]>
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 "internal.h" | |
34 | ||
35 | ||
36 | LIST_HEAD(super_blocks); | |
37 | DEFINE_SPINLOCK(sb_lock); | |
38 | ||
39 | /** | |
40 | * alloc_super - create new superblock | |
41 | * @type: filesystem type superblock should belong to | |
42 | * | |
43 | * Allocates and initializes a new &struct super_block. alloc_super() | |
44 | * returns a pointer new superblock or %NULL if allocation had failed. | |
45 | */ | |
46 | static struct super_block *alloc_super(struct file_system_type *type) | |
47 | { | |
48 | struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER); | |
49 | static const struct super_operations default_op; | |
50 | ||
51 | if (s) { | |
52 | if (security_sb_alloc(s)) { | |
53 | kfree(s); | |
54 | s = NULL; | |
55 | goto out; | |
56 | } | |
57 | #ifdef CONFIG_SMP | |
58 | s->s_files = alloc_percpu(struct list_head); | |
59 | if (!s->s_files) { | |
60 | security_sb_free(s); | |
61 | kfree(s); | |
62 | s = NULL; | |
63 | goto out; | |
64 | } else { | |
65 | int i; | |
66 | ||
67 | for_each_possible_cpu(i) | |
68 | INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i)); | |
69 | } | |
70 | #else | |
71 | INIT_LIST_HEAD(&s->s_files); | |
72 | #endif | |
73 | INIT_LIST_HEAD(&s->s_instances); | |
74 | INIT_HLIST_HEAD(&s->s_anon); | |
75 | INIT_LIST_HEAD(&s->s_inodes); | |
76 | INIT_LIST_HEAD(&s->s_dentry_lru); | |
77 | init_rwsem(&s->s_umount); | |
78 | mutex_init(&s->s_lock); | |
79 | lockdep_set_class(&s->s_umount, &type->s_umount_key); | |
80 | /* | |
81 | * The locking rules for s_lock are up to the | |
82 | * filesystem. For example ext3fs has different | |
83 | * lock ordering than usbfs: | |
84 | */ | |
85 | lockdep_set_class(&s->s_lock, &type->s_lock_key); | |
86 | /* | |
87 | * sget() can have s_umount recursion. | |
88 | * | |
89 | * When it cannot find a suitable sb, it allocates a new | |
90 | * one (this one), and tries again to find a suitable old | |
91 | * one. | |
92 | * | |
93 | * In case that succeeds, it will acquire the s_umount | |
94 | * lock of the old one. Since these are clearly distrinct | |
95 | * locks, and this object isn't exposed yet, there's no | |
96 | * risk of deadlocks. | |
97 | * | |
98 | * Annotate this by putting this lock in a different | |
99 | * subclass. | |
100 | */ | |
101 | down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING); | |
102 | s->s_count = 1; | |
103 | atomic_set(&s->s_active, 1); | |
104 | mutex_init(&s->s_vfs_rename_mutex); | |
105 | lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key); | |
106 | mutex_init(&s->s_dquot.dqio_mutex); | |
107 | mutex_init(&s->s_dquot.dqonoff_mutex); | |
108 | init_rwsem(&s->s_dquot.dqptr_sem); | |
109 | init_waitqueue_head(&s->s_wait_unfrozen); | |
110 | s->s_maxbytes = MAX_NON_LFS; | |
111 | s->s_op = &default_op; | |
112 | s->s_time_gran = 1000000000; | |
113 | } | |
114 | out: | |
115 | return s; | |
116 | } | |
117 | ||
118 | /** | |
119 | * destroy_super - frees a superblock | |
120 | * @s: superblock to free | |
121 | * | |
122 | * Frees a superblock. | |
123 | */ | |
124 | static inline void destroy_super(struct super_block *s) | |
125 | { | |
126 | #ifdef CONFIG_SMP | |
127 | free_percpu(s->s_files); | |
128 | #endif | |
129 | security_sb_free(s); | |
130 | kfree(s->s_subtype); | |
131 | kfree(s->s_options); | |
132 | kfree(s); | |
133 | } | |
134 | ||
135 | /* Superblock refcounting */ | |
136 | ||
137 | /* | |
138 | * Drop a superblock's refcount. The caller must hold sb_lock. | |
139 | */ | |
140 | void __put_super(struct super_block *sb) | |
141 | { | |
142 | if (!--sb->s_count) { | |
143 | list_del_init(&sb->s_list); | |
144 | destroy_super(sb); | |
145 | } | |
146 | } | |
147 | ||
148 | /** | |
149 | * put_super - drop a temporary reference to superblock | |
150 | * @sb: superblock in question | |
151 | * | |
152 | * Drops a temporary reference, frees superblock if there's no | |
153 | * references left. | |
154 | */ | |
155 | void put_super(struct super_block *sb) | |
156 | { | |
157 | spin_lock(&sb_lock); | |
158 | __put_super(sb); | |
159 | spin_unlock(&sb_lock); | |
160 | } | |
161 | ||
162 | ||
163 | /** | |
164 | * deactivate_locked_super - drop an active reference to superblock | |
165 | * @s: superblock to deactivate | |
166 | * | |
167 | * Drops an active reference to superblock, converting it into a temprory | |
168 | * one if there is no other active references left. In that case we | |
