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/fscrypt.h>
35	#include <linux/fsnotify.h>
36	#include <linux/lockdep.h>
37	#include <linux/user_namespace.h>
38	#include <linux/fs_context.h>
39	#include <uapi/linux/mount.h>
40	#include "internal.h"
41
42	static int thaw_super_locked(struct super_block *sb, enum freeze_holder who,
43	const void *freeze_owner);
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	static inline void __super_lock(struct super_block *sb, bool excl)
55	{
56	if (excl)
57	down_write(sem: &sb->s_umount);
58	else
59	down_read(sem: &sb->s_umount);
60	}
61
62	static inline void super_unlock(struct super_block *sb, bool excl)
63	{
64	if (excl)
65	up_write(sem: &sb->s_umount);
66	else
67	up_read(sem: &sb->s_umount);
68	}
69
70	static inline void __super_lock_excl(struct super_block *sb)
71	{
72	__super_lock(sb, excl: true);
73	}
74
75	static inline void super_unlock_excl(struct super_block *sb)
76	{
77	super_unlock(sb, excl: true);
78	}
79
80	static inline void super_unlock_shared(struct super_block *sb)
81	{
82	super_unlock(sb, excl: false);
83	}
84
85	static bool super_flags(const struct super_block *sb, unsigned int flags)
86	{
87	/*
88	* Pairs with smp_store_release() in super_wake() and ensures
89	* that we see @flags after we're woken.
90	*/
91	return smp_load_acquire(&sb->s_flags) & flags;
92	}
93
94	/**
95	* super_lock - wait for superblock to become ready and lock it
96	* @sb: superblock to wait for
97	* @excl: whether exclusive access is required
98	*
99	* If the superblock has neither passed through vfs_get_tree() or
100	* generic_shutdown_super() yet wait for it to happen. Either superblock
101	* creation will succeed and SB_BORN is set by vfs_get_tree() or we're
102	* woken and we'll see SB_DYING.
103	*
104	* The caller must have acquired a temporary reference on @sb->s_count.
105	*
106	* Return: The function returns true if SB_BORN was set and with
107	* s_umount held. The function returns false if SB_DYING was
108	* set and without s_umount held.
109	*/
110	static __must_check bool super_lock(struct super_block *sb, bool excl)
111	{
112	lockdep_assert_not_held(&sb->s_umount);
113
114	/* wait until the superblock is ready or dying */
115	wait_var_event(&sb->s_flags, super_flags(sb, SB_BORN | SB_DYING));
116
117	/* Don't pointlessly acquire s_umount. */
118	if (super_flags(sb, SB_DYING))
119	return false;
120
121	__super_lock(sb, excl);
122
123	/*
124	* Has gone through generic_shutdown_super() in the meantime.
125	* @sb->s_root is NULL and @sb->s_active is 0. No one needs to
126	* grab a reference to this. Tell them so.
127	*/
128	if (sb->s_flags & SB_DYING) {
129	super_unlock(sb, excl);
130	return false;
131	}
132
133	WARN_ON_ONCE(!(sb->s_flags & SB_BORN));
134	return true;
135	}
136
137	/* wait and try to acquire read-side of @sb->s_umount */
138	static inline bool super_lock_shared(struct super_block *sb)
139	{
140	return super_lock(sb, excl: false);
141	}
142
143	/* wait and try to acquire write-side of @sb->s_umount */
144	static inline bool super_lock_excl(struct super_block *sb)
145	{
146	return super_lock(sb, excl: true);
147	}
148
149	/* wake waiters */
150	#define SUPER_WAKE_FLAGS (SB_BORN | SB_DYING | SB_DEAD)
151	static void super_wake(struct super_block *sb, unsigned int flag)
152	{
153	WARN_ON_ONCE((flag & ~SUPER_WAKE_FLAGS));
154	WARN_ON_ONCE(hweight32(flag & SUPER_WAKE_FLAGS) > 1);
155
156	/*
157	* Pairs with smp_load_acquire() in super_lock() to make sure
158	* all initializations in the superblock are seen by the user
159	* seeing SB_BORN sent.
160	*/
161	smp_store_release(&sb->s_flags, sb->s_flags | flag);
162	/*
163	* Pairs with the barrier in prepare_to_wait_event() to make sure
164	* ___wait_var_event() either sees SB_BORN set or
165	* waitqueue_active() check in wake_up_var() sees the waiter.
166	*/
167	smp_mb();
168	wake_up_var(var: &sb->s_flags);
169	}
170
171	/*
172	* One thing we have to be careful of with a per-sb shrinker is that we don't
173	* drop the last active reference to the superblock from within the shrinker.
174	* If that happens we could trigger unregistering the shrinker from within the
175	* shrinker path and that leads to deadlock on the shrinker_mutex. Hence we
176	* take a passive reference to the superblock to avoid this from occurring.
177	*/
178	static unsigned long super_cache_scan(struct shrinker *shrink,
179	struct shrink_control *sc)
180	{
181	struct super_block *sb;
182	long fs_objects = 0;
183	long total_objects;
184	long freed = 0;
185	long dentries;
186	long inodes;
187
188	sb = shrink->private_data;
189
190	/*
191	* Deadlock avoidance. We may hold various FS locks, and we don't want
192	* to recurse into the FS that called us in clear_inode() and friends..
193	*/
194	if (!(sc->gfp_mask & __GFP_FS))
195	return SHRINK_STOP;
196
197	if (!super_trylock_shared(sb))
198	return SHRINK_STOP;
199
200	if (sb->s_op->nr_cached_objects)
201	fs_objects = sb->s_op->nr_cached_objects(sb, sc);
202
203	inodes = list_lru_shrink_count(lru: &sb->s_inode_lru, sc);
204	dentries = list_lru_shrink_count(lru: &sb->s_dentry_lru, sc);
205	total_objects = dentries + inodes + fs_objects;
206	if (!total_objects)
207	total_objects = 1;
208
209	/* proportion the scan between the caches */
210	dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
211	inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
212	fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
213
214	/*
215	* prune the dcache first as the icache is pinned by it, then
216	* prune the icache, followed by the filesystem specific caches
217	*
218	* Ensure that we always scan at least one object - memcg kmem
219	* accounting uses this to fully empty the caches.
220	*/
221	sc->nr_to_scan = dentries + 1;
222	freed = prune_dcache_sb(sb, sc);
223	sc->nr_to_scan = inodes + 1;
224	freed += prune_icache_sb(sb, sc);
225
226	if (fs_objects) {
227	sc->nr_to_scan = fs_objects + 1;
228	freed += sb->s_op->free_cached_objects(sb, sc);
229	}
230
231	super_unlock_shared(sb);
232	return freed;
233	}
234
235	static unsigned long super_cache_count(struct shrinker *shrink,
236	struct shrink_control *sc)
237	{
238	struct super_block *sb;
239	long total_objects = 0;
240
241	sb = shrink->private_data;
242
243	/*
244	* We don't call super_trylock_shared() here as it is a scalability
245	* bottleneck, so we're exposed to partial setup state. The shrinker
246	* rwsem does not protect filesystem operations backing
247	* list_lru_shrink_count() or s_op->nr_cached_objects(). Counts can
248	* change between super_cache_count and super_cache_scan, so we really
249	* don't need locks here.
250	*
251	* However, if we are currently mounting the superblock, the underlying
252	* filesystem might be in a state of partial construction and hence it
253	* is dangerous to access it. super_trylock_shared() uses a SB_BORN check
254	* to avoid this situation, so do the same here. The memory barrier is
255	* matched with the one in mount_fs() as we don't hold locks here.
256	*/
257	if (!(sb->s_flags & SB_BORN))
258	return 0;
259	smp_rmb();
260
261	if (sb->s_op && sb->s_op->nr_cached_objects)
262	total_objects = sb->s_op->nr_cached_objects(sb, sc);
263
264	total_objects += list_lru_shrink_count(lru: &sb->s_dentry_lru, sc);
265	total_objects += list_lru_shrink_count(lru: &sb->s_inode_lru, sc);
266
267	if (!total_objects)
268	return SHRINK_EMPTY;
269
270	total_objects = vfs_pressure_ratio(val: total_objects);
271	return total_objects;
272	}
273
274	static void destroy_super_work(struct work_struct *work)
275	{
276	struct super_block *s = container_of(work, struct super_block,
277	destroy_work);
278	fsnotify_sb_free(sb: s);
279	security_sb_free(sb: s);
280	put_user_ns(ns: s->s_user_ns);
281	kfree(objp: s->s_subtype);
282	for (int i = 0; i < SB_FREEZE_LEVELS; i++)
283	percpu_free_rwsem(&s->s_writers.rw_sem[i]);
284	kfree(objp: s);
285	}
286
287	static void destroy_super_rcu(struct rcu_head *head)
288	{
289	struct super_block *s = container_of(head, struct super_block, rcu);
290	INIT_WORK(&s->destroy_work, destroy_super_work);
291	schedule_work(work: &s->destroy_work);
292	}
293
294	/* Free a superblock that has never been seen by anyone */
295	static void destroy_unused_super(struct super_block *s)
296	{
297	if (!s)
298	return;
299	super_unlock_excl(sb: s);
300	list_lru_destroy(lru: &s->s_dentry_lru);
301	list_lru_destroy(lru: &s->s_inode_lru);
302	shrinker_free(shrinker: s->s_shrink);
303	/* no delays needed */
304	destroy_super_work(work: &s->destroy_work);
305	}
306
307	/**
308	* alloc_super - create new superblock
309	* @type: filesystem type superblock should belong to
310	* @flags: the mount flags
311	* @user_ns: User namespace for the super_block
312	*
313	* Allocates and initializes a new &struct super_block. alloc_super()
314	* returns a pointer new superblock or %NULL if allocation had failed.
315	*/
316	static struct super_block *alloc_super(struct file_system_type *type, int flags,
317	struct user_namespace *user_ns)
318	{
319	struct super_block *s = kzalloc(sizeof(struct super_block), GFP_KERNEL);
320	static const struct super_operations default_op;
321	int i;
322
323	if (!s)
324	return NULL;
325
326	INIT_LIST_HEAD(list: &s->s_mounts);
327	s->s_user_ns = get_user_ns(ns: user_ns);
328	init_rwsem(&s->s_umount);
329	lockdep_set_class(&s->s_umount, &type->s_umount_key);
330	/*
331	* sget() can have s_umount recursion.
