]>
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/module.h> | |
24 | #include <linux/slab.h> | |
25 | #include <linux/acct.h> | |
26 | #include <linux/blkdev.h> | |
27 | #include <linux/mount.h> | |
28 | #include <linux/security.h> | |
29 | #include <linux/writeback.h> /* for the emergency remount stuff */ | |
30 | #include <linux/idr.h> | |
31 | #include <linux/mutex.h> | |
32 | #include <linux/backing-dev.h> | |
33 | #include <linux/rculist_bl.h> | |
34 | #include <linux/cleancache.h> | |
35 | #include "internal.h" | |
36 | ||
37 | ||
38 | LIST_HEAD(super_blocks); | |
39 | DEFINE_SPINLOCK(sb_lock); | |
40 | ||
41 | /* | |
42 | * One thing we have to be careful of with a per-sb shrinker is that we don't | |
43 | * drop the last active reference to the superblock from within the shrinker. | |
44 | * If that happens we could trigger unregistering the shrinker from within the | |
45 | * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we | |
46 | * take a passive reference to the superblock to avoid this from occurring. | |
47 | */ | |
48 | static int prune_super(struct shrinker *shrink, struct shrink_control *sc) | |
49 | { | |
50 | struct super_block *sb; | |
51 | int fs_objects = 0; | |
52 | int total_objects; | |
53 | ||
54 | sb = container_of(shrink, struct super_block, s_shrink); | |
55 | ||
56 | /* | |
57 | * Deadlock avoidance. We may hold various FS locks, and we don't want | |
58 | * to recurse into the FS that called us in clear_inode() and friends.. | |
59 | */ | |
60 | if (sc->nr_to_scan && !(sc->gfp_mask & __GFP_FS)) | |
61 | return -1; | |
62 | ||
63 | if (!grab_super_passive(sb)) | |
64 | return !sc->nr_to_scan ? 0 : -1; | |
65 | ||
66 | if (sb->s_op && sb->s_op->nr_cached_objects) | |
67 | fs_objects = sb->s_op->nr_cached_objects(sb); | |
68 | ||
69 | total_objects = sb->s_nr_dentry_unused + | |
70 | sb->s_nr_inodes_unused + fs_objects + 1; | |
71 | ||
72 | if (sc->nr_to_scan) { | |
73 | int dentries; | |
74 | int inodes; | |
75 | ||
76 | /* proportion the scan between the caches */ | |
77 | dentries = (sc->nr_to_scan * sb->s_nr_dentry_unused) / | |
78 | total_objects; | |
79 | inodes = (sc->nr_to_scan * sb->s_nr_inodes_unused) / | |
80 | total_objects; | |
81 | if (fs_objects) | |
82 | fs_objects = (sc->nr_to_scan * fs_objects) / | |
83 | total_objects; | |
84 | /* | |
85 | * prune the dcache first as the icache is pinned by it, then | |
86 | * prune the icache, followed by the filesystem specific caches | |
87 | */ | |
88 | prune_dcache_sb(sb, dentries); | |
89 | prune_icache_sb(sb, inodes); | |
90 | ||
91 | if (fs_objects && sb->s_op->free_cached_objects) { | |
92 | sb->s_op->free_cached_objects(sb, fs_objects); | |
93 | fs_objects = sb->s_op->nr_cached_objects(sb); | |
94 | } | |
95 | total_objects = sb->s_nr_dentry_unused + | |
96 | sb->s_nr_inodes_unused + fs_objects; | |
97 | } | |
98 | ||
99 | total_objects = (total_objects / 100) * sysctl_vfs_cache_pressure; | |
100 | drop_super(sb); | |
101 | return total_objects; | |
102 | } | |
103 | ||
104 | /** | |
105 | * alloc_super - create new superblock | |
106 | * @type: filesystem type superblock should belong to | |
107 | * | |
108 | * Allocates and initializes a new &struct super_block. alloc_super() | |
109 | * returns a pointer new superblock or %NULL if allocation had failed. | |
110 | */ | |
111 | static struct super_block *alloc_super(struct file_system_type *type) | |
112 | { | |
113 | struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER); | |
114 | static const struct super_operations default_op; | |
115 | ||
116 | if (s) { | |
117 | if (security_sb_alloc(s)) { | |
118 | kfree(s); | |
119 | s = NULL; | |
120 | goto out; | |
121 | } | |
122 | #ifdef CONFIG_SMP | |
123 | s->s_files = alloc_percpu(struct list_head); | |
124 | if (!s->s_files) { | |
125 | security_sb_free(s); | |
126 | kfree(s); | |
127 | s = NULL; | |
128 | goto out; | |
129 | } else { | |
130 | int i; | |
131 | ||
132 | for_each_possible_cpu(i) | |
133 | INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i)); | |
134 | } | |
135 | #else | |
136 | INIT_LIST_HEAD(&s->s_files); | |
137 | #endif | |
138 | s->s_bdi = &default_backing_dev_info; | |
139 | INIT_HLIST_NODE(&s->s_instances); | |
140 | INIT_HLIST_BL_HEAD(&s->s_anon); | |
141 | INIT_LIST_HEAD(&s->s_inodes); | |
142 | INIT_LIST_HEAD(&s->s_dentry_lru); | |
143 | INIT_LIST_HEAD(&s->s_inode_lru); | |
144 | spin_lock_init(&s->s_inode_lru_lock); | |
145 | INIT_LIST_HEAD(&s->s_mounts); | |
146 | init_rwsem(&s->s_umount); | |
147 | mutex_init(&s->s_lock); | |
148 | lockdep_set_class(&s->s_umount, &type->s_umount_key); | |
149 | /* | |
150 | * The locking rules for s_lock are up to the | |
151 | * filesystem. For example ext3fs has different | |
152 | * lock ordering than usbfs: | |
153 | */ | |
154 | lockdep_set_class(&s->s_lock, &type->s_lock_key); | |
155 | /* | |
156 | * sget() can have s_umount recursion. | |
