xref: /linux/fs/super.c (revision 5bf2b19320ec31d094d7370fdf536f7fd91fd799)
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 		INIT_LIST_HEAD(&s->s_files);
58 		INIT_LIST_HEAD(&s->s_instances);
59 		INIT_HLIST_HEAD(&s->s_anon);
60 		INIT_LIST_HEAD(&s->s_inodes);
61 		INIT_LIST_HEAD(&s->s_dentry_lru);
62 		init_rwsem(&s->s_umount);
63 		mutex_init(&s->s_lock);
64 		lockdep_set_class(&s->s_umount, &type->s_umount_key);
65 		/*
66 		 * The locking rules for s_lock are up to the
67 		 * filesystem. For example ext3fs has different
68 		 * lock ordering than usbfs:
69 		 */
70 		lockdep_set_class(&s->s_lock, &type->s_lock_key);
71 		/*
72 		 * sget() can have s_umount recursion.
73 		 *
74 		 * When it cannot find a suitable sb, it allocates a new
75 		 * one (this one), and tries again to find a suitable old
76 		 * one.
77 		 *
78 		 * In case that succeeds, it will acquire the s_umount
79 		 * lock of the old one. Since these are clearly distrinct
80 		 * locks, and this object isn't exposed yet, there's no
81 		 * risk of deadlocks.
82 		 *
83 		 * Annotate this by putting this lock in a different
84 		 * subclass.
85 		 */
86 		down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
87 		s->s_count = 1;
88 		atomic_set(&s->s_active, 1);
89 		mutex_init(&s->s_vfs_rename_mutex);
90 		lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
91 		mutex_init(&s->s_dquot.dqio_mutex);
92 		mutex_init(&s->s_dquot.dqonoff_mutex);
93 		init_rwsem(&s->s_dquot.dqptr_sem);
94 		init_waitqueue_head(&s->s_wait_unfrozen);
95 		s->s_maxbytes = MAX_NON_LFS;
96 		s->s_op = &default_op;
97 		s->s_time_gran = 1000000000;
98 	}
99 out:
100 	return s;
101 }
102 
103 /**
104  *	destroy_super	-	frees a superblock
105  *	@s: superblock to free
106  *
107  *	Frees a superblock.
108  */
109 static inline void destroy_super(struct super_block *s)
110 {
111 	security_sb_free(s);
112 	kfree(s->s_subtype);
113 	kfree(s->s_options);
114 	kfree(s);
115 }
116 
117 /* Superblock refcounting  */
118 
119 /*
120  * Drop a superblock's refcount.  The caller must hold sb_lock.
121  */
122 void __put_super(struct super_block *sb)
123 {
124 	if (!--sb->s_count) {
125 		list_del_init(&sb->s_list);
126 		destroy_super(sb);
127 	}
128 }
129 
130 /**
131  *	put_super	-	drop a temporary reference to superblock
132  *	@sb: superblock in question
133  *
134  *	Drops a temporary reference, frees superblock if there's no
135  *	references left.
136  */
137 void put_super(struct super_block *sb)
138 {
139 	spin_lock(&sb_lock);
140 	__put_super(sb);
141 	spin_unlock(&sb_lock);
142 }
143 
144 
145 /**
146  *	deactivate_locked_super	-	drop an active reference to superblock
147  *	@s: superblock to deactivate
148  *
149  *	Drops an active reference to superblock, converting it into a temprory
150  *	one if there is no other active references left.  In that case we
151  *	tell fs driver to shut it down and drop the temporary reference we
152  *	had just acquired.
153  *
154  *	Caller holds exclusive lock on superblock; that lock is released.
155  */
156 void deactivate_locked_super(struct super_block *s)
157 {
158 	struct file_system_type *fs = s->s_type;
159 	if (atomic_dec_and_test(&s->s_active)) {
160 		fs->kill_sb(s);
161 		put_filesystem(fs);
162 		put_super(s);
163 	} else {
164 		up_write(&s->s_umount);
165 	}
166 }
167 
168 EXPORT_SYMBOL(deactivate_locked_super);
169 
170 /**
171  *	deactivate_super	-	drop an active reference to superblock
172  *	@s: superblock to deactivate
173  *
174  *	Variant of deactivate_locked_super(), except that superblock is *not*
175  *	locked by caller.  If we are going to drop the final active reference,
176  *	lock will be acquired prior to that.
