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
  42static int thaw_super_locked(struct super_block *sb, enum freeze_holder who);
  43
  44static LIST_HEAD(super_blocks);
  45static DEFINE_SPINLOCK(sb_lock);
  46
  47static char *sb_writers_name[SB_FREEZE_LEVELS] = {
  48	"sb_writers",
  49	"sb_pagefaults",
  50	"sb_internal",
  51};
  52
  53static inline void __super_lock(struct super_block *sb, bool excl)
  54{
  55	if (excl)
  56		down_write(&sb->s_umount);
  57	else
  58		down_read(&sb->s_umount);
  59}
  60
  61static inline void super_unlock(struct super_block *sb, bool excl)
  62{
  63	if (excl)
  64		up_write(&sb->s_umount);
  65	else
  66		up_read(&sb->s_umount);
  67}
  68
  69static inline void __super_lock_excl(struct super_block *sb)
  70{
  71	__super_lock(sb, true);
  72}
  73
  74static inline void super_unlock_excl(struct super_block *sb)
  75{
  76	super_unlock(sb, true);
  77}
  78
  79static inline void super_unlock_shared(struct super_block *sb)
  80{
  81	super_unlock(sb, false);
  82}
  83
  84static bool super_flags(const struct super_block *sb, unsigned int flags)
  85{
  86	/*
  87	 * Pairs with smp_store_release() in super_wake() and ensures
  88	 * that we see @flags after we're woken.
  89	 */
  90	return smp_load_acquire(&sb->s_flags) & flags;
  91}
  92
  93/**
  94 * super_lock - wait for superblock to become ready and lock it
  95 * @sb: superblock to wait for
  96 * @excl: whether exclusive access is required
  97 *
  98 * If the superblock has neither passed through vfs_get_tree() or
  99 * generic_shutdown_super() yet wait for it to happen. Either superblock
 100 * creation will succeed and SB_BORN is set by vfs_get_tree() or we're
 101 * woken and we'll see SB_DYING.
 102 *
 103 * The caller must have acquired a temporary reference on @sb->s_count.
 104 *
 105 * Return: The function returns true if SB_BORN was set and with
 106 *         s_umount held. The function returns false if SB_DYING was
 107 *         set and without s_umount held.
 108 */
 109static __must_check bool super_lock(struct super_block *sb, bool excl)
 110{
 111	lockdep_assert_not_held(&sb->s_umount);
 112
 113	/* wait until the superblock is ready or dying */
 114	wait_var_event(&sb->s_flags, super_flags(sb, SB_BORN | SB_DYING));
 115
 116	/* Don't pointlessly acquire s_umount. */
 117	if (super_flags(sb, SB_DYING))
 118		return false;
 119
 120	__super_lock(sb, excl);
 121
 122	/*
 123	 * Has gone through generic_shutdown_super() in the meantime.
 124	 * @sb->s_root is NULL and @sb->s_active is 0. No one needs to
 125	 * grab a reference to this. Tell them so.
 126	 */
 127	if (sb->s_flags & SB_DYING) {
 128		super_unlock(sb, excl);
 129		return false;
 130	}
 131
 132	WARN_ON_ONCE(!(sb->s_flags & SB_BORN));
 133	return true;
 134}
 135
 136/* wait and try to acquire read-side of @sb->s_umount */
 137static inline bool super_lock_shared(struct super_block *sb)
 138{
 139	return super_lock(sb, false);
 140}
 141
 142/* wait and try to acquire write-side of @sb->s_umount */
 143static inline bool super_lock_excl(struct super_block *sb)
 144{
 145	return super_lock(sb, true);
 146}
 147
 148/* wake waiters */
 149#define SUPER_WAKE_FLAGS (SB_BORN | SB_DYING | SB_DEAD)
 150static void super_wake(struct super_block *sb, unsigned int flag)
 151{
 152	WARN_ON_ONCE((flag & ~SUPER_WAKE_FLAGS));
 153	WARN_ON_ONCE(hweight32(flag & SUPER_WAKE_FLAGS) > 1);
 154
 155	/*
 156	 * Pairs with smp_load_acquire() in super_lock() to make sure
 157	 * all initializations in the superblock are seen by the user
 158	 * seeing SB_BORN sent.
 159	 */
 160	smp_store_release(&sb->s_flags, sb->s_flags | flag);
 161	/*
 162	 * Pairs with the barrier in prepare_to_wait_event() to make sure
 163	 * ___wait_var_event() either sees SB_BORN set or
 164	 * waitqueue_active() check in wake_up_var() sees the waiter.
 165	 */
 166	smp_mb();
 167	wake_up_var(&sb->s_flags);
 168}
 169
 170/*
 171 * One thing we have to be careful of with a per-sb shrinker is that we don't
 172 * drop the last active reference to the superblock from within the shrinker.
 173 * If that happens we could trigger unregistering the shrinker from within the
 174 * shrinker path and that leads to deadlock on the shrinker_mutex. Hence we
 175 * take a passive reference to the superblock to avoid this from occurring.
 176 */
 177static unsigned long super_cache_scan(struct shrinker *shrink,
 178				      struct shrink_control *sc)
 179{
 180	struct super_block *sb;
 181	long	fs_objects = 0;
 182	long	total_objects;
 183	long	freed = 0;
 184	long	dentries;
 185	long	inodes;
 186
 187	sb = shrink->private_data;
 188
 189	/*
 190	 * Deadlock avoidance.  We may hold various FS locks, and we don't want
 191	 * to recurse into the FS that called us in clear_inode() and friends..
 192	 */
 193	if (!(sc->gfp_mask & __GFP_FS))
 194		return SHRINK_STOP;
 195
 196	if (!super_trylock_shared(sb))
 197		return SHRINK_STOP;
 198
 199	if (sb->s_op->nr_cached_objects)
 200		fs_objects = sb->s_op->nr_cached_objects(sb, sc);
 201
 202	inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
 203	dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
 204	total_objects = dentries + inodes + fs_objects + 1;
 205	if (!total_objects)
 206		total_objects = 1;
 207
 208	/* proportion the scan between the caches */
 209	dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
 210	inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
 211	fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
 212
 213	/*
 214	 * prune the dcache first as the icache is pinned by it, then
 215	 * prune the icache, followed by the filesystem specific caches
 216	 *
 217	 * Ensure that we always scan at least one object - memcg kmem
 218	 * accounting uses this to fully empty the caches.
 219	 */
 220	sc->nr_to_scan = dentries + 1;
 221	freed = prune_dcache_sb(sb, sc);
 222	sc->nr_to_scan = inodes + 1;
 223	freed += prune_icache_sb(sb, sc);
 224
 225	if (fs_objects) {
 226		sc->nr_to_scan = fs_objects + 1;
 227		freed += sb->s_op->free_cached_objects(sb, sc);
 228	}
 229
 230	super_unlock_shared(sb);
 231	return freed;
 232}
 233
 234static unsigned long super_cache_count(struct shrinker *shrink,
 235				       struct shrink_control *sc)
 236{
 237	struct super_block *sb;
 238	long	total_objects = 0;
 239
 240	sb = shrink->private_data;
 241
 242	/*
 243	 * We don't call super_trylock_shared() here as it is a scalability
 244	 * bottleneck, so we're exposed to partial setup state. The shrinker
 245	 * rwsem does not protect filesystem operations backing
 246	 * list_lru_shrink_count() or s_op->nr_cached_objects(). Counts can
 247	 * change between super_cache_count and super_cache_scan, so we really
 248	 * don't need locks here.
 249	 *
 250	 * However, if we are currently mounting the superblock, the underlying
 251	 * filesystem might be in a state of partial construction and hence it
 252	 * is dangerous to access it.  super_trylock_shared() uses a SB_BORN check
 253	 * to avoid this situation, so do the same here. The memory barrier is
 254	 * matched with the one in mount_fs() as we don't hold locks here.
 255	 */
 256	if (!(sb->s_flags & SB_BORN))
 257		return 0;
 258	smp_rmb();
 259
 260	if (sb->s_op && sb->s_op->nr_cached_objects)
 261		total_objects = sb->s_op->nr_cached_objects(sb, sc);
 262
 263	total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
 264	total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
 265
 266	if (!total_objects)
 267		return SHRINK_EMPTY;
 268
 269	total_objects = vfs_pressure_ratio(total_objects);
 270	return total_objects;
 271}
 272
 273static void destroy_super_work(struct work_struct *work)
 274{
 275	struct super_block *s = container_of(work, struct super_block,
 276							destroy_work);
 277	security_sb_free(s);
 278	put_user_ns(s->s_user_ns);
 279	kfree(s->s_subtype);
 280	for (int i = 0; i < SB_FREEZE_LEVELS; i++)
 281		percpu_free_rwsem(&s->s_writers.rw_sem[i]);
 282	kfree(s);
 283}
 284
 285static void destroy_super_rcu(struct rcu_head *head)
 286{
 287	struct super_block *s = container_of(head, struct super_block, rcu);
 288	INIT_WORK(&s->destroy_work, destroy_super_work);
 289	schedule_work(&s->destroy_work);
 290}
 291
 292/* Free a superblock that has never been seen by anyone */
 293static void destroy_unused_super(struct super_block *s)
 
