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 (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 | 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 | ||
5e874914 CB |
84 | static 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 | * | |
f0cd9880 CB |
108 | * Return: The function returns true if SB_BORN was set and with |
109 | * s_umount held. The function returns false if SB_DYING was | |
110 | * set and without s_umount held. | |
5e874914 CB |
111 | */ |
112 | static __must_check bool super_lock(struct super_block *sb, bool excl) | |
113 | { | |
114 | ||
115 | lockdep_assert_not_held(&sb->s_umount); | |
116 | ||
117 | relock: | |
118 | __super_lock(sb, excl); | |
119 | ||
120 | /* | |
121 | * Has gone through generic_shutdown_super() in the meantime. | |
122 | * @sb->s_root is NULL and @sb->s_active is 0. No one needs to | |
123 | * grab a reference to this. Tell them so. | |
124 | */ | |
f0cd9880 CB |
125 | if (sb->s_flags & SB_DYING) { |
126 | super_unlock(sb, excl); | |
5e874914 | 127 | return false; |
f0cd9880 | 128 | } |
5e874914 CB |
129 | |
130 | /* Has called ->get_tree() successfully. */ | |
131 | if (sb->s_flags & SB_BORN) | |
132 | return true; | |
133 | ||
134 | super_unlock(sb, excl); | |
135 | ||
136 | /* wait until the superblock is ready or dying */ | |
137 | wait_var_event(&sb->s_flags, wait_born(sb)); | |
138 | ||
139 | /* | |
140 | * Neither SB_BORN nor SB_DYING are ever unset so we never loop. | |
141 | * Just reacquire @sb->s_umount for the caller. | |
142 | */ | |
143 | goto relock; | |
144 | } | |
145 | ||
f0cd9880 | 146 | /* wait and try to acquire read-side of @sb->s_umount */ |
5e874914 CB |
147 | static inline bool super_lock_shared(struct super_block *sb) |
148 | { | |
149 | return super_lock(sb, false); | |
150 | } | |
151 | ||
f0cd9880 | 152 | /* wait and try to acquire write-side of @sb->s_umount */ |
5e874914 CB |
153 | static inline bool super_lock_excl(struct super_block *sb) |
154 | { | |
155 | return super_lock(sb, true); | |
156 | } | |
157 | ||
158 | /* wake waiters */ | |
2c18a63b | 159 | #define SUPER_WAKE_FLAGS (SB_BORN | SB_DYING | SB_DEAD) |
5e874914 CB |
160 | static void super_wake(struct super_block *sb, unsigned int flag) |
161 | { | |
162 | WARN_ON_ONCE((flag & ~SUPER_WAKE_FLAGS)); | |
163 | WARN_ON_ONCE(hweight32(flag & SUPER_WAKE_FLAGS) > 1); | |
164 | ||
165 | /* | |
166 | * Pairs with smp_load_acquire() in super_lock() to make sure | |
167 | * all initializations in the superblock are seen by the user | |
168 | * seeing SB_BORN sent. | |
169 | */ | |
170 | smp_store_release(&sb->s_flags, sb->s_flags | flag); | |
171 | /* | |
172 | * Pairs with the barrier in prepare_to_wait_event() to make sure | |
173 | * ___wait_var_event() either sees SB_BORN set or | |
174 | * waitqueue_active() check in wake_up_var() sees the waiter. | |
175 | */ | |
176 | smp_mb(); | |
177 | wake_up_var(&sb->s_flags); | |
178 | } | |
179 | ||
b0d40c92 DC |
180 | /* |
181 | * One thing we have to be careful of with a per-sb shrinker is that we don't | |
182 | * drop the last active reference to the superblock from within the shrinker. | |
183 | * If that happens we could trigger unregistering the shrinker from within the | |
8a0e8bb1 | 184 | * shrinker path and that leads to deadlock on the shrinker_mutex. Hence we |
b0d40c92 DC |
185 | * take a passive reference to the superblock to avoid this from occurring. |
186 | */ | |
0a234c6d DC |
187 | static unsigned long super_cache_scan(struct shrinker *shrink, |
188 | struct shrink_control *sc) | |
b0d40c92 DC |
189 | { |
190 | struct super_block *sb; | |
0a234c6d DC |
191 | long fs_objects = 0; |
192 | long total_objects; | |
193 | long freed = 0; | |
194 | long dentries; | |
195 | long inodes; | |
b0d40c92 | 196 | |
1720f5dd | 197 | sb = shrink->private_data; |
b0d40c92 DC |
198 | |
199 | /* | |
200 | * Deadlock avoidance. We may hold various FS locks, and we don't want | |
201 | * to recurse into the FS that called us in clear_inode() and friends.. | |
202 | */ | |
0a234c6d DC |
203 | if (!(sc->gfp_mask & __GFP_FS)) |
204 | return SHRINK_STOP; | |
b0d40c92 | 205 | |
d8ce82ef | 206 | if (!super_trylock_shared(sb)) |
0a234c6d | 207 | return SHRINK_STOP; |
b0d40c92 | 208 | |
d0407903 | 209 | if (sb->s_op->nr_cached_objects) |
4101b624 | 210 | fs_objects = sb->s_op->nr_cached_objects(sb, sc); |
0e1fdafd | 211 | |
503c358c VD |
212 | inodes = list_lru_shrink_count(&sb->s_inode_lru, sc); |
213 | dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc); | |
f6041567 | 214 | total_objects = dentries + inodes + fs_objects + 1; |
475d0db7 TH |
215 | if (!total_objects) |
216 | total_objects = 1; | |
0e1fdafd | 217 | |
0a234c6d | 218 | /* proportion the scan between the caches */ |
f6041567 | 219 | dentries = mult_frac(sc->nr_to_scan, dentries, total_objects); |
bc3b14cb | 220 | inodes = mult_frac(sc->nr_to_scan, inodes, total_objects); |
503c358c | 221 | fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects); |
b0d40c92 | 222 | |
0a234c6d DC |
223 | /* |
224 | * prune the dcache first as the icache is pinned by it, then | |
225 | * prune the icache, followed by the filesystem specific caches | |
49e7e7ff VD |
226 | * |
227 | * Ensure that we always scan at least one object - memcg kmem | |
228 | * accounting uses this to fully empty the caches. | |
0a234c6d | 229 | */ |
49e7e7ff | 230 | sc->nr_to_scan = dentries + 1; |
503c358c | 231 | freed = prune_dcache_sb(sb, sc); |
49e7e7ff | 232 | sc->nr_to_scan = inodes + 1; |
503c358c | 233 | freed += prune_icache_sb(sb, sc); |
0a234c6d DC |
234 | |
235 | if (fs_objects) { | |
49e7e7ff | 236 | sc->nr_to_scan = fs_objects + 1; |
4101b624 | 237 | freed += sb->s_op->free_cached_objects(sb, sc); |
b0d40c92 DC |
238 | } |
239 | ||
0ed33598 | 240 | super_unlock_shared(sb); |
0a234c6d DC |
241 | return freed; |
242 | } | |
243 | ||
244 | static unsigned long super_cache_count(struct shrinker *shrink, | |
245 | struct shrink_control *sc) | |
246 | { | |
247 | struct super_block *sb; | |
248 | long total_objects = 0; | |
249 | ||
1720f5dd | 250 | sb = shrink->private_data; |
0a234c6d | 251 | |
d23da150 | 252 | /* |
d8ce82ef CB |
253 | * We don't call super_trylock_shared() here as it is a scalability |
254 | * bottleneck, so we're exposed to partial setup state. The shrinker | |
255 | * rwsem does not protect filesystem operations backing | |
256 | * list_lru_shrink_count() or s_op->nr_cached_objects(). Counts can | |
257 | * change between super_cache_count and super_cache_scan, so we really | |
258 | * don't need locks here. | |
79f546a6 DC |
259 | * |
260 | * However, if we are currently mounting the superblock, the underlying | |
261 | * filesystem might be in a state of partial construction and hence it | |
d8ce82ef CB |
262 | * is dangerous to access it. super_trylock_shared() uses a SB_BORN check |
263 | * to avoid this situation, so do the same here. The memory barrier is | |
79f546a6 | 264 | * matched with the one in mount_fs() as we don't hold locks here. |
d23da150 | 265 | */ |
79f546a6 DC |
266 | if (!(sb->s_flags & SB_BORN)) |
267 | return 0; | |
268 | smp_rmb(); | |
269 | ||
0a234c6d | 270 | if (sb->s_op && sb->s_op->nr_cached_objects) |
4101b624 | 271 | total_objects = sb->s_op->nr_cached_objects(sb, sc); |
0a234c6d | 272 | |
503c358c VD |
273 | total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc); |
274 | total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc); | |
0a234c6d | 275 | |
9b996468 KT |
276 | if (!total_objects) |
277 | return SHRINK_EMPTY; | |
278 | ||
55f841ce | 279 | total_objects = vfs_pressure_ratio(total_objects); |
0e1fdafd | 280 | return total_objects; |
b0d40c92 DC |
281 | } |
282 | ||
853b39a7 ON |
283 | static void destroy_super_work(struct work_struct *work) |
284 | { | |
285 | struct super_block *s = container_of(work, struct super_block, | |
286 | destroy_work); | |
287 | int i; | |
288 | ||
289 | for (i = 0; i < SB_FREEZE_LEVELS; i++) | |
8129ed29 | 290 | percpu_free_rwsem(&s->s_writers.rw_sem[i]); |
853b39a7 ON |
291 | kfree(s); |
292 | } | |
293 | ||
294 | static void destroy_super_rcu(struct rcu_head *head) | |
295 | { | |
296 | struct super_block *s = container_of(head, struct super_block, rcu); | |
297 | INIT_WORK(&s->destroy_work, destroy_super_work); | |
298 | schedule_work(&s->destroy_work); | |
299 | } | |
300 | ||
0200894d AV |
301 | /* Free a superblock that has never been seen by anyone */ |
302 | static void destroy_unused_super(struct super_block *s) | |
5accdf82 | 303 | { |
0200894d AV |
304 | if (!s) |
305 | return; | |
0ed33598 | 306 | super_unlock_excl(s); |
7eb5e882 AV |
307 | list_lru_destroy(&s->s_dentry_lru); |
308 | list_lru_destroy(&s->s_inode_lru); | |
7eb5e882 | 309 | security_sb_free(s); |
