]>
Commit | Line | Data |
---|---|---|
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 ([email protected]), April 14, 1996. |
1da177e4 LT |
20 | * Added devfs support: Richard Gooch <[email protected]>, 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 | 42 | static int thaw_super_locked(struct super_block *sb, enum freeze_holder who); |
1da177e4 | 43 | |
15d0f5ea AV |
44 | static LIST_HEAD(super_blocks); |
45 | static DEFINE_SPINLOCK(sb_lock); | |
1da177e4 | 46 | |
5accdf82 JK |
47 | static char *sb_writers_name[SB_FREEZE_LEVELS] = { |
48 | "sb_writers", | |
49 | "sb_pagefaults", | |
50 | "sb_internal", | |
51 | }; | |
52 | ||
5e874914 | 53 | static 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 | ||
61 | static 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 | 69 | static inline void __super_lock_excl(struct super_block *sb) |
0ed33598 | 70 | { |
5e874914 | 71 | __super_lock(sb, true); |
0ed33598 CB |
72 | } |
73 | ||
74 | static inline void super_unlock_excl(struct super_block *sb) | |
75 | { | |
76 | super_unlock(sb, true); | |
77 | } | |
78 | ||
79 | static inline void super_unlock_shared(struct super_block *sb) | |
80 | { | |
81 | super_unlock(sb, false); | |
82 | } | |
83 | ||
b30850c5 | 84 | static bool super_flags(const struct super_block *sb, unsigned int flags) |
5e874914 | 85 | { |
5e874914 CB |
86 | /* |
87 | * Pairs with smp_store_release() in super_wake() and ensures | |
b30850c5 | 88 | * that we see @flags after we're woken. |
5e874914 | 89 | */ |
b30850c5 | 90 | return smp_load_acquire(&sb->s_flags) & flags; |
5e874914 CB |
91 | } |
92 | ||
93 | /** | |
94 | * super_lock - wait for superblock to become ready and lock it | |
95 | * @sb: superblock to wait for | |
96 | * @excl: whether exclusive access is required | |
97 | * | |
98 | * If the superblock has neither passed through vfs_get_tree() or | |
99 | * generic_shutdown_super() yet wait for it to happen. Either superblock | |
100 | * creation will succeed and SB_BORN is set by vfs_get_tree() or we're | |
101 | * woken and we'll see SB_DYING. | |
102 | * | |
103 | * The caller must have acquired a temporary reference on @sb->s_count. | |
104 | * | |
f0cd9880 CB |
105 | * Return: The function returns true if SB_BORN was set and with |
106 | * s_umount held. The function returns false if SB_DYING was | |
107 | * set and without s_umount held. | |
5e874914 CB |
108 | */ |
109 | static __must_check bool super_lock(struct super_block *sb, bool excl) | |
110 | { | |
5e874914 CB |
111 | lockdep_assert_not_held(&sb->s_umount); |
112 | ||
b30850c5 CB |
113 | /* wait until the superblock is ready or dying */ |
114 | wait_var_event(&sb->s_flags, super_flags(sb, SB_BORN | SB_DYING)); | |
115 | ||
116 | /* Don't pointlessly acquire s_umount. */ | |
117 | if (super_flags(sb, SB_DYING)) | |
118 | return false; | |
119 | ||
5e874914 CB |
120 | __super_lock(sb, excl); |
121 | ||
122 | /* | |
123 | * Has gone through generic_shutdown_super() in the meantime. | |
124 | * @sb->s_root is NULL and @sb->s_active is 0. No one needs to | |
125 | * grab a reference to this. Tell them so. | |
126 | */ | |
f0cd9880 CB |
127 | if (sb->s_flags & SB_DYING) { |
128 | super_unlock(sb, excl); | |
5e874914 | 129 | return false; |
f0cd9880 | 130 | } |
5e874914 | 131 | |
b30850c5 CB |
132 | WARN_ON_ONCE(!(sb->s_flags & SB_BORN)); |
133 | return true; | |
5e874914 CB |
134 | } |
135 | ||
f0cd9880 | 136 | /* wait and try to acquire read-side of @sb->s_umount */ |
5e874914 CB |
137 | static inline bool super_lock_shared(struct super_block *sb) |
138 | { | |
139 | return super_lock(sb, false); | |
140 | } | |
141 | ||
f0cd9880 | 142 | /* wait and try to acquire write-side of @sb->s_umount */ |
5e874914 CB |
143 | static inline bool super_lock_excl(struct super_block *sb) |
144 | { | |
145 | return super_lock(sb, true); | |
146 | } | |
147 | ||
148 | /* wake waiters */ | |
2c18a63b | 149 | #define SUPER_WAKE_FLAGS (SB_BORN | SB_DYING | SB_DEAD) |
5e874914 CB |
150 | static void super_wake(struct super_block *sb, unsigned int flag) |
151 | { | |
152 | WARN_ON_ONCE((flag & ~SUPER_WAKE_FLAGS)); | |
153 | WARN_ON_ONCE(hweight32(flag & SUPER_WAKE_FLAGS) > 1); | |
154 | ||
155 | /* | |
156 | * Pairs with smp_load_acquire() in super_lock() to make sure | |
157 | * all initializations in the superblock are seen by the user | |
158 | * seeing SB_BORN sent. | |
159 | */ | |
160 | smp_store_release(&sb->s_flags, sb->s_flags | flag); | |
161 | /* | |
162 | * Pairs with the barrier in prepare_to_wait_event() to make sure | |
163 | * ___wait_var_event() either sees SB_BORN set or | |
164 | * waitqueue_active() check in wake_up_var() sees the waiter. | |
165 | */ | |
166 | smp_mb(); | |
167 | wake_up_var(&sb->s_flags); | |
168 | } | |
169 | ||
b0d40c92 DC |
170 | /* |
171 | * One thing we have to be careful of with a per-sb shrinker is that we don't | |
172 | * drop the last active reference to the superblock from within the shrinker. | |
173 | * If that happens we could trigger unregistering the shrinker from within the | |
8a0e8bb1 | 174 | * shrinker path and that leads to deadlock on the shrinker_mutex. Hence we |
b0d40c92 DC |
175 | * take a passive reference to the superblock to avoid this from occurring. |
176 | */ | |
0a234c6d DC |
177 | static unsigned long super_cache_scan(struct shrinker *shrink, |
178 | struct shrink_control *sc) | |
b0d40c92 DC |
179 | { |
180 | struct super_block *sb; | |
0a234c6d DC |
181 | long fs_objects = 0; |
182 | long total_objects; | |
183 | long freed = 0; | |
184 | long dentries; | |
185 | long inodes; | |
b0d40c92 | 186 | |
1720f5dd | 187 | sb = shrink->private_data; |
b0d40c92 DC |
188 | |
189 | /* | |
190 | * Deadlock avoidance. We may hold various FS locks, and we don't want | |
191 | * to recurse into the FS that called us in clear_inode() and friends.. | |
192 | */ | |
0a234c6d DC |
193 | if (!(sc->gfp_mask & __GFP_FS)) |
194 | return SHRINK_STOP; | |
b0d40c92 | 195 | |
d8ce82ef | 196 | if (!super_trylock_shared(sb)) |
0a234c6d | 197 | return SHRINK_STOP; |
b0d40c92 | 198 | |
d0407903 | 199 | if (sb->s_op->nr_cached_objects) |
4101b624 | 200 | fs_objects = sb->s_op->nr_cached_objects(sb, sc); |
0e1fdafd | 201 | |
503c358c VD |
202 | inodes = list_lru_shrink_count(&sb->s_inode_lru, sc); |
203 | dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc); | |
f6041567 | 204 | total_objects = dentries + inodes + fs_objects + 1; |
475d0db7 TH |
205 | if (!total_objects) |
206 | total_objects = 1; | |
0e1fdafd | 207 | |
0a234c6d | 208 | /* proportion the scan between the caches */ |
f6041567 | 209 | dentries = mult_frac(sc->nr_to_scan, dentries, total_objects); |
bc3b14cb | 210 | inodes = mult_frac(sc->nr_to_scan, inodes, total_objects); |
503c358c | 211 | fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects); |
b0d40c92 | 212 | |
0a234c6d DC |
213 | /* |
214 | * prune the dcache first as the icache is pinned by it, then | |
215 | * prune the icache, followed by the filesystem specific caches | |
49e7e7ff VD |
216 | * |
217 | * Ensure that we always scan at least one object - memcg kmem | |
218 | * accounting uses this to fully empty the caches. | |
0a234c6d | 219 | */ |
49e7e7ff | 220 | sc->nr_to_scan = dentries + 1; |
503c358c | 221 | freed = prune_dcache_sb(sb, sc); |
49e7e7ff | 222 | sc->nr_to_scan = inodes + 1; |
503c358c | 223 | freed += prune_icache_sb(sb, sc); |
0a234c6d DC |
224 | |
225 | if (fs_objects) { | |
49e7e7ff | 226 | sc->nr_to_scan = fs_objects + 1; |
4101b624 | 227 | freed += sb->s_op->free_cached_objects(sb, sc); |
b0d40c92 DC |
228 | } |
229 | ||
0ed33598 | 230 | super_unlock_shared(sb); |
0a234c6d DC |
231 | return freed; |
232 | } | |
233 | ||
234 | static unsigned long super_cache_count(struct shrinker *shrink, | |
235 | struct shrink_control *sc) | |
236 | { | |
237 | struct super_block *sb; | |
238 | long total_objects = 0; | |
239 | ||
1720f5dd | 240 | sb = shrink->private_data; |
0a234c6d | 241 | |
d23da150 | 242 | /* |
d8ce82ef CB |
243 | * We don't call super_trylock_shared() here as it is a scalability |
244 | * bottleneck, so we're exposed to partial setup state. The shrinker | |
245 | * rwsem does not protect filesystem operations backing | |
246 | * list_lru_shrink_count() or s_op->nr_cached_objects(). Counts can | |
247 | * change between super_cache_count and super_cache_scan, so we really | |
248 | * don't need locks here. | |
79f546a6 DC |
249 | * |
250 | * However, if we are currently mounting the superblock, the underlying | |
251 | * filesystem might be in a state of partial construction and hence it | |
d8ce82ef CB |
252 | * is dangerous to access it. super_trylock_shared() uses a SB_BORN check |
253 | * to avoid this situation, so do the same here. The memory barrier is | |
79f546a6 | 254 | * matched with the one in mount_fs() as we don't hold locks here. |
d23da150 | 255 | */ |
79f546a6 DC |
256 | if (!(sb->s_flags & SB_BORN)) |
257 | return 0; | |
258 | smp_rmb(); | |
259 | ||
0a234c6d | 260 | if (sb->s_op && sb->s_op->nr_cached_objects) |
4101b624 | 261 | total_objects = sb->s_op->nr_cached_objects(sb, sc); |
0a234c6d | 262 | |
503c358c VD |
263 | total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc); |
264 | total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc); | |
0a234c6d | 265 | |
9b996468 KT |
266 | if (!total_objects) |
267 | return SHRINK_EMPTY; | |
268 | ||
55f841ce | 269 | total_objects = vfs_pressure_ratio(total_objects); |
0e1fdafd | 270 | return total_objects; |
b0d40c92 DC |
271 | } |
272 | ||
853b39a7 ON |
273 | static void destroy_super_work(struct work_struct *work) |
274 | { | |
275 | struct super_block *s = container_of(work, struct super_block, | |
276 | destroy_work); | |
583340de AV |
277 | security_sb_free(s); |
278 | put_user_ns(s->s_user_ns); | |
279 | kfree(s->s_subtype); | |
280 | for (int i = 0; i < SB_FREEZE_LEVELS; i++) | |
8129ed29 | 281 | percpu_free_rwsem(&s->s_writers.rw_sem[i]); |
853b39a7 ON |
282 | kfree(s); |
283 | } | |
284 | ||
285 | static void destroy_super_rcu(struct rcu_head *head) | |
286 | { | |
287 | struct super_block *s = container_of(head, struct super_block, rcu); | |
288 | INIT_WORK(&s->destroy_work, destroy_super_work); | |
289 | schedule_work(&s->destroy_work); | |
290 | } | |
291 | ||
0200894d AV |
292 | /* Free a superblock that has never been seen by anyone */ |
293 | static void destroy_unused_super(struct super_block *s) | |
5accdf82 | 294 | { |
0200894d AV |
295 | if (!s) |
296 | return; | |
0ed33598 | 297 | super_unlock_excl(s); |
7eb5e882 AV |
298 | list_lru_destroy(&s->s_dentry_lru); |
299 | list_lru_destroy(&s->s_inode_lru); | |
1720f5dd | 300 | shrinker_free(s->s_shrink); |
0200894d AV |
301 | /* no delays needed */ |
302 | destroy_super_work(&s->destroy_work); | |
5accdf82 JK |