 
 
 
 
 294{
 295	if (!s)
 296		return;
 297	super_unlock_excl(s);
 298	list_lru_destroy(&s->s_dentry_lru);
 299	list_lru_destroy(&s->s_inode_lru);
 300	shrinker_free(s->s_shrink);
 301	/* no delays needed */
 302	destroy_super_work(&s->destroy_work);
 
 
 
 303}
 304
 305/**
 306 *	alloc_super	-	create new superblock
 307 *	@type:	filesystem type superblock should belong to
 308 *	@flags: the mount flags
 309 *	@user_ns: User namespace for the super_block
 310 *
 311 *	Allocates and initializes a new &struct super_block.  alloc_super()
 312 *	returns a pointer new superblock or %NULL if allocation had failed.
 313 */
 314static struct super_block *alloc_super(struct file_system_type *type, int flags,
 315				       struct user_namespace *user_ns)
 316{
 317	struct super_block *s = kzalloc(sizeof(struct super_block), GFP_KERNEL);
 318	static const struct super_operations default_op;
 319	int i;
 320
 321	if (!s)
 322		return NULL;
 323
 324	INIT_LIST_HEAD(&s->s_mounts);
 325	s->s_user_ns = get_user_ns(user_ns);
 326	init_rwsem(&s->s_umount);
 327	lockdep_set_class(&s->s_umount, &type->s_umount_key);
 328	/*
 329	 * sget() can have s_umount recursion.
 330	 *
 331	 * When it cannot find a suitable sb, it allocates a new
 332	 * one (this one), and tries again to find a suitable old
 333	 * one.
 334	 *
 335	 * In case that succeeds, it will acquire the s_umount
 336	 * lock of the old one. Since these are clearly distrinct
 337	 * locks, and this object isn't exposed yet, there's no
 338	 * risk of deadlocks.
 339	 *
 340	 * Annotate this by putting this lock in a different
 341	 * subclass.
 342	 */
 343	down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
 344
 345	if (security_sb_alloc(s))
 346		goto fail;
 347
 348	for (i = 0; i < SB_FREEZE_LEVELS; i++) {
 349		if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
 350					sb_writers_name[i],
 351					&type->s_writers_key[i]))
 352			goto fail;
 353	}
 
 354	s->s_bdi = &noop_backing_dev_info;
 355	s->s_flags = flags;
 356	if (s->s_user_ns != &init_user_ns)
 357		s->s_iflags |= SB_I_NODEV;
 358	INIT_HLIST_NODE(&s->s_instances);
 359	INIT_HLIST_BL_HEAD(&s->s_roots);
 360	mutex_init(&s->s_sync_lock);
 361	INIT_LIST_HEAD(&s->s_inodes);
 362	spin_lock_init(&s->s_inode_list_lock);
 363	INIT_LIST_HEAD(&s->s_inodes_wb);
 364	spin_lock_init(&s->s_inode_wblist_lock);
 365
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 366	s->s_count = 1;
 367	atomic_set(&s->s_active, 1);
 368	mutex_init(&s->s_vfs_rename_mutex);
 369	lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
 370	init_rwsem(&s->s_dquot.dqio_sem);
 371	s->s_maxbytes = MAX_NON_LFS;
 372	s->s_op = &default_op;
 373	s->s_time_gran = 1000000000;
 374	s->s_time_min = TIME64_MIN;
 375	s->s_time_max = TIME64_MAX;
 376
 377	s->s_shrink = shrinker_alloc(SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE,
 378				     "sb-%s", type->name);
 379	if (!s->s_shrink)
 380		goto fail;
 381
 382	s->s_shrink->scan_objects = super_cache_scan;
 383	s->s_shrink->count_objects = super_cache_count;
 384	s->s_shrink->batch = 1024;
 385	s->s_shrink->private_data = s;
 386
 387	if (list_lru_init_memcg(&s->s_dentry_lru, s->s_shrink))
 388		goto fail;
 389	if (list_lru_init_memcg(&s->s_inode_lru, s->s_shrink))
 390		goto fail;
 
 391	return s;
 392
 393fail:
 394	destroy_unused_super(s);
 395	return NULL;
 396}
 397
 398/* Superblock refcounting  */
 399
 400/*
 401 * Drop a superblock's refcount.  The caller must hold sb_lock.
 402 */
 403static void __put_super(struct super_block *s)
 404{
 405	if (!--s->s_count) {
 406		list_del_init(&s->s_list);
 407		WARN_ON(s->s_dentry_lru.node);
 408		WARN_ON(s->s_inode_lru.node);
 409		WARN_ON(!list_empty(&s->s_mounts));
 410		call_rcu(&s->rcu, destroy_super_rcu);
 411	}
 412}
 413
 414/**
 415 *	put_super	-	drop a temporary reference to superblock
 416 *	@sb: superblock in question
 417 *
 418 *	Drops a temporary reference, frees superblock if there's no
 419 *	references left.
 420 */
 421void put_super(struct super_block *sb)
 422{
 423	spin_lock(&sb_lock);
 424	__put_super(sb);
 425	spin_unlock(&sb_lock);
 426}
 427
 428static void kill_super_notify(struct super_block *sb)
 429{
 430	lockdep_assert_not_held(&sb->s_umount);
 431
 432	/* already notified earlier */
 433	if (sb->s_flags & SB_DEAD)
 434		return;
 435
 436	/*
 437	 * Remove it from @fs_supers so it isn't found by new
 438	 * sget{_fc}() walkers anymore. Any concurrent mounter still
 439	 * managing to grab a temporary reference is guaranteed to
 440	 * already see SB_DYING and will wait until we notify them about
 441	 * SB_DEAD.
 442	 */
 443	spin_lock(&sb_lock);
 444	hlist_del_init(&sb->s_instances);
 445	spin_unlock(&sb_lock);
 446
 447	/*
 448	 * Let concurrent mounts know that this thing is really dead.
 449	 * We don't need @sb->s_umount here as every concurrent caller
 450	 * will see SB_DYING and either discard the superblock or wait
 451	 * for SB_DEAD.
 452	 */
 453	super_wake(sb, SB_DEAD);
 454}
 455
 456/**
 457 *	deactivate_locked_super	-	drop an active reference to superblock
 458 *	@s: superblock to deactivate
 459 *
 460 *	Drops an active reference to superblock, converting it into a temporary
 461 *	one if there is no other active references left.  In that case we
 462 *	tell fs driver to shut it down and drop the temporary reference we
 463 *	had just acquired.
 464 *
 465 *	Caller holds exclusive lock on superblock; that lock is released.
 466 */
 467void deactivate_locked_super(struct super_block *s)
 468{
 469	struct file_system_type *fs = s->s_type;
 470	if (atomic_dec_and_test(&s->s_active)) {
 471		shrinker_free(s->s_shrink);
 
 472		fs->kill_sb(s);
 473
 474		kill_super_notify(s);
 475
 476		/*
 477		 * Since list_lru_destroy() may sleep, we cannot call it from
 478		 * put_super(), where we hold the sb_lock. Therefore we destroy
 479		 * the lru lists right now.
 480		 */
 481		list_lru_destroy(&s->s_dentry_lru);
 482		list_lru_destroy(&s->s_inode_lru);
 483
 484		put_filesystem(fs);
 485		put_super(s);
 486	} else {
 487		super_unlock_excl(s);
 488	}
 489}
 490
 491EXPORT_SYMBOL(deactivate_locked_super);
 492
 493/**
 494 *	deactivate_super	-	drop an active reference to superblock
 495 *	@s: superblock to deactivate
 496 *
 497 *	Variant of deactivate_locked_super(), except that superblock is *not*
 498 *	locked by caller.  If we are going to drop the final active reference,
 499 *	lock will be acquired prior to that.
 500 */
 501void deactivate_super(struct super_block *s)
 502{
 503	if (!atomic_add_unless(&s->s_active, -1, 1)) {
 504		__super_lock_excl(s);
 505		deactivate_locked_super(s);
 506	}
 507}
 508
 509EXPORT_SYMBOL(deactivate_super);
 510
 511/**
 512 * grab_super - acquire an active reference to a superblock
 513 * @sb: superblock to acquire
 514 *
 515 * Acquire a temporary reference on a superblock and try to trade it for
 516 * an active reference. This is used in sget{_fc}() to wait for a
 517 * superblock to either become SB_BORN or for it to pass through
 518 * sb->kill() and be marked as SB_DEAD.
 519 *
 520 * Return: This returns true if an active reference could be acquired,
 521 *         false if not.
 