169 | * tell fs driver to shut it down and drop the temporary reference we | |
170 | * had just acquired. | |
171 | * | |
172 | * Caller holds exclusive lock on superblock; that lock is released. | |
173 | */ | |
174 | void deactivate_locked_super(struct super_block *s) | |
175 | { | |
176 | struct file_system_type *fs = s->s_type; | |
177 | if (atomic_dec_and_test(&s->s_active)) { | |
178 | fs->kill_sb(s); | |
179 | put_filesystem(fs); | |
180 | put_super(s); | |
181 | } else { | |
182 | up_write(&s->s_umount); | |
183 | } | |
184 | } | |
185 | ||
186 | EXPORT_SYMBOL(deactivate_locked_super); | |
187 | ||
188 | /** | |
189 | * deactivate_super - drop an active reference to superblock | |
190 | * @s: superblock to deactivate | |
191 | * | |
192 | * Variant of deactivate_locked_super(), except that superblock is *not* | |
193 | * locked by caller. If we are going to drop the final active reference, | |
194 | * lock will be acquired prior to that. | |
195 | */ | |
196 | void deactivate_super(struct super_block *s) | |
197 | { | |
198 | if (!atomic_add_unless(&s->s_active, -1, 1)) { | |
199 | down_write(&s->s_umount); | |
200 | deactivate_locked_super(s); | |
201 | } | |
202 | } | |
203 | ||
204 | EXPORT_SYMBOL(deactivate_super); | |
205 | ||
206 | /** | |
207 | * grab_super - acquire an active reference | |
208 | * @s: reference we are trying to make active | |
209 | * | |
210 | * Tries to acquire an active reference. grab_super() is used when we | |
211 | * had just found a superblock in super_blocks or fs_type->fs_supers | |
212 | * and want to turn it into a full-blown active reference. grab_super() | |
213 | * is called with sb_lock held and drops it. Returns 1 in case of | |
214 | * success, 0 if we had failed (superblock contents was already dead or | |
215 | * dying when grab_super() had been called). | |
216 | */ | |
217 | static int grab_super(struct super_block *s) __releases(sb_lock) | |
218 | { | |
219 | if (atomic_inc_not_zero(&s->s_active)) { | |
220 | spin_unlock(&sb_lock); | |
221 | return 1; | |
222 | } | |
223 | /* it's going away */ | |
224 | s->s_count++; | |
225 | spin_unlock(&sb_lock); | |
226 | /* wait for it to die */ | |
227 | down_write(&s->s_umount); | |
228 | up_write(&s->s_umount); | |
229 | put_super(s); | |
230 | return 0; | |
231 | } | |
232 | ||
233 | /* | |
234 | * Superblock locking. We really ought to get rid of these two. | |
235 | */ | |
236 | void lock_super(struct super_block * sb) | |
237 | { | |
238 | get_fs_excl(); | |
239 | mutex_lock(&sb->s_lock); | |
240 | } | |
241 | ||
242 | void unlock_super(struct super_block * sb) | |
243 | { | |
244 | put_fs_excl(); | |
245 | mutex_unlock(&sb->s_lock); | |
246 | } | |
247 | ||
248 | EXPORT_SYMBOL(lock_super); | |
249 | EXPORT_SYMBOL(unlock_super); | |
250 | ||
251 | /** | |
252 | * generic_shutdown_super - common helper for ->kill_sb() | |
253 | * @sb: superblock to kill | |
254 | * | |
255 | * generic_shutdown_super() does all fs-independent work on superblock | |
256 | * shutdown. Typical ->kill_sb() should pick all fs-specific objects | |
257 | * that need destruction out of superblock, call generic_shutdown_super() | |
258 | * and release aforementioned objects. Note: dentries and inodes _are_ | |
259 | * taken care of and do not need specific handling. | |
260 | * | |
261 | * Upon calling this function, the filesystem may no longer alter or | |
262 | * rearrange the set of dentries belonging to this super_block, nor may it | |
263 | * change the attachments of dentries to inodes. | |
264 | */ | |
265 | void generic_shutdown_super(struct super_block *sb) | |
266 | { | |
267 | const struct super_operations *sop = sb->s_op; | |
268 | ||
269 | ||
270 | if (sb->s_root) { | |
271 | shrink_dcache_for_umount(sb); | |
272 | sync_filesystem(sb); | |
273 | get_fs_excl(); | |
274 | sb->s_flags &= ~MS_ACTIVE; | |
275 | ||
276 | fsnotify_unmount_inodes(&sb->s_inodes); | |
277 | ||
278 | evict_inodes(sb); | |
279 | ||
280 | if (sop->put_super) | |
281 | sop->put_super(sb); | |
282 | ||
283 | if (!list_empty(&sb->s_inodes)) { | |
284 | printk("VFS: Busy inodes after unmount of %s. " | |
285 | "Self-destruct in 5 seconds. Have a nice day...\n", | |
286 | sb->s_id); | |
287 | } | |
288 | put_fs_excl(); | |
289 | } | |
290 | spin_lock(&sb_lock); | |
291 | /* should be initialized for __put_super_and_need_restart() */ | |
292 | list_del_init(&sb->s_instances); | |
293 | spin_unlock(&sb_lock); | |
294 | up_write(&sb->s_umount); | |
295 | } | |
296 | ||
297 | EXPORT_SYMBOL(generic_shutdown_super); | |