177  */
178 void deactivate_super(struct super_block *s)
179 {
180         if (!atomic_add_unless(&s->s_active, -1, 1)) {
181 		down_write(&s->s_umount);
182 		deactivate_locked_super(s);
183 	}
184 }
185 
186 EXPORT_SYMBOL(deactivate_super);
187 
188 /**
189  *	grab_super - acquire an active reference
190  *	@s: reference we are trying to make active
191  *
192  *	Tries to acquire an active reference.  grab_super() is used when we
193  * 	had just found a superblock in super_blocks or fs_type->fs_supers
194  *	and want to turn it into a full-blown active reference.  grab_super()
195  *	is called with sb_lock held and drops it.  Returns 1 in case of
196  *	success, 0 if we had failed (superblock contents was already dead or
197  *	dying when grab_super() had been called).
198  */
199 static int grab_super(struct super_block *s) __releases(sb_lock)
200 {
201 	if (atomic_inc_not_zero(&s->s_active)) {
202 		spin_unlock(&sb_lock);
203 		return 1;
204 	}
205 	/* it's going away */
206 	s->s_count++;
207 	spin_unlock(&sb_lock);
208 	/* wait for it to die */
209 	down_write(&s->s_umount);
210 	up_write(&s->s_umount);
211 	put_super(s);
212 	return 0;
213 }
214 
215 /*
216  * Superblock locking.  We really ought to get rid of these two.
217  */
218 void lock_super(struct super_block * sb)
219 {
220 	get_fs_excl();
221 	mutex_lock(&sb->s_lock);
222 }
223 
224 void unlock_super(struct super_block * sb)
225 {
226 	put_fs_excl();
227 	mutex_unlock(&sb->s_lock);
228 }
229 
230 EXPORT_SYMBOL(lock_super);
231 EXPORT_SYMBOL(unlock_super);
232 
233 /**
234  *	generic_shutdown_super	-	common helper for ->kill_sb()
235  *	@sb: superblock to kill
236  *
237  *	generic_shutdown_super() does all fs-independent work on superblock
238  *	shutdown.  Typical ->kill_sb() should pick all fs-specific objects
239  *	that need destruction out of superblock, call generic_shutdown_super()
240  *	and release aforementioned objects.  Note: dentries and inodes _are_
241  *	taken care of and do not need specific handling.
242  *
243  *	Upon calling this function, the filesystem may no longer alter or
244  *	rearrange the set of dentries belonging to this super_block, nor may it
245  *	change the attachments of dentries to inodes.
246  */
247 void generic_shutdown_super(struct super_block *sb)
248 {
249 	const struct super_operations *sop = sb->s_op;
250 
251 
252 	if (sb->s_root) {
253 		shrink_dcache_for_umount(sb);
254 		sync_filesystem(sb);
255 		get_fs_excl();
256 		sb->s_flags &= ~MS_ACTIVE;
257 
258 		/* bad name - it should be evict_inodes() */
259 		invalidate_inodes(sb);
260 
261 		if (sop->put_super)
262 			sop->put_super(sb);
263 
264 		/* Forget any remaining inodes */
265 		if (invalidate_inodes(sb)) {
266 			printk("VFS: Busy inodes after unmount of %s. "
267 			   "Self-destruct in 5 seconds.  Have a nice day...\n",
268 			   sb->s_id);
269 		}
270 		put_fs_excl();
271 	}
272 	spin_lock(&sb_lock);
273 	/* should be initialized for __put_super_and_need_restart() */
274 	list_del_init(&sb->s_instances);
275 	spin_unlock(&sb_lock);
276 	up_write(&sb->s_umount);
277 }
278 
279 EXPORT_SYMBOL(generic_shutdown_super);
280 
281 /**
282  *	sget	-	find or create a superblock
283  *	@type:	filesystem type superblock should belong to
284  *	@test:	comparison callback
285  *	@set:	setup callback
286  *	@data:	argument to each of them
287  */
288 struct super_block *sget(struct file_system_type *type,
289 			int (*test)(struct super_block *,void *),
290 			int (*set)(struct super_block *,void *),
291 			void *data)
292 {