303 | } |
304 | ||
1da177e4 LT |
305 | /** |
306 | * alloc_super - create new superblock | |
fe2bbc48 | 307 | * @type: filesystem type superblock should belong to |
9249e17f | 308 | * @flags: the mount flags |
6e4eab57 | 309 | * @user_ns: User namespace for the super_block |
1da177e4 LT |
310 | * |
311 | * Allocates and initializes a new &struct super_block. alloc_super() | |
312 | * returns a pointer new superblock or %NULL if allocation had failed. | |
313 | */ | |
6e4eab57 EB |
314 | static struct super_block *alloc_super(struct file_system_type *type, int flags, |
315 | struct user_namespace *user_ns) | |
1da177e4 | 316 | { |
2b46a19d | 317 | struct super_block *s = kzalloc(sizeof(struct super_block), GFP_KERNEL); |
b87221de | 318 | static const struct super_operations default_op; |
7eb5e882 AV |
319 | int i; |
320 | ||
321 | if (!s) | |
322 | return NULL; | |
1da177e4 | 323 | |
b5bd856a | 324 | INIT_LIST_HEAD(&s->s_mounts); |
6e4eab57 | 325 | s->s_user_ns = get_user_ns(user_ns); |
ca0168e8 AV |
326 | init_rwsem(&s->s_umount); |
327 | lockdep_set_class(&s->s_umount, &type->s_umount_key); | |
328 | /* | |
329 | * sget() can have s_umount recursion. | |
330 | * | |
331 | * When it cannot find a suitable sb, it allocates a new | |
332 | * one (this one), and tries again to find a suitable old | |
333 | * one. | |
334 | * | |
335 | * In case that succeeds, it will acquire the s_umount | |
336 | * lock of the old one. Since these are clearly distrinct | |
337 | * locks, and this object isn't exposed yet, there's no | |
338 | * risk of deadlocks. | |
339 | * | |
340 | * Annotate this by putting this lock in a different | |
341 | * subclass. | |
342 | */ | |
343 | down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING); | |
b5bd856a | 344 | |
7eb5e882 AV |
345 | if (security_sb_alloc(s)) |
346 | goto fail; | |
7b7a8665 | 347 | |
7eb5e882 | 348 | for (i = 0; i < SB_FREEZE_LEVELS; i++) { |
8129ed29 ON |
349 | if (__percpu_init_rwsem(&s->s_writers.rw_sem[i], |
350 | sb_writers_name[i], | |
351 | &type->s_writers_key[i])) | |
7eb5e882 | 352 | goto fail; |
1da177e4 | 353 | } |
df0ce26c | 354 | s->s_bdi = &noop_backing_dev_info; |
7eb5e882 | 355 | s->s_flags = flags; |
cc50a07a | 356 | if (s->s_user_ns != &init_user_ns) |
67690f93 | 357 | s->s_iflags |= SB_I_NODEV; |
7eb5e882 | 358 | INIT_HLIST_NODE(&s->s_instances); |
f1ee6162 | 359 | INIT_HLIST_BL_HEAD(&s->s_roots); |
e97fedb9 | 360 | mutex_init(&s->s_sync_lock); |
7eb5e882 | 361 | INIT_LIST_HEAD(&s->s_inodes); |
74278da9 | 362 | spin_lock_init(&s->s_inode_list_lock); |
6c60d2b5 DC |
363 | INIT_LIST_HEAD(&s->s_inodes_wb); |
364 | spin_lock_init(&s->s_inode_wblist_lock); | |
7eb5e882 | 365 | |
7eb5e882 AV |
366 | s->s_count = 1; |
367 | atomic_set(&s->s_active, 1); | |
368 | mutex_init(&s->s_vfs_rename_mutex); | |
369 | lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key); | |
bc8230ee | 370 | init_rwsem(&s->s_dquot.dqio_sem); |
7eb5e882 AV |
371 | s->s_maxbytes = MAX_NON_LFS; |
372 | s->s_op = &default_op; | |
373 | s->s_time_gran = 1000000000; | |
188d20bc DD |
374 | s->s_time_min = TIME64_MIN; |
375 | s->s_time_max = TIME64_MAX; | |
7eb5e882 | 376 | |
1720f5dd QZ |
377 | s->s_shrink = shrinker_alloc(SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE, |
378 | "sb-%s", type->name); | |
379 | if (!s->s_shrink) | |
8e04944f | 380 | goto fail; |
1720f5dd QZ |
381 | |
382 | s->s_shrink->scan_objects = super_cache_scan; | |
383 | s->s_shrink->count_objects = super_cache_count; | |
384 | s->s_shrink->batch = 1024; | |
385 | s->s_shrink->private_data = s; | |
386 | ||
387 | if (list_lru_init_memcg(&s->s_dentry_lru, s->s_shrink)) | |
2b3648a6 | 388 | goto fail; |
1720f5dd | 389 | if (list_lru_init_memcg(&s->s_inode_lru, s->s_shrink)) |
2b3648a6 | 390 | goto fail; |
1da177e4 | 391 | return s; |
5ca302c8 | 392 | |
7eb5e882 | 393 | fail: |
0200894d | 394 | destroy_unused_super(s); |
7eb5e882 | 395 | return NULL; |
1da177e4 LT |
396 | } |
397 | ||
398 | /* Superblock refcounting */ | |
399 | ||
400 | /* | |
35cf7ba0 | 401 | * Drop a superblock's refcount. The caller must hold sb_lock. |
1da177e4 | 402 | */ |
c645b930 | 403 | static void __put_super(struct super_block *s) |
1da177e4 | 404 | { |
c645b930 AV |
405 | if (!--s->s_count) { |
406 | list_del_init(&s->s_list); | |
407 | WARN_ON(s->s_dentry_lru.node); | |
408 | WARN_ON(s->s_inode_lru.node); | |
409 | WARN_ON(!list_empty(&s->s_mounts)); | |
c645b930 | 410 | call_rcu(&s->rcu, destroy_super_rcu); |
1da177e4 | 411 | } |
1da177e4 LT |
412 | } |
413 | ||
414 | /** | |
415 | * put_super - drop a temporary reference to superblock | |
416 | * @sb: superblock in question | |
417 | * | |
418 | * Drops a temporary reference, frees superblock if there's no | |
419 | * references left. | |
420 | */ | |
60b49885 | 421 | void put_super(struct super_block *sb) |
1da177e4 LT |
422 | { |
423 | spin_lock(&sb_lock); | |
424 | __put_super(sb); | |
425 | spin_unlock(&sb_lock); | |
426 | } | |
427 | ||
dc3216b1 CB |
428 | static void kill_super_notify(struct super_block *sb) |
429 | { | |
430 | lockdep_assert_not_held(&sb->s_umount); | |
431 | ||
432 | /* already notified earlier */ | |
433 | if (sb->s_flags & SB_DEAD) | |
434 | return; | |
435 | ||
436 | /* | |
437 | * Remove it from @fs_supers so it isn't found by new | |
438 | * sget{_fc}() walkers anymore. Any concurrent mounter still | |
439 | * managing to grab a temporary reference is guaranteed to | |
440 | * already see SB_DYING and will wait until we notify them about | |
441 | * SB_DEAD. | |
442 | */ | |
443 | spin_lock(&sb_lock); | |
444 | hlist_del_init(&sb->s_instances); | |
445 | spin_unlock(&sb_lock); | |
446 | ||
447 | /* | |
448 | * Let concurrent mounts know that this thing is really dead. | |
449 | * We don't need @sb->s_umount here as every concurrent caller | |
450 | * will see SB_DYING and either discard the superblock or wait | |
451 | * for SB_DEAD. | |
452 | */ | |
453 | super_wake(sb, SB_DEAD); | |
454 | } | |
1da177e4 LT |
455 | |
456 | /** | |
1712ac8f | 457 | * deactivate_locked_super - drop an active reference to superblock |
1da177e4 LT |
458 | * @s: superblock to deactivate |
459 | * | |
bd7ced98 | 460 | * Drops an active reference to superblock, converting it into a temporary |
1712ac8f | 461 | * one if there is no other active references left. In that case we |
1da177e4 LT |
462 | * tell fs driver to shut it down and drop the temporary reference we |
463 | * had just acquired. | |
1712ac8f AV |
464 | * |
465 | * Caller holds exclusive lock on superblock; that lock is released. | |
1da177e4 | 466 | */ |
1712ac8f | 467 | void deactivate_locked_super(struct super_block *s) |
1da177e4 LT |
468 | { |
469 | struct file_system_type *fs = s->s_type; | |
b20bd1a5 | 470 | if (atomic_dec_and_test(&s->s_active)) { |
1720f5dd | 471 | shrinker_free(s->s_shrink); |
28f2cd4f | 472 | fs->kill_sb(s); |
f5e1dd34 | 473 | |
dc3216b1 CB |
474 | kill_super_notify(s); |
475 | ||
c0a5b560 VD |
476 | /* |
477 | * Since list_lru_destroy() may sleep, we cannot call it from | |
478 | * put_super(), where we hold the sb_lock. Therefore we destroy | |
479 | * the lru lists right now. | |
480 | */ | |
481 | list_lru_destroy(&s->s_dentry_lru); | |
482 | list_lru_destroy(&s->s_inode_lru); | |
483 | ||
1da177e4 LT |
484 | put_filesystem(fs); |
485 | put_super(s); | |
1712ac8f | 486 | } else { |
0ed33598 | 487 | super_unlock_excl(s); |
1da177e4 LT |
488 | } |
489 | } | |
490 | ||
1712ac8f | 491 | EXPORT_SYMBOL(deactivate_locked_super); |
1da177e4 | 492 | |
74dbbdd7 | 493 | /** |
1712ac8f | 494 | * deactivate_super - drop an active reference to superblock |
74dbbdd7 AV |
495 | * @s: superblock to deactivate |
496 | * | |
1712ac8f AV |
497 | * Variant of deactivate_locked_super(), except that superblock is *not* |
498 | * locked by caller. If we are going to drop the final active reference, | |
499 | * lock will be acquired prior to that. | |
74dbbdd7 | 500 | */ |
1712ac8f | 501 | void deactivate_super(struct super_block *s) |
74dbbdd7 | 502 | { |
cc23402c | 503 | if (!atomic_add_unless(&s->s_active, -1, 1)) { |
5e874914 | 504 | __super_lock_excl(s); |
1712ac8f | 505 | deactivate_locked_super(s); |
74dbbdd7 AV |
506 | } |
507 | } | |
508 | ||
1712ac8f | 509 | EXPORT_SYMBOL(deactivate_super); |
74dbbdd7 | 510 | |
1da177e4 | 511 | /** |
97cbed04 | 512 | * grab_super - acquire an active reference to a superblock |
2c18a63b CB |
513 | * @sb: superblock to acquire |
514 | * | |
515 | * Acquire a temporary reference on a superblock and try to trade it for | |
516 | * an active reference. This is used in sget{_fc}() to wait for a | |
517 | * superblock to either become SB_BORN or for it to pass through | |
518 | * sb->kill() and be marked as SB_DEAD. | |
519 | * | |
520 | * Return: This returns true if an active reference could be acquired, | |
521 | * false if not. | |
522 | */ | |
97cbed04 | 523 | static bool grab_super(struct super_block *sb) |
2c18a63b | 524 | { |
97cbed04 | 525 | bool locked; |
2c18a63b CB |
526 | |
527 | sb->s_count++; | |
97cbed04 CB |
528 | spin_unlock(&sb_lock); |
529 | locked = super_lock_excl(sb); | |
530 | if (locked) { | |
531 | if (atomic_inc_not_zero(&sb->s_active)) { | |
532 | put_super(sb); | |
533 | return true; | |
534 | } | |
535 | super_unlock_excl(sb); | |
2c18a63b | 536 | } |
b30850c5 | 537 | wait_var_event(&sb->s_flags, super_flags(sb, SB_DEAD)); |
345a5c4a | 538 | put_super(sb); |
2c18a63b CB |
539 | return false; |
540 | } | |
541 | ||
12ad3ab6 | 542 | /* |
d8ce82ef | 543 | * super_trylock_shared - try to grab ->s_umount shared |
331cbdee | 544 | * @sb: reference we are trying to grab |
12ad3ab6 | 545 | * |
eb6ef3df | 546 | * Try to prevent fs shutdown. This is used in places where we |
12ad3ab6 | 547 | * cannot take an active reference but we need to ensure that the |
eb6ef3df KK |
548 | * filesystem is not shut down while we are working on it. It returns |
549 | * false if we cannot acquire s_umount or if we lose the race and | |
550 | * filesystem already got into shutdown, and returns true with the s_umount | |
551 | * lock held in read mode in case of success. On successful return, | |
552 | * the caller must drop the s_umount lock when done. | |
553 | * | |
554 | * Note that unlike get_super() et.al. this one does *not* bump ->s_count. | |
555 | * The reason why it's safe is that we are OK with doing trylock instead | |
556 | * of down_read(). There's a couple of places that are OK with that, but | |
557 | * it's very much not a general-purpose interface. | |
12ad3ab6 | 558 | */ |
d8ce82ef | 559 | bool super_trylock_shared(struct super_block *sb) |
12ad3ab6 | 560 | { |
12ad3ab6 | 561 | if (down_read_trylock(&sb->s_umount)) { |
5e874914 CB |
562 | if (!(sb->s_flags & SB_DYING) && sb->s_root && |
563 | (sb->s_flags & SB_BORN)) | |