157 | * | |
158 | * When it cannot find a suitable sb, it allocates a new | |
159 | * one (this one), and tries again to find a suitable old | |
160 | * one. | |
161 | * | |
162 | * In case that succeeds, it will acquire the s_umount | |
163 | * lock of the old one. Since these are clearly distrinct | |
164 | * locks, and this object isn't exposed yet, there's no | |
165 | * risk of deadlocks. | |
166 | * | |
167 | * Annotate this by putting this lock in a different | |
168 | * subclass. | |
169 | */ | |
170 | down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING); | |
171 | s->s_count = 1; | |
172 | atomic_set(&s->s_active, 1); | |
173 | mutex_init(&s->s_vfs_rename_mutex); | |
174 | lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key); | |
175 | mutex_init(&s->s_dquot.dqio_mutex); | |
176 | mutex_init(&s->s_dquot.dqonoff_mutex); | |
177 | init_rwsem(&s->s_dquot.dqptr_sem); | |
178 | init_waitqueue_head(&s->s_wait_unfrozen); | |
179 | s->s_maxbytes = MAX_NON_LFS; | |
180 | s->s_op = &default_op; | |
181 | s->s_time_gran = 1000000000; | |
182 | s->cleancache_poolid = -1; | |
183 | ||
184 | s->s_shrink.seeks = DEFAULT_SEEKS; | |
185 | s->s_shrink.shrink = prune_super; | |
186 | s->s_shrink.batch = 1024; | |
187 | } | |
188 | out: | |
189 | return s; | |
190 | } | |
191 | ||
192 | /** | |
193 | * destroy_super - frees a superblock | |
194 | * @s: superblock to free | |
195 | * | |
196 | * Frees a superblock. | |
197 | */ | |
198 | static inline void destroy_super(struct super_block *s) | |
199 | { | |
200 | #ifdef CONFIG_SMP | |
201 | free_percpu(s->s_files); | |
202 | #endif | |
203 | security_sb_free(s); | |
204 | WARN_ON(!list_empty(&s->s_mounts)); | |
205 | kfree(s->s_subtype); | |
206 | kfree(s->s_options); | |
207 | kfree(s); | |
208 | } | |
209 | ||
210 | /* Superblock refcounting */ | |
211 | ||
212 | /* | |
213 | * Drop a superblock's refcount. The caller must hold sb_lock. | |
214 | */ | |
215 | static void __put_super(struct super_block *sb) | |
216 | { | |
217 | if (!--sb->s_count) { | |
218 | list_del_init(&sb->s_list); | |
219 | destroy_super(sb); | |
220 | } | |
221 | } | |
222 | ||
223 | /** | |
224 | * put_super - drop a temporary reference to superblock | |
225 | * @sb: superblock in question | |
226 | * | |
227 | * Drops a temporary reference, frees superblock if there's no | |
228 | * references left. | |
229 | */ | |
230 | static void put_super(struct super_block *sb) | |
231 | { | |
232 | spin_lock(&sb_lock); | |
233 | __put_super(sb); | |
234 | spin_unlock(&sb_lock); | |
235 | } | |
236 | ||
237 | ||
238 | /** | |
239 | * deactivate_locked_super - drop an active reference to superblock | |
240 | * @s: superblock to deactivate | |
241 | * | |
242 | * Drops an active reference to superblock, converting it into a temprory | |
243 | * one if there is no other active references left. In that case we | |
244 | * tell fs driver to shut it down and drop the temporary reference we | |
245 | * had just acquired. | |
246 | * | |
247 | * Caller holds exclusive lock on superblock; that lock is released. | |
248 | */ | |
249 | void deactivate_locked_super(struct super_block *s) | |
250 | { | |
251 | struct file_system_type *fs = s->s_type; | |
252 | if (atomic_dec_and_test(&s->s_active)) { | |
253 | cleancache_flush_fs(s); | |
254 | fs->kill_sb(s); | |
255 | ||
256 | /* caches are now gone, we can safely kill the shrinker now */ | |
257 | unregister_shrinker(&s->s_shrink); | |
258 | ||
259 | /* | |
260 | * We need to call rcu_barrier so all the delayed rcu free | |
261 | * inodes are flushed before we release the fs module. | |
262 | */ | |
263 | rcu_barrier(); | |
264 | put_filesystem(fs); | |
265 | put_super(s); | |
266 | } else { | |
267 | up_write(&s->s_umount); | |
268 | } | |
269 | } | |
270 | ||
271 | EXPORT_SYMBOL(deactivate_locked_super); | |
272 | ||
273 | /** | |
274 | * deactivate_super - drop an active reference to superblock | |
275 | * @s: superblock to deactivate | |
276 | * | |
277 | * Variant of deactivate_locked_super(), except that superblock is *not* | |
278 | * locked by caller. If we are going to drop the final active reference, | |
279 | * lock will be acquired prior to that. | |
280 | */ | |
281 | void deactivate_super(struct super_block *s) | |
282 | { | |
283 | if (!atomic_add_unless(&s->s_active, -1, 1)) { | |
284 | down_write(&s->s_umount); | |
285 | deactivate_locked_super(s); | |
286 | } | |
287 | } | |
288 | ||
289 | EXPORT_SYMBOL(deactivate_super); | |
290 | ||
291 | /** | |
292 | * grab_super - acquire an active reference | |
293 | * @s: reference we are trying to make active | |
294 | * | |
295 | * Tries to acquire an active reference. grab_super() is used when we | |
296 | * had just found a superblock in super_blocks or fs_type->fs_supers | |
297 | * and want to turn it into a full-blown active reference. grab_super() | |