 522 */
 523static bool grab_super(struct super_block *sb)
 524{
 525	bool locked;
 526
 527	sb->s_count++;
 528	spin_unlock(&sb_lock);
 529	locked = super_lock_excl(sb);
 530	if (locked) {
 531		if (atomic_inc_not_zero(&sb->s_active)) {
 532			put_super(sb);
 533			return true;
 534		}
 535		super_unlock_excl(sb);
 536	}
 537	wait_var_event(&sb->s_flags, super_flags(sb, SB_DEAD));
 538	put_super(sb);
 539	return false;
 540}
 541
 542/*
 543 *	super_trylock_shared - try to grab ->s_umount shared
 544 *	@sb: reference we are trying to grab
 545 *
 546 *	Try to prevent fs shutdown.  This is used in places where we
 547 *	cannot take an active reference but we need to ensure that the
 548 *	filesystem is not shut down while we are working on it. It returns
 549 *	false if we cannot acquire s_umount or if we lose the race and
 550 *	filesystem already got into shutdown, and returns true with the s_umount
 551 *	lock held in read mode in case of success. On successful return,
 552 *	the caller must drop the s_umount lock when done.
 553 *
 554 *	Note that unlike get_super() et.al. this one does *not* bump ->s_count.
 555 *	The reason why it's safe is that we are OK with doing trylock instead
 556 *	of down_read().  There's a couple of places that are OK with that, but
 557 *	it's very much not a general-purpose interface.
 558 */
 559bool super_trylock_shared(struct super_block *sb)
 560{
 561	if (down_read_trylock(&sb->s_umount)) {
 562		if (!(sb->s_flags & SB_DYING) && sb->s_root &&
 563		    (sb->s_flags & SB_BORN))
 564			return true;
 565		super_unlock_shared(sb);
 566	}
 567
 568	return false;
 569}
 570
 571/**
 572 *	retire_super	-	prevents superblock from being reused
 573 *	@sb: superblock to retire
 574 *
 575 *	The function marks superblock to be ignored in superblock test, which
 576 *	prevents it from being reused for any new mounts.  If the superblock has
 577 *	a private bdi, it also unregisters it, but doesn't reduce the refcount
 578 *	of the superblock to prevent potential races.  The refcount is reduced
 579 *	by generic_shutdown_super().  The function can not be called
 580 *	concurrently with generic_shutdown_super().  It is safe to call the
 581 *	function multiple times, subsequent calls have no effect.
 582 *
 583 *	The marker will affect the re-use only for block-device-based
 584 *	superblocks.  Other superblocks will still get marked if this function
 585 *	is used, but that will not affect their reusability.
 586 */
 587void retire_super(struct super_block *sb)
 588{
 589	WARN_ON(!sb->s_bdev);
 590	__super_lock_excl(sb);
 591	if (sb->s_iflags & SB_I_PERSB_BDI) {
 592		bdi_unregister(sb->s_bdi);
 593		sb->s_iflags &= ~SB_I_PERSB_BDI;
 594	}
 595	sb->s_iflags |= SB_I_RETIRED;
 596	super_unlock_excl(sb);
 597}
 598EXPORT_SYMBOL(retire_super);
 599
 600/**
 601 *	generic_shutdown_super	-	common helper for ->kill_sb()
 602 *	@sb: superblock to kill
 603 *
 604 *	generic_shutdown_super() does all fs-independent work on superblock
 605 *	shutdown.  Typical ->kill_sb() should pick all fs-specific objects
 606 *	that need destruction out of superblock, call generic_shutdown_super()
 607 *	and release aforementioned objects.  Note: dentries and inodes _are_
 608 *	taken care of and do not need specific handling.
 609 *
 610 *	Upon calling this function, the filesystem may no longer alter or
 611 *	rearrange the set of dentries belonging to this super_block, nor may it
 612 *	change the attachments of dentries to inodes.
 613 */
 614void generic_shutdown_super(struct super_block *sb)
 615{
 616	const struct super_operations *sop = sb->s_op;
 617
 618	if (sb->s_root) {
 619		shrink_dcache_for_umount(sb);
 620		sync_filesystem(sb);
 621		sb->s_flags &= ~SB_ACTIVE;
 622
 
 623		cgroup_writeback_umount();
 624
 625		/* Evict all inodes with zero refcount. */
 626		evict_inodes(sb);
 627
 628		/*
 629		 * Clean up and evict any inodes that still have references due
 630		 * to fsnotify or the security policy.
 631		 */
 632		fsnotify_sb_delete(sb);
 633		security_sb_delete(sb);
 634
 635		if (sb->s_dio_done_wq) {
 636			destroy_workqueue(sb->s_dio_done_wq);
 637			sb->s_dio_done_wq = NULL;
 638		}
 639
 640		if (sop->put_super)
 641			sop->put_super(sb);
 642
 643		/*
 644		 * Now that all potentially-encrypted inodes have been evicted,
 645		 * the fscrypt keyring can be destroyed.
 646		 */
 647		fscrypt_destroy_keyring(sb);
 648
 649		if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes),
 650				"VFS: Busy inodes after unmount of %s (%s)",
 651				sb->s_id, sb->s_type->name)) {
 652			/*
 653			 * Adding a proper bailout path here would be hard, but
 654			 * we can at least make it more likely that a later
 655			 * iput_final() or such crashes cleanly.
 656			 */
 657			struct inode *inode;
 658
 659			spin_lock(&sb->s_inode_list_lock);
 660			list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
 661				inode->i_op = VFS_PTR_POISON;
 662				inode->i_sb = VFS_PTR_POISON;
 663				inode->i_mapping = VFS_PTR_POISON;
 664			}
 665			spin_unlock(&sb->s_inode_list_lock);
 666		}
 667	}
 668	/*
 669	 * Broadcast to everyone that grabbed a temporary reference to this
 670	 * superblock before we removed it from @fs_supers that the superblock
 671	 * is dying. Every walker of @fs_supers outside of sget{_fc}() will now
 672	 * discard this superblock and treat it as dead.
 673	 *
 674	 * We leave the superblock on @fs_supers so it can be found by
 675	 * sget{_fc}() until we passed sb->kill_sb().
 676	 */
 677	super_wake(sb, SB_DYING);
 678	super_unlock_excl(sb);
 679	if (sb->s_bdi != &noop_backing_dev_info) {
 680		if (sb->s_iflags & SB_I_PERSB_BDI)
 681			bdi_unregister(sb->s_bdi);
 682		bdi_put(sb->s_bdi);
 683		sb->s_bdi = &noop_backing_dev_info;
 684	}
 685}
 686
 687EXPORT_SYMBOL(generic_shutdown_super);
 688
 689bool mount_capable(struct fs_context *fc)
 690{
 691	if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
 692		return capable(CAP_SYS_ADMIN);
 693	else
 694		return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
 695}
 696
 697/**
 698 * sget_fc - Find or create a superblock
 699 * @fc:	Filesystem context.
 700 * @test: Comparison callback
 701 * @set: Setup callback
 702 *
 703 * Create a new superblock or find an existing one.
 704 *
 705 * The @test callback is used to find a matching existing superblock.
 706 * Whether or not the requested parameters in @fc are taken into account
 707 * is specific to the @test callback that is used. They may even be
 708 * completely ignored.
 709 *
 710 * If an extant superblock is matched, it will be returned unless:
 711 *
 712 * (1) the namespace the filesystem context @fc and the extant
 713 *     superblock's namespace differ
 714 *
 715 * (2) the filesystem context @fc has requested that reusing an extant
 716 *     superblock is not allowed
 717 *
 718 * In both cases EBUSY will be returned.
 719 *
 720 * If no match is made, a new superblock will be allocated and basic
 721 * initialisation will be performed (s_type, s_fs_info and s_id will be
 722 * set and the @set callback will be invoked), the superblock will be
 723 * published and it will be returned in a partially constructed state
 724 * with SB_BORN and SB_ACTIVE as yet unset.
 725 *
 726 * Return: On success, an extant or newly created superblock is
 727 *         returned. On failure an error pointer is returned.
 728 */
 729struct super_block *sget_fc(struct fs_context *fc,
 730			    int (*test)(struct super_block *, struct fs_context *),
 731			    int (*set)(struct super_block *, struct fs_context *))
 
 
 732{
 733	struct super_block *s = NULL;
 734	struct super_block *old;
 735	struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
 736	int err;
 737
 
 
 
 
 738retry:
 739	spin_lock(&sb_lock);
 740	if (test) {
 741		hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
 742			if (test(old, fc))
 743				goto share_extant_sb;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 744		}
 745	}
 746	if (!s) {
 747		spin_unlock(&sb_lock);
 748		s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
 749		if (!s)
 750			return ERR_PTR(-ENOMEM);
 751		goto retry;
 752	}
 753
 754	s->s_fs_info = fc->s_fs_info;
 755	err = set(s, fc);
 756	if (err) {
 757		s->s_fs_info = NULL;
 758		spin_unlock(&sb_lock);
 759		destroy_unused_super(s);
 
 760		return ERR_PTR(err);
 761	}
 762	fc->s_fs_info = NULL;
 763	s->s_type = fc->fs_type;
 764	s->s_iflags |= fc->s_iflags;
 765	strscpy(s->s_id, s->s_type->name, sizeof(s->s_id));
 766	/*
 767	 * Make the superblock visible on @super_blocks and @fs_supers.
 768	 * It's in a nascent state and users should wait on SB_BORN or
 769	 * SB_DYING to be set.
 770	 */
 771	list_add_tail(&s->s_list, &super_blocks);
 772	hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
 773	spin_unlock(&sb_lock);
 774	get_filesystem(s->s_type);
 775	shrinker_register(s->s_shrink);
 776	return s;
 777
 778share_extant_sb:
 779	if (user_ns != old->s_user_ns || fc->exclusive) {
 780		spin_unlock(&sb_lock);
 781		destroy_unused_super(s);
 782		if (fc->exclusive)
 783			warnfc(fc, "reusing existing filesystem not allowed");
 784		else
 785			warnfc(fc, "reusing existing filesystem in another namespace not allowed");
 786		return ERR_PTR(-EBUSY);
 787	}
 788	if (!grab_super(old))
 789		goto retry;
 790	destroy_unused_super(s);
 791	return old;
 792}
 793EXPORT_SYMBOL(sget_fc);
 