293 	struct super_block *s = NULL;
294 	struct super_block *old;
295 	int err;
296 
297 retry:
298 	spin_lock(&sb_lock);
299 	if (test) {
300 		list_for_each_entry(old, &type->fs_supers, s_instances) {
301 			if (!test(old, data))
302 				continue;
303 			if (!grab_super(old))
304 				goto retry;
305 			if (s) {
306 				up_write(&s->s_umount);
307 				destroy_super(s);
308 				s = NULL;
309 			}
310 			down_write(&old->s_umount);
311 			if (unlikely(!(old->s_flags & MS_BORN))) {
312 				deactivate_locked_super(old);
313 				goto retry;
314 			}
315 			return old;
316 		}
317 	}
318 	if (!s) {
319 		spin_unlock(&sb_lock);
320 		s = alloc_super(type);
321 		if (!s)
322 			return ERR_PTR(-ENOMEM);
323 		goto retry;
324 	}
325 
326 	err = set(s, data);
327 	if (err) {
328 		spin_unlock(&sb_lock);
329 		up_write(&s->s_umount);
330 		destroy_super(s);
331 		return ERR_PTR(err);
332 	}
333 	s->s_type = type;
334 	strlcpy(s->s_id, type->name, sizeof(s->s_id));
335 	list_add_tail(&s->s_list, &super_blocks);
336 	list_add(&s->s_instances, &type->fs_supers);
337 	spin_unlock(&sb_lock);
338 	get_filesystem(type);
339 	return s;
340 }
341 
342 EXPORT_SYMBOL(sget);
343 
344 void drop_super(struct super_block *sb)
345 {
346 	up_read(&sb->s_umount);
347 	put_super(sb);
348 }
349 
350 EXPORT_SYMBOL(drop_super);
351 
352 /**
353  * sync_supers - helper for periodic superblock writeback
354  *
355  * Call the write_super method if present on all dirty superblocks in
356  * the system.  This is for the periodic writeback used by most older
357  * filesystems.  For data integrity superblock writeback use
358  * sync_filesystems() instead.
359  *
360  * Note: check the dirty flag before waiting, so we don't
361  * hold up the sync while mounting a device. (The newly
362  * mounted device won't need syncing.)
363  */
364 void sync_supers(void)
365 {
366 	struct super_block *sb, *p = NULL;
367 
368 	spin_lock(&sb_lock);
369 	list_for_each_entry(sb, &super_blocks, s_list) {
370 		if (list_empty(&sb->s_instances))
371 			continue;
372 		if (sb->s_op->write_super && sb->s_dirt) {
373 			sb->s_count++;
374 			spin_unlock(&sb_lock);
375 
376 			down_read(&sb->s_umount);
377 			if (sb->s_root && sb->s_dirt)
378 				sb->s_op->write_super(sb);
379 			up_read(&sb->s_umount);
380 
381 			spin_lock(&sb_lock);
382 			if (p)
383 				__put_super(p);
384 			p = sb;
385 		}
386 	}
387 	if (p)
388 		__put_super(p);
389 	spin_unlock(&sb_lock);
390 }
391 
392 /**
393  *	iterate_supers - call function for all active superblocks
394  *	@f: function to call
395  *	@arg: argument to pass to it
396  *
397  *	Scans the superblock list and calls given function, passing it
398  *	locked superblock and given argument.
399  */
400 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
401 {
402 	struct super_block *sb, *p = NULL;
403 
404 	spin_lock(&sb_lock);
405 	list_for_each_entry(sb, &super_blocks, s_list) {
406 		if (list_empty(&sb->s_instances))
407 			continue;
408 		sb->s_count++;
409 		spin_unlock(&sb_lock);
410 
411 		down_read(&sb->s_umount);
412 		if (sb->s_root)
413 			f(sb, arg);
414 		up_read(&sb->s_umount);
415 
416 		spin_lock(&sb_lock);
417 		if (p)
418 			__put_super(p);
419 		p = sb;
420 	}
421 	if (p)
422 		__put_super(p);
423 	spin_unlock(&sb_lock);
424 }
425 
426 /**
427  *	get_super - get the superblock of a device
428  *	@bdev: device to get the superblock for
429  *
430  *	Scans the superblock list and finds the superblock of the file system
431  *	mounted on the device given. %NULL is returned if no match is found.