12ad3ab6 | 564 | return true; |
0ed33598 | 565 | super_unlock_shared(sb); |
12ad3ab6 DC |
566 | } |
567 | ||
12ad3ab6 DC |
568 | return false; |
569 | } | |
570 | ||
04b94071 DL |
571 | /** |
572 | * retire_super - prevents superblock from being reused | |
573 | * @sb: superblock to retire | |
574 | * | |
575 | * The function marks superblock to be ignored in superblock test, which | |
576 | * prevents it from being reused for any new mounts. If the superblock has | |
577 | * a private bdi, it also unregisters it, but doesn't reduce the refcount | |
578 | * of the superblock to prevent potential races. The refcount is reduced | |
579 | * by generic_shutdown_super(). The function can not be called | |
580 | * concurrently with generic_shutdown_super(). It is safe to call the | |
581 | * function multiple times, subsequent calls have no effect. | |
582 | * | |
583 | * The marker will affect the re-use only for block-device-based | |
584 | * superblocks. Other superblocks will still get marked if this function | |
585 | * is used, but that will not affect their reusability. | |
586 | */ | |
587 | void retire_super(struct super_block *sb) | |
588 | { | |
589 | WARN_ON(!sb->s_bdev); | |
5e874914 | 590 | __super_lock_excl(sb); |
04b94071 DL |
591 | if (sb->s_iflags & SB_I_PERSB_BDI) { |
592 | bdi_unregister(sb->s_bdi); | |
593 | sb->s_iflags &= ~SB_I_PERSB_BDI; | |
594 | } | |
595 | sb->s_iflags |= SB_I_RETIRED; | |
0ed33598 | 596 | super_unlock_excl(sb); |
04b94071 DL |
597 | } |
598 | EXPORT_SYMBOL(retire_super); | |
599 | ||
1da177e4 LT |
600 | /** |
601 | * generic_shutdown_super - common helper for ->kill_sb() | |
602 | * @sb: superblock to kill | |
603 | * | |
604 | * generic_shutdown_super() does all fs-independent work on superblock | |
605 | * shutdown. Typical ->kill_sb() should pick all fs-specific objects | |
606 | * that need destruction out of superblock, call generic_shutdown_super() | |
607 | * and release aforementioned objects. Note: dentries and inodes _are_ | |
608 | * taken care of and do not need specific handling. | |
c636ebdb DH |
609 | * |
610 | * Upon calling this function, the filesystem may no longer alter or | |
611 | * rearrange the set of dentries belonging to this super_block, nor may it | |
612 | * change the attachments of dentries to inodes. | |
1da177e4 LT |
613 | */ |
614 | void generic_shutdown_super(struct super_block *sb) | |
615 | { | |
ee9b6d61 | 616 | const struct super_operations *sop = sb->s_op; |
1da177e4 | 617 | |
c636ebdb DH |
618 | if (sb->s_root) { |
619 | shrink_dcache_for_umount(sb); | |
60b0680f | 620 | sync_filesystem(sb); |
e462ec50 | 621 | sb->s_flags &= ~SB_ACTIVE; |
efaee192 | 622 | |
a1a0e23e | 623 | cgroup_writeback_umount(); |
63997e98 | 624 | |
ccb820dc | 625 | /* Evict all inodes with zero refcount. */ |
63997e98 | 626 | evict_inodes(sb); |
ccb820dc EB |
627 | |
628 | /* | |
629 | * Clean up and evict any inodes that still have references due | |
630 | * to fsnotify or the security policy. | |
631 | */ | |
1edc8eb2 | 632 | fsnotify_sb_delete(sb); |
83e804f0 | 633 | security_sb_delete(sb); |
1da177e4 | 634 | |
7b7a8665 CH |
635 | if (sb->s_dio_done_wq) { |
636 | destroy_workqueue(sb->s_dio_done_wq); | |
637 | sb->s_dio_done_wq = NULL; | |
638 | } | |
639 | ||
1da177e4 LT |
640 | if (sop->put_super) |
641 | sop->put_super(sb); | |
642 | ||
2a0e8571 JB |
643 | /* |
644 | * Now that all potentially-encrypted inodes have been evicted, | |
645 | * the fscrypt keyring can be destroyed. | |
646 | */ | |
647 | fscrypt_destroy_keyring(sb); | |
648 | ||
47d58691 JH |
649 | if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes), |
650 | "VFS: Busy inodes after unmount of %s (%s)", | |
651 | sb->s_id, sb->s_type->name)) { | |
652 | /* | |
653 | * Adding a proper bailout path here would be hard, but | |
654 | * we can at least make it more likely that a later | |
655 | * iput_final() or such crashes cleanly. | |
656 | */ | |
657 | struct inode *inode; | |
658 | ||
659 | spin_lock(&sb->s_inode_list_lock); | |
660 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { | |
661 | inode->i_op = VFS_PTR_POISON; | |
662 | inode->i_sb = VFS_PTR_POISON; | |
663 | inode->i_mapping = VFS_PTR_POISON; | |
664 | } | |
665 | spin_unlock(&sb->s_inode_list_lock); | |
1da177e4 | 666 | } |
1da177e4 | 667 | } |
5e874914 CB |
668 | /* |
669 | * Broadcast to everyone that grabbed a temporary reference to this | |
670 | * superblock before we removed it from @fs_supers that the superblock | |
671 | * is dying. Every walker of @fs_supers outside of sget{_fc}() will now | |
672 | * discard this superblock and treat it as dead. | |
2c18a63b CB |
673 | * |
674 | * We leave the superblock on @fs_supers so it can be found by | |
675 | * sget{_fc}() until we passed sb->kill_sb(). | |
5e874914 CB |
676 | */ |
677 | super_wake(sb, SB_DYING); | |
0ed33598 | 678 | super_unlock_excl(sb); |
c1844d53 | 679 | if (sb->s_bdi != &noop_backing_dev_info) { |
0b3ea092 CH |
680 | if (sb->s_iflags & SB_I_PERSB_BDI) |
681 | bdi_unregister(sb->s_bdi); | |
fca39346 JK |
682 | bdi_put(sb->s_bdi); |
683 | sb->s_bdi = &noop_backing_dev_info; | |
fca39346 | 684 | } |
1da177e4 LT |
685 | } |
686 | ||
687 | EXPORT_SYMBOL(generic_shutdown_super); | |
688 | ||
20284ab7 | 689 | bool mount_capable(struct fs_context *fc) |
0ce0cf12 | 690 | { |
20284ab7 | 691 | if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT)) |
0ce0cf12 AV |
692 | return capable(CAP_SYS_ADMIN); |
693 | else | |
c2c44ec2 | 694 | return ns_capable(fc->user_ns, CAP_SYS_ADMIN); |
0ce0cf12 AV |
695 | } |
696 | ||
cb50b348 AV |
697 | /** |
698 | * sget_fc - Find or create a superblock | |
699 | * @fc: Filesystem context. | |
700 | * @test: Comparison callback | |
701 | * @set: Setup callback | |
702 | * | |
22ed7ecd | 703 | * Create a new superblock or find an existing one. |
cb50b348 | 704 | * |
22ed7ecd CB |
705 | * The @test callback is used to find a matching existing superblock. |
706 | * Whether or not the requested parameters in @fc are taken into account | |
707 | * is specific to the @test callback that is used. They may even be | |
708 | * completely ignored. | |
709 | * | |
710 | * If an extant superblock is matched, it will be returned unless: | |
711 | * | |
712 | * (1) the namespace the filesystem context @fc and the extant | |
713 | * superblock's namespace differ | |
714 | * | |
715 | * (2) the filesystem context @fc has requested that reusing an extant | |
716 | * superblock is not allowed | |
717 | * | |
718 | * In both cases EBUSY will be returned. | |
cb50b348 AV |
719 | * |
720 | * If no match is made, a new superblock will be allocated and basic | |
22ed7ecd CB |
721 | * initialisation will be performed (s_type, s_fs_info and s_id will be |
722 | * set and the @set callback will be invoked), the superblock will be | |
723 | * published and it will be returned in a partially constructed state | |
724 | * with SB_BORN and SB_ACTIVE as yet unset. | |
725 | * | |
726 | * Return: On success, an extant or newly created superblock is | |
727 | * returned. On failure an error pointer is returned. | |
cb50b348 AV |
728 | */ |
729 | struct super_block *sget_fc(struct fs_context *fc, | |
730 | int (*test)(struct super_block *, struct fs_context *), | |
731 | int (*set)(struct super_block *, struct fs_context *)) | |
732 | { | |
733 | struct super_block *s = NULL; | |
734 | struct super_block *old; | |
735 | struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns; | |
736 | int err; | |
737 | ||
cb50b348 AV |
738 | retry: |
739 | spin_lock(&sb_lock); | |
740 | if (test) { | |
741 | hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) { | |
742 | if (test(old, fc)) | |
743 | goto share_extant_sb; | |
744 | } | |
745 | } | |
746 | if (!s) { | |
747 | spin_unlock(&sb_lock); | |
748 | s = alloc_super(fc->fs_type, fc->sb_flags, user_ns); | |
749 | if (!s) | |
750 | return ERR_PTR(-ENOMEM); | |
751 | goto retry; | |
752 | } | |
753 | ||
754 | s->s_fs_info = fc->s_fs_info; | |
755 | err = set(s, fc); | |
756 | if (err) { | |
757 | s->s_fs_info = NULL; | |
758 | spin_unlock(&sb_lock); | |
759 | destroy_unused_super(s); | |
760 | return ERR_PTR(err); | |
761 | } | |
762 | fc->s_fs_info = NULL; | |
763 | s->s_type = fc->fs_type; | |
c80fa7c8 | 764 | s->s_iflags |= fc->s_iflags; |
c642256b | 765 | strscpy(s->s_id, s->s_type->name, sizeof(s->s_id)); |
5e874914 CB |
766 | /* |
767 | * Make the superblock visible on @super_blocks and @fs_supers. | |
768 | * It's in a nascent state and users should wait on SB_BORN or | |
769 | * SB_DYING to be set. | |
770 | */ | |
cb50b348 AV |
771 | list_add_tail(&s->s_list, &super_blocks); |
772 | hlist_add_head(&s->s_instances, &s->s_type->fs_supers); | |
773 | spin_unlock(&sb_lock); | |
774 | get_filesystem(s->s_type); | |
1720f5dd | 775 | shrinker_register(s->s_shrink); |
cb50b348 AV |
776 | return s; |
777 | ||
778 | share_extant_sb: | |
22ed7ecd | 779 | if (user_ns != old->s_user_ns || fc->exclusive) { |
cb50b348 AV |
780 | spin_unlock(&sb_lock); |
781 | destroy_unused_super(s); | |
22ed7ecd CB |
782 | if (fc->exclusive) |
783 | warnfc(fc, "reusing existing filesystem not allowed"); | |
784 | else | |
785 | warnfc(fc, "reusing existing filesystem in another namespace not allowed"); | |
cb50b348 AV |
786 | return ERR_PTR(-EBUSY); |
787 | } | |
97cbed04 | 788 | if (!grab_super(old)) |
cb50b348 AV |
789 | goto retry; |
790 | destroy_unused_super(s); | |
791 | return old; | |
792 | } | |
793 | EXPORT_SYMBOL(sget_fc); | |
794 | ||
1da177e4 | 795 | /** |
023d066a DH |
796 | * sget - find or create a superblock |
797 | * @type: filesystem type superblock should belong to | |
798 | * @test: comparison callback | |
799 | * @set: setup callback | |
800 | * @flags: mount flags | |
801 | * @data: argument to each of them | |
1da177e4 | 802 | */ |
023d066a | 803 | struct super_block *sget(struct file_system_type *type, |
1da177e4 LT |
804 | int (*test)(struct super_block *,void *), |
805 | int (*set)(struct super_block *,void *), | |
023d066a | 806 | int flags, |
1da177e4 LT |
807 | void *data) |
808 | { | |
023d066a | 809 | struct user_namespace *user_ns = current_user_ns(); |
1da177e4 | 810 | struct super_block *s = NULL; |
d4730127 | 811 | struct super_block *old; |
1da177e4 LT |
812 | int err; |
813 | ||
023d066a DH |
814 | /* We don't yet pass the user namespace of the parent |
815 | * mount through to here so always use &init_user_ns | |
816 | * until that changes. | |
817 | */ | |
818 | if (flags & SB_SUBMOUNT) | |
819 | user_ns = &init_user_ns; | |
820 | ||
1da177e4 LT |
821 | retry: |
822 | spin_lock(&sb_lock); | |
d4730127 | 823 | if (test) { |
b67bfe0d | 824 | hlist_for_each_entry(old, &type->fs_supers, s_instances) { |
d4730127 MK |
825 | if (!test(old, data)) |
826 | continue; | |
