xref: /openbmc/linux/fs/super.c (revision ff7b437f)
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/fscrypt.h>
36 #include <linux/fsnotify.h>
37 #include <linux/lockdep.h>
38 #include <linux/user_namespace.h>
39 #include <linux/fs_context.h>
40 #include <uapi/linux/mount.h>
41 #include "internal.h"
42 
43 static int thaw_super_locked(struct super_block *sb);
44 
45 static LIST_HEAD(super_blocks);
46 static DEFINE_SPINLOCK(sb_lock);
47 
48 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
49 	"sb_writers",
50 	"sb_pagefaults",
51 	"sb_internal",
52 };
53 
54 /*
55  * One thing we have to be careful of with a per-sb shrinker is that we don't
56  * drop the last active reference to the superblock from within the shrinker.
57  * If that happens we could trigger unregistering the shrinker from within the
58  * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
59  * take a passive reference to the superblock to avoid this from occurring.
60  */
61 static unsigned long super_cache_scan(struct shrinker *shrink,
62 				      struct shrink_control *sc)
63 {
64 	struct super_block *sb;
65 	long	fs_objects = 0;
66 	long	total_objects;
67 	long	freed = 0;
68 	long	dentries;
69 	long	inodes;
70 
71 	sb = container_of(shrink, struct super_block, s_shrink);
72 
73 	/*
74 	 * Deadlock avoidance.  We may hold various FS locks, and we don't want
75 	 * to recurse into the FS that called us in clear_inode() and friends..
76 	 */
77 	if (!(sc->gfp_mask & __GFP_FS))
78 		return SHRINK_STOP;
79 
80 	if (!trylock_super(sb))
81 		return SHRINK_STOP;
82 
83 	if (sb->s_op->nr_cached_objects)
84 		fs_objects = sb->s_op->nr_cached_objects(sb, sc);
85 
86 	inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
87 	dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
88 	total_objects = dentries + inodes + fs_objects + 1;
89 	if (!total_objects)
90 		total_objects = 1;
91 
92 	/* proportion the scan between the caches */
93 	dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
94 	inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
95 	fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
96 
97 	/*
98 	 * prune the dcache first as the icache is pinned by it, then
99 	 * prune the icache, followed by the filesystem specific caches
100 	 *
101 	 * Ensure that we always scan at least one object - memcg kmem
102 	 * accounting uses this to fully empty the caches.
103 	 */
104 	sc->nr_to_scan = dentries + 1;
105 	freed = prune_dcache_sb(sb, sc);
106 	sc->nr_to_scan = inodes + 1;
107 	freed += prune_icache_sb(sb, sc);
108 
109 	if (fs_objects) {
110 		sc->nr_to_scan = fs_objects + 1;
111 		freed += sb->s_op->free_cached_objects(sb, sc);
112 	}
113 
114 	up_read(&sb->s_umount);
115 	return freed;
116 }
117 
118 static unsigned long super_cache_count(struct shrinker *shrink,
119 				       struct shrink_control *sc)
120 {
121 	struct super_block *sb;
122 	long	total_objects = 0;
123 
124 	sb = container_of(shrink, struct super_block, s_shrink);
125 
126 	/*
127 	 * We don't call trylock_super() here as it is a scalability bottleneck,
128 	 * so we're exposed to partial setup state. The shrinker rwsem does not
129 	 * protect filesystem operations backing list_lru_shrink_count() or
130 	 * s_op->nr_cached_objects(). Counts can change between
131 	 * super_cache_count and super_cache_scan, so we really don't need locks
132 	 * here.
133 	 *
134 	 * However, if we are currently mounting the superblock, the underlying
135 	 * filesystem might be in a state of partial construction and hence it
136 	 * is dangerous to access it.  trylock_super() uses a SB_BORN check to
137 	 * avoid this situation, so do the same here. The memory barrier is
138 	 * matched with the one in mount_fs() as we don't hold locks here.
139 	 */
140 	if (!(sb->s_flags & SB_BORN))
141 		return 0;
142 	smp_rmb();
143 
144 	if (sb->s_op && sb->s_op->nr_cached_objects)
145 		total_objects = sb->s_op->nr_cached_objects(sb, sc);
146 
147 	total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
148 	total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
149 
150 	if (!total_objects)
151 		return SHRINK_EMPTY;
152 
153 	total_objects = vfs_pressure_ratio(total_objects);
154 	return total_objects;
155 }
156 
157 static void destroy_super_work(struct work_struct *work)
158 {
159 	struct super_block *s = container_of(work, struct super_block,
160 							destroy_work);
161 	int i;
162 
163 	for (i = 0; i < SB_FREEZE_LEVELS; i++)
164 		percpu_free_rwsem(&s->s_writers.rw_sem[i]);
165 	kfree(s);
166 }
167 
168 static void destroy_super_rcu(struct rcu_head *head)
169 {
170 	struct super_block *s = container_of(head, struct super_block, rcu);
171 	INIT_WORK(&s->destroy_work, destroy_super_work);
172 	schedule_work(&s->destroy_work);
173 }
174 
175 /* Free a superblock that has never been seen by anyone */
176 static void destroy_unused_super(struct super_block *s)
177 {
178 	if (!s)
179 		return;
180 	up_write(&s->s_umount);
181 	list_lru_destroy(&s->s_dentry_lru);
182 	list_lru_destroy(&s->s_inode_lru);
183 	security_sb_free(s);
184 	put_user_ns(s->s_user_ns);
185 	kfree(s->s_subtype);
186 	free_prealloced_shrinker(&s->s_shrink);
187 	/* no delays needed */
188 	destroy_super_work(&s->destroy_work);
189 }
190 
191 /**
192  *	alloc_super	-	create new superblock
193  *	@type:	filesystem type superblock should belong to
194  *	@flags: the mount flags
195  *	@user_ns: User namespace for the super_block
196  *
197  *	Allocates and initializes a new &struct super_block.  alloc_super()
198  *	returns a pointer new superblock or %NULL if allocation had failed.
199  */
200 static struct super_block *alloc_super(struct file_system_type *type, int flags,
201 				       struct user_namespace *user_ns)
202 {
203 	struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
204 	static const struct super_operations default_op;
205 	int i;
206 
207 	if (!s)
208 		return NULL;
209 
210 	INIT_LIST_HEAD(&s->s_mounts);
211 	s->s_user_ns = get_user_ns(user_ns);
212 	init_rwsem(&s->s_umount);
213 	lockdep_set_class(&s->s_umount, &type->s_umount_key);
214 	/*
215 	 * sget() can have s_umount recursion.
216 	 *
217 	 * When it cannot find a suitable sb, it allocates a new
218 	 * one (this one), and tries again to find a suitable old
219 	 * one.
220 	 *
221 	 * In case that succeeds, it will acquire the s_umount
222 	 * lock of the old one. Since these are clearly distrinct
223 	 * locks, and this object isn't exposed yet, there's no
224 	 * risk of deadlocks.
225 	 *
226 	 * Annotate this by putting this lock in a different
227 	 * subclass.