432  */
433 
434 struct super_block *get_super(struct block_device *bdev)
435 {
436 	struct super_block *sb;
437 
438 	if (!bdev)
439 		return NULL;
440 
441 	spin_lock(&sb_lock);
442 rescan:
443 	list_for_each_entry(sb, &super_blocks, s_list) {
444 		if (list_empty(&sb->s_instances))
445 			continue;
446 		if (sb->s_bdev == bdev) {
447 			sb->s_count++;
448 			spin_unlock(&sb_lock);
449 			down_read(&sb->s_umount);
450 			/* still alive? */
451 			if (sb->s_root)
452 				return sb;
453 			up_read(&sb->s_umount);
454 			/* nope, got unmounted */
455 			spin_lock(&sb_lock);
456 			__put_super(sb);
457 			goto rescan;
458 		}
459 	}
460 	spin_unlock(&sb_lock);
461 	return NULL;
462 }
463 
464 EXPORT_SYMBOL(get_super);
465 
466 /**
467  * get_active_super - get an active reference to the superblock of a device
468  * @bdev: device to get the superblock for
469  *
470  * Scans the superblock list and finds the superblock of the file system
471  * mounted on the device given.  Returns the superblock with an active
472  * reference or %NULL if none was found.
473  */
474 struct super_block *get_active_super(struct block_device *bdev)
475 {
476 	struct super_block *sb;
477 
478 	if (!bdev)
479 		return NULL;
480 
481 restart:
482 	spin_lock(&sb_lock);
483 	list_for_each_entry(sb, &super_blocks, s_list) {
484 		if (list_empty(&sb->s_instances))
485 			continue;
486 		if (sb->s_bdev == bdev) {
487 			if (grab_super(sb)) /* drops sb_lock */
488 				return sb;
489 			else
490 				goto restart;
491 		}
492 	}
493 	spin_unlock(&sb_lock);
494 	return NULL;
495 }
496 
497 struct super_block *user_get_super(dev_t dev)
498 {
499 	struct super_block *sb;
500 
501 	spin_lock(&sb_lock);
502 rescan:
503 	list_for_each_entry(sb, &super_blocks, s_list) {
504 		if (list_empty(&sb->s_instances))
505 			continue;
506 		if (sb->s_dev ==  dev) {
507 			sb->s_count++;
508 			spin_unlock(&sb_lock);
509 			down_read(&sb->s_umount);
510 			/* still alive? */
511 			if (sb->s_root)
512 				return sb;
513 			up_read(&sb->s_umount);
514 			/* nope, got unmounted */
515 			spin_lock(&sb_lock);
516 			__put_super(sb);
517 			goto rescan;
518 		}
519 	}
520 	spin_unlock(&sb_lock);
521 	return NULL;
522 }
523 
524 /**
525  *	do_remount_sb - asks filesystem to change mount options.
526  *	@sb:	superblock in question
527  *	@flags:	numeric part of options
528  *	@data:	the rest of options
529  *      @force: whether or not to force the change
530  *
531  *	Alters the mount options of a mounted file system.
532  */
533 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
534 {
535 	int retval;
536 	int remount_ro;
537 
538 	if (sb->s_frozen != SB_UNFROZEN)
539 		return -EBUSY;
540 
541 #ifdef CONFIG_BLOCK
542 	if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
543 		return -EACCES;
544 #endif
545 
546 	if (flags & MS_RDONLY)
547 		acct_auto_close(sb);
548 	shrink_dcache_sb(sb);
549 	sync_filesystem(sb);
550 
551 	remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
552 
553 	/* If we are remounting RDONLY and current sb is read/write,
554 	   make sure there are no rw files opened */
555 	if (remount_ro) {
556 		if (force)
557 			mark_files_ro(sb);
558 		else if (!fs_may_remount_ro(sb))
559 			return -EBUSY;
560 	}
561 
562 	if (sb->s_op->remount_fs) {
563 		retval = sb->s_op->remount_fs(sb, &flags, data);
564 		if (retval)
565 			return retval;
566 	}
567 	sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
568 
569 	/*
570 	 * Some filesystems modify their metadata via some other path than the
571 	 * bdev buffer cache (eg. use a private mapping, or directories in
572 	 * pagecache, etc). Also file data modifications go via their own
573 	 * mappings. So If we try to mount readonly then copy the filesystem
574 	 * from bdev, we could get stale data, so invalidate it to give a best
575 	 * effort at coherency.