298 | ||
299 | /** | |
300 | * sget - find or create a superblock | |
301 | * @type: filesystem type superblock should belong to | |
302 | * @test: comparison callback | |
303 | * @set: setup callback | |
304 | * @data: argument to each of them | |
305 | */ | |
306 | struct super_block *sget(struct file_system_type *type, | |
307 | int (*test)(struct super_block *,void *), | |
308 | int (*set)(struct super_block *,void *), | |
309 | void *data) | |
310 | { | |
311 | struct super_block *s = NULL; | |
312 | struct super_block *old; | |
313 | int err; | |
314 | ||
315 | retry: | |
316 | spin_lock(&sb_lock); | |
317 | if (test) { | |
318 | list_for_each_entry(old, &type->fs_supers, s_instances) { | |
319 | if (!test(old, data)) | |
320 | continue; | |
321 | if (!grab_super(old)) | |
322 | goto retry; | |
323 | if (s) { | |
324 | up_write(&s->s_umount); | |
325 | destroy_super(s); | |
326 | s = NULL; | |
327 | } | |
328 | down_write(&old->s_umount); | |
329 | if (unlikely(!(old->s_flags & MS_BORN))) { | |
330 | deactivate_locked_super(old); | |
331 | goto retry; | |
332 | } | |
333 | return old; | |
334 | } | |
335 | } | |
336 | if (!s) { | |
337 | spin_unlock(&sb_lock); | |
338 | s = alloc_super(type); | |
339 | if (!s) | |
340 | return ERR_PTR(-ENOMEM); | |
341 | goto retry; | |
342 | } | |
343 | ||
344 | err = set(s, data); | |
345 | if (err) { | |
346 | spin_unlock(&sb_lock); | |
347 | up_write(&s->s_umount); | |
348 | destroy_super(s); | |
349 | return ERR_PTR(err); | |
350 | } | |
351 | s->s_type = type; | |
352 | strlcpy(s->s_id, type->name, sizeof(s->s_id)); | |
353 | list_add_tail(&s->s_list, &super_blocks); | |
354 | list_add(&s->s_instances, &type->fs_supers); | |
355 | spin_unlock(&sb_lock); | |
356 | get_filesystem(type); | |
357 | return s; | |
358 | } | |
359 | ||
360 | EXPORT_SYMBOL(sget); | |
361 | ||
362 | void drop_super(struct super_block *sb) | |
363 | { | |
364 | up_read(&sb->s_umount); | |
365 | put_super(sb); | |
366 | } | |
367 | ||
368 | EXPORT_SYMBOL(drop_super); | |
369 | ||
370 | /** | |
371 | * sync_supers - helper for periodic superblock writeback | |
372 | * | |
373 | * Call the write_super method if present on all dirty superblocks in | |
374 | * the system. This is for the periodic writeback used by most older | |
375 | * filesystems. For data integrity superblock writeback use | |
376 | * sync_filesystems() instead. | |
377 | * | |
378 | * Note: check the dirty flag before waiting, so we don't | |
379 | * hold up the sync while mounting a device. (The newly | |
380 | * mounted device won't need syncing.) | |
381 | */ | |
382 | void sync_supers(void) | |
383 | { | |
384 | struct super_block *sb, *p = NULL; | |
385 | ||
386 | spin_lock(&sb_lock); | |
387 | list_for_each_entry(sb, &super_blocks, s_list) { | |
388 | if (list_empty(&sb->s_instances)) | |
389 | continue; | |
390 | if (sb->s_op->write_super && sb->s_dirt) { | |
391 | sb->s_count++; | |
392 | spin_unlock(&sb_lock); | |
393 | ||
394 | down_read(&sb->s_umount); | |
395 | if (sb->s_root && sb->s_dirt) | |
396 | sb->s_op->write_super(sb); | |
397 | up_read(&sb->s_umount); | |
398 | ||
399 | spin_lock(&sb_lock); | |
400 | if (p) | |
401 | __put_super(p); | |
402 | p = sb; | |
403 | } | |
404 | } | |
405 | if (p) | |
406 | __put_super(p); | |
407 | spin_unlock(&sb_lock); | |
408 | } | |
409 | ||
410 | /** | |
411 | * iterate_supers - call function for all active superblocks | |
412 | * @f: function to call | |
413 | * @arg: argument to pass to it | |
414 | * | |
415 | * Scans the superblock list and calls given function, passing it | |
416 | * locked superblock and given argument. | |
417 | */ | |
418 | void iterate_supers(void (*f)(struct super_block *, void *), void *arg) | |
419 | { | |
420 | struct super_block *sb, *p = NULL; | |
421 | ||
422 | spin_lock(&sb_lock); | |
423 | list_for_each_entry(sb, &super_blocks, s_list) { | |
424 | if (list_empty(&sb->s_instances)) | |
425 | continue; | |
426 | sb->s_count++; | |
427 | spin_unlock(&sb_lock); | |
428 | ||
429 | down_read(&sb->s_umount); | |
430 | if (sb->s_root) | |
431 | f(sb, arg); | |
432 | up_read(&sb->s_umount); | |
433 | ||
434 | spin_lock(&sb_lock); | |
435 | if (p) | |
436 | __put_super(p); | |
437 | p = sb; | |
438 | } | |
439 | if (p) | |
440 | __put_super(p); | |
441 | spin_unlock(&sb_lock); | |
442 | } | |
443 | ||
444 | /** | |
445 | * get_super - get the superblock of a device | |
446 | * @bdev: device to get the superblock for | |
447 | * | |