 794
 795/**
 796 *	sget	-	find or create a superblock
 797 *	@type:	  filesystem type superblock should belong to
 798 *	@test:	  comparison callback
 799 *	@set:	  setup callback
 800 *	@flags:	  mount flags
 801 *	@data:	  argument to each of them
 802 */
 803struct super_block *sget(struct file_system_type *type,
 804			int (*test)(struct super_block *,void *),
 805			int (*set)(struct super_block *,void *),
 806			int flags,
 807			void *data)
 808{
 809	struct user_namespace *user_ns = current_user_ns();
 810	struct super_block *s = NULL;
 811	struct super_block *old;
 812	int err;
 813
 814	/* We don't yet pass the user namespace of the parent
 815	 * mount through to here so always use &init_user_ns
 816	 * until that changes.
 817	 */
 818	if (flags & SB_SUBMOUNT)
 819		user_ns = &init_user_ns;
 820
 821retry:
 822	spin_lock(&sb_lock);
 823	if (test) {
 824		hlist_for_each_entry(old, &type->fs_supers, s_instances) {
 825			if (!test(old, data))
 826				continue;
 827			if (user_ns != old->s_user_ns) {
 828				spin_unlock(&sb_lock);
 829				destroy_unused_super(s);
 830				return ERR_PTR(-EBUSY);
 831			}
 832			if (!grab_super(old))
 833				goto retry;
 834			destroy_unused_super(s);
 835			return old;
 836		}
 837	}
 838	if (!s) {
 839		spin_unlock(&sb_lock);
 840		s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
 841		if (!s)
 842			return ERR_PTR(-ENOMEM);
 843		goto retry;
 844	}
 845
 846	err = set(s, data);
 847	if (err) {
 848		spin_unlock(&sb_lock);
 849		destroy_unused_super(s);
 850		return ERR_PTR(err);
 851	}
 852	s->s_type = type;
 853	strscpy(s->s_id, type->name, sizeof(s->s_id));
 854	list_add_tail(&s->s_list, &super_blocks);
 855	hlist_add_head(&s->s_instances, &type->fs_supers);
 856	spin_unlock(&sb_lock);
 857	get_filesystem(type);
 858	shrinker_register(s->s_shrink);
 859	return s;
 860}
 
 861EXPORT_SYMBOL(sget);
 862
 863void drop_super(struct super_block *sb)
 864{
 865	super_unlock_shared(sb);
 866	put_super(sb);
 867}
 868
 869EXPORT_SYMBOL(drop_super);
 870
 871void drop_super_exclusive(struct super_block *sb)
 872{
 873	super_unlock_excl(sb);
 874	put_super(sb);
 875}
 876EXPORT_SYMBOL(drop_super_exclusive);
 877
 878static void __iterate_supers(void (*f)(struct super_block *))
 879{
 880	struct super_block *sb, *p = NULL;
 881
 882	spin_lock(&sb_lock);
 883	list_for_each_entry(sb, &super_blocks, s_list) {
 884		if (super_flags(sb, SB_DYING))
 885			continue;
 886		sb->s_count++;
 887		spin_unlock(&sb_lock);
 888
 889		f(sb);
 890
 891		spin_lock(&sb_lock);
 892		if (p)
 893			__put_super(p);
 894		p = sb;
 895	}
 896	if (p)
 897		__put_super(p);
 898	spin_unlock(&sb_lock);
 899}
 900/**
 901 *	iterate_supers - call function for all active superblocks
 902 *	@f: function to call
 903 *	@arg: argument to pass to it
 904 *
 905 *	Scans the superblock list and calls given function, passing it
 906 *	locked superblock and given argument.
 907 */
 908void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
 909{
 910	struct super_block *sb, *p = NULL;
 911
 912	spin_lock(&sb_lock);
 913	list_for_each_entry(sb, &super_blocks, s_list) {
 914		bool locked;
 915
 916		sb->s_count++;
 917		spin_unlock(&sb_lock);
 918
 919		locked = super_lock_shared(sb);
 920		if (locked) {
 921			if (sb->s_root)
 922				f(sb, arg);
 923			super_unlock_shared(sb);
 924		}
 925
 926		spin_lock(&sb_lock);
 927		if (p)
 928			__put_super(p);
 929		p = sb;
 930	}
 931	if (p)
 932		__put_super(p);
 933	spin_unlock(&sb_lock);
 934}
 935
 936/**
 937 *	iterate_supers_type - call function for superblocks of given type
 938 *	@type: fs type
 939 *	@f: function to call
 940 *	@arg: argument to pass to it
 941 *
 942 *	Scans the superblock list and calls given function, passing it
 943 *	locked superblock and given argument.
 944 */
 945void iterate_supers_type(struct file_system_type *type,
 946	void (*f)(struct super_block *, void *), void *arg)
 947{
 948	struct super_block *sb, *p = NULL;
 949
 950	spin_lock(&sb_lock);
 951	hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
 952		bool locked;
 953
 954		sb->s_count++;
 955		spin_unlock(&sb_lock);
 956
 957		locked = super_lock_shared(sb);
 958		if (locked) {
 959			if (sb->s_root)
 960				f(sb, arg);
 961			super_unlock_shared(sb);
 962		}
 963
 964		spin_lock(&sb_lock);
 965		if (p)
 966			__put_super(p);
 967		p = sb;
 968	}
 969	if (p)
 970		__put_super(p);
 971	spin_unlock(&sb_lock);
 972}
 973
 974EXPORT_SYMBOL(iterate_supers_type);
 975
 976struct super_block *user_get_super(dev_t dev, bool excl)
 977{
 978	struct super_block *sb;
 979
 
 
 
 980	spin_lock(&sb_lock);
 
 981	list_for_each_entry(sb, &super_blocks, s_list) {
 982		if (sb->s_dev ==  dev) {
 983			bool locked;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 984
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 985			sb->s_count++;
 986			spin_unlock(&sb_lock);
 
 987			/* still alive? */
 988			locked = super_lock(sb, excl);
 989			if (locked) {
 990				if (sb->s_root)
 991					return sb;
 992				super_unlock(sb, excl);
 993			}
 994			/* nope, got unmounted */
 995			spin_lock(&sb_lock);
 996			__put_super(sb);
 997			break;
 998		}
 999	}
1000	spin_unlock(&sb_lock);
1001	return NULL;
1002}
1003
1004/**
1005 * reconfigure_super - asks filesystem to change superblock parameters
1006 * @fc: The superblock and configuration
 
 
 
1007 *
1008 * Alters the configuration parameters of a live superblock.
1009 */
1010int reconfigure_super(struct fs_context *fc)
1011{
1012	struct super_block *sb = fc->root->d_sb;
1013	int retval;
1014	bool remount_ro = false;
1015	bool remount_rw = false;
1016	bool force = fc->sb_flags & SB_FORCE;
1017
1018	if (fc->sb_flags_mask & ~MS_RMT_MASK)
1019		return -EINVAL;
1020	if (sb->s_writers.frozen != SB_UNFROZEN)
1021		return -EBUSY;
1022
1023	retval = security_sb_remount(sb, fc->security);
1024	if (retval)
1025		return retval;
1026
1027	if (fc->sb_flags_mask & SB_RDONLY) {
1028#ifdef CONFIG_BLOCK
1029		if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
1030		    bdev_read_only(sb->s_bdev))
1031			return -EACCES;
1032#endif
1033		remount_rw = !(fc->sb_flags & SB_RDONLY) && sb_rdonly(sb);
1034		remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
1035	}
1036
1037	if (remount_ro) {
1038		if (!hlist_empty(&sb->s_pins)) {
1039			super_unlock_excl(sb);
1040			group_pin_kill(&sb->s_pins);
1041			__super_lock_excl(sb);
1042			if (!sb->s_root)
1043				return 0;
1044			if (sb->s_writers.frozen != SB_UNFROZEN)
1045				return -EBUSY;
1046			remount_ro = !sb_rdonly(sb);
1047		}
1048	}
1049	shrink_dcache_sb(sb);
1050
1051	/* If we are reconfiguring to RDONLY and current sb is read/write,
1052	 * make sure there are no files open for writing.
1053	 */
1054	if (remount_ro) {
1055		if (force) {
1056			sb_start_ro_state_change(sb);
 