576 	 */
577 	if (remount_ro && sb->s_bdev)
578 		invalidate_bdev(sb->s_bdev);
579 	return 0;
580 }
581 
582 static void do_emergency_remount(struct work_struct *work)
583 {
584 	struct super_block *sb, *p = NULL;
585 
586 	spin_lock(&sb_lock);
587 	list_for_each_entry(sb, &super_blocks, s_list) {
588 		if (list_empty(&sb->s_instances))
589 			continue;
590 		sb->s_count++;
591 		spin_unlock(&sb_lock);
592 		down_write(&sb->s_umount);
593 		if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
594 			/*
595 			 * What lock protects sb->s_flags??
596 			 */
597 			do_remount_sb(sb, MS_RDONLY, NULL, 1);
598 		}
599 		up_write(&sb->s_umount);
600 		spin_lock(&sb_lock);
601 		if (p)
602 			__put_super(p);
603 		p = sb;
604 	}
605 	if (p)
606 		__put_super(p);
607 	spin_unlock(&sb_lock);
608 	kfree(work);
609 	printk("Emergency Remount complete\n");
610 }
611 
612 void emergency_remount(void)
613 {
614 	struct work_struct *work;
615 
616 	work = kmalloc(sizeof(*work), GFP_ATOMIC);
617 	if (work) {
618 		INIT_WORK(work, do_emergency_remount);
619 		schedule_work(work);
620 	}
621 }
622 
623 /*
624  * Unnamed block devices are dummy devices used by virtual
625  * filesystems which don't use real block-devices.  -- jrs
626  */
627 
628 static DEFINE_IDA(unnamed_dev_ida);
629 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
630 static int unnamed_dev_start = 0; /* don't bother trying below it */
631 
632 int set_anon_super(struct super_block *s, void *data)
633 {
634 	int dev;
635 	int error;
636 
637  retry:
638 	if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
639 		return -ENOMEM;
640 	spin_lock(&unnamed_dev_lock);
641 	error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
642 	if (!error)
643 		unnamed_dev_start = dev + 1;
644 	spin_unlock(&unnamed_dev_lock);
645 	if (error == -EAGAIN)
646 		/* We raced and lost with another CPU. */
647 		goto retry;
648 	else if (error)
649 		return -EAGAIN;
650 
651 	if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
652 		spin_lock(&unnamed_dev_lock);
653 		ida_remove(&unnamed_dev_ida, dev);
654 		if (unnamed_dev_start > dev)
655 			unnamed_dev_start = dev;
656 		spin_unlock(&unnamed_dev_lock);
657 		return -EMFILE;
658 	}
659 	s->s_dev = MKDEV(0, dev & MINORMASK);
660 	s->s_bdi = &noop_backing_dev_info;
661 	return 0;
662 }
663 
664 EXPORT_SYMBOL(set_anon_super);
665 
666 void kill_anon_super(struct super_block *sb)
667 {
668 	int slot = MINOR(sb->s_dev);
669 
670 	generic_shutdown_super(sb);
671 	spin_lock(&unnamed_dev_lock);
672 	ida_remove(&unnamed_dev_ida, slot);
673 	if (slot < unnamed_dev_start)
674 		unnamed_dev_start = slot;
675 	spin_unlock(&unnamed_dev_lock);
676 }
677 
678 EXPORT_SYMBOL(kill_anon_super);
679 
680 void kill_litter_super(struct super_block *sb)
681 {
682 	if (sb->s_root)
683 		d_genocide(sb->s_root);
684 	kill_anon_super(sb);
685 }
686 
687 EXPORT_SYMBOL(kill_litter_super);
