xref: /linux/fs/super.c (revision 791d3ef2e11100449837dc0b6fe884e60ca3a484)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/super.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  *
7  *  super.c contains code to handle: - mount structures
8  *                                   - super-block tables
9  *                                   - filesystem drivers list
10  *                                   - mount system call
11  *                                   - umount system call
12  *                                   - ustat system call
13  *
14  * GK 2/5/95  -  Changed to support mounting the root fs via NFS
15  *
16  *  Added kerneld support: Jacques Gelinas and Bjorn Ekwall
17  *  Added change_root: Werner Almesberger & Hans Lermen, Feb '96
18  *  Added options to /proc/mounts:
19  *    Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
20  *  Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
21  *  Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
22  */
23 
24 #include <linux/export.h>
25 #include <linux/slab.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 <linux/fsnotify.h>
36 #include <linux/lockdep.h>
37 #include <linux/user_namespace.h>
38 #include "internal.h"
39 
40 static int thaw_super_locked(struct super_block *sb);
41 
42 static LIST_HEAD(super_blocks);
43 static DEFINE_SPINLOCK(sb_lock);
44 
45 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
46 	"sb_writers",
47 	"sb_pagefaults",
48 	"sb_internal",
49 };
50 
51 /*
52  * One thing we have to be careful of with a per-sb shrinker is that we don't
53  * drop the last active reference to the superblock from within the shrinker.
54  * If that happens we could trigger unregistering the shrinker from within the
55  * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
56  * take a passive reference to the superblock to avoid this from occurring.
57  */
58 static unsigned long super_cache_scan(struct shrinker *shrink,
59 				      struct shrink_control *sc)
60 {
61 	struct super_block *sb;
62 	long	fs_objects = 0;
63 	long	total_objects;
64 	long	freed = 0;
65 	long	dentries;
66 	long	inodes;
67 
68 	sb = container_of(shrink, struct super_block, s_shrink);
69 
70 	/*
71 	 * Deadlock avoidance.  We may hold various FS locks, and we don't want
72 	 * to recurse into the FS that called us in clear_inode() and friends..
73 	 */
74 	if (!(sc->gfp_mask & __GFP_FS))
75 		return SHRINK_STOP;
76 
77 	if (!trylock_super(sb))
78 		return SHRINK_STOP;
79 
80 	if (sb->s_op->nr_cached_objects)
81 		fs_objects = sb->s_op->nr_cached_objects(sb, sc);
82 
83 	inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
84 	dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
85 	total_objects = dentries + inodes + fs_objects + 1;
86 	if (!total_objects)
87 		total_objects = 1;
88 
89 	/* proportion the scan between the caches */
90 	dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
91 	inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
92 	fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
93 
94 	/*
95 	 * prune the dcache first as the icache is pinned by it, then
96 	 * prune the icache, followed by the filesystem specific caches
97 	 *
98 	 * Ensure that we always scan at least one object - memcg kmem
99 	 * accounting uses this to fully empty the caches.
100 	 */
101 	sc->nr_to_scan = dentries + 1;
102 	freed = prune_dcache_sb(sb, sc);
103 	sc->nr_to_scan = inodes + 1;
104 	freed += prune_icache_sb(sb, sc);
105 
106 	if (fs_objects) {
107 		sc->nr_to_scan = fs_objects + 1;
108 		freed += sb->s_op->free_cached_objects(sb, sc);
109 	}
110 
111 	up_read(&sb->s_umount);
112 	return freed;
113 }
114 
115 static unsigned long super_cache_count(struct shrinker *shrink,
116 				       struct shrink_control *sc)
117 {
118 	struct super_block *sb;
119 	long	total_objects = 0;
120 
121 	sb = container_of(shrink, struct super_block, s_shrink);
122 
123 	/*
124 	 * We don't call trylock_super() here as it is a scalability bottleneck,
125 	 * so we're exposed to partial setup state. The shrinker rwsem does not
126 	 * protect filesystem operations backing list_lru_shrink_count() or
127 	 * s_op->nr_cached_objects(). Counts can change between
128 	 * super_cache_count and super_cache_scan, so we really don't need locks
129 	 * here.
130 	 *
131 	 * However, if we are currently mounting the superblock, the underlying
132 	 * filesystem might be in a state of partial construction and hence it
133 	 * is dangerous to access it.  trylock_super() uses a SB_BORN check to
134 	 * avoid this situation, so do the same here. The memory barrier is
135 	 * matched with the one in mount_fs() as we don't hold locks here.
136 	 */
137 	if (!(sb->s_flags & SB_BORN))
138 		return 0;
139 	smp_rmb();
140 
141 	if (sb->s_op && sb->s_op->nr_cached_objects)
142 		total_objects = sb->s_op->nr_cached_objects(sb, sc);
143 
144 	total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
145 	total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
146 
147 	total_objects = vfs_pressure_ratio(total_objects);
148 	return total_objects;
149 }
150 
151 static void destroy_super_work(struct work_struct *work)
152 {
153 	struct super_block *s = container_of(work, struct super_block,
154 							destroy_work);
155 	int i;
156 
157 	for (i = 0; i < SB_FREEZE_LEVELS; i++)
158 		percpu_free_rwsem(&s->s_writers.rw_sem[i]);
159 	kfree(s);
160 }
161 
162 static void destroy_super_rcu(struct rcu_head *head)
163 {
164 	struct super_block *s = container_of(head, struct super_block, rcu);
165 	INIT_WORK(&s->destroy_work, destroy_super_work);
166 	schedule_work(&s->destroy_work);
167 }
168 
169 /* Free a superblock that has never been seen by anyone */
170 static void destroy_unused_super(struct super_block *s)
171 {
172 	if (!s)
173 		return;
174 	up_write(&s->s_umount);
175 	list_lru_destroy(&s->s_dentry_lru);
176 	list_lru_destroy(&s->s_inode_lru);
177 	security_sb_free(s);
178 	put_user_ns(s->s_user_ns);
179 	kfree(s->s_subtype);
180 	free_prealloced_shrinker(&s->s_shrink);
181 	/* no delays needed */
182 	destroy_super_work(&s->destroy_work);
183 }
184 
185 /**
186  *	alloc_super	-	create new superblock
187  *	@type:	filesystem type superblock should belong to
188  *	@flags: the mount flags
189  *	@user_ns: User namespace for the super_block
190  *
191  *	Allocates and initializes a new &struct super_block.  alloc_super()
192  *	returns a pointer new superblock or %NULL if allocation had failed.