1057		} else {
1058			retval = sb_prepare_remount_readonly(sb);
1059			if (retval)
1060				return retval;
1061		}
1062	} else if (remount_rw) {
1063		/*
1064		 * Protect filesystem's reconfigure code from writes from
1065		 * userspace until reconfigure finishes.
1066		 */
1067		sb_start_ro_state_change(sb);
1068	}
1069
1070	if (fc->ops->reconfigure) {
1071		retval = fc->ops->reconfigure(fc);
1072		if (retval) {
1073			if (!force)
1074				goto cancel_readonly;
1075			/* If forced remount, go ahead despite any errors */
1076			WARN(1, "forced remount of a %s fs returned %i\n",
1077			     sb->s_type->name, retval);
1078		}
1079	}
1080
1081	WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
1082				 (fc->sb_flags & fc->sb_flags_mask)));
1083	sb_end_ro_state_change(sb);
1084
1085	/*
1086	 * Some filesystems modify their metadata via some other path than the
1087	 * bdev buffer cache (eg. use a private mapping, or directories in
1088	 * pagecache, etc). Also file data modifications go via their own
1089	 * mappings. So If we try to mount readonly then copy the filesystem
1090	 * from bdev, we could get stale data, so invalidate it to give a best
1091	 * effort at coherency.
1092	 */
1093	if (remount_ro && sb->s_bdev)
1094		invalidate_bdev(sb->s_bdev);
1095	return 0;
1096
1097cancel_readonly:
1098	sb_end_ro_state_change(sb);
1099	return retval;
1100}
1101
1102static void do_emergency_remount_callback(struct super_block *sb)
1103{
1104	bool locked = super_lock_excl(sb);
1105
1106	if (locked && sb->s_root && sb->s_bdev && !sb_rdonly(sb)) {
1107		struct fs_context *fc;
1108
1109		fc = fs_context_for_reconfigure(sb->s_root,
1110					SB_RDONLY | SB_FORCE, SB_RDONLY);
1111		if (!IS_ERR(fc)) {
1112			if (parse_monolithic_mount_data(fc, NULL) == 0)
1113				(void)reconfigure_super(fc);
1114			put_fs_context(fc);
 
 
 
 
 
 
 
1115		}
 
 
 
 
 
1116	}
1117	if (locked)
1118		super_unlock_excl(sb);
1119}
1120
1121static void do_emergency_remount(struct work_struct *work)
1122{
1123	__iterate_supers(do_emergency_remount_callback);
1124	kfree(work);
1125	printk("Emergency Remount complete\n");
1126}
1127
1128void emergency_remount(void)
1129{
1130	struct work_struct *work;
1131
1132	work = kmalloc(sizeof(*work), GFP_ATOMIC);
1133	if (work) {
1134		INIT_WORK(work, do_emergency_remount);
1135		schedule_work(work);
1136	}
1137}
1138
1139static void do_thaw_all_callback(struct super_block *sb)
1140{
1141	bool locked = super_lock_excl(sb);
1142
1143	if (locked && sb->s_root) {
1144		if (IS_ENABLED(CONFIG_BLOCK))
1145			while (sb->s_bdev && !bdev_thaw(sb->s_bdev))
1146				pr_warn("Emergency Thaw on %pg\n", sb->s_bdev);
1147		thaw_super_locked(sb, FREEZE_HOLDER_USERSPACE);
1148		return;
1149	}
1150	if (locked)
1151		super_unlock_excl(sb);
1152}
1153
1154static void do_thaw_all(struct work_struct *work)
1155{
1156	__iterate_supers(do_thaw_all_callback);
1157	kfree(work);
1158	printk(KERN_WARNING "Emergency Thaw complete\n");
1159}
1160
1161/**
1162 * emergency_thaw_all -- forcibly thaw every frozen filesystem
1163 *
1164 * Used for emergency unfreeze of all filesystems via SysRq
1165 */
1166void emergency_thaw_all(void)
1167{
1168	struct work_struct *work;
1169
1170	work = kmalloc(sizeof(*work), GFP_ATOMIC);
1171	if (work) {
1172		INIT_WORK(work, do_thaw_all);
1173		schedule_work(work);
1174	}
1175}
1176
1177static DEFINE_IDA(unnamed_dev_ida);
1178
1179/**
1180 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1181 * @p: Pointer to a dev_t.
1182 *
1183 * Filesystems which don't use real block devices can call this function
1184 * to allocate a virtual block device.
1185 *
1186 * Context: Any context.  Frequently called while holding sb_lock.
1187 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1188 * or -ENOMEM if memory allocation failed.
1189 */
 
 
1190int get_anon_bdev(dev_t *p)
1191{
1192	int dev;
 
1193
1194	/*
1195	 * Many userspace utilities consider an FSID of 0 invalid.
1196	 * Always return at least 1 from get_anon_bdev.
1197	 */
1198	dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1199			GFP_ATOMIC);
1200	if (dev == -ENOSPC)
1201		dev = -EMFILE;
1202	if (dev < 0)
1203		return dev;
 
 
 
1204
1205	*p = MKDEV(0, dev);
 
 
 
 
 
 
 
 
1206	return 0;
1207}
1208EXPORT_SYMBOL(get_anon_bdev);
1209
1210void free_anon_bdev(dev_t dev)
1211{
1212	ida_free(&unnamed_dev_ida, MINOR(dev));
 
 
 
 
 
1213}
1214EXPORT_SYMBOL(free_anon_bdev);
1215
1216int set_anon_super(struct super_block *s, void *data)
1217{
1218	return get_anon_bdev(&s->s_dev);
1219}
 
1220EXPORT_SYMBOL(set_anon_super);
1221
1222void kill_anon_super(struct super_block *sb)
1223{
1224	dev_t dev = sb->s_dev;
1225	generic_shutdown_super(sb);
1226	kill_super_notify(sb);
1227	free_anon_bdev(dev);
1228}
 
1229EXPORT_SYMBOL(kill_anon_super);
1230
1231void kill_litter_super(struct super_block *sb)
1232{
1233	if (sb->s_root)
1234		d_genocide(sb->s_root);
1235	kill_anon_super(sb);
1236}
1237EXPORT_SYMBOL(kill_litter_super);
1238
1239int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1240{
1241	return set_anon_super(sb, NULL);
1242}
1243EXPORT_SYMBOL(set_anon_super_fc);
1244
1245static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1246{
1247	return sb->s_fs_info == fc->s_fs_info;
1248}
1249
1250static int test_single_super(struct super_block *s, struct fs_context *fc)
1251{
1252	return 1;
 
1253}
1254
1255static int vfs_get_super(struct fs_context *fc,
1256		int (*test)(struct super_block *, struct fs_context *),
1257		int (*fill_super)(struct super_block *sb,
1258				  struct fs_context *fc))
1259{
1260	struct super_block *sb;
1261	int err;
1262
1263	sb = sget_fc(fc, test, set_anon_super_fc);
 
 
 
 
 
 
 
1264	if (IS_ERR(sb))
1265		return PTR_ERR(sb);
1266
1267	if (!sb->s_root) {
1268		err = fill_super(sb, fc);
1269		if (err)
1270			goto error;
 
 
 
