xref: /openbmc/linux/fs/super.c (revision b9ddb2500e7e544410f38476ab928fc2fe01e381)
1  // SPDX-License-Identifier: GPL-2.0
2  /*
3   *  linux/fs/super.c
4   *
5   *  Copyright (C) 1991, 1992  Linus Torvalds
6   *
7   *  super.c contains code to handle: - mount structures
8   *                                   - super-block tables
9   *                                   - filesystem drivers list
10   *                                   - mount system call
11   *                                   - umount system call
12   *                                   - ustat system call
13   *
14   * GK 2/5/95  -  Changed to support mounting the root fs via NFS
15   *
16   *  Added kerneld support: Jacques Gelinas and Bjorn Ekwall
17   *  Added change_root: Werner Almesberger & Hans Lermen, Feb '96
18   *  Added options to /proc/mounts:
19   *    Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
20   *  Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
21   *  Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
22   */
23  
24  #include <linux/export.h>
25  #include <linux/slab.h>
26  #include <linux/blkdev.h>
27  #include <linux/mount.h>
28  #include <linux/security.h>
29  #include <linux/writeback.h>		/* for the emergency remount stuff */
30  #include <linux/idr.h>
31  #include <linux/mutex.h>
32  #include <linux/backing-dev.h>
33  #include <linux/rculist_bl.h>
34  #include <linux/cleancache.h>
35  #include <linux/fscrypt.h>
36  #include <linux/fsnotify.h>
37  #include <linux/lockdep.h>
38  #include <linux/user_namespace.h>
39  #include <linux/fs_context.h>
40  #include <uapi/linux/mount.h>
41  #include "internal.h"
42  
43  static int thaw_super_locked(struct super_block *sb);
44  
45  static LIST_HEAD(super_blocks);
46  static DEFINE_SPINLOCK(sb_lock);
47  
48  static char *sb_writers_name[SB_FREEZE_LEVELS] = {
49  	"sb_writers",
50  	"sb_pagefaults",
51  	"sb_internal",
52  };
53  
54  /*
55   * One thing we have to be careful of with a per-sb shrinker is that we don't
56   * drop the last active reference to the superblock from within the shrinker.
57   * If that happens we could trigger unregistering the shrinker from within the
58   * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
59   * take a passive reference to the superblock to avoid this from occurring.
60   */
61  static unsigned long super_cache_scan(struct shrinker *shrink,
62  				      struct shrink_control *sc)
63  {
64  	struct super_block *sb;
65  	long	fs_objects = 0;
66  	long	total_objects;
67  	long	freed = 0;
68  	long	dentries;
69  	long	inodes;
70  
71  	sb = container_of(shrink, struct super_block, s_shrink);
72  
73  	/*
74  	 * Deadlock avoidance.  We may hold various FS locks, and we don't want
75  	 * to recurse into the FS that called us in clear_inode() and friends..
76  	 */
77  	if (!(sc->gfp_mask & __GFP_FS))
78  		return SHRINK_STOP;
79  
80  	if (!trylock_super(sb))
81  		return SHRINK_STOP;
82  
83  	if (sb->s_op->nr_cached_objects)
84  		fs_objects = sb->s_op->nr_cached_objects(sb, sc);
85  
86  	inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
87  	dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
88  	total_objects = dentries + inodes + fs_objects + 1;
89  	if (!total_objects)
90  		total_objects = 1;
91  
92  	/* proportion the scan between the caches */
93  	dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
94  	inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
95  	fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
96  
97  	/*
98  	 * prune the dcache first as the icache is pinned by it, then
99  	 * prune the icache, followed by the filesystem specific caches
100  	 *
101  	 * Ensure that we always scan at least one object - memcg kmem
102  	 * accounting uses this to fully empty the caches.
103  	 */
104  	sc->nr_to_scan = dentries + 1;
105  	freed = prune_dcache_sb(sb, sc);
106  	sc->nr_to_scan = inodes + 1;
107  	freed += prune_icache_sb(sb, sc);
108  
109  	if (fs_objects) {
110  		sc->nr_to_scan = fs_objects + 1;
111  		freed += sb->s_op->free_cached_objects(sb, sc);
112  	}
113  
114  	up_read(&sb->s_umount);
115  	return freed;
116  }
117  
118  static unsigned long super_cache_count(struct shrinker *shrink,
119  				       struct shrink_control *sc)
120  {
121  	struct super_block *sb;
122  	long	total_objects = 0;
123  
124  	sb = container_of(shrink, struct super_block, s_shrink);
125  
126  	/*
127  	 * We don't call trylock_super() here as it is a scalability bottleneck,
128  	 * so we're exposed to partial setup state. The shrinker rwsem does not
129  	 * protect filesystem operations backing list_lru_shrink_count() or
130  	 * s_op->nr_cached_objects(). Counts can change between
131  	 * super_cache_count and super_cache_scan, so we really don't need locks
132  	 * here.
133  	 *
134  	 * However, if we are currently mounting the superblock, the underlying
135  	 * filesystem might be in a state of partial construction and hence it
136  	 * is dangerous to access it.  trylock_super() uses a SB_BORN check to
137  	 * avoid this situation, so do the same here. The memory barrier is
138  	 * matched with the one in mount_fs() as we don't hold locks here.