228 	 */
229 	down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
230 
231 	if (security_sb_alloc(s))
232 		goto fail;
233 
234 	for (i = 0; i < SB_FREEZE_LEVELS; i++) {
235 		if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
236 					sb_writers_name[i],
237 					&type->s_writers_key[i]))
238 			goto fail;
239 	}
240 	init_waitqueue_head(&s->s_writers.wait_unfrozen);
241 	s->s_bdi = &noop_backing_dev_info;
242 	s->s_flags = flags;
243 	if (s->s_user_ns != &init_user_ns)
244 		s->s_iflags |= SB_I_NODEV;
245 	INIT_HLIST_NODE(&s->s_instances);
246 	INIT_HLIST_BL_HEAD(&s->s_roots);
247 	mutex_init(&s->s_sync_lock);
248 	INIT_LIST_HEAD(&s->s_inodes);
249 	spin_lock_init(&s->s_inode_list_lock);
250 	INIT_LIST_HEAD(&s->s_inodes_wb);
251 	spin_lock_init(&s->s_inode_wblist_lock);
252 
253 	s->s_count = 1;
254 	atomic_set(&s->s_active, 1);
255 	mutex_init(&s->s_vfs_rename_mutex);
256 	lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
257 	init_rwsem(&s->s_dquot.dqio_sem);
258 	s->s_maxbytes = MAX_NON_LFS;
259 	s->s_op = &default_op;
260 	s->s_time_gran = 1000000000;
261 	s->s_time_min = TIME64_MIN;
262 	s->s_time_max = TIME64_MAX;
263 	s->cleancache_poolid = CLEANCACHE_NO_POOL;
264 
265 	s->s_shrink.seeks = DEFAULT_SEEKS;
266 	s->s_shrink.scan_objects = super_cache_scan;
267 	s->s_shrink.count_objects = super_cache_count;
268 	s->s_shrink.batch = 1024;
269 	s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
270 	if (prealloc_shrinker(&s->s_shrink))
271 		goto fail;
272 	if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
273 		goto fail;
274 	if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
275 		goto fail;
276 	return s;
277 
278 fail:
279 	destroy_unused_super(s);
280 	return NULL;
281 }
282 
283 /* Superblock refcounting  */
284 
285 /*
286  * Drop a superblock's refcount.  The caller must hold sb_lock.
287  */
288 static void __put_super(struct super_block *s)
289 {
290 	if (!--s->s_count) {
291 		list_del_init(&s->s_list);
292 		WARN_ON(s->s_dentry_lru.node);
293 		WARN_ON(s->s_inode_lru.node);
294 		WARN_ON(!list_empty(&s->s_mounts));
295 		security_sb_free(s);
296 		fscrypt_sb_free(s);
297 		put_user_ns(s->s_user_ns);
298 		kfree(s->s_subtype);
299 		call_rcu(&s->rcu, destroy_super_rcu);
300 	}
301 }
302 
303 /**
304  *	put_super	-	drop a temporary reference to superblock
305  *	@sb: superblock in question
306  *
307  *	Drops a temporary reference, frees superblock if there's no
308  *	references left.
309  */
310 static void put_super(struct super_block *sb)
311 {
312 	spin_lock(&sb_lock);
313 	__put_super(sb);
314 	spin_unlock(&sb_lock);
315 }
316 
317 
318 /**
319  *	deactivate_locked_super	-	drop an active reference to superblock
320  *	@s: superblock to deactivate
321  *
322  *	Drops an active reference to superblock, converting it into a temporary
323  *	one if there is no other active references left.  In that case we
324  *	tell fs driver to shut it down and drop the temporary reference we
325  *	had just acquired.
326  *
327  *	Caller holds exclusive lock on superblock; that lock is released.
328  */
329 void deactivate_locked_super(struct super_block *s)
330 {
331 	struct file_system_type *fs = s->s_type;
332 	if (atomic_dec_and_test(&s->s_active)) {
333 		cleancache_invalidate_fs(s);
334 		unregister_shrinker(&s->s_shrink);
335 		fs->kill_sb(s);
336 
337 		/*
338 		 * Since list_lru_destroy() may sleep, we cannot call it from
339 		 * put_super(), where we hold the sb_lock. Therefore we destroy
340 		 * the lru lists right now.
341 		 */
342 		list_lru_destroy(&s->s_dentry_lru);
343 		list_lru_destroy(&s->s_inode_lru);
344 
345 		put_filesystem(fs);
346 		put_super(s);
347 	} else {
348 		up_write(&s->s_umount);
349 	}
350 }
351 
352 EXPORT_SYMBOL(deactivate_locked_super);
353 
354 /**
355  *	deactivate_super	-	drop an active reference to superblock
356  *	@s: superblock to deactivate
357  *
358  *	Variant of deactivate_locked_super(), except that superblock is *not*
359  *	locked by caller.  If we are going to drop the final active reference,
360  *	lock will be acquired prior to that.
361  */
362 void deactivate_super(struct super_block *s)
363 {
364         if (!atomic_add_unless(&s->s_active, -1, 1)) {
365 		down_write(&s->s_umount);
366 		deactivate_locked_super(s);
367 	}
368 }
369 
370 EXPORT_SYMBOL(deactivate_super);
371 
372 /**
373  *	grab_super - acquire an active reference
374  *	@s: reference we are trying to make active
375  *
376  *	Tries to acquire an active reference.  grab_super() is used when we
377  * 	had just found a superblock in super_blocks or fs_type->fs_supers
378  *	and want to turn it into a full-blown active reference.  grab_super()
379  *	is called with sb_lock held and drops it.  Returns 1 in case of
380  *	success, 0 if we had failed (superblock contents was already dead or
381  *	dying when grab_super() had been called).  Note that this is only
382  *	called for superblocks not in rundown mode (== ones still on ->fs_supers
383  *	of their type), so increment of ->s_count is OK here.
384  */
385 static int grab_super(struct super_block *s) __releases(sb_lock)
386 {
387 	s->s_count++;
388 	spin_unlock(&sb_lock);
389 	down_write(&s->s_umount);
390 	if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
391 		put_super(s);
392 		return 1;
393 	}
394 	up_write(&s->s_umount);
395 	put_super(s);
396 	return 0;
397 }
398 
399 /*
400  *	trylock_super - try to grab ->s_umount shared
401  *	@sb: reference we are trying to grab
402  *
403  *	Try to prevent fs shutdown.  This is used in places where we
404  *	cannot take an active reference but we need to ensure that the
405  *	filesystem is not shut down while we are working on it. It returns
406  *	false if we cannot acquire s_umount or if we lose the race and
407  *	filesystem already got into shutdown, and returns true with the s_umount
408  *	lock held in read mode in case of success. On successful return,
409  *	the caller must drop the s_umount lock when done.
410  *
411  *	Note that unlike get_super() et.al. this one does *not* bump ->s_count.
412  *	The reason why it's safe is that we are OK with doing trylock instead
413  *	of down_read().  There's a couple of places that are OK with that, but
414  *	it's very much not a general-purpose interface.
415  */
416 bool trylock_super(struct super_block *sb)
417 {
418 	if (down_read_trylock(&sb->s_umount)) {
419 		if (!hlist_unhashed(&sb->s_instances) &&
420 		    sb->s_root && (sb->s_flags & SB_BORN))
421 			return true;
422 		up_read(&sb->s_umount);
423 	}
424 
425 	return false;
426 }
427 
428 /**
429  *	generic_shutdown_super	-	common helper for ->kill_sb()
430  *	@sb: superblock to kill
431  *
432  *	generic_shutdown_super() does all fs-independent work on superblock
433  *	shutdown.  Typical ->kill_sb() should pick all fs-specific objects
434  *	that need destruction out of superblock, call generic_shutdown_super()
435  *	and release aforementioned objects.  Note: dentries and inodes _are_
436  *	taken care of and do not need specific handling.
437  *
438  *	Upon calling this function, the filesystem may no longer alter or
439  *	rearrange the set of dentries belonging to this super_block, nor may it
440  *	change the attachments of dentries to inodes.