6e4eab57 | 310 | put_user_ns(s->s_user_ns); |
7eb5e882 | 311 | kfree(s->s_subtype); |
1720f5dd | 312 | shrinker_free(s->s_shrink); |
0200894d AV |
313 | /* no delays needed */ |
314 | destroy_super_work(&s->destroy_work); | |
5accdf82 JK |
315 | } |
316 | ||
1da177e4 LT |
317 | /** |
318 | * alloc_super - create new superblock | |
fe2bbc48 | 319 | * @type: filesystem type superblock should belong to |
9249e17f | 320 | * @flags: the mount flags |
6e4eab57 | 321 | * @user_ns: User namespace for the super_block |
1da177e4 LT |
322 | * |
323 | * Allocates and initializes a new &struct super_block. alloc_super() | |
324 | * returns a pointer new superblock or %NULL if allocation had failed. | |
325 | */ | |
6e4eab57 EB |
326 | static struct super_block *alloc_super(struct file_system_type *type, int flags, |
327 | struct user_namespace *user_ns) | |
1da177e4 | 328 | { |
11b0b5ab | 329 | struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER); |
b87221de | 330 | static const struct super_operations default_op; |
7eb5e882 AV |
331 | int i; |
332 | ||
333 | if (!s) | |
334 | return NULL; | |
1da177e4 | 335 | |
b5bd856a | 336 | INIT_LIST_HEAD(&s->s_mounts); |
6e4eab57 | 337 | s->s_user_ns = get_user_ns(user_ns); |
ca0168e8 AV |
338 | init_rwsem(&s->s_umount); |
339 | lockdep_set_class(&s->s_umount, &type->s_umount_key); | |
340 | /* | |
341 | * sget() can have s_umount recursion. | |
342 | * | |
343 | * When it cannot find a suitable sb, it allocates a new | |
344 | * one (this one), and tries again to find a suitable old | |
345 | * one. | |
346 | * | |
347 | * In case that succeeds, it will acquire the s_umount | |
348 | * lock of the old one. Since these are clearly distrinct | |
349 | * locks, and this object isn't exposed yet, there's no | |
350 | * risk of deadlocks. | |
351 | * | |
352 | * Annotate this by putting this lock in a different | |
353 | * subclass. | |
354 | */ | |
355 | down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING); | |
b5bd856a | 356 | |
7eb5e882 AV |
357 | if (security_sb_alloc(s)) |
358 | goto fail; | |
7b7a8665 | 359 | |
7eb5e882 | 360 | for (i = 0; i < SB_FREEZE_LEVELS; i++) { |
8129ed29 ON |
361 | if (__percpu_init_rwsem(&s->s_writers.rw_sem[i], |
362 | sb_writers_name[i], | |
363 | &type->s_writers_key[i])) | |
7eb5e882 | 364 | goto fail; |
1da177e4 | 365 | } |
df0ce26c | 366 | s->s_bdi = &noop_backing_dev_info; |
7eb5e882 | 367 | s->s_flags = flags; |
cc50a07a | 368 | if (s->s_user_ns != &init_user_ns) |
67690f93 | 369 | s->s_iflags |= SB_I_NODEV; |
7eb5e882 | 370 | INIT_HLIST_NODE(&s->s_instances); |
f1ee6162 | 371 | INIT_HLIST_BL_HEAD(&s->s_roots); |
e97fedb9 | 372 | mutex_init(&s->s_sync_lock); |
7eb5e882 | 373 | INIT_LIST_HEAD(&s->s_inodes); |
74278da9 | 374 | spin_lock_init(&s->s_inode_list_lock); |
6c60d2b5 DC |
375 | INIT_LIST_HEAD(&s->s_inodes_wb); |
376 | spin_lock_init(&s->s_inode_wblist_lock); | |
7eb5e882 | 377 | |
7eb5e882 AV |
378 | s->s_count = 1; |
379 | atomic_set(&s->s_active, 1); | |
380 | mutex_init(&s->s_vfs_rename_mutex); | |
381 | lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key); | |
bc8230ee | 382 | init_rwsem(&s->s_dquot.dqio_sem); |
7eb5e882 AV |
383 | s->s_maxbytes = MAX_NON_LFS; |
384 | s->s_op = &default_op; | |
385 | s->s_time_gran = 1000000000; | |
188d20bc DD |
386 | s->s_time_min = TIME64_MIN; |
387 | s->s_time_max = TIME64_MAX; | |
7eb5e882 | 388 | |
1720f5dd QZ |
389 | s->s_shrink = shrinker_alloc(SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE, |
390 | "sb-%s", type->name); | |
391 | if (!s->s_shrink) | |
8e04944f | 392 | goto fail; |
1720f5dd QZ |
393 | |
394 | s->s_shrink->scan_objects = super_cache_scan; | |
395 | s->s_shrink->count_objects = super_cache_count; | |
396 | s->s_shrink->batch = 1024; | |
397 | s->s_shrink->private_data = s; | |
398 | ||
399 | if (list_lru_init_memcg(&s->s_dentry_lru, s->s_shrink)) | |
2b3648a6 | 400 | goto fail; |
1720f5dd | 401 | if (list_lru_init_memcg(&s->s_inode_lru, s->s_shrink)) |
2b3648a6 | 402 | goto fail; |
1da177e4 | 403 | return s; |
5ca302c8 | 404 | |
7eb5e882 | 405 | fail: |
0200894d | 406 | destroy_unused_super(s); |
7eb5e882 | 407 | return NULL; |
1da177e4 LT |
408 | } |
409 | ||
410 | /* Superblock refcounting */ | |
411 | ||
412 | /* | |
35cf7ba0 | 413 | * Drop a superblock's refcount. The caller must hold sb_lock. |
1da177e4 | 414 | */ |
c645b930 | 415 | static void __put_super(struct super_block *s) |
1da177e4 | 416 | { |
c645b930 AV |
417 | if (!--s->s_count) { |
418 | list_del_init(&s->s_list); | |
419 | WARN_ON(s->s_dentry_lru.node); | |
420 | WARN_ON(s->s_inode_lru.node); | |
421 | WARN_ON(!list_empty(&s->s_mounts)); | |
422 | security_sb_free(s); | |
423 | put_user_ns(s->s_user_ns); | |
424 | kfree(s->s_subtype); | |
425 | call_rcu(&s->rcu, destroy_super_rcu); | |
1da177e4 | 426 | } |
1da177e4 LT |
427 | } |
428 | ||
429 | /** | |
430 | * put_super - drop a temporary reference to superblock | |
431 | * @sb: superblock in question | |
432 | * | |
433 | * Drops a temporary reference, frees superblock if there's no | |
434 | * references left. | |
435 | */ | |
60b49885 | 436 | void put_super(struct super_block *sb) |
1da177e4 LT |
437 | { |
438 | spin_lock(&sb_lock); | |
439 | __put_super(sb); | |
440 | spin_unlock(&sb_lock); | |
441 | } | |
442 | ||
dc3216b1 CB |
443 | static void kill_super_notify(struct super_block *sb) |
444 | { | |
445 | lockdep_assert_not_held(&sb->s_umount); | |
446 | ||
447 | /* already notified earlier */ | |
448 | if (sb->s_flags & SB_DEAD) | |
449 | return; | |
450 | ||
451 | /* | |
452 | * Remove it from @fs_supers so it isn't found by new | |
453 | * sget{_fc}() walkers anymore. Any concurrent mounter still | |
454 | * managing to grab a temporary reference is guaranteed to | |
455 | * already see SB_DYING and will wait until we notify them about | |
456 | * SB_DEAD. | |
457 | */ | |
458 | spin_lock(&sb_lock); | |
459 | hlist_del_init(&sb->s_instances); | |
460 | spin_unlock(&sb_lock); | |
461 | ||
462 | /* | |
463 | * Let concurrent mounts know that this thing is really dead. | |
464 | * We don't need @sb->s_umount here as every concurrent caller | |
465 | * will see SB_DYING and either discard the superblock or wait | |
466 | * for SB_DEAD. | |
467 | */ | |
468 | super_wake(sb, SB_DEAD); | |
469 | } | |
1da177e4 LT |
470 | |
471 | /** | |
1712ac8f | 472 | * deactivate_locked_super - drop an active reference to superblock |
1da177e4 LT |
473 | * @s: superblock to deactivate |
474 | * | |
bd7ced98 | 475 | * Drops an active reference to superblock, converting it into a temporary |
1712ac8f | 476 | * one if there is no other active references left. In that case we |
1da177e4 LT |
477 | * tell fs driver to shut it down and drop the temporary reference we |
478 | * had just acquired. | |
1712ac8f AV |
479 | * |
480 | * Caller holds exclusive lock on superblock; that lock is released. | |
1da177e4 | 481 | */ |
1712ac8f | 482 | void deactivate_locked_super(struct super_block *s) |
1da177e4 LT |
483 | { |
484 | struct file_system_type *fs = s->s_type; | |
b20bd1a5 | 485 | if (atomic_dec_and_test(&s->s_active)) { |
1720f5dd | 486 | shrinker_free(s->s_shrink); |
28f2cd4f | 487 | fs->kill_sb(s); |
f5e1dd34 | 488 | |
dc3216b1 CB |
489 | kill_super_notify(s); |
490 | ||
c0a5b560 VD |
491 | /* |
492 | * Since list_lru_destroy() may sleep, we cannot call it from | |
493 | * put_super(), where we hold the sb_lock. Therefore we destroy | |
494 | * the lru lists right now. | |
495 | */ | |
496 | list_lru_destroy(&s->s_dentry_lru); | |
497 | list_lru_destroy(&s->s_inode_lru); | |
498 | ||
1da177e4 LT |
499 | put_filesystem(fs); |
500 | put_super(s); | |
1712ac8f | 501 | } else { |
0ed33598 | 502 | super_unlock_excl(s); |
1da177e4 LT |
503 | } |
504 | } | |
505 | ||
1712ac8f | 506 | EXPORT_SYMBOL(deactivate_locked_super); |
1da177e4 | 507 | |
74dbbdd7 | 508 | /** |
1712ac8f | 509 | * deactivate_super - drop an active reference to superblock |
74dbbdd7 AV |
510 | * @s: superblock to deactivate |
511 | * | |
1712ac8f AV |
512 | * Variant of deactivate_locked_super(), except that superblock is *not* |
513 | * locked by caller. If we are going to drop the final active reference, | |
514 | * lock will be acquired prior to that. | |
74dbbdd7 | 515 | */ |
1712ac8f | 516 | void deactivate_super(struct super_block *s) |
74dbbdd7 | 517 | { |
cc23402c | 518 | if (!atomic_add_unless(&s->s_active, -1, 1)) { |
5e874914 | 519 | __super_lock_excl(s); |
1712ac8f | 520 | deactivate_locked_super(s); |
74dbbdd7 AV |
521 | } |
522 | } | |
523 | ||
1712ac8f | 524 | EXPORT_SYMBOL(deactivate_super); |
74dbbdd7 | 525 | |
2c18a63b CB |
526 | static inline bool wait_dead(struct super_block *sb) |
527 | { | |
528 | unsigned int flags; | |
529 | ||