298 | * is called with sb_lock held and drops it. Returns 1 in case of | |
299 | * success, 0 if we had failed (superblock contents was already dead or | |
300 | * dying when grab_super() had been called). | |
301 | */ | |
302 | static int grab_super(struct super_block *s) __releases(sb_lock) | |
303 | { | |
304 | if (atomic_inc_not_zero(&s->s_active)) { | |
305 | spin_unlock(&sb_lock); | |
306 | return 1; | |
307 | } | |
308 | /* it's going away */ | |
309 | s->s_count++; | |
310 | spin_unlock(&sb_lock); | |
311 | /* wait for it to die */ | |
312 | down_write(&s->s_umount); | |
313 | up_write(&s->s_umount); | |
314 | put_super(s); | |
315 | return 0; | |
316 | } | |
317 | ||
318 | /* | |
319 | * grab_super_passive - acquire a passive reference | |
320 | * @s: reference we are trying to grab | |
321 | * | |
322 | * Tries to acquire a passive reference. This is used in places where we | |
323 | * cannot take an active reference but we need to ensure that the | |
324 | * superblock does not go away while we are working on it. It returns | |
325 | * false if a reference was not gained, and returns true with the s_umount | |
326 | * lock held in read mode if a reference is gained. On successful return, | |
327 | * the caller must drop the s_umount lock and the passive reference when | |
328 | * done. | |
329 | */ | |
330 | bool grab_super_passive(struct super_block *sb) | |
331 | { | |
332 | spin_lock(&sb_lock); | |
333 | if (hlist_unhashed(&sb->s_instances)) { | |
334 | spin_unlock(&sb_lock); | |
335 | return false; | |
336 | } | |
337 | ||
338 | sb->s_count++; | |
339 | spin_unlock(&sb_lock); | |
340 | ||
341 | if (down_read_trylock(&sb->s_umount)) { | |
342 | if (sb->s_root && (sb->s_flags & MS_BORN)) | |
343 | return true; | |
344 | up_read(&sb->s_umount); | |
345 | } | |
346 | ||
347 | put_super(sb); | |
348 | return false; | |
349 | } | |
350 | ||
351 | /* | |
352 | * Superblock locking. We really ought to get rid of these two. | |
353 | */ | |
354 | void lock_super(struct super_block * sb) | |
355 | { | |
356 | mutex_lock(&sb->s_lock); | |
357 | } | |
358 | ||
359 | void unlock_super(struct super_block * sb) | |
360 | { | |
361 | mutex_unlock(&sb->s_lock); | |
362 | } | |
363 | ||
364 | EXPORT_SYMBOL(lock_super); | |
365 | EXPORT_SYMBOL(unlock_super); | |
366 | ||
367 | /** | |
368 | * generic_shutdown_super - common helper for ->kill_sb() | |
369 | * @sb: superblock to kill | |
370 | * | |
371 | * generic_shutdown_super() does all fs-independent work on superblock | |
372 | * shutdown. Typical ->kill_sb() should pick all fs-specific objects | |
373 | * that need destruction out of superblock, call generic_shutdown_super() | |
374 | * and release aforementioned objects. Note: dentries and inodes _are_ | |
375 | * taken care of and do not need specific handling. | |
376 | * | |
377 | * Upon calling this function, the filesystem may no longer alter or | |
378 | * rearrange the set of dentries belonging to this super_block, nor may it | |
379 | * change the attachments of dentries to inodes. | |
380 | */ | |
381 | void generic_shutdown_super(struct super_block *sb) | |
382 | { | |
383 | const struct super_operations *sop = sb->s_op; | |
384 | ||
385 | if (sb->s_root) { | |
386 | shrink_dcache_for_umount(sb); | |
387 | sync_filesystem(sb); | |
388 | sb->s_flags &= ~MS_ACTIVE; | |
389 | ||
390 | fsnotify_unmount_inodes(&sb->s_inodes); | |
391 | ||
392 | evict_inodes(sb); | |
393 | ||
394 | if (sop->put_super) | |
395 | sop->put_super(sb); | |
396 | ||
397 | if (!list_empty(&sb->s_inodes)) { | |
398 | printk("VFS: Busy inodes after unmount of %s. " | |
399 | "Self-destruct in 5 seconds. Have a nice day...\n", | |
400 | sb->s_id); | |
401 | } | |
402 | } | |
403 | spin_lock(&sb_lock); | |
404 | /* should be initialized for __put_super_and_need_restart() */ | |
405 | hlist_del_init(&sb->s_instances); | |
406 | spin_unlock(&sb_lock); | |
407 | up_write(&sb->s_umount); | |
408 | } | |
409 | ||
410 | EXPORT_SYMBOL(generic_shutdown_super); | |
411 | ||
412 | /** | |
413 | * sget - find or create a superblock | |
414 | * @type: filesystem type superblock should belong to | |
415 | * @test: comparison callback | |
416 | * @set: setup callback | |
417 | * @data: argument to each of them | |
418 | */ | |
419 | struct super_block *sget(struct file_system_type *type, | |
420 | int (*test)(struct super_block *,void *), | |
421 | int (*set)(struct super_block *,void *), | |
422 | void *data) | |
423 | { | |
424 | struct super_block *s = NULL; | |
425 | struct hlist_node *node; | |
426 | struct super_block *old; | |
427 | int err; | |
428 | ||
429 | retry: | |
430 | spin_lock(&sb_lock); | |
431 | if (test) { | |
432 | hlist_for_each_entry(old, node, &type->fs_supers, s_instances) { | |
433 | if (!test(old, data)) | |
434 | continue; | |
435 | if (!grab_super(old)) | |
436 | goto retry; | |
437 | if (s) { | |
438 | up_write(&s->s_umount); | |
439 | destroy_super(s); | |
