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 | |
08fdc8a0 | 42 | static int thaw_super_locked(struct super_block *sb); |
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 | * | |
108 | * Return: This returns true if SB_BORN was set, false if SB_DYING was | |
109 | * set. The function acquires s_umount and returns with it held. | |
110 | */ | |
111 | static __must_check bool super_lock(struct super_block *sb, bool excl) | |
112 | { | |
113 | ||
114 | lockdep_assert_not_held(&sb->s_umount); | |
115 | ||
116 | relock: | |
117 | __super_lock(sb, excl); | |
118 | ||
119 | /* | |
120 | * Has gone through generic_shutdown_super() in the meantime. | |
121 | * @sb->s_root is NULL and @sb->s_active is 0. No one needs to | |
122 | * grab a reference to this. Tell them so. | |
123 | */ | |
124 | if (sb->s_flags & SB_DYING) | |
125 | return false; | |
126 | ||
127 | /* Has called ->get_tree() successfully. */ | |
128 | if (sb->s_flags & SB_BORN) | |
129 | return true; | |
130 | ||
131 | super_unlock(sb, excl); | |
132 | ||
133 | /* wait until the superblock is ready or dying */ | |
134 | wait_var_event(&sb->s_flags, wait_born(sb)); | |
135 | ||
136 | /* | |
137 | * Neither SB_BORN nor SB_DYING are ever unset so we never loop. | |
138 | * Just reacquire @sb->s_umount for the caller. | |
139 | */ | |
140 | goto relock; | |
141 | } | |
142 | ||
143 | /* wait and acquire read-side of @sb->s_umount */ | |
144 | static inline bool super_lock_shared(struct super_block *sb) | |
145 | { | |
146 | return super_lock(sb, false); | |
147 | } | |
148 | ||
149 | /* wait and acquire write-side of @sb->s_umount */ | |
150 | static inline bool super_lock_excl(struct super_block *sb) | |
151 | { | |
152 | return super_lock(sb, true); | |
153 | } | |
154 | ||
155 | /* wake waiters */ | |
156 | #define SUPER_WAKE_FLAGS (SB_BORN | SB_DYING) | |
157 | static void super_wake(struct super_block *sb, unsigned int flag) | |
158 | { | |
159 | WARN_ON_ONCE((flag & ~SUPER_WAKE_FLAGS)); | |
160 | WARN_ON_ONCE(hweight32(flag & SUPER_WAKE_FLAGS) > 1); | |
161 | ||
162 | /* | |
163 | * Pairs with smp_load_acquire() in super_lock() to make sure | |
164 | * all initializations in the superblock are seen by the user | |
165 | * seeing SB_BORN sent. | |
166 | */ | |
167 | smp_store_release(&sb->s_flags, sb->s_flags | flag); | |
168 | /* | |
169 | * Pairs with the barrier in prepare_to_wait_event() to make sure | |
170 | * ___wait_var_event() either sees SB_BORN set or | |
171 | * waitqueue_active() check in wake_up_var() sees the waiter. | |
172 | */ | |
173 | smp_mb(); | |
174 | wake_up_var(&sb->s_flags); | |
175 | } | |
176 | ||
b0d40c92 DC |
177 | /* |
178 | * One thing we have to be careful of with a per-sb shrinker is that we don't | |
179 | * drop the last active reference to the superblock from within the shrinker. | |
180 | * If that happens we could trigger unregistering the shrinker from within the | |
47a7c01c | 181 | * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we |
b0d40c92 DC |
182 | * take a passive reference to the superblock to avoid this from occurring. |
183 | */ | |
0a234c6d DC |
184 | static unsigned long super_cache_scan(struct shrinker *shrink, |
185 | struct shrink_control *sc) | |
b0d40c92 DC |
186 | { |
187 | struct super_block *sb; | |
0a234c6d DC |
188 | long fs_objects = 0; |
189 | long total_objects; | |
190 | long freed = 0; | |
191 | long dentries; | |
192 | long inodes; | |
b0d40c92 DC |
193 | |
194 | sb = container_of(shrink, struct super_block, s_shrink); | |
195 | ||
196 | /* | |
197 | * Deadlock avoidance. We may hold various FS locks, and we don't want | |
198 | * to recurse into the FS that called us in clear_inode() and friends.. | |
199 | */ | |
0a234c6d DC |
200 | if (!(sc->gfp_mask & __GFP_FS)) |
201 | return SHRINK_STOP; | |
b0d40c92 | 202 | |
d8ce82ef | 203 | if (!super_trylock_shared(sb)) |
0a234c6d | 204 | return SHRINK_STOP; |
b0d40c92 | 205 | |
d0407903 | 206 | if (sb->s_op->nr_cached_objects) |
4101b624 | 207 | fs_objects = sb->s_op->nr_cached_objects(sb, sc); |
0e1fdafd | 208 | |
503c358c VD |
209 | inodes = list_lru_shrink_count(&sb->s_inode_lru, sc); |
210 | dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc); | |
f6041567 | 211 | total_objects = dentries + inodes + fs_objects + 1; |
475d0db7 TH |
212 | if (!total_objects) |
213 | total_objects = 1; | |
0e1fdafd | 214 | |
0a234c6d | 215 | /* proportion the scan between the caches */ |
f6041567 | 216 | dentries = mult_frac(sc->nr_to_scan, dentries, total_objects); |
bc3b14cb | 217 | inodes = mult_frac(sc->nr_to_scan, inodes, total_objects); |
503c358c | 218 | fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects); |
b0d40c92 | 219 | |
0a234c6d DC |
220 | /* |
221 | * prune the dcache first as the icache is pinned by it, then | |
222 | * prune the icache, followed by the filesystem specific caches | |
49e7e7ff VD |
223 | * |
224 | * Ensure that we always scan at least one object - memcg kmem | |
225 | * accounting uses this to fully empty the caches. | |
0a234c6d | 226 | */ |
49e7e7ff | 227 | sc->nr_to_scan = dentries + 1; |
503c358c | 228 | freed = prune_dcache_sb(sb, sc); |
49e7e7ff | 229 | sc->nr_to_scan = inodes + 1; |
503c358c | 230 | freed += prune_icache_sb(sb, sc); |
0a234c6d DC |
231 | |
232 | if (fs_objects) { | |
49e7e7ff | 233 | sc->nr_to_scan = fs_objects + 1; |
4101b624 | 234 | freed += sb->s_op->free_cached_objects(sb, sc); |
b0d40c92 DC |
235 | } |
236 | ||
0ed33598 | 237 | super_unlock_shared(sb); |
0a234c6d DC |
238 | return freed; |
239 | } | |
240 | ||
241 | static unsigned long super_cache_count(struct shrinker *shrink, | |
242 | struct shrink_control *sc) | |
243 | { | |
244 | struct super_block *sb; | |
245 | long total_objects = 0; | |
246 | ||
247 | sb = container_of(shrink, struct super_block, s_shrink); | |
248 | ||
d23da150 | 249 | /* |
d8ce82ef CB |
250 | * We don't call super_trylock_shared() here as it is a scalability |
251 | * bottleneck, so we're exposed to partial setup state. The shrinker | |
252 | * rwsem does not protect filesystem operations backing | |
253 | * list_lru_shrink_count() or s_op->nr_cached_objects(). Counts can | |
254 | * change between super_cache_count and super_cache_scan, so we really | |
255 | * don't need locks here. | |
79f546a6 DC |
256 | * |
257 | * However, if we are currently mounting the superblock, the underlying | |
258 | * filesystem might be in a state of partial construction and hence it | |
d8ce82ef CB |
259 | * is dangerous to access it. super_trylock_shared() uses a SB_BORN check |
260 | * to avoid this situation, so do the same here. The memory barrier is | |
79f546a6 | 261 | * matched with the one in mount_fs() as we don't hold locks here. |
d23da150 | 262 | */ |
79f546a6 DC |
263 | if (!(sb->s_flags & SB_BORN)) |
264 | return 0; | |
265 | smp_rmb(); | |
266 | ||
0a234c6d | 267 | if (sb->s_op && sb->s_op->nr_cached_objects) |
4101b624 | 268 | total_objects = sb->s_op->nr_cached_objects(sb, sc); |
0a234c6d | 269 | |
503c358c VD |
270 | total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc); |
271 | total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc); | |
0a234c6d | 272 | |
9b996468 KT |
273 | if (!total_objects) |
274 | return SHRINK_EMPTY; | |
275 | ||
55f841ce | 276 | total_objects = vfs_pressure_ratio(total_objects); |
0e1fdafd | 277 | return total_objects; |
b0d40c92 DC |
278 | } |
279 | ||
853b39a7 ON |
280 | static void destroy_super_work(struct work_struct *work) |
281 | { | |
282 | struct super_block *s = container_of(work, struct super_block, | |
283 | destroy_work); | |
284 | int i; | |
285 | ||
286 | for (i = 0; i < SB_FREEZE_LEVELS; i++) | |
8129ed29 | 287 | percpu_free_rwsem(&s->s_writers.rw_sem[i]); |
853b39a7 ON |
288 | kfree(s); |
289 | } | |
290 | ||
291 | static void destroy_super_rcu(struct rcu_head *head) | |
292 | { | |
293 | struct super_block *s = container_of(head, struct super_block, rcu); | |
294 | INIT_WORK(&s->destroy_work, destroy_super_work); | |
295 | schedule_work(&s->destroy_work); | |
296 | } | |
297 | ||
0200894d AV |
298 | /* Free a superblock that has never been seen by anyone */ |
299 | static void destroy_unused_super(struct super_block *s) | |
5accdf82 | 300 | { |
0200894d AV |
301 | if (!s) |
302 | return; | |
0ed33598 | 303 | super_unlock_excl(s); |
7eb5e882 AV |
304 | list_lru_destroy(&s->s_dentry_lru); |
305 | list_lru_destroy(&s->s_inode_lru); | |
7eb5e882 | 306 | security_sb_free(s); |
6e4eab57 | 307 | put_user_ns(s->s_user_ns); |
7eb5e882 | 308 | kfree(s->s_subtype); |
8e04944f | 309 | free_prealloced_shrinker(&s->s_shrink); |
0200894d AV |
310 | /* no delays needed */ |
311 | destroy_super_work(&s->destroy_work); | |
5accdf82 JK |
312 | } |
313 | ||
1da177e4 LT |
314 | /** |
315 | * alloc_super - create new superblock | |
fe2bbc48 | 316 | * @type: filesystem type superblock should belong to |
9249e17f | 317 | * @flags: the mount flags |
6e4eab57 | 318 | * @user_ns: User namespace for the super_block |
1da177e4 LT |
319 | * |
320 | * Allocates and initializes a new &struct super_block. alloc_super() | |