441  */
442 void generic_shutdown_super(struct super_block *sb)
443 {
444 	const struct super_operations *sop = sb->s_op;
445 
446 	if (sb->s_root) {
447 		shrink_dcache_for_umount(sb);
448 		sync_filesystem(sb);
449 		sb->s_flags &= ~SB_ACTIVE;
450 
451 		fsnotify_sb_delete(sb);
452 		cgroup_writeback_umount();
453 
454 		evict_inodes(sb);
455 
456 		if (sb->s_dio_done_wq) {
457 			destroy_workqueue(sb->s_dio_done_wq);
458 			sb->s_dio_done_wq = NULL;
459 		}
460 
461 		if (sop->put_super)
462 			sop->put_super(sb);
463 
464 		if (!list_empty(&sb->s_inodes)) {
465 			printk("VFS: Busy inodes after unmount of %s. "
466 			   "Self-destruct in 5 seconds.  Have a nice day...\n",
467 			   sb->s_id);
468 		}
469 	}
470 	spin_lock(&sb_lock);
471 	/* should be initialized for __put_super_and_need_restart() */
472 	hlist_del_init(&sb->s_instances);
473 	spin_unlock(&sb_lock);
474 	up_write(&sb->s_umount);
475 	if (sb->s_bdi != &noop_backing_dev_info) {
476 		bdi_put(sb->s_bdi);
477 		sb->s_bdi = &noop_backing_dev_info;
478 	}
479 }
480 
481 EXPORT_SYMBOL(generic_shutdown_super);
482 
483 bool mount_capable(struct fs_context *fc)
484 {
485 	if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
486 		return capable(CAP_SYS_ADMIN);
487 	else
488 		return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
489 }
490 
491 /**
492  * sget_fc - Find or create a superblock
493  * @fc:	Filesystem context.
494  * @test: Comparison callback
495  * @set: Setup callback
496  *
497  * Find or create a superblock using the parameters stored in the filesystem
498  * context and the two callback functions.
499  *
500  * If an extant superblock is matched, then that will be returned with an
501  * elevated reference count that the caller must transfer or discard.
502  *
503  * If no match is made, a new superblock will be allocated and basic
504  * initialisation will be performed (s_type, s_fs_info and s_id will be set and
505  * the set() callback will be invoked), the superblock will be published and it
506  * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
507  * as yet unset.
508  */
509 struct super_block *sget_fc(struct fs_context *fc,
510 			    int (*test)(struct super_block *, struct fs_context *),
511 			    int (*set)(struct super_block *, struct fs_context *))
512 {
513 	struct super_block *s = NULL;
514 	struct super_block *old;
515 	struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
516 	int err;
517 
518 retry:
519 	spin_lock(&sb_lock);
520 	if (test) {
521 		hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
522 			if (test(old, fc))
523 				goto share_extant_sb;
524 		}
525 	}
526 	if (!s) {
527 		spin_unlock(&sb_lock);
528 		s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
529 		if (!s)
530 			return ERR_PTR(-ENOMEM);
531 		goto retry;
532 	}
533 
534 	s->s_fs_info = fc->s_fs_info;
535 	err = set(s, fc);
536 	if (err) {
537 		s->s_fs_info = NULL;
538 		spin_unlock(&sb_lock);
539 		destroy_unused_super(s);
540 		return ERR_PTR(err);
541 	}
542 	fc->s_fs_info = NULL;
543 	s->s_type = fc->fs_type;
544 	s->s_iflags |= fc->s_iflags;
545 	strlcpy(s->s_id, s->s_type->name, sizeof(s->s_id));
546 	list_add_tail(&s->s_list, &super_blocks);
547 	hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
548 	spin_unlock(&sb_lock);
549 	get_filesystem(s->s_type);
550 	register_shrinker_prepared(&s->s_shrink);
551 	return s;
552 
553 share_extant_sb:
554 	if (user_ns != old->s_user_ns) {
555 		spin_unlock(&sb_lock);
556 		destroy_unused_super(s);
557 		return ERR_PTR(-EBUSY);
558 	}
559 	if (!grab_super(old))
560 		goto retry;
561 	destroy_unused_super(s);
562 	return old;
563 }
564 EXPORT_SYMBOL(sget_fc);
565 
566 /**
567  *	sget	-	find or create a superblock
568  *	@type:	  filesystem type superblock should belong to
569  *	@test:	  comparison callback
570  *	@set:	  setup callback
571  *	@flags:	  mount flags
572  *	@data:	  argument to each of them
573  */
574 struct super_block *sget(struct file_system_type *type,
575 			int (*test)(struct super_block *,void *),
576 			int (*set)(struct super_block *,void *),
577 			int flags,
578 			void *data)
579 {
580 	struct user_namespace *user_ns = current_user_ns();
581 	struct super_block *s = NULL;
582 	struct super_block *old;
583 	int err;
584 
585 	/* We don't yet pass the user namespace of the parent
586 	 * mount through to here so always use &init_user_ns
587 	 * until that changes.
588 	 */
589 	if (flags & SB_SUBMOUNT)
590 		user_ns = &init_user_ns;
591 
592 retry:
593 	spin_lock(&sb_lock);
594 	if (test) {
595 		hlist_for_each_entry(old, &type->fs_supers, s_instances) {
596 			if (!test(old, data))
597 				continue;
598 			if (user_ns != old->s_user_ns) {
599 				spin_unlock(&sb_lock);
600 				destroy_unused_super(s);
601 				return ERR_PTR(-EBUSY);
602 			}
603 			if (!grab_super(old))
604 				goto retry;
605 			destroy_unused_super(s);
606 			return old;
607 		}
608 	}
609 	if (!s) {
610 		spin_unlock(&sb_lock);
611 		s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
612 		if (!s)
613 			return ERR_PTR(-ENOMEM);
614 		goto retry;
615 	}
616 
617 	err = set(s, data);
618 	if (err) {
619 		spin_unlock(&sb_lock);
620 		destroy_unused_super(s);
621 		return ERR_PTR(err);
622 	}
623 	s->s_type = type;
624 	strlcpy(s->s_id, type->name, sizeof(s->s_id));
625 	list_add_tail(&s->s_list, &super_blocks);
626 	hlist_add_head(&s->s_instances, &type->fs_supers);
627 	spin_unlock(&sb_lock);
628 	get_filesystem(type);
629 	register_shrinker_prepared(&s->s_shrink);
630 	return s;
631 }
632 EXPORT_SYMBOL(sget);
633 
634 void drop_super(struct super_block *sb)
635 {
636 	up_read(&sb->s_umount);
637 	put_super(sb);
638 }
639 
640 EXPORT_SYMBOL(drop_super);
641 
642 void drop_super_exclusive(struct super_block *sb)
643 {
644 	up_write(&sb->s_umount);
645 	put_super(sb);
646 }
647 EXPORT_SYMBOL(drop_super_exclusive);
648 
649 static void __iterate_supers(void (*f)(struct super_block *))
650 {
651 	struct super_block *sb, *p = NULL;
652 
653 	spin_lock(&sb_lock);
654 	list_for_each_entry(sb, &super_blocks, s_list) {
655 		if (hlist_unhashed(&sb->s_instances))
656 			continue;
657 		sb->s_count++;
658 		spin_unlock(&sb_lock);
659 
660 		f(sb);
661 
662 		spin_lock(&sb_lock);
663 		if (p)
664 			__put_super(p);
665 		p = sb;
666 	}
667 	if (p)
668 		__put_super(p);
669 	spin_unlock(&sb_lock);
670 }
671 /**
672  *	iterate_supers - call function for all active superblocks
673  *	@f: function to call
674  *	@arg: argument to pass to it
675  *
676  *	Scans the superblock list and calls given function, passing it
677  *	locked superblock and given argument.
678  */
679 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
680 {
681 	struct super_block *sb, *p = NULL;
682 
683 	spin_lock(&sb_lock);
684 	list_for_each_entry(sb, &super_blocks, s_list) {
685 		if (hlist_unhashed(&sb->s_instances))
686 			continue;
687 		sb->s_count++;
688 		spin_unlock(&sb_lock);
689 
690 		down_read(&sb->s_umount);
691 		if (sb->s_root && (sb->s_flags & SB_BORN))
692 			f(sb, arg);
693 		up_read(&sb->s_umount);
694 
695 		spin_lock(&sb_lock);
696 		if (p)
697 			__put_super(p);
698 		p = sb;
699 	}
700 	if (p)
701 		__put_super(p);
702 	spin_unlock(&sb_lock);
703 }
704 
705 /**
706  *	iterate_supers_type - call function for superblocks of given type
707  *	@type: fs type
708  *	@f: function to call
709  *	@arg: argument to pass to it
710  *
711  *	Scans the superblock list and calls given function, passing it
712  *	locked superblock and given argument.