139  	 */
140  	if (!(sb->s_flags & SB_BORN))
141  		return 0;
142  	smp_rmb();
143  
144  	if (sb->s_op && sb->s_op->nr_cached_objects)
145  		total_objects = sb->s_op->nr_cached_objects(sb, sc);
146  
147  	total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
148  	total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
149  
150  	if (!total_objects)
151  		return SHRINK_EMPTY;
152  
153  	total_objects = vfs_pressure_ratio(total_objects);
154  	return total_objects;
155  }
156  
157  static void destroy_super_work(struct work_struct *work)
158  {
159  	struct super_block *s = container_of(work, struct super_block,
160  							destroy_work);
161  	int i;
162  
163  	for (i = 0; i < SB_FREEZE_LEVELS; i++)
164  		percpu_free_rwsem(&s->s_writers.rw_sem[i]);
165  	kfree(s);
166  }
167  
168  static void destroy_super_rcu(struct rcu_head *head)
169  {
170  	struct super_block *s = container_of(head, struct super_block, rcu);
171  	INIT_WORK(&s->destroy_work, destroy_super_work);
172  	schedule_work(&s->destroy_work);
173  }
174  
175  /* Free a superblock that has never been seen by anyone */
176  static void destroy_unused_super(struct super_block *s)
177  {
178  	if (!s)
179  		return;
180  	up_write(&s->s_umount);
181  	list_lru_destroy(&s->s_dentry_lru);
182  	list_lru_destroy(&s->s_inode_lru);
183  	security_sb_free(s);
184  	put_user_ns(s->s_user_ns);
185  	kfree(s->s_subtype);
186  	free_prealloced_shrinker(&s->s_shrink);
187  	/* no delays needed */
188  	destroy_super_work(&s->destroy_work);
189  }
190  
191  /**
192   *	alloc_super	-	create new superblock
193   *	@type:	filesystem type superblock should belong to
194   *	@flags: the mount flags
195   *	@user_ns: User namespace for the super_block
196   *
197   *	Allocates and initializes a new &struct super_block.  alloc_super()
198   *	returns a pointer new superblock or %NULL if allocation had failed.
199   */
200  static struct super_block *alloc_super(struct file_system_type *type, int flags,
201  				       struct user_namespace *user_ns)
202  {
203  	struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
204  	static const struct super_operations default_op;
205  	int i;
206  
207  	if (!s)
208  		return NULL;
209  
210  	INIT_LIST_HEAD(&s->s_mounts);
211  	s->s_user_ns = get_user_ns(user_ns);
212  	init_rwsem(&s->s_umount);
213  	lockdep_set_class(&s->s_umount, &type->s_umount_key);
214  	/*
215  	 * sget() can have s_umount recursion.
216  	 *
217  	 * When it cannot find a suitable sb, it allocates a new
218  	 * one (this one), and tries again to find a suitable old
219  	 * one.
220  	 *
221  	 * In case that succeeds, it will acquire the s_umount
222  	 * lock of the old one. Since these are clearly distrinct
223  	 * locks, and this object isn't exposed yet, there's no
224  	 * risk of deadlocks.
225  	 *
226  	 * Annotate this by putting this lock in a different
227  	 * subclass.
228  	 */
229  	down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
230  
231  	if (security_sb_alloc(s))
232  		goto fail;
233  
234  	for (i = 0; i < SB_FREEZE_LEVELS; i++) {
235  		if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
236  					sb_writers_name[i],
237  					&type->s_writers_key[i]))
238  			goto fail;
239  	}
240  	init_waitqueue_head(&s->s_writers.wait_unfrozen);
241  	s->s_bdi = &noop_backing_dev_info;
242  	s->s_flags = flags;
243  	if (s->s_user_ns != &init_user_ns)
244  		s->s_iflags |= SB_I_NODEV;
245  	INIT_HLIST_NODE(&s->s_instances);
246  	INIT_HLIST_BL_HEAD(&s->s_roots);
247  	mutex_init(&s->s_sync_lock);
248  	INIT_LIST_HEAD(&s->s_inodes);
249  	spin_lock_init(&s->s_inode_list_lock);
250  	INIT_LIST_HEAD(&s->s_inodes_wb);
251  	spin_lock_init(&s->s_inode_wblist_lock);
252  
253  	s->s_count = 1;
254  	atomic_set(&s->s_active, 1);
255  	mutex_init(&s->s_vfs_rename_mutex);
256  	lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
257  	init_rwsem(&s->s_dquot.dqio_sem);
258  	s->s_maxbytes = MAX_NON_LFS;
259  	s->s_op = &default_op;
260  	s->s_time_gran = 1000000000;
261  	s->s_time_min = TIME64_MIN;
262  	s->s_time_max = TIME64_MAX;
263  	s->cleancache_poolid = CLEANCACHE_NO_POOL;
264  
265  	s->s_shrink.seeks = DEFAULT_SEEKS;
266  	s->s_shrink.scan_objects = super_cache_scan;
267  	s->s_shrink.count_objects = super_cache_count;
268  	s->s_shrink.batch = 1024;
269  	s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
270  	if (prealloc_shrinker(&s->s_shrink))
271  		goto fail;
272  	if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
273  		goto fail;
274  	if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
275  		goto fail;
276  	return s;
277  
278  fail:
279  	destroy_unused_super(s);
280  	return NULL;
281  }
282  
283  /* Superblock refcounting  */
284  
285  /*
286   * Drop a superblock's refcount.  The caller must hold sb_lock.
287   */
288  static void __put_super(struct super_block *s)
289  {
290  	if (!--s->s_count) {
291  		list_del_init(&s->s_list);
292  		WARN_ON(s->s_dentry_lru.node);
293  		WARN_ON(s->s_inode_lru.node);
294  		WARN_ON(!list_empty(&s->s_mounts));
295  		security_sb_free(s);
296  		fscrypt_sb_free(s);
297  		put_user_ns(s->s_user_ns);
298  		kfree(s->s_subtype);
299  		call_rcu(&s->rcu, destroy_super_rcu);
300  	}
301  }
302  
303  /**
304   *	put_super	-	drop a temporary reference to superblock
305   *	@sb: superblock in question
306   *
307   *	Drops a temporary reference, frees superblock if there's no
308   *	references left.