321 | * returns a pointer new superblock or %NULL if allocation had failed. | |
322 | */ | |
6e4eab57 EB |
323 | static struct super_block *alloc_super(struct file_system_type *type, int flags, |
324 | struct user_namespace *user_ns) | |
1da177e4 | 325 | { |
11b0b5ab | 326 | struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER); |
b87221de | 327 | static const struct super_operations default_op; |
7eb5e882 AV |
328 | int i; |
329 | ||
330 | if (!s) | |
331 | return NULL; | |
1da177e4 | 332 | |
b5bd856a | 333 | INIT_LIST_HEAD(&s->s_mounts); |
6e4eab57 | 334 | s->s_user_ns = get_user_ns(user_ns); |
ca0168e8 AV |
335 | init_rwsem(&s->s_umount); |
336 | lockdep_set_class(&s->s_umount, &type->s_umount_key); | |
337 | /* | |
338 | * sget() can have s_umount recursion. | |
339 | * | |
340 | * When it cannot find a suitable sb, it allocates a new | |
341 | * one (this one), and tries again to find a suitable old | |
342 | * one. | |
343 | * | |
344 | * In case that succeeds, it will acquire the s_umount | |
345 | * lock of the old one. Since these are clearly distrinct | |
346 | * locks, and this object isn't exposed yet, there's no | |
347 | * risk of deadlocks. | |
348 | * | |
349 | * Annotate this by putting this lock in a different | |
350 | * subclass. | |
351 | */ | |
352 | down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING); | |
b5bd856a | 353 | |
7eb5e882 AV |
354 | if (security_sb_alloc(s)) |
355 | goto fail; | |
7b7a8665 | 356 | |
7eb5e882 | 357 | for (i = 0; i < SB_FREEZE_LEVELS; i++) { |
8129ed29 ON |
358 | if (__percpu_init_rwsem(&s->s_writers.rw_sem[i], |
359 | sb_writers_name[i], | |
360 | &type->s_writers_key[i])) | |
7eb5e882 | 361 | goto fail; |
1da177e4 | 362 | } |
df0ce26c | 363 | s->s_bdi = &noop_backing_dev_info; |
7eb5e882 | 364 | s->s_flags = flags; |
cc50a07a | 365 | if (s->s_user_ns != &init_user_ns) |
67690f93 | 366 | s->s_iflags |= SB_I_NODEV; |
7eb5e882 | 367 | INIT_HLIST_NODE(&s->s_instances); |
f1ee6162 | 368 | INIT_HLIST_BL_HEAD(&s->s_roots); |
e97fedb9 | 369 | mutex_init(&s->s_sync_lock); |
7eb5e882 | 370 | INIT_LIST_HEAD(&s->s_inodes); |
74278da9 | 371 | spin_lock_init(&s->s_inode_list_lock); |
6c60d2b5 DC |
372 | INIT_LIST_HEAD(&s->s_inodes_wb); |
373 | spin_lock_init(&s->s_inode_wblist_lock); | |
7eb5e882 | 374 | |
7eb5e882 AV |
375 | s->s_count = 1; |
376 | atomic_set(&s->s_active, 1); | |
377 | mutex_init(&s->s_vfs_rename_mutex); | |
378 | lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key); | |
bc8230ee | 379 | init_rwsem(&s->s_dquot.dqio_sem); |
7eb5e882 AV |
380 | s->s_maxbytes = MAX_NON_LFS; |
381 | s->s_op = &default_op; | |
382 | s->s_time_gran = 1000000000; | |
188d20bc DD |
383 | s->s_time_min = TIME64_MIN; |
384 | s->s_time_max = TIME64_MAX; | |
7eb5e882 AV |
385 | |
386 | s->s_shrink.seeks = DEFAULT_SEEKS; | |
387 | s->s_shrink.scan_objects = super_cache_scan; | |
388 | s->s_shrink.count_objects = super_cache_count; | |
389 | s->s_shrink.batch = 1024; | |
2acb60a0 | 390 | s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE; |
e33c267a | 391 | if (prealloc_shrinker(&s->s_shrink, "sb-%s", type->name)) |
8e04944f | 392 | goto fail; |
c92e8e10 | 393 | if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink)) |
2b3648a6 | 394 | goto fail; |
c92e8e10 | 395 | if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink)) |
2b3648a6 | 396 | goto fail; |
1da177e4 | 397 | return s; |
5ca302c8 | 398 | |
7eb5e882 | 399 | fail: |
0200894d | 400 | destroy_unused_super(s); |
7eb5e882 | 401 | return NULL; |
1da177e4 LT |
402 | } |
403 | ||
404 | /* Superblock refcounting */ | |
405 | ||
406 | /* | |
35cf7ba0 | 407 | * Drop a superblock's refcount. The caller must hold sb_lock. |
1da177e4 | 408 | */ |
c645b930 | 409 | static void __put_super(struct super_block *s) |
1da177e4 | 410 | { |
c645b930 AV |
411 | if (!--s->s_count) { |
412 | list_del_init(&s->s_list); | |
413 | WARN_ON(s->s_dentry_lru.node); | |
414 | WARN_ON(s->s_inode_lru.node); | |
415 | WARN_ON(!list_empty(&s->s_mounts)); | |
416 | security_sb_free(s); | |
417 | put_user_ns(s->s_user_ns); | |
418 | kfree(s->s_subtype); | |
419 | call_rcu(&s->rcu, destroy_super_rcu); | |
1da177e4 | 420 | } |
1da177e4 LT |
421 | } |
422 | ||
423 | /** | |
424 | * put_super - drop a temporary reference to superblock | |
425 | * @sb: superblock in question | |
426 | * | |
427 | * Drops a temporary reference, frees superblock if there's no | |
428 | * references left. | |
429 | */ | |
60b49885 | 430 | void put_super(struct super_block *sb) |
1da177e4 LT |
431 | { |
432 | spin_lock(&sb_lock); | |
433 | __put_super(sb); | |
434 | spin_unlock(&sb_lock); | |
435 | } | |
436 | ||
437 | ||
438 | /** | |
1712ac8f | 439 | * deactivate_locked_super - drop an active reference to superblock |
1da177e4 LT |
440 | * @s: superblock to deactivate |
441 | * | |
bd7ced98 | 442 | * Drops an active reference to superblock, converting it into a temporary |
1712ac8f | 443 | * one if there is no other active references left. In that case we |
1da177e4 LT |
444 | * tell fs driver to shut it down and drop the temporary reference we |
445 | * had just acquired. | |
1712ac8f AV |
446 | * |
447 | * Caller holds exclusive lock on superblock; that lock is released. | |
1da177e4 | 448 | */ |
1712ac8f | 449 | void deactivate_locked_super(struct super_block *s) |
1da177e4 LT |
450 | { |
451 | struct file_system_type *fs = s->s_type; | |
b20bd1a5 | 452 | if (atomic_dec_and_test(&s->s_active)) { |
b0d40c92 | 453 | unregister_shrinker(&s->s_shrink); |
28f2cd4f | 454 | fs->kill_sb(s); |
f5e1dd34 | 455 | |
c0a5b560 VD |
456 | /* |
457 | * Since list_lru_destroy() may sleep, we cannot call it from | |
458 | * put_super(), where we hold the sb_lock. Therefore we destroy | |
459 | * the lru lists right now. | |
460 | */ | |
461 | list_lru_destroy(&s->s_dentry_lru); | |
462 | list_lru_destroy(&s->s_inode_lru); | |
463 | ||
1da177e4 LT |
464 | put_filesystem(fs); |
465 | put_super(s); | |
1712ac8f | 466 | } else { |
0ed33598 | 467 | super_unlock_excl(s); |
1da177e4 LT |
468 | } |
469 | } | |
470 | ||
1712ac8f | 471 | EXPORT_SYMBOL(deactivate_locked_super); |
1da177e4 | 472 | |
74dbbdd7 | 473 | /** |
1712ac8f | 474 | * deactivate_super - drop an active reference to superblock |
74dbbdd7 AV |
475 | * @s: superblock to deactivate |
476 | * | |
1712ac8f AV |
477 | * Variant of deactivate_locked_super(), except that superblock is *not* |
478 | * locked by caller. If we are going to drop the final active reference, | |
479 | * lock will be acquired prior to that. | |
74dbbdd7 | 480 | */ |
1712ac8f | 481 | void deactivate_super(struct super_block *s) |
74dbbdd7 | 482 | { |
cc23402c | 483 | if (!atomic_add_unless(&s->s_active, -1, 1)) { |
5e874914 | 484 | __super_lock_excl(s); |
1712ac8f | 485 | deactivate_locked_super(s); |
74dbbdd7 AV |
486 | } |
487 | } | |
488 | ||
1712ac8f | 489 | EXPORT_SYMBOL(deactivate_super); |
74dbbdd7 | 490 | |
1da177e4 LT |
491 | /** |
492 | * grab_super - acquire an active reference | |
493 | * @s: reference we are trying to make active | |
494 | * | |
495 | * Tries to acquire an active reference. grab_super() is used when we | |
496 | * had just found a superblock in super_blocks or fs_type->fs_supers | |
497 | * and want to turn it into a full-blown active reference. grab_super() | |
498 | * is called with sb_lock held and drops it. Returns 1 in case of | |
499 | * success, 0 if we had failed (superblock contents was already dead or | |
acfec9a5 AV |
500 | * dying when grab_super() had been called). Note that this is only |
501 | * called for superblocks not in rundown mode (== ones still on ->fs_supers | |
502 | * of their type), so increment of ->s_count is OK here. | |
1da177e4 | 503 | */ |
9c4dbee7 | 504 | static int grab_super(struct super_block *s) __releases(sb_lock) |
1da177e4 | 505 | { |
5e874914 CB |
506 | bool born; |
507 | ||
1da177e4 LT |
508 | s->s_count++; |
509 | spin_unlock(&sb_lock); | |
5e874914 CB |
510 | born = super_lock_excl(s); |
511 | if (born && atomic_inc_not_zero(&s->s_active)) { | |
acfec9a5 AV |
512 | put_super(s); |
513 | return 1; | |
514 | } | |
0ed33598 | 515 | super_unlock_excl(s); |
1da177e4 | 516 | put_super(s); |
1da177e4 LT |
517 | return 0; |
518 | } | |
519 | ||
12ad3ab6 | 520 | /* |
d8ce82ef | 521 | * super_trylock_shared - try to grab ->s_umount shared |
331cbdee | 522 | * @sb: reference we are trying to grab |
12ad3ab6 | 523 | * |
eb6ef3df | 524 | * Try to prevent fs shutdown. This is used in places where we |
12ad3ab6 | 525 | * cannot take an active reference but we need to ensure that the |
eb6ef3df KK |
526 | * filesystem is not shut down while we are working on it. It returns |
527 | * false if we cannot acquire s_umount or if we lose the race and | |
528 | * filesystem already got into shutdown, and returns true with the s_umount | |
529 | * lock held in read mode in case of success. On successful return, | |
530 | * the caller must drop the s_umount lock when done. | |
531 | * | |
532 | * Note that unlike get_super() et.al. this one does *not* bump ->s_count. | |