713  */
714 void iterate_supers_type(struct file_system_type *type,
715 	void (*f)(struct super_block *, void *), void *arg)
716 {
717 	struct super_block *sb, *p = NULL;
718 
719 	spin_lock(&sb_lock);
720 	hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
721 		sb->s_count++;
722 		spin_unlock(&sb_lock);
723 
724 		down_read(&sb->s_umount);
725 		if (sb->s_root && (sb->s_flags & SB_BORN))
726 			f(sb, arg);
727 		up_read(&sb->s_umount);
728 
729 		spin_lock(&sb_lock);
730 		if (p)
731 			__put_super(p);
732 		p = sb;
733 	}
734 	if (p)
735 		__put_super(p);
736 	spin_unlock(&sb_lock);
737 }
738 
739 EXPORT_SYMBOL(iterate_supers_type);
740 
741 static struct super_block *__get_super(struct block_device *bdev, bool excl)
742 {
743 	struct super_block *sb;
744 
745 	if (!bdev)
746 		return NULL;
747 
748 	spin_lock(&sb_lock);
749 rescan:
750 	list_for_each_entry(sb, &super_blocks, s_list) {
751 		if (hlist_unhashed(&sb->s_instances))
752 			continue;
753 		if (sb->s_bdev == bdev) {
754 			sb->s_count++;
755 			spin_unlock(&sb_lock);
756 			if (!excl)
757 				down_read(&sb->s_umount);
758 			else
759 				down_write(&sb->s_umount);
760 			/* still alive? */
761 			if (sb->s_root && (sb->s_flags & SB_BORN))
762 				return sb;
763 			if (!excl)
764 				up_read(&sb->s_umount);
765 			else
766 				up_write(&sb->s_umount);
767 			/* nope, got unmounted */
768 			spin_lock(&sb_lock);
769 			__put_super(sb);
770 			goto rescan;
771 		}
772 	}
773 	spin_unlock(&sb_lock);
774 	return NULL;
775 }
776 
777 /**
778  *	get_super - get the superblock of a device
779  *	@bdev: device to get the superblock for
780  *
781  *	Scans the superblock list and finds the superblock of the file system
782  *	mounted on the device given. %NULL is returned if no match is found.
783  */
784 struct super_block *get_super(struct block_device *bdev)
785 {
786 	return __get_super(bdev, false);
787 }
788 EXPORT_SYMBOL(get_super);
789 
790 static struct super_block *__get_super_thawed(struct block_device *bdev,
791 					      bool excl)
792 {
793 	while (1) {
794 		struct super_block *s = __get_super(bdev, excl);
795 		if (!s || s->s_writers.frozen == SB_UNFROZEN)
796 			return s;
797 		if (!excl)
798 			up_read(&s->s_umount);
799 		else
800 			up_write(&s->s_umount);
801 		wait_event(s->s_writers.wait_unfrozen,
802 			   s->s_writers.frozen == SB_UNFROZEN);
803 		put_super(s);
804 	}
805 }
806 
807 /**
808  *	get_super_thawed - get thawed superblock of a device
809  *	@bdev: device to get the superblock for
810  *
811  *	Scans the superblock list and finds the superblock of the file system
812  *	mounted on the device. The superblock is returned once it is thawed
813  *	(or immediately if it was not frozen). %NULL is returned if no match
814  *	is found.
815  */
816 struct super_block *get_super_thawed(struct block_device *bdev)
817 {
818 	return __get_super_thawed(bdev, false);
819 }
820 EXPORT_SYMBOL(get_super_thawed);
821 
822 /**
823  *	get_super_exclusive_thawed - get thawed superblock of a device
824  *	@bdev: device to get the superblock for
825  *
826  *	Scans the superblock list and finds the superblock of the file system
827  *	mounted on the device. The superblock is returned once it is thawed
828  *	(or immediately if it was not frozen) and s_umount semaphore is held
829  *	in exclusive mode. %NULL is returned if no match is found.
830  */
831 struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
832 {
833 	return __get_super_thawed(bdev, true);
834 }
835 EXPORT_SYMBOL(get_super_exclusive_thawed);
836 
837 /**
838  * get_active_super - get an active reference to the superblock of a device
839  * @bdev: device to get the superblock for
840  *
841  * Scans the superblock list and finds the superblock of the file system
842  * mounted on the device given.  Returns the superblock with an active
843  * reference or %NULL if none was found.
844  */
845 struct super_block *get_active_super(struct block_device *bdev)
846 {
847 	struct super_block *sb;
848 
849 	if (!bdev)
850 		return NULL;
851 
852 restart:
853 	spin_lock(&sb_lock);
854 	list_for_each_entry(sb, &super_blocks, s_list) {
855 		if (hlist_unhashed(&sb->s_instances))
856 			continue;
857 		if (sb->s_bdev == bdev) {
858 			if (!grab_super(sb))
859 				goto restart;
860 			up_write(&sb->s_umount);
861 			return sb;
862 		}
863 	}
864 	spin_unlock(&sb_lock);
865 	return NULL;
866 }
867 
868 struct super_block *user_get_super(dev_t dev)
869 {
870 	struct super_block *sb;
871 
872 	spin_lock(&sb_lock);
873 rescan:
874 	list_for_each_entry(sb, &super_blocks, s_list) {
875 		if (hlist_unhashed(&sb->s_instances))
876 			continue;
877 		if (sb->s_dev ==  dev) {
878 			sb->s_count++;
879 			spin_unlock(&sb_lock);
880 			down_read(&sb->s_umount);
881 			/* still alive? */
882 			if (sb->s_root && (sb->s_flags & SB_BORN))
883 				return sb;
884 			up_read(&sb->s_umount);
885 			/* nope, got unmounted */
886 			spin_lock(&sb_lock);
887 			__put_super(sb);
888 			goto rescan;
889 		}
890 	}
891 	spin_unlock(&sb_lock);
892 	return NULL;
893 }
894 
895 /**
896  * reconfigure_super - asks filesystem to change superblock parameters
897  * @fc: The superblock and configuration
898  *
899  * Alters the configuration parameters of a live superblock.
900  */
901 int reconfigure_super(struct fs_context *fc)
902 {
903 	struct super_block *sb = fc->root->d_sb;
904 	int retval;
905 	bool remount_ro = false;
906 	bool force = fc->sb_flags & SB_FORCE;
907 
908 	if (fc->sb_flags_mask & ~MS_RMT_MASK)
909 		return -EINVAL;
910 	if (sb->s_writers.frozen != SB_UNFROZEN)
911 		return -EBUSY;
912 
913 	retval = security_sb_remount(sb, fc->security);
914 	if (retval)
915 		return retval;
916 
917 	if (fc->sb_flags_mask & SB_RDONLY) {
918 #ifdef CONFIG_BLOCK
919 		if (!(fc->sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
920 			return -EACCES;
921 #endif
922 
923 		remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
924 	}
925 
926 	if (remount_ro) {
927 		if (!hlist_empty(&sb->s_pins)) {
928 			up_write(&sb->s_umount);
929 			group_pin_kill(&sb->s_pins);
930 			down_write(&sb->s_umount);
931 			if (!sb->s_root)
932 				return 0;
933 			if (sb->s_writers.frozen != SB_UNFROZEN)
934 				return -EBUSY;
935 			remount_ro = !sb_rdonly(sb);
936 		}
937 	}
938 	shrink_dcache_sb(sb);
939 
940 	/* If we are reconfiguring to RDONLY and current sb is read/write,
941 	 * make sure there are no files open for writing.