309   */
310  void put_super(struct super_block *sb)
311  {
312  	spin_lock(&sb_lock);
313  	__put_super(sb);
314  	spin_unlock(&sb_lock);
315  }
316  
317  
318  /**
319   *	deactivate_locked_super	-	drop an active reference to superblock
320   *	@s: superblock to deactivate
321   *
322   *	Drops an active reference to superblock, converting it into a temporary
323   *	one if there is no other active references left.  In that case we
324   *	tell fs driver to shut it down and drop the temporary reference we
325   *	had just acquired.
326   *
327   *	Caller holds exclusive lock on superblock; that lock is released.
328   */
329  void deactivate_locked_super(struct super_block *s)
330  {
331  	struct file_system_type *fs = s->s_type;
332  	if (atomic_dec_and_test(&s->s_active)) {
333  		cleancache_invalidate_fs(s);
334  		unregister_shrinker(&s->s_shrink);
335  		fs->kill_sb(s);
336  
337  		/*
338  		 * Since list_lru_destroy() may sleep, we cannot call it from
339  		 * put_super(), where we hold the sb_lock. Therefore we destroy
340  		 * the lru lists right now.
341  		 */
342  		list_lru_destroy(&s->s_dentry_lru);
343  		list_lru_destroy(&s->s_inode_lru);
344  
345  		put_filesystem(fs);
346  		put_super(s);
347  	} else {
348  		up_write(&s->s_umount);
349  	}
350  }
351  
352  EXPORT_SYMBOL(deactivate_locked_super);
353  
354  /**
355   *	deactivate_super	-	drop an active reference to superblock
356   *	@s: superblock to deactivate
357   *
358   *	Variant of deactivate_locked_super(), except that superblock is *not*
359   *	locked by caller.  If we are going to drop the final active reference,
360   *	lock will be acquired prior to that.
361   */
362  void deactivate_super(struct super_block *s)
363  {
364  	if (!atomic_add_unless(&s->s_active, -1, 1)) {
365  		down_write(&s->s_umount);
366  		deactivate_locked_super(s);
367  	}
368  }
369  
370  EXPORT_SYMBOL(deactivate_super);
371  
372  /**
373   *	grab_super - acquire an active reference
374   *	@s: reference we are trying to make active
375   *
376   *	Tries to acquire an active reference.  grab_super() is used when we
377   * 	had just found a superblock in super_blocks or fs_type->fs_supers
378   *	and want to turn it into a full-blown active reference.  grab_super()
379   *	is called with sb_lock held and drops it.  Returns 1 in case of
380   *	success, 0 if we had failed (superblock contents was already dead or
381   *	dying when grab_super() had been called).  Note that this is only
382   *	called for superblocks not in rundown mode (== ones still on ->fs_supers
383   *	of their type), so increment of ->s_count is OK here.
384   */
385  static int grab_super(struct super_block *s) __releases(sb_lock)
386  {
387  	s->s_count++;
388  	spin_unlock(&sb_lock);
389  	down_write(&s->s_umount);
390  	if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
391  		put_super(s);
392  		return 1;
393  	}
394  	up_write(&s->s_umount);
395  	put_super(s);
396  	return 0;
397  }
398  
399  /*
400   *	trylock_super - try to grab ->s_umount shared
401   *	@sb: reference we are trying to grab
402   *
403   *	Try to prevent fs shutdown.  This is used in places where we
404   *	cannot take an active reference but we need to ensure that the
405   *	filesystem is not shut down while we are working on it. It returns
406   *	false if we cannot acquire s_umount or if we lose the race and
407   *	filesystem already got into shutdown, and returns true with the s_umount
408   *	lock held in read mode in case of success. On successful return,
409   *	the caller must drop the s_umount lock when done.
410   *
411   *	Note that unlike get_super() et.al. this one does *not* bump ->s_count.
412   *	The reason why it's safe is that we are OK with doing trylock instead
413   *	of down_read().  There's a couple of places that are OK with that, but
414   *	it's very much not a general-purpose interface.
415   */
416  bool trylock_super(struct super_block *sb)
417  {
418  	if (down_read_trylock(&sb->s_umount)) {
419  		if (!hlist_unhashed(&sb->s_instances) &&
420  		    sb->s_root && (sb->s_flags & SB_BORN))
421  			return true;
422  		up_read(&sb->s_umount);
423  	}
424  
425  	return false;
426  }
427  
428  /**
429   *	generic_shutdown_super	-	common helper for ->kill_sb()
430   *	@sb: superblock to kill
431   *
432   *	generic_shutdown_super() does all fs-independent work on superblock
433   *	shutdown.  Typical ->kill_sb() should pick all fs-specific objects
434   *	that need destruction out of superblock, call generic_shutdown_super()
435   *	and release aforementioned objects.  Note: dentries and inodes _are_
436   *	taken care of and do not need specific handling.
437   *
438   *	Upon calling this function, the filesystem may no longer alter or
439   *	rearrange the set of dentries belonging to this super_block, nor may it
440   *	change the attachments of dentries to inodes.