1271
1272		sb->s_flags |= SB_ACTIVE;
1273	}
1274
1275	fc->root = dget(sb->s_root);
1276	return 0;
1277
1278error:
1279	deactivate_locked_super(sb);
1280	return err;
1281}
1282
1283int get_tree_nodev(struct fs_context *fc,
1284		  int (*fill_super)(struct super_block *sb,
1285				    struct fs_context *fc))
1286{
1287	return vfs_get_super(fc, NULL, fill_super);
1288}
1289EXPORT_SYMBOL(get_tree_nodev);
1290
1291int get_tree_single(struct fs_context *fc,
1292		  int (*fill_super)(struct super_block *sb,
1293				    struct fs_context *fc))
1294{
1295	return vfs_get_super(fc, test_single_super, fill_super);
1296}
1297EXPORT_SYMBOL(get_tree_single);
1298
1299int get_tree_keyed(struct fs_context *fc,
1300		  int (*fill_super)(struct super_block *sb,
1301				    struct fs_context *fc),
1302		void *key)
1303{
1304	fc->s_fs_info = key;
1305	return vfs_get_super(fc, test_keyed_super, fill_super);
1306}
1307EXPORT_SYMBOL(get_tree_keyed);
1308
1309static int set_bdev_super(struct super_block *s, void *data)
1310{
1311	s->s_dev = *(dev_t *)data;
1312	return 0;
1313}
1314
1315static int super_s_dev_set(struct super_block *s, struct fs_context *fc)
1316{
1317	return set_bdev_super(s, fc->sget_key);
1318}
1319
1320static int super_s_dev_test(struct super_block *s, struct fs_context *fc)
1321{
1322	return !(s->s_iflags & SB_I_RETIRED) &&
1323		s->s_dev == *(dev_t *)fc->sget_key;
1324}
1325
1326/**
1327 * sget_dev - Find or create a superblock by device number
1328 * @fc: Filesystem context.
1329 * @dev: device number
1330 *
1331 * Find or create a superblock using the provided device number that
1332 * will be stored in fc->sget_key.
1333 *
1334 * If an extant superblock is matched, then that will be returned with
1335 * an elevated reference count that the caller must transfer or discard.
1336 *
1337 * If no match is made, a new superblock will be allocated and basic
1338 * initialisation will be performed (s_type, s_fs_info, s_id, s_dev will
1339 * be set). The superblock will be published and it will be returned in
1340 * a partially constructed state with SB_BORN and SB_ACTIVE as yet
1341 * unset.
1342 *
1343 * Return: an existing or newly created superblock on success, an error
1344 *         pointer on failure.
1345 */
1346struct super_block *sget_dev(struct fs_context *fc, dev_t dev)
1347{
1348	fc->sget_key = &dev;
1349	return sget_fc(fc, super_s_dev_test, super_s_dev_set);
1350}
1351EXPORT_SYMBOL(sget_dev);
1352
1353#ifdef CONFIG_BLOCK
1354/*
1355 * Lock the superblock that is holder of the bdev. Returns the superblock
1356 * pointer if we successfully locked the superblock and it is alive. Otherwise
1357 * we return NULL and just unlock bdev->bd_holder_lock.
1358 *
1359 * The function must be called with bdev->bd_holder_lock and releases it.
1360 */
1361static struct super_block *bdev_super_lock(struct block_device *bdev, bool excl)
1362	__releases(&bdev->bd_holder_lock)
1363{
1364	struct super_block *sb = bdev->bd_holder;
1365	bool locked;
1366
1367	lockdep_assert_held(&bdev->bd_holder_lock);
1368	lockdep_assert_not_held(&sb->s_umount);
1369	lockdep_assert_not_held(&bdev->bd_disk->open_mutex);
1370
1371	/* Make sure sb doesn't go away from under us */
1372	spin_lock(&sb_lock);
1373	sb->s_count++;
1374	spin_unlock(&sb_lock);
1375
1376	mutex_unlock(&bdev->bd_holder_lock);
1377
1378	locked = super_lock(sb, excl);
1379
1380	/*
1381	 * If the superblock wasn't already SB_DYING then we hold
1382	 * s_umount and can safely drop our temporary reference.
1383         */
1384	put_super(sb);
1385
1386	if (!locked)
1387		return NULL;
1388
1389	if (!sb->s_root || !(sb->s_flags & SB_ACTIVE)) {
1390		super_unlock(sb, excl);
1391		return NULL;
1392	}
1393
1394	return sb;
1395}
1396
1397static void fs_bdev_mark_dead(struct block_device *bdev, bool surprise)
1398{
1399	struct super_block *sb;
1400
1401	sb = bdev_super_lock(bdev, false);
1402	if (!sb)
1403		return;
1404
1405	if (!surprise)
1406		sync_filesystem(sb);
1407	shrink_dcache_sb(sb);
1408	invalidate_inodes(sb);
1409	if (sb->s_op->shutdown)
1410		sb->s_op->shutdown(sb);
1411
1412	super_unlock_shared(sb);
1413}
1414
1415static void fs_bdev_sync(struct block_device *bdev)
1416{
1417	struct super_block *sb;
1418
1419	sb = bdev_super_lock(bdev, false);
1420	if (!sb)
1421		return;
1422
1423	sync_filesystem(sb);
1424	super_unlock_shared(sb);
1425}
1426
1427static struct super_block *get_bdev_super(struct block_device *bdev)
1428{
1429	bool active = false;
1430	struct super_block *sb;
1431
1432	sb = bdev_super_lock(bdev, true);
1433	if (sb) {
1434		active = atomic_inc_not_zero(&sb->s_active);
1435		super_unlock_excl(sb);
1436	}
1437	if (!active)
1438		return NULL;
1439	return sb;
1440}
1441
1442/**
1443 * fs_bdev_freeze - freeze owning filesystem of block device
1444 * @bdev: block device
1445 *
1446 * Freeze the filesystem that owns this block device if it is still
1447 * active.
1448 *
1449 * A filesystem that owns multiple block devices may be frozen from each
1450 * block device and won't be unfrozen until all block devices are
1451 * unfrozen. Each block device can only freeze the filesystem once as we
1452 * nest freezes for block devices in the block layer.
1453 *
1454 * Return: If the freeze was successful zero is returned. If the freeze
1455 *         failed a negative error code is returned.
1456 */
1457static int fs_bdev_freeze(struct block_device *bdev)
1458{
1459	struct super_block *sb;
1460	int error = 0;
1461
1462	lockdep_assert_held(&bdev->bd_fsfreeze_mutex);
1463
1464	sb = get_bdev_super(bdev);
1465	if (!sb)
1466		return -EINVAL;
1467
1468	if (sb->s_op->freeze_super)
1469		error = sb->s_op->freeze_super(sb,
1470				FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE);
1471	else
1472		error = freeze_super(sb,
1473				FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE);
1474	if (!error)
1475		error = sync_blockdev(bdev);
1476	deactivate_super(sb);
1477	return error;
1478}
1479
1480/**
1481 * fs_bdev_thaw - thaw owning filesystem of block device
1482 * @bdev: block device
1483 *
1484 * Thaw the filesystem that owns this block device.
1485 *
1486 * A filesystem that owns multiple block devices may be frozen from each
1487 * block device and won't be unfrozen until all block devices are
1488 * unfrozen. Each block device can only freeze the filesystem once as we
1489 * nest freezes for block devices in the block layer.
1490 *
1491 * Return: If the thaw was successful zero is returned. If the thaw
1492 *         failed a negative error code is returned. If this function
1493 *         returns zero it doesn't mean that the filesystem is unfrozen
1494 *         as it may have been frozen multiple times (kernel may hold a
1495 *         freeze or might be frozen from other block devices).
1496 */
1497static int fs_bdev_thaw(struct block_device *bdev)
1498{
1499	struct super_block *sb;
1500	int error;
1501
1502	lockdep_assert_held(&bdev->bd_fsfreeze_mutex);
1503
1504	sb = get_bdev_super(bdev);
1505	if (WARN_ON_ONCE(!sb))
1506		return -EINVAL;
1507
1508	if (sb->s_op->thaw_super)
1509		error = sb->s_op->thaw_super(sb,
1510				FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE);
1511	else
1512		error = thaw_super(sb,
1513				FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE);
1514	deactivate_super(sb);
1515	return error;
1516}
1517
1518const struct blk_holder_ops fs_holder_ops = {
1519	.mark_dead		= fs_bdev_mark_dead,
1520	.sync			= fs_bdev_sync,
1521	.freeze			= fs_bdev_freeze,
1522	.thaw			= fs_bdev_thaw,
1523};
1524EXPORT_SYMBOL_GPL(fs_holder_ops);
1525
1526int setup_bdev_super(struct super_block *sb, int sb_flags,
1527		struct fs_context *fc)
1528{
1529	blk_mode_t mode = sb_open_mode(sb_flags);
1530	struct bdev_handle *bdev_handle;
1531	struct block_device *bdev;
1532
1533	bdev_handle = bdev_open_by_dev(sb->s_dev, mode, sb, &fs_holder_ops);
1534	if (IS_ERR(bdev_handle)) {
1535		if (fc)
1536			errorf(fc, "%s: Can't open blockdev", fc->source);
1537		return PTR_ERR(bdev_handle);
1538	}
1539	bdev = bdev_handle->bdev;
1540
1541	/*
1542	 * This really should be in blkdev_get_by_dev, but right now can't due
1543	 * to legacy issues that require us to allow opening a block device node
1544	 * writable from userspace even for a read-only block device.
1545	 */
1546	if ((mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
1547		bdev_release(bdev_handle);
1548		return -EACCES;
1549	}
1550
1551	/*
1552	 * It is enough to check bdev was not frozen before we set
1553	 * s_bdev as freezing will wait until SB_BORN is set.
1554	 */
1555	if (atomic_read(&bdev->bd_fsfreeze_count) > 0) {
1556		if (fc)
1557			warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1558		bdev_release(bdev_handle);
1559		return -EBUSY;
1560	}
1561	spin_lock(&sb_lock);
1562	sb->s_bdev_handle = bdev_handle;
1563	sb->s_bdev = bdev;
1564	sb->s_bdi = bdi_get(bdev->bd_disk->bdi);
1565	if (bdev_stable_writes(bdev))
1566		sb->s_iflags |= SB_I_STABLE_WRITES;
1567	spin_unlock(&sb_lock);
1568
1569	snprintf(sb->s_id, sizeof(sb->s_id), "%pg", bdev);
1570	shrinker_debugfs_rename(sb->s_shrink, "sb-%s:%s", sb->s_type->name,
1571				sb->s_id);
1572	sb_set_blocksize(sb, block_size(bdev));
1573	return 0;
1574}
1575EXPORT_SYMBOL_GPL(setup_bdev_super);
1576
1577/**
1578 * get_tree_bdev - Get a superblock based on a single block device
1579 * @fc: The filesystem context holding the parameters
1580 * @fill_super: Helper to initialise a new superblock
1581 */
1582int get_tree_bdev(struct fs_context *fc,
1583		int (*fill_super)(struct super_block *,
1584				  struct fs_context *))
1585{
1586	struct super_block *s;
1587	int error = 0;
1588	dev_t dev;
1589
1590	if (!fc->source)
1591		return invalf(fc, "No source specified");
1592
1593	error = lookup_bdev(fc->source, &dev);
1594	if (error) {
1595		errorf(fc, "%s: Can't lookup blockdev", fc->source);
1596		return error;
1597	}
1598
1599	fc->sb_flags |= SB_NOSEC;
1600	s = sget_dev(fc, dev);
1601	if (IS_ERR(s))
1602		return PTR_ERR(s);
1603
1604	if (s->s_root) {
1605		/* Don't summarily change the RO/RW state. */
1606		if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1607			warnf(fc, "%pg: Can't mount, would change RO state", s->s_bdev);
1608			deactivate_locked_super(s);
1609			return -EBUSY;
1610		}
1611	} else {
1612		error = setup_bdev_super(s, fc->sb_flags, fc);
1613		if (!error)
1614			error = fill_super(s, fc);
1615		if (error) {
1616			deactivate_locked_super(s);
1617			return error;
1618		}
1619		s->s_flags |= SB_ACTIVE;
1620	}
1621
1622	BUG_ON(fc->root);
1623	fc->root = dget(s->s_root);
1624	return 0;
1625}
1626EXPORT_SYMBOL(get_tree_bdev);
1627
1628static int test_bdev_super(struct super_block *s, void *data)
1629{
1630	return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data;
1631}
1632
1633struct dentry *mount_bdev(struct file_system_type *fs_type,
1634	int flags, const char *dev_name, void *data,
1635	int (*fill_super)(struct super_block *, void *, int))
1636{
 