332	*
333	* When it cannot find a suitable sb, it allocates a new
334	* one (this one), and tries again to find a suitable old
335	* one.
336	*
337	* In case that succeeds, it will acquire the s_umount
338	* lock of the old one. Since these are clearly distrinct
339	* locks, and this object isn't exposed yet, there's no
340	* risk of deadlocks.
341	*
342	* Annotate this by putting this lock in a different
343	* subclass.
344	*/
345	down_write_nested(sem: &s->s_umount, SINGLE_DEPTH_NESTING);
346
347	if (security_sb_alloc(sb: s))
348	goto fail;
349
350	for (i = 0; i < SB_FREEZE_LEVELS; i++) {
351	if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
352	sb_writers_name[i],
353	&type->s_writers_key[i]))
354	goto fail;
355	}
356	s->s_bdi = &noop_backing_dev_info;
357	s->s_flags = flags;
358	if (s->s_user_ns != &init_user_ns)
359	s->s_iflags |= SB_I_NODEV;
360	INIT_HLIST_NODE(h: &s->s_instances);
361	INIT_HLIST_BL_HEAD(&s->s_roots);
362	mutex_init(&s->s_sync_lock);
363	INIT_LIST_HEAD(list: &s->s_inodes);
364	spin_lock_init(&s->s_inode_list_lock);
365	INIT_LIST_HEAD(list: &s->s_inodes_wb);
366	spin_lock_init(&s->s_inode_wblist_lock);
367
368	s->s_count = 1;
369	atomic_set(v: &s->s_active, i: 1);
370	mutex_init(&s->s_vfs_rename_mutex);
371	lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
372	init_rwsem(&s->s_dquot.dqio_sem);
373	s->s_maxbytes = MAX_NON_LFS;
374	s->s_op = &default_op;
375	s->s_time_gran = 1000000000;
376	s->s_time_min = TIME64_MIN;
377	s->s_time_max = TIME64_MAX;
378
379	s->s_shrink = shrinker_alloc(SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE,
380	fmt: "sb-%s", type->name);
381	if (!s->s_shrink)
382	goto fail;
383
384	s->s_shrink->scan_objects = super_cache_scan;
385	s->s_shrink->count_objects = super_cache_count;
386	s->s_shrink->batch = 1024;
387	s->s_shrink->private_data = s;
388
389	if (list_lru_init_memcg(&s->s_dentry_lru, s->s_shrink))
390	goto fail;
391	if (list_lru_init_memcg(&s->s_inode_lru, s->s_shrink))
392	goto fail;
393	return s;
394
395	fail:
396	destroy_unused_super(s);
397	return NULL;
398	}
399
400	/* Superblock refcounting */
401
402	/*
403	* Drop a superblock's refcount. The caller must hold sb_lock.
404	*/
405	static void __put_super(struct super_block *s)
406	{
407	if (!--s->s_count) {
408	list_del_init(entry: &s->s_list);
409	WARN_ON(s->s_dentry_lru.node);
410	WARN_ON(s->s_inode_lru.node);
411	WARN_ON(!list_empty(&s->s_mounts));
412	call_rcu(head: &s->rcu, func: destroy_super_rcu);
413	}
414	}
415
416	/**
417	* put_super - drop a temporary reference to superblock
418	* @sb: superblock in question
419	*
420	* Drops a temporary reference, frees superblock if there's no
421	* references left.
422	*/
423	void put_super(struct super_block *sb)
424	{
425	spin_lock(lock: &sb_lock);
426	__put_super(s: sb);
427	spin_unlock(lock: &sb_lock);
428	}
429
430	static void kill_super_notify(struct super_block *sb)
431	{
432	lockdep_assert_not_held(&sb->s_umount);
433
434	/* already notified earlier */
435	if (sb->s_flags & SB_DEAD)
436	return;
437
438	/*
439	* Remove it from @fs_supers so it isn't found by new
440	* sget{_fc}() walkers anymore. Any concurrent mounter still
441	* managing to grab a temporary reference is guaranteed to
442	* already see SB_DYING and will wait until we notify them about
443	* SB_DEAD.
444	*/
445	spin_lock(lock: &sb_lock);
446	hlist_del_init(n: &sb->s_instances);
447	spin_unlock(lock: &sb_lock);
448
449	/*
450	* Let concurrent mounts know that this thing is really dead.
451	* We don't need @sb->s_umount here as every concurrent caller
452	* will see SB_DYING and either discard the superblock or wait
453	* for SB_DEAD.
454	*/
455	super_wake(sb, SB_DEAD);
456	}
457
458	/**
459	* deactivate_locked_super - drop an active reference to superblock
460	* @s: superblock to deactivate
461	*
462	* Drops an active reference to superblock, converting it into a temporary
463	* one if there is no other active references left. In that case we
464	* tell fs driver to shut it down and drop the temporary reference we
465	* had just acquired.
466	*
467	* Caller holds exclusive lock on superblock; that lock is released.
468	*/
469	void deactivate_locked_super(struct super_block *s)
470	{
471	struct file_system_type *fs = s->s_type;
472	if (atomic_dec_and_test(v: &s->s_active)) {
473	shrinker_free(shrinker: s->s_shrink);
474	fs->kill_sb(s);
475
476	kill_super_notify(sb: s);
477
478	/*
479	* Since list_lru_destroy() may sleep, we cannot call it from
480	* put_super(), where we hold the sb_lock. Therefore we destroy
481	* the lru lists right now.
482	*/
483	list_lru_destroy(lru: &s->s_dentry_lru);
484	list_lru_destroy(lru: &s->s_inode_lru);
485
486	put_filesystem(fs);
487	put_super(sb: s);
488	} else {
489	super_unlock_excl(sb: s);
490	}
491	}
492
493	EXPORT_SYMBOL(deactivate_locked_super);
494
495	/**
496	* deactivate_super - drop an active reference to superblock
497	* @s: superblock to deactivate
498	*
499	* Variant of deactivate_locked_super(), except that superblock is *not*
500	* locked by caller. If we are going to drop the final active reference,
501	* lock will be acquired prior to that.
502	*/
503	void deactivate_super(struct super_block *s)
504	{
505	if (!atomic_add_unless(v: &s->s_active, a: -1, u: 1)) {
506	__super_lock_excl(sb: s);
507	deactivate_locked_super(s);
508	}
509	}
510
511	EXPORT_SYMBOL(deactivate_super);
512
513	/**
514	* grab_super - acquire an active reference to a superblock
515	* @sb: superblock to acquire
516	*
517	* Acquire a temporary reference on a superblock and try to trade it for
518	* an active reference. This is used in sget{_fc}() to wait for a
519	* superblock to either become SB_BORN or for it to pass through
520	* sb->kill() and be marked as SB_DEAD.
521	*
522	* Return: This returns true if an active reference could be acquired,
523	* false if not.
524	*/
525	static bool grab_super(struct super_block *sb)
526	{
527	bool locked;
528
529	sb->s_count++;
530	spin_unlock(lock: &sb_lock);
531	locked = super_lock_excl(sb);
532	if (locked) {
533	if (atomic_inc_not_zero(v: &sb->s_active)) {
534	put_super(sb);
535	return true;
536	}
537	super_unlock_excl(sb);
538	}
539	wait_var_event(&sb->s_flags, super_flags(sb, SB_DEAD));
540	put_super(sb);
541	return false;
542	}
543
544	/*
545	* super_trylock_shared - try to grab ->s_umount shared
546	* @sb: reference we are trying to grab
547	*
548	* Try to prevent fs shutdown. This is used in places where we
549	* cannot take an active reference but we need to ensure that the
550	* filesystem is not shut down while we are working on it. It returns
551	* false if we cannot acquire s_umount or if we lose the race and
552	* filesystem already got into shutdown, and returns true with the s_umount
553	* lock held in read mode in case of success. On successful return,
554	* the caller must drop the s_umount lock when done.
555	*
556	* Note that unlike get_super() et.al. this one does *not* bump ->s_count.
557	* The reason why it's safe is that we are OK with doing trylock instead
558	* of down_read(). There's a couple of places that are OK with that, but
559	* it's very much not a general-purpose interface.
560	*/
561	bool super_trylock_shared(struct super_block *sb)
562	{
563	if (down_read_trylock(sem: &sb->s_umount)) {
564	if (!(sb->s_flags & SB_DYING) && sb->s_root &&
565	(sb->s_flags & SB_BORN))
566	return true;
567	super_unlock_shared(sb);
568	}
569
570	return false;
571	}
572
573	/**
574	* retire_super - prevents superblock from being reused
575	* @sb: superblock to retire
576	*
577	* The function marks superblock to be ignored in superblock test, which
578	* prevents it from being reused for any new mounts. If the superblock has
579	* a private bdi, it also unregisters it, but doesn't reduce the refcount
580	* of the superblock to prevent potential races. The refcount is reduced
581	* by generic_shutdown_super(). The function can not be called
582	* concurrently with generic_shutdown_super(). It is safe to call the
583	* function multiple times, subsequent calls have no effect.
584	*
585	* The marker will affect the re-use only for block-device-based
586	* superblocks. Other superblocks will still get marked if this function
587	* is used, but that will not affect their reusability.
588	*/
589	void retire_super(struct super_block *sb)
590	{
591	WARN_ON(!sb->s_bdev);
592	__super_lock_excl(sb);
593	if (sb->s_iflags & SB_I_PERSB_BDI) {
594	bdi_unregister(bdi: sb->s_bdi);
595	sb->s_iflags &= ~SB_I_PERSB_BDI;
596	}
597	sb->s_iflags |= SB_I_RETIRED;
598	super_unlock_excl(sb);
599	}
600	EXPORT_SYMBOL(retire_super);
601
602	/**
603	* generic_shutdown_super - common helper for ->kill_sb()
604	* @sb: superblock to kill
605	*
606	* generic_shutdown_super() does all fs-independent work on superblock
607	* shutdown. Typical ->kill_sb() should pick all fs-specific objects
608	* that need destruction out of superblock, call generic_shutdown_super()
609	* and release aforementioned objects. Note: dentries and inodes _are_
610	* taken care of and do not need specific handling.
611	*
612	* Upon calling this function, the filesystem may no longer alter or
613	* rearrange the set of dentries belonging to this super_block, nor may it
614	* change the attachments of dentries to inodes.