193  */
194 static struct super_block *alloc_super(struct file_system_type *type, int flags,
195 				       struct user_namespace *user_ns)
196 {
197 	struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
198 	static const struct super_operations default_op;
199 	int i;
200 
201 	if (!s)
202 		return NULL;
203 
204 	INIT_LIST_HEAD(&s->s_mounts);
205 	s->s_user_ns = get_user_ns(user_ns);
206 	init_rwsem(&s->s_umount);
207 	lockdep_set_class(&s->s_umount, &type->s_umount_key);
208 	/*
209 	 * sget() can have s_umount recursion.
210 	 *
211 	 * When it cannot find a suitable sb, it allocates a new
212 	 * one (this one), and tries again to find a suitable old
213 	 * one.
214 	 *
215 	 * In case that succeeds, it will acquire the s_umount
216 	 * lock of the old one. Since these are clearly distrinct
217 	 * locks, and this object isn't exposed yet, there's no
218 	 * risk of deadlocks.
219 	 *
220 	 * Annotate this by putting this lock in a different
221 	 * subclass.
222 	 */
223 	down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
224 
225 	if (security_sb_alloc(s))
226 		goto fail;
227 
228 	for (i = 0; i < SB_FREEZE_LEVELS; i++) {
229 		if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
230 					sb_writers_name[i],
231 					&type->s_writers_key[i]))
232 			goto fail;
233 	}
234 	init_waitqueue_head(&s->s_writers.wait_unfrozen);
235 	s->s_bdi = &noop_backing_dev_info;
236 	s->s_flags = flags;
237 	if (s->s_user_ns != &init_user_ns)
238 		s->s_iflags |= SB_I_NODEV;
239 	INIT_HLIST_NODE(&s->s_instances);
240 	INIT_HLIST_BL_HEAD(&s->s_roots);
241 	mutex_init(&s->s_sync_lock);
242 	INIT_LIST_HEAD(&s->s_inodes);
243 	spin_lock_init(&s->s_inode_list_lock);
244 	INIT_LIST_HEAD(&s->s_inodes_wb);
245 	spin_lock_init(&s->s_inode_wblist_lock);
246 
247 	if (list_lru_init_memcg(&s->s_dentry_lru))
248 		goto fail;
249 	if (list_lru_init_memcg(&s->s_inode_lru))
250 		goto fail;
251 	s->s_count = 1;
252 	atomic_set(&s->s_active, 1);
253 	mutex_init(&s->s_vfs_rename_mutex);
254 	lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
255 	init_rwsem(&s->s_dquot.dqio_sem);
256 	s->s_maxbytes = MAX_NON_LFS;
257 	s->s_op = &default_op;
258 	s->s_time_gran = 1000000000;
259 	s->cleancache_poolid = CLEANCACHE_NO_POOL;
260 
261 	s->s_shrink.seeks = DEFAULT_SEEKS;
262 	s->s_shrink.scan_objects = super_cache_scan;
263 	s->s_shrink.count_objects = super_cache_count;
264 	s->s_shrink.batch = 1024;
265 	s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
266 	if (prealloc_shrinker(&s->s_shrink))
267 		goto fail;
268 	return s;
269 
270 fail:
271 	destroy_unused_super(s);
272 	return NULL;
273 }
274 
275 /* Superblock refcounting  */
276 
277 /*
278  * Drop a superblock's refcount.  The caller must hold sb_lock.
279  */
280 static void __put_super(struct super_block *s)
281 {
282 	if (!--s->s_count) {
283 		list_del_init(&s->s_list);
284 		WARN_ON(s->s_dentry_lru.node);
285 		WARN_ON(s->s_inode_lru.node);
286 		WARN_ON(!list_empty(&s->s_mounts));
287 		security_sb_free(s);
288 		put_user_ns(s->s_user_ns);
289 		kfree(s->s_subtype);
290 		call_rcu(&s->rcu, destroy_super_rcu);
291 	}
292 }
293 
294 /**
295  *	put_super	-	drop a temporary reference to superblock
296  *	@sb: superblock in question
297  *
298  *	Drops a temporary reference, frees superblock if there's no
299  *	references left.
300  */
301 static void put_super(struct super_block *sb)
302 {
303 	spin_lock(&sb_lock);
304 	__put_super(sb);
305 	spin_unlock(&sb_lock);
306 }
307 
308 
309 /**
310  *	deactivate_locked_super	-	drop an active reference to superblock
311  *	@s: superblock to deactivate
312  *
313  *	Drops an active reference to superblock, converting it into a temporary
314  *	one if there is no other active references left.  In that case we
315  *	tell fs driver to shut it down and drop the temporary reference we
316  *	had just acquired.
317  *
318  *	Caller holds exclusive lock on superblock; that lock is released.
319  */
320 void deactivate_locked_super(struct super_block *s)
321 {
322 	struct file_system_type *fs = s->s_type;
323 	if (atomic_dec_and_test(&s->s_active)) {
324 		cleancache_invalidate_fs(s);
325 		unregister_shrinker(&s->s_shrink);
326 		fs->kill_sb(s);
327 
328 		/*
329 		 * Since list_lru_destroy() may sleep, we cannot call it from
330 		 * put_super(), where we hold the sb_lock. Therefore we destroy
331 		 * the lru lists right now.
332 		 */
333 		list_lru_destroy(&s->s_dentry_lru);
334 		list_lru_destroy(&s->s_inode_lru);
335 
336 		put_filesystem(fs);
337 		put_super(s);
338 	} else {
339 		up_write(&s->s_umount);
340 	}
341 }
342 
343 EXPORT_SYMBOL(deactivate_locked_super);
344 
345 /**
346  *	deactivate_super	-	drop an active reference to superblock
347  *	@s: superblock to deactivate
348  *
349  *	Variant of deactivate_locked_super(), except that superblock is *not*
350  *	locked by caller.  If we are going to drop the final active reference,
351  *	lock will be acquired prior to that.
352  */
353 void deactivate_super(struct super_block *s)
354 {
355         if (!atomic_add_unless(&s->s_active, -1, 1)) {
356 		down_write(&s->s_umount);
357 		deactivate_locked_super(s);
358 	}
359 }
360 
361 EXPORT_SYMBOL(deactivate_super);
362 
363 /**
364  *	grab_super - acquire an active reference
365  *	@s: reference we are trying to make active
366  *
367  *	Tries to acquire an active reference.  grab_super() is used when we
368  * 	had just found a superblock in super_blocks or fs_type->fs_supers
369  *	and want to turn it into a full-blown active reference.  grab_super()
370  *	is called with sb_lock held and drops it.  Returns 1 in case of
371  *	success, 0 if we had failed (superblock contents was already dead or
372  *	dying when grab_super() had been called).  Note that this is only
373  *	called for superblocks not in rundown mode (== ones still on ->fs_supers
374  *	of their type), so increment of ->s_count is OK here.