942 	 */
943 	if (remount_ro) {
944 		if (force) {
945 			sb->s_readonly_remount = 1;
946 			smp_wmb();
947 		} else {
948 			retval = sb_prepare_remount_readonly(sb);
949 			if (retval)
950 				return retval;
951 		}
952 	}
953 
954 	if (fc->ops->reconfigure) {
955 		retval = fc->ops->reconfigure(fc);
956 		if (retval) {
957 			if (!force)
958 				goto cancel_readonly;
959 			/* If forced remount, go ahead despite any errors */
960 			WARN(1, "forced remount of a %s fs returned %i\n",
961 			     sb->s_type->name, retval);
962 		}
963 	}
964 
965 	WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
966 				 (fc->sb_flags & fc->sb_flags_mask)));
967 	/* Needs to be ordered wrt mnt_is_readonly() */
968 	smp_wmb();
969 	sb->s_readonly_remount = 0;
970 
971 	/*
972 	 * Some filesystems modify their metadata via some other path than the
973 	 * bdev buffer cache (eg. use a private mapping, or directories in
974 	 * pagecache, etc). Also file data modifications go via their own
975 	 * mappings. So If we try to mount readonly then copy the filesystem
976 	 * from bdev, we could get stale data, so invalidate it to give a best
977 	 * effort at coherency.
978 	 */
979 	if (remount_ro && sb->s_bdev)
980 		invalidate_bdev(sb->s_bdev);
981 	return 0;
982 
983 cancel_readonly:
984 	sb->s_readonly_remount = 0;
985 	return retval;
986 }
987 
988 static void do_emergency_remount_callback(struct super_block *sb)
989 {
990 	down_write(&sb->s_umount);
991 	if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
992 	    !sb_rdonly(sb)) {
993 		struct fs_context *fc;
994 
995 		fc = fs_context_for_reconfigure(sb->s_root,
996 					SB_RDONLY | SB_FORCE, SB_RDONLY);
997 		if (!IS_ERR(fc)) {
998 			if (parse_monolithic_mount_data(fc, NULL) == 0)
999 				(void)reconfigure_super(fc);
1000 			put_fs_context(fc);
1001 		}
1002 	}
1003 	up_write(&sb->s_umount);
1004 }
1005 
1006 static void do_emergency_remount(struct work_struct *work)
1007 {
1008 	__iterate_supers(do_emergency_remount_callback);
1009 	kfree(work);
1010 	printk("Emergency Remount complete\n");
1011 }
1012 
1013 void emergency_remount(void)
1014 {
1015 	struct work_struct *work;
1016 
1017 	work = kmalloc(sizeof(*work), GFP_ATOMIC);
1018 	if (work) {
1019 		INIT_WORK(work, do_emergency_remount);
1020 		schedule_work(work);
1021 	}
1022 }
1023 
1024 static void do_thaw_all_callback(struct super_block *sb)
1025 {
1026 	down_write(&sb->s_umount);
1027 	if (sb->s_root && sb->s_flags & SB_BORN) {
1028 		emergency_thaw_bdev(sb);
1029 		thaw_super_locked(sb);
1030 	} else {
1031 		up_write(&sb->s_umount);
1032 	}
1033 }
1034 
1035 static void do_thaw_all(struct work_struct *work)
1036 {
1037 	__iterate_supers(do_thaw_all_callback);
1038 	kfree(work);
1039 	printk(KERN_WARNING "Emergency Thaw complete\n");
1040 }
1041 
1042 /**
1043  * emergency_thaw_all -- forcibly thaw every frozen filesystem
1044  *
1045  * Used for emergency unfreeze of all filesystems via SysRq
1046  */
1047 void emergency_thaw_all(void)
1048 {
1049 	struct work_struct *work;
1050 
1051 	work = kmalloc(sizeof(*work), GFP_ATOMIC);
1052 	if (work) {
1053 		INIT_WORK(work, do_thaw_all);
1054 		schedule_work(work);
1055 	}
1056 }
1057 
1058 static DEFINE_IDA(unnamed_dev_ida);
1059 
1060 /**
1061  * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1062  * @p: Pointer to a dev_t.
1063  *
1064  * Filesystems which don't use real block devices can call this function
1065  * to allocate a virtual block device.
1066  *
1067  * Context: Any context.  Frequently called while holding sb_lock.
1068  * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1069  * or -ENOMEM if memory allocation failed.
1070  */
1071 int get_anon_bdev(dev_t *p)
1072 {
1073 	int dev;
1074 
1075 	/*
1076 	 * Many userspace utilities consider an FSID of 0 invalid.
1077 	 * Always return at least 1 from get_anon_bdev.
1078 	 */
1079 	dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1080 			GFP_ATOMIC);
1081 	if (dev == -ENOSPC)
1082 		dev = -EMFILE;
1083 	if (dev < 0)
1084 		return dev;
1085 
1086 	*p = MKDEV(0, dev);
1087 	return 0;
1088 }
1089 EXPORT_SYMBOL(get_anon_bdev);
1090 
1091 void free_anon_bdev(dev_t dev)
1092 {
1093 	ida_free(&unnamed_dev_ida, MINOR(dev));
1094 }
1095 EXPORT_SYMBOL(free_anon_bdev);
1096 
1097 int set_anon_super(struct super_block *s, void *data)
1098 {
1099 	return get_anon_bdev(&s->s_dev);
1100 }
1101 EXPORT_SYMBOL(set_anon_super);
1102 
1103 void kill_anon_super(struct super_block *sb)
1104 {
1105 	dev_t dev = sb->s_dev;
1106 	generic_shutdown_super(sb);
1107 	free_anon_bdev(dev);
1108 }
1109 EXPORT_SYMBOL(kill_anon_super);
1110 
1111 void kill_litter_super(struct super_block *sb)
1112 {
1113 	if (sb->s_root)
1114 		d_genocide(sb->s_root);
1115 	kill_anon_super(sb);
1116 }
1117 EXPORT_SYMBOL(kill_litter_super);
1118 
1119 int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1120 {
1121 	return set_anon_super(sb, NULL);
1122 }
1123 EXPORT_SYMBOL(set_anon_super_fc);
1124 
1125 static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1126 {
1127 	return sb->s_fs_info == fc->s_fs_info;
1128 }
1129 
1130 static int test_single_super(struct super_block *s, struct fs_context *fc)
1131 {
1132 	return 1;
1133 }
1134 
1135 /**
1136  * vfs_get_super - Get a superblock with a search key set in s_fs_info.
1137  * @fc: The filesystem context holding the parameters
1138  * @keying: How to distinguish superblocks
1139  * @fill_super: Helper to initialise a new superblock
1140  *
1141  * Search for a superblock and create a new one if not found.  The search
1142  * criterion is controlled by @keying.  If the search fails, a new superblock
1143  * is created and @fill_super() is called to initialise it.
1144  *
1145  * @keying can take one of a number of values:
1146  *
1147  * (1) vfs_get_single_super - Only one superblock of this type may exist on the
1148  *     system.  This is typically used for special system filesystems.
1149  *
1150  * (2) vfs_get_keyed_super - Multiple superblocks may exist, but they must have
1151  *     distinct keys (where the key is in s_fs_info).  Searching for the same
1152  *     key again will turn up the superblock for that key.
1153  *
1154  * (3) vfs_get_independent_super - Multiple superblocks may exist and are
1155  *     unkeyed.  Each call will get a new superblock.
1156  *
1157  * A permissions check is made by sget_fc() unless we're getting a superblock
1158  * for a kernel-internal mount or a submount.