615	*/
616	void generic_shutdown_super(struct super_block *sb)
617	{
618	const struct super_operations *sop = sb->s_op;
619
620	if (sb->s_root) {
621	shrink_dcache_for_umount(sb);
622	sync_filesystem(sb);
623	sb->s_flags &= ~SB_ACTIVE;
624
625	cgroup_writeback_umount(sb);
626
627	/* Evict all inodes with zero refcount. */
628	evict_inodes(sb);
629
630	/*
631	* Clean up and evict any inodes that still have references due
632	* to fsnotify or the security policy.
633	*/
634	fsnotify_sb_delete(sb);
635	security_sb_delete(sb);
636
637	if (sb->s_dio_done_wq) {
638	destroy_workqueue(wq: sb->s_dio_done_wq);
639	sb->s_dio_done_wq = NULL;
640	}
641
642	if (sop->put_super)
643	sop->put_super(sb);
644
645	/*
646	* Now that all potentially-encrypted inodes have been evicted,
647	* the fscrypt keyring can be destroyed.
648	*/
649	fscrypt_destroy_keyring(sb);
650
651	if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes), NULL,
652	"VFS: Busy inodes after unmount of %s (%s)",
653	sb->s_id, sb->s_type->name)) {
654	/*
655	* Adding a proper bailout path here would be hard, but
656	* we can at least make it more likely that a later
657	* iput_final() or such crashes cleanly.
658	*/
659	struct inode *inode;
660
661	spin_lock(lock: &sb->s_inode_list_lock);
662	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
663	inode->i_op = VFS_PTR_POISON;
664	inode->i_sb = VFS_PTR_POISON;
665	inode->i_mapping = VFS_PTR_POISON;
666	}
667	spin_unlock(lock: &sb->s_inode_list_lock);
668	}
669	}
670	/*
671	* Broadcast to everyone that grabbed a temporary reference to this
672	* superblock before we removed it from @fs_supers that the superblock
673	* is dying. Every walker of @fs_supers outside of sget{_fc}() will now
674	* discard this superblock and treat it as dead.
675	*
676	* We leave the superblock on @fs_supers so it can be found by
677	* sget{_fc}() until we passed sb->kill_sb().
678	*/
679	super_wake(sb, SB_DYING);
680	super_unlock_excl(sb);
681	if (sb->s_bdi != &noop_backing_dev_info) {
682	if (sb->s_iflags & SB_I_PERSB_BDI)
683	bdi_unregister(bdi: sb->s_bdi);
684	bdi_put(bdi: sb->s_bdi);
685	sb->s_bdi = &noop_backing_dev_info;
686	}
687	}
688
689	EXPORT_SYMBOL(generic_shutdown_super);
690
691	bool mount_capable(struct fs_context *fc)
692	{
693	if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
694	return capable(CAP_SYS_ADMIN);
695	else
696	return ns_capable(ns: fc->user_ns, CAP_SYS_ADMIN);
697	}
698
699	/**
700	* sget_fc - Find or create a superblock
701	* @fc: Filesystem context.
702	* @test: Comparison callback
703	* @set: Setup callback
704	*
705	* Create a new superblock or find an existing one.
706	*
707	* The @test callback is used to find a matching existing superblock.
708	* Whether or not the requested parameters in @fc are taken into account
709	* is specific to the @test callback that is used. They may even be
710	* completely ignored.
711	*
712	* If an extant superblock is matched, it will be returned unless:
713	*
714	* (1) the namespace the filesystem context @fc and the extant
715	* superblock's namespace differ
716	*
717	* (2) the filesystem context @fc has requested that reusing an extant
718	* superblock is not allowed
719	*
720	* In both cases EBUSY will be returned.
721	*
722	* If no match is made, a new superblock will be allocated and basic
723	* initialisation will be performed (s_type, s_fs_info and s_id will be
724	* set and the @set callback will be invoked), the superblock will be
725	* published and it will be returned in a partially constructed state
726	* with SB_BORN and SB_ACTIVE as yet unset.
727	*
728	* Return: On success, an extant or newly created superblock is
729	* returned. On failure an error pointer is returned.
730	*/
731	struct super_block *sget_fc(struct fs_context *fc,
732	int (*test)(struct super_block *, struct fs_context *),
733	int (*set)(struct super_block *, struct fs_context *))
734	{
735	struct super_block *s = NULL;
736	struct super_block *old;
737	struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
738	int err;
739
740	/*
741	* Never allow s_user_ns != &init_user_ns when FS_USERNS_MOUNT is
742	* not set, as the filesystem is likely unprepared to handle it.
743	* This can happen when fsconfig() is called from init_user_ns with
744	* an fs_fd opened in another user namespace.
745	*/
746	if (user_ns != &init_user_ns && !(fc->fs_type->fs_flags & FS_USERNS_MOUNT)) {
747	errorfc(fc, "VFS: Mounting from non-initial user namespace is not allowed");
748	return ERR_PTR(error: -EPERM);
749	}
750
751	retry:
752	spin_lock(lock: &sb_lock);
753	if (test) {
754	hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
755	if (test(old, fc))
756	goto share_extant_sb;
757	}
758	}
759	if (!s) {
760	spin_unlock(lock: &sb_lock);
761	s = alloc_super(type: fc->fs_type, flags: fc->sb_flags, user_ns);
762	if (!s)
763	return ERR_PTR(error: -ENOMEM);
764	goto retry;
765	}
766
767	s->s_fs_info = fc->s_fs_info;
768	err = set(s, fc);
769	if (err) {
770	s->s_fs_info = NULL;
771	spin_unlock(lock: &sb_lock);
772	destroy_unused_super(s);
773	return ERR_PTR(error: err);
774	}
775	fc->s_fs_info = NULL;
776	s->s_type = fc->fs_type;
777	s->s_iflags |= fc->s_iflags;
778	strscpy(s->s_id, s->s_type->name, sizeof(s->s_id));
779	/*
780	* Make the superblock visible on @super_blocks and @fs_supers.
781	* It's in a nascent state and users should wait on SB_BORN or
782	* SB_DYING to be set.
783	*/
784	list_add_tail(new: &s->s_list, head: &super_blocks);
785	hlist_add_head(n: &s->s_instances, h: &s->s_type->fs_supers);
786	spin_unlock(lock: &sb_lock);
787	get_filesystem(fs: s->s_type);
788	shrinker_register(shrinker: s->s_shrink);
789	return s;
790
791	share_extant_sb:
792	if (user_ns != old->s_user_ns || fc->exclusive) {
793	spin_unlock(lock: &sb_lock);
794	destroy_unused_super(s);
795	if (fc->exclusive)
796	warnfc(fc, "reusing existing filesystem not allowed");
797	else
798	warnfc(fc, "reusing existing filesystem in another namespace not allowed");
799	return ERR_PTR(error: -EBUSY);
800	}
801	if (!grab_super(sb: old))
802	goto retry;
803	destroy_unused_super(s);
804	return old;
805	}
806	EXPORT_SYMBOL(sget_fc);
807
808	/**
809	* sget - find or create a superblock
810	* @type: filesystem type superblock should belong to
811	* @test: comparison callback
812	* @set: setup callback
813	* @flags: mount flags
814	* @data: argument to each of them
815	*/
816	struct super_block *sget(struct file_system_type *type,
817	int (*test)(struct super_block *,void *),
818	int (*set)(struct super_block *,void *),
819	int flags,
820	void *data)
821	{
822	struct user_namespace *user_ns = current_user_ns();
823	struct super_block *s = NULL;
824	struct super_block *old;
825	int err;
826
827	retry:
828	spin_lock(lock: &sb_lock);
829	if (test) {
830	hlist_for_each_entry(old, &type->fs_supers, s_instances) {
831	if (!test(old, data))
832	continue;
833	if (user_ns != old->s_user_ns) {
834	spin_unlock(lock: &sb_lock);
835	destroy_unused_super(s);
836	return ERR_PTR(error: -EBUSY);
837	}
838	if (!grab_super(sb: old))
839	goto retry;
840	destroy_unused_super(s);
841	return old;
842	}
843	}
844	if (!s) {
845	spin_unlock(lock: &sb_lock);
846	s = alloc_super(type, flags, user_ns);
847	if (!s)
848	return ERR_PTR(error: -ENOMEM);
849	goto retry;
850	}
851
852	err = set(s, data);
853	if (err) {
854	spin_unlock(lock: &sb_lock);
855	destroy_unused_super(s);
856	return ERR_PTR(error: err);
857	}
858	s->s_type = type;
859	strscpy(s->s_id, type->name, sizeof(s->s_id));
860	list_add_tail(new: &s->s_list, head: &super_blocks);
861	hlist_add_head(n: &s->s_instances, h: &type->fs_supers);
862	spin_unlock(lock: &sb_lock);
863	get_filesystem(fs: type);
864	shrinker_register(shrinker: s->s_shrink);
865	return s;
866	}
867	EXPORT_SYMBOL(sget);
868
869	void drop_super(struct super_block *sb)
870	{
871	super_unlock_shared(sb);
872	put_super(sb);
873	}
874
875	EXPORT_SYMBOL(drop_super);
876
877	void drop_super_exclusive(struct super_block *sb)
878	{
879	super_unlock_excl(sb);
880	put_super(sb);
881	}
882	EXPORT_SYMBOL(drop_super_exclusive);
883
884	enum super_iter_flags_t {
885	SUPER_ITER_EXCL = (1U << 0),
886	SUPER_ITER_UNLOCKED = (1U << 1),
887	SUPER_ITER_REVERSE = (1U << 2),
888	};
889
890	static inline struct super_block *first_super(enum super_iter_flags_t flags)
891	{
892	if (flags & SUPER_ITER_REVERSE)
893	return list_last_entry(&super_blocks, struct super_block, s_list);
894	return list_first_entry(&super_blocks, struct super_block, s_list);
895	}
896
897	static inline struct super_block *next_super(struct super_block *sb,
898	enum super_iter_flags_t flags)
899	{
900	if (flags & SUPER_ITER_REVERSE)
901	return list_prev_entry(sb, s_list);
902	return list_next_entry(sb, s_list);
903	}
904
905	static void __iterate_supers(void (*f)(struct super_block *, void *), void *arg,
906	enum super_iter_flags_t flags)
907	{
908	struct super_block *sb, *p = NULL;
909	bool excl = flags & SUPER_ITER_EXCL;
910
911	guard(spinlock)(l: &sb_lock);
912
913	for (sb = first_super(flags);
914	!list_entry_is_head(sb, &super_blocks, s_list);
915	sb = next_super(sb, flags)) {
916	if (super_flags(sb, SB_DYING))
917	continue;
918	sb->s_count++;
919	spin_unlock(lock: &sb_lock);
920
921	if (flags & SUPER_ITER_UNLOCKED) {
922	f(sb, arg);
923	} else if (super_lock(sb, excl)) {
924	f(sb, arg);
925	super_unlock(sb, excl);
926	}
927
928	spin_lock(lock: &sb_lock);
929	if (p)
930	__put_super(s: p);
931	p = sb;
932	}
933	if (p)
934	__put_super(s: p);
935	}
936
937	void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
938	{
939	__iterate_supers(f, arg, flags: 0);
940	}
941
942	/**
943	* iterate_supers_type - call function for superblocks of given type
944	* @type: fs type
945	* @f: function to call
946	* @arg: argument to pass to it
947	*
948	* Scans the superblock list and calls given function, passing it
949	* locked superblock and given argument.