375  */
376 static int grab_super(struct super_block *s) __releases(sb_lock)
377 {
378 	s->s_count++;
379 	spin_unlock(&sb_lock);
380 	down_write(&s->s_umount);
381 	if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
382 		put_super(s);
383 		return 1;
384 	}
385 	up_write(&s->s_umount);
386 	put_super(s);
387 	return 0;
388 }
389 
390 /*
391  *	trylock_super - try to grab ->s_umount shared
392  *	@sb: reference we are trying to grab
393  *
394  *	Try to prevent fs shutdown.  This is used in places where we
395  *	cannot take an active reference but we need to ensure that the
396  *	filesystem is not shut down while we are working on it. It returns
397  *	false if we cannot acquire s_umount or if we lose the race and
398  *	filesystem already got into shutdown, and returns true with the s_umount
399  *	lock held in read mode in case of success. On successful return,
400  *	the caller must drop the s_umount lock when done.
401  *
402  *	Note that unlike get_super() et.al. this one does *not* bump ->s_count.
403  *	The reason why it's safe is that we are OK with doing trylock instead
404  *	of down_read().  There's a couple of places that are OK with that, but
405  *	it's very much not a general-purpose interface.
406  */
407 bool trylock_super(struct super_block *sb)
408 {
409 	if (down_read_trylock(&sb->s_umount)) {
410 		if (!hlist_unhashed(&sb->s_instances) &&
411 		    sb->s_root && (sb->s_flags & SB_BORN))
412 			return true;
413 		up_read(&sb->s_umount);
414 	}
415 
416 	return false;
417 }
418 
419 /**
420  *	generic_shutdown_super	-	common helper for ->kill_sb()
421  *	@sb: superblock to kill
422  *
423  *	generic_shutdown_super() does all fs-independent work on superblock
424  *	shutdown.  Typical ->kill_sb() should pick all fs-specific objects
425  *	that need destruction out of superblock, call generic_shutdown_super()
426  *	and release aforementioned objects.  Note: dentries and inodes _are_
427  *	taken care of and do not need specific handling.
428  *
429  *	Upon calling this function, the filesystem may no longer alter or
430  *	rearrange the set of dentries belonging to this super_block, nor may it
431  *	change the attachments of dentries to inodes.
432  */
433 void generic_shutdown_super(struct super_block *sb)
434 {
435 	const struct super_operations *sop = sb->s_op;
436 
437 	if (sb->s_root) {
438 		shrink_dcache_for_umount(sb);
439 		sync_filesystem(sb);
440 		sb->s_flags &= ~SB_ACTIVE;
441 
442 		fsnotify_unmount_inodes(sb);
443 		cgroup_writeback_umount();
444 
445 		evict_inodes(sb);
446 
447 		if (sb->s_dio_done_wq) {
448 			destroy_workqueue(sb->s_dio_done_wq);
449 			sb->s_dio_done_wq = NULL;
450 		}
451 
452 		if (sop->put_super)
453 			sop->put_super(sb);
454 
455 		if (!list_empty(&sb->s_inodes)) {
456 			printk("VFS: Busy inodes after unmount of %s. "
457 			   "Self-destruct in 5 seconds.  Have a nice day...\n",
458 			   sb->s_id);
459 		}
460 	}
461 	spin_lock(&sb_lock);
462 	/* should be initialized for __put_super_and_need_restart() */
463 	hlist_del_init(&sb->s_instances);
464 	spin_unlock(&sb_lock);
465 	up_write(&sb->s_umount);
466 	if (sb->s_bdi != &noop_backing_dev_info) {
467 		bdi_put(sb->s_bdi);
468 		sb->s_bdi = &noop_backing_dev_info;
469 	}
470 }
471 
472 EXPORT_SYMBOL(generic_shutdown_super);
473 
474 /**
475  *	sget_userns -	find or create a superblock
476  *	@type:	filesystem type superblock should belong to
477  *	@test:	comparison callback
478  *	@set:	setup callback
479  *	@flags:	mount flags
480  *	@user_ns: User namespace for the super_block
481  *	@data:	argument to each of them
482  */
483 struct super_block *sget_userns(struct file_system_type *type,
484 			int (*test)(struct super_block *,void *),
485 			int (*set)(struct super_block *,void *),
486 			int flags, struct user_namespace *user_ns,
487 			void *data)
488 {
489 	struct super_block *s = NULL;
490 	struct super_block *old;
491 	int err;
492 
493 	if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) &&
494 	    !(type->fs_flags & FS_USERNS_MOUNT) &&
495 	    !capable(CAP_SYS_ADMIN))
496 		return ERR_PTR(-EPERM);
497 retry:
498 	spin_lock(&sb_lock);
499 	if (test) {
500 		hlist_for_each_entry(old, &type->fs_supers, s_instances) {
501 			if (!test(old, data))
502 				continue;
503 			if (user_ns != old->s_user_ns) {
504 				spin_unlock(&sb_lock);
505 				destroy_unused_super(s);
506 				return ERR_PTR(-EBUSY);
507 			}
508 			if (!grab_super(old))
509 				goto retry;
510 			destroy_unused_super(s);
511 			return old;
512 		}
513 	}
514 	if (!s) {
515 		spin_unlock(&sb_lock);
516 		s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
517 		if (!s)
518 			return ERR_PTR(-ENOMEM);
519 		goto retry;
520 	}
521 
522 	err = set(s, data);
523 	if (err) {
524 		spin_unlock(&sb_lock);
525 		destroy_unused_super(s);
526 		return ERR_PTR(err);
527 	}
528 	s->s_type = type;
529 	strlcpy(s->s_id, type->name, sizeof(s->s_id));
530 	list_add_tail(&s->s_list, &super_blocks);
531 	hlist_add_head(&s->s_instances, &type->fs_supers);
532 	spin_unlock(&sb_lock);
533 	get_filesystem(type);
534 	register_shrinker_prepared(&s->s_shrink);
535 	return s;
536 }
537 
538 EXPORT_SYMBOL(sget_userns);
539 
540 /**
541  *	sget	-	find or create a superblock
542  *	@type:	  filesystem type superblock should belong to
543  *	@test:	  comparison callback
544  *	@set:	  setup callback
545  *	@flags:	  mount flags
546  *	@data:	  argument to each of them
547  */
548 struct super_block *sget(struct file_system_type *type,
549 			int (*test)(struct super_block *,void *),
550 			int (*set)(struct super_block *,void *),
551 			int flags,
552 			void *data)
553 {
554 	struct user_namespace *user_ns = current_user_ns();
555 
556 	/* We don't yet pass the user namespace of the parent
557 	 * mount through to here so always use &init_user_ns
558 	 * until that changes.