530 | /* | |
531 | * Pairs with memory barrier in super_wake() and ensures | |
532 | * that we see SB_DEAD after we're woken. | |
533 | */ | |
534 | flags = smp_load_acquire(&sb->s_flags); | |
535 | return flags & SB_DEAD; | |
536 | } | |
537 | ||
538 | /** | |
97cbed04 | 539 | * grab_super - acquire an active reference to a superblock |
2c18a63b CB |
540 | * @sb: superblock to acquire |
541 | * | |
542 | * Acquire a temporary reference on a superblock and try to trade it for | |
543 | * an active reference. This is used in sget{_fc}() to wait for a | |
544 | * superblock to either become SB_BORN or for it to pass through | |
545 | * sb->kill() and be marked as SB_DEAD. | |
546 | * | |
547 | * Return: This returns true if an active reference could be acquired, | |
548 | * false if not. | |
549 | */ | |
97cbed04 | 550 | static bool grab_super(struct super_block *sb) |
2c18a63b | 551 | { |
97cbed04 CB |
552 | bool locked; |
553 | ||
2c18a63b | 554 | sb->s_count++; |
97cbed04 CB |
555 | spin_unlock(&sb_lock); |
556 | locked = super_lock_excl(sb); | |
557 | if (locked) { | |
558 | if (atomic_inc_not_zero(&sb->s_active)) { | |
559 | put_super(sb); | |
560 | return true; | |
561 | } | |
562 | super_unlock_excl(sb); | |
2c18a63b CB |
563 | } |
564 | wait_var_event(&sb->s_flags, wait_dead(sb)); | |
345a5c4a | 565 | put_super(sb); |
2c18a63b CB |
566 | return false; |
567 | } | |
568 | ||
12ad3ab6 | 569 | /* |
d8ce82ef | 570 | * super_trylock_shared - try to grab ->s_umount shared |
331cbdee | 571 | * @sb: reference we are trying to grab |
12ad3ab6 | 572 | * |
eb6ef3df | 573 | * Try to prevent fs shutdown. This is used in places where we |
12ad3ab6 | 574 | * cannot take an active reference but we need to ensure that the |
eb6ef3df KK |
575 | * filesystem is not shut down while we are working on it. It returns |
576 | * false if we cannot acquire s_umount or if we lose the race and | |
577 | * filesystem already got into shutdown, and returns true with the s_umount | |
578 | * lock held in read mode in case of success. On successful return, | |
579 | * the caller must drop the s_umount lock when done. | |
580 | * | |
581 | * Note that unlike get_super() et.al. this one does *not* bump ->s_count. | |
582 | * The reason why it's safe is that we are OK with doing trylock instead | |
583 | * of down_read(). There's a couple of places that are OK with that, but | |
584 | * it's very much not a general-purpose interface. | |
12ad3ab6 | 585 | */ |
d8ce82ef | 586 | bool super_trylock_shared(struct super_block *sb) |
12ad3ab6 | 587 | { |
12ad3ab6 | 588 | if (down_read_trylock(&sb->s_umount)) { |
5e874914 CB |
589 | if (!(sb->s_flags & SB_DYING) && sb->s_root && |
590 | (sb->s_flags & SB_BORN)) | |
12ad3ab6 | 591 | return true; |
0ed33598 | 592 | super_unlock_shared(sb); |
12ad3ab6 DC |
593 | } |
594 | ||
12ad3ab6 DC |
595 | return false; |
596 | } | |
597 | ||
04b94071 DL |
598 | /** |
599 | * retire_super - prevents superblock from being reused | |
600 | * @sb: superblock to retire | |
601 | * | |
602 | * The function marks superblock to be ignored in superblock test, which | |
603 | * prevents it from being reused for any new mounts. If the superblock has | |
604 | * a private bdi, it also unregisters it, but doesn't reduce the refcount | |
605 | * of the superblock to prevent potential races. The refcount is reduced | |
606 | * by generic_shutdown_super(). The function can not be called | |
607 | * concurrently with generic_shutdown_super(). It is safe to call the | |
608 | * function multiple times, subsequent calls have no effect. | |
609 | * | |
610 | * The marker will affect the re-use only for block-device-based | |
611 | * superblocks. Other superblocks will still get marked if this function | |
612 | * is used, but that will not affect their reusability. | |
613 | */ | |
614 | void retire_super(struct super_block *sb) | |
615 | { | |
616 | WARN_ON(!sb->s_bdev); | |
5e874914 | 617 | __super_lock_excl(sb); |
04b94071 DL |
618 | if (sb->s_iflags & SB_I_PERSB_BDI) { |
619 | bdi_unregister(sb->s_bdi); | |
620 | sb->s_iflags &= ~SB_I_PERSB_BDI; | |
621 | } | |
622 | sb->s_iflags |= SB_I_RETIRED; | |
0ed33598 | 623 | super_unlock_excl(sb); |
04b94071 DL |
624 | } |
625 | EXPORT_SYMBOL(retire_super); | |
626 | ||
1da177e4 LT |
627 | /** |
628 | * generic_shutdown_super - common helper for ->kill_sb() | |
629 | * @sb: superblock to kill | |
630 | * | |
631 | * generic_shutdown_super() does all fs-independent work on superblock | |
632 | * shutdown. Typical ->kill_sb() should pick all fs-specific objects | |
633 | * that need destruction out of superblock, call generic_shutdown_super() | |
634 | * and release aforementioned objects. Note: dentries and inodes _are_ | |
635 | * taken care of and do not need specific handling. | |
c636ebdb DH |
636 | * |
637 | * Upon calling this function, the filesystem may no longer alter or | |
638 | * rearrange the set of dentries belonging to this super_block, nor may it | |
639 | * change the attachments of dentries to inodes. | |
1da177e4 LT |
640 | */ |
641 | void generic_shutdown_super(struct super_block *sb) | |
642 | { | |
ee9b6d61 | 643 | const struct super_operations *sop = sb->s_op; |
1da177e4 | 644 | |
c636ebdb DH |
645 | if (sb->s_root) { |
646 | shrink_dcache_for_umount(sb); | |
60b0680f | 647 | sync_filesystem(sb); |
e462ec50 | 648 | sb->s_flags &= ~SB_ACTIVE; |
efaee192 | 649 | |
a1a0e23e | 650 | cgroup_writeback_umount(); |
63997e98 | 651 | |
ccb820dc | 652 | /* Evict all inodes with zero refcount. */ |
63997e98 | 653 | evict_inodes(sb); |
ccb820dc EB |
654 | |
655 | /* | |
656 | * Clean up and evict any inodes that still have references due | |
657 | * to fsnotify or the security policy. | |
658 | */ | |
1edc8eb2 | 659 | fsnotify_sb_delete(sb); |
83e804f0 | 660 | security_sb_delete(sb); |
1da177e4 | 661 | |
ccb820dc EB |
662 | /* |
663 | * Now that all potentially-encrypted inodes have been evicted, | |
664 | * the fscrypt keyring can be destroyed. | |
665 | */ | |
666 | fscrypt_destroy_keyring(sb); | |
667 | ||
7b7a8665 CH |
668 | if (sb->s_dio_done_wq) { |
669 | destroy_workqueue(sb->s_dio_done_wq); | |
670 | sb->s_dio_done_wq = NULL; | |
671 | } | |
672 | ||
1da177e4 LT |
673 | if (sop->put_super) |
674 | sop->put_super(sb); | |
675 | ||
47d58691 JH |
676 | if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes), |
677 | "VFS: Busy inodes after unmount of %s (%s)", | |
678 | sb->s_id, sb->s_type->name)) { | |
679 | /* | |
680 | * Adding a proper bailout path here would be hard, but | |
681 | * we can at least make it more likely that a later | |
682 | * iput_final() or such crashes cleanly. | |
683 | */ | |
684 | struct inode *inode; | |
685 | ||
686 | spin_lock(&sb->s_inode_list_lock); | |
687 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { | |
688 | inode->i_op = VFS_PTR_POISON; | |
689 | inode->i_sb = VFS_PTR_POISON; | |
690 | inode->i_mapping = VFS_PTR_POISON; | |
691 | } | |
692 | spin_unlock(&sb->s_inode_list_lock); | |
1da177e4 | 693 | } |
1da177e4 | 694 | } |
5e874914 CB |
695 | /* |
696 | * Broadcast to everyone that grabbed a temporary reference to this | |
697 | * superblock before we removed it from @fs_supers that the superblock | |
698 | * is dying. Every walker of @fs_supers outside of sget{_fc}() will now | |
699 | * discard this superblock and treat it as dead. | |
2c18a63b CB |
700 | * |
701 | * We leave the superblock on @fs_supers so it can be found by | |
702 | * sget{_fc}() until we passed sb->kill_sb(). | |
5e874914 CB |
703 | */ |
704 | super_wake(sb, SB_DYING); | |
0ed33598 | 705 | super_unlock_excl(sb); |
c1844d53 | 706 | if (sb->s_bdi != &noop_backing_dev_info) { |
0b3ea092 CH |
707 | if (sb->s_iflags & SB_I_PERSB_BDI) |
708 | bdi_unregister(sb->s_bdi); | |
fca39346 JK |
709 | bdi_put(sb->s_bdi); |
710 | sb->s_bdi = &noop_backing_dev_info; | |
fca39346 | 711 | } |
1da177e4 LT |
712 | } |
713 | ||
714 | EXPORT_SYMBOL(generic_shutdown_super); | |
715 | ||
20284ab7 | 716 | bool mount_capable(struct fs_context *fc) |
0ce0cf12 | 717 | { |
20284ab7 | 718 | if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT)) |
0ce0cf12 AV |
719 | return capable(CAP_SYS_ADMIN); |
720 | else | |
c2c44ec2 | 721 | return ns_capable(fc->user_ns, CAP_SYS_ADMIN); |
0ce0cf12 AV |
722 | } |
723 | ||
cb50b348 AV |
724 | /** |
725 | * sget_fc - Find or create a superblock | |
726 | * @fc: Filesystem context. | |
727 | * @test: Comparison callback | |
728 | * @set: Setup callback | |
729 | * | |
22ed7ecd | 730 | * Create a new superblock or find an existing one. |
cb50b348 | 731 | * |
22ed7ecd CB |
732 | * The @test callback is used to find a matching existing superblock. |
733 | * Whether or not the requested parameters in @fc are taken into account | |