533 | * The reason why it's safe is that we are OK with doing trylock instead | |
534 | * of down_read(). There's a couple of places that are OK with that, but | |
535 | * it's very much not a general-purpose interface. | |
12ad3ab6 | 536 | */ |
d8ce82ef | 537 | bool super_trylock_shared(struct super_block *sb) |
12ad3ab6 | 538 | { |
12ad3ab6 | 539 | if (down_read_trylock(&sb->s_umount)) { |
5e874914 CB |
540 | if (!(sb->s_flags & SB_DYING) && sb->s_root && |
541 | (sb->s_flags & SB_BORN)) | |
12ad3ab6 | 542 | return true; |
0ed33598 | 543 | super_unlock_shared(sb); |
12ad3ab6 DC |
544 | } |
545 | ||
12ad3ab6 DC |
546 | return false; |
547 | } | |
548 | ||
04b94071 DL |
549 | /** |
550 | * retire_super - prevents superblock from being reused | |
551 | * @sb: superblock to retire | |
552 | * | |
553 | * The function marks superblock to be ignored in superblock test, which | |
554 | * prevents it from being reused for any new mounts. If the superblock has | |
555 | * a private bdi, it also unregisters it, but doesn't reduce the refcount | |
556 | * of the superblock to prevent potential races. The refcount is reduced | |
557 | * by generic_shutdown_super(). The function can not be called | |
558 | * concurrently with generic_shutdown_super(). It is safe to call the | |
559 | * function multiple times, subsequent calls have no effect. | |
560 | * | |
561 | * The marker will affect the re-use only for block-device-based | |
562 | * superblocks. Other superblocks will still get marked if this function | |
563 | * is used, but that will not affect their reusability. | |
564 | */ | |
565 | void retire_super(struct super_block *sb) | |
566 | { | |
567 | WARN_ON(!sb->s_bdev); | |
5e874914 | 568 | __super_lock_excl(sb); |
04b94071 DL |
569 | if (sb->s_iflags & SB_I_PERSB_BDI) { |
570 | bdi_unregister(sb->s_bdi); | |
571 | sb->s_iflags &= ~SB_I_PERSB_BDI; | |
572 | } | |
573 | sb->s_iflags |= SB_I_RETIRED; | |
0ed33598 | 574 | super_unlock_excl(sb); |
04b94071 DL |
575 | } |
576 | EXPORT_SYMBOL(retire_super); | |
577 | ||
1da177e4 LT |
578 | /** |
579 | * generic_shutdown_super - common helper for ->kill_sb() | |
580 | * @sb: superblock to kill | |
581 | * | |
582 | * generic_shutdown_super() does all fs-independent work on superblock | |
583 | * shutdown. Typical ->kill_sb() should pick all fs-specific objects | |
584 | * that need destruction out of superblock, call generic_shutdown_super() | |
585 | * and release aforementioned objects. Note: dentries and inodes _are_ | |
586 | * taken care of and do not need specific handling. | |
c636ebdb DH |
587 | * |
588 | * Upon calling this function, the filesystem may no longer alter or | |
589 | * rearrange the set of dentries belonging to this super_block, nor may it | |
590 | * change the attachments of dentries to inodes. | |
1da177e4 LT |
591 | */ |
592 | void generic_shutdown_super(struct super_block *sb) | |
593 | { | |
ee9b6d61 | 594 | const struct super_operations *sop = sb->s_op; |
1da177e4 | 595 | |
c636ebdb DH |
596 | if (sb->s_root) { |
597 | shrink_dcache_for_umount(sb); | |
60b0680f | 598 | sync_filesystem(sb); |
e462ec50 | 599 | sb->s_flags &= ~SB_ACTIVE; |
efaee192 | 600 | |
a1a0e23e | 601 | cgroup_writeback_umount(); |
63997e98 | 602 | |
ccb820dc | 603 | /* Evict all inodes with zero refcount. */ |
63997e98 | 604 | evict_inodes(sb); |
ccb820dc EB |
605 | |
606 | /* | |
607 | * Clean up and evict any inodes that still have references due | |
608 | * to fsnotify or the security policy. | |
609 | */ | |
1edc8eb2 | 610 | fsnotify_sb_delete(sb); |
83e804f0 | 611 | security_sb_delete(sb); |
1da177e4 | 612 | |
ccb820dc EB |
613 | /* |
614 | * Now that all potentially-encrypted inodes have been evicted, | |
615 | * the fscrypt keyring can be destroyed. | |
616 | */ | |
617 | fscrypt_destroy_keyring(sb); | |
618 | ||
7b7a8665 CH |
619 | if (sb->s_dio_done_wq) { |
620 | destroy_workqueue(sb->s_dio_done_wq); | |
621 | sb->s_dio_done_wq = NULL; | |
622 | } | |
623 | ||
1da177e4 LT |
624 | if (sop->put_super) |
625 | sop->put_super(sb); | |
626 | ||
47d58691 JH |
627 | if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes), |
628 | "VFS: Busy inodes after unmount of %s (%s)", | |
629 | sb->s_id, sb->s_type->name)) { | |
630 | /* | |
631 | * Adding a proper bailout path here would be hard, but | |
632 | * we can at least make it more likely that a later | |
633 | * iput_final() or such crashes cleanly. | |
634 | */ | |
635 | struct inode *inode; | |
636 | ||
637 | spin_lock(&sb->s_inode_list_lock); | |
638 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { | |
639 | inode->i_op = VFS_PTR_POISON; | |
640 | inode->i_sb = VFS_PTR_POISON; | |
641 | inode->i_mapping = VFS_PTR_POISON; | |
642 | } | |
643 | spin_unlock(&sb->s_inode_list_lock); | |
1da177e4 | 644 | } |
1da177e4 LT |
645 | } |
646 | spin_lock(&sb_lock); | |
647 | /* should be initialized for __put_super_and_need_restart() */ | |
a5166169 | 648 | hlist_del_init(&sb->s_instances); |
1da177e4 | 649 | spin_unlock(&sb_lock); |
5e874914 CB |
650 | /* |
651 | * Broadcast to everyone that grabbed a temporary reference to this | |
652 | * superblock before we removed it from @fs_supers that the superblock | |
653 | * is dying. Every walker of @fs_supers outside of sget{_fc}() will now | |
654 | * discard this superblock and treat it as dead. | |
655 | */ | |
656 | super_wake(sb, SB_DYING); | |
0ed33598 | 657 | super_unlock_excl(sb); |
c1844d53 | 658 | if (sb->s_bdi != &noop_backing_dev_info) { |
0b3ea092 CH |
659 | if (sb->s_iflags & SB_I_PERSB_BDI) |
660 | bdi_unregister(sb->s_bdi); | |
fca39346 JK |
661 | bdi_put(sb->s_bdi); |
662 | sb->s_bdi = &noop_backing_dev_info; | |
fca39346 | 663 | } |
1da177e4 LT |
664 | } |
665 | ||
666 | EXPORT_SYMBOL(generic_shutdown_super); | |
667 | ||
20284ab7 | 668 | bool mount_capable(struct fs_context *fc) |
0ce0cf12 | 669 | { |
20284ab7 | 670 | if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT)) |
0ce0cf12 AV |
671 | return capable(CAP_SYS_ADMIN); |
672 | else | |
c2c44ec2 | 673 | return ns_capable(fc->user_ns, CAP_SYS_ADMIN); |
0ce0cf12 AV |
674 | } |
675 | ||
cb50b348 AV |
676 | /** |
677 | * sget_fc - Find or create a superblock | |
678 | * @fc: Filesystem context. | |
679 | * @test: Comparison callback | |
680 | * @set: Setup callback | |
681 | * | |
682 | * Find or create a superblock using the parameters stored in the filesystem | |
683 | * context and the two callback functions. | |
684 | * | |
685 | * If an extant superblock is matched, then that will be returned with an | |
686 | * elevated reference count that the caller must transfer or discard. | |
687 | * | |
688 | * If no match is made, a new superblock will be allocated and basic | |
689 | * initialisation will be performed (s_type, s_fs_info and s_id will be set and | |
690 | * the set() callback will be invoked), the superblock will be published and it | |
691 | * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE | |
692 | * as yet unset. | |
693 | */ | |
694 | struct super_block *sget_fc(struct fs_context *fc, | |
695 | int (*test)(struct super_block *, struct fs_context *), | |
696 | int (*set)(struct super_block *, struct fs_context *)) | |
697 | { | |
698 | struct super_block *s = NULL; | |
699 | struct super_block *old; | |
700 | struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns; | |
701 | int err; | |
702 | ||
cb50b348 AV |
703 | retry: |
704 | spin_lock(&sb_lock); | |
705 | if (test) { | |
706 | hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) { | |
707 | if (test(old, fc)) | |
708 | goto share_extant_sb; | |
709 | } | |
710 | } | |
711 | if (!s) { | |
712 | spin_unlock(&sb_lock); | |
713 | s = alloc_super(fc->fs_type, fc->sb_flags, user_ns); | |
714 | if (!s) | |
715 | return ERR_PTR(-ENOMEM); | |
716 | goto retry; | |
717 | } | |
718 | ||
719 | s->s_fs_info = fc->s_fs_info; | |
720 | err = set(s, fc); | |
721 | if (err) { | |
722 | s->s_fs_info = NULL; | |
723 | spin_unlock(&sb_lock); | |
724 | destroy_unused_super(s); | |
725 | return ERR_PTR(err); | |
726 | } | |
727 | fc->s_fs_info = NULL; | |
728 | s->s_type = fc->fs_type; | |
c80fa7c8 | 729 | s->s_iflags |= fc->s_iflags; |
c642256b | 730 | strscpy(s->s_id, s->s_type->name, sizeof(s->s_id)); |
5e874914 CB |
731 | /* |
732 | * Make the superblock visible on @super_blocks and @fs_supers. | |
733 | * It's in a nascent state and users should wait on SB_BORN or | |
734 | * SB_DYING to be set. | |
735 | */ | |
cb50b348 AV |
736 | list_add_tail(&s->s_list, &super_blocks); |
737 | hlist_add_head(&s->s_instances, &s->s_type->fs_supers); | |
738 | spin_unlock(&sb_lock); | |
739 | get_filesystem(s->s_type); | |
740 | register_shrinker_prepared(&s->s_shrink); | |
741 | return s; | |
742 | ||
743 | share_extant_sb: | |
744 | if (user_ns != old->s_user_ns) { | |
745 | spin_unlock(&sb_lock); | |
746 | destroy_unused_super(s); | |
747 | return ERR_PTR(-EBUSY); | |
748 | } | |
749 | if (!grab_super(old)) | |
750 | goto retry; | |
751 | destroy_unused_super(s); | |
752 | return old; | |
753 | } | |
754 | EXPORT_SYMBOL(sget_fc); | |
755 | ||
1da177e4 | 756 | /** |
023d066a DH |