559 	 */
560 	if (flags & SB_SUBMOUNT)
561 		user_ns = &init_user_ns;
562 
563 	/* Ensure the requestor has permissions over the target filesystem */
564 	if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
565 		return ERR_PTR(-EPERM);
566 
567 	return sget_userns(type, test, set, flags, user_ns, data);
568 }
569 
570 EXPORT_SYMBOL(sget);
571 
572 void drop_super(struct super_block *sb)
573 {
574 	up_read(&sb->s_umount);
575 	put_super(sb);
576 }
577 
578 EXPORT_SYMBOL(drop_super);
579 
580 void drop_super_exclusive(struct super_block *sb)
581 {
582 	up_write(&sb->s_umount);
583 	put_super(sb);
584 }
585 EXPORT_SYMBOL(drop_super_exclusive);
586 
587 static void __iterate_supers(void (*f)(struct super_block *))
588 {
589 	struct super_block *sb, *p = NULL;
590 
591 	spin_lock(&sb_lock);
592 	list_for_each_entry(sb, &super_blocks, s_list) {
593 		if (hlist_unhashed(&sb->s_instances))
594 			continue;
595 		sb->s_count++;
596 		spin_unlock(&sb_lock);
597 
598 		f(sb);
599 
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 }
609 /**
610  *	iterate_supers - call function for all active superblocks
611  *	@f: function to call
612  *	@arg: argument to pass to it
613  *
614  *	Scans the superblock list and calls given function, passing it
615  *	locked superblock and given argument.
616  */
617 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
618 {
619 	struct super_block *sb, *p = NULL;
620 
621 	spin_lock(&sb_lock);
622 	list_for_each_entry(sb, &super_blocks, s_list) {
623 		if (hlist_unhashed(&sb->s_instances))
624 			continue;
625 		sb->s_count++;
626 		spin_unlock(&sb_lock);
627 
628 		down_read(&sb->s_umount);
629 		if (sb->s_root && (sb->s_flags & SB_BORN))
630 			f(sb, arg);
631 		up_read(&sb->s_umount);
632 
633 		spin_lock(&sb_lock);
634 		if (p)
635 			__put_super(p);
636 		p = sb;
637 	}
638 	if (p)
639 		__put_super(p);
640 	spin_unlock(&sb_lock);
641 }
642 
643 /**
644  *	iterate_supers_type - call function for superblocks of given type
645  *	@type: fs type
646  *	@f: function to call
647  *	@arg: argument to pass to it
648  *
649  *	Scans the superblock list and calls given function, passing it
650  *	locked superblock and given argument.
651  */
652 void iterate_supers_type(struct file_system_type *type,
653 	void (*f)(struct super_block *, void *), void *arg)
654 {
655 	struct super_block *sb, *p = NULL;
656 
657 	spin_lock(&sb_lock);
658 	hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
659 		sb->s_count++;
660 		spin_unlock(&sb_lock);
661 
662 		down_read(&sb->s_umount);
663 		if (sb->s_root && (sb->s_flags & SB_BORN))
664 			f(sb, arg);
665 		up_read(&sb->s_umount);
666 
667 		spin_lock(&sb_lock);
668 		if (p)
669 			__put_super(p);
670 		p = sb;
671 	}
672 	if (p)
673 		__put_super(p);
674 	spin_unlock(&sb_lock);
675 }
676 
677 EXPORT_SYMBOL(iterate_supers_type);
678 
679 static struct super_block *__get_super(struct block_device *bdev, bool excl)
680 {
681 	struct super_block *sb;
682 
683 	if (!bdev)
684 		return NULL;
685 
686 	spin_lock(&sb_lock);
687 rescan:
688 	list_for_each_entry(sb, &super_blocks, s_list) {
689 		if (hlist_unhashed(&sb->s_instances))
690 			continue;
691 		if (sb->s_bdev == bdev) {
692 			sb->s_count++;
693 			spin_unlock(&sb_lock);
694 			if (!excl)
695 				down_read(&sb->s_umount);
696 			else
697 				down_write(&sb->s_umount);
698 			/* still alive? */
699 			if (sb->s_root && (sb->s_flags & SB_BORN))
700 				return sb;
701 			if (!excl)
702 				up_read(&sb->s_umount);
703 			else
704 				up_write(&sb->s_umount);
705 			/* nope, got unmounted */
706 			spin_lock(&sb_lock);
707 			__put_super(sb);
708 			goto rescan;
709 		}
710 	}
711 	spin_unlock(&sb_lock);
712 	return NULL;
713 }
714 
715 /**
716  *	get_super - get the superblock of a device
717  *	@bdev: device to get the superblock for
718  *
719  *	Scans the superblock list and finds the superblock of the file system
720  *	mounted on the device given. %NULL is returned if no match is found.
721  */
722 struct super_block *get_super(struct block_device *bdev)
723 {
724 	return __get_super(bdev, false);
725 }
726 EXPORT_SYMBOL(get_super);
727 
728 static struct super_block *__get_super_thawed(struct block_device *bdev,
729 					      bool excl)
730 {
731 	while (1) {
732 		struct super_block *s = __get_super(bdev, excl);
733 		if (!s || s->s_writers.frozen == SB_UNFROZEN)
734 			return s;
735 		if (!excl)
736 			up_read(&s->s_umount);
737 		else
738 			up_write(&s->s_umount);
739 		wait_event(s->s_writers.wait_unfrozen,
740 			   s->s_writers.frozen == SB_UNFROZEN);
741 		put_super(s);
742 	}
743 }
744 
745 /**
746  *	get_super_thawed - get thawed superblock of a device
747  *	@bdev: device to get the superblock for
748  *
749  *	Scans the superblock list and finds the superblock of the file system
750  *	mounted on the device. The superblock is returned once it is thawed
751  *	(or immediately if it was not frozen). %NULL is returned if no match
752  *	is found.
753  */
754 struct super_block *get_super_thawed(struct block_device *bdev)
755 {
756 	return __get_super_thawed(bdev, false);
757 }
758 EXPORT_SYMBOL(get_super_thawed);
759 
760 /**
761  *	get_super_exclusive_thawed - get thawed superblock of a device
762  *	@bdev: device to get the superblock for
763  *
764  *	Scans the superblock list and finds the superblock of the file system
765  *	mounted on the device. The superblock is returned once it is thawed
766  *	(or immediately if it was not frozen) and s_umount semaphore is held
767  *	in exclusive mode. %NULL is returned if no match is found.
768  */
769 struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
770 {
771 	return __get_super_thawed(bdev, true);
772 }
773 EXPORT_SYMBOL(get_super_exclusive_thawed);
774 
775 /**
776  * get_active_super - get an active reference to the superblock of a device
777  * @bdev: device to get the superblock for
778  *
779  * Scans the superblock list and finds the superblock of the file system
780  * mounted on the device given.  Returns the superblock with an active
781  * reference or %NULL if none was found.