448 | * Scans the superblock list and finds the superblock of the file system | |
449 | * mounted on the device given. %NULL is returned if no match is found. | |
450 | */ | |
451 | ||
452 | struct super_block *get_super(struct block_device *bdev) | |
453 | { | |
454 | struct super_block *sb; | |
455 | ||
456 | if (!bdev) | |
457 | return NULL; | |
458 | ||
459 | spin_lock(&sb_lock); | |
460 | rescan: | |
461 | list_for_each_entry(sb, &super_blocks, s_list) { | |
462 | if (list_empty(&sb->s_instances)) | |
463 | continue; | |
464 | if (sb->s_bdev == bdev) { | |
465 | sb->s_count++; | |
466 | spin_unlock(&sb_lock); | |
467 | down_read(&sb->s_umount); | |
468 | /* still alive? */ | |
469 | if (sb->s_root) | |
470 | return sb; | |
471 | up_read(&sb->s_umount); | |
472 | /* nope, got unmounted */ | |
473 | spin_lock(&sb_lock); | |
474 | __put_super(sb); | |
475 | goto rescan; | |
476 | } | |
477 | } | |
478 | spin_unlock(&sb_lock); | |
479 | return NULL; | |
480 | } | |
481 | ||
482 | EXPORT_SYMBOL(get_super); | |
483 | ||
484 | /** | |
485 | * get_active_super - get an active reference to the superblock of a device | |
486 | * @bdev: device to get the superblock for | |
487 | * | |
488 | * Scans the superblock list and finds the superblock of the file system | |
489 | * mounted on the device given. Returns the superblock with an active | |
490 | * reference or %NULL if none was found. | |
491 | */ | |
492 | struct super_block *get_active_super(struct block_device *bdev) | |
493 | { | |
494 | struct super_block *sb; | |
495 | ||
496 | if (!bdev) | |
497 | return NULL; | |
498 | ||
499 | restart: | |
500 | spin_lock(&sb_lock); | |
501 | list_for_each_entry(sb, &super_blocks, s_list) { | |
502 | if (list_empty(&sb->s_instances)) | |
503 | continue; | |
504 | if (sb->s_bdev == bdev) { | |
505 | if (grab_super(sb)) /* drops sb_lock */ | |
506 | return sb; | |
507 | else | |
508 | goto restart; | |
509 | } | |
510 | } | |
511 | spin_unlock(&sb_lock); | |
512 | return NULL; | |
513 | } | |
514 | ||
515 | struct super_block *user_get_super(dev_t dev) | |
516 | { | |
517 | struct super_block *sb; | |
518 | ||
519 | spin_lock(&sb_lock); | |
520 | rescan: | |
521 | list_for_each_entry(sb, &super_blocks, s_list) { | |
522 | if (list_empty(&sb->s_instances)) | |
523 | continue; | |
524 | if (sb->s_dev == dev) { | |
525 | sb->s_count++; | |
526 | spin_unlock(&sb_lock); | |
527 | down_read(&sb->s_umount); | |
528 | /* still alive? */ | |
529 | if (sb->s_root) | |
530 | return sb; | |
531 | up_read(&sb->s_umount); | |
532 | /* nope, got unmounted */ | |
533 | spin_lock(&sb_lock); | |
534 | __put_super(sb); | |
535 | goto rescan; | |
536 | } | |
537 | } | |
538 | spin_unlock(&sb_lock); | |
539 | return NULL; | |
540 | } | |
541 | ||
542 | /** | |
543 | * do_remount_sb - asks filesystem to change mount options. | |
544 | * @sb: superblock in question | |
545 | * @flags: numeric part of options | |
546 | * @data: the rest of options | |
547 | * @force: whether or not to force the change | |
548 | * | |
549 | * Alters the mount options of a mounted file system. | |
550 | */ | |
551 | int do_remount_sb(struct super_block *sb, int flags, void *data, int force) | |
552 | { | |
553 | int retval; | |
554 | int remount_ro; | |
555 | ||
556 | if (sb->s_frozen != SB_UNFROZEN) | |
557 | return -EBUSY; | |
558 | ||
559 | #ifdef CONFIG_BLOCK | |
560 | if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev)) | |
561 | return -EACCES; | |
562 | #endif | |
563 | ||
564 | if (flags & MS_RDONLY) | |
565 | acct_auto_close(sb); | |
566 | shrink_dcache_sb(sb); | |
567 | sync_filesystem(sb); | |
568 | ||
569 | remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY); | |
570 | ||
571 | /* If we are remounting RDONLY and current sb is read/write, | |
572 | make sure there are no rw files opened */ | |
573 | if (remount_ro) { | |
574 | if (force) | |
575 | mark_files_ro(sb); | |
576 | else if (!fs_may_remount_ro(sb)) | |
577 | return -EBUSY; | |
578 | } | |
579 | ||
580 | if (sb->s_op->remount_fs) { | |
581 | retval = sb->s_op->remount_fs(sb, &flags, data); | |
582 | if (retval) | |
583 | return retval; | |
584 | } | |
585 | sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK); | |
586 | ||
587 | /* | |
588 | * Some filesystems modify their metadata via some other path than the | |
589 | * bdev buffer cache (eg. use a private mapping, or directories in | |
590 | * pagecache, etc). Also file data modifications go via their own | |