757 | * sget - find or create a superblock |
758 | * @type: filesystem type superblock should belong to | |
759 | * @test: comparison callback | |
760 | * @set: setup callback | |
761 | * @flags: mount flags | |
762 | * @data: argument to each of them | |
1da177e4 | 763 | */ |
023d066a | 764 | struct super_block *sget(struct file_system_type *type, |
1da177e4 LT |
765 | int (*test)(struct super_block *,void *), |
766 | int (*set)(struct super_block *,void *), | |
023d066a | 767 | int flags, |
1da177e4 LT |
768 | void *data) |
769 | { | |
023d066a | 770 | struct user_namespace *user_ns = current_user_ns(); |
1da177e4 | 771 | struct super_block *s = NULL; |
d4730127 | 772 | struct super_block *old; |
1da177e4 LT |
773 | int err; |
774 | ||
023d066a DH |
775 | /* We don't yet pass the user namespace of the parent |
776 | * mount through to here so always use &init_user_ns | |
777 | * until that changes. | |
778 | */ | |
779 | if (flags & SB_SUBMOUNT) | |
780 | user_ns = &init_user_ns; | |
781 | ||
1da177e4 LT |
782 | retry: |
783 | spin_lock(&sb_lock); | |
d4730127 | 784 | if (test) { |
b67bfe0d | 785 | hlist_for_each_entry(old, &type->fs_supers, s_instances) { |
d4730127 MK |
786 | if (!test(old, data)) |
787 | continue; | |
6e4eab57 EB |
788 | if (user_ns != old->s_user_ns) { |
789 | spin_unlock(&sb_lock); | |
0200894d | 790 | destroy_unused_super(s); |
6e4eab57 EB |
791 | return ERR_PTR(-EBUSY); |
792 | } | |
d4730127 MK |
793 | if (!grab_super(old)) |
794 | goto retry; | |
0200894d | 795 | destroy_unused_super(s); |
d4730127 MK |
796 | return old; |
797 | } | |
1da177e4 LT |
798 | } |
799 | if (!s) { | |
800 | spin_unlock(&sb_lock); | |
e462ec50 | 801 | s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns); |
1da177e4 LT |
802 | if (!s) |
803 | return ERR_PTR(-ENOMEM); | |
804 | goto retry; | |
805 | } | |
dd111b31 | 806 | |
1da177e4 LT |
807 | err = set(s, data); |
808 | if (err) { | |
809 | spin_unlock(&sb_lock); | |
0200894d | 810 | destroy_unused_super(s); |
1da177e4 LT |
811 | return ERR_PTR(err); |
812 | } | |
813 | s->s_type = type; | |
c642256b | 814 | strscpy(s->s_id, type->name, sizeof(s->s_id)); |
1da177e4 | 815 | list_add_tail(&s->s_list, &super_blocks); |
a5166169 | 816 | hlist_add_head(&s->s_instances, &type->fs_supers); |
1da177e4 LT |
817 | spin_unlock(&sb_lock); |
818 | get_filesystem(type); | |
8e04944f | 819 | register_shrinker_prepared(&s->s_shrink); |
1da177e4 LT |
820 | return s; |
821 | } | |
1da177e4 LT |
822 | EXPORT_SYMBOL(sget); |
823 | ||
824 | void drop_super(struct super_block *sb) | |
825 | { | |
0ed33598 | 826 | super_unlock_shared(sb); |
1da177e4 LT |
827 | put_super(sb); |
828 | } | |
829 | ||
830 | EXPORT_SYMBOL(drop_super); | |
831 | ||
ba6379f7 JK |
832 | void drop_super_exclusive(struct super_block *sb) |
833 | { | |
0ed33598 | 834 | super_unlock_excl(sb); |
ba6379f7 JK |
835 | put_super(sb); |
836 | } | |
837 | EXPORT_SYMBOL(drop_super_exclusive); | |
838 | ||
fa7c1d50 MG |
839 | static void __iterate_supers(void (*f)(struct super_block *)) |
840 | { | |
841 | struct super_block *sb, *p = NULL; | |
842 | ||
843 | spin_lock(&sb_lock); | |
844 | list_for_each_entry(sb, &super_blocks, s_list) { | |
5e874914 CB |
845 | /* Pairs with memory marrier in super_wake(). */ |
846 | if (smp_load_acquire(&sb->s_flags) & SB_DYING) | |
fa7c1d50 MG |
847 | continue; |
848 | sb->s_count++; | |
849 | spin_unlock(&sb_lock); | |
850 | ||
851 | f(sb); | |
852 | ||
853 | spin_lock(&sb_lock); | |
854 | if (p) | |
855 | __put_super(p); | |
856 | p = sb; | |
857 | } | |
858 | if (p) | |
859 | __put_super(p); | |
860 | spin_unlock(&sb_lock); | |
861 | } | |
01a05b33 AV |
862 | /** |
863 | * iterate_supers - call function for all active superblocks | |
864 | * @f: function to call | |
865 | * @arg: argument to pass to it | |
866 | * | |
867 | * Scans the superblock list and calls given function, passing it | |
868 | * locked superblock and given argument. | |
869 | */ | |
870 | void iterate_supers(void (*f)(struct super_block *, void *), void *arg) | |
871 | { | |
dca33252 | 872 | struct super_block *sb, *p = NULL; |
01a05b33 AV |
873 | |
874 | spin_lock(&sb_lock); | |
dca33252 | 875 | list_for_each_entry(sb, &super_blocks, s_list) { |
5e874914 CB |
876 | bool born; |
877 | ||
01a05b33 AV |
878 | sb->s_count++; |
879 | spin_unlock(&sb_lock); | |
880 | ||
5e874914 CB |
881 | born = super_lock_shared(sb); |
882 | if (born && sb->s_root) | |
01a05b33 | 883 | f(sb, arg); |
0ed33598 | 884 | super_unlock_shared(sb); |
01a05b33 AV |
885 | |
886 | spin_lock(&sb_lock); | |
dca33252 AV |
887 | if (p) |
888 | __put_super(p); | |
889 | p = sb; | |
01a05b33 | 890 | } |
dca33252 AV |
891 | if (p) |
892 | __put_super(p); | |
01a05b33 AV |
893 | spin_unlock(&sb_lock); |
894 | } | |
895 | ||
43e15cdb AV |
896 | /** |
897 | * iterate_supers_type - call function for superblocks of given type | |
898 | * @type: fs type | |
899 | * @f: function to call | |
900 | * @arg: argument to pass to it | |
901 | * | |
902 | * Scans the superblock list and calls given function, passing it | |
903 | * locked superblock and given argument. | |
904 | */ | |
905 | void iterate_supers_type(struct file_system_type *type, | |
906 | void (*f)(struct super_block *, void *), void *arg) | |
907 | { | |
908 | struct super_block *sb, *p = NULL; | |
909 | ||
910 | spin_lock(&sb_lock); | |
b67bfe0d | 911 | hlist_for_each_entry(sb, &type->fs_supers, s_instances) { |
5e874914 CB |
912 | bool born; |
913 | ||
43e15cdb AV |
914 | sb->s_count++; |
915 | spin_unlock(&sb_lock); | |
916 | ||
5e874914 CB |
917 | born = super_lock_shared(sb); |
918 | if (born && sb->s_root) | |
43e15cdb | 919 | f(sb, arg); |
0ed33598 | 920 | super_unlock_shared(sb); |
43e15cdb AV |
921 | |
922 | spin_lock(&sb_lock); | |
923 | if (p) | |
924 | __put_super(p); | |
925 | p = sb; | |
926 | } | |
927 | if (p) | |
928 | __put_super(p); | |
929 | spin_unlock(&sb_lock); | |
930 | } | |
931 | ||
932 | EXPORT_SYMBOL(iterate_supers_type); | |
933 | ||
4504230a CH |
934 | /** |
935 | * get_active_super - get an active reference to the superblock of a device | |
936 | * @bdev: device to get the superblock for | |
937 | * | |
938 | * Scans the superblock list and finds the superblock of the file system | |
939 | * mounted on the device given. Returns the superblock with an active | |
d3f21473 | 940 | * reference or %NULL if none was found. |
4504230a CH |
941 | */ |
942 | struct super_block *get_active_super(struct block_device *bdev) | |
943 | { | |
944 | struct super_block *sb; | |
945 | ||
946 | if (!bdev) | |
947 | return NULL; | |
948 | ||
949 | spin_lock(&sb_lock); | |
950 | list_for_each_entry(sb, &super_blocks, s_list) { | |
1494583d | 951 | if (sb->s_bdev == bdev) { |
acfec9a5 | 952 | if (!grab_super(sb)) |
5e874914 | 953 | return NULL; |
0ed33598 | 954 | super_unlock_excl(sb); |
acfec9a5 | 955 | return sb; |
1494583d | 956 | } |
4504230a CH |
957 | } |
958 | spin_unlock(&sb_lock); | |
959 | return NULL; | |
960 | } | |
dd111b31 | 961 | |
4e7b5671 | 962 | struct super_block *user_get_super(dev_t dev, bool excl) |
1da177e4 | 963 | { |
618f0636 | 964 | struct super_block *sb; |
1da177e4 | 965 | |
1da177e4 | 966 | spin_lock(&sb_lock); |
618f0636 KK |
967 | list_for_each_entry(sb, &super_blocks, s_list) { |
968 | if (sb->s_dev == dev) { | |
5e874914 CB |
969 | bool born; |
970 | ||
618f0636 | 971 | sb->s_count++; |
1da177e4 | 972 | spin_unlock(&sb_lock); |
df40c01a | 973 | /* still alive? */ |
5e874914 CB |
974 | born = super_lock(sb, excl); |
975 | if (born && sb->s_root) | |
618f0636 | 976 | return sb; |
0ed33598 | 977 | super_unlock(sb, excl); |
df40c01a | 978 | /* nope, got unmounted */ |
618f0636 | 979 | spin_lock(&sb_lock); |
df40c01a | 980 | __put_super(sb); |
5e874914 | 981 | break; |
1da177e4 LT |
982 | } |
983 | } | |
984 | spin_unlock(&sb_lock); | |
985 | return NULL; | |
986 | } | |
987 | ||
1da177e4 | 988 | /** |
8d0347f6 DH |
989 | * reconfigure_super - asks filesystem to change superblock parameters |
990 | * @fc: The superblock and configuration | |
1da177e4 | 991 | * |
8d0347f6 | 992 | * Alters the configuration parameters of a live superblock. |
1da177e4 | 993 | */ |
8d0347f6 | 994 | int reconfigure_super(struct fs_context *fc) |
1da177e4 | 995 | { |
8d0347f6 | 996 | struct super_block *sb = fc->root->d_sb; |
1da177e4 | 997 | int retval; |
8d0347f6 | 998 | bool remount_ro = false; |
c541dce8 | 999 | bool remount_rw = false; |
8d0347f6 | 1000 | bool force = fc->sb_flags & SB_FORCE; |
4504230a | 1001 | |
8d0347f6 DH |
1002 | if (fc->sb_flags_mask & ~MS_RMT_MASK) |
1003 | return -EINVAL; | |
5accdf82 | 1004 | if (sb->s_writers.frozen != SB_UNFROZEN) |
4504230a CH |
1005 | return -EBUSY; |
1006 | ||
8d0347f6 DH |
1007 | retval = security_sb_remount(sb, fc->security); |
1008 | if (retval) | |
1009 | return retval; | |
1010 | ||
1011 | if (fc->sb_flags_mask & SB_RDONLY) { | |
9361401e | 1012 | #ifdef CONFIG_BLOCK |
6f0d9689 CH |
1013 | if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev && |
1014 | bdev_read_only(sb->s_bdev)) | |
8d0347f6 | 1015 | return -EACCES; |