782  */
783 struct super_block *get_active_super(struct block_device *bdev)
784 {
785 	struct super_block *sb;
786 
787 	if (!bdev)
788 		return NULL;
789 
790 restart:
791 	spin_lock(&sb_lock);
792 	list_for_each_entry(sb, &super_blocks, s_list) {
793 		if (hlist_unhashed(&sb->s_instances))
794 			continue;
795 		if (sb->s_bdev == bdev) {
796 			if (!grab_super(sb))
797 				goto restart;
798 			up_write(&sb->s_umount);
799 			return sb;
800 		}
801 	}
802 	spin_unlock(&sb_lock);
803 	return NULL;
804 }
805 
806 struct super_block *user_get_super(dev_t dev)
807 {
808 	struct super_block *sb;
809 
810 	spin_lock(&sb_lock);
811 rescan:
812 	list_for_each_entry(sb, &super_blocks, s_list) {
813 		if (hlist_unhashed(&sb->s_instances))
814 			continue;
815 		if (sb->s_dev ==  dev) {
816 			sb->s_count++;
817 			spin_unlock(&sb_lock);
818 			down_read(&sb->s_umount);
819 			/* still alive? */
820 			if (sb->s_root && (sb->s_flags & SB_BORN))
821 				return sb;
822 			up_read(&sb->s_umount);
823 			/* nope, got unmounted */
824 			spin_lock(&sb_lock);
825 			__put_super(sb);
826 			goto rescan;
827 		}
828 	}
829 	spin_unlock(&sb_lock);
830 	return NULL;
831 }
832 
833 /**
834  *	do_remount_sb - asks filesystem to change mount options.
835  *	@sb:	superblock in question
836  *	@sb_flags: revised superblock flags
837  *	@data:	the rest of options
838  *      @force: whether or not to force the change
839  *
840  *	Alters the mount options of a mounted file system.
841  */
842 int do_remount_sb(struct super_block *sb, int sb_flags, void *data, int force)
843 {
844 	int retval;
845 	int remount_ro;
846 
847 	if (sb->s_writers.frozen != SB_UNFROZEN)
848 		return -EBUSY;
849 
850 #ifdef CONFIG_BLOCK
851 	if (!(sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
852 		return -EACCES;
853 #endif
854 
855 	remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
856 
857 	if (remount_ro) {
858 		if (!hlist_empty(&sb->s_pins)) {
859 			up_write(&sb->s_umount);
860 			group_pin_kill(&sb->s_pins);
861 			down_write(&sb->s_umount);
862 			if (!sb->s_root)
863 				return 0;
864 			if (sb->s_writers.frozen != SB_UNFROZEN)
865 				return -EBUSY;
866 			remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
867 		}
868 	}
869 	shrink_dcache_sb(sb);
870 
871 	/* If we are remounting RDONLY and current sb is read/write,
872 	   make sure there are no rw files opened */
873 	if (remount_ro) {
874 		if (force) {
875 			sb->s_readonly_remount = 1;
876 			smp_wmb();
877 		} else {
878 			retval = sb_prepare_remount_readonly(sb);
879 			if (retval)
880 				return retval;
881 		}
882 	}
883 
884 	if (sb->s_op->remount_fs) {
885 		retval = sb->s_op->remount_fs(sb, &sb_flags, data);
886 		if (retval) {
887 			if (!force)
888 				goto cancel_readonly;
889 			/* If forced remount, go ahead despite any errors */
890 			WARN(1, "forced remount of a %s fs returned %i\n",
891 			     sb->s_type->name, retval);
892 		}
893 	}
894 	sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (sb_flags & MS_RMT_MASK);
895 	/* Needs to be ordered wrt mnt_is_readonly() */
896 	smp_wmb();
897 	sb->s_readonly_remount = 0;
898 
899 	/*
900 	 * Some filesystems modify their metadata via some other path than the
901 	 * bdev buffer cache (eg. use a private mapping, or directories in
902 	 * pagecache, etc). Also file data modifications go via their own
903 	 * mappings. So If we try to mount readonly then copy the filesystem
904 	 * from bdev, we could get stale data, so invalidate it to give a best
905 	 * effort at coherency.
906 	 */
907 	if (remount_ro && sb->s_bdev)
908 		invalidate_bdev(sb->s_bdev);
909 	return 0;
910 
911 cancel_readonly:
912 	sb->s_readonly_remount = 0;
913 	return retval;
914 }
915 
916 static void do_emergency_remount_callback(struct super_block *sb)
917 {
918 	down_write(&sb->s_umount);
919 	if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
920 	    !sb_rdonly(sb)) {
921 		/*
922 		 * What lock protects sb->s_flags??
923 		 */
924 		do_remount_sb(sb, SB_RDONLY, NULL, 1);
925 	}
926 	up_write(&sb->s_umount);
927 }
928 
929 static void do_emergency_remount(struct work_struct *work)
930 {
931 	__iterate_supers(do_emergency_remount_callback);
932 	kfree(work);
933 	printk("Emergency Remount complete\n");
934 }
935 
936 void emergency_remount(void)
937 {
938 	struct work_struct *work;
939 
940 	work = kmalloc(sizeof(*work), GFP_ATOMIC);
941 	if (work) {
942 		INIT_WORK(work, do_emergency_remount);
943 		schedule_work(work);
944 	}
945 }
946 
947 static void do_thaw_all_callback(struct super_block *sb)
948 {
949 	down_write(&sb->s_umount);
950 	if (sb->s_root && sb->s_flags & SB_BORN) {
951 		emergency_thaw_bdev(sb);
952 		thaw_super_locked(sb);
953 	} else {
954 		up_write(&sb->s_umount);
955 	}
956 }
957 
958 static void do_thaw_all(struct work_struct *work)
959 {
960 	__iterate_supers(do_thaw_all_callback);
961 	kfree(work);
962 	printk(KERN_WARNING "Emergency Thaw complete\n");
963 }
964 
965 /**
966  * emergency_thaw_all -- forcibly thaw every frozen filesystem
967  *
968  * Used for emergency unfreeze of all filesystems via SysRq
969  */
970 void emergency_thaw_all(void)
971 {
972 	struct work_struct *work;
973 
974 	work = kmalloc(sizeof(*work), GFP_ATOMIC);
975 	if (work) {
976 		INIT_WORK(work, do_thaw_all);
977 		schedule_work(work);
978 	}
979 }
980 
981 /*
982  * Unnamed block devices are dummy devices used by virtual
983  * filesystems which don't use real block-devices.  -- jrs
984  */
985 
986 static DEFINE_IDA(unnamed_dev_ida);
987 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
988 /* Many userspace utilities consider an FSID of 0 invalid.
989  * Always return at least 1 from get_anon_bdev.
