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