1637	struct super_block *s;
1638	int error;
1639	dev_t dev;
1640
1641	error = lookup_bdev(dev_name, &dev);
1642	if (error)
1643		return ERR_PTR(error);
1644
1645	flags |= SB_NOSEC;
1646	s = sget(fs_type, test_bdev_super, set_bdev_super, flags, &dev);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1647	if (IS_ERR(s))
1648		return ERR_CAST(s);
1649
1650	if (s->s_root) {
1651		if ((flags ^ s->s_flags) & SB_RDONLY) {
1652			deactivate_locked_super(s);
1653			return ERR_PTR(-EBUSY);
 
1654		}
 
 
 
 
 
 
 
 
 
 
 
1655	} else {
1656		error = setup_bdev_super(s, flags, NULL);
1657		if (!error)
1658			error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
 
1659		if (error) {
1660			deactivate_locked_super(s);
1661			return ERR_PTR(error);
1662		}
1663
1664		s->s_flags |= SB_ACTIVE;
 
1665	}
1666
1667	return dget(s->s_root);
 
 
 
 
 
 
 
1668}
1669EXPORT_SYMBOL(mount_bdev);
1670
1671void kill_block_super(struct super_block *sb)
1672{
1673	struct block_device *bdev = sb->s_bdev;
 
1674
 
1675	generic_shutdown_super(sb);
1676	if (bdev) {
1677		sync_blockdev(bdev);
1678		bdev_release(sb->s_bdev_handle);
1679	}
1680}
1681
1682EXPORT_SYMBOL(kill_block_super);
1683#endif
1684
1685struct dentry *mount_nodev(struct file_system_type *fs_type,
1686	int flags, void *data,
1687	int (*fill_super)(struct super_block *, void *, int))
1688{
1689	int error;
1690	struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1691
1692	if (IS_ERR(s))
1693		return ERR_CAST(s);
1694
1695	error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1696	if (error) {
1697		deactivate_locked_super(s);
1698		return ERR_PTR(error);
1699	}
1700	s->s_flags |= SB_ACTIVE;
1701	return dget(s->s_root);
1702}
1703EXPORT_SYMBOL(mount_nodev);
1704
1705int reconfigure_single(struct super_block *s,
1706		       int flags, void *data)
1707{
1708	struct fs_context *fc;
1709	int ret;
1710
1711	/* The caller really need to be passing fc down into mount_single(),
1712	 * then a chunk of this can be removed.  [Bollocks -- AV]
1713	 * Better yet, reconfiguration shouldn't happen, but rather the second
1714	 * mount should be rejected if the parameters are not compatible.
1715	 */
1716	fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1717	if (IS_ERR(fc))
1718		return PTR_ERR(fc);
1719
1720	ret = parse_monolithic_mount_data(fc, data);
1721	if (ret < 0)
1722		goto out;
1723
1724	ret = reconfigure_super(fc);
1725out:
1726	put_fs_context(fc);
1727	return ret;
1728}
1729
1730static int compare_single(struct super_block *s, void *p)
1731{
1732	return 1;
1733}
1734
1735struct dentry *mount_single(struct file_system_type *fs_type,
1736	int flags, void *data,
1737	int (*fill_super)(struct super_block *, void *, int))
1738{
1739	struct super_block *s;
1740	int error;
1741
1742	s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1743	if (IS_ERR(s))
1744		return ERR_CAST(s);
1745	if (!s->s_root) {
1746		error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1747		if (!error)
1748			s->s_flags |= SB_ACTIVE;
 
 
 
1749	} else {
1750		error = reconfigure_single(s, flags, data);
1751	}
1752	if (unlikely(error)) {
1753		deactivate_locked_super(s);
1754		return ERR_PTR(error);
1755	}
1756	return dget(s->s_root);
1757}
1758EXPORT_SYMBOL(mount_single);
1759
1760/**
1761 * vfs_get_tree - Get the mountable root
1762 * @fc: The superblock configuration context.
1763 *
1764 * The filesystem is invoked to get or create a superblock which can then later
1765 * be used for mounting.  The filesystem places a pointer to the root to be
1766 * used for mounting in @fc->root.
1767 */
1768int vfs_get_tree(struct fs_context *fc)
1769{
 
1770	struct super_block *sb;
1771	int error;
1772
1773	if (fc->root)
1774		return -EBUSY;
1775
1776	/* Get the mountable root in fc->root, with a ref on the root and a ref
1777	 * on the superblock.
1778	 */
1779	error = fc->ops->get_tree(fc);
1780	if (error < 0)
1781		return error;
1782
1783	if (!fc->root) {
1784		pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1785		       fc->fs_type->name);
1786		/* We don't know what the locking state of the superblock is -
1787		 * if there is a superblock.
1788		 */
1789		BUG();
1790	}
1791
1792	sb = fc->root->d_sb;
1793	WARN_ON(!sb->s_bdi);
 
1794
1795	/*
1796	 * super_wake() contains a memory barrier which also care of
1797	 * ordering for super_cache_count(). We place it before setting
1798	 * SB_BORN as the data dependency between the two functions is
1799	 * the superblock structure contents that we just set up, not
1800	 * the SB_BORN flag.
1801	 */
1802	super_wake(sb, SB_BORN);
1803
1804	error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1805	if (unlikely(error)) {
1806		fc_drop_locked(fc);
1807		return error;
1808	}
1809
1810	/*
1811	 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1812	 * but s_maxbytes was an unsigned long long for many releases. Throw
1813	 * this warning for a little while to try and catch filesystems that
1814	 * violate this rule.
1815	 */
1816	WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1817		"negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1818
1819	return 0;
 
 
 
 
 
 
 
 
 
1820}
1821EXPORT_SYMBOL(vfs_get_tree);
1822
1823/*
1824 * Setup private BDI for given superblock. It gets automatically cleaned up
1825 * in generic_shutdown_super().
1826 */
1827int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1828{
1829	struct backing_dev_info *bdi;
1830	int err;
1831	va_list args;
1832
1833	bdi = bdi_alloc(NUMA_NO_NODE);
1834	if (!bdi)
1835		return -ENOMEM;
1836
1837	va_start(args, fmt);
1838	err = bdi_register_va(bdi, fmt, args);
1839	va_end(args);
1840	if (err) {
1841		bdi_put(bdi);
1842		return err;
1843	}
1844	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1845	sb->s_bdi = bdi;
1846	sb->s_iflags |= SB_I_PERSB_BDI;
1847
1848	return 0;
1849}
1850EXPORT_SYMBOL(super_setup_bdi_name);
1851
1852/*
1853 * Setup private BDI for given superblock. I gets automatically cleaned up
1854 * in generic_shutdown_super().
1855 */
1856int super_setup_bdi(struct super_block *sb)
1857{
1858	static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
 