734 | * is specific to the @test callback that is used. They may even be | |
735 | * completely ignored. | |
736 | * | |
737 | * If an extant superblock is matched, it will be returned unless: | |
738 | * | |
739 | * (1) the namespace the filesystem context @fc and the extant | |
740 | * superblock's namespace differ | |
741 | * | |
742 | * (2) the filesystem context @fc has requested that reusing an extant | |
743 | * superblock is not allowed | |
744 | * | |
745 | * In both cases EBUSY will be returned. | |
cb50b348 AV |
746 | * |
747 | * If no match is made, a new superblock will be allocated and basic | |
22ed7ecd CB |
748 | * initialisation will be performed (s_type, s_fs_info and s_id will be |
749 | * set and the @set callback will be invoked), the superblock will be | |
750 | * published and it will be returned in a partially constructed state | |
751 | * with SB_BORN and SB_ACTIVE as yet unset. | |
752 | * | |
753 | * Return: On success, an extant or newly created superblock is | |
754 | * returned. On failure an error pointer is returned. | |
cb50b348 AV |
755 | */ |
756 | struct super_block *sget_fc(struct fs_context *fc, | |
757 | int (*test)(struct super_block *, struct fs_context *), | |
758 | int (*set)(struct super_block *, struct fs_context *)) | |
759 | { | |
760 | struct super_block *s = NULL; | |
761 | struct super_block *old; | |
762 | struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns; | |
763 | int err; | |
764 | ||
cb50b348 AV |
765 | retry: |
766 | spin_lock(&sb_lock); | |
767 | if (test) { | |
768 | hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) { | |
769 | if (test(old, fc)) | |
770 | goto share_extant_sb; | |
771 | } | |
772 | } | |
773 | if (!s) { | |
774 | spin_unlock(&sb_lock); | |
775 | s = alloc_super(fc->fs_type, fc->sb_flags, user_ns); | |
776 | if (!s) | |
777 | return ERR_PTR(-ENOMEM); | |
778 | goto retry; | |
779 | } | |
780 | ||
781 | s->s_fs_info = fc->s_fs_info; | |
782 | err = set(s, fc); | |
783 | if (err) { | |
784 | s->s_fs_info = NULL; | |
785 | spin_unlock(&sb_lock); | |
786 | destroy_unused_super(s); | |
787 | return ERR_PTR(err); | |
788 | } | |
789 | fc->s_fs_info = NULL; | |
790 | s->s_type = fc->fs_type; | |
c80fa7c8 | 791 | s->s_iflags |= fc->s_iflags; |
c642256b | 792 | strscpy(s->s_id, s->s_type->name, sizeof(s->s_id)); |
5e874914 CB |
793 | /* |
794 | * Make the superblock visible on @super_blocks and @fs_supers. | |
795 | * It's in a nascent state and users should wait on SB_BORN or | |
796 | * SB_DYING to be set. | |
797 | */ | |
cb50b348 AV |
798 | list_add_tail(&s->s_list, &super_blocks); |
799 | hlist_add_head(&s->s_instances, &s->s_type->fs_supers); | |
800 | spin_unlock(&sb_lock); | |
801 | get_filesystem(s->s_type); | |
1720f5dd | 802 | shrinker_register(s->s_shrink); |
cb50b348 AV |
803 | return s; |
804 | ||
805 | share_extant_sb: | |
22ed7ecd | 806 | if (user_ns != old->s_user_ns || fc->exclusive) { |
cb50b348 AV |
807 | spin_unlock(&sb_lock); |
808 | destroy_unused_super(s); | |
22ed7ecd CB |
809 | if (fc->exclusive) |
810 | warnfc(fc, "reusing existing filesystem not allowed"); | |
811 | else | |
812 | warnfc(fc, "reusing existing filesystem in another namespace not allowed"); | |
cb50b348 AV |
813 | return ERR_PTR(-EBUSY); |
814 | } | |
97cbed04 | 815 | if (!grab_super(old)) |
cb50b348 AV |
816 | goto retry; |
817 | destroy_unused_super(s); | |
818 | return old; | |
819 | } | |
820 | EXPORT_SYMBOL(sget_fc); | |
821 | ||
1da177e4 | 822 | /** |
023d066a DH |
823 | * sget - find or create a superblock |
824 | * @type: filesystem type superblock should belong to | |
825 | * @test: comparison callback | |
826 | * @set: setup callback | |
827 | * @flags: mount flags | |
828 | * @data: argument to each of them | |
1da177e4 | 829 | */ |
023d066a | 830 | struct super_block *sget(struct file_system_type *type, |
1da177e4 LT |
831 | int (*test)(struct super_block *,void *), |
832 | int (*set)(struct super_block *,void *), | |
023d066a | 833 | int flags, |
1da177e4 LT |
834 | void *data) |
835 | { | |
023d066a | 836 | struct user_namespace *user_ns = current_user_ns(); |
1da177e4 | 837 | struct super_block *s = NULL; |
d4730127 | 838 | struct super_block *old; |
1da177e4 LT |
839 | int err; |
840 | ||
023d066a DH |
841 | /* We don't yet pass the user namespace of the parent |
842 | * mount through to here so always use &init_user_ns | |
843 | * until that changes. | |
844 | */ | |
845 | if (flags & SB_SUBMOUNT) | |
846 | user_ns = &init_user_ns; | |
847 | ||
1da177e4 LT |
848 | retry: |
849 | spin_lock(&sb_lock); | |
d4730127 | 850 | if (test) { |
b67bfe0d | 851 | hlist_for_each_entry(old, &type->fs_supers, s_instances) { |
d4730127 MK |
852 | if (!test(old, data)) |
853 | continue; | |
6e4eab57 EB |
854 | if (user_ns != old->s_user_ns) { |
855 | spin_unlock(&sb_lock); | |
0200894d | 856 | destroy_unused_super(s); |
6e4eab57 EB |
857 | return ERR_PTR(-EBUSY); |
858 | } | |
97cbed04 | 859 | if (!grab_super(old)) |
d4730127 | 860 | goto retry; |
0200894d | 861 | destroy_unused_super(s); |
d4730127 MK |
862 | return old; |
863 | } | |
1da177e4 LT |
864 | } |
865 | if (!s) { | |
866 | spin_unlock(&sb_lock); | |
e462ec50 | 867 | s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns); |
1da177e4 LT |
868 | if (!s) |
869 | return ERR_PTR(-ENOMEM); | |
870 | goto retry; | |
871 | } | |
dd111b31 | 872 | |
1da177e4 LT |
873 | err = set(s, data); |
874 | if (err) { | |
875 | spin_unlock(&sb_lock); | |
0200894d | 876 | destroy_unused_super(s); |
1da177e4 LT |
877 | return ERR_PTR(err); |
878 | } | |
879 | s->s_type = type; | |
c642256b | 880 | strscpy(s->s_id, type->name, sizeof(s->s_id)); |
1da177e4 | 881 | list_add_tail(&s->s_list, &super_blocks); |
a5166169 | 882 | hlist_add_head(&s->s_instances, &type->fs_supers); |
1da177e4 LT |
883 | spin_unlock(&sb_lock); |
884 | get_filesystem(type); | |
1720f5dd | 885 | shrinker_register(s->s_shrink); |
1da177e4 LT |
886 | return s; |
887 | } | |
1da177e4 LT |
888 | EXPORT_SYMBOL(sget); |
889 | ||
890 | void drop_super(struct super_block *sb) | |
891 | { | |
0ed33598 | 892 | super_unlock_shared(sb); |
1da177e4 LT |
893 | put_super(sb); |
894 | } | |
895 | ||
896 | EXPORT_SYMBOL(drop_super); | |
897 | ||
ba6379f7 JK |
898 | void drop_super_exclusive(struct super_block *sb) |
899 | { | |
0ed33598 | 900 | super_unlock_excl(sb); |
ba6379f7 JK |
901 | put_super(sb); |
902 | } | |
903 | EXPORT_SYMBOL(drop_super_exclusive); | |
904 | ||
fa7c1d50 MG |
905 | static void __iterate_supers(void (*f)(struct super_block *)) |
906 | { | |
907 | struct super_block *sb, *p = NULL; | |
908 | ||
909 | spin_lock(&sb_lock); | |
910 | list_for_each_entry(sb, &super_blocks, s_list) { | |
5e874914 CB |
911 | /* Pairs with memory marrier in super_wake(). */ |
912 | if (smp_load_acquire(&sb->s_flags) & SB_DYING) | |
fa7c1d50 MG |
913 | continue; |
914 | sb->s_count++; | |
915 | spin_unlock(&sb_lock); | |
916 | ||
917 | f(sb); | |
918 | ||
919 | spin_lock(&sb_lock); | |
920 | if (p) | |
921 | __put_super(p); | |
922 | p = sb; | |
923 | } | |
924 | if (p) | |
925 | __put_super(p); | |
926 | spin_unlock(&sb_lock); | |
927 | } | |
01a05b33 AV |
928 | /** |
929 | * iterate_supers - call function for all active superblocks | |
930 | * @f: function to call | |
931 | * @arg: argument to pass to it | |
932 | * | |
933 | * Scans the superblock list and calls given function, passing it | |
934 | * locked superblock and given argument. | |
935 | */ | |
936 | void iterate_supers(void (*f)(struct super_block *, void *), void *arg) | |
937 | { | |
dca33252 | 938 | struct super_block *sb, *p = NULL; |
01a05b33 AV |
939 | |
940 | spin_lock(&sb_lock); | |
dca33252 | 941 | list_for_each_entry(sb, &super_blocks, s_list) { |
f0cd9880 | 942 | bool locked; |
5e874914 | 943 | |
01a05b33 AV |
944 | sb->s_count++; |
945 | spin_unlock(&sb_lock); | |
946 | ||
f0cd9880 CB |
947 | locked = super_lock_shared(sb); |
948 | if (locked) { | |
949 | if (sb->s_root) | |
950 | f(sb, arg); | |
951 | super_unlock_shared(sb); | |
952 | } | |
01a05b33 AV |
953 | |
954 | spin_lock(&sb_lock); | |
dca33252 AV |
955 | if (p) |
956 | __put_super(p); | |
957 | p = sb; | |
01a05b33 | 958 | } |
dca33252 AV |
959 | if (p) |
960 | __put_super(p); | |
01a05b33 AV |
961 | spin_unlock(&sb_lock); |
962 | } | |
963 | ||
43e15cdb AV |
964 | /** |
965 | * iterate_supers_type - call function for superblocks of given type | |
966 | * @type: fs type | |
967 | * @f: function to call | |
968 | * @arg: argument to pass to it | |
969 | * | |
970 | * Scans the superblock list and calls given function, passing it | |
971 | * locked superblock and given argument. | |
972 | */ | |
973 | void iterate_supers_type(struct file_system_type *type, | |
