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