8129ed29
ON
1889 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1890 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
f1a96220
ON
1891}
1892
2719c716 1893static void sb_freeze_unlock(struct super_block *sb, int level)
f1a96220 1894{
2719c716 1895 for (level--; level >= 0; level--)
8129ed29 1896 percpu_up_write(sb->s_writers.rw_sem + level);
5accdf82
JK
1897}
1898
59ba4fdd
DW
1899static int wait_for_partially_frozen(struct super_block *sb)
1900{
1901 int ret = 0;
1902
1903 do {
1904 unsigned short old = sb->s_writers.frozen;
1905
1906 up_write(&sb->s_umount);
1907 ret = wait_var_event_killable(&sb->s_writers.frozen,
1908 sb->s_writers.frozen != old);
1909 down_write(&sb->s_umount);
1910 } while (ret == 0 &&
1911 sb->s_writers.frozen != SB_UNFROZEN &&
1912 sb->s_writers.frozen != SB_FREEZE_COMPLETE);
1913
1914 return ret;
1915}
1916
18e9e510 1917/**
7000d3c4
RD
1918 * freeze_super - lock the filesystem and force it into a consistent state
1919 * @sb: the super to lock
880b9577 1920 * @who: context that wants to freeze
18e9e510
JB
1921 *
1922 * Syncs the super to make sure the filesystem is consistent and calls the fs's
880b9577 1923 * freeze_fs. Subsequent calls to this without first thawing the fs may return
18e9e510 1924 * -EBUSY.
5accdf82 1925 *
880b9577
DW
1926 * @who should be:
1927 * * %FREEZE_HOLDER_USERSPACE if userspace wants to freeze the fs;
1928 * * %FREEZE_HOLDER_KERNEL if the kernel wants to freeze the fs.
1929 *
1930 * The @who argument distinguishes between the kernel and userspace trying to
1931 * freeze the filesystem. Although there cannot be multiple kernel freezes or
1932 * multiple userspace freezes in effect at any given time, the kernel and
1933 * userspace can both hold a filesystem frozen. The filesystem remains frozen
1934 * until there are no kernel or userspace freezes in effect.
1935 *
5accdf82
JK
1936 * During this function, sb->s_writers.frozen goes through these values:
1937 *
1938 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1939 *
1940 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1941 * writes should be blocked, though page faults are still allowed. We wait for
1942 * all writes to complete and then proceed to the next stage.
1943 *
1944 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1945 * but internal fs threads can still modify the filesystem (although they
1946 * should not dirty new pages or inodes), writeback can run etc. After waiting
1947 * for all running page faults we sync the filesystem which will clean all
1948 * dirty pages and inodes (no new dirty pages or inodes can be created when
1949 * sync is running).
1950 *
1951 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1952 * modification are blocked (e.g. XFS preallocation truncation on inode
1953 * reclaim). This is usually implemented by blocking new transactions for
1954 * filesystems that have them and need this additional guard. After all
1955 * internal writers are finished we call ->freeze_fs() to finish filesystem
1956 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1957 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1958 *
1959 * sb->s_writers.frozen is protected by sb->s_umount.
18e9e510 1960 */
880b9577 1961int freeze_super(struct super_block *sb, enum freeze_holder who)
18e9e510
JB
1962{
1963 int ret;
1964
1965 atomic_inc(&sb->s_active);
051178c3
CB
1966 if (!super_lock_excl(sb))
1967 WARN(1, "Dying superblock while freezing!");
1968
59ba4fdd 1969retry:
880b9577
DW
1970 if (sb->s_writers.frozen == SB_FREEZE_COMPLETE) {
1971 if (sb->s_writers.freeze_holders & who) {
1972 deactivate_locked_super(sb);
1973 return -EBUSY;
1974 }
1975
1976 WARN_ON(sb->s_writers.freeze_holders == 0);
1977
1978 /*
1979 * Someone else already holds this type of freeze; share the
1980 * freeze and assign the active ref to the freeze.