974 | void (*f)(struct super_block *, void *), void *arg) | |
975 | { | |
976 | struct super_block *sb, *p = NULL; | |
977 | ||
978 | spin_lock(&sb_lock); | |
b67bfe0d | 979 | hlist_for_each_entry(sb, &type->fs_supers, s_instances) { |
f0cd9880 | 980 | bool locked; |
5e874914 | 981 | |
43e15cdb AV |
982 | sb->s_count++; |
983 | spin_unlock(&sb_lock); | |
984 | ||
f0cd9880 CB |
985 | locked = super_lock_shared(sb); |
986 | if (locked) { | |
987 | if (sb->s_root) | |
988 | f(sb, arg); | |
989 | super_unlock_shared(sb); | |
990 | } | |
43e15cdb AV |
991 | |
992 | spin_lock(&sb_lock); | |
993 | if (p) | |
994 | __put_super(p); | |
995 | p = sb; | |
996 | } | |
997 | if (p) | |
998 | __put_super(p); | |
999 | spin_unlock(&sb_lock); | |
1000 | } | |
1001 | ||
1002 | EXPORT_SYMBOL(iterate_supers_type); | |
1003 | ||
4e7b5671 | 1004 | struct super_block *user_get_super(dev_t dev, bool excl) |
1da177e4 | 1005 | { |
618f0636 | 1006 | struct super_block *sb; |
1da177e4 | 1007 | |
1da177e4 | 1008 | spin_lock(&sb_lock); |
618f0636 KK |
1009 | list_for_each_entry(sb, &super_blocks, s_list) { |
1010 | if (sb->s_dev == dev) { | |
f0cd9880 | 1011 | bool locked; |
5e874914 | 1012 | |
618f0636 | 1013 | sb->s_count++; |
1da177e4 | 1014 | spin_unlock(&sb_lock); |
df40c01a | 1015 | /* still alive? */ |
f0cd9880 CB |
1016 | locked = super_lock(sb, excl); |
1017 | if (locked) { | |
1018 | if (sb->s_root) | |
1019 | return sb; | |
1020 | super_unlock(sb, excl); | |
1021 | } | |
df40c01a | 1022 | /* nope, got unmounted */ |
618f0636 | 1023 | spin_lock(&sb_lock); |
df40c01a | 1024 | __put_super(sb); |
5e874914 | 1025 | break; |
1da177e4 LT |
1026 | } |
1027 | } | |
1028 | spin_unlock(&sb_lock); | |
1029 | return NULL; | |
1030 | } | |
1031 | ||
1da177e4 | 1032 | /** |
8d0347f6 DH |
1033 | * reconfigure_super - asks filesystem to change superblock parameters |
1034 | * @fc: The superblock and configuration | |
1da177e4 | 1035 | * |
8d0347f6 | 1036 | * Alters the configuration parameters of a live superblock. |
1da177e4 | 1037 | */ |
8d0347f6 | 1038 | int reconfigure_super(struct fs_context *fc) |
1da177e4 | 1039 | { |
8d0347f6 | 1040 | struct super_block *sb = fc->root->d_sb; |
1da177e4 | 1041 | int retval; |
8d0347f6 | 1042 | bool remount_ro = false; |
c541dce8 | 1043 | bool remount_rw = false; |
8d0347f6 | 1044 | bool force = fc->sb_flags & SB_FORCE; |
4504230a | 1045 | |
8d0347f6 DH |
1046 | if (fc->sb_flags_mask & ~MS_RMT_MASK) |
1047 | return -EINVAL; | |
5accdf82 | 1048 | if (sb->s_writers.frozen != SB_UNFROZEN) |
4504230a CH |
1049 | return -EBUSY; |
1050 | ||
8d0347f6 DH |
1051 | retval = security_sb_remount(sb, fc->security); |
1052 | if (retval) | |
1053 | return retval; | |
1054 | ||
1055 | if (fc->sb_flags_mask & SB_RDONLY) { | |
9361401e | 1056 | #ifdef CONFIG_BLOCK |
6f0d9689 CH |
1057 | if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev && |
1058 | bdev_read_only(sb->s_bdev)) | |
8d0347f6 | 1059 | return -EACCES; |
9361401e | 1060 | #endif |
c541dce8 | 1061 | remount_rw = !(fc->sb_flags & SB_RDONLY) && sb_rdonly(sb); |
8d0347f6 DH |
1062 | remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb); |
1063 | } | |
d208bbdd | 1064 | |
0aec09d0 | 1065 | if (remount_ro) { |
fdab684d | 1066 | if (!hlist_empty(&sb->s_pins)) { |
0ed33598 | 1067 | super_unlock_excl(sb); |
fdab684d | 1068 | group_pin_kill(&sb->s_pins); |
5e874914 | 1069 | __super_lock_excl(sb); |
0aec09d0 AV |
1070 | if (!sb->s_root) |
1071 | return 0; | |
1072 | if (sb->s_writers.frozen != SB_UNFROZEN) | |
1073 | return -EBUSY; | |
8d0347f6 | 1074 | remount_ro = !sb_rdonly(sb); |
0aec09d0 AV |
1075 | } |
1076 | } | |
1077 | shrink_dcache_sb(sb); | |
1078 | ||
8d0347f6 DH |
1079 | /* If we are reconfiguring to RDONLY and current sb is read/write, |
1080 | * make sure there are no files open for writing. | |
1081 | */ | |
d208bbdd | 1082 | if (remount_ro) { |
4ed5e82f | 1083 | if (force) { |
d7439fb1 | 1084 | sb_start_ro_state_change(sb); |
4ed5e82f MS |
1085 | } else { |
1086 | retval = sb_prepare_remount_readonly(sb); | |
1087 | if (retval) | |
1088 | return retval; | |
4ed5e82f | 1089 | } |
c541dce8 JK |
1090 | } else if (remount_rw) { |
1091 | /* | |
d7439fb1 JK |
1092 | * Protect filesystem's reconfigure code from writes from |
1093 | * userspace until reconfigure finishes. | |
c541dce8 | 1094 | */ |
d7439fb1 | 1095 | sb_start_ro_state_change(sb); |
1da177e4 LT |
1096 | } |
1097 | ||
f3a09c92 AV |
1098 | if (fc->ops->reconfigure) { |
1099 | retval = fc->ops->reconfigure(fc); | |
1100 | if (retval) { | |
1101 | if (!force) | |
1102 | goto cancel_readonly; | |
1103 | /* If forced remount, go ahead despite any errors */ | |
1104 | WARN(1, "forced remount of a %s fs returned %i\n", | |
1105 | sb->s_type->name, retval); | |
1106 | } | |
1da177e4 | 1107 | } |
8d0347f6 DH |
1108 | |
1109 | WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) | | |
1110 | (fc->sb_flags & fc->sb_flags_mask))); | |
d7439fb1 | 1111 | sb_end_ro_state_change(sb); |
c79d967d | 1112 | |
d208bbdd NP |
1113 | /* |
1114 | * Some filesystems modify their metadata via some other path than the | |
1115 | * bdev buffer cache (eg. use a private mapping, or directories in | |
1116 | * pagecache, etc). Also file data modifications go via their own | |
1117 | * mappings. So If we try to mount readonly then copy the filesystem | |
1118 | * from bdev, we could get stale data, so invalidate it to give a best | |
1119 | * effort at coherency. | |
1120 | */ | |
1121 | if (remount_ro && sb->s_bdev) | |
1122 | invalidate_bdev(sb->s_bdev); | |
1da177e4 | 1123 | return 0; |
4ed5e82f MS |
1124 | |
1125 | cancel_readonly: | |
d7439fb1 | 1126 | sb_end_ro_state_change(sb); |
4ed5e82f | 1127 | return retval; |
1da177e4 LT |
1128 | } |
1129 | ||
fa7c1d50 | 1130 | static void do_emergency_remount_callback(struct super_block *sb) |
1da177e4 | 1131 | { |
f0cd9880 | 1132 | bool locked = super_lock_excl(sb); |
5e874914 | 1133 | |
f0cd9880 | 1134 | if (locked && sb->s_root && sb->s_bdev && !sb_rdonly(sb)) { |
8d0347f6 DH |
1135 | struct fs_context *fc; |
1136 | ||
1137 | fc = fs_context_for_reconfigure(sb->s_root, | |
1138 | SB_RDONLY | SB_FORCE, SB_RDONLY); | |
1139 | if (!IS_ERR(fc)) { | |
1140 | if (parse_monolithic_mount_data(fc, NULL) == 0) | |
1141 | (void)reconfigure_super(fc); | |
1142 | put_fs_context(fc); | |
1143 | } | |
1da177e4 | 1144 | } |
f0cd9880 CB |
1145 | if (locked) |
1146 | super_unlock_excl(sb); | |
fa7c1d50 MG |
1147 | } |
1148 | ||
1149 | static void do_emergency_remount(struct work_struct *work) | |
1150 | { | |
1151 | __iterate_supers(do_emergency_remount_callback); | |
a2a9537a | 1152 | kfree(work); |
1da177e4 LT |
1153 | printk("Emergency Remount complete\n"); |
1154 | } | |
1155 | ||
1156 | void emergency_remount(void) | |
1157 | { | |
a2a9537a JA |
1158 | struct work_struct *work; |
1159 | ||
1160 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
1161 | if (work) { | |
1162 | INIT_WORK(work, do_emergency_remount); | |
1163 | schedule_work(work); | |
1164 | } | |
1da177e4 LT |
1165 | } |
1166 | ||
08fdc8a0 MG |
1167 | static void do_thaw_all_callback(struct super_block *sb) |
1168 | { | |
f0cd9880 | 1169 | bool locked = super_lock_excl(sb); |
5e874914 | 1170 | |
f0cd9880 | 1171 | if (locked && sb->s_root) { |
4a8b719f | 1172 | if (IS_ENABLED(CONFIG_BLOCK)) |
982c3b30 | 1173 | while (sb->s_bdev && !bdev_thaw(sb->s_bdev)) |
4a8b719f | 1174 | pr_warn("Emergency Thaw on %pg\n", sb->s_bdev); |
880b9577 | 1175 | thaw_super_locked(sb, FREEZE_HOLDER_USERSPACE); |
f0cd9880 | 1176 | return; |
08fdc8a0 | 1177 | } |
f0cd9880 CB |
1178 | if (locked) |
1179 | super_unlock_excl(sb); | |
08fdc8a0 MG |
1180 | } |
1181 | ||
1182 | static void do_thaw_all(struct work_struct *work) | |
1183 | { | |
1184 | __iterate_supers(do_thaw_all_callback); | |
1185 | kfree(work); | |
1186 | printk(KERN_WARNING "Emergency Thaw complete\n"); | |
1187 | } | |
1188 | ||
1189 | /** | |
1190 | * emergency_thaw_all -- forcibly thaw every frozen filesystem | |
1191 | * | |
1192 | * Used for emergency unfreeze of all filesystems via SysRq | |
1193 | */ | |
1194 | void emergency_thaw_all(void) | |
1195 | { | |
1196 | struct work_struct *work; | |
1197 | ||
1198 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
1199 | if (work) { | |
1200 | INIT_WORK(work, do_thaw_all); | |
1201 | schedule_work(work); | |
1202 | } | |
1203 | } | |
1204 | ||
ad76cbc6 | 1205 | static DEFINE_IDA(unnamed_dev_ida); |
1da177e4 | 1206 | |
5a66847e MW |
1207 | /** |
1208 | * get_anon_bdev - Allocate a block device for filesystems which don't have one. | |
1209 | * @p: Pointer to a dev_t. | |
1210 | * | |
1211 | * Filesystems which don't use real block devices can call this function | |
1212 | * to allocate a virtual block device. | |