990  */
991 static int unnamed_dev_start = 1;
992 
993 int get_anon_bdev(dev_t *p)
994 {
995 	int dev;
996 	int error;
997 
998  retry:
999 	if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
1000 		return -ENOMEM;
1001 	spin_lock(&unnamed_dev_lock);
1002 	error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
1003 	if (!error)
1004 		unnamed_dev_start = dev + 1;
1005 	spin_unlock(&unnamed_dev_lock);
1006 	if (error == -EAGAIN)
1007 		/* We raced and lost with another CPU. */
1008 		goto retry;
1009 	else if (error)
1010 		return -EAGAIN;
1011 
1012 	if (dev >= (1 << MINORBITS)) {
1013 		spin_lock(&unnamed_dev_lock);
1014 		ida_remove(&unnamed_dev_ida, dev);
1015 		if (unnamed_dev_start > dev)
1016 			unnamed_dev_start = dev;
1017 		spin_unlock(&unnamed_dev_lock);
1018 		return -EMFILE;
1019 	}
1020 	*p = MKDEV(0, dev & MINORMASK);
1021 	return 0;
1022 }
1023 EXPORT_SYMBOL(get_anon_bdev);
1024 
1025 void free_anon_bdev(dev_t dev)
1026 {
1027 	int slot = MINOR(dev);
1028 	spin_lock(&unnamed_dev_lock);
1029 	ida_remove(&unnamed_dev_ida, slot);
1030 	if (slot < unnamed_dev_start)
1031 		unnamed_dev_start = slot;
1032 	spin_unlock(&unnamed_dev_lock);
1033 }
1034 EXPORT_SYMBOL(free_anon_bdev);
1035 
1036 int set_anon_super(struct super_block *s, void *data)
1037 {
1038 	return get_anon_bdev(&s->s_dev);
1039 }
1040 
1041 EXPORT_SYMBOL(set_anon_super);
1042 
1043 void kill_anon_super(struct super_block *sb)
1044 {
1045 	dev_t dev = sb->s_dev;
1046 	generic_shutdown_super(sb);
1047 	free_anon_bdev(dev);
1048 }
1049 
1050 EXPORT_SYMBOL(kill_anon_super);
1051 
1052 void kill_litter_super(struct super_block *sb)
1053 {
1054 	if (sb->s_root)
1055 		d_genocide(sb->s_root);
1056 	kill_anon_super(sb);
1057 }
1058 
1059 EXPORT_SYMBOL(kill_litter_super);
1060 
1061 static int ns_test_super(struct super_block *sb, void *data)
1062 {
1063 	return sb->s_fs_info == data;
1064 }
1065 
1066 static int ns_set_super(struct super_block *sb, void *data)
1067 {
1068 	sb->s_fs_info = data;
1069 	return set_anon_super(sb, NULL);
1070 }
1071 
1072 struct dentry *mount_ns(struct file_system_type *fs_type,
1073 	int flags, void *data, void *ns, struct user_namespace *user_ns,
1074 	int (*fill_super)(struct super_block *, void *, int))
1075 {
1076 	struct super_block *sb;
1077 
1078 	/* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1079 	 * over the namespace.
1080 	 */
1081 	if (!(flags & SB_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
1082 		return ERR_PTR(-EPERM);
1083 
1084 	sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1085 			 user_ns, ns);
1086 	if (IS_ERR(sb))
1087 		return ERR_CAST(sb);
1088 
1089 	if (!sb->s_root) {
1090 		int err;
1091 		err = fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
1092 		if (err) {
1093 			deactivate_locked_super(sb);
1094 			return ERR_PTR(err);
1095 		}
1096 
1097 		sb->s_flags |= SB_ACTIVE;
1098 	}
1099 
1100 	return dget(sb->s_root);
1101 }
1102 
1103 EXPORT_SYMBOL(mount_ns);
1104 
1105 #ifdef CONFIG_BLOCK
1106 static int set_bdev_super(struct super_block *s, void *data)
1107 {
1108 	s->s_bdev = data;
1109 	s->s_dev = s->s_bdev->bd_dev;
1110 	s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1111 
1112 	return 0;
1113 }
1114 
1115 static int test_bdev_super(struct super_block *s, void *data)
1116 {
1117 	return (void *)s->s_bdev == data;
1118 }
1119 
1120 struct dentry *mount_bdev(struct file_system_type *fs_type,
1121 	int flags, const char *dev_name, void *data,
1122 	int (*fill_super)(struct super_block *, void *, int))
1123 {
1124 	struct block_device *bdev;
1125 	struct super_block *s;
1126 	fmode_t mode = FMODE_READ | FMODE_EXCL;
1127 	int error = 0;
1128 
1129 	if (!(flags & SB_RDONLY))
1130 		mode |= FMODE_WRITE;
1131 
1132 	bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1133 	if (IS_ERR(bdev))
1134 		return ERR_CAST(bdev);
1135 
1136 	/*
1137 	 * once the super is inserted into the list by sget, s_umount
1138 	 * will protect the lockfs code from trying to start a snapshot
1139 	 * while we are mounting
1140 	 */
1141 	mutex_lock(&bdev->bd_fsfreeze_mutex);
1142 	if (bdev->bd_fsfreeze_count > 0) {
1143 		mutex_unlock(&bdev->bd_fsfreeze_mutex);
1144 		error = -EBUSY;
1145 		goto error_bdev;
1146 	}
1147 	s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1148 		 bdev);
1149 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
1150 	if (IS_ERR(s))
1151 		goto error_s;
1152 
1153 	if (s->s_root) {
1154 		if ((flags ^ s->s_flags) & SB_RDONLY) {
1155 			deactivate_locked_super(s);
1156 			error = -EBUSY;
1157 			goto error_bdev;
1158 		}
1159 
1160 		/*
1161 		 * s_umount nests inside bd_mutex during
1162 		 * __invalidate_device().  blkdev_put() acquires
1163 		 * bd_mutex and can't be called under s_umount.  Drop
1164 		 * s_umount temporarily.  This is safe as we're
1165 		 * holding an active reference.