1981 */
1982 sb->s_writers.freeze_holders |= who;
3fb5a656 1983 super_unlock_excl(sb);
880b9577
DW
1984 return 0;
1985 }
1986
5accdf82 1987 if (sb->s_writers.frozen != SB_UNFROZEN) {
59ba4fdd
DW
1988 ret = wait_for_partially_frozen(sb);
1989 if (ret) {
1990 deactivate_locked_super(sb);
1991 return ret;
1992 }
1993
1994 goto retry;
18e9e510
JB
1995 }
1996
e462ec50 1997 if (!(sb->s_flags & SB_BORN)) {
0ed33598 1998 super_unlock_excl(sb);
dabe0dc1
AV
1999 return 0; /* sic - it's "nothing to do" */
2000 }
2001
bc98a42c 2002 if (sb_rdonly(sb)) {
5accdf82 2003 /* Nothing to do really... */
880b9577 2004 sb->s_writers.freeze_holders |= who;
5accdf82 2005 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
59ba4fdd 2006 wake_up_var(&sb->s_writers.frozen);
0ed33598 2007 super_unlock_excl(sb);
18e9e510
JB
2008 return 0;
2009 }
2010
5accdf82 2011 sb->s_writers.frozen = SB_FREEZE_WRITE;
5accdf82 2012 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
0ed33598 2013 super_unlock_excl(sb);
5accdf82 2014 sb_wait_write(sb, SB_FREEZE_WRITE);
051178c3
CB
2015 if (!super_lock_excl(sb))
2016 WARN(1, "Dying superblock while freezing!");
5accdf82
JK
2017
2018 /* Now we go and block page faults... */
5accdf82 2019 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
5accdf82
JK
2020 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
2021
2022 /* All writers are done so after syncing there won't be dirty data */
2719c716
DW
2023 ret = sync_filesystem(sb);
2024 if (ret) {
2025 sb->s_writers.frozen = SB_UNFROZEN;
2026 sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
59ba4fdd 2027 wake_up_var(&sb->s_writers.frozen);
2719c716
DW
2028 deactivate_locked_super(sb);
2029 return ret;
2030 }
18e9e510 2031
5accdf82
JK
2032 /* Now wait for internal filesystem counter */
2033 sb->s_writers.frozen = SB_FREEZE_FS;
5accdf82 2034 sb_wait_write(sb, SB_FREEZE_FS);
18e9e510 2035
18e9e510
JB
2036 if (sb->s_op->freeze_fs) {
2037 ret = sb->s_op->freeze_fs(sb);
2038 if (ret) {
2039 printk(KERN_ERR
2040 "VFS:Filesystem freeze failed\n");
5accdf82 2041 sb->s_writers.frozen = SB_UNFROZEN;
2719c716 2042 sb_freeze_unlock(sb, SB_FREEZE_FS);
59ba4fdd 2043 wake_up_var(&sb->s_writers.frozen);
18e9e510
JB
2044 deactivate_locked_super(sb);
2045 return ret;
2046 }
2047 }
5accdf82 2048 /*
89f39af1
ON
2049 * For debugging purposes so that fs can warn if it sees write activity
2050 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
5accdf82 2051 */
880b9577 2052 sb->s_writers.freeze_holders |= who;
5accdf82 2053 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
59ba4fdd 2054 wake_up_var(&sb->s_writers.frozen);
f1a96220 2055 lockdep_sb_freeze_release(sb);
0ed33598 2056 super_unlock_excl(sb);
18e9e510
JB
2057 return 0;
2058}
2059EXPORT_SYMBOL(freeze_super);
2060
880b9577
DW
2061/*
2062 * Undoes the effect of a freeze_super_locked call. If the filesystem is
2063 * frozen both by userspace and the kernel, a thaw call from either source
2064 * removes that state without releasing the other state or unlocking the
2065 * filesystem.
2066 */
2067static int thaw_super_locked(struct super_block *sb, enum freeze_holder who)
18e9e510
JB
2068{
2069 int error;
2070
880b9577
DW
2071 if (sb->s_writers.frozen == SB_FREEZE_COMPLETE) {
2072 if (!(sb->s_writers.freeze_holders & who)) {
3fb5a656 2073 super_unlock_excl(sb);
880b9577
DW
2074 return -EINVAL;
2075 }
2076
2077 /*
2078 * Freeze is shared with someone else. Release our hold and
2079 * drop the active ref that freeze_super assigned to the
2080 * freezer.