440 | s = NULL; | |
441 | } | |
442 | down_write(&old->s_umount); | |
443 | if (unlikely(!(old->s_flags & MS_BORN))) { | |
444 | deactivate_locked_super(old); | |
445 | goto retry; | |
446 | } | |
447 | return old; | |
448 | } | |
449 | } | |
450 | if (!s) { | |
451 | spin_unlock(&sb_lock); | |
452 | s = alloc_super(type); | |
453 | if (!s) | |
454 | return ERR_PTR(-ENOMEM); | |
455 | goto retry; | |
456 | } | |
457 | ||
458 | err = set(s, data); | |
459 | if (err) { | |
460 | spin_unlock(&sb_lock); | |
461 | up_write(&s->s_umount); | |
462 | destroy_super(s); | |
463 | return ERR_PTR(err); | |
464 | } | |
465 | s->s_type = type; | |
466 | strlcpy(s->s_id, type->name, sizeof(s->s_id)); | |
467 | list_add_tail(&s->s_list, &super_blocks); | |
468 | hlist_add_head(&s->s_instances, &type->fs_supers); | |
469 | spin_unlock(&sb_lock); | |
470 | get_filesystem(type); | |
471 | register_shrinker(&s->s_shrink); | |
472 | return s; | |
473 | } | |
474 | ||
475 | EXPORT_SYMBOL(sget); | |
476 | ||
477 | void drop_super(struct super_block *sb) | |
478 | { | |
479 | up_read(&sb->s_umount); | |
480 | put_super(sb); | |
481 | } | |
482 | ||
483 | EXPORT_SYMBOL(drop_super); | |
484 | ||
485 | /** | |
486 | * sync_supers - helper for periodic superblock writeback | |
487 | * | |
488 | * Call the write_super method if present on all dirty superblocks in | |
489 | * the system. This is for the periodic writeback used by most older | |
490 | * filesystems. For data integrity superblock writeback use | |
491 | * sync_filesystems() instead. | |
492 | * | |
493 | * Note: check the dirty flag before waiting, so we don't | |
494 | * hold up the sync while mounting a device. (The newly | |
495 | * mounted device won't need syncing.) | |
496 | */ | |
497 | void sync_supers(void) | |
498 | { | |
499 | struct super_block *sb, *p = NULL; | |
500 | ||
501 | spin_lock(&sb_lock); | |
502 | list_for_each_entry(sb, &super_blocks, s_list) { | |
503 | if (hlist_unhashed(&sb->s_instances)) | |
504 | continue; | |
505 | if (sb->s_op->write_super && sb->s_dirt) { | |
506 | sb->s_count++; | |
507 | spin_unlock(&sb_lock); | |
508 | ||
509 | down_read(&sb->s_umount); | |
510 | if (sb->s_root && sb->s_dirt && (sb->s_flags & MS_BORN)) | |
511 | sb->s_op->write_super(sb); | |
512 | up_read(&sb->s_umount); | |
513 | ||
514 | spin_lock(&sb_lock); | |
515 | if (p) | |
516 | __put_super(p); | |
517 | p = sb; | |
518 | } | |
519 | } | |
520 | if (p) | |
521 | __put_super(p); | |
522 | spin_unlock(&sb_lock); | |
523 | } | |
524 | ||
525 | /** | |
526 | * iterate_supers - call function for all active superblocks | |
527 | * @f: function to call | |
528 | * @arg: argument to pass to it | |
529 | * | |
530 | * Scans the superblock list and calls given function, passing it | |
531 | * locked superblock and given argument. | |
532 | */ | |
533 | void iterate_supers(void (*f)(struct super_block *, void *), void *arg) | |
534 | { | |
535 | struct super_block *sb, *p = NULL; | |
536 | ||
537 | spin_lock(&sb_lock); | |
538 | list_for_each_entry(sb, &super_blocks, s_list) { | |
539 | if (hlist_unhashed(&sb->s_instances)) | |
540 | continue; | |
541 | sb->s_count++; | |
542 | spin_unlock(&sb_lock); | |
543 | ||
544 | down_read(&sb->s_umount); | |
545 | if (sb->s_root && (sb->s_flags & MS_BORN)) | |
546 | f(sb, arg); | |
547 | up_read(&sb->s_umount); | |
548 | ||
549 | spin_lock(&sb_lock); | |
550 | if (p) | |
551 | __put_super(p); | |
552 | p = sb; | |
553 | } | |
554 | if (p) | |
555 | __put_super(p); | |
556 | spin_unlock(&sb_lock); | |
557 | } | |
558 | ||
559 | /** | |
560 | * iterate_supers_type - call function for superblocks of given type | |
561 | * @type: fs type | |
562 | * @f: function to call | |
563 | * @arg: argument to pass to it | |
564 | * | |
565 | * Scans the superblock list and calls given function, passing it | |
566 | * locked superblock and given argument. | |
567 | */ | |
568 | void iterate_supers_type(struct file_system_type *type, | |
569 | void (*f)(struct super_block *, void *), void *arg) | |
570 | { | |
571 | struct super_block *sb, *p = NULL; | |
572 | struct hlist_node *node; | |
573 | ||
574 | spin_lock(&sb_lock); | |
575 | hlist_for_each_entry(sb, node, &type->fs_supers, s_instances) { | |
576 | sb->s_count++; | |
577 | spin_unlock(&sb_lock); | |
578 | ||
579 | down_read(&sb->s_umount); | |
580 | if (sb->s_root && (sb->s_flags & MS_BORN)) | |
581 | f(sb, arg); | |
582 | up_read(&sb->s_umount); | |
583 | ||
584 | spin_lock(&sb_lock); | |
585 | if (p) | |
586 | __put_super(p); | |
587 | p = sb; | |
588 | } | |
589 | if (p) | |
590 | __put_super(p); | |
591 | spin_unlock(&sb_lock); | |
592 | } | |
593 | ||
594 | EXPORT_SYMBOL(iterate_supers_type); | |
595 | ||
596 | /** | |
597 | * get_super - get the superblock of a device | |
598 | * @bdev: device to get the superblock for | |
599 | * | |
600 | * Scans the superblock list and finds the superblock of the file system | |