441   */
442  void generic_shutdown_super(struct super_block *sb)
443  {
444  	const struct super_operations *sop = sb->s_op;
445  
446  	if (sb->s_root) {
447  		shrink_dcache_for_umount(sb);
448  		sync_filesystem(sb);
449  		sb->s_flags &= ~SB_ACTIVE;
450  
451  		cgroup_writeback_umount();
452  
453  		/* evict all inodes with zero refcount */
454  		evict_inodes(sb);
455  		/* only nonzero refcount inodes can have marks */
456  		fsnotify_sb_delete(sb);
457  
458  		if (sb->s_dio_done_wq) {
459  			destroy_workqueue(sb->s_dio_done_wq);
460  			sb->s_dio_done_wq = NULL;
461  		}
462  
463  		if (sop->put_super)
464  			sop->put_super(sb);
465  
466  		if (!list_empty(&sb->s_inodes)) {
467  			printk("VFS: Busy inodes after unmount of %s. "
468  			   "Self-destruct in 5 seconds.  Have a nice day...\n",
469  			   sb->s_id);
470  		}
471  	}
472  	spin_lock(&sb_lock);
473  	/* should be initialized for __put_super_and_need_restart() */
474  	hlist_del_init(&sb->s_instances);
475  	spin_unlock(&sb_lock);
476  	up_write(&sb->s_umount);
477  	if (sb->s_bdi != &noop_backing_dev_info) {
478  		bdi_put(sb->s_bdi);
479  		sb->s_bdi = &noop_backing_dev_info;
480  	}
481  }
482  
483  EXPORT_SYMBOL(generic_shutdown_super);
484  
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   */
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   */
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  
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  
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  
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   */
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   */
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   */
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   */
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  
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   */
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) && bdev_read_only(sb->s_bdev))
869  			return -EACCES;
870  #endif
871  
872  		remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
873  	}
874  
875  	if (remount_ro) {
876  		if (!hlist_empty(&sb->s_pins)) {
877  			up_write(&sb->s_umount);
878  			group_pin_kill(&sb->s_pins);
879  			down_write(&sb->s_umount);
880  			if (!sb->s_root)
881  				return 0;
882  			if (sb->s_writers.frozen != SB_UNFROZEN)
883  				return -EBUSY;
884  			remount_ro = !sb_rdonly(sb);
885  		}
886  	}
887  	shrink_dcache_sb(sb);
888  
889  	/* If we are reconfiguring to RDONLY and current sb is read/write,
890  	 * make sure there are no files open for writing.
891  	 */
892  	if (remount_ro) {
893  		if (force) {
894  			sb->s_readonly_remount = 1;
895  			smp_wmb();
896  		} else {
897  			retval = sb_prepare_remount_readonly(sb);
898  			if (retval)
899  				return retval;
900  		}
901  	}
902  
903  	if (fc->ops->reconfigure) {
904  		retval = fc->ops->reconfigure(fc);
905  		if (retval) {
906  			if (!force)
907  				goto cancel_readonly;
908  			/* If forced remount, go ahead despite any errors */
909  			WARN(1, "forced remount of a %s fs returned %i\n",
910  			     sb->s_type->name, retval);
911  		}
912  	}
913  
914  	WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
915  				 (fc->sb_flags & fc->sb_flags_mask)));
916  	/* Needs to be ordered wrt mnt_is_readonly() */
917  	smp_wmb();
918  	sb->s_readonly_remount = 0;
919  
920  	/*
921  	 * Some filesystems modify their metadata via some other path than the
922  	 * bdev buffer cache (eg. use a private mapping, or directories in
923  	 * pagecache, etc). Also file data modifications go via their own
924  	 * mappings. So If we try to mount readonly then copy the filesystem
925  	 * from bdev, we could get stale data, so invalidate it to give a best
926  	 * effort at coherency.
927  	 */
928  	if (remount_ro && sb->s_bdev)
929  		invalidate_bdev(sb->s_bdev);
930  	return 0;
931  
932  cancel_readonly:
933  	sb->s_readonly_remount = 0;
934  	return retval;
935  }
936  
937  static void do_emergency_remount_callback(struct super_block *sb)
938  {
939  	down_write(&sb->s_umount);
940  	if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
941  	    !sb_rdonly(sb)) {
942  		struct fs_context *fc;
943  
944  		fc = fs_context_for_reconfigure(sb->s_root,
945  					SB_RDONLY | SB_FORCE, SB_RDONLY);
946  		if (!IS_ERR(fc)) {
947  			if (parse_monolithic_mount_data(fc, NULL) == 0)
948  				(void)reconfigure_super(fc);
949  			put_fs_context(fc);
950  		}
951  	}
952  	up_write(&sb->s_umount);
953  }
954  
955  static void do_emergency_remount(struct work_struct *work)
956  {
957  	__iterate_supers(do_emergency_remount_callback);
958  	kfree(work);
959  	printk("Emergency Remount complete\n");
960  }
961  
962  void emergency_remount(void)
963  {
964  	struct work_struct *work;
965  
966  	work = kmalloc(sizeof(*work), GFP_ATOMIC);
967  	if (work) {
968  		INIT_WORK(work, do_emergency_remount);
969  		schedule_work(work);
970  	}
971  }
972  
973  static void do_thaw_all_callback(struct super_block *sb)
974  {
975  	down_write(&sb->s_umount);
976  	if (sb->s_root && sb->s_flags & SB_BORN) {
977  		emergency_thaw_bdev(sb);
978  		thaw_super_locked(sb);
979  	} else {
980  		up_write(&sb->s_umount);
981  	}
982  }
983  
984  static void do_thaw_all(struct work_struct *work)
985  {
986  	__iterate_supers(do_thaw_all_callback);
987  	kfree(work);
988  	printk(KERN_WARNING "Emergency Thaw complete\n");
989  }
990  
991  /**
992   * emergency_thaw_all -- forcibly thaw every frozen filesystem
993   *
994   * Used for emergency unfreeze of all filesystems via SysRq
995   */
996  void emergency_thaw_all(void)
997  {
998  	struct work_struct *work;
999  
1000  	work = kmalloc(sizeof(*work), GFP_ATOMIC);
1001  	if (work) {
1002  		INIT_WORK(work, do_thaw_all);
1003  		schedule_work(work);
1004  	}
1005  }
1006  
1007  static DEFINE_IDA(unnamed_dev_ida);
1008  
1009  /**
1010   * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1011   * @p: Pointer to a dev_t.
1012   *
1013   * Filesystems which don't use real block devices can call this function
1014   * to allocate a virtual block device.
1015   *
1016   * Context: Any context.  Frequently called while holding sb_lock.
1017   * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1018   * or -ENOMEM if memory allocation failed.