688 
689 static int ns_test_super(struct super_block *sb, void *data)
690 {
691 	return sb->s_fs_info == data;
692 }
693 
694 static int ns_set_super(struct super_block *sb, void *data)
695 {
696 	sb->s_fs_info = data;
697 	return set_anon_super(sb, NULL);
698 }
699 
700 int get_sb_ns(struct file_system_type *fs_type, int flags, void *data,
701 	int (*fill_super)(struct super_block *, void *, int),
702 	struct vfsmount *mnt)
703 {
704 	struct super_block *sb;
705 
706 	sb = sget(fs_type, ns_test_super, ns_set_super, data);
707 	if (IS_ERR(sb))
708 		return PTR_ERR(sb);
709 
710 	if (!sb->s_root) {
711 		int err;
712 		sb->s_flags = flags;
713 		err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
714 		if (err) {
715 			deactivate_locked_super(sb);
716 			return err;
717 		}
718 
719 		sb->s_flags |= MS_ACTIVE;
720 	}
721 
722 	simple_set_mnt(mnt, sb);
723 	return 0;
724 }
725 
726 EXPORT_SYMBOL(get_sb_ns);
727 
728 #ifdef CONFIG_BLOCK
729 static int set_bdev_super(struct super_block *s, void *data)
730 {
731 	s->s_bdev = data;
732 	s->s_dev = s->s_bdev->bd_dev;
733 
734 	/*
735 	 * We set the bdi here to the queue backing, file systems can
736 	 * overwrite this in ->fill_super()
737 	 */
738 	s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
739 	return 0;
740 }
741 
742 static int test_bdev_super(struct super_block *s, void *data)
743 {
744 	return (void *)s->s_bdev == data;
745 }
746 
747 int get_sb_bdev(struct file_system_type *fs_type,
748 	int flags, const char *dev_name, void *data,
749 	int (*fill_super)(struct super_block *, void *, int),
750 	struct vfsmount *mnt)
751 {
752 	struct block_device *bdev;
753 	struct super_block *s;
754 	fmode_t mode = FMODE_READ;
755 	int error = 0;
756 
757 	if (!(flags & MS_RDONLY))
758 		mode |= FMODE_WRITE;
759 
760 	bdev = open_bdev_exclusive(dev_name, mode, fs_type);
761 	if (IS_ERR(bdev))
762 		return PTR_ERR(bdev);
763 
764 	/*
765 	 * once the super is inserted into the list by sget, s_umount
766 	 * will protect the lockfs code from trying to start a snapshot
767 	 * while we are mounting
768 	 */
769 	mutex_lock(&bdev->bd_fsfreeze_mutex);
770 	if (bdev->bd_fsfreeze_count > 0) {
771 		mutex_unlock(&bdev->bd_fsfreeze_mutex);
772 		error = -EBUSY;
773 		goto error_bdev;
774 	}
775 	s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
776 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
777 	if (IS_ERR(s))
778 		goto error_s;
779 
780 	if (s->s_root) {
781 		if ((flags ^ s->s_flags) & MS_RDONLY) {
782 			deactivate_locked_super(s);
783 			error = -EBUSY;
784 			goto error_bdev;
785 		}
786 
787 		/*
788 		 * s_umount nests inside bd_mutex during
789 		 * __invalidate_device().  close_bdev_exclusive()
790 		 * acquires bd_mutex and can't be called under
791 		 * s_umount.  Drop s_umount temporarily.  This is safe
792 		 * as we're holding an active reference.