601 | * mounted on the device given. %NULL is returned if no match is found. | |
602 | */ | |
603 | ||
604 | struct super_block *get_super(struct block_device *bdev) | |
605 | { | |
606 | struct super_block *sb; | |
607 | ||
608 | if (!bdev) | |
609 | return NULL; | |
610 | ||
611 | spin_lock(&sb_lock); | |
612 | rescan: | |
613 | list_for_each_entry(sb, &super_blocks, s_list) { | |
614 | if (hlist_unhashed(&sb->s_instances)) | |
615 | continue; | |
616 | if (sb->s_bdev == bdev) { | |
617 | sb->s_count++; | |
618 | spin_unlock(&sb_lock); | |
619 | down_read(&sb->s_umount); | |
620 | /* still alive? */ | |
621 | if (sb->s_root && (sb->s_flags & MS_BORN)) | |
622 | return sb; | |
623 | up_read(&sb->s_umount); | |
624 | /* nope, got unmounted */ | |
625 | spin_lock(&sb_lock); | |
626 | __put_super(sb); | |
627 | goto rescan; | |
628 | } | |
629 | } | |
630 | spin_unlock(&sb_lock); | |
631 | return NULL; | |
632 | } | |
633 | ||
634 | EXPORT_SYMBOL(get_super); | |
635 | ||
636 | /** | |
637 | * get_active_super - get an active reference to the superblock of a device | |
638 | * @bdev: device to get the superblock for | |
639 | * | |
640 | * Scans the superblock list and finds the superblock of the file system | |
641 | * mounted on the device given. Returns the superblock with an active | |
642 | * reference or %NULL if none was found. | |
643 | */ | |
644 | struct super_block *get_active_super(struct block_device *bdev) | |
645 | { | |
646 | struct super_block *sb; | |
647 | ||
648 | if (!bdev) | |
649 | return NULL; | |
650 | ||
651 | restart: | |
652 | spin_lock(&sb_lock); | |
653 | list_for_each_entry(sb, &super_blocks, s_list) { | |
654 | if (hlist_unhashed(&sb->s_instances)) | |
655 | continue; | |
656 | if (sb->s_bdev == bdev) { | |
657 | if (grab_super(sb)) /* drops sb_lock */ | |
658 | return sb; | |
659 | else | |
660 | goto restart; | |
661 | } | |
662 | } | |
663 | spin_unlock(&sb_lock); | |
664 | return NULL; | |
665 | } | |
666 | ||
667 | struct super_block *user_get_super(dev_t dev) | |
668 | { | |
669 | struct super_block *sb; | |
670 | ||
671 | spin_lock(&sb_lock); | |
672 | rescan: | |
673 | list_for_each_entry(sb, &super_blocks, s_list) { | |
674 | if (hlist_unhashed(&sb->s_instances)) | |
675 | continue; | |
676 | if (sb->s_dev == dev) { | |
677 | sb->s_count++; | |
678 | spin_unlock(&sb_lock); | |
679 | down_read(&sb->s_umount); | |
680 | /* still alive? */ | |
681 | if (sb->s_root && (sb->s_flags & MS_BORN)) | |
682 | return sb; | |
683 | up_read(&sb->s_umount); | |
684 | /* nope, got unmounted */ | |
685 | spin_lock(&sb_lock); | |
686 | __put_super(sb); | |
687 | goto rescan; | |
688 | } | |
689 | } | |
690 | spin_unlock(&sb_lock); | |
691 | return NULL; | |
692 | } | |
693 | ||
694 | /** | |
695 | * do_remount_sb - asks filesystem to change mount options. | |
696 | * @sb: superblock in question | |
697 | * @flags: numeric part of options | |
698 | * @data: the rest of options | |
699 | * @force: whether or not to force the change | |
700 | * | |
701 | * Alters the mount options of a mounted file system. | |
702 | */ | |
703 | int do_remount_sb(struct super_block *sb, int flags, void *data, int force) | |
704 | { | |
705 | int retval; | |
706 | int remount_ro; | |
707 | ||
708 | if (sb->s_frozen != SB_UNFROZEN) | |
709 | return -EBUSY; | |
710 | ||
711 | #ifdef CONFIG_BLOCK | |
712 | if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev)) | |
713 | return -EACCES; | |
714 | #endif | |
715 | ||
716 | if (flags & MS_RDONLY) | |
717 | acct_auto_close(sb); | |
718 | shrink_dcache_sb(sb); | |
719 | sync_filesystem(sb); | |
720 | ||
721 | remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY); | |
722 | ||
723 | /* If we are remounting RDONLY and current sb is read/write, | |
724 | make sure there are no rw files opened */ | |
725 | if (remount_ro) { | |
726 | if (force) | |
727 | mark_files_ro(sb); | |
728 | else if (!fs_may_remount_ro(sb)) | |
729 | return -EBUSY; | |
730 | } | |
731 | ||
732 | if (sb->s_op->remount_fs) { | |
733 | retval = sb->s_op->remount_fs(sb, &flags, data); | |
734 | if (retval) { | |
735 | if (!force) | |
736 | return retval; | |
737 | /* If forced remount, go ahead despite any errors */ | |
738 | WARN(1, "forced remount of a %s fs returned %i\n", | |
739 | sb->s_type->name, retval); | |
740 | } | |
741 | } | |
742 | sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK); | |
743 | ||
744 | /* | |
745 | * Some filesystems modify their metadata via some other path than the | |
746 | * bdev buffer cache (eg. use a private mapping, or directories in | |
747 | * pagecache, etc). Also file data modifications go via their own | |
748 | * mappings. So If we try to mount readonly then copy the filesystem | |
749 | * from bdev, we could get stale data, so invalidate it to give a best | |
750 | * effort at coherency. | |
751 | */ | |
752 | if (remount_ro && sb->s_bdev) | |