950	*/
951	void iterate_supers_type(struct file_system_type *type,
952	void (*f)(struct super_block *, void *), void *arg)
953	{
954	struct super_block *sb, *p = NULL;
955
956	spin_lock(lock: &sb_lock);
957	hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
958	bool locked;
959
960	if (super_flags(sb, SB_DYING))
961	continue;
962
963	sb->s_count++;
964	spin_unlock(lock: &sb_lock);
965
966	locked = super_lock_shared(sb);
967	if (locked)
968	f(sb, arg);
969
970	spin_lock(lock: &sb_lock);
971	if (p)
972	__put_super(s: p);
973	p = sb;
974	}
975	if (p)
976	__put_super(s: p);
977	spin_unlock(lock: &sb_lock);
978	}
979
980	EXPORT_SYMBOL(iterate_supers_type);
981
982	struct super_block *user_get_super(dev_t dev, bool excl)
983	{
984	struct super_block *sb;
985
986	spin_lock(lock: &sb_lock);
987	list_for_each_entry(sb, &super_blocks, s_list) {
988	bool locked;
989
990	if (sb->s_dev != dev)
991	continue;
992
993	sb->s_count++;
994	spin_unlock(lock: &sb_lock);
995
996	locked = super_lock(sb, excl);
997	if (locked)
998	return sb;
999
1000	spin_lock(lock: &sb_lock);
1001	__put_super(s: sb);
1002	break;
1003	}
1004	spin_unlock(lock: &sb_lock);
1005	return NULL;
1006	}
1007
1008	/**
1009	* reconfigure_super - asks filesystem to change superblock parameters
1010	* @fc: The superblock and configuration
1011	*
1012	* Alters the configuration parameters of a live superblock.
1013	*/
1014	int reconfigure_super(struct fs_context *fc)
1015	{
1016	struct super_block *sb = fc->root->d_sb;
1017	int retval;
1018	bool remount_ro = false;
1019	bool remount_rw = false;
1020	bool force = fc->sb_flags & SB_FORCE;
1021
1022	if (fc->sb_flags_mask & ~MS_RMT_MASK)
1023	return -EINVAL;
1024	if (sb->s_writers.frozen != SB_UNFROZEN)
1025	return -EBUSY;
1026
1027	retval = security_sb_remount(sb, mnt_opts: fc->security);
1028	if (retval)
1029	return retval;
1030
1031	if (fc->sb_flags_mask & SB_RDONLY) {
1032	#ifdef CONFIG_BLOCK
1033	if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
1034	bdev_read_only(bdev: sb->s_bdev))
1035	return -EACCES;
1036	#endif
1037	remount_rw = !(fc->sb_flags & SB_RDONLY) && sb_rdonly(sb);
1038	remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
1039	}
1040
1041	if (remount_ro) {
1042	if (!hlist_empty(h: &sb->s_pins)) {
1043	super_unlock_excl(sb);
1044	group_pin_kill(p: &sb->s_pins);
1045	__super_lock_excl(sb);
1046	if (!sb->s_root)
1047	return 0;
1048	if (sb->s_writers.frozen != SB_UNFROZEN)
1049	return -EBUSY;
1050	remount_ro = !sb_rdonly(sb);
1051	}
1052	}
1053	shrink_dcache_sb(sb);
1054
1055	/* If we are reconfiguring to RDONLY and current sb is read/write,
1056	* make sure there are no files open for writing.
1057	*/
1058	if (remount_ro) {
1059	if (force) {
1060	sb_start_ro_state_change(sb);
1061	} else {
1062	retval = sb_prepare_remount_readonly(sb);
1063	if (retval)
1064	return retval;
1065	}
1066	} else if (remount_rw) {
1067	/*
1068	* Protect filesystem's reconfigure code from writes from
1069	* userspace until reconfigure finishes.
1070	*/
1071	sb_start_ro_state_change(sb);
1072	}
1073
1074	if (fc->ops->reconfigure) {
1075	retval = fc->ops->reconfigure(fc);
1076	if (retval) {
1077	if (!force)
1078	goto cancel_readonly;
1079	/* If forced remount, go ahead despite any errors */
1080	WARN(1, "forced remount of a %s fs returned %i\n",
1081	sb->s_type->name, retval);
1082	}
1083	}
1084
1085	WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
1086	(fc->sb_flags & fc->sb_flags_mask)));
1087	sb_end_ro_state_change(sb);
1088
1089	/*
1090	* Some filesystems modify their metadata via some other path than the
1091	* bdev buffer cache (eg. use a private mapping, or directories in
1092	* pagecache, etc). Also file data modifications go via their own
1093	* mappings. So If we try to mount readonly then copy the filesystem
1094	* from bdev, we could get stale data, so invalidate it to give a best
1095	* effort at coherency.
1096	*/
1097	if (remount_ro && sb->s_bdev)
1098	invalidate_bdev(bdev: sb->s_bdev);
1099	return 0;
1100
1101	cancel_readonly:
1102	sb_end_ro_state_change(sb);
1103	return retval;
1104	}
1105
1106	static void do_emergency_remount_callback(struct super_block *sb, void *unused)
1107	{
1108	if (sb->s_bdev && !sb_rdonly(sb)) {
1109	struct fs_context *fc;
1110
1111	fc = fs_context_for_reconfigure(dentry: sb->s_root,
1112	SB_RDONLY | SB_FORCE, SB_RDONLY);
1113	if (!IS_ERR(ptr: fc)) {
1114	if (parse_monolithic_mount_data(fc, NULL) == 0)
1115	(void)reconfigure_super(fc);
1116	put_fs_context(fc);
1117	}
1118	}
1119	}
1120
1121	static void do_emergency_remount(struct work_struct *work)
1122	{
1123	__iterate_supers(f: do_emergency_remount_callback, NULL,
1124	flags: SUPER_ITER_EXCL | SUPER_ITER_REVERSE);
1125	kfree(objp: work);
1126	printk("Emergency Remount complete\n");
1127	}
1128
1129	void emergency_remount(void)
1130	{
1131	struct work_struct *work;
1132
1133	work = kmalloc(sizeof(*work), GFP_ATOMIC);
1134	if (work) {
1135	INIT_WORK(work, do_emergency_remount);
1136	schedule_work(work);
1137	}
1138	}
1139
1140	static void do_thaw_all_callback(struct super_block *sb, void *unused)
1141	{
1142	if (IS_ENABLED(CONFIG_BLOCK))
1143	while (sb->s_bdev && !bdev_thaw(bdev: sb->s_bdev))
1144	pr_warn("Emergency Thaw on %pg\n", sb->s_bdev);
1145	thaw_super_locked(sb, who: FREEZE_HOLDER_USERSPACE, NULL);
1146	return;
1147	}
1148
1149	static void do_thaw_all(struct work_struct *work)
1150	{
1151	__iterate_supers(f: do_thaw_all_callback, NULL, flags: SUPER_ITER_EXCL);
1152	kfree(objp: work);
1153	printk(KERN_WARNING "Emergency Thaw complete\n");
1154	}
1155
1156	/**
1157	* emergency_thaw_all -- forcibly thaw every frozen filesystem
1158	*
1159	* Used for emergency unfreeze of all filesystems via SysRq
1160	*/
1161	void emergency_thaw_all(void)
1162	{
1163	struct work_struct *work;
1164
1165	work = kmalloc(sizeof(*work), GFP_ATOMIC);
1166	if (work) {
1167	INIT_WORK(work, do_thaw_all);
1168	schedule_work(work);
1169	}
1170	}
1171
1172	static inline bool get_active_super(struct super_block *sb)
1173	{
1174	bool active = false;
1175
1176	if (super_lock_excl(sb)) {
1177	active = atomic_inc_not_zero(v: &sb->s_active);
1178	super_unlock_excl(sb);
1179	}
1180	return active;
1181	}
1182
1183	static const char *filesystems_freeze_ptr = "filesystems_freeze";
1184
1185	static void filesystems_freeze_callback(struct super_block *sb, void *unused)
1186	{
1187	if (!sb->s_op->freeze_fs && !sb->s_op->freeze_super)
1188	return;
1189
1190	if (!get_active_super(sb))
1191	return;
1192
1193	if (sb->s_op->freeze_super)
1194	sb->s_op->freeze_super(sb, FREEZE_EXCL | FREEZE_HOLDER_KERNEL,
1195	filesystems_freeze_ptr);
1196	else
1197	freeze_super(super: sb, who: FREEZE_EXCL | FREEZE_HOLDER_KERNEL,
1198	freeze_owner: filesystems_freeze_ptr);
1199
1200	deactivate_super(sb);
1201	}
1202
1203	void filesystems_freeze(void)
1204	{
1205	__iterate_supers(f: filesystems_freeze_callback, NULL,
1206	flags: SUPER_ITER_UNLOCKED | SUPER_ITER_REVERSE);
1207	}
1208
1209	static void filesystems_thaw_callback(struct super_block *sb, void *unused)
1210	{
1211	if (!sb->s_op->freeze_fs && !sb->s_op->freeze_super)
1212	return;
1213
1214	if (!get_active_super(sb))
1215	return;
1216
1217	if (sb->s_op->thaw_super)
1218	sb->s_op->thaw_super(sb, FREEZE_EXCL | FREEZE_HOLDER_KERNEL,
1219	filesystems_freeze_ptr);
1220	else
1221	thaw_super(super: sb, who: FREEZE_EXCL | FREEZE_HOLDER_KERNEL,
1222	freeze_owner: filesystems_freeze_ptr);
1223
1224	deactivate_super(sb);
1225	}
1226
1227	void filesystems_thaw(void)
1228	{
1229	__iterate_supers(f: filesystems_thaw_callback, NULL, flags: SUPER_ITER_UNLOCKED);
1230	}
1231
1232	static DEFINE_IDA(unnamed_dev_ida);
1233
1234	/**
1235	* get_anon_bdev - Allocate a block device for filesystems which don't have one.