793 		 */
794 		up_write(&s->s_umount);
795 		close_bdev_exclusive(bdev, mode);
796 		down_write(&s->s_umount);
797 	} else {
798 		char b[BDEVNAME_SIZE];
799 
800 		s->s_flags = flags;
801 		s->s_mode = mode;
802 		strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
803 		sb_set_blocksize(s, block_size(bdev));
804 		error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
805 		if (error) {
806 			deactivate_locked_super(s);
807 			goto error;
808 		}
809 
810 		s->s_flags |= MS_ACTIVE;
811 		bdev->bd_super = s;
812 	}
813 
814 	simple_set_mnt(mnt, s);
815 	return 0;
816 
817 error_s:
818 	error = PTR_ERR(s);
819 error_bdev:
820 	close_bdev_exclusive(bdev, mode);
821 error:
822 	return error;
823 }
824 
825 EXPORT_SYMBOL(get_sb_bdev);
826 
827 void kill_block_super(struct super_block *sb)
828 {
829 	struct block_device *bdev = sb->s_bdev;
830 	fmode_t mode = sb->s_mode;
831 
832 	bdev->bd_super = NULL;
833 	generic_shutdown_super(sb);
834 	sync_blockdev(bdev);
835 	close_bdev_exclusive(bdev, mode);
836 }
837 
838 EXPORT_SYMBOL(kill_block_super);
839 #endif
840 
841 int get_sb_nodev(struct file_system_type *fs_type,
842 	int flags, void *data,
843 	int (*fill_super)(struct super_block *, void *, int),
844 	struct vfsmount *mnt)
845 {
846 	int error;
847 	struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
848 
849 	if (IS_ERR(s))
850 		return PTR_ERR(s);
851 
852 	s->s_flags = flags;
853 
854 	error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
855 	if (error) {
856 		deactivate_locked_super(s);
857 		return error;
858 	}
859 	s->s_flags |= MS_ACTIVE;
860 	simple_set_mnt(mnt, s);
861 	return 0;
862 }
863 
864 EXPORT_SYMBOL(get_sb_nodev);
865 
866 static int compare_single(struct super_block *s, void *p)
867 {
868 	return 1;
869 }
870 
871 int get_sb_single(struct file_system_type *fs_type,
872 	int flags, void *data,
873 	int (*fill_super)(struct super_block *, void *, int),
874 	struct vfsmount *mnt)
875 {
876 	struct super_block *s;
877 	int error;
878 
879 	s = sget(fs_type, compare_single, set_anon_super, NULL);
880 	if (IS_ERR(s))
881 		return PTR_ERR(s);
882 	if (!s->s_root) {
883 		s->s_flags = flags;
884 		error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
885 		if (error) {
886 			deactivate_locked_super(s);
887 			return error;
888 		}
889 		s->s_flags |= MS_ACTIVE;
890 	} else {
891 		do_remount_sb(s, flags, data, 0);
892 	}
893 	simple_set_mnt(mnt, s);
894 	return 0;
895 }
896 
897 EXPORT_SYMBOL(get_sb_single);
898 
899 struct vfsmount *
900 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
901 {
902 	struct vfsmount *mnt;
903 	char *secdata = NULL;
904 	int error;
905 
906 	if (!type)
907 		return ERR_PTR(-ENODEV);
908 
909 	error = -ENOMEM;
910 	mnt = alloc_vfsmnt(name);
911 	if (!mnt)
912 		goto out;
913 
914 	if (flags & MS_KERNMOUNT)
915 		mnt->mnt_flags = MNT_INTERNAL;
916 
917 	if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
918 		secdata = alloc_secdata();
919 		if (!secdata)
920 			goto out_mnt;
921 
922 		error = security_sb_copy_data(data, secdata);
923 		if (error)
924 			goto out_free_secdata;
925 	}
926 
927 	error = type->get_sb(type, flags, name, data, mnt);
928 	if (error < 0)
929 		goto out_free_secdata;
930 	BUG_ON(!mnt->mnt_sb);
931 	WARN_ON(!mnt->mnt_sb->s_bdi);
932 	mnt->mnt_sb->s_flags |= MS_BORN;
933 
934 	error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata);
935 	if (error)
936 		goto out_sb;
937 
938 	/*
939 	 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
940 	 * but s_maxbytes was an unsigned long long for many releases. Throw
941 	 * this warning for a little while to try and catch filesystems that
942 	 * violate this rule. This warning should be either removed or
943 	 * converted to a BUG() in 2.6.34.
944 	 */
945 	WARN((mnt->mnt_sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
946 		"negative value (%lld)\n", type->name, mnt->mnt_sb->s_maxbytes);
947 
948 	mnt->mnt_mountpoint = mnt->mnt_root;
949 	mnt->mnt_parent = mnt;
950 	up_write(&mnt->mnt_sb->s_umount);
951 	free_secdata(secdata);
952 	return mnt;
953 out_sb:
954 	dput(mnt->mnt_root);
955 	deactivate_locked_super(mnt->mnt_sb);
956 out_free_secdata:
957 	free_secdata(secdata);
958 out_mnt:
959 	free_vfsmnt(mnt);
960 out:
961 	return ERR_PTR(error);
962 }
963 
964 EXPORT_SYMBOL_GPL(vfs_kern_mount);
965 
966 /**
967  * freeze_super - lock the filesystem and force it into a consistent state
968  * @sb: the super to lock
969  *
970  * Syncs the super to make sure the filesystem is consistent and calls the fs's
971  * freeze_fs.  Subsequent calls to this without first thawing the fs will return
972  * -EBUSY.