6e4eab57 EB |
827 | if (user_ns != old->s_user_ns) { |
828 | spin_unlock(&sb_lock); | |
0200894d | 829 | destroy_unused_super(s); |
6e4eab57 EB |
830 | return ERR_PTR(-EBUSY); |
831 | } | |
97cbed04 | 832 | if (!grab_super(old)) |
d4730127 | 833 | goto retry; |
0200894d | 834 | destroy_unused_super(s); |
d4730127 MK |
835 | return old; |
836 | } | |
1da177e4 LT |
837 | } |
838 | if (!s) { | |
839 | spin_unlock(&sb_lock); | |
e462ec50 | 840 | s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns); |
1da177e4 LT |
841 | if (!s) |
842 | return ERR_PTR(-ENOMEM); | |
843 | goto retry; | |
844 | } | |
dd111b31 | 845 | |
1da177e4 LT |
846 | err = set(s, data); |
847 | if (err) { | |
848 | spin_unlock(&sb_lock); | |
0200894d | 849 | destroy_unused_super(s); |
1da177e4 LT |
850 | return ERR_PTR(err); |
851 | } | |
852 | s->s_type = type; | |
c642256b | 853 | strscpy(s->s_id, type->name, sizeof(s->s_id)); |
1da177e4 | 854 | list_add_tail(&s->s_list, &super_blocks); |
a5166169 | 855 | hlist_add_head(&s->s_instances, &type->fs_supers); |
1da177e4 LT |
856 | spin_unlock(&sb_lock); |
857 | get_filesystem(type); | |
1720f5dd | 858 | shrinker_register(s->s_shrink); |
1da177e4 LT |
859 | return s; |
860 | } | |
1da177e4 LT |
861 | EXPORT_SYMBOL(sget); |
862 | ||
863 | void drop_super(struct super_block *sb) | |
864 | { | |
0ed33598 | 865 | super_unlock_shared(sb); |
1da177e4 LT |
866 | put_super(sb); |
867 | } | |
868 | ||
869 | EXPORT_SYMBOL(drop_super); | |
870 | ||
ba6379f7 JK |
871 | void drop_super_exclusive(struct super_block *sb) |
872 | { | |
0ed33598 | 873 | super_unlock_excl(sb); |
ba6379f7 JK |
874 | put_super(sb); |
875 | } | |
876 | EXPORT_SYMBOL(drop_super_exclusive); | |
877 | ||
fa7c1d50 MG |
878 | static void __iterate_supers(void (*f)(struct super_block *)) |
879 | { | |
880 | struct super_block *sb, *p = NULL; | |
881 | ||
882 | spin_lock(&sb_lock); | |
883 | list_for_each_entry(sb, &super_blocks, s_list) { | |
b30850c5 | 884 | if (super_flags(sb, SB_DYING)) |
fa7c1d50 MG |
885 | continue; |
886 | sb->s_count++; | |
887 | spin_unlock(&sb_lock); | |
888 | ||
889 | f(sb); | |
890 | ||
891 | spin_lock(&sb_lock); | |
892 | if (p) | |
893 | __put_super(p); | |
894 | p = sb; | |
895 | } | |
896 | if (p) | |
897 | __put_super(p); | |
898 | spin_unlock(&sb_lock); | |
899 | } | |
01a05b33 AV |
900 | /** |
901 | * iterate_supers - call function for all active superblocks | |
902 | * @f: function to call | |
903 | * @arg: argument to pass to it | |
904 | * | |
905 | * Scans the superblock list and calls given function, passing it | |
906 | * locked superblock and given argument. | |
907 | */ | |
908 | void iterate_supers(void (*f)(struct super_block *, void *), void *arg) | |
909 | { | |
dca33252 | 910 | struct super_block *sb, *p = NULL; |
01a05b33 AV |
911 | |
912 | spin_lock(&sb_lock); | |
dca33252 | 913 | list_for_each_entry(sb, &super_blocks, s_list) { |
f0cd9880 | 914 | bool locked; |
5e874914 | 915 | |
01a05b33 AV |
916 | sb->s_count++; |
917 | spin_unlock(&sb_lock); | |
918 | ||
f0cd9880 CB |
919 | locked = super_lock_shared(sb); |
920 | if (locked) { | |
921 | if (sb->s_root) | |
922 | f(sb, arg); | |
923 | super_unlock_shared(sb); | |
924 | } | |
01a05b33 AV |
925 | |
926 | spin_lock(&sb_lock); | |
dca33252 AV |
927 | if (p) |
928 | __put_super(p); | |
929 | p = sb; | |
01a05b33 | 930 | } |
dca33252 AV |
931 | if (p) |
932 | __put_super(p); | |
01a05b33 AV |
933 | spin_unlock(&sb_lock); |
934 | } | |
935 | ||
43e15cdb AV |
936 | /** |
937 | * iterate_supers_type - call function for superblocks of given type | |
938 | * @type: fs type | |
939 | * @f: function to call | |
940 | * @arg: argument to pass to it | |
941 | * | |
942 | * Scans the superblock list and calls given function, passing it | |
943 | * locked superblock and given argument. | |
944 | */ | |
945 | void iterate_supers_type(struct file_system_type *type, | |
946 | void (*f)(struct super_block *, void *), void *arg) | |
947 | { | |
948 | struct super_block *sb, *p = NULL; | |
949 | ||
950 | spin_lock(&sb_lock); | |
b67bfe0d | 951 | hlist_for_each_entry(sb, &type->fs_supers, s_instances) { |
f0cd9880 | 952 | bool locked; |
5e874914 | 953 | |
43e15cdb AV |
954 | sb->s_count++; |
955 | spin_unlock(&sb_lock); | |
956 | ||
f0cd9880 CB |
957 | locked = super_lock_shared(sb); |
958 | if (locked) { | |
959 | if (sb->s_root) | |
960 | f(sb, arg); | |
961 | super_unlock_shared(sb); | |
962 | } | |
43e15cdb AV |
963 | |
964 | spin_lock(&sb_lock); | |
965 | if (p) | |
966 | __put_super(p); | |
967 | p = sb; | |
968 | } | |
969 | if (p) | |
970 | __put_super(p); | |
971 | spin_unlock(&sb_lock); | |
972 | } | |
973 | ||
974 | EXPORT_SYMBOL(iterate_supers_type); | |
975 | ||
4e7b5671 | 976 | struct super_block *user_get_super(dev_t dev, bool excl) |
1da177e4 | 977 | { |
618f0636 | 978 | struct super_block *sb; |
1da177e4 | 979 | |
1da177e4 | 980 | spin_lock(&sb_lock); |
618f0636 KK |
981 | list_for_each_entry(sb, &super_blocks, s_list) { |
982 | if (sb->s_dev == dev) { | |
f0cd9880 | 983 | bool locked; |
5e874914 | 984 | |
618f0636 | 985 | sb->s_count++; |
1da177e4 | 986 | spin_unlock(&sb_lock); |
df40c01a | 987 | /* still alive? */ |
f0cd9880 CB |
988 | locked = super_lock(sb, excl); |
989 | if (locked) { | |
990 | if (sb->s_root) | |
991 | return sb; | |
992 | super_unlock(sb, excl); | |
993 | } | |
df40c01a | 994 | /* nope, got unmounted */ |
618f0636 | 995 | spin_lock(&sb_lock); |
df40c01a | 996 | __put_super(sb); |
5e874914 | 997 | break; |
1da177e4 LT |
998 | } |
999 | } | |
1000 | spin_unlock(&sb_lock); | |
1001 | return NULL; | |
1002 | } | |
1003 | ||
1da177e4 | 1004 | /** |
8d0347f6 DH |
1005 | * reconfigure_super - asks filesystem to change superblock parameters |
1006 | * @fc: The superblock and configuration | |
1da177e4 | 1007 | * |
8d0347f6 | 1008 | * Alters the configuration parameters of a live superblock. |
1da177e4 | 1009 | */ |
8d0347f6 | 1010 | int reconfigure_super(struct fs_context *fc) |
1da177e4 | 1011 | { |
8d0347f6 | 1012 | struct super_block *sb = fc->root->d_sb; |
1da177e4 | 1013 | int retval; |
8d0347f6 | 1014 | bool remount_ro = false; |
c541dce8 | 1015 | bool remount_rw = false; |
8d0347f6 | 1016 | bool force = fc->sb_flags & SB_FORCE; |
4504230a | 1017 | |
8d0347f6 DH |
1018 | if (fc->sb_flags_mask & ~MS_RMT_MASK) |
1019 | return -EINVAL; | |
5accdf82 | 1020 | if (sb->s_writers.frozen != SB_UNFROZEN) |
4504230a CH |
1021 | return -EBUSY; |
1022 | ||
8d0347f6 DH |
1023 | retval = security_sb_remount(sb, fc->security); |
1024 | if (retval) | |
1025 | return retval; | |
1026 | ||
1027 | if (fc->sb_flags_mask & SB_RDONLY) { | |
9361401e | 1028 | #ifdef CONFIG_BLOCK |
6f0d9689 CH |
1029 | if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev && |
1030 | bdev_read_only(sb->s_bdev)) | |
8d0347f6 | 1031 | return -EACCES; |
9361401e | 1032 | #endif |
c541dce8 | 1033 | remount_rw = !(fc->sb_flags & SB_RDONLY) && sb_rdonly(sb); |
8d0347f6 DH |
1034 | remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb); |
1035 | } | |
d208bbdd | 1036 | |
0aec09d0 | 1037 | if (remount_ro) { |
fdab684d | 1038 | if (!hlist_empty(&sb->s_pins)) { |
0ed33598 | 1039 | super_unlock_excl(sb); |
fdab684d | 1040 | group_pin_kill(&sb->s_pins); |
5e874914 | 1041 | __super_lock_excl(sb); |
0aec09d0 AV |
1042 | if (!sb->s_root) |
1043 | return 0; | |
1044 | if (sb->s_writers.frozen != SB_UNFROZEN) | |
1045 | return -EBUSY; | |
8d0347f6 | 1046 | remount_ro = !sb_rdonly(sb); |
0aec09d0 AV |
1047 | } |
1048 | } | |
1049 | shrink_dcache_sb(sb); | |
1050 | ||
8d0347f6 DH |
1051 | /* If we are reconfiguring to RDONLY and current sb is read/write, |
1052 | * make sure there are no files open for writing. | |
1053 | */ | |
d208bbdd | 1054 | if (remount_ro) { |
4ed5e82f | 1055 | if (force) { |
d7439fb1 | 1056 | sb_start_ro_state_change(sb); |
4ed5e82f MS |
1057 | } else { |
1058 | retval = sb_prepare_remount_readonly(sb); | |
1059 | if (retval) | |
1060 | return retval; | |
4ed5e82f | 1061 | } |
c541dce8 JK |
1062 | } else if (remount_rw) { |
1063 | /* | |
d7439fb1 JK |
1064 | * Protect filesystem's reconfigure code from writes from |
1065 | * userspace until reconfigure finishes. | |
c541dce8 | 1066 | */ |
d7439fb1 | 1067 | sb_start_ro_state_change(sb); |
1da177e4 LT |
1068 | } |
1069 | ||
f3a09c92 AV |
1070 | if (fc->ops->reconfigure) { |
1071 | retval = fc->ops->reconfigure(fc); | |
1072 | if (retval) { | |
1073 | if (!force) | |
1074 | goto cancel_readonly; | |
1075 | /* If forced remount, go ahead despite any errors */ | |
1076 | WARN(1, "forced remount of a %s fs returned %i\n", | |
1077 | sb->s_type->name, retval); | |
1078 | } | |
1da177e4 | 1079 | } |
8d0347f6 DH |
1080 | |
1081 | WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) | | |
1082 | (fc->sb_flags & fc->sb_flags_mask))); | |
d7439fb1 | 1083 | sb_end_ro_state_change(sb); |
c79d967d | 1084 | |
d208bbdd NP |
1085 | /* |
1086 | * Some filesystems modify their metadata via some other path than the | |
1087 | * bdev buffer cache (eg. use a private mapping, or directories in | |
1088 | * pagecache, etc). Also file data modifications go via their own | |
1089 | * mappings. So If we try to mount readonly then copy the filesystem | |
1090 | * from bdev, we could get stale data, so invalidate it to give a best | |
1091 | * effort at coherency. | |
1092 | */ | |
1093 | if (remount_ro && sb->s_bdev) | |
1094 | invalidate_bdev(sb->s_bdev); | |
1da177e4 | 1095 | return 0; |
4ed5e82f MS |
1096 | |
1097 | cancel_readonly: | |
d7439fb1 | 1098 | sb_end_ro_state_change(sb); |
4ed5e82f | 1099 | return retval; |
1da177e4 LT |
1100 | } |
1101 | ||
fa7c1d50 | 1102 | static void do_emergency_remount_callback(struct super_block *sb) |
1da177e4 | 1103 | { |
f0cd9880 | 1104 | bool locked = super_lock_excl(sb); |
5e874914 | 1105 | |
f0cd9880 | 1106 | if (locked && sb->s_root && sb->s_bdev && !sb_rdonly(sb)) { |
8d0347f6 DH |
1107 | struct fs_context *fc; |
1108 | ||
1109 | fc = fs_context_for_reconfigure(sb->s_root, | |
1110 | SB_RDONLY | SB_FORCE, SB_RDONLY); | |
1111 | if (!IS_ERR(fc)) { | |
1112 | if (parse_monolithic_mount_data(fc, NULL) == 0) | |
1113 | (void)reconfigure_super(fc); | |
1114 | put_fs_context(fc); | |
1115 | } | |
1da177e4 | 1116 | } |
f0cd9880 CB |
1117 | if (locked) |
1118 | super_unlock_excl(sb); | |
fa7c1d50 MG |
1119 | } |
1120 | ||
1121 | static void do_emergency_remount(struct work_struct *work) | |
1122 | { | |
1123 | __iterate_supers(do_emergency_remount_callback); | |
a2a9537a | 1124 | kfree(work); |
1da177e4 LT |
1125 | printk("Emergency Remount complete\n"); |
1126 | } | |
1127 | ||
1128 | void emergency_remount(void) | |
1129 | { | |
a2a9537a JA |
1130 | struct work_struct *work; |
1131 | ||
1132 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