1236	* @p: Pointer to a dev_t.
1237	*
1238	* Filesystems which don't use real block devices can call this function
1239	* to allocate a virtual block device.
1240	*
1241	* Context: Any context. Frequently called while holding sb_lock.
1242	* Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1243	* or -ENOMEM if memory allocation failed.
1244	*/
1245	int get_anon_bdev(dev_t *p)
1246	{
1247	int dev;
1248
1249	/*
1250	* Many userspace utilities consider an FSID of 0 invalid.
1251	* Always return at least 1 from get_anon_bdev.
1252	*/
1253	dev = ida_alloc_range(&unnamed_dev_ida, min: 1, max: (1 << MINORBITS) - 1,
1254	GFP_ATOMIC);
1255	if (dev == -ENOSPC)
1256	dev = -EMFILE;
1257	if (dev < 0)
1258	return dev;
1259
1260	*p = MKDEV(0, dev);
1261	return 0;
1262	}
1263	EXPORT_SYMBOL(get_anon_bdev);
1264
1265	void free_anon_bdev(dev_t dev)
1266	{
1267	ida_free(&unnamed_dev_ida, MINOR(dev));
1268	}
1269	EXPORT_SYMBOL(free_anon_bdev);
1270
1271	int set_anon_super(struct super_block *s, void *data)
1272	{
1273	return get_anon_bdev(&s->s_dev);
1274	}
1275	EXPORT_SYMBOL(set_anon_super);
1276
1277	void kill_anon_super(struct super_block *sb)
1278	{
1279	dev_t dev = sb->s_dev;
1280	generic_shutdown_super(sb);
1281	kill_super_notify(sb);
1282	free_anon_bdev(dev);
1283	}
1284	EXPORT_SYMBOL(kill_anon_super);
1285
1286	void kill_litter_super(struct super_block *sb)
1287	{
1288	if (sb->s_root)
1289	d_genocide(sb->s_root);
1290	kill_anon_super(sb);
1291	}
1292	EXPORT_SYMBOL(kill_litter_super);
1293
1294	int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1295	{
1296	return set_anon_super(sb, NULL);
1297	}
1298	EXPORT_SYMBOL(set_anon_super_fc);
1299
1300	static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1301	{
1302	return sb->s_fs_info == fc->s_fs_info;
1303	}
1304
1305	static int test_single_super(struct super_block *s, struct fs_context *fc)
1306	{
1307	return 1;
1308	}
1309
1310	static int vfs_get_super(struct fs_context *fc,
1311	int (*test)(struct super_block *, struct fs_context *),
1312	int (*fill_super)(struct super_block *sb,
1313	struct fs_context *fc))
1314	{
1315	struct super_block *sb;
1316	int err;
1317
1318	sb = sget_fc(fc, test, set_anon_super_fc);
1319	if (IS_ERR(ptr: sb))
1320	return PTR_ERR(ptr: sb);
1321
1322	if (!sb->s_root) {
1323	err = fill_super(sb, fc);
1324	if (err)
1325	goto error;
1326
1327	sb->s_flags |= SB_ACTIVE;
1328	}
1329
1330	fc->root = dget(dentry: sb->s_root);
1331	return 0;
1332
1333	error:
1334	deactivate_locked_super(sb);
1335	return err;
1336	}
1337
1338	int get_tree_nodev(struct fs_context *fc,
1339	int (*fill_super)(struct super_block *sb,
1340	struct fs_context *fc))
1341	{
1342	return vfs_get_super(fc, NULL, fill_super);
1343	}
1344	EXPORT_SYMBOL(get_tree_nodev);
1345
1346	int get_tree_single(struct fs_context *fc,
1347	int (*fill_super)(struct super_block *sb,
1348	struct fs_context *fc))
1349	{
1350	return vfs_get_super(fc, test: test_single_super, fill_super);
1351	}
1352	EXPORT_SYMBOL(get_tree_single);
1353
1354	int get_tree_keyed(struct fs_context *fc,
1355	int (*fill_super)(struct super_block *sb,
1356	struct fs_context *fc),
1357	void *key)
1358	{
1359	fc->s_fs_info = key;
1360	return vfs_get_super(fc, test: test_keyed_super, fill_super);
1361	}
1362	EXPORT_SYMBOL(get_tree_keyed);
1363
1364	static int set_bdev_super(struct super_block *s, void *data)
1365	{
1366	s->s_dev = *(dev_t *)data;
1367	return 0;
1368	}
1369
1370	static int super_s_dev_set(struct super_block *s, struct fs_context *fc)
1371	{
1372	return set_bdev_super(s, data: fc->sget_key);
1373	}
1374
1375	static int super_s_dev_test(struct super_block *s, struct fs_context *fc)
1376	{
1377	return !(s->s_iflags & SB_I_RETIRED) &&
1378	s->s_dev == *(dev_t *)fc->sget_key;
1379	}
1380
1381	/**
1382	* sget_dev - Find or create a superblock by device number
1383	* @fc: Filesystem context.
1384	* @dev: device number
1385	*
1386	* Find or create a superblock using the provided device number that
1387	* will be stored in fc->sget_key.
1388	*
1389	* If an extant superblock is matched, then that will be returned with
1390	* an elevated reference count that the caller must transfer or discard.
1391	*
1392	* If no match is made, a new superblock will be allocated and basic
1393	* initialisation will be performed (s_type, s_fs_info, s_id, s_dev will
1394	* be set). The superblock will be published and it will be returned in
1395	* a partially constructed state with SB_BORN and SB_ACTIVE as yet
1396	* unset.
1397	*
1398	* Return: an existing or newly created superblock on success, an error
1399	* pointer on failure.
1400	*/
1401	struct super_block *sget_dev(struct fs_context *fc, dev_t dev)
1402	{
1403	fc->sget_key = &dev;
1404	return sget_fc(fc, super_s_dev_test, super_s_dev_set);
1405	}
1406	EXPORT_SYMBOL(sget_dev);
1407
1408	#ifdef CONFIG_BLOCK
1409	/*
1410	* Lock the superblock that is holder of the bdev. Returns the superblock
1411	* pointer if we successfully locked the superblock and it is alive. Otherwise
1412	* we return NULL and just unlock bdev->bd_holder_lock.
1413	*
1414	* The function must be called with bdev->bd_holder_lock and releases it.
1415	*/
1416	static struct super_block *bdev_super_lock(struct block_device *bdev, bool excl)
1417	__releases(&bdev->bd_holder_lock)
1418	{
1419	struct super_block *sb = bdev->bd_holder;
1420	bool locked;
1421
1422	lockdep_assert_held(&bdev->bd_holder_lock);
1423	lockdep_assert_not_held(&sb->s_umount);
1424	lockdep_assert_not_held(&bdev->bd_disk->open_mutex);
1425
1426	/* Make sure sb doesn't go away from under us */
1427	spin_lock(lock: &sb_lock);
1428	sb->s_count++;
1429	spin_unlock(lock: &sb_lock);
1430
1431	mutex_unlock(lock: &bdev->bd_holder_lock);
1432
1433	locked = super_lock(sb, excl);
1434
1435	/*
1436	* If the superblock wasn't already SB_DYING then we hold
1437	* s_umount and can safely drop our temporary reference.
1438	*/
1439	put_super(sb);
1440
1441	if (!locked)
1442	return NULL;
1443
1444	if (!sb->s_root || !(sb->s_flags & SB_ACTIVE)) {
1445	super_unlock(sb, excl);
1446	return NULL;
1447	}
1448
1449	return sb;
1450	}
1451
1452	static void fs_bdev_mark_dead(struct block_device *bdev, bool surprise)
1453	{
1454	struct super_block *sb;
1455
1456	sb = bdev_super_lock(bdev, excl: false);
1457	if (!sb)
1458	return;
1459
1460	if (!surprise)
1461	sync_filesystem(sb);
1462	shrink_dcache_sb(sb);
1463	evict_inodes(sb);
1464	if (sb->s_op->shutdown)
1465	sb->s_op->shutdown(sb);
1466
1467	super_unlock_shared(sb);
1468	}
1469
1470	static void fs_bdev_sync(struct block_device *bdev)
1471	{
1472	struct super_block *sb;
1473
1474	sb = bdev_super_lock(bdev, excl: false);
1475	if (!sb)
1476	return;
1477
1478	sync_filesystem(sb);
1479	super_unlock_shared(sb);
1480	}
1481
1482	static struct super_block *get_bdev_super(struct block_device *bdev)
1483	{
1484	bool active = false;
1485	struct super_block *sb;
1486
1487	sb = bdev_super_lock(bdev, excl: true);
1488	if (sb) {
1489	active = atomic_inc_not_zero(v: &sb->s_active);
1490	super_unlock_excl(sb);
1491	}
1492	if (!active)
1493	return NULL;
1494	return sb;
1495	}
1496
1497	/**
1498	* fs_bdev_freeze - freeze owning filesystem of block device
1499	* @bdev: block device
1500	*
1501	* Freeze the filesystem that owns this block device if it is still
1502	* active.
1503	*
1504	* A filesystem that owns multiple block devices may be frozen from each
1505	* block device and won't be unfrozen until all block devices are
1506	* unfrozen. Each block device can only freeze the filesystem once as we
1507	* nest freezes for block devices in the block layer.
1508	*
1509	* Return: If the freeze was successful zero is returned. If the freeze
1510	* failed a negative error code is returned.