1166 		 */
1167 		up_write(&s->s_umount);
1168 		blkdev_put(bdev, mode);
1169 		down_write(&s->s_umount);
1170 	} else {
1171 		s->s_mode = mode;
1172 		snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1173 		sb_set_blocksize(s, block_size(bdev));
1174 		error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1175 		if (error) {
1176 			deactivate_locked_super(s);
1177 			goto error;
1178 		}
1179 
1180 		s->s_flags |= SB_ACTIVE;
1181 		bdev->bd_super = s;
1182 	}
1183 
1184 	return dget(s->s_root);
1185 
1186 error_s:
1187 	error = PTR_ERR(s);
1188 error_bdev:
1189 	blkdev_put(bdev, mode);
1190 error:
1191 	return ERR_PTR(error);
1192 }
1193 EXPORT_SYMBOL(mount_bdev);
1194 
1195 void kill_block_super(struct super_block *sb)
1196 {
1197 	struct block_device *bdev = sb->s_bdev;
1198 	fmode_t mode = sb->s_mode;
1199 
1200 	bdev->bd_super = NULL;
1201 	generic_shutdown_super(sb);
1202 	sync_blockdev(bdev);
1203 	WARN_ON_ONCE(!(mode & FMODE_EXCL));
1204 	blkdev_put(bdev, mode | FMODE_EXCL);
1205 }
1206 
1207 EXPORT_SYMBOL(kill_block_super);
1208 #endif
1209 
1210 struct dentry *mount_nodev(struct file_system_type *fs_type,
1211 	int flags, void *data,
1212 	int (*fill_super)(struct super_block *, void *, int))
1213 {
1214 	int error;
1215 	struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1216 
1217 	if (IS_ERR(s))
1218 		return ERR_CAST(s);
1219 
1220 	error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1221 	if (error) {
1222 		deactivate_locked_super(s);
1223 		return ERR_PTR(error);
1224 	}
1225 	s->s_flags |= SB_ACTIVE;
1226 	return dget(s->s_root);
1227 }
1228 EXPORT_SYMBOL(mount_nodev);
1229 
1230 static int compare_single(struct super_block *s, void *p)
1231 {
1232 	return 1;
1233 }
1234 
1235 struct dentry *mount_single(struct file_system_type *fs_type,
1236 	int flags, void *data,
1237 	int (*fill_super)(struct super_block *, void *, int))
1238 {
1239 	struct super_block *s;
1240 	int error;
1241 
1242 	s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1243 	if (IS_ERR(s))
1244 		return ERR_CAST(s);
1245 	if (!s->s_root) {
1246 		error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1247 		if (error) {
1248 			deactivate_locked_super(s);
1249 			return ERR_PTR(error);
1250 		}
1251 		s->s_flags |= SB_ACTIVE;
1252 	} else {
1253 		do_remount_sb(s, flags, data, 0);
1254 	}
1255 	return dget(s->s_root);
1256 }
1257 EXPORT_SYMBOL(mount_single);
1258 
1259 struct dentry *
1260 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1261 {
1262 	struct dentry *root;
1263 	struct super_block *sb;
1264 	char *secdata = NULL;
1265 	int error = -ENOMEM;
1266 
1267 	if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1268 		secdata = alloc_secdata();
1269 		if (!secdata)
1270 			goto out;
1271 
1272 		error = security_sb_copy_data(data, secdata);
1273 		if (error)
1274 			goto out_free_secdata;
1275 	}
1276 
1277 	root = type->mount(type, flags, name, data);
1278 	if (IS_ERR(root)) {
1279 		error = PTR_ERR(root);
1280 		goto out_free_secdata;
1281 	}
1282 	sb = root->d_sb;
1283 	BUG_ON(!sb);
1284 	WARN_ON(!sb->s_bdi);
1285 
1286 	/*
1287 	 * Write barrier is for super_cache_count(). We place it before setting
1288 	 * SB_BORN as the data dependency between the two functions is the
1289 	 * superblock structure contents that we just set up, not the SB_BORN
1290 	 * flag.
1291 	 */
1292 	smp_wmb();
1293 	sb->s_flags |= SB_BORN;
1294 
1295 	error = security_sb_kern_mount(sb, flags, secdata);
1296 	if (error)
1297 		goto out_sb;
1298 
1299 	/*
1300 	 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1301 	 * but s_maxbytes was an unsigned long long for many releases. Throw
1302 	 * this warning for a little while to try and catch filesystems that
1303 	 * violate this rule.
1304 	 */
1305 	WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1306 		"negative value (%lld)\n", type->name, sb->s_maxbytes);
1307 
1308 	up_write(&sb->s_umount);
1309 	free_secdata(secdata);
1310 	return root;
1311 out_sb:
1312 	dput(root);
1313 	deactivate_locked_super(sb);
1314 out_free_secdata:
1315 	free_secdata(secdata);
1316 out:
1317 	return ERR_PTR(error);
1318 }
1319 
1320 /*
1321  * Setup private BDI for given superblock. It gets automatically cleaned up
1322  * in generic_shutdown_super().
1323  */
1324 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1325 {
1326 	struct backing_dev_info *bdi;
1327 	int err;
1328 	va_list args;
1329 
1330 	bdi = bdi_alloc(GFP_KERNEL);
1331 	if (!bdi)
1332 		return -ENOMEM;
1333 
1334 	bdi->name = sb->s_type->name;
1335 
1336 	va_start(args, fmt);
1337 	err = bdi_register_va(bdi, fmt, args);
1338 	va_end(args);
1339 	if (err) {
1340 		bdi_put(bdi);
1341 		return err;
1342 	}
1343 	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1344 	sb->s_bdi = bdi;
1345 
1346 	return 0;
1347 }
1348 EXPORT_SYMBOL(super_setup_bdi_name);
1349 
1350 /*
1351  * Setup private BDI for given superblock. I gets automatically cleaned up
1352  * in generic_shutdown_super().
1353  */
1354 int super_setup_bdi(struct super_block *sb)
1355 {
1356 	static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1357 
1358 	return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1359 				    atomic_long_inc_return(&bdi_seq));
1360 }
1361 EXPORT_SYMBOL(super_setup_bdi);
1362 
1363 /*
1364  * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1365  * instead.
1366  */
1367 void __sb_end_write(struct super_block *sb, int level)
1368 {
1369 	percpu_up_read(sb->s_writers.rw_sem + level-1);
1370 }
1371 EXPORT_SYMBOL(__sb_end_write);
1372 
1373 /*
1374  * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1375  * instead.
1376  */
1377 int __sb_start_write(struct super_block *sb, int level, bool wait)
1378 {
1379 	bool force_trylock = false;
1380 	int ret = 1;
1381 
1382 #ifdef CONFIG_LOCKDEP
1383 	/*
1384 	 * We want lockdep to tell us about possible deadlocks with freezing
1385 	 * but it's it bit tricky to properly instrument it. Getting a freeze
1386 	 * protection works as getting a read lock but there are subtle
1387 	 * problems. XFS for example gets freeze protection on internal level
1388 	 * twice in some cases, which is OK only because we already hold a
1389 	 * freeze protection also on higher level. Due to these cases we have
1390 	 * to use wait == F (trylock mode) which must not fail.