753 | invalidate_bdev(sb->s_bdev); | |
754 | return 0; | |
755 | } | |
756 | ||
757 | static void do_emergency_remount(struct work_struct *work) | |
758 | { | |
759 | struct super_block *sb, *p = NULL; | |
760 | ||
761 | spin_lock(&sb_lock); | |
762 | list_for_each_entry(sb, &super_blocks, s_list) { | |
763 | if (hlist_unhashed(&sb->s_instances)) | |
764 | continue; | |
765 | sb->s_count++; | |
766 | spin_unlock(&sb_lock); | |
767 | down_write(&sb->s_umount); | |
768 | if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) && | |
769 | !(sb->s_flags & MS_RDONLY)) { | |
770 | /* | |
771 | * What lock protects sb->s_flags?? | |
772 | */ | |
773 | do_remount_sb(sb, MS_RDONLY, NULL, 1); | |
774 | } | |
775 | up_write(&sb->s_umount); | |
776 | spin_lock(&sb_lock); | |
777 | if (p) | |
778 | __put_super(p); | |
779 | p = sb; | |
780 | } | |
781 | if (p) | |
782 | __put_super(p); | |
783 | spin_unlock(&sb_lock); | |
784 | kfree(work); | |
785 | printk("Emergency Remount complete\n"); | |
786 | } | |
787 | ||
788 | void emergency_remount(void) | |
789 | { | |
790 | struct work_struct *work; | |
791 | ||
792 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
793 | if (work) { | |
794 | INIT_WORK(work, do_emergency_remount); | |
795 | schedule_work(work); | |
796 | } | |
797 | } | |
798 | ||
799 | /* | |
800 | * Unnamed block devices are dummy devices used by virtual | |
801 | * filesystems which don't use real block-devices. -- jrs | |
802 | */ | |
803 | ||
804 | static DEFINE_IDA(unnamed_dev_ida); | |
805 | static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */ | |
806 | static int unnamed_dev_start = 0; /* don't bother trying below it */ | |
807 | ||
808 | int get_anon_bdev(dev_t *p) | |
809 | { | |
810 | int dev; | |
811 | int error; | |
812 | ||
813 | retry: | |
814 | if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0) | |
815 | return -ENOMEM; | |
816 | spin_lock(&unnamed_dev_lock); | |
817 | error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev); | |
818 | if (!error) | |
819 | unnamed_dev_start = dev + 1; | |
820 | spin_unlock(&unnamed_dev_lock); | |
821 | if (error == -EAGAIN) | |
822 | /* We raced and lost with another CPU. */ | |
823 | goto retry; | |
824 | else if (error) | |
825 | return -EAGAIN; | |
826 | ||
827 | if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) { | |
828 | spin_lock(&unnamed_dev_lock); | |
829 | ida_remove(&unnamed_dev_ida, dev); | |
830 | if (unnamed_dev_start > dev) | |
831 | unnamed_dev_start = dev; | |
832 | spin_unlock(&unnamed_dev_lock); | |
833 | return -EMFILE; | |
834 | } | |
835 | *p = MKDEV(0, dev & MINORMASK); | |
836 | return 0; | |
837 | } | |
838 | EXPORT_SYMBOL(get_anon_bdev); | |
839 | ||
840 | void free_anon_bdev(dev_t dev) | |
841 | { | |
842 | int slot = MINOR(dev); | |
843 | spin_lock(&unnamed_dev_lock); | |
844 | ida_remove(&unnamed_dev_ida, slot); | |
845 | if (slot < unnamed_dev_start) | |
846 | unnamed_dev_start = slot; | |
847 | spin_unlock(&unnamed_dev_lock); | |
848 | } | |
849 | EXPORT_SYMBOL(free_anon_bdev); | |
850 | ||
851 | int set_anon_super(struct super_block *s, void *data) | |
852 | { | |
853 | int error = get_anon_bdev(&s->s_dev); | |
854 | if (!error) | |
855 | s->s_bdi = &noop_backing_dev_info; | |
856 | return error; | |
857 | } | |
858 | ||
859 | EXPORT_SYMBOL(set_anon_super); | |
860 | ||
861 | void kill_anon_super(struct super_block *sb) | |
862 | { | |
863 | dev_t dev = sb->s_dev; | |
864 | generic_shutdown_super(sb); | |
865 | free_anon_bdev(dev); | |
866 | } | |
867 | ||
868 | EXPORT_SYMBOL(kill_anon_super); | |
869 | ||
870 | void kill_litter_super(struct super_block *sb) | |
871 | { | |
872 | if (sb->s_root) | |
873 | d_genocide(sb->s_root); | |
874 | kill_anon_super(sb); | |
875 | } | |
876 | ||
877 | EXPORT_SYMBOL(kill_litter_super); | |
878 | ||
879 | static int ns_test_super(struct super_block *sb, void *data) | |
880 | { | |
881 | return sb->s_fs_info == data; | |
882 | } | |
883 | ||
884 | static int ns_set_super(struct super_block *sb, void *data) | |
885 | { | |
886 | sb->s_fs_info = data; | |
887 | return set_anon_super(sb, NULL); | |
888 | } | |
889 | ||
890 | struct dentry *mount_ns(struct file_system_type *fs_type, int flags, | |
891 | void *data, int (*fill_super)(struct super_block *, void *, int)) | |
892 | { | |
893 | struct super_block *sb; | |
894 | ||
895 | sb = sget(fs_type, ns_test_super, ns_set_super, data); | |
896 | if (IS_ERR(sb)) | |
897 | return ERR_CAST(sb); | |
898 | ||
899 | if (!sb->s_root) { | |
900 | int err; | |
901 | sb->s_flags = flags; | |
902 | err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0); | |
903 | if (err) { | |
904 | deactivate_locked_super(sb); | |
905 | return ERR_PTR(err); | |
906 | } | |
907 | ||
908 | sb->s_flags |= MS_ACTIVE; | |
909 | } | |
910 | ||
911 | return dget(sb->s_root); | |
912 | } | |
913 | ||
914 | EXPORT_SYMBOL(mount_ns); | |