1159  */
1160 int vfs_get_super(struct fs_context *fc,
1161 		  enum vfs_get_super_keying keying,
1162 		  int (*fill_super)(struct super_block *sb,
1163 				    struct fs_context *fc))
1164 {
1165 	int (*test)(struct super_block *, struct fs_context *);
1166 	struct super_block *sb;
1167 	int err;
1168 
1169 	switch (keying) {
1170 	case vfs_get_single_super:
1171 	case vfs_get_single_reconf_super:
1172 		test = test_single_super;
1173 		break;
1174 	case vfs_get_keyed_super:
1175 		test = test_keyed_super;
1176 		break;
1177 	case vfs_get_independent_super:
1178 		test = NULL;
1179 		break;
1180 	default:
1181 		BUG();
1182 	}
1183 
1184 	sb = sget_fc(fc, test, set_anon_super_fc);
1185 	if (IS_ERR(sb))
1186 		return PTR_ERR(sb);
1187 
1188 	if (!sb->s_root) {
1189 		err = fill_super(sb, fc);
1190 		if (err)
1191 			goto error;
1192 
1193 		sb->s_flags |= SB_ACTIVE;
1194 		fc->root = dget(sb->s_root);
1195 	} else {
1196 		fc->root = dget(sb->s_root);
1197 		if (keying == vfs_get_single_reconf_super) {
1198 			err = reconfigure_super(fc);
1199 			if (err < 0) {
1200 				dput(fc->root);
1201 				fc->root = NULL;
1202 				goto error;
1203 			}
1204 		}
1205 	}
1206 
1207 	return 0;
1208 
1209 error:
1210 	deactivate_locked_super(sb);
1211 	return err;
1212 }
1213 EXPORT_SYMBOL(vfs_get_super);
1214 
1215 int get_tree_nodev(struct fs_context *fc,
1216 		  int (*fill_super)(struct super_block *sb,
1217 				    struct fs_context *fc))
1218 {
1219 	return vfs_get_super(fc, vfs_get_independent_super, fill_super);
1220 }
1221 EXPORT_SYMBOL(get_tree_nodev);
1222 
1223 int get_tree_single(struct fs_context *fc,
1224 		  int (*fill_super)(struct super_block *sb,
1225 				    struct fs_context *fc))
1226 {
1227 	return vfs_get_super(fc, vfs_get_single_super, fill_super);
1228 }
1229 EXPORT_SYMBOL(get_tree_single);
1230 
1231 int get_tree_single_reconf(struct fs_context *fc,
1232 		  int (*fill_super)(struct super_block *sb,
1233 				    struct fs_context *fc))
1234 {
1235 	return vfs_get_super(fc, vfs_get_single_reconf_super, fill_super);
1236 }
1237 EXPORT_SYMBOL(get_tree_single_reconf);
1238 
1239 int get_tree_keyed(struct fs_context *fc,
1240 		  int (*fill_super)(struct super_block *sb,
1241 				    struct fs_context *fc),
1242 		void *key)
1243 {
1244 	fc->s_fs_info = key;
1245 	return vfs_get_super(fc, vfs_get_keyed_super, fill_super);
1246 }
1247 EXPORT_SYMBOL(get_tree_keyed);
1248 
1249 #ifdef CONFIG_BLOCK
1250 
1251 static int set_bdev_super(struct super_block *s, void *data)
1252 {
1253 	s->s_bdev = data;
1254 	s->s_dev = s->s_bdev->bd_dev;
1255 	s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1256 
1257 	return 0;
1258 }
1259 
1260 static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1261 {
1262 	return set_bdev_super(s, fc->sget_key);
1263 }
1264 
1265 static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1266 {
1267 	return s->s_bdev == fc->sget_key;
1268 }
1269 
1270 /**
1271  * get_tree_bdev - Get a superblock based on a single block device
1272  * @fc: The filesystem context holding the parameters
1273  * @fill_super: Helper to initialise a new superblock
1274  */
1275 int get_tree_bdev(struct fs_context *fc,
1276 		int (*fill_super)(struct super_block *,
1277 				  struct fs_context *))
1278 {
1279 	struct block_device *bdev;
1280 	struct super_block *s;
1281 	fmode_t mode = FMODE_READ | FMODE_EXCL;
1282 	int error = 0;
1283 
1284 	if (!(fc->sb_flags & SB_RDONLY))
1285 		mode |= FMODE_WRITE;
1286 
1287 	if (!fc->source)
1288 		return invalf(fc, "No source specified");
1289 
1290 	bdev = blkdev_get_by_path(fc->source, mode, fc->fs_type);
1291 	if (IS_ERR(bdev)) {
1292 		errorf(fc, "%s: Can't open blockdev", fc->source);
1293 		return PTR_ERR(bdev);
1294 	}
1295 
1296 	/* Once the superblock is inserted into the list by sget_fc(), s_umount
1297 	 * will protect the lockfs code from trying to start a snapshot while
1298 	 * we are mounting
1299 	 */
1300 	mutex_lock(&bdev->bd_fsfreeze_mutex);
1301 	if (bdev->bd_fsfreeze_count > 0) {
1302 		mutex_unlock(&bdev->bd_fsfreeze_mutex);
1303 		warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1304 		return -EBUSY;
1305 	}
1306 
1307 	fc->sb_flags |= SB_NOSEC;
1308 	fc->sget_key = bdev;
1309 	s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc);
1310 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
1311 	if (IS_ERR(s))
1312 		return PTR_ERR(s);
1313 
1314 	if (s->s_root) {
1315 		/* Don't summarily change the RO/RW state. */
1316 		if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1317 			warnf(fc, "%pg: Can't mount, would change RO state", bdev);
1318 			deactivate_locked_super(s);
1319 			blkdev_put(bdev, mode);
1320 			return -EBUSY;
1321 		}
1322 
1323 		/*
1324 		 * s_umount nests inside bd_mutex during
1325 		 * __invalidate_device().  blkdev_put() acquires
1326 		 * bd_mutex and can't be called under s_umount.  Drop
1327 		 * s_umount temporarily.  This is safe as we're
1328 		 * holding an active reference.
1329 		 */
1330 		up_write(&s->s_umount);
1331 		blkdev_put(bdev, mode);
1332 		down_write(&s->s_umount);
1333 	} else {
1334 		s->s_mode = mode;
1335 		snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1336 		sb_set_blocksize(s, block_size(bdev));
1337 		error = fill_super(s, fc);
1338 		if (error) {
1339 			deactivate_locked_super(s);
1340 			return error;
1341 		}
1342 
1343 		s->s_flags |= SB_ACTIVE;
1344 		bdev->bd_super = s;
1345 	}
1346 
1347 	BUG_ON(fc->root);
1348 	fc->root = dget(s->s_root);
1349 	return 0;
1350 }
1351 EXPORT_SYMBOL(get_tree_bdev);
1352 
1353 static int test_bdev_super(struct super_block *s, void *data)
1354 {
1355 	return (void *)s->s_bdev == data;
1356 }
1357 
1358 struct dentry *mount_bdev(struct file_system_type *fs_type,
1359 	int flags, const char *dev_name, void *data,
1360 	int (*fill_super)(struct super_block *, void *, int))
1361 {
1362 	struct block_device *bdev;
1363 	struct super_block *s;
1364 	fmode_t mode = FMODE_READ | FMODE_EXCL;
1365 	int error = 0;
1366 
1367 	if (!(flags & SB_RDONLY))
1368 		mode |= FMODE_WRITE;
1369 
1370 	bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1371 	if (IS_ERR(bdev))
1372 		return ERR_CAST(bdev);
1373 
1374 	/*
1375 	 * once the super is inserted into the list by sget, s_umount
1376 	 * will protect the lockfs code from trying to start a snapshot
1377 	 * while we are mounting
1378 	 */
1379 	mutex_lock(&bdev->bd_fsfreeze_mutex);
1380 	if (bdev->bd_fsfreeze_count > 0) {
1381 		mutex_unlock(&bdev->bd_fsfreeze_mutex);
1382 		error = -EBUSY;
1383 		goto error_bdev;
1384 	}
1385 	s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1386 		 bdev);
1387 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
1388 	if (IS_ERR(s))
1389 		goto error_s;
1390 
1391 	if (s->s_root) {
1392 		if ((flags ^ s->s_flags) & SB_RDONLY) {
1393 			deactivate_locked_super(s);
1394 			error = -EBUSY;
1395 			goto error_bdev;
1396 		}
1397 
1398 		/*
1399 		 * s_umount nests inside bd_mutex during
1400 		 * __invalidate_device().  blkdev_put() acquires
1401 		 * bd_mutex and can't be called under s_umount.  Drop
1402 		 * s_umount temporarily.  This is safe as we're
1403 		 * holding an active reference.
