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