915 | ||
916 | #ifdef CONFIG_BLOCK | |
917 | static int set_bdev_super(struct super_block *s, void *data) | |
918 | { | |
919 | s->s_bdev = data; | |
920 | s->s_dev = s->s_bdev->bd_dev; | |
921 | ||
922 | /* | |
923 | * We set the bdi here to the queue backing, file systems can | |
924 | * overwrite this in ->fill_super() | |
925 | */ | |
926 | s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info; | |
927 | return 0; | |
928 | } | |
929 | ||
930 | static int test_bdev_super(struct super_block *s, void *data) | |
931 | { | |
932 | return (void *)s->s_bdev == data; | |
933 | } | |
934 | ||
935 | struct dentry *mount_bdev(struct file_system_type *fs_type, | |
936 | int flags, const char *dev_name, void *data, | |
937 | int (*fill_super)(struct super_block *, void *, int)) | |
938 | { | |
939 | struct block_device *bdev; | |
940 | struct super_block *s; | |
941 | fmode_t mode = FMODE_READ | FMODE_EXCL; | |
942 | int error = 0; | |
943 | ||
944 | if (!(flags & MS_RDONLY)) | |
945 | mode |= FMODE_WRITE; | |
946 | ||
947 | bdev = blkdev_get_by_path(dev_name, mode, fs_type); | |
948 | if (IS_ERR(bdev)) | |
949 | return ERR_CAST(bdev); | |
950 | ||
951 | /* | |
952 | * once the super is inserted into the list by sget, s_umount | |
953 | * will protect the lockfs code from trying to start a snapshot | |
954 | * while we are mounting | |
955 | */ | |
956 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
957 | if (bdev->bd_fsfreeze_count > 0) { | |
958 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
959 | error = -EBUSY; | |
960 | goto error_bdev; | |
961 | } | |
962 | s = sget(fs_type, test_bdev_super, set_bdev_super, bdev); | |
963 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
964 | if (IS_ERR(s)) | |
965 | goto error_s; | |
966 | ||
967 | if (s->s_root) { | |
968 | if ((flags ^ s->s_flags) & MS_RDONLY) { | |
969 | deactivate_locked_super(s); | |
970 | error = -EBUSY; | |
971 | goto error_bdev; | |
972 | } | |
973 | ||
974 | /* | |
975 | * s_umount nests inside bd_mutex during | |
976 | * __invalidate_device(). blkdev_put() acquires | |
977 | * bd_mutex and can't be called under s_umount. Drop | |
978 | * s_umount temporarily. This is safe as we're | |
979 | * holding an active reference. | |
980 | */ | |
981 | up_write(&s->s_umount); | |
982 | blkdev_put(bdev, mode); | |
983 | down_write(&s->s_umount); | |
984 | } else { | |
985 | char b[BDEVNAME_SIZE]; | |
986 | ||
987 | s->s_flags = flags | MS_NOSEC; | |
988 | s->s_mode = mode; | |
989 | strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id)); | |
990 | sb_set_blocksize(s, block_size(bdev)); | |
991 | error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); | |
992 | if (error) { | |
993 | deactivate_locked_super(s); | |
994 | goto error; | |
995 | } | |
996 | ||
997 | s->s_flags |= MS_ACTIVE; | |
998 | bdev->bd_super = s; | |
999 | } | |
1000 | ||
1001 | return dget(s->s_root); | |
1002 | ||
1003 | error_s: | |
1004 | error = PTR_ERR(s); | |
1005 | error_bdev: | |
1006 | blkdev_put(bdev, mode); | |
1007 | error: | |
1008 | return ERR_PTR(error); | |
1009 | } | |
1010 | EXPORT_SYMBOL(mount_bdev); | |
1011 | ||
1012 | void kill_block_super(struct super_block *sb) | |
1013 | { | |
1014 | struct block_device *bdev = sb->s_bdev; | |
1015 | fmode_t mode = sb->s_mode; | |
1016 | ||
1017 | bdev->bd_super = NULL; | |
1018 | generic_shutdown_super(sb); | |
1019 | sync_blockdev(bdev); | |
1020 | WARN_ON_ONCE(!(mode & FMODE_EXCL)); | |
1021 | blkdev_put(bdev, mode | FMODE_EXCL); | |
1022 | } | |
1023 | ||
1024 | EXPORT_SYMBOL(kill_block_super); | |
1025 | #endif | |
1026 | ||
1027 | struct dentry *mount_nodev(struct file_system_type *fs_type, | |
1028 | int flags, void *data, | |
1029 | int (*fill_super)(struct super_block *, void *, int)) | |
1030 | { | |
1031 | int error; | |
1032 | struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL); | |
1033 | ||
1034 | if (IS_ERR(s)) | |
1035 | return ERR_CAST(s); | |
1036 | ||
1037 | s->s_flags = flags; | |
1038 | ||
1039 | error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); | |
1040 | if (error) { | |
1041 | deactivate_locked_super(s); | |
1042 | return ERR_PTR(error); | |
1043 | } | |
1044 | s->s_flags |= MS_ACTIVE; | |
1045 | return dget(s->s_root); | |
1046 | } | |
1047 | EXPORT_SYMBOL(mount_nodev); | |
1048 | ||
1049 | static int compare_single(struct super_block *s, void *p) | |
1050 | { | |
1051 | return 1; | |
1052 | } | |
1053 | ||
1054 | struct dentry *mount_single(struct file_system_type *fs_type, | |
1055 | int flags, void *data, | |
1056 | int (*fill_super)(struct super_block *, void *, int)) | |
1057 | { | |
1058 | struct super_block *s; | |
1059 | int error; | |
1060 | ||
1061 | s = sget(fs_type, compare_single, set_anon_super, NULL); | |
1062 | if (IS_ERR(s)) | |
1063 | return ERR_CAST(s); | |
1064 | if (!s->s_root) { | |
1065 | s->s_flags = flags; | |