1511	*/
1512	static int fs_bdev_freeze(struct block_device *bdev)
1513	{
1514	struct super_block *sb;
1515	int error = 0;
1516
1517	lockdep_assert_held(&bdev->bd_fsfreeze_mutex);
1518
1519	sb = get_bdev_super(bdev);
1520	if (!sb)
1521	return -EINVAL;
1522
1523	if (sb->s_op->freeze_super)
1524	error = sb->s_op->freeze_super(sb,
1525	FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE, NULL);
1526	else
1527	error = freeze_super(super: sb,
1528	who: FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE, NULL);
1529	if (!error)
1530	error = sync_blockdev(bdev);
1531	deactivate_super(sb);
1532	return error;
1533	}
1534
1535	/**
1536	* fs_bdev_thaw - thaw owning filesystem of block device
1537	* @bdev: block device
1538	*
1539	* Thaw the filesystem that owns this block device.
1540	*
1541	* A filesystem that owns multiple block devices may be frozen from each
1542	* block device and won't be unfrozen until all block devices are
1543	* unfrozen. Each block device can only freeze the filesystem once as we
1544	* nest freezes for block devices in the block layer.
1545	*
1546	* Return: If the thaw was successful zero is returned. If the thaw
1547	* failed a negative error code is returned. If this function
1548	* returns zero it doesn't mean that the filesystem is unfrozen
1549	* as it may have been frozen multiple times (kernel may hold a
1550	* freeze or might be frozen from other block devices).
1551	*/
1552	static int fs_bdev_thaw(struct block_device *bdev)
1553	{
1554	struct super_block *sb;
1555	int error;
1556
1557	lockdep_assert_held(&bdev->bd_fsfreeze_mutex);
1558
1559	/*
1560	* The block device may have been frozen before it was claimed by a
1561	* filesystem. Concurrently another process might try to mount that
1562	* frozen block device and has temporarily claimed the block device for
1563	* that purpose causing a concurrent fs_bdev_thaw() to end up here. The
1564	* mounter is already about to abort mounting because they still saw an
1565	* elevanted bdev->bd_fsfreeze_count so get_bdev_super() will return
1566	* NULL in that case.
1567	*/
1568	sb = get_bdev_super(bdev);
1569	if (!sb)
1570	return -EINVAL;
1571
1572	if (sb->s_op->thaw_super)
1573	error = sb->s_op->thaw_super(sb,
1574	FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE, NULL);
1575	else
1576	error = thaw_super(super: sb,
1577	who: FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE, NULL);
1578	deactivate_super(sb);
1579	return error;
1580	}
1581
1582	const struct blk_holder_ops fs_holder_ops = {
1583	.mark_dead = fs_bdev_mark_dead,
1584	.sync = fs_bdev_sync,
1585	.freeze = fs_bdev_freeze,
1586	.thaw = fs_bdev_thaw,
1587	};
1588	EXPORT_SYMBOL_GPL(fs_holder_ops);
1589
1590	int setup_bdev_super(struct super_block *sb, int sb_flags,
1591	struct fs_context *fc)
1592	{
1593	blk_mode_t mode = sb_open_mode(sb_flags);
1594	struct file *bdev_file;
1595	struct block_device *bdev;
1596
1597	bdev_file = bdev_file_open_by_dev(dev: sb->s_dev, mode, holder: sb, hops: &fs_holder_ops);
1598	if (IS_ERR(ptr: bdev_file)) {
1599	if (fc)
1600	errorf(fc, "%s: Can't open blockdev", fc->source);
1601	return PTR_ERR(ptr: bdev_file);
1602	}
1603	bdev = file_bdev(bdev_file);
1604
1605	/*
1606	* This really should be in blkdev_get_by_dev, but right now can't due
1607	* to legacy issues that require us to allow opening a block device node
1608	* writable from userspace even for a read-only block device.
1609	*/
1610	if ((mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
1611	bdev_fput(bdev_file);
1612	return -EACCES;
1613	}
1614
1615	/*
1616	* It is enough to check bdev was not frozen before we set
1617	* s_bdev as freezing will wait until SB_BORN is set.
1618	*/
1619	if (atomic_read(v: &bdev->bd_fsfreeze_count) > 0) {
1620	if (fc)
1621	warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1622	bdev_fput(bdev_file);
1623	return -EBUSY;
1624	}
1625	spin_lock(lock: &sb_lock);
1626	sb->s_bdev_file = bdev_file;
1627	sb->s_bdev = bdev;
1628	sb->s_bdi = bdi_get(bdi: bdev->bd_disk->bdi);
1629	if (bdev_stable_writes(bdev))
1630	sb->s_iflags |= SB_I_STABLE_WRITES;
1631	spin_unlock(lock: &sb_lock);
1632
1633	snprintf(buf: sb->s_id, size: sizeof(sb->s_id), fmt: "%pg", bdev);
1634	shrinker_debugfs_rename(shrinker: sb->s_shrink, fmt: "sb-%s:%s", sb->s_type->name,
1635	sb->s_id);
1636	sb_set_blocksize(sb, block_size(bdev));
1637	return 0;
1638	}
1639	EXPORT_SYMBOL_GPL(setup_bdev_super);
1640
1641	/**
1642	* get_tree_bdev_flags - Get a superblock based on a single block device
1643	* @fc: The filesystem context holding the parameters
1644	* @fill_super: Helper to initialise a new superblock
1645	* @flags: GET_TREE_BDEV_* flags
1646	*/
1647	int get_tree_bdev_flags(struct fs_context *fc,
1648	int (*fill_super)(struct super_block *sb,
1649	struct fs_context *fc), unsigned int flags)
1650	{
1651	struct super_block *s;
1652	int error = 0;
1653	dev_t dev;
1654
1655	if (!fc->source)
1656	return invalf(fc, "No source specified");
1657
1658	error = lookup_bdev(pathname: fc->source, dev: &dev);
1659	if (error) {
1660	if (!(flags & GET_TREE_BDEV_QUIET_LOOKUP))
1661	errorf(fc, "%s: Can't lookup blockdev", fc->source);
1662	return error;
1663	}
1664	fc->sb_flags |= SB_NOSEC;
1665	s = sget_dev(fc, dev);
1666	if (IS_ERR(ptr: s))
1667	return PTR_ERR(ptr: s);
1668
1669	if (s->s_root) {
1670	/* Don't summarily change the RO/RW state. */
1671	if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1672	warnf(fc, "%pg: Can't mount, would change RO state", s->s_bdev);
1673	deactivate_locked_super(s);
1674	return -EBUSY;
1675	}
1676	} else {
1677	error = setup_bdev_super(s, fc->sb_flags, fc);
1678	if (!error)
1679	error = fill_super(s, fc);
1680	if (error) {
1681	deactivate_locked_super(s);
1682	return error;
1683	}
1684	s->s_flags |= SB_ACTIVE;
1685	}
1686
1687	BUG_ON(fc->root);
1688	fc->root = dget(dentry: s->s_root);
1689	return 0;
1690	}
1691	EXPORT_SYMBOL_GPL(get_tree_bdev_flags);
1692
1693	/**
1694	* get_tree_bdev - Get a superblock based on a single block device
1695	* @fc: The filesystem context holding the parameters
1696	* @fill_super: Helper to initialise a new superblock
1697	*/
1698	int get_tree_bdev(struct fs_context *fc,
1699	int (*fill_super)(struct super_block *,
1700	struct fs_context *))
1701	{
1702	return get_tree_bdev_flags(fc, fill_super, 0);
1703	}
1704	EXPORT_SYMBOL(get_tree_bdev);
1705
1706	static int test_bdev_super(struct super_block *s, void *data)
1707	{
1708	return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data;
1709	}
1710
1711	struct dentry *mount_bdev(struct file_system_type *fs_type,
1712	int flags, const char *dev_name, void *data,
1713	int (*fill_super)(struct super_block *, void *, int))
1714	{
1715	struct super_block *s;
1716	int error;
1717	dev_t dev;
1718
1719	error = lookup_bdev(pathname: dev_name, dev: &dev);
1720	if (error)
1721	return ERR_PTR(error);
1722
1723	flags |= SB_NOSEC;
1724	s = sget(fs_type, test_bdev_super, set_bdev_super, flags, &dev);
1725	if (IS_ERR(ptr: s))
1726	return ERR_CAST(ptr: s);
1727
1728	if (s->s_root) {
1729	if ((flags ^ s->s_flags) & SB_RDONLY) {
1730	deactivate_locked_super(s);
1731	return ERR_PTR(error: -EBUSY);
1732	}
1733	} else {
1734	error = setup_bdev_super(s, flags, NULL);
1735	if (!error)
1736	error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1737	if (error) {
1738	deactivate_locked_super(s);
1739	return ERR_PTR(error);
1740	}
1741
1742	s->s_flags |= SB_ACTIVE;
1743	}
1744
1745	return dget(dentry: s->s_root);
1746	}
1747	EXPORT_SYMBOL(mount_bdev);
1748
1749	void kill_block_super(struct super_block *sb)
1750	{
1751	struct block_device *bdev = sb->s_bdev;
1752
1753	generic_shutdown_super(sb);
1754	if (bdev) {
1755	sync_blockdev(bdev);
1756	bdev_fput(bdev_file: sb->s_bdev_file);
1757	}
1758	}
1759
1760	EXPORT_SYMBOL(kill_block_super);
1761	#endif
1762
1763	struct dentry *mount_nodev(struct file_system_type *fs_type,
1764	int flags, void *data,
1765	int (*fill_super)(struct super_block *, void *, int))
1766	{
1767	int error;
1768	struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1769
1770	if (IS_ERR(ptr: s))
1771	return ERR_CAST(ptr: s);
1772
1773	error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1774	if (error) {
1775	deactivate_locked_super(s);
1776	return ERR_PTR(error);
1777	}
1778	s->s_flags |= SB_ACTIVE;
1779	return dget(dentry: s->s_root);
1780	}
1781	EXPORT_SYMBOL(mount_nodev);
1782
1783	/**
1784	* vfs_get_tree - Get the mountable root
1785	* @fc: The superblock configuration context.
1786	*
1787	* The filesystem is invoked to get or create a superblock which can then later
1788	* be used for mounting. The filesystem places a pointer to the root to be
1789	* used for mounting in @fc->root.
1790	*/
1791	int vfs_get_tree(struct fs_context *fc)
1792	{
1793	struct super_block *sb;
1794	int error;
1795
1796	if (fc->root)
1797	return -EBUSY;
1798
1799	/* Get the mountable root in fc->root, with a ref on the root and a ref
1800	* on the superblock.