1133 | if (work) { | |
1134 | INIT_WORK(work, do_emergency_remount); | |
1135 | schedule_work(work); | |
1136 | } | |
1da177e4 LT |
1137 | } |
1138 | ||
08fdc8a0 MG |
1139 | static void do_thaw_all_callback(struct super_block *sb) |
1140 | { | |
f0cd9880 | 1141 | bool locked = super_lock_excl(sb); |
5e874914 | 1142 | |
f0cd9880 | 1143 | if (locked && sb->s_root) { |
4a8b719f | 1144 | if (IS_ENABLED(CONFIG_BLOCK)) |
982c3b30 | 1145 | while (sb->s_bdev && !bdev_thaw(sb->s_bdev)) |
4a8b719f | 1146 | pr_warn("Emergency Thaw on %pg\n", sb->s_bdev); |
880b9577 | 1147 | thaw_super_locked(sb, FREEZE_HOLDER_USERSPACE); |
f0cd9880 | 1148 | return; |
08fdc8a0 | 1149 | } |
f0cd9880 CB |
1150 | if (locked) |
1151 | super_unlock_excl(sb); | |
08fdc8a0 MG |
1152 | } |
1153 | ||
1154 | static void do_thaw_all(struct work_struct *work) | |
1155 | { | |
1156 | __iterate_supers(do_thaw_all_callback); | |
1157 | kfree(work); | |
1158 | printk(KERN_WARNING "Emergency Thaw complete\n"); | |
1159 | } | |
1160 | ||
1161 | /** | |
1162 | * emergency_thaw_all -- forcibly thaw every frozen filesystem | |
1163 | * | |
1164 | * Used for emergency unfreeze of all filesystems via SysRq | |
1165 | */ | |
1166 | void emergency_thaw_all(void) | |
1167 | { | |
1168 | struct work_struct *work; | |
1169 | ||
1170 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
1171 | if (work) { | |
1172 | INIT_WORK(work, do_thaw_all); | |
1173 | schedule_work(work); | |
1174 | } | |
1175 | } | |
1176 | ||
ad76cbc6 | 1177 | static DEFINE_IDA(unnamed_dev_ida); |
1da177e4 | 1178 | |
5a66847e MW |
1179 | /** |
1180 | * get_anon_bdev - Allocate a block device for filesystems which don't have one. | |
1181 | * @p: Pointer to a dev_t. | |
1182 | * | |
1183 | * Filesystems which don't use real block devices can call this function | |
1184 | * to allocate a virtual block device. | |
1185 | * | |
1186 | * Context: Any context. Frequently called while holding sb_lock. | |
1187 | * Return: 0 on success, -EMFILE if there are no anonymous bdevs left | |
1188 | * or -ENOMEM if memory allocation failed. | |
1189 | */ | |
0ee5dc67 | 1190 | int get_anon_bdev(dev_t *p) |
1da177e4 LT |
1191 | { |
1192 | int dev; | |
5a66847e MW |
1193 | |
1194 | /* | |
1195 | * Many userspace utilities consider an FSID of 0 invalid. | |
1196 | * Always return at least 1 from get_anon_bdev. | |
1197 | */ | |
1198 | dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1, | |
1199 | GFP_ATOMIC); | |
1200 | if (dev == -ENOSPC) | |
1201 | dev = -EMFILE; | |
1202 | if (dev < 0) | |
1203 | return dev; | |
1204 | ||
1205 | *p = MKDEV(0, dev); | |
1da177e4 LT |
1206 | return 0; |
1207 | } | |
0ee5dc67 | 1208 | EXPORT_SYMBOL(get_anon_bdev); |
1da177e4 | 1209 | |
0ee5dc67 | 1210 | void free_anon_bdev(dev_t dev) |
1da177e4 | 1211 | { |
5a66847e | 1212 | ida_free(&unnamed_dev_ida, MINOR(dev)); |
1da177e4 | 1213 | } |
0ee5dc67 AV |
1214 | EXPORT_SYMBOL(free_anon_bdev); |
1215 | ||
1216 | int set_anon_super(struct super_block *s, void *data) | |
1217 | { | |
df0ce26c | 1218 | return get_anon_bdev(&s->s_dev); |
0ee5dc67 | 1219 | } |
0ee5dc67 AV |
1220 | EXPORT_SYMBOL(set_anon_super); |
1221 | ||
1222 | void kill_anon_super(struct super_block *sb) | |
1223 | { | |
1224 | dev_t dev = sb->s_dev; | |
1225 | generic_shutdown_super(sb); | |
dc3216b1 | 1226 | kill_super_notify(sb); |
0ee5dc67 AV |
1227 | free_anon_bdev(dev); |
1228 | } | |
1da177e4 LT |
1229 | EXPORT_SYMBOL(kill_anon_super); |
1230 | ||
1da177e4 LT |
1231 | void kill_litter_super(struct super_block *sb) |
1232 | { | |
1233 | if (sb->s_root) | |
1234 | d_genocide(sb->s_root); | |
1235 | kill_anon_super(sb); | |
1236 | } | |
1da177e4 LT |
1237 | EXPORT_SYMBOL(kill_litter_super); |
1238 | ||
cb50b348 AV |
1239 | int set_anon_super_fc(struct super_block *sb, struct fs_context *fc) |
1240 | { | |
1241 | return set_anon_super(sb, NULL); | |
1242 | } | |
1243 | EXPORT_SYMBOL(set_anon_super_fc); | |
1244 | ||
1245 | static int test_keyed_super(struct super_block *sb, struct fs_context *fc) | |
1246 | { | |
1247 | return sb->s_fs_info == fc->s_fs_info; | |
1248 | } | |
1249 | ||
1250 | static int test_single_super(struct super_block *s, struct fs_context *fc) | |
1251 | { | |
1252 | return 1; | |
1253 | } | |
1254 | ||
e062abae | 1255 | static int vfs_get_super(struct fs_context *fc, |
cda2ed05 CH |
1256 | int (*test)(struct super_block *, struct fs_context *), |
1257 | int (*fill_super)(struct super_block *sb, | |
1258 | struct fs_context *fc)) | |
cb50b348 | 1259 | { |
cb50b348 | 1260 | struct super_block *sb; |
43ce4c1f | 1261 | int err; |
cb50b348 | 1262 | |
cb50b348 AV |
1263 | sb = sget_fc(fc, test, set_anon_super_fc); |
1264 | if (IS_ERR(sb)) | |
1265 | return PTR_ERR(sb); | |
1266 | ||
1267 | if (!sb->s_root) { | |
43ce4c1f DH |
1268 | err = fill_super(sb, fc); |
1269 | if (err) | |
1270 | goto error; | |
cb50b348 AV |
1271 | |
1272 | sb->s_flags |= SB_ACTIVE; | |
1273 | } | |
1274 | ||
e062abae | 1275 | fc->root = dget(sb->s_root); |
cb50b348 | 1276 | return 0; |
43ce4c1f DH |
1277 | |
1278 | error: | |
1279 | deactivate_locked_super(sb); | |
1280 | return err; | |
cb50b348 | 1281 | } |
cb50b348 | 1282 | |
2ac295d4 AV |
1283 | int get_tree_nodev(struct fs_context *fc, |
1284 | int (*fill_super)(struct super_block *sb, | |
1285 | struct fs_context *fc)) | |
1286 | { | |
e062abae | 1287 | return vfs_get_super(fc, NULL, fill_super); |
2ac295d4 AV |
1288 | } |
1289 | EXPORT_SYMBOL(get_tree_nodev); | |
1290 | ||
c23a0bba AV |
1291 | int get_tree_single(struct fs_context *fc, |
1292 | int (*fill_super)(struct super_block *sb, | |
1293 | struct fs_context *fc)) | |
1294 | { | |
e062abae | 1295 | return vfs_get_super(fc, test_single_super, fill_super); |
c23a0bba AV |
1296 | } |
1297 | EXPORT_SYMBOL(get_tree_single); | |
1298 | ||
533770cc AV |
1299 | int get_tree_keyed(struct fs_context *fc, |
1300 | int (*fill_super)(struct super_block *sb, | |
1301 | struct fs_context *fc), | |
1302 | void *key) | |
1303 | { | |
1304 | fc->s_fs_info = key; | |
e062abae | 1305 | return vfs_get_super(fc, test_keyed_super, fill_super); |
533770cc AV |
1306 | } |
1307 | EXPORT_SYMBOL(get_tree_keyed); | |
1308 | ||
69881be3 CB |
1309 | static int set_bdev_super(struct super_block *s, void *data) |
1310 | { | |
1311 | s->s_dev = *(dev_t *)data; | |
1312 | return 0; | |
1313 | } | |
1314 | ||
1315 | static int super_s_dev_set(struct super_block *s, struct fs_context *fc) | |
1316 | { | |
1317 | return set_bdev_super(s, fc->sget_key); | |
1318 | } | |
1319 | ||
1320 | static int super_s_dev_test(struct super_block *s, struct fs_context *fc) | |
1321 | { | |
1322 | return !(s->s_iflags & SB_I_RETIRED) && | |
1323 | s->s_dev == *(dev_t *)fc->sget_key; | |
1324 | } | |
1325 | ||
1326 | /** | |
1327 | * sget_dev - Find or create a superblock by device number | |
1328 | * @fc: Filesystem context. | |
1329 | * @dev: device number | |
1330 | * | |
1331 | * Find or create a superblock using the provided device number that | |
1332 | * will be stored in fc->sget_key. | |
1333 | * | |
1334 | * If an extant superblock is matched, then that will be returned with | |
1335 | * an elevated reference count that the caller must transfer or discard. | |
1336 | * | |
1337 | * If no match is made, a new superblock will be allocated and basic | |
1338 | * initialisation will be performed (s_type, s_fs_info, s_id, s_dev will | |
1339 | * be set). The superblock will be published and it will be returned in | |
1340 | * a partially constructed state with SB_BORN and SB_ACTIVE as yet | |
1341 | * unset. | |
1342 | * | |
1343 | * Return: an existing or newly created superblock on success, an error | |
1344 | * pointer on failure. | |
1345 | */ | |
1346 | struct super_block *sget_dev(struct fs_context *fc, dev_t dev) | |
1347 | { | |
1348 | fc->sget_key = &dev; | |
1349 | return sget_fc(fc, super_s_dev_test, super_s_dev_set); | |
1350 | } | |
1351 | EXPORT_SYMBOL(sget_dev); | |
1352 | ||
9361401e | 1353 | #ifdef CONFIG_BLOCK |
9c09a7cf | 1354 | /* |
fd146410 JK |
1355 | * Lock the superblock that is holder of the bdev. Returns the superblock |
1356 | * pointer if we successfully locked the superblock and it is alive. Otherwise | |
1357 | * we return NULL and just unlock bdev->bd_holder_lock. | |
9c09a7cf | 1358 | * |
fd146410 | 1359 | * The function must be called with bdev->bd_holder_lock and releases it. |
9c09a7cf | 1360 | */ |
49ef8832 | 1361 | static struct super_block *bdev_super_lock(struct block_device *bdev, bool excl) |
fd146410 | 1362 | __releases(&bdev->bd_holder_lock) |
87efb390 | 1363 | { |
fd146410 | 1364 | struct super_block *sb = bdev->bd_holder; |
f0cd9880 | 1365 | bool locked; |
fd146410 JK |
1366 | |
1367 | lockdep_assert_held(&bdev->bd_holder_lock); | |
1368 | lockdep_assert_not_held(&sb->s_umount); | |
3b224e1d | 1369 | lockdep_assert_not_held(&bdev->bd_disk->open_mutex); |
fd146410 JK |
1370 | |
1371 | /* Make sure sb doesn't go away from under us */ | |
1372 | spin_lock(&sb_lock); | |
1373 | sb->s_count++; | |
1374 | spin_unlock(&sb_lock); | |
49ef8832 | 1375 | |
fd146410 | 1376 | mutex_unlock(&bdev->bd_holder_lock); |
87efb390 | 1377 | |
49ef8832 CB |
1378 | locked = super_lock(sb, excl); |
1379 | ||
fd146410 | 1380 | /* |
49ef8832 CB |
1381 | * If the superblock wasn't already SB_DYING then we hold |
1382 | * s_umount and can safely drop our temporary reference. | |
1383 | */ | |
fd146410 | 1384 | put_super(sb); |
49ef8832 CB |
1385 | |
1386 | if (!locked) | |
1387 | return NULL; | |
1388 | ||
1389 | if (!sb->s_root || !(sb->s_flags & SB_ACTIVE)) { | |
1390 | super_unlock(sb, excl); | |
1391 | return NULL; | |
1392 | } | |
1393 | ||
fd146410 | 1394 | return sb; |
9c09a7cf CH |
1395 | } |
1396 | ||
d8530de5 | 1397 | static void fs_bdev_mark_dead(struct block_device *bdev, bool surprise) |
87efb390 | 1398 | { |
fd146410 | 1399 | struct super_block *sb; |
9c09a7cf | 1400 | |
49ef8832 | 1401 | sb = bdev_super_lock(bdev, false); |
fd146410 | 1402 | if (!sb) |
87efb390 CH |
1403 | return; |
1404 | ||
d8530de5 CH |
1405 | if (!surprise) |
1406 | sync_filesystem(sb); | |
1407 | shrink_dcache_sb(sb); | |
e127b9bc | 1408 | invalidate_inodes(sb); |
87efb390 CH |
1409 | if (sb->s_op->shutdown) |
1410 | sb->s_op->shutdown(sb); | |
9c09a7cf | 1411 | |
0ed33598 | 1412 | super_unlock_shared(sb); |
87efb390 CH |
1413 | } |
1414 | ||
2142b88c CH |
1415 | static void fs_bdev_sync(struct block_device *bdev) |
1416 | { | |
fd146410 | 1417 | struct super_block *sb; |
2142b88c | 1418 | |
49ef8832 | 1419 | sb = bdev_super_lock(bdev, false); |
fd146410 | 1420 | if (!sb) |
2142b88c | 1421 | return; |
49ef8832 | 1422 | |
2142b88c | 1423 | sync_filesystem(sb); |
0ed33598 | 1424 | super_unlock_shared(sb); |
2142b88c CH |
1425 | } |
1426 | ||
49ef8832 CB |