9361401e | 1016 | #endif |
c541dce8 | 1017 | remount_rw = !(fc->sb_flags & SB_RDONLY) && sb_rdonly(sb); |
8d0347f6 DH |
1018 | remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb); |
1019 | } | |
d208bbdd | 1020 | |
0aec09d0 | 1021 | if (remount_ro) { |
fdab684d | 1022 | if (!hlist_empty(&sb->s_pins)) { |
0ed33598 | 1023 | super_unlock_excl(sb); |
fdab684d | 1024 | group_pin_kill(&sb->s_pins); |
5e874914 | 1025 | __super_lock_excl(sb); |
0aec09d0 AV |
1026 | if (!sb->s_root) |
1027 | return 0; | |
1028 | if (sb->s_writers.frozen != SB_UNFROZEN) | |
1029 | return -EBUSY; | |
8d0347f6 | 1030 | remount_ro = !sb_rdonly(sb); |
0aec09d0 AV |
1031 | } |
1032 | } | |
1033 | shrink_dcache_sb(sb); | |
1034 | ||
8d0347f6 DH |
1035 | /* If we are reconfiguring to RDONLY and current sb is read/write, |
1036 | * make sure there are no files open for writing. | |
1037 | */ | |
d208bbdd | 1038 | if (remount_ro) { |
4ed5e82f | 1039 | if (force) { |
d7439fb1 | 1040 | sb_start_ro_state_change(sb); |
4ed5e82f MS |
1041 | } else { |
1042 | retval = sb_prepare_remount_readonly(sb); | |
1043 | if (retval) | |
1044 | return retval; | |
4ed5e82f | 1045 | } |
c541dce8 JK |
1046 | } else if (remount_rw) { |
1047 | /* | |
d7439fb1 JK |
1048 | * Protect filesystem's reconfigure code from writes from |
1049 | * userspace until reconfigure finishes. | |
c541dce8 | 1050 | */ |
d7439fb1 | 1051 | sb_start_ro_state_change(sb); |
1da177e4 LT |
1052 | } |
1053 | ||
f3a09c92 AV |
1054 | if (fc->ops->reconfigure) { |
1055 | retval = fc->ops->reconfigure(fc); | |
1056 | if (retval) { | |
1057 | if (!force) | |
1058 | goto cancel_readonly; | |
1059 | /* If forced remount, go ahead despite any errors */ | |
1060 | WARN(1, "forced remount of a %s fs returned %i\n", | |
1061 | sb->s_type->name, retval); | |
1062 | } | |
1da177e4 | 1063 | } |
8d0347f6 DH |
1064 | |
1065 | WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) | | |
1066 | (fc->sb_flags & fc->sb_flags_mask))); | |
d7439fb1 | 1067 | sb_end_ro_state_change(sb); |
c79d967d | 1068 | |
d208bbdd NP |
1069 | /* |
1070 | * Some filesystems modify their metadata via some other path than the | |
1071 | * bdev buffer cache (eg. use a private mapping, or directories in | |
1072 | * pagecache, etc). Also file data modifications go via their own | |
1073 | * mappings. So If we try to mount readonly then copy the filesystem | |
1074 | * from bdev, we could get stale data, so invalidate it to give a best | |
1075 | * effort at coherency. | |
1076 | */ | |
1077 | if (remount_ro && sb->s_bdev) | |
1078 | invalidate_bdev(sb->s_bdev); | |
1da177e4 | 1079 | return 0; |
4ed5e82f MS |
1080 | |
1081 | cancel_readonly: | |
d7439fb1 | 1082 | sb_end_ro_state_change(sb); |
4ed5e82f | 1083 | return retval; |
1da177e4 LT |
1084 | } |
1085 | ||
fa7c1d50 | 1086 | static void do_emergency_remount_callback(struct super_block *sb) |
1da177e4 | 1087 | { |
5e874914 CB |
1088 | bool born = super_lock_excl(sb); |
1089 | ||
1090 | if (born && sb->s_root && sb->s_bdev && !sb_rdonly(sb)) { | |
8d0347f6 DH |
1091 | struct fs_context *fc; |
1092 | ||
1093 | fc = fs_context_for_reconfigure(sb->s_root, | |
1094 | SB_RDONLY | SB_FORCE, SB_RDONLY); | |
1095 | if (!IS_ERR(fc)) { | |
1096 | if (parse_monolithic_mount_data(fc, NULL) == 0) | |
1097 | (void)reconfigure_super(fc); | |
1098 | put_fs_context(fc); | |
1099 | } | |
1da177e4 | 1100 | } |
0ed33598 | 1101 | super_unlock_excl(sb); |
fa7c1d50 MG |
1102 | } |
1103 | ||
1104 | static void do_emergency_remount(struct work_struct *work) | |
1105 | { | |
1106 | __iterate_supers(do_emergency_remount_callback); | |
a2a9537a | 1107 | kfree(work); |
1da177e4 LT |
1108 | printk("Emergency Remount complete\n"); |
1109 | } | |
1110 | ||
1111 | void emergency_remount(void) | |
1112 | { | |
a2a9537a JA |
1113 | struct work_struct *work; |
1114 | ||
1115 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
1116 | if (work) { | |
1117 | INIT_WORK(work, do_emergency_remount); | |
1118 | schedule_work(work); | |
1119 | } | |
1da177e4 LT |
1120 | } |
1121 | ||
08fdc8a0 MG |
1122 | static void do_thaw_all_callback(struct super_block *sb) |
1123 | { | |
5e874914 CB |
1124 | bool born = super_lock_excl(sb); |
1125 | ||
1126 | if (born && sb->s_root) { | |
08fdc8a0 MG |
1127 | emergency_thaw_bdev(sb); |
1128 | thaw_super_locked(sb); | |
1129 | } else { | |
0ed33598 | 1130 | super_unlock_excl(sb); |
08fdc8a0 MG |
1131 | } |
1132 | } | |
1133 | ||
1134 | static void do_thaw_all(struct work_struct *work) | |
1135 | { | |
1136 | __iterate_supers(do_thaw_all_callback); | |
1137 | kfree(work); | |
1138 | printk(KERN_WARNING "Emergency Thaw complete\n"); | |
1139 | } | |
1140 | ||
1141 | /** | |
1142 | * emergency_thaw_all -- forcibly thaw every frozen filesystem | |
1143 | * | |
1144 | * Used for emergency unfreeze of all filesystems via SysRq | |
1145 | */ | |
1146 | void emergency_thaw_all(void) | |
1147 | { | |
1148 | struct work_struct *work; | |
1149 | ||
1150 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
1151 | if (work) { | |
1152 | INIT_WORK(work, do_thaw_all); | |
1153 | schedule_work(work); | |
1154 | } | |
1155 | } | |
1156 | ||
ad76cbc6 | 1157 | static DEFINE_IDA(unnamed_dev_ida); |
1da177e4 | 1158 | |
5a66847e MW |
1159 | /** |
1160 | * get_anon_bdev - Allocate a block device for filesystems which don't have one. | |
1161 | * @p: Pointer to a dev_t. | |
1162 | * | |
1163 | * Filesystems which don't use real block devices can call this function | |
1164 | * to allocate a virtual block device. | |
1165 | * | |
1166 | * Context: Any context. Frequently called while holding sb_lock. | |
1167 | * Return: 0 on success, -EMFILE if there are no anonymous bdevs left | |
1168 | * or -ENOMEM if memory allocation failed. | |
1169 | */ | |
0ee5dc67 | 1170 | int get_anon_bdev(dev_t *p) |
1da177e4 LT |
1171 | { |
1172 | int dev; | |
5a66847e MW |
1173 | |
1174 | /* | |
1175 | * Many userspace utilities consider an FSID of 0 invalid. | |
1176 | * Always return at least 1 from get_anon_bdev. | |
1177 | */ | |
1178 | dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1, | |
1179 | GFP_ATOMIC); | |
1180 | if (dev == -ENOSPC) | |
1181 | dev = -EMFILE; | |
1182 | if (dev < 0) | |
1183 | return dev; | |
1184 | ||
1185 | *p = MKDEV(0, dev); | |
1da177e4 LT |
1186 | return 0; |
1187 | } | |
0ee5dc67 | 1188 | EXPORT_SYMBOL(get_anon_bdev); |
1da177e4 | 1189 | |
0ee5dc67 | 1190 | void free_anon_bdev(dev_t dev) |
1da177e4 | 1191 | { |
5a66847e | 1192 | ida_free(&unnamed_dev_ida, MINOR(dev)); |
1da177e4 | 1193 | } |
0ee5dc67 AV |
1194 | EXPORT_SYMBOL(free_anon_bdev); |
1195 | ||
1196 | int set_anon_super(struct super_block *s, void *data) | |
1197 | { | |
df0ce26c | 1198 | return get_anon_bdev(&s->s_dev); |
0ee5dc67 | 1199 | } |
0ee5dc67 AV |
1200 | EXPORT_SYMBOL(set_anon_super); |
1201 | ||
1202 | void kill_anon_super(struct super_block *sb) | |
1203 | { | |
1204 | dev_t dev = sb->s_dev; | |
1205 | generic_shutdown_super(sb); | |
1206 | free_anon_bdev(dev); | |
1207 | } | |
1da177e4 LT |
1208 | EXPORT_SYMBOL(kill_anon_super); |
1209 | ||
1da177e4 LT |
1210 | void kill_litter_super(struct super_block *sb) |
1211 | { | |
1212 | if (sb->s_root) | |
1213 | d_genocide(sb->s_root); | |
1214 | kill_anon_super(sb); | |
1215 | } | |
1da177e4 LT |
1216 | EXPORT_SYMBOL(kill_litter_super); |
1217 | ||
cb50b348 AV |
1218 | int set_anon_super_fc(struct super_block *sb, struct fs_context *fc) |
1219 | { | |
1220 | return set_anon_super(sb, NULL); | |
1221 | } | |
1222 | EXPORT_SYMBOL(set_anon_super_fc); | |
1223 | ||
1224 | static int test_keyed_super(struct super_block *sb, struct fs_context *fc) | |
1225 | { | |
1226 | return sb->s_fs_info == fc->s_fs_info; | |
1227 | } | |
1228 | ||
1229 | static int test_single_super(struct super_block *s, struct fs_context *fc) | |
1230 | { | |
1231 | return 1; | |
1232 | } | |
1233 | ||
cda2ed05 CH |
1234 | static int vfs_get_super(struct fs_context *fc, bool reconf, |
1235 | int (*test)(struct super_block *, struct fs_context *), | |
1236 | int (*fill_super)(struct super_block *sb, | |
1237 | struct fs_context *fc)) | |
cb50b348 | 1238 | { |
cb50b348 | 1239 | struct super_block *sb; |
43ce4c1f | 1240 | int err; |
cb50b348 | 1241 | |
cb50b348 AV |
1242 | sb = sget_fc(fc, test, set_anon_super_fc); |
1243 | if (IS_ERR(sb)) | |
1244 | return PTR_ERR(sb); | |
1245 | ||
1246 | if (!sb->s_root) { | |
43ce4c1f DH |
1247 | err = fill_super(sb, fc); |
1248 | if (err) | |
1249 | goto error; | |
cb50b348 AV |
1250 | |
1251 | sb->s_flags |= SB_ACTIVE; | |
43ce4c1f DH |
1252 | fc->root = dget(sb->s_root); |
1253 | } else { | |
1254 | fc->root = dget(sb->s_root); | |
cda2ed05 | 1255 | if (reconf) { |
43ce4c1f DH |
1256 | err = reconfigure_super(fc); |
1257 | if (err < 0) { | |
1258 | dput(fc->root); | |
1259 | fc->root = NULL; | |
1260 | goto error; | |
1261 | } | |
1262 | } | |
cb50b348 AV |
1263 | } |
1264 | ||
cb50b348 | 1265 | return 0; |
43ce4c1f DH |
1266 | |
1267 | error: | |
1268 | deactivate_locked_super(sb); | |
1269 | return err; | |
cb50b348 | 1270 | } |
cb50b348 | 1271 | |
2ac295d4 AV |