1801	*/
1802	error = fc->ops->get_tree(fc);
1803	if (error < 0)
1804	return error;
1805
1806	if (!fc->root) {
1807	pr_err("Filesystem %s get_tree() didn't set fc->root, returned %i\n",
1808	fc->fs_type->name, error);
1809	/* We don't know what the locking state of the superblock is -
1810	* if there is a superblock.
1811	*/
1812	BUG();
1813	}
1814
1815	sb = fc->root->d_sb;
1816	WARN_ON(!sb->s_bdi);
1817
1818	/*
1819	* super_wake() contains a memory barrier which also care of
1820	* ordering for super_cache_count(). We place it before setting
1821	* SB_BORN as the data dependency between the two functions is
1822	* the superblock structure contents that we just set up, not
1823	* the SB_BORN flag.
1824	*/
1825	super_wake(sb, SB_BORN);
1826
1827	error = security_sb_set_mnt_opts(sb, mnt_opts: fc->security, kern_flags: 0, NULL);
1828	if (unlikely(error)) {
1829	fc_drop_locked(fc);
1830	return error;
1831	}
1832
1833	/*
1834	* filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1835	* but s_maxbytes was an unsigned long long for many releases. Throw
1836	* this warning for a little while to try and catch filesystems that
1837	* violate this rule.
1838	*/
1839	WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1840	"negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1841
1842	return 0;
1843	}
1844	EXPORT_SYMBOL(vfs_get_tree);
1845
1846	/*
1847	* Setup private BDI for given superblock. It gets automatically cleaned up
1848	* in generic_shutdown_super().
1849	*/
1850	int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1851	{
1852	struct backing_dev_info *bdi;
1853	int err;
1854	va_list args;
1855
1856	bdi = bdi_alloc(NUMA_NO_NODE);
1857	if (!bdi)
1858	return -ENOMEM;
1859
1860	va_start(args, fmt);
1861	err = bdi_register_va(bdi, fmt, args);
1862	va_end(args);
1863	if (err) {
1864	bdi_put(bdi);
1865	return err;
1866	}
1867	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1868	sb->s_bdi = bdi;
1869	sb->s_iflags |= SB_I_PERSB_BDI;
1870
1871	return 0;
1872	}
1873	EXPORT_SYMBOL(super_setup_bdi_name);
1874
1875	/*
1876	* Setup private BDI for given superblock. I gets automatically cleaned up
1877	* in generic_shutdown_super().
1878	*/
1879	int super_setup_bdi(struct super_block *sb)
1880	{
1881	static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1882
1883	return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1884	atomic_long_inc_return(v: &bdi_seq));
1885	}
1886	EXPORT_SYMBOL(super_setup_bdi);
1887
1888	/**
1889	* sb_wait_write - wait until all writers to given file system finish
1890	* @sb: the super for which we wait
1891	* @level: type of writers we wait for (normal vs page fault)
1892	*
1893	* This function waits until there are no writers of given type to given file
1894	* system.
1895	*/
1896	static void sb_wait_write(struct super_block *sb, int level)
1897	{
1898	percpu_down_write(sb->s_writers.rw_sem + level-1);
1899	}
1900
1901	/*
1902	* We are going to return to userspace and forget about these locks, the
1903	* ownership goes to the caller of thaw_super() which does unlock().
1904	*/
1905	static void lockdep_sb_freeze_release(struct super_block *sb)
1906	{
1907	int level;
1908
1909	for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1910	percpu_rwsem_release(sem: sb->s_writers.rw_sem + level, _THIS_IP_);
1911	}
1912
1913	/*
1914	* Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1915	*/
1916	static void lockdep_sb_freeze_acquire(struct super_block *sb)
1917	{
1918	int level;
1919
1920	for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1921	percpu_rwsem_acquire(sem: sb->s_writers.rw_sem + level, read: 0, _THIS_IP_);
1922	}
1923
1924	static void sb_freeze_unlock(struct super_block *sb, int level)
1925	{
1926	for (level--; level >= 0; level--)
1927	percpu_up_write(sb->s_writers.rw_sem + level);
1928	}
1929
1930	static int wait_for_partially_frozen(struct super_block *sb)
1931	{
1932	int ret = 0;
1933
1934	do {
1935	unsigned short old = sb->s_writers.frozen;
1936
1937	up_write(sem: &sb->s_umount);
1938	ret = wait_var_event_killable(&sb->s_writers.frozen,
1939	sb->s_writers.frozen != old);
1940	down_write(sem: &sb->s_umount);
1941	} while (ret == 0 &&
1942	sb->s_writers.frozen != SB_UNFROZEN &&
1943	sb->s_writers.frozen != SB_FREEZE_COMPLETE);
1944
1945	return ret;
1946	}
1947
1948	#define FREEZE_HOLDERS (FREEZE_HOLDER_KERNEL | FREEZE_HOLDER_USERSPACE)
1949	#define FREEZE_FLAGS (FREEZE_HOLDERS | FREEZE_MAY_NEST | FREEZE_EXCL)
1950
1951	static inline int freeze_inc(struct super_block *sb, enum freeze_holder who)
1952	{
1953	WARN_ON_ONCE((who & ~FREEZE_FLAGS));
1954	WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
1955
1956	if (who & FREEZE_HOLDER_KERNEL)
1957	++sb->s_writers.freeze_kcount;
1958	if (who & FREEZE_HOLDER_USERSPACE)
1959	++sb->s_writers.freeze_ucount;
1960	return sb->s_writers.freeze_kcount + sb->s_writers.freeze_ucount;
1961	}
1962
1963	static inline int freeze_dec(struct super_block *sb, enum freeze_holder who)
1964	{
1965	WARN_ON_ONCE((who & ~FREEZE_FLAGS));
1966	WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
1967
1968	if ((who & FREEZE_HOLDER_KERNEL) && sb->s_writers.freeze_kcount)
1969	--sb->s_writers.freeze_kcount;
1970	if ((who & FREEZE_HOLDER_USERSPACE) && sb->s_writers.freeze_ucount)
1971	--sb->s_writers.freeze_ucount;
1972	return sb->s_writers.freeze_kcount + sb->s_writers.freeze_ucount;
1973	}
1974
1975	static inline bool may_freeze(struct super_block *sb, enum freeze_holder who,
1976	const void *freeze_owner)
1977	{
1978	lockdep_assert_held(&sb->s_umount);
1979
1980	WARN_ON_ONCE((who & ~FREEZE_FLAGS));
1981	WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
1982
1983	if (who & FREEZE_EXCL) {
1984	if (WARN_ON_ONCE(!(who & FREEZE_HOLDER_KERNEL)))
1985	return false;
1986	if (WARN_ON_ONCE(who & ~(FREEZE_EXCL | FREEZE_HOLDER_KERNEL)))
1987	return false;
1988	if (WARN_ON_ONCE(!freeze_owner))
1989	return false;
1990	/* This freeze already has a specific owner. */
1991	if (sb->s_writers.freeze_owner)
1992	return false;
1993	/*
1994	* This is already frozen multiple times so we're just
1995	* going to take a reference count and mark the freeze as
1996	* being owned by the caller.
1997	*/
1998	if (sb->s_writers.freeze_kcount + sb->s_writers.freeze_ucount)
1999	sb->s_writers.freeze_owner = freeze_owner;
2000	return true;
2001	}
2002
2003	if (who & FREEZE_HOLDER_KERNEL)
2004	return (who & FREEZE_MAY_NEST) ||
2005	sb->s_writers.freeze_kcount == 0;
2006	if (who & FREEZE_HOLDER_USERSPACE)
2007	return (who & FREEZE_MAY_NEST) ||
2008	sb->s_writers.freeze_ucount == 0;
2009	return false;
2010	}
2011
2012	static inline bool may_unfreeze(struct super_block *sb, enum freeze_holder who,
2013	const void *freeze_owner)
2014	{
2015	lockdep_assert_held(&sb->s_umount);
2016
2017	WARN_ON_ONCE((who & ~FREEZE_FLAGS));
2018	WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
2019
2020	if (who & FREEZE_EXCL) {
2021	if (WARN_ON_ONCE(!(who & FREEZE_HOLDER_KERNEL)))
2022	return false;
2023	if (WARN_ON_ONCE(who & ~(FREEZE_EXCL | FREEZE_HOLDER_KERNEL)))
2024	return false;
2025	if (WARN_ON_ONCE(!freeze_owner))
2026	return false;
2027	if (WARN_ON_ONCE(sb->s_writers.freeze_kcount == 0))
2028	return false;
2029	/* This isn't exclusively frozen. */
2030	if (!sb->s_writers.freeze_owner)
2031	return false;
2032	/* This isn't exclusively frozen by us. */
2033	if (sb->s_writers.freeze_owner != freeze_owner)
2034	return false;
2035	/*
2036	* This is still frozen multiple times so we're just
2037	* going to drop our reference count and undo our
2038	* exclusive freeze.
2039	*/
2040	if ((sb->s_writers.freeze_kcount + sb->s_writers.freeze_ucount) > 1)
2041	sb->s_writers.freeze_owner = NULL;
2042	return true;
2043	}
2044
2045	if (who & FREEZE_HOLDER_KERNEL) {
2046	/*
2047	* Someone's trying to steal the reference belonging to
2048	* @sb->s_writers.freeze_owner.
2049	*/
2050	if (sb->s_writers.freeze_kcount == 1 &&
2051	sb->s_writers.freeze_owner)
2052	return false;
2053	return sb->s_writers.freeze_kcount > 0;
2054	}
2055
2056	if (who & FREEZE_HOLDER_USERSPACE)
2057	return sb->s_writers.freeze_ucount > 0;
2058
2059	return false;
2060	}
2061
2062	/**
2063	* freeze_super - lock the filesystem and force it into a consistent state
2064	* @sb: the super to lock
2065	* @who: context that wants to freeze
2066	* @freeze_owner: owner of the freeze
2067	*
2068	* Syncs the super to make sure the filesystem is consistent and calls the fs's
2069	* freeze_fs. Subsequent calls to this without first thawing the fs may return
2070	* -EBUSY.
2071	*
2072	* @who should be:
2073	* * %FREEZE_HOLDER_USERSPACE if userspace wants to freeze the fs;
2074	* * %FREEZE_HOLDER_KERNEL if the kernel wants to freeze the fs.