1019   */
1020  int get_anon_bdev(dev_t *p)
1021  {
1022  	int dev;
1023  
1024  	/*
1025  	 * Many userspace utilities consider an FSID of 0 invalid.
1026  	 * Always return at least 1 from get_anon_bdev.
1027  	 */
1028  	dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1029  			GFP_ATOMIC);
1030  	if (dev == -ENOSPC)
1031  		dev = -EMFILE;
1032  	if (dev < 0)
1033  		return dev;
1034  
1035  	*p = MKDEV(0, dev);
1036  	return 0;
1037  }
1038  EXPORT_SYMBOL(get_anon_bdev);
1039  
1040  void free_anon_bdev(dev_t dev)
1041  {
1042  	ida_free(&unnamed_dev_ida, MINOR(dev));
1043  }
1044  EXPORT_SYMBOL(free_anon_bdev);
1045  
1046  int set_anon_super(struct super_block *s, void *data)
1047  {
1048  	return get_anon_bdev(&s->s_dev);
1049  }
1050  EXPORT_SYMBOL(set_anon_super);
1051  
1052  void kill_anon_super(struct super_block *sb)
1053  {
1054  	dev_t dev = sb->s_dev;
1055  	generic_shutdown_super(sb);
1056  	free_anon_bdev(dev);
1057  }
1058  EXPORT_SYMBOL(kill_anon_super);
1059  
1060  void kill_litter_super(struct super_block *sb)
1061  {
1062  	if (sb->s_root)
1063  		d_genocide(sb->s_root);
1064  	kill_anon_super(sb);
1065  }
1066  EXPORT_SYMBOL(kill_litter_super);
1067  
1068  int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1069  {
1070  	return set_anon_super(sb, NULL);
1071  }
1072  EXPORT_SYMBOL(set_anon_super_fc);
1073  
1074  static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1075  {
1076  	return sb->s_fs_info == fc->s_fs_info;
1077  }
1078  
1079  static int test_single_super(struct super_block *s, struct fs_context *fc)
1080  {
1081  	return 1;
1082  }
1083  
1084  /**
1085   * vfs_get_super - Get a superblock with a search key set in s_fs_info.
1086   * @fc: The filesystem context holding the parameters
1087   * @keying: How to distinguish superblocks
1088   * @fill_super: Helper to initialise a new superblock
1089   *
1090   * Search for a superblock and create a new one if not found.  The search
1091   * criterion is controlled by @keying.  If the search fails, a new superblock
1092   * is created and @fill_super() is called to initialise it.
1093   *
1094   * @keying can take one of a number of values:
1095   *
1096   * (1) vfs_get_single_super - Only one superblock of this type may exist on the
1097   *     system.  This is typically used for special system filesystems.
1098   *
1099   * (2) vfs_get_keyed_super - Multiple superblocks may exist, but they must have
1100   *     distinct keys (where the key is in s_fs_info).  Searching for the same
1101   *     key again will turn up the superblock for that key.
1102   *
1103   * (3) vfs_get_independent_super - Multiple superblocks may exist and are
1104   *     unkeyed.  Each call will get a new superblock.
1105   *
1106   * A permissions check is made by sget_fc() unless we're getting a superblock
1107   * for a kernel-internal mount or a submount.
1108   */
1109  int vfs_get_super(struct fs_context *fc,
1110  		  enum vfs_get_super_keying keying,
1111  		  int (*fill_super)(struct super_block *sb,
1112  				    struct fs_context *fc))
1113  {
1114  	int (*test)(struct super_block *, struct fs_context *);
1115  	struct super_block *sb;
1116  	int err;
1117  
1118  	switch (keying) {
1119  	case vfs_get_single_super:
1120  	case vfs_get_single_reconf_super:
1121  		test = test_single_super;
1122  		break;
1123  	case vfs_get_keyed_super:
1124  		test = test_keyed_super;
1125  		break;
1126  	case vfs_get_independent_super:
1127  		test = NULL;
1128  		break;
1129  	default:
1130  		BUG();
1131  	}
1132  
1133  	sb = sget_fc(fc, test, set_anon_super_fc);
1134  	if (IS_ERR(sb))
1135  		return PTR_ERR(sb);
1136  
1137  	if (!sb->s_root) {
1138  		err = fill_super(sb, fc);
1139  		if (err)
1140  			goto error;
1141  
1142  		sb->s_flags |= SB_ACTIVE;
1143  		fc->root = dget(sb->s_root);
1144  	} else {
1145  		fc->root = dget(sb->s_root);
1146  		if (keying == vfs_get_single_reconf_super) {
1147  			err = reconfigure_super(fc);
1148  			if (err < 0) {
1149  				dput(fc->root);
1150  				fc->root = NULL;
1151  				goto error;
1152  			}
1153  		}
1154  	}
1155  
1156  	return 0;
1157  
1158  error:
1159  	deactivate_locked_super(sb);
1160  	return err;
1161  }
1162  EXPORT_SYMBOL(vfs_get_super);
1163  
1164  int get_tree_nodev(struct fs_context *fc,
1165  		  int (*fill_super)(struct super_block *sb,
1166  				    struct fs_context *fc))
1167  {
1168  	return vfs_get_super(fc, vfs_get_independent_super, fill_super);
1169  }
1170  EXPORT_SYMBOL(get_tree_nodev);
1171  
1172  int get_tree_single(struct fs_context *fc,
1173  		  int (*fill_super)(struct super_block *sb,
1174  				    struct fs_context *fc))
1175  {
1176  	return vfs_get_super(fc, vfs_get_single_super, fill_super);
1177  }
1178  EXPORT_SYMBOL(get_tree_single);
1179  
1180  int get_tree_single_reconf(struct fs_context *fc,
1181  		  int (*fill_super)(struct super_block *sb,
1182  				    struct fs_context *fc))
1183  {
1184  	return vfs_get_super(fc, vfs_get_single_reconf_super, fill_super);
1185  }
1186  EXPORT_SYMBOL(get_tree_single_reconf);
1187  
1188  int get_tree_keyed(struct fs_context *fc,
1189  		  int (*fill_super)(struct super_block *sb,
1190  				    struct fs_context *fc),
1191  		void *key)