1272 | int get_tree_nodev(struct fs_context *fc, |
1273 | int (*fill_super)(struct super_block *sb, | |
1274 | struct fs_context *fc)) | |
1275 | { | |
cda2ed05 | 1276 | return vfs_get_super(fc, false, NULL, fill_super); |
2ac295d4 AV |
1277 | } |
1278 | EXPORT_SYMBOL(get_tree_nodev); | |
1279 | ||
c23a0bba AV |
1280 | int get_tree_single(struct fs_context *fc, |
1281 | int (*fill_super)(struct super_block *sb, | |
1282 | struct fs_context *fc)) | |
1283 | { | |
cda2ed05 | 1284 | return vfs_get_super(fc, false, test_single_super, fill_super); |
c23a0bba AV |
1285 | } |
1286 | EXPORT_SYMBOL(get_tree_single); | |
1287 | ||
43ce4c1f DH |
1288 | int get_tree_single_reconf(struct fs_context *fc, |
1289 | int (*fill_super)(struct super_block *sb, | |
1290 | struct fs_context *fc)) | |
1291 | { | |
cda2ed05 | 1292 | return vfs_get_super(fc, true, test_single_super, fill_super); |
43ce4c1f DH |
1293 | } |
1294 | EXPORT_SYMBOL(get_tree_single_reconf); | |
1295 | ||
533770cc AV |
1296 | int get_tree_keyed(struct fs_context *fc, |
1297 | int (*fill_super)(struct super_block *sb, | |
1298 | struct fs_context *fc), | |
1299 | void *key) | |
1300 | { | |
1301 | fc->s_fs_info = key; | |
cda2ed05 | 1302 | return vfs_get_super(fc, false, test_keyed_super, fill_super); |
533770cc AV |
1303 | } |
1304 | EXPORT_SYMBOL(get_tree_keyed); | |
1305 | ||
9361401e | 1306 | #ifdef CONFIG_BLOCK |
9c09a7cf CH |
1307 | /* |
1308 | * Lock a super block that the callers holds a reference to. | |
1309 | * | |
1310 | * The caller needs to ensure that the super_block isn't being freed while | |
1311 | * calling this function, e.g. by holding a lock over the call to this function | |
1312 | * and the place that clears the pointer to the superblock used by this function | |
1313 | * before freeing the superblock. | |
1314 | */ | |
d8ce82ef | 1315 | static bool super_lock_shared_active(struct super_block *sb) |
9c09a7cf | 1316 | { |
5e874914 CB |
1317 | bool born = super_lock_shared(sb); |
1318 | ||
1319 | if (!born || !sb->s_root || !(sb->s_flags & SB_ACTIVE)) { | |
0ed33598 | 1320 | super_unlock_shared(sb); |
9c09a7cf CH |
1321 | return false; |
1322 | } | |
1323 | return true; | |
1324 | } | |
1325 | ||
d8530de5 | 1326 | static void fs_bdev_mark_dead(struct block_device *bdev, bool surprise) |
87efb390 | 1327 | { |
9c09a7cf | 1328 | struct super_block *sb = bdev->bd_holder; |
87efb390 | 1329 | |
9c09a7cf CH |
1330 | /* bd_holder_lock ensures that the sb isn't freed */ |
1331 | lockdep_assert_held(&bdev->bd_holder_lock); | |
1332 | ||
d8ce82ef | 1333 | if (!super_lock_shared_active(sb)) |
87efb390 CH |
1334 | return; |
1335 | ||
d8530de5 CH |
1336 | if (!surprise) |
1337 | sync_filesystem(sb); | |
1338 | shrink_dcache_sb(sb); | |
e127b9bc | 1339 | invalidate_inodes(sb); |
87efb390 CH |
1340 | if (sb->s_op->shutdown) |
1341 | sb->s_op->shutdown(sb); | |
9c09a7cf | 1342 | |
0ed33598 | 1343 | super_unlock_shared(sb); |
87efb390 CH |
1344 | } |
1345 | ||
2142b88c CH |
1346 | static void fs_bdev_sync(struct block_device *bdev) |
1347 | { | |
1348 | struct super_block *sb = bdev->bd_holder; | |
1349 | ||
1350 | lockdep_assert_held(&bdev->bd_holder_lock); | |
1351 | ||
d8ce82ef | 1352 | if (!super_lock_shared_active(sb)) |
2142b88c CH |
1353 | return; |
1354 | sync_filesystem(sb); | |
0ed33598 | 1355 | super_unlock_shared(sb); |
2142b88c CH |
1356 | } |
1357 | ||
7ecd0b6f | 1358 | const struct blk_holder_ops fs_holder_ops = { |
d8530de5 | 1359 | .mark_dead = fs_bdev_mark_dead, |
2142b88c | 1360 | .sync = fs_bdev_sync, |
87efb390 | 1361 | }; |
7ecd0b6f | 1362 | EXPORT_SYMBOL_GPL(fs_holder_ops); |
fe62c3a4 | 1363 | |
1da177e4 LT |
1364 | static int set_bdev_super(struct super_block *s, void *data) |
1365 | { | |
aca740ce | 1366 | s->s_dev = *(dev_t *)data; |
1da177e4 LT |
1367 | return 0; |
1368 | } | |
1369 | ||
fe62c3a4 DH |
1370 | static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc) |
1371 | { | |
1372 | return set_bdev_super(s, fc->sget_key); | |
1373 | } | |
1374 | ||
1375 | static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc) | |
1376 | { | |
aca740ce JK |
1377 | return !(s->s_iflags & SB_I_RETIRED) && |
1378 | s->s_dev == *(dev_t *)fc->sget_key; | |
1379 | } | |
1380 | ||
cf6da236 | 1381 | int setup_bdev_super(struct super_block *sb, int sb_flags, |
aca740ce JK |
1382 | struct fs_context *fc) |
1383 | { | |
1384 | blk_mode_t mode = sb_open_mode(sb_flags); | |
1385 | struct block_device *bdev; | |
1386 | ||
2ea6f689 | 1387 | bdev = blkdev_get_by_dev(sb->s_dev, mode, sb, &fs_holder_ops); |
aca740ce JK |
1388 | if (IS_ERR(bdev)) { |
1389 | if (fc) | |
1390 | errorf(fc, "%s: Can't open blockdev", fc->source); | |
1391 | return PTR_ERR(bdev); | |
1392 | } | |
1393 | ||
1394 | /* | |
1395 | * This really should be in blkdev_get_by_dev, but right now can't due | |
1396 | * to legacy issues that require us to allow opening a block device node | |
1397 | * writable from userspace even for a read-only block device. | |
1398 | */ | |
1399 | if ((mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) { | |
2ea6f689 | 1400 | blkdev_put(bdev, sb); |
aca740ce JK |
1401 | return -EACCES; |
1402 | } | |
1403 | ||
1404 | /* | |
1405 | * Until SB_BORN flag is set, there can be no active superblock | |
1406 | * references and thus no filesystem freezing. get_active_super() will | |
1407 | * just loop waiting for SB_BORN so even freeze_bdev() cannot proceed. | |
1408 | * | |
1409 | * It is enough to check bdev was not frozen before we set s_bdev. | |
1410 | */ | |
1411 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
1412 | if (bdev->bd_fsfreeze_count > 0) { | |
1413 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
1414 | if (fc) | |
1415 | warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev); | |
2ea6f689 | 1416 | blkdev_put(bdev, sb); |
aca740ce JK |
1417 | return -EBUSY; |
1418 | } | |
1419 | spin_lock(&sb_lock); | |
1420 | sb->s_bdev = bdev; | |
1421 | sb->s_bdi = bdi_get(bdev->bd_disk->bdi); | |
1422 | if (bdev_stable_writes(bdev)) | |
1423 | sb->s_iflags |= SB_I_STABLE_WRITES; | |
1424 | spin_unlock(&sb_lock); | |
1425 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
1426 | ||
1427 | snprintf(sb->s_id, sizeof(sb->s_id), "%pg", bdev); | |
1428 | shrinker_debugfs_rename(&sb->s_shrink, "sb-%s:%s", sb->s_type->name, | |
1429 | sb->s_id); | |
1430 | sb_set_blocksize(sb, block_size(bdev)); | |
1431 | return 0; | |
fe62c3a4 | 1432 | } |
cf6da236 | 1433 | EXPORT_SYMBOL_GPL(setup_bdev_super); |
fe62c3a4 DH |
1434 | |
1435 | /** | |
1436 | * get_tree_bdev - Get a superblock based on a single block device | |
1437 | * @fc: The filesystem context holding the parameters | |
1438 | * @fill_super: Helper to initialise a new superblock | |
1439 | */ | |
1440 | int get_tree_bdev(struct fs_context *fc, | |
1441 | int (*fill_super)(struct super_block *, | |
1442 | struct fs_context *)) | |
1443 | { | |
fe62c3a4 | 1444 | struct super_block *s; |
fe62c3a4 | 1445 | int error = 0; |
aca740ce | 1446 | dev_t dev; |
fe62c3a4 | 1447 | |
fe62c3a4 DH |
1448 | if (!fc->source) |
1449 | return invalf(fc, "No source specified"); | |
1450 | ||
aca740ce JK |
1451 | error = lookup_bdev(fc->source, &dev); |
1452 | if (error) { | |
1453 | errorf(fc, "%s: Can't lookup blockdev", fc->source); | |
1454 | return error; | |
fe62c3a4 DH |
1455 | } |
1456 | ||
1457 | fc->sb_flags |= SB_NOSEC; | |
aca740ce | 1458 | fc->sget_key = &dev; |
fe62c3a4 | 1459 | s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc); |
aca740ce | 1460 | if (IS_ERR(s)) |
fe62c3a4 DH |
1461 | return PTR_ERR(s); |
1462 | ||
1463 | if (s->s_root) { | |
1464 | /* Don't summarily change the RO/RW state. */ | |
1465 | if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) { | |
aca740ce | 1466 | warnf(fc, "%pg: Can't mount, would change RO state", s->s_bdev); |
fe62c3a4 | 1467 | deactivate_locked_super(s); |
fe62c3a4 DH |
1468 | return -EBUSY; |
1469 | } | |
aca740ce | 1470 | } else { |
fe62c3a4 | 1471 | /* |
aca740ce JK |
1472 | * We drop s_umount here because we need to open the bdev and |
1473 | * bdev->open_mutex ranks above s_umount (blkdev_put() -> | |
560e20e4 | 1474 | * bdev_mark_dead()). It is safe because we have active sb |
aca740ce | 1475 | * reference and SB_BORN is not set yet. |
fe62c3a4 | 1476 | */ |
0ed33598 | 1477 | super_unlock_excl(s); |
aca740ce | 1478 | error = setup_bdev_super(s, fc->sb_flags, fc); |
5e874914 | 1479 | __super_lock_excl(s); |
aca740ce JK |
1480 | if (!error) |
1481 | error = fill_super(s, fc); | |
fe62c3a4 DH |
1482 | if (error) { |
1483 | deactivate_locked_super(s); | |
1484 | return error; | |
1485 | } | |
fe62c3a4 | 1486 | s->s_flags |= SB_ACTIVE; |
fe62c3a4 DH |
1487 | } |
1488 | ||
1489 | BUG_ON(fc->root); | |
1490 | fc->root = dget(s->s_root); | |
1491 | return 0; | |
1492 | } | |
1493 | EXPORT_SYMBOL(get_tree_bdev); | |
1494 | ||
1da177e4 LT |
1495 | static int test_bdev_super(struct super_block *s, void *data) |
1496 | { | |
aca740ce | 1497 | return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data; |
1da177e4 LT |
1498 | } |
1499 | ||