2075	* * %FREEZE_MAY_NEST whether nesting freeze and thaw requests is allowed.
2076	*
2077	* The @who argument distinguishes between the kernel and userspace trying to
2078	* freeze the filesystem. Although there cannot be multiple kernel freezes or
2079	* multiple userspace freezes in effect at any given time, the kernel and
2080	* userspace can both hold a filesystem frozen. The filesystem remains frozen
2081	* until there are no kernel or userspace freezes in effect.
2082	*
2083	* A filesystem may hold multiple devices and thus a filesystems may be
2084	* frozen through the block layer via multiple block devices. In this
2085	* case the request is marked as being allowed to nest by passing
2086	* FREEZE_MAY_NEST. The filesystem remains frozen until all block
2087	* devices are unfrozen. If multiple freezes are attempted without
2088	* FREEZE_MAY_NEST -EBUSY will be returned.
2089	*
2090	* During this function, sb->s_writers.frozen goes through these values:
2091	*
2092	* SB_UNFROZEN: File system is normal, all writes progress as usual.
2093	*
2094	* SB_FREEZE_WRITE: The file system is in the process of being frozen. New
2095	* writes should be blocked, though page faults are still allowed. We wait for
2096	* all writes to complete and then proceed to the next stage.
2097	*
2098	* SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
2099	* but internal fs threads can still modify the filesystem (although they
2100	* should not dirty new pages or inodes), writeback can run etc. After waiting
2101	* for all running page faults we sync the filesystem which will clean all
2102	* dirty pages and inodes (no new dirty pages or inodes can be created when
2103	* sync is running).
2104	*
2105	* SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
2106	* modification are blocked (e.g. XFS preallocation truncation on inode
2107	* reclaim). This is usually implemented by blocking new transactions for
2108	* filesystems that have them and need this additional guard. After all
2109	* internal writers are finished we call ->freeze_fs() to finish filesystem
2110	* freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
2111	* mostly auxiliary for filesystems to verify they do not modify frozen fs.
2112	*
2113	* sb->s_writers.frozen is protected by sb->s_umount.
2114	*
2115	* Return: If the freeze was successful zero is returned. If the freeze
2116	* failed a negative error code is returned.
2117	*/
2118	int freeze_super(struct super_block *sb, enum freeze_holder who, const void *freeze_owner)
2119	{
2120	int ret;
2121
2122	if (!super_lock_excl(sb)) {
2123	WARN_ON_ONCE("Dying superblock while freezing!");
2124	return -EINVAL;
2125	}
2126	atomic_inc(v: &sb->s_active);
2127
2128	retry:
2129	if (sb->s_writers.frozen == SB_FREEZE_COMPLETE) {
2130	if (may_freeze(sb, who, freeze_owner))
2131	ret = !!WARN_ON_ONCE(freeze_inc(sb, who) == 1);
2132	else
2133	ret = -EBUSY;
2134	/* All freezers share a single active reference. */
2135	deactivate_locked_super(sb);
2136	return ret;
2137	}
2138
2139	if (sb->s_writers.frozen != SB_UNFROZEN) {
2140	ret = wait_for_partially_frozen(sb);
2141	if (ret) {
2142	deactivate_locked_super(sb);
2143	return ret;
2144	}
2145
2146	goto retry;
2147	}
2148
2149	if (sb_rdonly(sb)) {
2150	/* Nothing to do really... */
2151	WARN_ON_ONCE(freeze_inc(sb, who) > 1);
2152	sb->s_writers.freeze_owner = freeze_owner;
2153	sb->s_writers.frozen = SB_FREEZE_COMPLETE;
2154	wake_up_var(var: &sb->s_writers.frozen);
2155	super_unlock_excl(sb);
2156	return 0;
2157	}
2158
2159	sb->s_writers.frozen = SB_FREEZE_WRITE;
2160	/* Release s_umount to preserve sb_start_write -> s_umount ordering */
2161	super_unlock_excl(sb);
2162	sb_wait_write(sb, level: SB_FREEZE_WRITE);
2163	__super_lock_excl(sb);
2164
2165	/* Now we go and block page faults... */
2166	sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
2167	sb_wait_write(sb, level: SB_FREEZE_PAGEFAULT);
2168
2169	/* All writers are done so after syncing there won't be dirty data */
2170	ret = sync_filesystem(sb);
2171	if (ret) {
2172	sb->s_writers.frozen = SB_UNFROZEN;
2173	sb_freeze_unlock(sb, level: SB_FREEZE_PAGEFAULT);
2174	wake_up_var(var: &sb->s_writers.frozen);
2175	deactivate_locked_super(sb);
2176	return ret;
2177	}
2178
2179	/* Now wait for internal filesystem counter */
2180	sb->s_writers.frozen = SB_FREEZE_FS;
2181	sb_wait_write(sb, level: SB_FREEZE_FS);
2182
2183	if (sb->s_op->freeze_fs) {
2184	ret = sb->s_op->freeze_fs(sb);
2185	if (ret) {
2186	printk(KERN_ERR
2187	"VFS:Filesystem freeze failed\n");
2188	sb->s_writers.frozen = SB_UNFROZEN;
2189	sb_freeze_unlock(sb, level: SB_FREEZE_FS);
2190	wake_up_var(var: &sb->s_writers.frozen);
2191	deactivate_locked_super(sb);
2192	return ret;
2193	}
2194	}
2195	/*
2196	* For debugging purposes so that fs can warn if it sees write activity
2197	* when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
2198	*/
2199	WARN_ON_ONCE(freeze_inc(sb, who) > 1);
2200	sb->s_writers.freeze_owner = freeze_owner;
2201	sb->s_writers.frozen = SB_FREEZE_COMPLETE;
2202	wake_up_var(var: &sb->s_writers.frozen);
2203	lockdep_sb_freeze_release(sb);
2204	super_unlock_excl(sb);
2205	return 0;
2206	}
2207	EXPORT_SYMBOL(freeze_super);
2208
2209	/*
2210	* Undoes the effect of a freeze_super_locked call. If the filesystem is
2211	* frozen both by userspace and the kernel, a thaw call from either source
2212	* removes that state without releasing the other state or unlocking the
2213	* filesystem.
2214	*/
2215	static int thaw_super_locked(struct super_block *sb, enum freeze_holder who,
2216	const void *freeze_owner)
2217	{
2218	int error = -EINVAL;
2219
2220	if (sb->s_writers.frozen != SB_FREEZE_COMPLETE)
2221	goto out_unlock;
2222
2223	if (!may_unfreeze(sb, who, freeze_owner))
2224	goto out_unlock;
2225
2226	/*
2227	* All freezers share a single active reference.
2228	* So just unlock in case there are any left.
2229	*/
2230	if (freeze_dec(sb, who))
2231	goto out_unlock;
2232
2233	if (sb_rdonly(sb)) {
2234	sb->s_writers.frozen = SB_UNFROZEN;
2235	sb->s_writers.freeze_owner = NULL;
2236	wake_up_var(var: &sb->s_writers.frozen);
2237	goto out_deactivate;
2238	}
2239
2240	lockdep_sb_freeze_acquire(sb);
2241
2242	if (sb->s_op->unfreeze_fs) {
2243	error = sb->s_op->unfreeze_fs(sb);
2244	if (error) {
2245	pr_err("VFS: Filesystem thaw failed\n");
2246	freeze_inc(sb, who);
2247	lockdep_sb_freeze_release(sb);
2248	goto out_unlock;
2249	}
2250	}
2251
2252	sb->s_writers.frozen = SB_UNFROZEN;
2253	sb->s_writers.freeze_owner = NULL;
2254	wake_up_var(var: &sb->s_writers.frozen);
2255	sb_freeze_unlock(sb, level: SB_FREEZE_FS);
2256	out_deactivate:
2257	deactivate_locked_super(sb);
2258	return 0;
2259
2260	out_unlock:
2261	super_unlock_excl(sb);
2262	return error;
2263	}
2264
2265	/**
2266	* thaw_super -- unlock filesystem
2267	* @sb: the super to thaw
2268	* @who: context that wants to freeze
2269	* @freeze_owner: owner of the freeze
2270	*
2271	* Unlocks the filesystem and marks it writeable again after freeze_super()
2272	* if there are no remaining freezes on the filesystem.
2273	*
2274	* @who should be:
2275	* * %FREEZE_HOLDER_USERSPACE if userspace wants to thaw the fs;
2276	* * %FREEZE_HOLDER_KERNEL if the kernel wants to thaw the fs.
2277	* * %FREEZE_MAY_NEST whether nesting freeze and thaw requests is allowed
2278	*
2279	* A filesystem may hold multiple devices and thus a filesystems may
2280	* have been frozen through the block layer via multiple block devices.
2281	* The filesystem remains frozen until all block devices are unfrozen.
2282	*/
2283	int thaw_super(struct super_block *sb, enum freeze_holder who,
2284	const void *freeze_owner)
2285	{
2286	if (!super_lock_excl(sb)) {
2287	WARN_ON_ONCE("Dying superblock while thawing!");
2288	return -EINVAL;
2289	}
2290	return thaw_super_locked(sb, who, freeze_owner);
2291	}
2292	EXPORT_SYMBOL(thaw_super);
2293
2294	/*
2295	* Create workqueue for deferred direct IO completions. We allocate the
2296	* workqueue when it's first needed. This avoids creating workqueue for
2297	* filesystems that don't need it and also allows us to create the workqueue
2298	* late enough so the we can include s_id in the name of the workqueue.
2299	*/
2300	int sb_init_dio_done_wq(struct super_block *sb)
2301	{
2302	struct workqueue_struct *old;
2303	struct workqueue_struct *wq = alloc_workqueue(fmt: "dio/%s",
2304	flags: WQ_MEM_RECLAIM, max_active: 0,
2305	sb->s_id);
2306	if (!wq)
2307	return -ENOMEM;
2308	/*
2309	* This has to be atomic as more DIOs can race to create the workqueue
2310	*/
2311	old = cmpxchg(&sb->s_dio_done_wq, NULL, wq);
2312	/* Someone created workqueue before us? Free ours... */
2313	if (old)
2314	destroy_workqueue(wq);
2315	return 0;
2316	}
2317	EXPORT_SYMBOL_GPL(sb_init_dio_done_wq);
2318