591 | * mappings. So If we try to mount readonly then copy the filesystem | |
592 | * from bdev, we could get stale data, so invalidate it to give a best | |
593 | * effort at coherency. | |
594 | */ | |
595 | if (remount_ro && sb->s_bdev) | |
596 | invalidate_bdev(sb->s_bdev); | |
597 | return 0; | |
598 | } | |
599 | ||
600 | static void do_emergency_remount(struct work_struct *work) | |
601 | { | |
602 | struct super_block *sb, *p = NULL; | |
603 | ||
604 | spin_lock(&sb_lock); | |
605 | list_for_each_entry(sb, &super_blocks, s_list) { | |
606 | if (list_empty(&sb->s_instances)) | |
607 | continue; | |
608 | sb->s_count++; | |
609 | spin_unlock(&sb_lock); | |
610 | down_write(&sb->s_umount); | |
611 | if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) { | |
612 | /* | |
613 | * What lock protects sb->s_flags?? | |
614 | */ | |
615 | do_remount_sb(sb, MS_RDONLY, NULL, 1); | |
616 | } | |
617 | up_write(&sb->s_umount); | |
618 | spin_lock(&sb_lock); | |
619 | if (p) | |
620 | __put_super(p); | |
621 | p = sb; | |
622 | } | |
623 | if (p) | |
624 | __put_super(p); | |
625 | spin_unlock(&sb_lock); | |
626 | kfree(work); | |
627 | printk("Emergency Remount complete\n"); | |
628 | } | |
629 | ||
630 | void emergency_remount(void) | |
631 | { | |
632 | struct work_struct *work; | |
633 | ||
634 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
635 | if (work) { | |
636 | INIT_WORK(work, do_emergency_remount); | |
637 | schedule_work(work); | |
638 | } | |
639 | } | |
640 | ||
641 | /* | |
642 | * Unnamed block devices are dummy devices used by virtual | |
643 | * filesystems which don't use real block-devices. -- jrs | |
644 | */ | |
645 | ||
646 | static DEFINE_IDA(unnamed_dev_ida); | |
647 | static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */ | |
648 | static int unnamed_dev_start = 0; /* don't bother trying below it */ | |
649 | ||
650 | int set_anon_super(struct super_block *s, void *data) | |
651 | { | |
652 | int dev; | |
653 | int error; | |
654 | ||
655 | retry: | |
656 | if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0) | |
657 | return -ENOMEM; | |
658 | spin_lock(&unnamed_dev_lock); | |
659 | error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev); | |
660 | if (!error) | |
661 | unnamed_dev_start = dev + 1; | |
662 | spin_unlock(&unnamed_dev_lock); | |
663 | if (error == -EAGAIN) | |
664 | /* We raced and lost with another CPU. */ | |
665 | goto retry; | |
666 | else if (error) | |
667 | return -EAGAIN; | |
668 | ||
669 | if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) { | |
670 | spin_lock(&unnamed_dev_lock); | |
671 | ida_remove(&unnamed_dev_ida, dev); | |
672 | if (unnamed_dev_start > dev) | |
673 | unnamed_dev_start = dev; | |
674 | spin_unlock(&unnamed_dev_lock); | |
675 | return -EMFILE; | |
676 | } | |
677 | s->s_dev = MKDEV(0, dev & MINORMASK); | |
678 | s->s_bdi = &noop_backing_dev_info; | |
679 | return 0; | |
680 | } | |
681 | ||
682 | EXPORT_SYMBOL(set_anon_super); | |
683 | ||
684 | void kill_anon_super(struct super_block *sb) | |
685 | { | |
686 | int slot = MINOR(sb->s_dev); | |
687 | ||
688 | generic_shutdown_super(sb); | |
689 | spin_lock(&unnamed_dev_lock); | |
690 | ida_remove(&unnamed_dev_ida, slot); | |
691 | if (slot < unnamed_dev_start) | |
692 | unnamed_dev_start = slot; | |
693 | spin_unlock(&unnamed_dev_lock); | |
694 | } | |
695 | ||
696 | EXPORT_SYMBOL(kill_anon_super); | |
697 | ||
698 | void kill_litter_super(struct super_block *sb) | |
699 | { | |
700 | if (sb->s_root) | |
701 | d_genocide(sb->s_root); | |
702 | kill_anon_super(sb); | |
703 | } | |
704 | ||
705 | EXPORT_SYMBOL(kill_litter_super); | |
706 | ||
707 | static int ns_test_super(struct super_block *sb, void *data) | |
708 | { | |
709 | return sb->s_fs_info == data; | |
710 | } | |
711 | ||
712 | static int ns_set_super(struct super_block *sb, void *data) | |
713 | { | |
714 | sb->s_fs_info = data; | |
715 | return set_anon_super(sb, NULL); | |
716 | } | |
717 | ||
718 | struct dentry *mount_ns(struct file_system_type *fs_type, int flags, | |
719 | void *data, int (*fill_super)(struct super_block *, void *, int)) | |
720 | { | |
721 | struct super_block *sb; | |
722 | ||
723 | sb = sget(fs_type, ns_test_super, ns_set_super, data); | |
724 | if (IS_ERR(sb)) | |
725 | return ERR_CAST(sb); | |
726 | ||
727 | if (!sb->s_root) { | |
728 | int err; | |
729 | sb->s_flags = flags; | |
730 | err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0); | |
731 | if (err) { | |
732 | deactivate_locked_super(sb); | |