2081 */
2082 if (sb->s_writers.freeze_holders & ~who) {
2083 sb->s_writers.freeze_holders &= ~who;
2084 deactivate_locked_super(sb);
2085 return 0;
2086 }
2087 } else {
0ed33598 2088 super_unlock_excl(sb);
18e9e510
JB
2089 return -EINVAL;
2090 }
2091
bc98a42c 2092 if (sb_rdonly(sb)) {
880b9577 2093 sb->s_writers.freeze_holders &= ~who;
8129ed29 2094 sb->s_writers.frozen = SB_UNFROZEN;
59ba4fdd 2095 wake_up_var(&sb->s_writers.frozen);
18e9e510 2096 goto out;
8129ed29 2097 }
18e9e510 2098
f1a96220
ON
2099 lockdep_sb_freeze_acquire(sb);
2100
18e9e510
JB
2101 if (sb->s_op->unfreeze_fs) {
2102 error = sb->s_op->unfreeze_fs(sb);
2103 if (error) {
3fb5a656 2104 printk(KERN_ERR "VFS:Filesystem thaw failed\n");
f1a96220 2105 lockdep_sb_freeze_release(sb);
0ed33598 2106 super_unlock_excl(sb);
18e9e510
JB
2107 return error;
2108 }
2109 }
2110
880b9577 2111 sb->s_writers.freeze_holders &= ~who;
5accdf82 2112 sb->s_writers.frozen = SB_UNFROZEN;
59ba4fdd 2113 wake_up_var(&sb->s_writers.frozen);
2719c716 2114 sb_freeze_unlock(sb, SB_FREEZE_FS);
8129ed29 2115out:
18e9e510 2116 deactivate_locked_super(sb);
18e9e510
JB
2117 return 0;
2118}
08fdc8a0 2119
961f3c89
MCC
2120/**
2121 * thaw_super -- unlock filesystem
2122 * @sb: the super to thaw
880b9577
DW
2123 * @who: context that wants to freeze
2124 *
2125 * Unlocks the filesystem and marks it writeable again after freeze_super()
2126 * if there are no remaining freezes on the filesystem.
961f3c89 2127 *
880b9577
DW
2128 * @who should be:
2129 * * %FREEZE_HOLDER_USERSPACE if userspace wants to thaw the fs;
2130 * * %FREEZE_HOLDER_KERNEL if the kernel wants to thaw the fs.
961f3c89 2131 */
880b9577 2132int thaw_super(struct super_block *sb, enum freeze_holder who)
08fdc8a0 2133{
051178c3
CB
2134 if (!super_lock_excl(sb))
2135 WARN(1, "Dying superblock while thawing!");
880b9577 2136 return thaw_super_locked(sb, who);
08fdc8a0 2137}
18e9e510 2138EXPORT_SYMBOL(thaw_super);
439bc39b
CH
2139
2140/*
2141 * Create workqueue for deferred direct IO completions. We allocate the
2142 * workqueue when it's first needed. This avoids creating workqueue for
2143 * filesystems that don't need it and also allows us to create the workqueue
2144 * late enough so the we can include s_id in the name of the workqueue.
2145 */
2146int sb_init_dio_done_wq(struct super_block *sb)
2147{
2148 struct workqueue_struct *old;
2149 struct workqueue_struct *wq = alloc_workqueue("dio/%s",
2150 WQ_MEM_RECLAIM, 0,
2151 sb->s_id);
2152 if (!wq)
2153 return -ENOMEM;
2154 /*
2155 * This has to be atomic as more DIOs can race to create the workqueue
2156 */
2157 old = cmpxchg(&sb->s_dio_done_wq, NULL, wq);
2158 /* Someone created workqueue before us? Free ours... */
2159 if (old)
2160 destroy_workqueue(wq);
2161 return 0;
2162}
389a4a4a 2163EXPORT_SYMBOL_GPL(sb_init_dio_done_wq);