152a0836 | 1500 | struct dentry *mount_bdev(struct file_system_type *fs_type, |
1da177e4 | 1501 | int flags, const char *dev_name, void *data, |
152a0836 | 1502 | int (*fill_super)(struct super_block *, void *, int)) |
1da177e4 | 1503 | { |
1da177e4 | 1504 | struct super_block *s; |
aca740ce JK |
1505 | int error; |
1506 | dev_t dev; | |
1da177e4 | 1507 | |
aca740ce JK |
1508 | error = lookup_bdev(dev_name, &dev); |
1509 | if (error) | |
1510 | return ERR_PTR(error); | |
1da177e4 | 1511 | |
aca740ce JK |
1512 | flags |= SB_NOSEC; |
1513 | s = sget(fs_type, test_bdev_super, set_bdev_super, flags, &dev); | |
1da177e4 | 1514 | if (IS_ERR(s)) |
aca740ce | 1515 | return ERR_CAST(s); |
1da177e4 LT |
1516 | |
1517 | if (s->s_root) { | |
e462ec50 | 1518 | if ((flags ^ s->s_flags) & SB_RDONLY) { |
74dbbdd7 | 1519 | deactivate_locked_super(s); |
aca740ce | 1520 | return ERR_PTR(-EBUSY); |
1da177e4 | 1521 | } |
aca740ce | 1522 | } else { |
4f331f01 | 1523 | /* |
aca740ce JK |
1524 | * We drop s_umount here because we need to open the bdev and |
1525 | * bdev->open_mutex ranks above s_umount (blkdev_put() -> | |
560e20e4 | 1526 | * bdev_mark_dead()). It is safe because we have active sb |
aca740ce | 1527 | * reference and SB_BORN is not set yet. |
4f331f01 | 1528 | */ |
0ed33598 | 1529 | super_unlock_excl(s); |
aca740ce | 1530 | error = setup_bdev_super(s, flags, NULL); |
5e874914 | 1531 | __super_lock_excl(s); |
aca740ce JK |
1532 | if (!error) |
1533 | error = fill_super(s, data, flags & SB_SILENT ? 1 : 0); | |
1da177e4 | 1534 | if (error) { |
74dbbdd7 | 1535 | deactivate_locked_super(s); |
aca740ce | 1536 | return ERR_PTR(error); |
fa675765 | 1537 | } |
454e2398 | 1538 | |
e462ec50 | 1539 | s->s_flags |= SB_ACTIVE; |
1da177e4 LT |
1540 | } |
1541 | ||
152a0836 | 1542 | return dget(s->s_root); |
152a0836 AV |
1543 | } |
1544 | EXPORT_SYMBOL(mount_bdev); | |
1545 | ||
1da177e4 LT |
1546 | void kill_block_super(struct super_block *sb) |
1547 | { | |
1548 | struct block_device *bdev = sb->s_bdev; | |
1549 | ||
1da177e4 | 1550 | generic_shutdown_super(sb); |
aca740ce JK |
1551 | if (bdev) { |
1552 | sync_blockdev(bdev); | |
2ea6f689 | 1553 | blkdev_put(bdev, sb); |
aca740ce | 1554 | } |
1da177e4 LT |
1555 | } |
1556 | ||
1557 | EXPORT_SYMBOL(kill_block_super); | |
9361401e | 1558 | #endif |
1da177e4 | 1559 | |
3c26ff6e | 1560 | struct dentry *mount_nodev(struct file_system_type *fs_type, |
1da177e4 | 1561 | int flags, void *data, |
3c26ff6e | 1562 | int (*fill_super)(struct super_block *, void *, int)) |
1da177e4 LT |
1563 | { |
1564 | int error; | |
9249e17f | 1565 | struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL); |
1da177e4 LT |
1566 | |
1567 | if (IS_ERR(s)) | |
3c26ff6e | 1568 | return ERR_CAST(s); |
1da177e4 | 1569 | |
e462ec50 | 1570 | error = fill_super(s, data, flags & SB_SILENT ? 1 : 0); |
1da177e4 | 1571 | if (error) { |
74dbbdd7 | 1572 | deactivate_locked_super(s); |
3c26ff6e | 1573 | return ERR_PTR(error); |
1da177e4 | 1574 | } |
e462ec50 | 1575 | s->s_flags |= SB_ACTIVE; |
3c26ff6e | 1576 | return dget(s->s_root); |
1da177e4 | 1577 | } |
3c26ff6e AV |
1578 | EXPORT_SYMBOL(mount_nodev); |
1579 | ||
a6097180 N |
1580 | int reconfigure_single(struct super_block *s, |
1581 | int flags, void *data) | |
8d0347f6 DH |
1582 | { |
1583 | struct fs_context *fc; | |
1584 | int ret; | |
1585 | ||
1586 | /* The caller really need to be passing fc down into mount_single(), | |
1587 | * then a chunk of this can be removed. [Bollocks -- AV] | |
1588 | * Better yet, reconfiguration shouldn't happen, but rather the second | |
1589 | * mount should be rejected if the parameters are not compatible. | |
1590 | */ | |
1591 | fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK); | |
1592 | if (IS_ERR(fc)) | |
1593 | return PTR_ERR(fc); | |
1594 | ||
1595 | ret = parse_monolithic_mount_data(fc, data); | |
1596 | if (ret < 0) | |
1597 | goto out; | |
1598 | ||
1599 | ret = reconfigure_super(fc); | |
1600 | out: | |
1601 | put_fs_context(fc); | |
1602 | return ret; | |
1603 | } | |
1604 | ||
1da177e4 LT |
1605 | static int compare_single(struct super_block *s, void *p) |
1606 | { | |
1607 | return 1; | |
1608 | } | |
1609 | ||
fc14f2fe | 1610 | struct dentry *mount_single(struct file_system_type *fs_type, |
1da177e4 | 1611 | int flags, void *data, |
fc14f2fe | 1612 | int (*fill_super)(struct super_block *, void *, int)) |
1da177e4 LT |
1613 | { |
1614 | struct super_block *s; | |
1615 | int error; | |
1616 | ||
9249e17f | 1617 | s = sget(fs_type, compare_single, set_anon_super, flags, NULL); |
1da177e4 | 1618 | if (IS_ERR(s)) |
fc14f2fe | 1619 | return ERR_CAST(s); |
1da177e4 | 1620 | if (!s->s_root) { |
e462ec50 | 1621 | error = fill_super(s, data, flags & SB_SILENT ? 1 : 0); |
8d0347f6 DH |
1622 | if (!error) |
1623 | s->s_flags |= SB_ACTIVE; | |
9329d1be | 1624 | } else { |
8d0347f6 DH |
1625 | error = reconfigure_single(s, flags, data); |
1626 | } | |
1627 | if (unlikely(error)) { | |
1628 | deactivate_locked_super(s); | |
1629 | return ERR_PTR(error); | |
1da177e4 | 1630 | } |
fc14f2fe AV |
1631 | return dget(s->s_root); |
1632 | } | |
1633 | EXPORT_SYMBOL(mount_single); | |
1634 | ||
9bc61ab1 DH |
1635 | /** |
1636 | * vfs_get_tree - Get the mountable root | |
1637 | * @fc: The superblock configuration context. | |
1638 | * | |
1639 | * The filesystem is invoked to get or create a superblock which can then later | |
1640 | * be used for mounting. The filesystem places a pointer to the root to be | |
1641 | * used for mounting in @fc->root. | |
1642 | */ | |
1643 | int vfs_get_tree(struct fs_context *fc) | |
1da177e4 | 1644 | { |
9d412a43 | 1645 | struct super_block *sb; |
9bc61ab1 | 1646 | int error; |
8089352a | 1647 | |
f3a09c92 AV |
1648 | if (fc->root) |
1649 | return -EBUSY; | |
1650 | ||
1651 | /* Get the mountable root in fc->root, with a ref on the root and a ref | |
1652 | * on the superblock. | |
1653 | */ | |
1654 | error = fc->ops->get_tree(fc); | |
9bc61ab1 DH |
1655 | if (error < 0) |
1656 | return error; | |
1da177e4 | 1657 | |
f3a09c92 AV |
1658 | if (!fc->root) { |
1659 | pr_err("Filesystem %s get_tree() didn't set fc->root\n", | |
1660 | fc->fs_type->name); | |
1661 | /* We don't know what the locking state of the superblock is - | |
1662 | * if there is a superblock. | |
1663 | */ | |
1664 | BUG(); | |
1665 | } | |
1666 | ||
9bc61ab1 | 1667 | sb = fc->root->d_sb; |
9d412a43 | 1668 | WARN_ON(!sb->s_bdi); |
79f546a6 DC |
1669 | |
1670 | /* | |
5e874914 CB |
1671 | * super_wake() contains a memory barrier which also care of |
1672 | * ordering for super_cache_count(). We place it before setting | |
1673 | * SB_BORN as the data dependency between the two functions is | |
1674 | * the superblock structure contents that we just set up, not | |
1675 | * the SB_BORN flag. | |
79f546a6 | 1676 | */ |
5e874914 | 1677 | super_wake(sb, SB_BORN); |
454e2398 | 1678 | |
9bc61ab1 | 1679 | error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL); |
c9ce29ed AV |
1680 | if (unlikely(error)) { |
1681 | fc_drop_locked(fc); | |
1682 | return error; | |
a10d7c22 AV |
1683 | } |
1684 | ||
42cb56ae JL |
1685 | /* |
1686 | * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE | |
1687 | * but s_maxbytes was an unsigned long long for many releases. Throw | |
1688 | * this warning for a little while to try and catch filesystems that | |
4358b567 | 1689 | * violate this rule. |
42cb56ae | 1690 | */ |
9d412a43 | 1691 | WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to " |
9bc61ab1 | 1692 | "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes); |
42cb56ae | 1693 | |
9bc61ab1 | 1694 | return 0; |
1da177e4 | 1695 | } |
9bc61ab1 | 1696 | EXPORT_SYMBOL(vfs_get_tree); |
1da177e4 | 1697 | |
fca39346 JK |
1698 | /* |
1699 | * Setup private BDI for given superblock. It gets automatically cleaned up | |
1700 | * in generic_shutdown_super(). | |
1701 | */ | |
1702 | int super_setup_bdi_name(struct super_block *sb, char *fmt, ...) | |
1703 | { | |
1704 | struct backing_dev_info *bdi; | |
1705 | int err; | |
1706 | va_list args; | |
1707 | ||
aef33c2f | 1708 | bdi = bdi_alloc(NUMA_NO_NODE); |
fca39346 JK |
1709 | if (!bdi) |
1710 | return -ENOMEM; | |
1711 | ||
fca39346 | 1712 | va_start(args, fmt); |
7c4cc300 | 1713 | err = bdi_register_va(bdi, fmt, args); |
fca39346 JK |
1714 | va_end(args); |
1715 | if (err) { | |
1716 | bdi_put(bdi); | |
1717 | return err; | |
1718 | } | |
1719 | WARN_ON(sb->s_bdi != &noop_backing_dev_info); | |
1720 | sb->s_bdi = bdi; | |
0b3ea092 | 1721 | sb->s_iflags |= SB_I_PERSB_BDI; |
fca39346 JK |
1722 | |
1723 | return 0; | |
1724 | } | |
1725 | EXPORT_SYMBOL(super_setup_bdi_name); | |
1726 | ||
1727 | /* | |
1728 | * Setup private BDI for given superblock. I gets automatically cleaned up | |
1729 | * in generic_shutdown_super(). | |
1730 | */ | |
1731 | int super_setup_bdi(struct super_block *sb) | |
1732 | { | |
1733 | static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); | |
1734 | ||
1735 | return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name, | |
1736 | atomic_long_inc_return(&bdi_seq)); | |