1404 		 */
1405 		up_write(&s->s_umount);
1406 		blkdev_put(bdev, mode);
1407 		down_write(&s->s_umount);
1408 	} else {
1409 		s->s_mode = mode;
1410 		snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1411 		sb_set_blocksize(s, block_size(bdev));
1412 		error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1413 		if (error) {
1414 			deactivate_locked_super(s);
1415 			goto error;
1416 		}
1417 
1418 		s->s_flags |= SB_ACTIVE;
1419 		bdev->bd_super = s;
1420 	}
1421 
1422 	return dget(s->s_root);
1423 
1424 error_s:
1425 	error = PTR_ERR(s);
1426 error_bdev:
1427 	blkdev_put(bdev, mode);
1428 error:
1429 	return ERR_PTR(error);
1430 }
1431 EXPORT_SYMBOL(mount_bdev);
1432 
1433 void kill_block_super(struct super_block *sb)
1434 {
1435 	struct block_device *bdev = sb->s_bdev;
1436 	fmode_t mode = sb->s_mode;
1437 
1438 	bdev->bd_super = NULL;
1439 	generic_shutdown_super(sb);
1440 	sync_blockdev(bdev);
1441 	WARN_ON_ONCE(!(mode & FMODE_EXCL));
1442 	blkdev_put(bdev, mode | FMODE_EXCL);
1443 }
1444 
1445 EXPORT_SYMBOL(kill_block_super);
1446 #endif
1447 
1448 struct dentry *mount_nodev(struct file_system_type *fs_type,
1449 	int flags, void *data,
1450 	int (*fill_super)(struct super_block *, void *, int))
1451 {
1452 	int error;
1453 	struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1454 
1455 	if (IS_ERR(s))
1456 		return ERR_CAST(s);
1457 
1458 	error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1459 	if (error) {
1460 		deactivate_locked_super(s);
1461 		return ERR_PTR(error);
1462 	}
1463 	s->s_flags |= SB_ACTIVE;
1464 	return dget(s->s_root);
1465 }
1466 EXPORT_SYMBOL(mount_nodev);
1467 
1468 static int reconfigure_single(struct super_block *s,
1469 			      int flags, void *data)
1470 {
1471 	struct fs_context *fc;
1472 	int ret;
1473 
1474 	/* The caller really need to be passing fc down into mount_single(),
1475 	 * then a chunk of this can be removed.  [Bollocks -- AV]
1476 	 * Better yet, reconfiguration shouldn't happen, but rather the second
1477 	 * mount should be rejected if the parameters are not compatible.
1478 	 */
1479 	fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1480 	if (IS_ERR(fc))
1481 		return PTR_ERR(fc);
1482 
1483 	ret = parse_monolithic_mount_data(fc, data);
1484 	if (ret < 0)
1485 		goto out;
1486 
1487 	ret = reconfigure_super(fc);
1488 out:
1489 	put_fs_context(fc);
1490 	return ret;
1491 }
1492 
1493 static int compare_single(struct super_block *s, void *p)
1494 {
1495 	return 1;
1496 }
1497 
1498 struct dentry *mount_single(struct file_system_type *fs_type,
1499 	int flags, void *data,
1500 	int (*fill_super)(struct super_block *, void *, int))
1501 {
1502 	struct super_block *s;
1503 	int error;
1504 
1505 	s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1506 	if (IS_ERR(s))
1507 		return ERR_CAST(s);
1508 	if (!s->s_root) {
1509 		error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1510 		if (!error)
1511 			s->s_flags |= SB_ACTIVE;
1512 	} else {
1513 		error = reconfigure_single(s, flags, data);
1514 	}
1515 	if (unlikely(error)) {
1516 		deactivate_locked_super(s);
1517 		return ERR_PTR(error);
1518 	}
1519 	return dget(s->s_root);
1520 }
1521 EXPORT_SYMBOL(mount_single);
1522 
1523 /**
1524  * vfs_get_tree - Get the mountable root
1525  * @fc: The superblock configuration context.
1526  *
1527  * The filesystem is invoked to get or create a superblock which can then later
1528  * be used for mounting.  The filesystem places a pointer to the root to be
1529  * used for mounting in @fc->root.
1530  */
1531 int vfs_get_tree(struct fs_context *fc)
1532 {
1533 	struct super_block *sb;
1534 	int error;
1535 
1536 	if (fc->root)
1537 		return -EBUSY;
1538 
1539 	/* Get the mountable root in fc->root, with a ref on the root and a ref
1540 	 * on the superblock.
1541 	 */
1542 	error = fc->ops->get_tree(fc);
1543 	if (error < 0)
1544 		return error;
1545 
1546 	if (!fc->root) {
1547 		pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1548 		       fc->fs_type->name);
1549 		/* We don't know what the locking state of the superblock is -
1550 		 * if there is a superblock.
1551 		 */
1552 		BUG();
1553 	}
1554 
1555 	sb = fc->root->d_sb;
1556 	WARN_ON(!sb->s_bdi);
1557 
1558 	/*
1559 	 * Write barrier is for super_cache_count(). We place it before setting
1560 	 * SB_BORN as the data dependency between the two functions is the
1561 	 * superblock structure contents that we just set up, not the SB_BORN
1562 	 * flag.
1563 	 */
1564 	smp_wmb();
1565 	sb->s_flags |= SB_BORN;
1566 
1567 	error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1568 	if (unlikely(error)) {
1569 		fc_drop_locked(fc);
1570 		return error;
1571 	}
1572 
1573 	/*
1574 	 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1575 	 * but s_maxbytes was an unsigned long long for many releases. Throw
1576 	 * this warning for a little while to try and catch filesystems that
1577 	 * violate this rule.
1578 	 */
1579 	WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1580 		"negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1581 
1582 	return 0;
1583 }
1584 EXPORT_SYMBOL(vfs_get_tree);
1585 
1586 /*
1587  * Setup private BDI for given superblock. It gets automatically cleaned up
1588  * in generic_shutdown_super().
1589  */
1590 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1591 {
1592 	struct backing_dev_info *bdi;
1593 	int err;
1594 	va_list args;
1595 
1596 	bdi = bdi_alloc(GFP_KERNEL);
1597 	if (!bdi)
1598 		return -ENOMEM;
1599 
1600 	bdi->name = sb->s_type->name;
1601 
1602 	va_start(args, fmt);
1603 	err = bdi_register_va(bdi, fmt, args);
1604 	va_end(args);
1605 	if (err) {
1606 		bdi_put(bdi);
1607 		return err;
1608 	}
1609 	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1610 	sb->s_bdi = bdi;
1611 
1612 	return 0;
1613 }
1614 EXPORT_SYMBOL(super_setup_bdi_name);
1615 
1616 /*
1617  * Setup private BDI for given superblock. I gets automatically cleaned up
1618  * in generic_shutdown_super().
1619  */
1620 int super_setup_bdi(struct super_block *sb)
1621 {
1622 	static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1623 
1624 	return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1625 				    atomic_long_inc_return(&bdi_seq));
1626 }
1627 EXPORT_SYMBOL(super_setup_bdi);
1628 
1629 /*
1630  * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1631  * instead.
1632  */
1633 void __sb_end_write(struct super_block *sb, int level)
1634 {
1635 	percpu_up_read(sb->s_writers.rw_sem + level-1);
1636 }
1637 EXPORT_SYMBOL(__sb_end_write);
1638 
1639 /*
1640  * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1641  * instead.