1213 | * | |
1214 | * Context: Any context. Frequently called while holding sb_lock. | |
1215 | * Return: 0 on success, -EMFILE if there are no anonymous bdevs left | |
1216 | * or -ENOMEM if memory allocation failed. | |
1217 | */ | |
0ee5dc67 | 1218 | int get_anon_bdev(dev_t *p) |
1da177e4 LT |
1219 | { |
1220 | int dev; | |
5a66847e MW |
1221 | |
1222 | /* | |
1223 | * Many userspace utilities consider an FSID of 0 invalid. | |
1224 | * Always return at least 1 from get_anon_bdev. | |
1225 | */ | |
1226 | dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1, | |
1227 | GFP_ATOMIC); | |
1228 | if (dev == -ENOSPC) | |
1229 | dev = -EMFILE; | |
1230 | if (dev < 0) | |
1231 | return dev; | |
1232 | ||
1233 | *p = MKDEV(0, dev); | |
1da177e4 LT |
1234 | return 0; |
1235 | } | |
0ee5dc67 | 1236 | EXPORT_SYMBOL(get_anon_bdev); |
1da177e4 | 1237 | |
0ee5dc67 | 1238 | void free_anon_bdev(dev_t dev) |
1da177e4 | 1239 | { |
5a66847e | 1240 | ida_free(&unnamed_dev_ida, MINOR(dev)); |
1da177e4 | 1241 | } |
0ee5dc67 AV |
1242 | EXPORT_SYMBOL(free_anon_bdev); |
1243 | ||
1244 | int set_anon_super(struct super_block *s, void *data) | |
1245 | { | |
df0ce26c | 1246 | return get_anon_bdev(&s->s_dev); |
0ee5dc67 | 1247 | } |
0ee5dc67 AV |
1248 | EXPORT_SYMBOL(set_anon_super); |
1249 | ||
1250 | void kill_anon_super(struct super_block *sb) | |
1251 | { | |
1252 | dev_t dev = sb->s_dev; | |
1253 | generic_shutdown_super(sb); | |
dc3216b1 | 1254 | kill_super_notify(sb); |
0ee5dc67 AV |
1255 | free_anon_bdev(dev); |
1256 | } | |
1da177e4 LT |
1257 | EXPORT_SYMBOL(kill_anon_super); |
1258 | ||
1da177e4 LT |
1259 | void kill_litter_super(struct super_block *sb) |
1260 | { | |
1261 | if (sb->s_root) | |
1262 | d_genocide(sb->s_root); | |
1263 | kill_anon_super(sb); | |
1264 | } | |
1da177e4 LT |
1265 | EXPORT_SYMBOL(kill_litter_super); |
1266 | ||
cb50b348 AV |
1267 | int set_anon_super_fc(struct super_block *sb, struct fs_context *fc) |
1268 | { | |
1269 | return set_anon_super(sb, NULL); | |
1270 | } | |
1271 | EXPORT_SYMBOL(set_anon_super_fc); | |
1272 | ||
1273 | static int test_keyed_super(struct super_block *sb, struct fs_context *fc) | |
1274 | { | |
1275 | return sb->s_fs_info == fc->s_fs_info; | |
1276 | } | |
1277 | ||
1278 | static int test_single_super(struct super_block *s, struct fs_context *fc) | |
1279 | { | |
1280 | return 1; | |
1281 | } | |
1282 | ||
e062abae | 1283 | static int vfs_get_super(struct fs_context *fc, |
cda2ed05 CH |
1284 | int (*test)(struct super_block *, struct fs_context *), |
1285 | int (*fill_super)(struct super_block *sb, | |
1286 | struct fs_context *fc)) | |
cb50b348 | 1287 | { |
cb50b348 | 1288 | struct super_block *sb; |
43ce4c1f | 1289 | int err; |
cb50b348 | 1290 | |
cb50b348 AV |
1291 | sb = sget_fc(fc, test, set_anon_super_fc); |
1292 | if (IS_ERR(sb)) | |
1293 | return PTR_ERR(sb); | |
1294 | ||
1295 | if (!sb->s_root) { | |
43ce4c1f DH |
1296 | err = fill_super(sb, fc); |
1297 | if (err) | |
1298 | goto error; | |
cb50b348 AV |
1299 | |
1300 | sb->s_flags |= SB_ACTIVE; | |
1301 | } | |
1302 | ||
e062abae | 1303 | fc->root = dget(sb->s_root); |
cb50b348 | 1304 | return 0; |
43ce4c1f DH |
1305 | |
1306 | error: | |
1307 | deactivate_locked_super(sb); | |
1308 | return err; | |
cb50b348 | 1309 | } |
cb50b348 | 1310 | |
2ac295d4 AV |
1311 | int get_tree_nodev(struct fs_context *fc, |
1312 | int (*fill_super)(struct super_block *sb, | |
1313 | struct fs_context *fc)) | |
1314 | { | |
e062abae | 1315 | return vfs_get_super(fc, NULL, fill_super); |
2ac295d4 AV |
1316 | } |
1317 | EXPORT_SYMBOL(get_tree_nodev); | |
1318 | ||
c23a0bba AV |
1319 | int get_tree_single(struct fs_context *fc, |
1320 | int (*fill_super)(struct super_block *sb, | |
1321 | struct fs_context *fc)) | |
1322 | { | |
e062abae | 1323 | return vfs_get_super(fc, test_single_super, fill_super); |
c23a0bba AV |
1324 | } |
1325 | EXPORT_SYMBOL(get_tree_single); | |
1326 | ||
533770cc AV |
1327 | int get_tree_keyed(struct fs_context *fc, |
1328 | int (*fill_super)(struct super_block *sb, | |
1329 | struct fs_context *fc), | |
1330 | void *key) | |
1331 | { | |
1332 | fc->s_fs_info = key; | |
e062abae | 1333 | return vfs_get_super(fc, test_keyed_super, fill_super); |
533770cc AV |
1334 | } |
1335 | EXPORT_SYMBOL(get_tree_keyed); | |
1336 | ||
69881be3 CB |
1337 | static int set_bdev_super(struct super_block *s, void *data) |
1338 | { | |
1339 | s->s_dev = *(dev_t *)data; | |
1340 | return 0; | |
1341 | } | |
1342 | ||
1343 | static int super_s_dev_set(struct super_block *s, struct fs_context *fc) | |
1344 | { | |
1345 | return set_bdev_super(s, fc->sget_key); | |
1346 | } | |
1347 | ||
1348 | static int super_s_dev_test(struct super_block *s, struct fs_context *fc) | |
1349 | { | |
1350 | return !(s->s_iflags & SB_I_RETIRED) && | |
1351 | s->s_dev == *(dev_t *)fc->sget_key; | |
1352 | } | |
1353 | ||
1354 | /** | |
1355 | * sget_dev - Find or create a superblock by device number | |
1356 | * @fc: Filesystem context. | |
1357 | * @dev: device number | |
1358 | * | |
1359 | * Find or create a superblock using the provided device number that | |
1360 | * will be stored in fc->sget_key. | |
1361 | * | |
1362 | * If an extant superblock is matched, then that will be returned with | |
1363 | * an elevated reference count that the caller must transfer or discard. | |
1364 | * | |
1365 | * If no match is made, a new superblock will be allocated and basic | |
1366 | * initialisation will be performed (s_type, s_fs_info, s_id, s_dev will | |
1367 | * be set). The superblock will be published and it will be returned in | |
1368 | * a partially constructed state with SB_BORN and SB_ACTIVE as yet | |
1369 | * unset. | |
1370 | * | |
1371 | * Return: an existing or newly created superblock on success, an error | |
1372 | * pointer on failure. | |
1373 | */ | |
1374 | struct super_block *sget_dev(struct fs_context *fc, dev_t dev) | |
1375 | { | |
1376 | fc->sget_key = &dev; | |
1377 | return sget_fc(fc, super_s_dev_test, super_s_dev_set); | |
1378 | } | |
1379 | EXPORT_SYMBOL(sget_dev); | |
1380 | ||
9361401e | 1381 | #ifdef CONFIG_BLOCK |
9c09a7cf | 1382 | /* |
fd146410 JK |
1383 | * Lock the superblock that is holder of the bdev. Returns the superblock |
1384 | * pointer if we successfully locked the superblock and it is alive. Otherwise | |
1385 | * we return NULL and just unlock bdev->bd_holder_lock. | |
9c09a7cf | 1386 | * |
fd146410 | 1387 | * The function must be called with bdev->bd_holder_lock and releases it. |
9c09a7cf | 1388 | */ |
49ef8832 | 1389 | static struct super_block *bdev_super_lock(struct block_device *bdev, bool excl) |
fd146410 | 1390 | __releases(&bdev->bd_holder_lock) |
87efb390 | 1391 | { |
fd146410 | 1392 | struct super_block *sb = bdev->bd_holder; |
f0cd9880 | 1393 | bool locked; |
fd146410 JK |
1394 | |
1395 | lockdep_assert_held(&bdev->bd_holder_lock); | |
1396 | lockdep_assert_not_held(&sb->s_umount); | |
3b224e1d | 1397 | lockdep_assert_not_held(&bdev->bd_disk->open_mutex); |
fd146410 JK |
1398 | |
1399 | /* Make sure sb doesn't go away from under us */ | |
1400 | spin_lock(&sb_lock); | |
1401 | sb->s_count++; | |
1402 | spin_unlock(&sb_lock); | |
49ef8832 | 1403 | |
fd146410 | 1404 | mutex_unlock(&bdev->bd_holder_lock); |
87efb390 | 1405 | |
49ef8832 CB |
1406 | locked = super_lock(sb, excl); |
1407 | ||
fd146410 | 1408 | /* |
49ef8832 CB |
1409 | * If the superblock wasn't already SB_DYING then we hold |
1410 | * s_umount and can safely drop our temporary reference. | |
1411 | */ | |
fd146410 | 1412 | put_super(sb); |
49ef8832 CB |
1413 | |
1414 | if (!locked) | |
1415 | return NULL; | |
1416 | ||
1417 | if (!sb->s_root || !(sb->s_flags & SB_ACTIVE)) { | |
1418 | super_unlock(sb, excl); | |
1419 | return NULL; | |
1420 | } | |
1421 | ||
fd146410 | 1422 | return sb; |
9c09a7cf CH |
1423 | } |
1424 | ||
d8530de5 | 1425 | static void fs_bdev_mark_dead(struct block_device *bdev, bool surprise) |
87efb390 | 1426 | { |
fd146410 | 1427 | struct super_block *sb; |
9c09a7cf | 1428 | |
49ef8832 | 1429 | sb = bdev_super_lock(bdev, false); |
fd146410 | 1430 | if (!sb) |
87efb390 CH |
1431 | return; |
1432 | ||
d8530de5 CH |
1433 | if (!surprise) |
1434 | sync_filesystem(sb); | |
1435 | shrink_dcache_sb(sb); | |
e127b9bc | 1436 | invalidate_inodes(sb); |
87efb390 CH |
1437 | if (sb->s_op->shutdown) |
1438 | sb->s_op->shutdown(sb); | |
9c09a7cf | 1439 | |
0ed33598 | 1440 | super_unlock_shared(sb); |
87efb390 CH |
1441 | } |
1442 | ||
2142b88c CH |
1443 | static void fs_bdev_sync(struct block_device *bdev) |
1444 | { | |
fd146410 | 1445 | struct super_block *sb; |
2142b88c | 1446 | |
49ef8832 | 1447 | sb = bdev_super_lock(bdev, false); |
fd146410 | 1448 | if (!sb) |