1192  {
1193  	fc->s_fs_info = key;
1194  	return vfs_get_super(fc, vfs_get_keyed_super, fill_super);
1195  }
1196  EXPORT_SYMBOL(get_tree_keyed);
1197  
1198  #ifdef CONFIG_BLOCK
1199  
1200  static int set_bdev_super(struct super_block *s, void *data)
1201  {
1202  	s->s_bdev = data;
1203  	s->s_dev = s->s_bdev->bd_dev;
1204  	s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1205  
1206  	if (blk_queue_stable_writes(s->s_bdev->bd_disk->queue))
1207  		s->s_iflags |= SB_I_STABLE_WRITES;
1208  	return 0;
1209  }
1210  
1211  static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1212  {
1213  	return set_bdev_super(s, fc->sget_key);
1214  }
1215  
1216  static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1217  {
1218  	return s->s_bdev == fc->sget_key;
1219  }
1220  
1221  /**
1222   * get_tree_bdev - Get a superblock based on a single block device
1223   * @fc: The filesystem context holding the parameters
1224   * @fill_super: Helper to initialise a new superblock
1225   */
1226  int get_tree_bdev(struct fs_context *fc,
1227  		int (*fill_super)(struct super_block *,
1228  				  struct fs_context *))
1229  {
1230  	struct block_device *bdev;
1231  	struct super_block *s;
1232  	fmode_t mode = FMODE_READ | FMODE_EXCL;
1233  	int error = 0;
1234  
1235  	if (!(fc->sb_flags & SB_RDONLY))
1236  		mode |= FMODE_WRITE;
1237  
1238  	if (!fc->source)
1239  		return invalf(fc, "No source specified");
1240  
1241  	bdev = blkdev_get_by_path(fc->source, mode, fc->fs_type);
1242  	if (IS_ERR(bdev)) {
1243  		errorf(fc, "%s: Can't open blockdev", fc->source);
1244  		return PTR_ERR(bdev);
1245  	}
1246  
1247  	/* Once the superblock is inserted into the list by sget_fc(), s_umount
1248  	 * will protect the lockfs code from trying to start a snapshot while
1249  	 * we are mounting
1250  	 */
1251  	mutex_lock(&bdev->bd_fsfreeze_mutex);
1252  	if (bdev->bd_fsfreeze_count > 0) {
1253  		mutex_unlock(&bdev->bd_fsfreeze_mutex);
1254  		warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1255  		blkdev_put(bdev, mode);
1256  		return -EBUSY;
1257  	}
1258  
1259  	fc->sb_flags |= SB_NOSEC;
1260  	fc->sget_key = bdev;
1261  	s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc);
1262  	mutex_unlock(&bdev->bd_fsfreeze_mutex);
1263  	if (IS_ERR(s)) {
1264  		blkdev_put(bdev, mode);
1265  		return PTR_ERR(s);
1266  	}
1267  
1268  	if (s->s_root) {
1269  		/* Don't summarily change the RO/RW state. */
1270  		if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1271  			warnf(fc, "%pg: Can't mount, would change RO state", bdev);
1272  			deactivate_locked_super(s);
1273  			blkdev_put(bdev, mode);
1274  			return -EBUSY;
1275  		}
1276  
1277  		/*
1278  		 * s_umount nests inside bd_mutex during
1279  		 * __invalidate_device().  blkdev_put() acquires
1280  		 * bd_mutex and can't be called under s_umount.  Drop
1281  		 * s_umount temporarily.  This is safe as we're
1282  		 * holding an active reference.
1283  		 */
1284  		up_write(&s->s_umount);
1285  		blkdev_put(bdev, mode);
1286  		down_write(&s->s_umount);
1287  	} else {
1288  		s->s_mode = mode;
1289  		snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1290  		sb_set_blocksize(s, block_size(bdev));
1291  		error = fill_super(s, fc);
1292  		if (error) {
1293  			deactivate_locked_super(s);
1294  			return error;
1295  		}
1296  
1297  		s->s_flags |= SB_ACTIVE;
1298  		bdev->bd_super = s;
1299  	}
1300  
1301  	BUG_ON(fc->root);
1302  	fc->root = dget(s->s_root);
1303  	return 0;
1304  }
1305  EXPORT_SYMBOL(get_tree_bdev);
1306  
1307  static int test_bdev_super(struct super_block *s, void *data)
1308  {
1309  	return (void *)s->s_bdev == data;
1310  }
1311  
1312  struct dentry *mount_bdev(struct file_system_type *fs_type,
1313  	int flags, const char *dev_name, void *data,
1314  	int (*fill_super)(struct super_block *, void *, int))
1315  {
1316  	struct block_device *bdev;
1317  	struct super_block *s;
1318  	fmode_t mode = FMODE_READ | FMODE_EXCL;
1319  	int error = 0;
1320  
1321  	if (!(flags & SB_RDONLY))
1322  		mode |= FMODE_WRITE;
1323  
1324  	bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1325  	if (IS_ERR(bdev))
1326  		return ERR_CAST(bdev);
1327  
1328  	/*
1329  	 * once the super is inserted into the list by sget, s_umount
1330  	 * will protect the lockfs code from trying to start a snapshot
1331  	 * while we are mounting
1332  	 */
1333  	mutex_lock(&bdev->bd_fsfreeze_mutex);
1334  	if (bdev->bd_fsfreeze_count > 0) {
1335  		mutex_unlock(&bdev->bd_fsfreeze_mutex);
1336  		error = -EBUSY;
1337  		goto error_bdev;
1338  	}
1339  	s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1340  		 bdev);
1341  	mutex_unlock(&bdev->bd_fsfreeze_mutex);
1342  	if (IS_ERR(s))
1343  		goto error_s;
1344  
1345  	if (s->s_root) {
1346  		if ((flags ^ s->s_flags) & SB_RDONLY) {
1347  			deactivate_locked_super(s);
1348  			error = -EBUSY;
1349  			goto error_bdev;
1350  		}
1351  
1352  		/*
1353  		 * s_umount nests inside bd_mutex during
1354  		 * __invalidate_device().  blkdev_put() acquires
1355  		 * bd_mutex and can't be called under s_umount.  Drop
1356  		 * s_umount temporarily.  This is safe as we're
1357  		 * holding an active reference.