1427 | static struct super_block *get_bdev_super(struct block_device *bdev) |
1428 | { | |
1429 | bool active = false; | |
1430 | struct super_block *sb; | |
1431 | ||
1432 | sb = bdev_super_lock(bdev, true); | |
1433 | if (sb) { | |
1434 | active = atomic_inc_not_zero(&sb->s_active); | |
1435 | super_unlock_excl(sb); | |
1436 | } | |
1437 | if (!active) | |
1438 | return NULL; | |
1439 | return sb; | |
1440 | } | |
1441 | ||
7366f8b6 CB |
1442 | /** |
1443 | * fs_bdev_freeze - freeze owning filesystem of block device | |
1444 | * @bdev: block device | |
1445 | * | |
1446 | * Freeze the filesystem that owns this block device if it is still | |
1447 | * active. | |
1448 | * | |
1449 | * A filesystem that owns multiple block devices may be frozen from each | |
1450 | * block device and won't be unfrozen until all block devices are | |
1451 | * unfrozen. Each block device can only freeze the filesystem once as we | |
1452 | * nest freezes for block devices in the block layer. | |
1453 | * | |
1454 | * Return: If the freeze was successful zero is returned. If the freeze | |
1455 | * failed a negative error code is returned. | |
1456 | */ | |
49ef8832 CB |
1457 | static int fs_bdev_freeze(struct block_device *bdev) |
1458 | { | |
1459 | struct super_block *sb; | |
1460 | int error = 0; | |
1461 | ||
1462 | lockdep_assert_held(&bdev->bd_fsfreeze_mutex); | |
1463 | ||
1464 | sb = get_bdev_super(bdev); | |
1465 | if (!sb) | |
1466 | return -EINVAL; | |
1467 | ||
1468 | if (sb->s_op->freeze_super) | |
7366f8b6 CB |
1469 | error = sb->s_op->freeze_super(sb, |
1470 | FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE); | |
49ef8832 | 1471 | else |
7366f8b6 CB |
1472 | error = freeze_super(sb, |
1473 | FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE); | |
49ef8832 CB |
1474 | if (!error) |
1475 | error = sync_blockdev(bdev); | |
1476 | deactivate_super(sb); | |
1477 | return error; | |
1478 | } | |
1479 | ||
7366f8b6 CB |
1480 | /** |
1481 | * fs_bdev_thaw - thaw owning filesystem of block device | |
1482 | * @bdev: block device | |
1483 | * | |
1484 | * Thaw the filesystem that owns this block device. | |
1485 | * | |
1486 | * A filesystem that owns multiple block devices may be frozen from each | |
1487 | * block device and won't be unfrozen until all block devices are | |
1488 | * unfrozen. Each block device can only freeze the filesystem once as we | |
1489 | * nest freezes for block devices in the block layer. | |
1490 | * | |
1491 | * Return: If the thaw was successful zero is returned. If the thaw | |
1492 | * failed a negative error code is returned. If this function | |
1493 | * returns zero it doesn't mean that the filesystem is unfrozen | |
1494 | * as it may have been frozen multiple times (kernel may hold a | |
1495 | * freeze or might be frozen from other block devices). | |
1496 | */ | |
49ef8832 CB |
1497 | static int fs_bdev_thaw(struct block_device *bdev) |
1498 | { | |
1499 | struct super_block *sb; | |
1500 | int error; | |
1501 | ||
1502 | lockdep_assert_held(&bdev->bd_fsfreeze_mutex); | |
1503 | ||
1504 | sb = get_bdev_super(bdev); | |
1505 | if (WARN_ON_ONCE(!sb)) | |
1506 | return -EINVAL; | |
1507 | ||
1508 | if (sb->s_op->thaw_super) | |
7366f8b6 CB |
1509 | error = sb->s_op->thaw_super(sb, |
1510 | FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE); | |
49ef8832 | 1511 | else |
7366f8b6 CB |
1512 | error = thaw_super(sb, |
1513 | FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE); | |
49ef8832 CB |
1514 | deactivate_super(sb); |
1515 | return error; | |
1516 | } | |
1517 | ||
7ecd0b6f | 1518 | const struct blk_holder_ops fs_holder_ops = { |
d8530de5 | 1519 | .mark_dead = fs_bdev_mark_dead, |
2142b88c | 1520 | .sync = fs_bdev_sync, |
49ef8832 CB |
1521 | .freeze = fs_bdev_freeze, |
1522 | .thaw = fs_bdev_thaw, | |
87efb390 | 1523 | }; |
7ecd0b6f | 1524 | EXPORT_SYMBOL_GPL(fs_holder_ops); |
fe62c3a4 | 1525 | |
cf6da236 | 1526 | int setup_bdev_super(struct super_block *sb, int sb_flags, |
aca740ce JK |
1527 | struct fs_context *fc) |
1528 | { | |
1529 | blk_mode_t mode = sb_open_mode(sb_flags); | |
f3a60882 | 1530 | struct file *bdev_file; |
aca740ce JK |
1531 | struct block_device *bdev; |
1532 | ||
f3a60882 CB |
1533 | bdev_file = bdev_file_open_by_dev(sb->s_dev, mode, sb, &fs_holder_ops); |
1534 | if (IS_ERR(bdev_file)) { | |
aca740ce JK |
1535 | if (fc) |
1536 | errorf(fc, "%s: Can't open blockdev", fc->source); | |
f3a60882 | 1537 | return PTR_ERR(bdev_file); |
aca740ce | 1538 | } |
f3a60882 | 1539 | bdev = file_bdev(bdev_file); |
aca740ce JK |
1540 | |
1541 | /* | |
1542 | * This really should be in blkdev_get_by_dev, but right now can't due | |
1543 | * to legacy issues that require us to allow opening a block device node | |
1544 | * writable from userspace even for a read-only block device. | |
1545 | */ | |
1546 | if ((mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) { | |
22650a99 | 1547 | bdev_fput(bdev_file); |
aca740ce JK |
1548 | return -EACCES; |
1549 | } | |
1550 | ||
1551 | /* | |
49ef8832 CB |
1552 | * It is enough to check bdev was not frozen before we set |
1553 | * s_bdev as freezing will wait until SB_BORN is set. | |
aca740ce | 1554 | */ |
49ef8832 | 1555 | if (atomic_read(&bdev->bd_fsfreeze_count) > 0) { |
aca740ce JK |
1556 | if (fc) |
1557 | warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev); | |
22650a99 | 1558 | bdev_fput(bdev_file); |
aca740ce JK |
1559 | return -EBUSY; |
1560 | } | |
1561 | spin_lock(&sb_lock); | |
f3a60882 | 1562 | sb->s_bdev_file = bdev_file; |
aca740ce JK |
1563 | sb->s_bdev = bdev; |
1564 | sb->s_bdi = bdi_get(bdev->bd_disk->bdi); | |
1565 | if (bdev_stable_writes(bdev)) | |
1566 | sb->s_iflags |= SB_I_STABLE_WRITES; | |
1567 | spin_unlock(&sb_lock); | |
aca740ce JK |
1568 | |
1569 | snprintf(sb->s_id, sizeof(sb->s_id), "%pg", bdev); | |
1720f5dd | 1570 | shrinker_debugfs_rename(sb->s_shrink, "sb-%s:%s", sb->s_type->name, |
aca740ce JK |
1571 | sb->s_id); |
1572 | sb_set_blocksize(sb, block_size(bdev)); | |
1573 | return 0; | |
fe62c3a4 | 1574 | } |
cf6da236 | 1575 | EXPORT_SYMBOL_GPL(setup_bdev_super); |
fe62c3a4 DH |
1576 | |
1577 | /** | |
1578 | * get_tree_bdev - Get a superblock based on a single block device | |
1579 | * @fc: The filesystem context holding the parameters | |
1580 | * @fill_super: Helper to initialise a new superblock | |
1581 | */ | |
1582 | int get_tree_bdev(struct fs_context *fc, | |
1583 | int (*fill_super)(struct super_block *, | |
1584 | struct fs_context *)) | |
1585 | { | |
fe62c3a4 | 1586 | struct super_block *s; |
fe62c3a4 | 1587 | int error = 0; |
aca740ce | 1588 | dev_t dev; |
fe62c3a4 | 1589 | |
fe62c3a4 DH |
1590 | if (!fc->source) |
1591 | return invalf(fc, "No source specified"); | |
1592 | ||
aca740ce JK |
1593 | error = lookup_bdev(fc->source, &dev); |
1594 | if (error) { | |
1595 | errorf(fc, "%s: Can't lookup blockdev", fc->source); | |
1596 | return error; | |
fe62c3a4 DH |
1597 | } |
1598 | ||
1599 | fc->sb_flags |= SB_NOSEC; | |
69881be3 | 1600 | s = sget_dev(fc, dev); |
aca740ce | 1601 | if (IS_ERR(s)) |
fe62c3a4 DH |
1602 | return PTR_ERR(s); |
1603 | ||
1604 | if (s->s_root) { | |
1605 | /* Don't summarily change the RO/RW state. */ | |
1606 | if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) { | |
aca740ce | 1607 | warnf(fc, "%pg: Can't mount, would change RO state", s->s_bdev); |
fe62c3a4 | 1608 | deactivate_locked_super(s); |
fe62c3a4 DH |
1609 | return -EBUSY; |
1610 | } | |
aca740ce | 1611 | } else { |
aca740ce | 1612 | error = setup_bdev_super(s, fc->sb_flags, fc); |
aca740ce JK |
1613 | if (!error) |
1614 | error = fill_super(s, fc); | |
fe62c3a4 DH |
1615 | if (error) { |
1616 | deactivate_locked_super(s); | |
1617 | return error; | |
1618 | } | |
fe62c3a4 | 1619 | s->s_flags |= SB_ACTIVE; |
fe62c3a4 DH |
1620 | } |
1621 | ||
1622 | BUG_ON(fc->root); | |
1623 | fc->root = dget(s->s_root); | |
1624 | return 0; | |
1625 | } | |
1626 | EXPORT_SYMBOL(get_tree_bdev); | |
1627 | ||
1da177e4 LT |
1628 | static int test_bdev_super(struct super_block *s, void *data) |
1629 | { | |
aca740ce | 1630 | return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data; |
1da177e4 LT |
1631 | } |
1632 | ||
152a0836 | 1633 | struct dentry *mount_bdev(struct file_system_type *fs_type, |
1da177e4 | 1634 | int flags, const char *dev_name, void *data, |
152a0836 | 1635 | int (*fill_super)(struct super_block *, void *, int)) |
1da177e4 | 1636 | { |
1da177e4 | 1637 | struct super_block *s; |
aca740ce JK |
1638 | int error; |
1639 | dev_t dev; | |
1da177e4 | 1640 | |
aca740ce JK |
1641 | error = lookup_bdev(dev_name, &dev); |
1642 | if (error) | |
1643 | return ERR_PTR(error); | |
1da177e4 | 1644 | |
aca740ce JK |
1645 | flags |= SB_NOSEC; |
1646 | s = sget(fs_type, test_bdev_super, set_bdev_super, flags, &dev); | |
1da177e4 | 1647 | if (IS_ERR(s)) |
aca740ce | 1648 | return ERR_CAST(s); |
1da177e4 LT |
1649 | |
1650 | if (s->s_root) { | |
e462ec50 | 1651 | if ((flags ^ s->s_flags) & SB_RDONLY) { |
74dbbdd7 | 1652 | deactivate_locked_super(s); |
aca740ce | 1653 | return ERR_PTR(-EBUSY); |
1da177e4 | 1654 | } |
aca740ce | 1655 | } else { |
aca740ce | 1656 | error = setup_bdev_super(s, flags, NULL); |
aca740ce JK |
1657 | if (!error) |
1658 | error = fill_super(s, data, flags & SB_SILENT ? 1 : 0); | |
1da177e4 | 1659 | if (error) { |
74dbbdd7 | 1660 | deactivate_locked_super(s); |
aca740ce | 1661 | return ERR_PTR(error); |
fa675765 | 1662 | } |
454e2398 | 1663 | |
e462ec50 | 1664 | s->s_flags |= SB_ACTIVE; |
1da177e4 LT |
1665 | } |
1666 | ||
152a0836 | 1667 | return dget(s->s_root); |
152a0836 AV |
1668 | } |
1669 | EXPORT_SYMBOL(mount_bdev); | |
1670 | ||
1da177e4 LT |
1671 | void kill_block_super(struct super_block *sb) |
1672 | { | |
1673 | struct block_device *bdev = sb->s_bdev; | |
1674 | ||
1da177e4 | 1675 | generic_shutdown_super(sb); |
aca740ce JK |
1676 | if (bdev) { |
1677 | sync_blockdev(bdev); | |
22650a99 | 1678 | bdev_fput(sb->s_bdev_file); |
aca740ce | 1679 | } |
1da177e4 LT |
1680 | } |
1681 | ||
1682 | EXPORT_SYMBOL(kill_block_super); | |
9361401e | 1683 | #endif |
1da177e4 | 1684 | |
3c26ff6e | 1685 | struct dentry *mount_nodev(struct file_system_type *fs_type, |
1da177e4 | 1686 | int flags, void *data, |
3c26ff6e | 1687 | int (*fill_super)(struct super_block *, void *, int)) |
1da177e4 LT |
1688 | { |
1689 | int error; | |
9249e17f | 1690 | struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL); |
1da177e4 LT |
1691 | |
1692 | if (IS_ERR(s)) | |
3c26ff6e | 1693 | return ERR_CAST(s); |
1da177e4 | 1694 | |
e462ec50 | 1695 | error = fill_super(s, data, flags & SB_SILENT ? 1 : 0); |
1da177e4 | 1696 | if (error) { |
74dbbdd7 | 1697 | deactivate_locked_super(s); |
3c26ff6e | 1698 | return ERR_PTR(error); |
1da177e4 | 1699 | } |
e462ec50 | 1700 | s->s_flags |= SB_ACTIVE; |