1737 | } | |
1738 | EXPORT_SYMBOL(super_setup_bdi); | |
1739 | ||
5accdf82 JK |
1740 | /** |
1741 | * sb_wait_write - wait until all writers to given file system finish | |
1742 | * @sb: the super for which we wait | |
1743 | * @level: type of writers we wait for (normal vs page fault) | |
1744 | * | |
1745 | * This function waits until there are no writers of given type to given file | |
8129ed29 | 1746 | * system. |
5accdf82 JK |
1747 | */ |
1748 | static void sb_wait_write(struct super_block *sb, int level) | |
1749 | { | |
8129ed29 | 1750 | percpu_down_write(sb->s_writers.rw_sem + level-1); |
8129ed29 | 1751 | } |
5accdf82 | 1752 | |
f1a96220 ON |
1753 | /* |
1754 | * We are going to return to userspace and forget about these locks, the | |
1755 | * ownership goes to the caller of thaw_super() which does unlock(). | |
1756 | */ | |
1757 | static void lockdep_sb_freeze_release(struct super_block *sb) | |
1758 | { | |
1759 | int level; | |
1760 | ||
1761 | for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--) | |
1762 | percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_); | |
1763 | } | |
1764 | ||
1765 | /* | |
1766 | * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb). | |
1767 | */ | |
1768 | static void lockdep_sb_freeze_acquire(struct super_block *sb) | |
8129ed29 ON |
1769 | { |
1770 | int level; | |
5accdf82 | 1771 | |
8129ed29 ON |
1772 | for (level = 0; level < SB_FREEZE_LEVELS; ++level) |
1773 | percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_); | |
f1a96220 ON |
1774 | } |
1775 | ||
2719c716 | 1776 | static void sb_freeze_unlock(struct super_block *sb, int level) |
f1a96220 | 1777 | { |
2719c716 | 1778 | for (level--; level >= 0; level--) |
8129ed29 | 1779 | percpu_up_write(sb->s_writers.rw_sem + level); |
5accdf82 JK |
1780 | } |
1781 | ||
18e9e510 | 1782 | /** |
7000d3c4 RD |
1783 | * freeze_super - lock the filesystem and force it into a consistent state |
1784 | * @sb: the super to lock | |
18e9e510 JB |
1785 | * |
1786 | * Syncs the super to make sure the filesystem is consistent and calls the fs's | |
1787 | * freeze_fs. Subsequent calls to this without first thawing the fs will return | |
1788 | * -EBUSY. | |
5accdf82 JK |
1789 | * |
1790 | * During this function, sb->s_writers.frozen goes through these values: | |
1791 | * | |
1792 | * SB_UNFROZEN: File system is normal, all writes progress as usual. | |
1793 | * | |
1794 | * SB_FREEZE_WRITE: The file system is in the process of being frozen. New | |
1795 | * writes should be blocked, though page faults are still allowed. We wait for | |
1796 | * all writes to complete and then proceed to the next stage. | |
1797 | * | |
1798 | * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked | |
1799 | * but internal fs threads can still modify the filesystem (although they | |
1800 | * should not dirty new pages or inodes), writeback can run etc. After waiting | |
1801 | * for all running page faults we sync the filesystem which will clean all | |
1802 | * dirty pages and inodes (no new dirty pages or inodes can be created when | |
1803 | * sync is running). | |
1804 | * | |
1805 | * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs | |
1806 | * modification are blocked (e.g. XFS preallocation truncation on inode | |
1807 | * reclaim). This is usually implemented by blocking new transactions for | |
1808 | * filesystems that have them and need this additional guard. After all | |
1809 | * internal writers are finished we call ->freeze_fs() to finish filesystem | |
1810 | * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is | |
1811 | * mostly auxiliary for filesystems to verify they do not modify frozen fs. | |
1812 | * | |
1813 | * sb->s_writers.frozen is protected by sb->s_umount. | |
18e9e510 JB |
1814 | */ |
1815 | int freeze_super(struct super_block *sb) | |
1816 | { | |
1817 | int ret; | |
1818 | ||
1819 | atomic_inc(&sb->s_active); | |
5e874914 | 1820 | __super_lock_excl(sb); |
5accdf82 | 1821 | if (sb->s_writers.frozen != SB_UNFROZEN) { |
18e9e510 JB |
1822 | deactivate_locked_super(sb); |
1823 | return -EBUSY; | |
1824 | } | |
1825 | ||
e462ec50 | 1826 | if (!(sb->s_flags & SB_BORN)) { |
0ed33598 | 1827 | super_unlock_excl(sb); |
dabe0dc1 AV |
1828 | return 0; /* sic - it's "nothing to do" */ |
1829 | } | |
1830 | ||
bc98a42c | 1831 | if (sb_rdonly(sb)) { |
5accdf82 JK |
1832 | /* Nothing to do really... */ |
1833 | sb->s_writers.frozen = SB_FREEZE_COMPLETE; | |
0ed33598 | 1834 | super_unlock_excl(sb); |
18e9e510 JB |
1835 | return 0; |
1836 | } | |
1837 | ||
5accdf82 | 1838 | sb->s_writers.frozen = SB_FREEZE_WRITE; |
5accdf82 | 1839 | /* Release s_umount to preserve sb_start_write -> s_umount ordering */ |
0ed33598 | 1840 | super_unlock_excl(sb); |
5accdf82 | 1841 | sb_wait_write(sb, SB_FREEZE_WRITE); |
5e874914 | 1842 | __super_lock_excl(sb); |
5accdf82 JK |
1843 | |
1844 | /* Now we go and block page faults... */ | |
5accdf82 | 1845 | sb->s_writers.frozen = SB_FREEZE_PAGEFAULT; |
5accdf82 JK |
1846 | sb_wait_write(sb, SB_FREEZE_PAGEFAULT); |
1847 | ||
1848 | /* All writers are done so after syncing there won't be dirty data */ | |
2719c716 DW |
1849 | ret = sync_filesystem(sb); |
1850 | if (ret) { | |
1851 | sb->s_writers.frozen = SB_UNFROZEN; | |
1852 | sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT); | |
2719c716 DW |
1853 | deactivate_locked_super(sb); |
1854 | return ret; | |
1855 | } | |
18e9e510 | 1856 | |
5accdf82 JK |
1857 | /* Now wait for internal filesystem counter */ |
1858 | sb->s_writers.frozen = SB_FREEZE_FS; | |
5accdf82 | 1859 | sb_wait_write(sb, SB_FREEZE_FS); |
18e9e510 | 1860 | |
18e9e510 JB |
1861 | if (sb->s_op->freeze_fs) { |
1862 | ret = sb->s_op->freeze_fs(sb); | |
1863 | if (ret) { | |
1864 | printk(KERN_ERR | |
1865 | "VFS:Filesystem freeze failed\n"); | |
5accdf82 | 1866 | sb->s_writers.frozen = SB_UNFROZEN; |
2719c716 | 1867 | sb_freeze_unlock(sb, SB_FREEZE_FS); |
18e9e510 JB |
1868 | deactivate_locked_super(sb); |
1869 | return ret; | |
1870 | } | |
1871 | } | |
5accdf82 | 1872 | /* |
89f39af1 ON |
1873 | * For debugging purposes so that fs can warn if it sees write activity |
1874 | * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super(). | |
5accdf82 JK |
1875 | */ |
1876 | sb->s_writers.frozen = SB_FREEZE_COMPLETE; | |
f1a96220 | 1877 | lockdep_sb_freeze_release(sb); |
0ed33598 | 1878 | super_unlock_excl(sb); |
18e9e510 JB |
1879 | return 0; |
1880 | } | |
1881 | EXPORT_SYMBOL(freeze_super); | |
1882 | ||
08fdc8a0 | 1883 | static int thaw_super_locked(struct super_block *sb) |
18e9e510 JB |
1884 | { |
1885 | int error; | |
1886 | ||
89f39af1 | 1887 | if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) { |
0ed33598 | 1888 | super_unlock_excl(sb); |
18e9e510 JB |
1889 | return -EINVAL; |
1890 | } | |
1891 | ||
bc98a42c | 1892 | if (sb_rdonly(sb)) { |
8129ed29 | 1893 | sb->s_writers.frozen = SB_UNFROZEN; |
18e9e510 | 1894 | goto out; |
8129ed29 | 1895 | } |
18e9e510 | 1896 | |
f1a96220 ON |
1897 | lockdep_sb_freeze_acquire(sb); |
1898 | ||
18e9e510 JB |
1899 | if (sb->s_op->unfreeze_fs) { |
1900 | error = sb->s_op->unfreeze_fs(sb); | |
1901 | if (error) { | |
1902 | printk(KERN_ERR | |
1903 | "VFS:Filesystem thaw failed\n"); | |
f1a96220 | 1904 | lockdep_sb_freeze_release(sb); |
0ed33598 | 1905 | super_unlock_excl(sb); |
18e9e510 JB |
1906 | return error; |
1907 | } | |
1908 | } | |
1909 | ||
5accdf82 | 1910 | sb->s_writers.frozen = SB_UNFROZEN; |
2719c716 | 1911 | sb_freeze_unlock(sb, SB_FREEZE_FS); |
8129ed29 | 1912 | out: |
18e9e510 | 1913 | deactivate_locked_super(sb); |
18e9e510 JB |
1914 | return 0; |
1915 | } | |
08fdc8a0 | 1916 | |
961f3c89 MCC |
1917 | /** |
1918 | * thaw_super -- unlock filesystem | |
1919 | * @sb: the super to thaw | |
1920 | * | |
1921 | * Unlocks the filesystem and marks it writeable again after freeze_super(). | |
1922 | */ | |
08fdc8a0 MG |
1923 | int thaw_super(struct super_block *sb) |
1924 | { | |
5e874914 | 1925 | __super_lock_excl(sb); |
08fdc8a0 MG |
1926 | return thaw_super_locked(sb); |
1927 | } | |
18e9e510 | 1928 | EXPORT_SYMBOL(thaw_super); |
439bc39b CH |
1929 | |
1930 | /* | |
1931 | * Create workqueue for deferred direct IO completions. We allocate the | |
1932 | * workqueue when it's first needed. This avoids creating workqueue for | |
1933 | * filesystems that don't need it and also allows us to create the workqueue | |
1934 | * late enough so the we can include s_id in the name of the workqueue. | |
1935 | */ | |
1936 | int sb_init_dio_done_wq(struct super_block *sb) | |
1937 | { | |
1938 | struct workqueue_struct *old; | |
1939 | struct workqueue_struct *wq = alloc_workqueue("dio/%s", | |
1940 | WQ_MEM_RECLAIM, 0, | |
1941 | sb->s_id); | |
1942 | if (!wq) | |
1943 | return -ENOMEM; | |
1944 | /* | |
1945 | * This has to be atomic as more DIOs can race to create the workqueue | |
1946 | */ | |
1947 | old = cmpxchg(&sb->s_dio_done_wq, NULL, wq); | |
1948 | /* Someone created workqueue before us? Free ours... */ | |
1949 | if (old) | |
1950 | destroy_workqueue(wq); | |
1951 | return 0; | |
1952 | } |