733 | return ERR_PTR(err); | |
734 | } | |
735 | ||
736 | sb->s_flags |= MS_ACTIVE; | |
737 | } | |
738 | ||
739 | return dget(sb->s_root); | |
740 | } | |
741 | ||
742 | EXPORT_SYMBOL(mount_ns); | |
743 | ||
744 | #ifdef CONFIG_BLOCK | |
745 | static int set_bdev_super(struct super_block *s, void *data) | |
746 | { | |
747 | s->s_bdev = data; | |
748 | s->s_dev = s->s_bdev->bd_dev; | |
749 | ||
750 | /* | |
751 | * We set the bdi here to the queue backing, file systems can | |
752 | * overwrite this in ->fill_super() | |
753 | */ | |
754 | s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info; | |
755 | return 0; | |
756 | } | |
757 | ||
758 | static int test_bdev_super(struct super_block *s, void *data) | |
759 | { | |
760 | return (void *)s->s_bdev == data; | |
761 | } | |
762 | ||
763 | struct dentry *mount_bdev(struct file_system_type *fs_type, | |
764 | int flags, const char *dev_name, void *data, | |
765 | int (*fill_super)(struct super_block *, void *, int)) | |
766 | { | |
767 | struct block_device *bdev; | |
768 | struct super_block *s; | |
769 | fmode_t mode = FMODE_READ; | |
770 | int error = 0; | |
771 | ||
772 | if (!(flags & MS_RDONLY)) | |
773 | mode |= FMODE_WRITE; | |
774 | ||
775 | bdev = open_bdev_exclusive(dev_name, mode, fs_type); | |
776 | if (IS_ERR(bdev)) | |
777 | return ERR_CAST(bdev); | |
778 | ||
779 | /* | |
780 | * once the super is inserted into the list by sget, s_umount | |
781 | * will protect the lockfs code from trying to start a snapshot | |
782 | * while we are mounting | |
783 | */ | |
784 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
785 | if (bdev->bd_fsfreeze_count > 0) { | |
786 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
787 | error = -EBUSY; | |
788 | goto error_bdev; | |
789 | } | |
790 | s = sget(fs_type, test_bdev_super, set_bdev_super, bdev); | |
791 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
792 | if (IS_ERR(s)) | |
793 | goto error_s; | |
794 | ||
795 | if (s->s_root) { | |
796 | if ((flags ^ s->s_flags) & MS_RDONLY) { | |
797 | deactivate_locked_super(s); | |
798 | error = -EBUSY; | |
799 | goto error_bdev; | |
800 | } | |
801 | ||
802 | /* | |
803 | * s_umount nests inside bd_mutex during | |
804 | * __invalidate_device(). close_bdev_exclusive() | |
805 | * acquires bd_mutex and can't be called under | |
806 | * s_umount. Drop s_umount temporarily. This is safe | |
807 | * as we're holding an active reference. | |
808 | */ | |
809 | up_write(&s->s_umount); | |
810 | close_bdev_exclusive(bdev, mode); | |
811 | down_write(&s->s_umount); | |
812 | } else { | |
813 | char b[BDEVNAME_SIZE]; | |
814 | ||
815 | s->s_flags = flags; | |
816 | s->s_mode = mode; | |
817 | strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id)); | |
818 | sb_set_blocksize(s, block_size(bdev)); | |
819 | error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); | |
820 | if (error) { | |
821 | deactivate_locked_super(s); | |
822 | goto error; | |
823 | } | |
824 | ||
825 | s->s_flags |= MS_ACTIVE; | |
826 | bdev->bd_super = s; | |
827 | } | |
828 | ||
829 | return dget(s->s_root); | |
830 | ||
831 | error_s: | |
832 | error = PTR_ERR(s); | |
833 | error_bdev: | |
834 | close_bdev_exclusive(bdev, mode); | |
835 | error: | |
836 | return ERR_PTR(error); | |
837 | } | |
838 | EXPORT_SYMBOL(mount_bdev); | |
839 | ||
840 | int get_sb_bdev(struct file_system_type *fs_type, | |
841 | int flags, const char *dev_name, void *data, | |
842 | int (*fill_super)(struct super_block *, void *, int), | |
843 | struct vfsmount *mnt) | |
844 | { | |
845 | struct dentry *root; | |
846 | ||
847 | root = mount_bdev(fs_type, flags, dev_name, data, fill_super); | |
848 | if (IS_ERR(root)) | |
849 | return PTR_ERR(root); | |
850 | mnt->mnt_root = root; | |
851 | mnt->mnt_sb = root->d_sb; | |
852 | return 0; | |
853 | } | |
854 | ||
855 | EXPORT_SYMBOL(get_sb_bdev); | |
856 | ||
857 | void kill_block_super(struct super_block *sb) | |
858 | { | |
859 | struct block_device *bdev = sb->s_bdev; | |
860 | fmode_t mode = sb->s_mode; | |
861 | ||
862 | bdev->bd_super = NULL; | |
863 | generic_shutdown_super(sb); | |
864 | sync_blockdev(bdev); | |
865 | close_bdev_exclusive(bdev, mode); | |
866 | } | |
867 | ||
868 | EXPORT_SYMBOL(kill_block_super); | |
869 | #endif | |
870 | ||
871 | struct dentry *mount_nodev(struct file_system_type *fs_type, | |
872 | int flags, void *data, | |
873 | int (*fill_super)(struct super_block *, void *, int)) | |
874 | { | |
875 | int error; | |