1642  */
1643 int __sb_start_write(struct super_block *sb, int level, bool wait)
1644 {
1645 	bool force_trylock = false;
1646 	int ret = 1;
1647 
1648 #ifdef CONFIG_LOCKDEP
1649 	/*
1650 	 * We want lockdep to tell us about possible deadlocks with freezing
1651 	 * but it's it bit tricky to properly instrument it. Getting a freeze
1652 	 * protection works as getting a read lock but there are subtle
1653 	 * problems. XFS for example gets freeze protection on internal level
1654 	 * twice in some cases, which is OK only because we already hold a
1655 	 * freeze protection also on higher level. Due to these cases we have
1656 	 * to use wait == F (trylock mode) which must not fail.
1657 	 */
1658 	if (wait) {
1659 		int i;
1660 
1661 		for (i = 0; i < level - 1; i++)
1662 			if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1663 				force_trylock = true;
1664 				break;
1665 			}
1666 	}
1667 #endif
1668 	if (wait && !force_trylock)
1669 		percpu_down_read(sb->s_writers.rw_sem + level-1);
1670 	else
1671 		ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1672 
1673 	WARN_ON(force_trylock && !ret);
1674 	return ret;
1675 }
1676 EXPORT_SYMBOL(__sb_start_write);
1677 
1678 /**
1679  * sb_wait_write - wait until all writers to given file system finish
1680  * @sb: the super for which we wait
1681  * @level: type of writers we wait for (normal vs page fault)
1682  *
1683  * This function waits until there are no writers of given type to given file
1684  * system.
1685  */
1686 static void sb_wait_write(struct super_block *sb, int level)
1687 {
1688 	percpu_down_write(sb->s_writers.rw_sem + level-1);
1689 }
1690 
1691 /*
1692  * We are going to return to userspace and forget about these locks, the
1693  * ownership goes to the caller of thaw_super() which does unlock().
1694  */
1695 static void lockdep_sb_freeze_release(struct super_block *sb)
1696 {
1697 	int level;
1698 
1699 	for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1700 		percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1701 }
1702 
1703 /*
1704  * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1705  */
1706 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1707 {
1708 	int level;
1709 
1710 	for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1711 		percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1712 }
1713 
1714 static void sb_freeze_unlock(struct super_block *sb)
1715 {
1716 	int level;
1717 
1718 	for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1719 		percpu_up_write(sb->s_writers.rw_sem + level);
1720 }
1721 
1722 /**
1723  * freeze_super - lock the filesystem and force it into a consistent state
1724  * @sb: the super to lock
1725  *
1726  * Syncs the super to make sure the filesystem is consistent and calls the fs's
1727  * freeze_fs.  Subsequent calls to this without first thawing the fs will return
1728  * -EBUSY.
1729  *
1730  * During this function, sb->s_writers.frozen goes through these values:
1731  *
1732  * SB_UNFROZEN: File system is normal, all writes progress as usual.
1733  *
1734  * SB_FREEZE_WRITE: The file system is in the process of being frozen.  New
1735  * writes should be blocked, though page faults are still allowed. We wait for
1736  * all writes to complete and then proceed to the next stage.
1737  *
1738  * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1739  * but internal fs threads can still modify the filesystem (although they
1740  * should not dirty new pages or inodes), writeback can run etc. After waiting
1741  * for all running page faults we sync the filesystem which will clean all
1742  * dirty pages and inodes (no new dirty pages or inodes can be created when
1743  * sync is running).
1744  *
1745  * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1746  * modification are blocked (e.g. XFS preallocation truncation on inode
1747  * reclaim). This is usually implemented by blocking new transactions for
1748  * filesystems that have them and need this additional guard. After all
1749  * internal writers are finished we call ->freeze_fs() to finish filesystem
1750  * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1751  * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1752  *
1753  * sb->s_writers.frozen is protected by sb->s_umount.
1754  */
1755 int freeze_super(struct super_block *sb)
1756 {
1757 	int ret;
1758 
1759 	atomic_inc(&sb->s_active);
1760 	down_write(&sb->s_umount);
1761 	if (sb->s_writers.frozen != SB_UNFROZEN) {
1762 		deactivate_locked_super(sb);
1763 		return -EBUSY;
1764 	}
1765 
1766 	if (!(sb->s_flags & SB_BORN)) {
1767 		up_write(&sb->s_umount);
1768 		return 0;	/* sic - it's "nothing to do" */
1769 	}
1770 
1771 	if (sb_rdonly(sb)) {
1772 		/* Nothing to do really... */
1773 		sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1774 		up_write(&sb->s_umount);
1775 		return 0;
1776 	}
1777 
1778 	sb->s_writers.frozen = SB_FREEZE_WRITE;
1779 	/* Release s_umount to preserve sb_start_write -> s_umount ordering */
1780 	up_write(&sb->s_umount);
1781 	sb_wait_write(sb, SB_FREEZE_WRITE);
1782 	down_write(&sb->s_umount);
1783 
1784 	/* Now we go and block page faults... */
1785 	sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1786 	sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1787 
1788 	/* All writers are done so after syncing there won't be dirty data */
1789 	sync_filesystem(sb);
1790 
1791 	/* Now wait for internal filesystem counter */
1792 	sb->s_writers.frozen = SB_FREEZE_FS;
1793 	sb_wait_write(sb, SB_FREEZE_FS);
1794 
1795 	if (sb->s_op->freeze_fs) {
1796 		ret = sb->s_op->freeze_fs(sb);
1797 		if (ret) {
1798 			printk(KERN_ERR
1799 				"VFS:Filesystem freeze failed\n");
1800 			sb->s_writers.frozen = SB_UNFROZEN;
1801 			sb_freeze_unlock(sb);
1802 			wake_up(&sb->s_writers.wait_unfrozen);
1803 			deactivate_locked_super(sb);
1804 			return ret;
1805 		}
1806 	}
1807 	/*
1808 	 * For debugging purposes so that fs can warn if it sees write activity
1809 	 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1810 	 */
1811 	sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1812 	lockdep_sb_freeze_release(sb);
1813 	up_write(&sb->s_umount);
1814 	return 0;
1815 }
1816 EXPORT_SYMBOL(freeze_super);
1817 
1818 /**
1819  * thaw_super -- unlock filesystem
1820  * @sb: the super to thaw
1821  *
1822  * Unlocks the filesystem and marks it writeable again after freeze_super().
1823  */
1824 static int thaw_super_locked(struct super_block *sb)
1825 {
1826 	int error;
1827 
1828 	if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1829 		up_write(&sb->s_umount);
1830 		return -EINVAL;
1831 	}
1832 
1833 	if (sb_rdonly(sb)) {
1834 		sb->s_writers.frozen = SB_UNFROZEN;
1835 		goto out;
1836 	}
1837 
1838 	lockdep_sb_freeze_acquire(sb);
1839 
1840 	if (sb->s_op->unfreeze_fs) {
1841 		error = sb->s_op->unfreeze_fs(sb);
1842 		if (error) {
1843 			printk(KERN_ERR
1844 				"VFS:Filesystem thaw failed\n");
1845 			lockdep_sb_freeze_release(sb);
1846 			up_write(&sb->s_umount);
1847 			return error;
1848 		}
1849 	}
1850 
1851 	sb->s_writers.frozen = SB_UNFROZEN;
1852 	sb_freeze_unlock(sb);
1853 out:
1854 	wake_up(&sb->s_writers.wait_unfrozen);
1855 	deactivate_locked_super(sb);
1856 	return 0;
1857 }
1858 
1859 int thaw_super(struct super_block *sb)
1860 {
1861 	down_write(&sb->s_umount);
1862 	return thaw_super_locked(sb);
1863 }
1864 EXPORT_SYMBOL(thaw_super);
1865