3c26ff6e | 1701 | return dget(s->s_root); |
1da177e4 | 1702 | } |
3c26ff6e AV |
1703 | EXPORT_SYMBOL(mount_nodev); |
1704 | ||
a6097180 N |
1705 | int reconfigure_single(struct super_block *s, |
1706 | int flags, void *data) | |
8d0347f6 DH |
1707 | { |
1708 | struct fs_context *fc; | |
1709 | int ret; | |
1710 | ||
1711 | /* The caller really need to be passing fc down into mount_single(), | |
1712 | * then a chunk of this can be removed. [Bollocks -- AV] | |
1713 | * Better yet, reconfiguration shouldn't happen, but rather the second | |
1714 | * mount should be rejected if the parameters are not compatible. | |
1715 | */ | |
1716 | fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK); | |
1717 | if (IS_ERR(fc)) | |
1718 | return PTR_ERR(fc); | |
1719 | ||
1720 | ret = parse_monolithic_mount_data(fc, data); | |
1721 | if (ret < 0) | |
1722 | goto out; | |
1723 | ||
1724 | ret = reconfigure_super(fc); | |
1725 | out: | |
1726 | put_fs_context(fc); | |
1727 | return ret; | |
1728 | } | |
1729 | ||
1da177e4 LT |
1730 | static int compare_single(struct super_block *s, void *p) |
1731 | { | |
1732 | return 1; | |
1733 | } | |
1734 | ||
fc14f2fe | 1735 | struct dentry *mount_single(struct file_system_type *fs_type, |
1da177e4 | 1736 | int flags, void *data, |
fc14f2fe | 1737 | int (*fill_super)(struct super_block *, void *, int)) |
1da177e4 LT |
1738 | { |
1739 | struct super_block *s; | |
1740 | int error; | |
1741 | ||
9249e17f | 1742 | s = sget(fs_type, compare_single, set_anon_super, flags, NULL); |
1da177e4 | 1743 | if (IS_ERR(s)) |
fc14f2fe | 1744 | return ERR_CAST(s); |
1da177e4 | 1745 | if (!s->s_root) { |
e462ec50 | 1746 | error = fill_super(s, data, flags & SB_SILENT ? 1 : 0); |
8d0347f6 DH |
1747 | if (!error) |
1748 | s->s_flags |= SB_ACTIVE; | |
9329d1be | 1749 | } else { |
8d0347f6 DH |
1750 | error = reconfigure_single(s, flags, data); |
1751 | } | |
1752 | if (unlikely(error)) { | |
1753 | deactivate_locked_super(s); | |
1754 | return ERR_PTR(error); | |
1da177e4 | 1755 | } |
fc14f2fe AV |
1756 | return dget(s->s_root); |
1757 | } | |
1758 | EXPORT_SYMBOL(mount_single); | |
1759 | ||
9bc61ab1 DH |
1760 | /** |
1761 | * vfs_get_tree - Get the mountable root | |
1762 | * @fc: The superblock configuration context. | |
1763 | * | |
1764 | * The filesystem is invoked to get or create a superblock which can then later | |
1765 | * be used for mounting. The filesystem places a pointer to the root to be | |
1766 | * used for mounting in @fc->root. | |
1767 | */ | |
1768 | int vfs_get_tree(struct fs_context *fc) | |
1da177e4 | 1769 | { |
9d412a43 | 1770 | struct super_block *sb; |
9bc61ab1 | 1771 | int error; |
8089352a | 1772 | |
f3a09c92 AV |
1773 | if (fc->root) |
1774 | return -EBUSY; | |
1775 | ||
1776 | /* Get the mountable root in fc->root, with a ref on the root and a ref | |
1777 | * on the superblock. | |
1778 | */ | |
1779 | error = fc->ops->get_tree(fc); | |
9bc61ab1 DH |
1780 | if (error < 0) |
1781 | return error; | |
1da177e4 | 1782 | |
f3a09c92 AV |
1783 | if (!fc->root) { |
1784 | pr_err("Filesystem %s get_tree() didn't set fc->root\n", | |
1785 | fc->fs_type->name); | |
1786 | /* We don't know what the locking state of the superblock is - | |
1787 | * if there is a superblock. | |
1788 | */ | |
1789 | BUG(); | |
1790 | } | |
1791 | ||
9bc61ab1 | 1792 | sb = fc->root->d_sb; |
9d412a43 | 1793 | WARN_ON(!sb->s_bdi); |
79f546a6 DC |
1794 | |
1795 | /* | |
5e874914 CB |
1796 | * super_wake() contains a memory barrier which also care of |
1797 | * ordering for super_cache_count(). We place it before setting | |
1798 | * SB_BORN as the data dependency between the two functions is | |
1799 | * the superblock structure contents that we just set up, not | |
1800 | * the SB_BORN flag. | |
79f546a6 | 1801 | */ |
5e874914 | 1802 | super_wake(sb, SB_BORN); |
454e2398 | 1803 | |
9bc61ab1 | 1804 | error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL); |
c9ce29ed AV |
1805 | if (unlikely(error)) { |
1806 | fc_drop_locked(fc); | |
1807 | return error; | |
a10d7c22 AV |
1808 | } |
1809 | ||
42cb56ae JL |
1810 | /* |
1811 | * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE | |
1812 | * but s_maxbytes was an unsigned long long for many releases. Throw | |
1813 | * this warning for a little while to try and catch filesystems that | |
4358b567 | 1814 | * violate this rule. |
42cb56ae | 1815 | */ |
9d412a43 | 1816 | WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to " |
9bc61ab1 | 1817 | "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes); |
42cb56ae | 1818 | |
9bc61ab1 | 1819 | return 0; |
1da177e4 | 1820 | } |
9bc61ab1 | 1821 | EXPORT_SYMBOL(vfs_get_tree); |
1da177e4 | 1822 | |
fca39346 JK |
1823 | /* |
1824 | * Setup private BDI for given superblock. It gets automatically cleaned up | |
1825 | * in generic_shutdown_super(). | |
1826 | */ | |
1827 | int super_setup_bdi_name(struct super_block *sb, char *fmt, ...) | |
1828 | { | |
1829 | struct backing_dev_info *bdi; | |
1830 | int err; | |
1831 | va_list args; | |
1832 | ||
aef33c2f | 1833 | bdi = bdi_alloc(NUMA_NO_NODE); |
fca39346 JK |
1834 | if (!bdi) |
1835 | return -ENOMEM; | |
1836 | ||
fca39346 | 1837 | va_start(args, fmt); |
7c4cc300 | 1838 | err = bdi_register_va(bdi, fmt, args); |
fca39346 JK |
1839 | va_end(args); |
1840 | if (err) { | |
1841 | bdi_put(bdi); | |
1842 | return err; | |
1843 | } | |
1844 | WARN_ON(sb->s_bdi != &noop_backing_dev_info); | |
1845 | sb->s_bdi = bdi; | |
0b3ea092 | 1846 | sb->s_iflags |= SB_I_PERSB_BDI; |
fca39346 JK |
1847 | |
1848 | return 0; | |
1849 | } | |
1850 | EXPORT_SYMBOL(super_setup_bdi_name); | |
1851 | ||
1852 | /* | |
1853 | * Setup private BDI for given superblock. I gets automatically cleaned up | |
1854 | * in generic_shutdown_super(). | |
1855 | */ | |
1856 | int super_setup_bdi(struct super_block *sb) | |
1857 | { | |
1858 | static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); | |
1859 | ||
1860 | return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name, | |
1861 | atomic_long_inc_return(&bdi_seq)); | |
1862 | } | |
1863 | EXPORT_SYMBOL(super_setup_bdi); | |
1864 | ||
5accdf82 JK |
1865 | /** |
1866 | * sb_wait_write - wait until all writers to given file system finish | |
1867 | * @sb: the super for which we wait | |
1868 | * @level: type of writers we wait for (normal vs page fault) | |
1869 | * | |
1870 | * This function waits until there are no writers of given type to given file | |
8129ed29 | 1871 | * system. |
5accdf82 JK |
1872 | */ |
1873 | static void sb_wait_write(struct super_block *sb, int level) | |
1874 | { | |
8129ed29 | 1875 | percpu_down_write(sb->s_writers.rw_sem + level-1); |
8129ed29 | 1876 | } |
5accdf82 | 1877 | |
f1a96220 ON |
1878 | /* |
1879 | * We are going to return to userspace and forget about these locks, the | |
1880 | * ownership goes to the caller of thaw_super() which does unlock(). | |
1881 | */ | |
1882 | static void lockdep_sb_freeze_release(struct super_block *sb) | |
1883 | { | |
1884 | int level; | |
1885 | ||
1886 | for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--) | |
1887 | percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_); | |
1888 | } | |
1889 | ||
1890 | /* | |
1891 | * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb). | |
1892 | */ | |
1893 | static void lockdep_sb_freeze_acquire(struct super_block *sb) | |
8129ed29 ON |
1894 | { |
1895 | int level; | |
5accdf82 | 1896 | |
8129ed29 ON |
1897 | for (level = 0; level < SB_FREEZE_LEVELS; ++level) |
1898 | percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_); | |
f1a96220 ON |
1899 | } |
1900 | ||
2719c716 | 1901 | static void sb_freeze_unlock(struct super_block *sb, int level) |
f1a96220 | 1902 | { |
2719c716 | 1903 | for (level--; level >= 0; level--) |
8129ed29 | 1904 | percpu_up_write(sb->s_writers.rw_sem + level); |
5accdf82 JK |
1905 | } |
1906 | ||
59ba4fdd DW |
1907 | static int wait_for_partially_frozen(struct super_block *sb) |
1908 | { | |
1909 | int ret = 0; | |
1910 | ||
1911 | do { | |
1912 | unsigned short old = sb->s_writers.frozen; | |
1913 | ||
1914 | up_write(&sb->s_umount); | |
1915 | ret = wait_var_event_killable(&sb->s_writers.frozen, | |
1916 | sb->s_writers.frozen != old); | |
1917 | down_write(&sb->s_umount); | |
1918 | } while (ret == 0 && | |
1919 | sb->s_writers.frozen != SB_UNFROZEN && | |
1920 | sb->s_writers.frozen != SB_FREEZE_COMPLETE); | |
1921 | ||
1922 | return ret; | |
1923 | } | |
1924 | ||
7366f8b6 CB |
1925 | #define FREEZE_HOLDERS (FREEZE_HOLDER_KERNEL | FREEZE_HOLDER_USERSPACE) |
1926 | #define FREEZE_FLAGS (FREEZE_HOLDERS | FREEZE_MAY_NEST) | |
1927 | ||
1928 | static inline int freeze_inc(struct super_block *sb, enum freeze_holder who) | |
1929 | { | |
1930 | WARN_ON_ONCE((who & ~FREEZE_FLAGS)); | |
1931 | WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1); | |
1932 | ||
1933 | if (who & FREEZE_HOLDER_KERNEL) | |
1934 | ++sb->s_writers.freeze_kcount; | |
1935 | if (who & FREEZE_HOLDER_USERSPACE) | |
1936 | ++sb->s_writers.freeze_ucount; | |
1937 | return sb->s_writers.freeze_kcount + sb->s_writers.freeze_ucount; | |
1938 | } | |
1939 | ||
1940 | static inline int freeze_dec(struct super_block *sb, enum freeze_holder who) | |
1941 | { | |
1942 | WARN_ON_ONCE((who & ~FREEZE_FLAGS)); | |
1943 | WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1); | |
1944 | ||
1945 | if ((who & FREEZE_HOLDER_KERNEL) && sb->s_writers.freeze_kcount) | |
1946 | --sb->s_writers.freeze_kcount; | |
1947 | if ((who & FREEZE_HOLDER_USERSPACE) && sb->s_writers.freeze_ucount) | |
1948 | --sb->s_writers.freeze_ucount; | |
1949 | return sb->s_writers.freeze_kcount + sb->s_writers.freeze_ucount; | |
1950 | } | |
1951 | ||
1952 | static inline bool may_freeze(struct super_block *sb, enum freeze_holder who) | |
1953 | { | |
1954 | WARN_ON_ONCE((who & ~FREEZE_FLAGS)); | |
1955 | WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1); | |
1956 | ||
1957 | if (who & FREEZE_HOLDER_KERNEL) | |
1958 | return (who & FREEZE_MAY_NEST) || | |
1959 | sb->s_writers.freeze_kcount == 0; | |
1960 | if (who & FREEZE_HOLDER_USERSPACE) | |
1961 | return (who & FREEZE_MAY_NEST) || | |
1962 | sb->s_writers.freeze_ucount == 0; | |
1963 | return false; | |
1964 | } | |
1965 | ||
18e9e510 | 1966 | /** |
7000d3c4 RD |
1967 | * freeze_super - lock the filesystem and force it into a consistent state |
1968 | * @sb: the super to lock | |
880b9577 | 1969 | * @who: context that wants to freeze |
18e9e510 JB |
1970 | * |
1971 | * Syncs the super to make sure the filesystem is consistent and calls the fs's | |
880b9577 | 1972 | * freeze_fs. Subsequent calls to this without first thawing the fs may return |
18e9e510 | 1973 | * -EBUSY. |
5accdf82 | 1974 | * |
880b9577 DW |
1975 | * @who should be: |