876 | struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL); | |
877 | ||
878 | if (IS_ERR(s)) | |
879 | return ERR_CAST(s); | |
880 | ||
881 | s->s_flags = flags; | |
882 | ||
883 | error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); | |
884 | if (error) { | |
885 | deactivate_locked_super(s); | |
886 | return ERR_PTR(error); | |
887 | } | |
888 | s->s_flags |= MS_ACTIVE; | |
889 | return dget(s->s_root); | |
890 | } | |
891 | EXPORT_SYMBOL(mount_nodev); | |
892 | ||
893 | int get_sb_nodev(struct file_system_type *fs_type, | |
894 | int flags, void *data, | |
895 | int (*fill_super)(struct super_block *, void *, int), | |
896 | struct vfsmount *mnt) | |
897 | { | |
898 | struct dentry *root; | |
899 | ||
900 | root = mount_nodev(fs_type, flags, data, fill_super); | |
901 | if (IS_ERR(root)) | |
902 | return PTR_ERR(root); | |
903 | mnt->mnt_root = root; | |
904 | mnt->mnt_sb = root->d_sb; | |
905 | return 0; | |
906 | } | |
907 | EXPORT_SYMBOL(get_sb_nodev); | |
908 | ||
909 | static int compare_single(struct super_block *s, void *p) | |
910 | { | |
911 | return 1; | |
912 | } | |
913 | ||
914 | struct dentry *mount_single(struct file_system_type *fs_type, | |
915 | int flags, void *data, | |
916 | int (*fill_super)(struct super_block *, void *, int)) | |
917 | { | |
918 | struct super_block *s; | |
919 | int error; | |
920 | ||
921 | s = sget(fs_type, compare_single, set_anon_super, NULL); | |
922 | if (IS_ERR(s)) | |
923 | return ERR_CAST(s); | |
924 | if (!s->s_root) { | |
925 | s->s_flags = flags; | |
926 | error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); | |
927 | if (error) { | |
928 | deactivate_locked_super(s); | |
929 | return ERR_PTR(error); | |
930 | } | |
931 | s->s_flags |= MS_ACTIVE; | |
932 | } else { | |
933 | do_remount_sb(s, flags, data, 0); | |
934 | } | |
935 | return dget(s->s_root); | |
936 | } | |
937 | EXPORT_SYMBOL(mount_single); | |
938 | ||
939 | int get_sb_single(struct file_system_type *fs_type, | |
940 | int flags, void *data, | |
941 | int (*fill_super)(struct super_block *, void *, int), | |
942 | struct vfsmount *mnt) | |
943 | { | |
944 | struct dentry *root; | |
945 | root = mount_single(fs_type, flags, data, fill_super); | |
946 | if (IS_ERR(root)) | |
947 | return PTR_ERR(root); | |
948 | mnt->mnt_root = root; | |
949 | mnt->mnt_sb = root->d_sb; | |
950 | return 0; | |
951 | } | |
952 | ||
953 | EXPORT_SYMBOL(get_sb_single); | |
954 | ||
955 | struct vfsmount * | |
956 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
957 | { | |
958 | struct vfsmount *mnt; | |
959 | struct dentry *root; | |
960 | char *secdata = NULL; | |
961 | int error; | |
962 | ||
963 | if (!type) | |
964 | return ERR_PTR(-ENODEV); | |
965 | ||
966 | error = -ENOMEM; | |
967 | mnt = alloc_vfsmnt(name); | |
968 | if (!mnt) | |
969 | goto out; | |
970 | ||
971 | if (flags & MS_KERNMOUNT) | |
972 | mnt->mnt_flags = MNT_INTERNAL; | |
973 | ||
974 | if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) { | |
975 | secdata = alloc_secdata(); | |
976 | if (!secdata) | |
977 | goto out_mnt; | |
978 | ||
979 | error = security_sb_copy_data(data, secdata); | |
980 | if (error) | |
981 | goto out_free_secdata; | |
982 | } | |
983 | ||
984 | if (type->mount) { | |
985 | root = type->mount(type, flags, name, data); | |
986 | if (IS_ERR(root)) { | |
987 | error = PTR_ERR(root); | |
988 | goto out_free_secdata; | |
989 | } | |
990 | mnt->mnt_root = root; | |
991 | mnt->mnt_sb = root->d_sb; | |
992 | } else { | |
993 | error = type->get_sb(type, flags, name, data, mnt); | |
994 | if (error < 0) | |
995 | goto out_free_secdata; | |
996 | } | |
997 | BUG_ON(!mnt->mnt_sb); | |
998 | WARN_ON(!mnt->mnt_sb->s_bdi); | |
999 | mnt->mnt_sb->s_flags |= MS_BORN; | |
1000 | ||
1001 | error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata); | |
1002 | if (error) | |
1003 | goto out_sb; | |
1004 | ||
1005 | /* | |
1006 | * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE | |
1007 | * but s_maxbytes was an unsigned long long for many releases. Throw | |
1008 | * this warning for a little while to try and catch filesystems that | |
1009 | * violate this rule. This warning should be either removed or | |
1010 | * converted to a BUG() in 2.6.34. | |
1011 | */ | |
1012 | WARN((mnt->mnt_sb->s_maxbytes < 0), "%s set sb->s_maxbytes to " | |
1013 | "negative value (%lld)\n", type->name, mnt->mnt_sb->s_maxbytes); | |
1014 | ||
1015 | mnt->mnt_mountpoint = mnt->mnt_root; | |
1016 | mnt->mnt_parent = mnt; | |
