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