2142b88c | 1449 | return; |
49ef8832 | 1450 | |
2142b88c | 1451 | sync_filesystem(sb); |
0ed33598 | 1452 | super_unlock_shared(sb); |
2142b88c CH |
1453 | } |
1454 | ||
49ef8832 CB |
1455 | static struct super_block *get_bdev_super(struct block_device *bdev) |
1456 | { | |
1457 | bool active = false; | |
1458 | struct super_block *sb; | |
1459 | ||
1460 | sb = bdev_super_lock(bdev, true); | |
1461 | if (sb) { | |
1462 | active = atomic_inc_not_zero(&sb->s_active); | |
1463 | super_unlock_excl(sb); | |
1464 | } | |
1465 | if (!active) | |
1466 | return NULL; | |
1467 | return sb; | |
1468 | } | |
1469 | ||
1470 | static int fs_bdev_freeze(struct block_device *bdev) | |
1471 | { | |
1472 | struct super_block *sb; | |
1473 | int error = 0; | |
1474 | ||
1475 | lockdep_assert_held(&bdev->bd_fsfreeze_mutex); | |
1476 | ||
1477 | sb = get_bdev_super(bdev); | |
1478 | if (!sb) | |
1479 | return -EINVAL; | |
1480 | ||
1481 | if (sb->s_op->freeze_super) | |
1482 | error = sb->s_op->freeze_super(sb, FREEZE_HOLDER_USERSPACE); | |
1483 | else | |
1484 | error = freeze_super(sb, FREEZE_HOLDER_USERSPACE); | |
1485 | if (!error) | |
1486 | error = sync_blockdev(bdev); | |
1487 | deactivate_super(sb); | |
1488 | return error; | |
1489 | } | |
1490 | ||
1491 | static int fs_bdev_thaw(struct block_device *bdev) | |
1492 | { | |
1493 | struct super_block *sb; | |
1494 | int error; | |
1495 | ||
1496 | lockdep_assert_held(&bdev->bd_fsfreeze_mutex); | |
1497 | ||
1498 | sb = get_bdev_super(bdev); | |
1499 | if (WARN_ON_ONCE(!sb)) | |
1500 | return -EINVAL; | |
1501 | ||
1502 | if (sb->s_op->thaw_super) | |
1503 | error = sb->s_op->thaw_super(sb, FREEZE_HOLDER_USERSPACE); | |
1504 | else | |
1505 | error = thaw_super(sb, FREEZE_HOLDER_USERSPACE); | |
1506 | deactivate_super(sb); | |
1507 | return error; | |
1508 | } | |
1509 | ||
7ecd0b6f | 1510 | const struct blk_holder_ops fs_holder_ops = { |
d8530de5 | 1511 | .mark_dead = fs_bdev_mark_dead, |
2142b88c | 1512 | .sync = fs_bdev_sync, |
49ef8832 CB |
1513 | .freeze = fs_bdev_freeze, |
1514 | .thaw = fs_bdev_thaw, | |
87efb390 | 1515 | }; |
7ecd0b6f | 1516 | EXPORT_SYMBOL_GPL(fs_holder_ops); |
fe62c3a4 | 1517 | |
cf6da236 | 1518 | int setup_bdev_super(struct super_block *sb, int sb_flags, |
aca740ce JK |
1519 | struct fs_context *fc) |
1520 | { | |
1521 | blk_mode_t mode = sb_open_mode(sb_flags); | |
f4a48bc3 | 1522 | struct bdev_handle *bdev_handle; |
aca740ce JK |
1523 | struct block_device *bdev; |
1524 | ||
f4a48bc3 JK |
1525 | bdev_handle = bdev_open_by_dev(sb->s_dev, mode, sb, &fs_holder_ops); |
1526 | if (IS_ERR(bdev_handle)) { | |
aca740ce JK |
1527 | if (fc) |
1528 | errorf(fc, "%s: Can't open blockdev", fc->source); | |
f4a48bc3 | 1529 | return PTR_ERR(bdev_handle); |
aca740ce | 1530 | } |
f4a48bc3 | 1531 | bdev = bdev_handle->bdev; |
aca740ce JK |
1532 | |
1533 | /* | |
1534 | * This really should be in blkdev_get_by_dev, but right now can't due | |
1535 | * to legacy issues that require us to allow opening a block device node | |
1536 | * writable from userspace even for a read-only block device. | |
1537 | */ | |
1538 | if ((mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) { | |
f4a48bc3 | 1539 | bdev_release(bdev_handle); |
aca740ce JK |
1540 | return -EACCES; |
1541 | } | |
1542 | ||
1543 | /* | |
49ef8832 CB |
1544 | * It is enough to check bdev was not frozen before we set |
1545 | * s_bdev as freezing will wait until SB_BORN is set. | |
aca740ce | 1546 | */ |
49ef8832 | 1547 | if (atomic_read(&bdev->bd_fsfreeze_count) > 0) { |
aca740ce JK |
1548 | if (fc) |
1549 | warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev); | |
f4a48bc3 | 1550 | bdev_release(bdev_handle); |
aca740ce JK |
1551 | return -EBUSY; |
1552 | } | |
1553 | spin_lock(&sb_lock); | |
f4a48bc3 | 1554 | sb->s_bdev_handle = bdev_handle; |
aca740ce JK |
1555 | sb->s_bdev = bdev; |
1556 | sb->s_bdi = bdi_get(bdev->bd_disk->bdi); | |
1557 | if (bdev_stable_writes(bdev)) | |
1558 | sb->s_iflags |= SB_I_STABLE_WRITES; | |
1559 | spin_unlock(&sb_lock); | |
aca740ce JK |
1560 | |
1561 | snprintf(sb->s_id, sizeof(sb->s_id), "%pg", bdev); | |
1720f5dd | 1562 | shrinker_debugfs_rename(sb->s_shrink, "sb-%s:%s", sb->s_type->name, |
aca740ce JK |
1563 | sb->s_id); |
1564 | sb_set_blocksize(sb, block_size(bdev)); | |
1565 | return 0; | |
fe62c3a4 | 1566 | } |
cf6da236 | 1567 | EXPORT_SYMBOL_GPL(setup_bdev_super); |
fe62c3a4 DH |
1568 | |
1569 | /** | |
1570 | * get_tree_bdev - Get a superblock based on a single block device | |
1571 | * @fc: The filesystem context holding the parameters | |
1572 | * @fill_super: Helper to initialise a new superblock | |
1573 | */ | |
1574 | int get_tree_bdev(struct fs_context *fc, | |
1575 | int (*fill_super)(struct super_block *, | |
1576 | struct fs_context *)) | |
1577 | { | |
fe62c3a4 | 1578 | struct super_block *s; |
fe62c3a4 | 1579 | int error = 0; |
aca740ce | 1580 | dev_t dev; |
fe62c3a4 | 1581 | |
fe62c3a4 DH |
1582 | if (!fc->source) |
1583 | return invalf(fc, "No source specified"); | |
1584 | ||
aca740ce JK |
1585 | error = lookup_bdev(fc->source, &dev); |
1586 | if (error) { | |
1587 | errorf(fc, "%s: Can't lookup blockdev", fc->source); | |
1588 | return error; | |
fe62c3a4 DH |
1589 | } |
1590 | ||
1591 | fc->sb_flags |= SB_NOSEC; | |
69881be3 | 1592 | s = sget_dev(fc, dev); |
aca740ce | 1593 | if (IS_ERR(s)) |
fe62c3a4 DH |
1594 | return PTR_ERR(s); |
1595 | ||
1596 | if (s->s_root) { | |
1597 | /* Don't summarily change the RO/RW state. */ | |
1598 | if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) { | |
aca740ce | 1599 | warnf(fc, "%pg: Can't mount, would change RO state", s->s_bdev); |
fe62c3a4 | 1600 | deactivate_locked_super(s); |
fe62c3a4 DH |
1601 | return -EBUSY; |
1602 | } | |
aca740ce | 1603 | } else { |
aca740ce | 1604 | error = setup_bdev_super(s, fc->sb_flags, fc); |
aca740ce JK |
1605 | if (!error) |
1606 | error = fill_super(s, fc); | |
fe62c3a4 DH |
1607 | if (error) { |
1608 | deactivate_locked_super(s); | |
1609 | return error; | |
1610 | } | |
fe62c3a4 | 1611 | s->s_flags |= SB_ACTIVE; |
fe62c3a4 DH |
1612 | } |
1613 | ||
1614 | BUG_ON(fc->root); | |
1615 | fc->root = dget(s->s_root); | |
1616 | return 0; | |
1617 | } | |
1618 | EXPORT_SYMBOL(get_tree_bdev); | |
1619 | ||
1da177e4 LT |
1620 | static int test_bdev_super(struct super_block *s, void *data) |
1621 | { | |
aca740ce | 1622 | return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data; |
1da177e4 LT |
1623 | } |
1624 | ||
152a0836 | 1625 | struct dentry *mount_bdev(struct file_system_type *fs_type, |
1da177e4 | 1626 | int flags, const char *dev_name, void *data, |
152a0836 | 1627 | int (*fill_super)(struct super_block *, void *, int)) |
1da177e4 | 1628 | { |
1da177e4 | 1629 | struct super_block *s; |
aca740ce JK |
1630 | int error; |
1631 | dev_t dev; | |
1da177e4 | 1632 | |
aca740ce JK |
1633 | error = lookup_bdev(dev_name, &dev); |
1634 | if (error) | |
1635 | return ERR_PTR(error); | |
1da177e4 | 1636 | |
aca740ce JK |
1637 | flags |= SB_NOSEC; |
1638 | s = sget(fs_type, test_bdev_super, set_bdev_super, flags, &dev); | |
1da177e4 | 1639 | if (IS_ERR(s)) |
aca740ce | 1640 | return ERR_CAST(s); |
1da177e4 LT |
1641 | |
1642 | if (s->s_root) { | |
e462ec50 | 1643 | if ((flags ^ s->s_flags) & SB_RDONLY) { |
74dbbdd7 | 1644 | deactivate_locked_super(s); |
aca740ce | 1645 | return ERR_PTR(-EBUSY); |
1da177e4 | 1646 | } |
aca740ce | 1647 | } else { |
aca740ce | 1648 | error = setup_bdev_super(s, flags, NULL); |
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 | } |
1661 | EXPORT_SYMBOL(mount_bdev); | |
1662 | ||
1da177e4 LT |
1663 | void 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 | ||
1674 | EXPORT_SYMBOL(kill_block_super); | |
9361401e | 1675 | #endif |
1da177e4 | 1676 | |
3c26ff6e | 1677 | struct 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 |
1695 | EXPORT_SYMBOL(mount_nodev); |
1696 | ||
a6097180 N |
1697 | int 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); | |
1717 | out: | |
1718 | put_fs_context(fc); | |
1719 | return ret; | |
1720 | } | |
1721 | ||
1da177e4 LT |
1722 | static int compare_single(struct super_block *s, void *p) |
1723 | { | |
1724 | return 1; | |
1725 | } | |
1726 | ||
fc14f2fe | 1727 | struct 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 | } | |
1750 | EXPORT_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 | */ | |
1760 | int 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 | 1813 | EXPORT_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 | */ | |
1819 | int 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 | } | |
1842 | EXPORT_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 | */ | |
1848 | int 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 | } | |