1017 | up_write(&mnt->mnt_sb->s_umount); | |
1018 | free_secdata(secdata); | |
1019 | return mnt; | |
1020 | out_sb: | |
1021 | dput(mnt->mnt_root); | |
1022 | deactivate_locked_super(mnt->mnt_sb); | |
1023 | out_free_secdata: | |
1024 | free_secdata(secdata); | |
1025 | out_mnt: | |
1026 | free_vfsmnt(mnt); | |
1027 | out: | |
1028 | return ERR_PTR(error); | |
1029 | } | |
1030 | ||
1031 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
1032 | ||
1033 | /** | |
1034 | * freeze_super - lock the filesystem and force it into a consistent state | |
1035 | * @sb: the super to lock | |
1036 | * | |
1037 | * Syncs the super to make sure the filesystem is consistent and calls the fs's | |
1038 | * freeze_fs. Subsequent calls to this without first thawing the fs will return | |
1039 | * -EBUSY. | |
1040 | */ | |
1041 | int freeze_super(struct super_block *sb) | |
1042 | { | |
1043 | int ret; | |
1044 | ||
1045 | atomic_inc(&sb->s_active); | |
1046 | down_write(&sb->s_umount); | |
1047 | if (sb->s_frozen) { | |
1048 | deactivate_locked_super(sb); | |
1049 | return -EBUSY; | |
1050 | } | |
1051 | ||
1052 | if (sb->s_flags & MS_RDONLY) { | |
1053 | sb->s_frozen = SB_FREEZE_TRANS; | |
1054 | smp_wmb(); | |
1055 | up_write(&sb->s_umount); | |
1056 | return 0; | |
1057 | } | |
1058 | ||
1059 | sb->s_frozen = SB_FREEZE_WRITE; | |
1060 | smp_wmb(); | |
1061 | ||
1062 | sync_filesystem(sb); | |
1063 | ||
1064 | sb->s_frozen = SB_FREEZE_TRANS; | |
1065 | smp_wmb(); | |
1066 | ||
1067 | sync_blockdev(sb->s_bdev); | |
1068 | if (sb->s_op->freeze_fs) { | |
1069 | ret = sb->s_op->freeze_fs(sb); | |
1070 | if (ret) { | |
1071 | printk(KERN_ERR | |
1072 | "VFS:Filesystem freeze failed\n"); | |
1073 | sb->s_frozen = SB_UNFROZEN; | |
1074 | deactivate_locked_super(sb); | |
1075 | return ret; | |
1076 | } | |
1077 | } | |
1078 | up_write(&sb->s_umount); | |
1079 | return 0; | |
1080 | } | |
1081 | EXPORT_SYMBOL(freeze_super); | |
1082 | ||
1083 | /** | |
1084 | * thaw_super -- unlock filesystem | |
1085 | * @sb: the super to thaw | |
1086 | * | |
1087 | * Unlocks the filesystem and marks it writeable again after freeze_super(). | |
1088 | */ | |
1089 | int thaw_super(struct super_block *sb) | |
1090 | { | |
1091 | int error; | |
1092 | ||
1093 | down_write(&sb->s_umount); | |
1094 | if (sb->s_frozen == SB_UNFROZEN) { | |
1095 | up_write(&sb->s_umount); | |
1096 | return -EINVAL; | |
1097 | } | |
1098 | ||
1099 | if (sb->s_flags & MS_RDONLY) | |
1100 | goto out; | |
1101 | ||
1102 | if (sb->s_op->unfreeze_fs) { | |
1103 | error = sb->s_op->unfreeze_fs(sb); | |
1104 | if (error) { | |
1105 | printk(KERN_ERR | |
1106 | "VFS:Filesystem thaw failed\n"); | |
1107 | sb->s_frozen = SB_FREEZE_TRANS; | |
1108 | up_write(&sb->s_umount); | |
1109 | return error; | |
1110 | } | |
1111 | } | |
1112 | ||
1113 | out: | |
1114 | sb->s_frozen = SB_UNFROZEN; | |
1115 | smp_wmb(); | |
1116 | wake_up(&sb->s_wait_unfrozen); | |
1117 | deactivate_locked_super(sb); | |
1118 | ||
1119 | return 0; | |
1120 | } | |
1121 | EXPORT_SYMBOL(thaw_super); | |
1122 | ||
1123 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) | |
1124 | { | |
1125 | int err; | |
1126 | const char *subtype = strchr(fstype, '.'); | |
1127 | if (subtype) { | |
1128 | subtype++; | |
1129 | err = -EINVAL; | |
1130 | if (!subtype[0]) | |
1131 | goto err; | |
1132 | } else | |
1133 | subtype = ""; | |
1134 | ||
1135 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
1136 | err = -ENOMEM; | |
1137 | if (!mnt->mnt_sb->s_subtype) | |
1138 | goto err; | |
1139 | return mnt; | |
1140 | ||
1141 | err: | |
1142 | mntput(mnt); | |
1143 | return ERR_PTR(err); | |
1144 | } | |
1145 | ||
1146 | struct vfsmount * | |
1147 | do_kern_mount(const char *fstype, int flags, const char *name, void *data) | |
1148 | { | |
1149 | struct file_system_type *type = get_fs_type(fstype); | |
1150 | struct vfsmount *mnt; | |
1151 | if (!type) | |
1152 | return ERR_PTR(-ENODEV); | |
1153 | mnt = vfs_kern_mount(type, flags, name, data); | |
1154 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
1155 | !mnt->mnt_sb->s_subtype) | |
1156 | mnt = fs_set_subtype(mnt, fstype); | |
1157 | put_filesystem(type); | |
1158 | return mnt; | |
1159 | } | |
1160 | EXPORT_SYMBOL_GPL(do_kern_mount); | |
1161 | ||
1162 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
1163 | { | |
1164 | return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
1165 | } | |
1166 | ||
1167 | EXPORT_SYMBOL_GPL(kern_mount_data); |