1358  		 */
1359  		up_write(&s->s_umount);
1360  		blkdev_put(bdev, mode);
1361  		down_write(&s->s_umount);
1362  	} else {
1363  		s->s_mode = mode;
1364  		snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1365  		sb_set_blocksize(s, block_size(bdev));
1366  		error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1367  		if (error) {
1368  			deactivate_locked_super(s);
1369  			goto error;
1370  		}
1371  
1372  		s->s_flags |= SB_ACTIVE;
1373  		bdev->bd_super = s;
1374  	}
1375  
1376  	return dget(s->s_root);
1377  
1378  error_s:
1379  	error = PTR_ERR(s);
1380  error_bdev:
1381  	blkdev_put(bdev, mode);
1382  error:
1383  	return ERR_PTR(error);
1384  }
1385  EXPORT_SYMBOL(mount_bdev);
1386  
1387  void kill_block_super(struct super_block *sb)
1388  {
1389  	struct block_device *bdev = sb->s_bdev;
1390  	fmode_t mode = sb->s_mode;
1391  
1392  	bdev->bd_super = NULL;
1393  	generic_shutdown_super(sb);
1394  	sync_blockdev(bdev);
1395  	WARN_ON_ONCE(!(mode & FMODE_EXCL));
1396  	blkdev_put(bdev, mode | FMODE_EXCL);
1397  }
1398  
1399  EXPORT_SYMBOL(kill_block_super);
1400  #endif
1401  
1402  struct dentry *mount_nodev(struct file_system_type *fs_type,
1403  	int flags, void *data,
1404  	int (*fill_super)(struct super_block *, void *, int))
1405  {
1406  	int error;
1407  	struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1408  
1409  	if (IS_ERR(s))
1410  		return ERR_CAST(s);
1411  
1412  	error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1413  	if (error) {
1414  		deactivate_locked_super(s);
1415  		return ERR_PTR(error);
1416  	}
1417  	s->s_flags |= SB_ACTIVE;
1418  	return dget(s->s_root);
1419  }
1420  EXPORT_SYMBOL(mount_nodev);
1421  
1422  static int reconfigure_single(struct super_block *s,
1423  			      int flags, void *data)
1424  {
1425  	struct fs_context *fc;
1426  	int ret;
1427  
1428  	/* The caller really need to be passing fc down into mount_single(),
1429  	 * then a chunk of this can be removed.  [Bollocks -- AV]
1430  	 * Better yet, reconfiguration shouldn't happen, but rather the second
1431  	 * mount should be rejected if the parameters are not compatible.
1432  	 */
1433  	fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1434  	if (IS_ERR(fc))
1435  		return PTR_ERR(fc);
1436  
1437  	ret = parse_monolithic_mount_data(fc, data);
1438  	if (ret < 0)
1439  		goto out;
1440  
1441  	ret = reconfigure_super(fc);
1442  out:
1443  	put_fs_context(fc);
1444  	return ret;
1445  }
1446  
1447  static int compare_single(struct super_block *s, void *p)
1448  {
1449  	return 1;
1450  }
1451  
1452  struct dentry *mount_single(struct file_system_type *fs_type,
1453  	int flags, void *data,
1454  	int (*fill_super)(struct super_block *, void *, int))
1455  {
1456  	struct super_block *s;
1457  	int error;
1458  
1459  	s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1460  	if (IS_ERR(s))
1461  		return ERR_CAST(s);
1462  	if (!s->s_root) {
1463  		error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1464  		if (!error)
1465  			s->s_flags |= SB_ACTIVE;
1466  	} else {
1467  		error = reconfigure_single(s, flags, data);
1468  	}
1469  	if (unlikely(error)) {
1470  		deactivate_locked_super(s);
1471  		return ERR_PTR(error);
1472  	}
1473  	return dget(s->s_root);
1474  }
1475  EXPORT_SYMBOL(mount_single);
1476  
1477  /**
1478   * vfs_get_tree - Get the mountable root
1479   * @fc: The superblock configuration context.
1480   *
1481   * The filesystem is invoked to get or create a superblock which can then later
1482   * be used for mounting.  The filesystem places a pointer to the root to be
1483   * used for mounting in @fc->root.
1484   */
1485  int vfs_get_tree(struct fs_context *fc)
1486  {
1487  	struct super_block *sb;
1488  	int error;
1489  
1490  	if (fc->root)
1491  		return -EBUSY;
1492  
1493  	/* Get the mountable root in fc->root, with a ref on the root and a ref
1494  	 * on the superblock.
1495  	 */
1496  	error = fc->ops->get_tree(fc);
1497  	if (error < 0)
1498  		return error;
1499  
1500  	if (!fc->root) {
1501  		pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1502  		       fc->fs_type->name);
1503  		/* We don't know what the locking state of the superblock is -
1504  		 * if there is a superblock.
1505  		 */
1506  		BUG();
1507  	}
1508  
1509  	sb = fc->root->d_sb;
1510  	WARN_ON(!sb->s_bdi);
1511  
1512  	/*
1513  	 * Write barrier is for super_cache_count(). We place it before setting
1514  	 * SB_BORN as the data dependency between the two functions is the
1515  	 * superblock structure contents that we just set up, not the SB_BORN
1516  	 * flag.