1855 | EXPORT_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 | */ |
1865 | static 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 | */ | |
1874 | static 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 | */ | |
1885 | static 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 | 1893 | static 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 |
1899 | static 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 | 1961 | int freeze_super(struct super_block *sb, enum freeze_holder who) |
18e9e510 JB |
1962 | { |
1963 | int ret; | |
1964 | ||
f0cd9880 CB |
1965 | if (!super_lock_excl(sb)) { |
1966 | WARN_ON_ONCE("Dying superblock while freezing!"); | |
1967 | return -EINVAL; | |
1968 | } | |
18e9e510 | 1969 | atomic_inc(&sb->s_active); |
051178c3 | 1970 | |
59ba4fdd | 1971 | retry: |
880b9577 DW |
1972 | if (sb->s_writers.frozen == SB_FREEZE_COMPLETE) { |
1973 | if (sb->s_writers.freeze_holders & who) { | |
1974 | deactivate_locked_super(sb); | |
1975 | return -EBUSY; | |
1976 | } | |
1977 | ||
1978 | WARN_ON(sb->s_writers.freeze_holders == 0); | |
1979 | ||
1980 | /* | |
1981 | * Someone else already holds this type of freeze; share the | |
1982 | * freeze and assign the active ref to the freeze. | |
1983 | */ | |
1984 | sb->s_writers.freeze_holders |= who; | |
3fb5a656 | 1985 | super_unlock_excl(sb); |
880b9577 DW |
1986 | return 0; |
1987 | } | |
1988 | ||
5accdf82 | 1989 | if (sb->s_writers.frozen != SB_UNFROZEN) { |
59ba4fdd DW |
1990 | ret = wait_for_partially_frozen(sb); |
1991 | if (ret) { | |
1992 | deactivate_locked_super(sb); | |
1993 | return ret; | |
1994 | } | |
1995 | ||
1996 | goto retry; | |
18e9e510 JB |
1997 | } |
1998 | ||
e462ec50 | 1999 | if (!(sb->s_flags & SB_BORN)) { |
0ed33598 | 2000 | super_unlock_excl(sb); |
dabe0dc1 AV |
2001 | return 0; /* sic - it's "nothing to do" */ |
2002 | } | |
2003 | ||
bc98a42c | 2004 | if (sb_rdonly(sb)) { |
5accdf82 | 2005 | /* Nothing to do really... */ |
880b9577 | 2006 | sb->s_writers.freeze_holders |= who; |
5accdf82 | 2007 | sb->s_writers.frozen = SB_FREEZE_COMPLETE; |
59ba4fdd | 2008 | wake_up_var(&sb->s_writers.frozen); |
0ed33598 | 2009 | super_unlock_excl(sb); |
18e9e510 JB |
2010 | return 0; |
2011 | } | |
2012 | ||
5accdf82 | 2013 | sb->s_writers.frozen = SB_FREEZE_WRITE; |
5accdf82 | 2014 | /* Release s_umount to preserve sb_start_write -> s_umount ordering */ |
0ed33598 | 2015 | super_unlock_excl(sb); |
5accdf82 | 2016 | sb_wait_write(sb, SB_FREEZE_WRITE); |
f0cd9880 CB |
2017 | if (!super_lock_excl(sb)) { |
2018 | WARN_ON_ONCE("Dying superblock while freezing!"); | |
2019 | return -EINVAL; | |
2020 | } | |
5accdf82 JK |
2021 | |
2022 | /* Now we go and block page faults... */ | |
5accdf82 | 2023 | sb->s_writers.frozen = SB_FREEZE_PAGEFAULT; |
5accdf82 JK |
2024 | sb_wait_write(sb, SB_FREEZE_PAGEFAULT); |
2025 | ||
2026 | /* All writers are done so after syncing there won't be dirty data */ | |
2719c716 DW |
2027 | ret = sync_filesystem(sb); |
2028 | if (ret) { | |
2029 | sb->s_writers.frozen = SB_UNFROZEN; | |
2030 | sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT); | |
59ba4fdd | 2031 | wake_up_var(&sb->s_writers.frozen); |
2719c716 DW |
2032 | deactivate_locked_super(sb); |
2033 | return ret; | |
2034 | } | |
18e9e510 | 2035 | |
5accdf82 JK |
2036 | /* Now wait for internal filesystem counter */ |
2037 | sb->s_writers.frozen = SB_FREEZE_FS; | |
5accdf82 | 2038 | sb_wait_write(sb, SB_FREEZE_FS); |
18e9e510 | 2039 | |
18e9e510 JB |
2040 | if (sb->s_op->freeze_fs) { |
2041 | ret = sb->s_op->freeze_fs(sb); | |
2042 | if (ret) { | |
2043 | printk(KERN_ERR | |
2044 | "VFS:Filesystem freeze failed\n"); | |
5accdf82 | 2045 | sb->s_writers.frozen = SB_UNFROZEN; |
2719c716 | 2046 | sb_freeze_unlock(sb, SB_FREEZE_FS); |
59ba4fdd | 2047 | wake_up_var(&sb->s_writers.frozen); |
18e9e510 JB |
2048 | deactivate_locked_super(sb); |
2049 | return ret; | |
2050 | } | |
2051 | } | |
5accdf82 | 2052 | /* |
89f39af1 ON |
2053 | * For debugging purposes so that fs can warn if it sees write activity |
2054 | * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super(). | |
5accdf82 | 2055 | */ |
880b9577 | 2056 | sb->s_writers.freeze_holders |= who; |
5accdf82 | 2057 | sb->s_writers.frozen = SB_FREEZE_COMPLETE; |
59ba4fdd | 2058 | wake_up_var(&sb->s_writers.frozen); |
f1a96220 | 2059 | lockdep_sb_freeze_release(sb); |
0ed33598 | 2060 | super_unlock_excl(sb); |
18e9e510 JB |
2061 | return 0; |
2062 | } | |
2063 | EXPORT_SYMBOL(freeze_super); | |
2064 | ||
880b9577 DW |
2065 | /* |
2066 | * Undoes the effect of a freeze_super_locked call. If the filesystem is | |
2067 | * frozen both by userspace and the kernel, a thaw call from either source | |
2068 | * removes that state without releasing the other state or unlocking the | |
2069 | * filesystem. | |
2070 | */ | |
2071 | static int thaw_super_locked(struct super_block *sb, enum freeze_holder who) | |
18e9e510 JB |
2072 | { |
2073 | int error; | |
2074 | ||
880b9577 DW |
2075 | if (sb->s_writers.frozen == SB_FREEZE_COMPLETE) { |
2076 | if (!(sb->s_writers.freeze_holders & who)) { | |
3fb5a656 | 2077 | super_unlock_excl(sb); |
880b9577 DW |
2078 | return -EINVAL; |
2079 | } | |
2080 | ||
2081 | /* | |
2082 | * Freeze is shared with someone else. Release our hold and | |
2083 | * drop the active ref that freeze_super assigned to the | |
2084 | * freezer. | |
2085 | */ | |
2086 | if (sb->s_writers.freeze_holders & ~who) { | |
2087 | sb->s_writers.freeze_holders &= ~who; | |
2088 | deactivate_locked_super(sb); | |
2089 | return 0; | |
2090 | } | |
2091 | } else { | |
0ed33598 | 2092 | super_unlock_excl(sb); |
18e9e510 JB |
2093 | return -EINVAL; |
2094 | } | |
2095 | ||
bc98a42c | 2096 | if (sb_rdonly(sb)) { |
880b9577 | 2097 | sb->s_writers.freeze_holders &= ~who; |
8129ed29 | 2098 | sb->s_writers.frozen = SB_UNFROZEN; |
59ba4fdd | 2099 | wake_up_var(&sb->s_writers.frozen); |
18e9e510 | 2100 | goto out; |
8129ed29 | 2101 | } |
18e9e510 | 2102 | |
f1a96220 ON |
2103 | lockdep_sb_freeze_acquire(sb); |
2104 | ||
18e9e510 JB |
2105 | if (sb->s_op->unfreeze_fs) { |
2106 | error = sb->s_op->unfreeze_fs(sb); | |
2107 | if (error) { | |
3fb5a656 | 2108 | printk(KERN_ERR "VFS:Filesystem thaw failed\n"); |
f1a96220 | 2109 | lockdep_sb_freeze_release(sb); |
0ed33598 | 2110 | super_unlock_excl(sb); |
18e9e510 JB |
2111 | return error; |
2112 | } | |
2113 | } | |
2114 | ||
880b9577 | 2115 | sb->s_writers.freeze_holders &= ~who; |
5accdf82 | 2116 | sb->s_writers.frozen = SB_UNFROZEN; |
59ba4fdd | 2117 | wake_up_var(&sb->s_writers.frozen); |
2719c716 | 2118 | sb_freeze_unlock(sb, SB_FREEZE_FS); |
8129ed29 | 2119 | out: |
18e9e510 | 2120 | deactivate_locked_super(sb); |
18e9e510 JB |
2121 | return 0; |
2122 | } | |
08fdc8a0 | 2123 | |
961f3c89 MCC |
2124 | /** |
2125 | * thaw_super -- unlock filesystem | |
2126 | * @sb: the super to thaw | |
880b9577 DW |
2127 | * @who: context that wants to freeze |
2128 | * | |
2129 | * Unlocks the filesystem and marks it writeable again after freeze_super() | |
2130 | * if there are no remaining freezes on the filesystem. | |
961f3c89 | 2131 | * |
880b9577 DW |
2132 | * @who should be: |
2133 | * * %FREEZE_HOLDER_USERSPACE if userspace wants to thaw the fs; | |
2134 | * * %FREEZE_HOLDER_KERNEL if the kernel wants to thaw the fs. | |
961f3c89 | 2135 | */ |
880b9577 | 2136 | int thaw_super(struct super_block *sb, enum freeze_holder who) |
08fdc8a0 | 2137 | { |
f0cd9880 CB |
2138 | if (!super_lock_excl(sb)) { |
2139 | WARN_ON_ONCE("Dying superblock while thawing!"); | |
2140 | return -EINVAL; | |
2141 | } | |
880b9577 | 2142 | return thaw_super_locked(sb, who); |
08fdc8a0 | 2143 | } |
18e9e510 | 2144 | EXPORT_SYMBOL(thaw_super); |
439bc39b CH |
2145 | |
2146 | /* | |
2147 | * Create workqueue for deferred direct IO completions. We allocate the | |
2148 | * workqueue when it's first needed. This avoids creating workqueue for | |
2149 | * filesystems that don't need it and also allows us to create the workqueue | |
2150 | * late enough so the we can include s_id in the name of the workqueue. | |
2151 | */ | |
2152 | int sb_init_dio_done_wq(struct super_block *sb) | |
2153 | { | |
2154 | struct workqueue_struct *old; | |
2155 | struct workqueue_struct *wq = alloc_workqueue("dio/%s", | |
2156 | WQ_MEM_RECLAIM, 0, | |
2157 | sb->s_id); | |
2158 | if (!wq) | |
2159 | return -ENOMEM; | |
2160 | /* | |
2161 | * This has to be atomic as more DIOs can race to create the workqueue | |
2162 | */ | |
2163 | old = cmpxchg(&sb->s_dio_done_wq, NULL, wq); | |
2164 | /* Someone created workqueue before us? Free ours... */ | |
2165 | if (old) | |
2166 | destroy_workqueue(wq); | |
2167 | return 0; | |
2168 | } | |
389a4a4a | 2169 | EXPORT_SYMBOL_GPL(sb_init_dio_done_wq); |