1976 | * * %FREEZE_HOLDER_USERSPACE if userspace wants to freeze the fs; | |
1977 | * * %FREEZE_HOLDER_KERNEL if the kernel wants to freeze the fs. | |
7366f8b6 | 1978 | * * %FREEZE_MAY_NEST whether nesting freeze and thaw requests is allowed. |
880b9577 DW |
1979 | * |
1980 | * The @who argument distinguishes between the kernel and userspace trying to | |
1981 | * freeze the filesystem. Although there cannot be multiple kernel freezes or | |
1982 | * multiple userspace freezes in effect at any given time, the kernel and | |
1983 | * userspace can both hold a filesystem frozen. The filesystem remains frozen | |
1984 | * until there are no kernel or userspace freezes in effect. | |
1985 | * | |
7366f8b6 CB |
1986 | * A filesystem may hold multiple devices and thus a filesystems may be |
1987 | * frozen through the block layer via multiple block devices. In this | |
1988 | * case the request is marked as being allowed to nest by passing | |
1989 | * FREEZE_MAY_NEST. The filesystem remains frozen until all block | |
1990 | * devices are unfrozen. If multiple freezes are attempted without | |
1991 | * FREEZE_MAY_NEST -EBUSY will be returned. | |
1992 | * | |
5accdf82 JK |
1993 | * During this function, sb->s_writers.frozen goes through these values: |
1994 | * | |
1995 | * SB_UNFROZEN: File system is normal, all writes progress as usual. | |
1996 | * | |
1997 | * SB_FREEZE_WRITE: The file system is in the process of being frozen. New | |
1998 | * writes should be blocked, though page faults are still allowed. We wait for | |
1999 | * all writes to complete and then proceed to the next stage. | |
2000 | * | |
2001 | * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked | |
2002 | * but internal fs threads can still modify the filesystem (although they | |
2003 | * should not dirty new pages or inodes), writeback can run etc. After waiting | |
2004 | * for all running page faults we sync the filesystem which will clean all | |
2005 | * dirty pages and inodes (no new dirty pages or inodes can be created when | |
2006 | * sync is running). | |
2007 | * | |
2008 | * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs | |
2009 | * modification are blocked (e.g. XFS preallocation truncation on inode | |
2010 | * reclaim). This is usually implemented by blocking new transactions for | |
2011 | * filesystems that have them and need this additional guard. After all | |
2012 | * internal writers are finished we call ->freeze_fs() to finish filesystem | |
2013 | * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is | |
2014 | * mostly auxiliary for filesystems to verify they do not modify frozen fs. | |
2015 | * | |
2016 | * sb->s_writers.frozen is protected by sb->s_umount. | |
7366f8b6 CB |
2017 | * |
2018 | * Return: If the freeze was successful zero is returned. If the freeze | |
2019 | * failed a negative error code is returned. | |
18e9e510 | 2020 | */ |
880b9577 | 2021 | int freeze_super(struct super_block *sb, enum freeze_holder who) |
18e9e510 JB |
2022 | { |
2023 | int ret; | |
2024 | ||
f0cd9880 CB |
2025 | if (!super_lock_excl(sb)) { |
2026 | WARN_ON_ONCE("Dying superblock while freezing!"); | |
2027 | return -EINVAL; | |
2028 | } | |
18e9e510 | 2029 | atomic_inc(&sb->s_active); |
051178c3 | 2030 | |
59ba4fdd | 2031 | retry: |
880b9577 | 2032 | if (sb->s_writers.frozen == SB_FREEZE_COMPLETE) { |
7366f8b6 CB |
2033 | if (may_freeze(sb, who)) |
2034 | ret = !!WARN_ON_ONCE(freeze_inc(sb, who) == 1); | |
2035 | else | |
2036 | ret = -EBUSY; | |
2037 | /* All freezers share a single active reference. */ | |
2038 | deactivate_locked_super(sb); | |
2039 | return ret; | |
880b9577 DW |
2040 | } |
2041 | ||
5accdf82 | 2042 | if (sb->s_writers.frozen != SB_UNFROZEN) { |
59ba4fdd DW |
2043 | ret = wait_for_partially_frozen(sb); |
2044 | if (ret) { | |
2045 | deactivate_locked_super(sb); | |
2046 | return ret; | |
2047 | } | |
2048 | ||
2049 | goto retry; | |
18e9e510 JB |
2050 | } |
2051 | ||
bc98a42c | 2052 | if (sb_rdonly(sb)) { |
5accdf82 | 2053 | /* Nothing to do really... */ |
7366f8b6 | 2054 | WARN_ON_ONCE(freeze_inc(sb, who) > 1); |
5accdf82 | 2055 | sb->s_writers.frozen = SB_FREEZE_COMPLETE; |
59ba4fdd | 2056 | wake_up_var(&sb->s_writers.frozen); |
0ed33598 | 2057 | super_unlock_excl(sb); |
18e9e510 JB |
2058 | return 0; |
2059 | } | |
2060 | ||
5accdf82 | 2061 | sb->s_writers.frozen = SB_FREEZE_WRITE; |
5accdf82 | 2062 | /* Release s_umount to preserve sb_start_write -> s_umount ordering */ |
0ed33598 | 2063 | super_unlock_excl(sb); |
5accdf82 | 2064 | sb_wait_write(sb, SB_FREEZE_WRITE); |
63513f85 | 2065 | __super_lock_excl(sb); |
5accdf82 JK |
2066 | |
2067 | /* Now we go and block page faults... */ | |
5accdf82 | 2068 | sb->s_writers.frozen = SB_FREEZE_PAGEFAULT; |
5accdf82 JK |
2069 | sb_wait_write(sb, SB_FREEZE_PAGEFAULT); |
2070 | ||
2071 | /* All writers are done so after syncing there won't be dirty data */ | |
2719c716 DW |
2072 | ret = sync_filesystem(sb); |
2073 | if (ret) { | |
2074 | sb->s_writers.frozen = SB_UNFROZEN; | |
2075 | sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT); | |
59ba4fdd | 2076 | wake_up_var(&sb->s_writers.frozen); |
2719c716 DW |
2077 | deactivate_locked_super(sb); |
2078 | return ret; | |
2079 | } | |
18e9e510 | 2080 | |
5accdf82 JK |
2081 | /* Now wait for internal filesystem counter */ |
2082 | sb->s_writers.frozen = SB_FREEZE_FS; | |
5accdf82 | 2083 | sb_wait_write(sb, SB_FREEZE_FS); |
18e9e510 | 2084 | |
18e9e510 JB |
2085 | if (sb->s_op->freeze_fs) { |
2086 | ret = sb->s_op->freeze_fs(sb); | |
2087 | if (ret) { | |
2088 | printk(KERN_ERR | |
2089 | "VFS:Filesystem freeze failed\n"); | |
5accdf82 | 2090 | sb->s_writers.frozen = SB_UNFROZEN; |
2719c716 | 2091 | sb_freeze_unlock(sb, SB_FREEZE_FS); |
59ba4fdd | 2092 | wake_up_var(&sb->s_writers.frozen); |
18e9e510 JB |
2093 | deactivate_locked_super(sb); |
2094 | return ret; | |
2095 | } | |
2096 | } | |
5accdf82 | 2097 | /* |
89f39af1 ON |
2098 | * For debugging purposes so that fs can warn if it sees write activity |
2099 | * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super(). | |
5accdf82 | 2100 | */ |
7366f8b6 | 2101 | WARN_ON_ONCE(freeze_inc(sb, who) > 1); |
5accdf82 | 2102 | sb->s_writers.frozen = SB_FREEZE_COMPLETE; |
59ba4fdd | 2103 | wake_up_var(&sb->s_writers.frozen); |
f1a96220 | 2104 | lockdep_sb_freeze_release(sb); |
0ed33598 | 2105 | super_unlock_excl(sb); |
18e9e510 JB |
2106 | return 0; |
2107 | } | |
2108 | EXPORT_SYMBOL(freeze_super); | |
2109 | ||
880b9577 DW |
2110 | /* |
2111 | * Undoes the effect of a freeze_super_locked call. If the filesystem is | |
2112 | * frozen both by userspace and the kernel, a thaw call from either source | |
2113 | * removes that state without releasing the other state or unlocking the | |
2114 | * filesystem. | |
2115 | */ | |
2116 | static int thaw_super_locked(struct super_block *sb, enum freeze_holder who) | |
18e9e510 | 2117 | { |
24c372d5 | 2118 | int error = -EINVAL; |
18e9e510 | 2119 | |
24c372d5 CH |
2120 | if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) |
2121 | goto out_unlock; | |
880b9577 | 2122 | |
24c372d5 | 2123 | /* |
7366f8b6 CB |
2124 | * All freezers share a single active reference. |
2125 | * So just unlock in case there are any left. | |
24c372d5 | 2126 | */ |
7366f8b6 CB |
2127 | if (freeze_dec(sb, who)) |
2128 | goto out_unlock; | |
18e9e510 | 2129 | |
bc98a42c | 2130 | if (sb_rdonly(sb)) { |
8129ed29 | 2131 | sb->s_writers.frozen = SB_UNFROZEN; |
59ba4fdd | 2132 | wake_up_var(&sb->s_writers.frozen); |
24c372d5 | 2133 | goto out_deactivate; |
8129ed29 | 2134 | } |
18e9e510 | 2135 | |
f1a96220 ON |
2136 | lockdep_sb_freeze_acquire(sb); |
2137 | ||
18e9e510 JB |
2138 | if (sb->s_op->unfreeze_fs) { |
2139 | error = sb->s_op->unfreeze_fs(sb); | |
2140 | if (error) { | |
7366f8b6 CB |
2141 | pr_err("VFS: Filesystem thaw failed\n"); |
2142 | freeze_inc(sb, who); | |
f1a96220 | 2143 | lockdep_sb_freeze_release(sb); |
24c372d5 | 2144 | goto out_unlock; |
18e9e510 JB |
2145 | } |
2146 | } | |
2147 | ||
5accdf82 | 2148 | sb->s_writers.frozen = SB_UNFROZEN; |
59ba4fdd | 2149 | wake_up_var(&sb->s_writers.frozen); |
2719c716 | 2150 | sb_freeze_unlock(sb, SB_FREEZE_FS); |
24c372d5 | 2151 | out_deactivate: |
18e9e510 | 2152 | deactivate_locked_super(sb); |
18e9e510 | 2153 | return 0; |
24c372d5 CH |
2154 | |
2155 | out_unlock: | |
2156 | super_unlock_excl(sb); | |
2157 | return error; | |
18e9e510 | 2158 | } |
08fdc8a0 | 2159 | |
961f3c89 MCC |
2160 | /** |
2161 | * thaw_super -- unlock filesystem | |
2162 | * @sb: the super to thaw | |
880b9577 DW |
2163 | * @who: context that wants to freeze |
2164 | * | |
2165 | * Unlocks the filesystem and marks it writeable again after freeze_super() | |
2166 | * if there are no remaining freezes on the filesystem. | |
961f3c89 | 2167 | * |
880b9577 DW |
2168 | * @who should be: |
2169 | * * %FREEZE_HOLDER_USERSPACE if userspace wants to thaw the fs; | |
2170 | * * %FREEZE_HOLDER_KERNEL if the kernel wants to thaw the fs. | |
7366f8b6 CB |
2171 | * * %FREEZE_MAY_NEST whether nesting freeze and thaw requests is allowed |
2172 | * | |
2173 | * A filesystem may hold multiple devices and thus a filesystems may | |
2174 | * have been frozen through the block layer via multiple block devices. | |
2175 | * The filesystem remains frozen until all block devices are unfrozen. | |
961f3c89 | 2176 | */ |
880b9577 | 2177 | int thaw_super(struct super_block *sb, enum freeze_holder who) |
08fdc8a0 | 2178 | { |
f0cd9880 CB |
2179 | if (!super_lock_excl(sb)) { |
2180 | WARN_ON_ONCE("Dying superblock while thawing!"); | |
2181 | return -EINVAL; | |
2182 | } | |
880b9577 | 2183 | return thaw_super_locked(sb, who); |
08fdc8a0 | 2184 | } |
18e9e510 | 2185 | EXPORT_SYMBOL(thaw_super); |
439bc39b CH |
2186 | |
2187 | /* | |
2188 | * Create workqueue for deferred direct IO completions. We allocate the | |
2189 | * workqueue when it's first needed. This avoids creating workqueue for | |
2190 | * filesystems that don't need it and also allows us to create the workqueue | |
2191 | * late enough so the we can include s_id in the name of the workqueue. | |
2192 | */ | |
2193 | int sb_init_dio_done_wq(struct super_block *sb) | |
2194 | { | |
2195 | struct workqueue_struct *old; | |
2196 | struct workqueue_struct *wq = alloc_workqueue("dio/%s", | |
2197 | WQ_MEM_RECLAIM, 0, | |
2198 | sb->s_id); | |
2199 | if (!wq) | |
2200 | return -ENOMEM; | |
2201 | /* | |
2202 | * This has to be atomic as more DIOs can race to create the workqueue | |
2203 | */ | |
2204 | old = cmpxchg(&sb->s_dio_done_wq, NULL, wq); | |
2205 | /* Someone created workqueue before us? Free ours... */ | |
2206 | if (old) | |
2207 | destroy_workqueue(wq); | |
2208 | return 0; | |
2209 | } | |
389a4a4a | 2210 | EXPORT_SYMBOL_GPL(sb_init_dio_done_wq); |