1066 | error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); | |
1067 | if (error) { | |
1068 | deactivate_locked_super(s); | |
1069 | return ERR_PTR(error); | |
1070 | } | |
1071 | s->s_flags |= MS_ACTIVE; | |
1072 | } else { | |
1073 | do_remount_sb(s, flags, data, 0); | |
1074 | } | |
1075 | return dget(s->s_root); | |
1076 | } | |
1077 | EXPORT_SYMBOL(mount_single); | |
1078 | ||
1079 | struct dentry * | |
1080 | mount_fs(struct file_system_type *type, int flags, const char *name, void *data) | |
1081 | { | |
1082 | struct dentry *root; | |
1083 | struct super_block *sb; | |
1084 | char *secdata = NULL; | |
1085 | int error = -ENOMEM; | |
1086 | ||
1087 | if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) { | |
1088 | secdata = alloc_secdata(); | |
1089 | if (!secdata) | |
1090 | goto out; | |
1091 | ||
1092 | error = security_sb_copy_data(data, secdata); | |
1093 | if (error) | |
1094 | goto out_free_secdata; | |
1095 | } | |
1096 | ||
1097 | root = type->mount(type, flags, name, data); | |
1098 | if (IS_ERR(root)) { | |
1099 | error = PTR_ERR(root); | |
1100 | goto out_free_secdata; | |
1101 | } | |
1102 | sb = root->d_sb; | |
1103 | BUG_ON(!sb); | |
1104 | WARN_ON(!sb->s_bdi); | |
1105 | WARN_ON(sb->s_bdi == &default_backing_dev_info); | |
1106 | sb->s_flags |= MS_BORN; | |
1107 | ||
1108 | error = security_sb_kern_mount(sb, flags, secdata); | |
1109 | if (error) | |
1110 | goto out_sb; | |
1111 | ||
1112 | /* | |
1113 | * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE | |
1114 | * but s_maxbytes was an unsigned long long for many releases. Throw | |
1115 | * this warning for a little while to try and catch filesystems that | |
1116 | * violate this rule. | |
1117 | */ | |
1118 | WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to " | |
1119 | "negative value (%lld)\n", type->name, sb->s_maxbytes); | |
1120 | ||
1121 | up_write(&sb->s_umount); | |
1122 | free_secdata(secdata); | |
1123 | return root; | |
1124 | out_sb: | |
1125 | dput(root); | |
1126 | deactivate_locked_super(sb); | |
1127 | out_free_secdata: | |
1128 | free_secdata(secdata); | |
1129 | out: | |
1130 | return ERR_PTR(error); | |
1131 | } | |
1132 | ||
1133 | /** | |
1134 | * freeze_super - lock the filesystem and force it into a consistent state | |
1135 | * @sb: the super to lock | |
1136 | * | |
1137 | * Syncs the super to make sure the filesystem is consistent and calls the fs's | |
1138 | * freeze_fs. Subsequent calls to this without first thawing the fs will return | |
1139 | * -EBUSY. | |
1140 | */ | |
1141 | int freeze_super(struct super_block *sb) | |
1142 | { | |
1143 | int ret; | |
1144 | ||
1145 | atomic_inc(&sb->s_active); | |
1146 | down_write(&sb->s_umount); | |
1147 | if (sb->s_frozen) { | |
1148 | deactivate_locked_super(sb); | |
1149 | return -EBUSY; | |
1150 | } | |
1151 | ||
1152 | if (!(sb->s_flags & MS_BORN)) { | |
1153 | up_write(&sb->s_umount); | |
1154 | return 0; /* sic - it's "nothing to do" */ | |
1155 | } | |
1156 | ||
1157 | if (sb->s_flags & MS_RDONLY) { | |
1158 | sb->s_frozen = SB_FREEZE_TRANS; | |
1159 | smp_wmb(); | |
1160 | up_write(&sb->s_umount); | |
1161 | return 0; | |
1162 | } | |
1163 | ||
1164 | sb->s_frozen = SB_FREEZE_WRITE; | |
1165 | smp_wmb(); | |
1166 | ||
1167 | sync_filesystem(sb); | |
1168 | ||
1169 | sb->s_frozen = SB_FREEZE_TRANS; | |
1170 | smp_wmb(); | |
1171 | ||
1172 | sync_blockdev(sb->s_bdev); | |
1173 | if (sb->s_op->freeze_fs) { | |
1174 | ret = sb->s_op->freeze_fs(sb); | |
1175 | if (ret) { | |
1176 | printk(KERN_ERR | |
1177 | "VFS:Filesystem freeze failed\n"); | |
1178 | sb->s_frozen = SB_UNFROZEN; | |
1179 | deactivate_locked_super(sb); | |
1180 | return ret; | |
1181 | } | |
1182 | } | |
1183 | up_write(&sb->s_umount); | |
1184 | return 0; | |
1185 | } | |
1186 | EXPORT_SYMBOL(freeze_super); | |
1187 | ||
1188 | /** | |
1189 | * thaw_super -- unlock filesystem | |
1190 | * @sb: the super to thaw | |
1191 | * | |
1192 | * Unlocks the filesystem and marks it writeable again after freeze_super(). | |
1193 | */ | |
1194 | int thaw_super(struct super_block *sb) | |
1195 | { | |
1196 | int error; | |
1197 | ||
1198 | down_write(&sb->s_umount); | |
1199 | if (sb->s_frozen == SB_UNFROZEN) { | |
1200 | up_write(&sb->s_umount); | |
1201 | return -EINVAL; | |
1202 | } | |
1203 | ||
1204 | if (sb->s_flags & MS_RDONLY) | |
1205 | goto out; | |
1206 | ||
1207 | if (sb->s_op->unfreeze_fs) { | |
1208 | error = sb->s_op->unfreeze_fs(sb); | |
1209 | if (error) { | |
1210 | printk(KERN_ERR | |
1211 | "VFS:Filesystem thaw failed\n"); | |
1212 | sb->s_frozen = SB_FREEZE_TRANS; | |
1213 | up_write(&sb->s_umount); | |
1214 | return error; | |
1215 | } | |
1216 | } | |
1217 | ||
1218 | out: | |
1219 | sb->s_frozen = SB_UNFROZEN; | |
1220 | smp_wmb(); | |
1221 | wake_up(&sb->s_wait_unfrozen); | |
1222 | deactivate_locked_super(sb); | |
1223 | ||
1224 | return 0; | |
1225 | } | |
1226 | EXPORT_SYMBOL(thaw_super); |