973  */
974 int freeze_super(struct super_block *sb)
975 {
976 	int ret;
977 
978 	atomic_inc(&sb->s_active);
979 	down_write(&sb->s_umount);
980 	if (sb->s_frozen) {
981 		deactivate_locked_super(sb);
982 		return -EBUSY;
983 	}
984 
985 	if (sb->s_flags & MS_RDONLY) {
986 		sb->s_frozen = SB_FREEZE_TRANS;
987 		smp_wmb();
988 		up_write(&sb->s_umount);
989 		return 0;
990 	}
991 
992 	sb->s_frozen = SB_FREEZE_WRITE;
993 	smp_wmb();
994 
995 	sync_filesystem(sb);
996 
997 	sb->s_frozen = SB_FREEZE_TRANS;
998 	smp_wmb();
999 
1000 	sync_blockdev(sb->s_bdev);
1001 	if (sb->s_op->freeze_fs) {
1002 		ret = sb->s_op->freeze_fs(sb);
1003 		if (ret) {
1004 			printk(KERN_ERR
1005 				"VFS:Filesystem freeze failed\n");
1006 			sb->s_frozen = SB_UNFROZEN;
1007 			deactivate_locked_super(sb);
1008 			return ret;
1009 		}
1010 	}
1011 	up_write(&sb->s_umount);
1012 	return 0;
1013 }
1014 EXPORT_SYMBOL(freeze_super);
1015 
1016 /**
1017  * thaw_super -- unlock filesystem
1018  * @sb: the super to thaw
1019  *
1020  * Unlocks the filesystem and marks it writeable again after freeze_super().
1021  */
1022 int thaw_super(struct super_block *sb)
1023 {
1024 	int error;
1025 
1026 	down_write(&sb->s_umount);
1027 	if (sb->s_frozen == SB_UNFROZEN) {
1028 		up_write(&sb->s_umount);
1029 		return -EINVAL;
1030 	}
1031 
1032 	if (sb->s_flags & MS_RDONLY)
1033 		goto out;
1034 
1035 	if (sb->s_op->unfreeze_fs) {
1036 		error = sb->s_op->unfreeze_fs(sb);
1037 		if (error) {
1038 			printk(KERN_ERR
1039 				"VFS:Filesystem thaw failed\n");
1040 			sb->s_frozen = SB_FREEZE_TRANS;
1041 			up_write(&sb->s_umount);
1042 			return error;
1043 		}
1044 	}
1045 
1046 out:
1047 	sb->s_frozen = SB_UNFROZEN;
1048 	smp_wmb();
1049 	wake_up(&sb->s_wait_unfrozen);
1050 	deactivate_locked_super(sb);
1051 
1052 	return 0;
1053 }
1054 EXPORT_SYMBOL(thaw_super);
1055 
1056 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1057 {
1058 	int err;
1059 	const char *subtype = strchr(fstype, '.');
1060 	if (subtype) {
1061 		subtype++;
1062 		err = -EINVAL;
1063 		if (!subtype[0])
1064 			goto err;
1065 	} else
1066 		subtype = "";
1067 
1068 	mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1069 	err = -ENOMEM;
1070 	if (!mnt->mnt_sb->s_subtype)
1071 		goto err;
1072 	return mnt;
1073 
1074  err:
1075 	mntput(mnt);
1076 	return ERR_PTR(err);
1077 }
1078 
1079 struct vfsmount *
1080 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1081 {
1082 	struct file_system_type *type = get_fs_type(fstype);
1083 	struct vfsmount *mnt;
1084 	if (!type)
1085 		return ERR_PTR(-ENODEV);
1086 	mnt = vfs_kern_mount(type, flags, name, data);
1087 	if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1088 	    !mnt->mnt_sb->s_subtype)
1089 		mnt = fs_set_subtype(mnt, fstype);
1090 	put_filesystem(type);
1091 	return mnt;
1092 }
1093 EXPORT_SYMBOL_GPL(do_kern_mount);
1094 
1095 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
1096 {
1097 	return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
1098 }
1099 
1100 EXPORT_SYMBOL_GPL(kern_mount_data);
1101