1391 	 */
1392 	if (wait) {
1393 		int i;
1394 
1395 		for (i = 0; i < level - 1; i++)
1396 			if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1397 				force_trylock = true;
1398 				break;
1399 			}
1400 	}
1401 #endif
1402 	if (wait && !force_trylock)
1403 		percpu_down_read(sb->s_writers.rw_sem + level-1);
1404 	else
1405 		ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1406 
1407 	WARN_ON(force_trylock && !ret);
1408 	return ret;
1409 }
1410 EXPORT_SYMBOL(__sb_start_write);
1411 
1412 /**
1413  * sb_wait_write - wait until all writers to given file system finish
1414  * @sb: the super for which we wait
1415  * @level: type of writers we wait for (normal vs page fault)
1416  *
1417  * This function waits until there are no writers of given type to given file
1418  * system.
1419  */
1420 static void sb_wait_write(struct super_block *sb, int level)
1421 {
1422 	percpu_down_write(sb->s_writers.rw_sem + level-1);
1423 }
1424 
1425 /*
1426  * We are going to return to userspace and forget about these locks, the
1427  * ownership goes to the caller of thaw_super() which does unlock().
1428  */
1429 static void lockdep_sb_freeze_release(struct super_block *sb)
1430 {
1431 	int level;
1432 
1433 	for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1434 		percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1435 }
1436 
1437 /*
1438  * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1439  */
1440 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1441 {
1442 	int level;
1443 
1444 	for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1445 		percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1446 }
1447 
1448 static void sb_freeze_unlock(struct super_block *sb)
1449 {
1450 	int level;
1451 
1452 	for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1453 		percpu_up_write(sb->s_writers.rw_sem + level);
1454 }
1455 
1456 /**
1457  * freeze_super - lock the filesystem and force it into a consistent state
1458  * @sb: the super to lock
1459  *
1460  * Syncs the super to make sure the filesystem is consistent and calls the fs's
1461  * freeze_fs.  Subsequent calls to this without first thawing the fs will return
1462  * -EBUSY.
1463  *
1464  * During this function, sb->s_writers.frozen goes through these values:
1465  *
1466  * SB_UNFROZEN: File system is normal, all writes progress as usual.
1467  *
1468  * SB_FREEZE_WRITE: The file system is in the process of being frozen.  New
1469  * writes should be blocked, though page faults are still allowed. We wait for
1470  * all writes to complete and then proceed to the next stage.
1471  *
1472  * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1473  * but internal fs threads can still modify the filesystem (although they
1474  * should not dirty new pages or inodes), writeback can run etc. After waiting
1475  * for all running page faults we sync the filesystem which will clean all
1476  * dirty pages and inodes (no new dirty pages or inodes can be created when
1477  * sync is running).
1478  *
1479  * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1480  * modification are blocked (e.g. XFS preallocation truncation on inode
1481  * reclaim). This is usually implemented by blocking new transactions for
1482  * filesystems that have them and need this additional guard. After all
1483  * internal writers are finished we call ->freeze_fs() to finish filesystem
1484  * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1485  * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1486  *
1487  * sb->s_writers.frozen is protected by sb->s_umount.
1488  */
1489 int freeze_super(struct super_block *sb)
1490 {
1491 	int ret;
1492 
1493 	atomic_inc(&sb->s_active);
1494 	down_write(&sb->s_umount);
1495 	if (sb->s_writers.frozen != SB_UNFROZEN) {
1496 		deactivate_locked_super(sb);
1497 		return -EBUSY;
1498 	}
1499 
1500 	if (!(sb->s_flags & SB_BORN)) {
1501 		up_write(&sb->s_umount);
1502 		return 0;	/* sic - it's "nothing to do" */
1503 	}
1504 
1505 	if (sb_rdonly(sb)) {
1506 		/* Nothing to do really... */
1507 		sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1508 		up_write(&sb->s_umount);
1509 		return 0;
1510 	}
1511 
1512 	sb->s_writers.frozen = SB_FREEZE_WRITE;
1513 	/* Release s_umount to preserve sb_start_write -> s_umount ordering */
1514 	up_write(&sb->s_umount);
1515 	sb_wait_write(sb, SB_FREEZE_WRITE);
1516 	down_write(&sb->s_umount);
1517 
1518 	/* Now we go and block page faults... */
1519 	sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1520 	sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1521 
1522 	/* All writers are done so after syncing there won't be dirty data */
1523 	sync_filesystem(sb);
1524 
1525 	/* Now wait for internal filesystem counter */
1526 	sb->s_writers.frozen = SB_FREEZE_FS;
1527 	sb_wait_write(sb, SB_FREEZE_FS);
1528 
1529 	if (sb->s_op->freeze_fs) {
1530 		ret = sb->s_op->freeze_fs(sb);
1531 		if (ret) {
1532 			printk(KERN_ERR
1533 				"VFS:Filesystem freeze failed\n");
1534 			sb->s_writers.frozen = SB_UNFROZEN;
1535 			sb_freeze_unlock(sb);
1536 			wake_up(&sb->s_writers.wait_unfrozen);
1537 			deactivate_locked_super(sb);
1538 			return ret;
1539 		}
1540 	}
1541 	/*
1542 	 * For debugging purposes so that fs can warn if it sees write activity
1543 	 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1544 	 */
1545 	sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1546 	lockdep_sb_freeze_release(sb);
1547 	up_write(&sb->s_umount);
1548 	return 0;
1549 }
1550 EXPORT_SYMBOL(freeze_super);
1551 
1552 /**
1553  * thaw_super -- unlock filesystem
1554  * @sb: the super to thaw
1555  *
1556  * Unlocks the filesystem and marks it writeable again after freeze_super().
1557  */
1558 static int thaw_super_locked(struct super_block *sb)
1559 {
1560 	int error;
1561 
1562 	if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1563 		up_write(&sb->s_umount);
1564 		return -EINVAL;
1565 	}
1566 
1567 	if (sb_rdonly(sb)) {
1568 		sb->s_writers.frozen = SB_UNFROZEN;
1569 		goto out;
1570 	}
1571 
1572 	lockdep_sb_freeze_acquire(sb);
1573 
1574 	if (sb->s_op->unfreeze_fs) {
1575 		error = sb->s_op->unfreeze_fs(sb);
1576 		if (error) {
1577 			printk(KERN_ERR
1578 				"VFS:Filesystem thaw failed\n");
1579 			lockdep_sb_freeze_release(sb);
1580 			up_write(&sb->s_umount);
1581 			return error;
1582 		}
1583 	}
1584 
1585 	sb->s_writers.frozen = SB_UNFROZEN;
1586 	sb_freeze_unlock(sb);
1587 out:
1588 	wake_up(&sb->s_writers.wait_unfrozen);
1589 	deactivate_locked_super(sb);
1590 	return 0;
1591 }
1592 
1593 int thaw_super(struct super_block *sb)
1594 {
1595 	down_write(&sb->s_umount);
1596 	return thaw_super_locked(sb);
1597 }
1598 EXPORT_SYMBOL(thaw_super);
1599