1517  	 */
1518  	smp_wmb();
1519  	sb->s_flags |= SB_BORN;
1520  
1521  	error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1522  	if (unlikely(error)) {
1523  		fc_drop_locked(fc);
1524  		return error;
1525  	}
1526  
1527  	/*
1528  	 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1529  	 * but s_maxbytes was an unsigned long long for many releases. Throw
1530  	 * this warning for a little while to try and catch filesystems that
1531  	 * violate this rule.
1532  	 */
1533  	WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1534  		"negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1535  
1536  	return 0;
1537  }
1538  EXPORT_SYMBOL(vfs_get_tree);
1539  
1540  /*
1541   * Setup private BDI for given superblock. It gets automatically cleaned up
1542   * in generic_shutdown_super().
1543   */
1544  int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1545  {
1546  	struct backing_dev_info *bdi;
1547  	int err;
1548  	va_list args;
1549  
1550  	bdi = bdi_alloc(NUMA_NO_NODE);
1551  	if (!bdi)
1552  		return -ENOMEM;
1553  
1554  	va_start(args, fmt);
1555  	err = bdi_register_va(bdi, fmt, args);
1556  	va_end(args);
1557  	if (err) {
1558  		bdi_put(bdi);
1559  		return err;
1560  	}
1561  	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1562  	sb->s_bdi = bdi;
1563  
1564  	return 0;
1565  }
1566  EXPORT_SYMBOL(super_setup_bdi_name);
1567  
1568  /*
1569   * Setup private BDI for given superblock. I gets automatically cleaned up
1570   * in generic_shutdown_super().
1571   */
1572  int super_setup_bdi(struct super_block *sb)
1573  {
1574  	static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1575  
1576  	return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1577  				    atomic_long_inc_return(&bdi_seq));
1578  }
1579  EXPORT_SYMBOL(super_setup_bdi);
1580  
1581  /**
1582   * sb_wait_write - wait until all writers to given file system finish
1583   * @sb: the super for which we wait
1584   * @level: type of writers we wait for (normal vs page fault)
1585   *
1586   * This function waits until there are no writers of given type to given file
1587   * system.
1588   */
1589  static void sb_wait_write(struct super_block *sb, int level)
1590  {
1591  	percpu_down_write(sb->s_writers.rw_sem + level-1);
1592  }
1593  
1594  /*
1595   * We are going to return to userspace and forget about these locks, the
1596   * ownership goes to the caller of thaw_super() which does unlock().
1597   */
1598  static void lockdep_sb_freeze_release(struct super_block *sb)
1599  {
1600  	int level;
1601  
1602  	for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1603  		percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1604  }
1605  
1606  /*
1607   * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1608   */
1609  static void lockdep_sb_freeze_acquire(struct super_block *sb)
1610  {
1611  	int level;
1612  
1613  	for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1614  		percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1615  }
1616  
1617  static void sb_freeze_unlock(struct super_block *sb)
1618  {
1619  	int level;
1620  
1621  	for (level = SB_FREEZE_LEVELS - 1; 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   */
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  	sync_filesystem(sb);
1693  
1694  	/* Now wait for internal filesystem counter */
1695  	sb->s_writers.frozen = SB_FREEZE_FS;
1696  	sb_wait_write(sb, SB_FREEZE_FS);
1697  
1698  	if (sb->s_op->freeze_fs) {
1699  		ret = sb->s_op->freeze_fs(sb);
1700  		if (ret) {
1701  			printk(KERN_ERR
1702  				"VFS:Filesystem freeze failed\n");
1703  			sb->s_writers.frozen = SB_UNFROZEN;
1704  			sb_freeze_unlock(sb);
1705  			wake_up(&sb->s_writers.wait_unfrozen);
1706  			deactivate_locked_super(sb);
1707  			return ret;
1708  		}
1709  	}
1710  	/*
1711  	 * For debugging purposes so that fs can warn if it sees write activity
1712  	 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1713  	 */
1714  	sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1715  	lockdep_sb_freeze_release(sb);
1716  	up_write(&sb->s_umount);
1717  	return 0;
1718  }
1719  EXPORT_SYMBOL(freeze_super);
1720  
1721  /**
1722   * thaw_super -- unlock filesystem
1723   * @sb: the super to thaw
1724   *
1725   * Unlocks the filesystem and marks it writeable again after freeze_super().
1726   */
1727  static int thaw_super_locked(struct super_block *sb)
1728  {
1729  	int error;
1730  
1731  	if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1732  		up_write(&sb->s_umount);
1733  		return -EINVAL;
1734  	}
1735  
1736  	if (sb_rdonly(sb)) {
1737  		sb->s_writers.frozen = SB_UNFROZEN;
1738  		goto out;
1739  	}
1740  
1741  	lockdep_sb_freeze_acquire(sb);
1742  
1743  	if (sb->s_op->unfreeze_fs) {
1744  		error = sb->s_op->unfreeze_fs(sb);
1745  		if (error) {
1746  			printk(KERN_ERR
1747  				"VFS:Filesystem thaw failed\n");
1748  			lockdep_sb_freeze_release(sb);
1749  			up_write(&sb->s_umount);
1750  			return error;
1751  		}
1752  	}
1753  
1754  	sb->s_writers.frozen = SB_UNFROZEN;
1755  	sb_freeze_unlock(sb);
1756  out:
1757  	wake_up(&sb->s_writers.wait_unfrozen);
1758  	deactivate_locked_super(sb);
1759  	return 0;
1760  }
1761  
1762  int thaw_super(struct super_block *sb)
1763  {
1764  	down_write(&sb->s_umount);
1765  	return thaw_super_locked(sb);
1766  }
1767  EXPORT_SYMBOL(thaw_super);
1768