1859
1860	return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1861				    atomic_long_inc_return(&bdi_seq));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1862}
1863EXPORT_SYMBOL(super_setup_bdi);
1864
1865/**
1866 * sb_wait_write - wait until all writers to given file system finish
1867 * @sb: the super for which we wait
1868 * @level: type of writers we wait for (normal vs page fault)
1869 *
1870 * This function waits until there are no writers of given type to given file
1871 * system.
1872 */
1873static void sb_wait_write(struct super_block *sb, int level)
1874{
1875	percpu_down_write(sb->s_writers.rw_sem + level-1);
1876}
1877
1878/*
1879 * We are going to return to userspace and forget about these locks, the
1880 * ownership goes to the caller of thaw_super() which does unlock().
1881 */
1882static void lockdep_sb_freeze_release(struct super_block *sb)
1883{
1884	int level;
1885
1886	for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1887		percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1888}
1889
1890/*
1891 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1892 */
1893static void lockdep_sb_freeze_acquire(struct super_block *sb)
1894{
1895	int level;
1896
1897	for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1898		percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1899}
1900
1901static void sb_freeze_unlock(struct super_block *sb, int level)
1902{
1903	for (level--; level >= 0; level--)
1904		percpu_up_write(sb->s_writers.rw_sem + level);
1905}
1906
1907static int wait_for_partially_frozen(struct super_block *sb)
1908{
1909	int ret = 0;
1910
1911	do {
1912		unsigned short old = sb->s_writers.frozen;
1913
1914		up_write(&sb->s_umount);
1915		ret = wait_var_event_killable(&sb->s_writers.frozen,
1916					       sb->s_writers.frozen != old);
1917		down_write(&sb->s_umount);
1918	} while (ret == 0 &&
1919		 sb->s_writers.frozen != SB_UNFROZEN &&
1920		 sb->s_writers.frozen != SB_FREEZE_COMPLETE);
1921
1922	return ret;
1923}
1924
1925#define FREEZE_HOLDERS (FREEZE_HOLDER_KERNEL | FREEZE_HOLDER_USERSPACE)
1926#define FREEZE_FLAGS (FREEZE_HOLDERS | FREEZE_MAY_NEST)
1927
1928static inline int freeze_inc(struct super_block *sb, enum freeze_holder who)
1929{
1930	WARN_ON_ONCE((who & ~FREEZE_FLAGS));
1931	WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
1932
1933	if (who & FREEZE_HOLDER_KERNEL)
1934		++sb->s_writers.freeze_kcount;
1935	if (who & FREEZE_HOLDER_USERSPACE)
1936		++sb->s_writers.freeze_ucount;
1937	return sb->s_writers.freeze_kcount + sb->s_writers.freeze_ucount;
1938}
1939
1940static inline int freeze_dec(struct super_block *sb, enum freeze_holder who)
1941{
1942	WARN_ON_ONCE((who & ~FREEZE_FLAGS));
1943	WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
1944
1945	if ((who & FREEZE_HOLDER_KERNEL) && sb->s_writers.freeze_kcount)
1946		--sb->s_writers.freeze_kcount;
1947	if ((who & FREEZE_HOLDER_USERSPACE) && sb->s_writers.freeze_ucount)
1948		--sb->s_writers.freeze_ucount;
1949	return sb->s_writers.freeze_kcount + sb->s_writers.freeze_ucount;
1950}
1951
1952static inline bool may_freeze(struct super_block *sb, enum freeze_holder who)
1953{
1954	WARN_ON_ONCE((who & ~FREEZE_FLAGS));
1955	WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
1956
1957	if (who & FREEZE_HOLDER_KERNEL)
1958		return (who & FREEZE_MAY_NEST) ||
1959		       sb->s_writers.freeze_kcount == 0;
1960	if (who & FREEZE_HOLDER_USERSPACE)
1961		return (who & FREEZE_MAY_NEST) ||
1962		       sb->s_writers.freeze_ucount == 0;
1963	return false;
1964}
1965
1966/**
1967 * freeze_super - lock the filesystem and force it into a consistent state
1968 * @sb: the super to lock
1969 * @who: context that wants to freeze
1970 *
1971 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1972 * freeze_fs.  Subsequent calls to this without first thawing the fs may return
1973 * -EBUSY.
1974 *
1975 * @who should be:
1976 * * %FREEZE_HOLDER_USERSPACE if userspace wants to freeze the fs;
1977 * * %FREEZE_HOLDER_KERNEL if the kernel wants to freeze the fs.
1978 * * %FREEZE_MAY_NEST whether nesting freeze and thaw requests is allowed.
1979 *
1980 * The @who argument distinguishes between the kernel and userspace trying to
1981 * freeze the filesystem.  Although there cannot be multiple kernel freezes or
1982 * multiple userspace freezes in effect at any given time, the kernel and
1983 * userspace can both hold a filesystem frozen.  The filesystem remains frozen
1984 * until there are no kernel or userspace freezes in effect.
1985 *
1986 * A filesystem may hold multiple devices and thus a filesystems may be
1987 * frozen through the block layer via multiple block devices. In this
1988 * case the request is marked as being allowed to nest by passing
1989 * FREEZE_MAY_NEST. The filesystem remains frozen until all block
1990 * devices are unfrozen. If multiple freezes are attempted without
1991 * FREEZE_MAY_NEST -EBUSY will be returned.
1992 *
1993 * During this function, sb->s_writers.frozen goes through these values:
1994 *
1995 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1996 *
1997 * SB_FREEZE_WRITE: The file system is in the process of being frozen.  New
1998 * writes should be blocked, though page faults are still allowed. We wait for
1999 * all writes to complete and then proceed to the next stage.
2000 *
2001 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
2002 * but internal fs threads can still modify the filesystem (although they
2003 * should not dirty new pages or inodes), writeback can run etc. After waiting
2004 * for all running page faults we sync the filesystem which will clean all
2005 * dirty pages and inodes (no new dirty pages or inodes can be created when
2006 * sync is running).
2007 *
2008 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
2009 * modification are blocked (e.g. XFS preallocation truncation on inode
2010 * reclaim). This is usually implemented by blocking new transactions for
2011 * filesystems that have them and need this additional guard. After all
2012 * internal writers are finished we call ->freeze_fs() to finish filesystem
2013 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
2014 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
2015 *
2016 * sb->s_writers.frozen is protected by sb->s_umount.
2017 *
2018 * Return: If the freeze was successful zero is returned. If the freeze
2019 *         failed a negative error code is returned.
2020 */
2021int freeze_super(struct super_block *sb, enum freeze_holder who)
2022{
2023	int ret;
2024
2025	if (!super_lock_excl(sb)) {
2026		WARN_ON_ONCE("Dying superblock while freezing!");
2027		return -EINVAL;
2028	}
2029	atomic_inc(&sb->s_active);
2030
2031retry:
2032	if (sb->s_writers.frozen == SB_FREEZE_COMPLETE) {
2033		if (may_freeze(sb, who))
2034			ret = !!WARN_ON_ONCE(freeze_inc(sb, who) == 1);
2035		else
2036			ret = -EBUSY;
2037		/* All freezers share a single active reference. */
2038		deactivate_locked_super(sb);
2039		return ret;
2040	}
2041
2042	if (sb->s_writers.frozen != SB_UNFROZEN) {
2043		ret = wait_for_partially_frozen(sb);
2044		if (ret) {
2045			deactivate_locked_super(sb);
2046			return ret;
2047		}
2048
2049		goto retry;
2050	}
2051
2052	if (sb_rdonly(sb)) {
2053		/* Nothing to do really... */
2054		WARN_ON_ONCE(freeze_inc(sb, who) > 1);
2055		sb->s_writers.frozen = SB_FREEZE_COMPLETE;
2056		wake_up_var(&sb->s_writers.frozen);
2057		super_unlock_excl(sb);
2058		return 0;
2059	}
2060
2061	sb->s_writers.frozen = SB_FREEZE_WRITE;
2062	/* Release s_umount to preserve sb_start_write -> s_umount ordering */
2063	super_unlock_excl(sb);
2064	sb_wait_write(sb, SB_FREEZE_WRITE);
2065	__super_lock_excl(sb);
2066
2067	/* Now we go and block page faults... */
2068	sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
2069	sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
2070
2071	/* All writers are done so after syncing there won't be dirty data */
2072	ret = sync_filesystem(sb);
2073	if (ret) {
2074		sb->s_writers.frozen = SB_UNFROZEN;
2075		sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
2076		wake_up_var(&sb->s_writers.frozen);
2077		deactivate_locked_super(sb);
2078		return ret;
2079	}
2080
2081	/* Now wait for internal filesystem counter */
2082	sb->s_writers.frozen = SB_FREEZE_FS;
2083	sb_wait_write(sb, SB_FREEZE_FS);
2084
2085	if (sb->s_op->freeze_fs) {
2086		ret = sb->s_op->freeze_fs(sb);
2087		if (ret) {
2088			printk(KERN_ERR
2089				"VFS:Filesystem freeze failed\n");
2090			sb->s_writers.frozen = SB_UNFROZEN;
2091			sb_freeze_unlock(sb, SB_FREEZE_FS);
2092			wake_up_var(&sb->s_writers.frozen);
2093			deactivate_locked_super(sb);
2094			return ret;
2095		}
2096	}
2097	/*
2098	 * For debugging purposes so that fs can warn if it sees write activity
2099	 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
2100	 */
2101	WARN_ON_ONCE(freeze_inc(sb, who) > 1);
2102	sb->s_writers.frozen = SB_FREEZE_COMPLETE;
2103	wake_up_var(&sb->s_writers.frozen);
2104	lockdep_sb_freeze_release(sb);
2105	super_unlock_excl(sb);
2106	return 0;
2107}
2108EXPORT_SYMBOL(freeze_super);
2109
2110/*
2111 * Undoes the effect of a freeze_super_locked call.  If the filesystem is
2112 * frozen both by userspace and the kernel, a thaw call from either source
2113 * removes that state without releasing the other state or unlocking the
2114 * filesystem.
2115 */
2116static int thaw_super_locked(struct super_block *sb, enum freeze_holder who)
2117{
2118	int error = -EINVAL;
2119
2120	if (sb->s_writers.frozen != SB_FREEZE_COMPLETE)
2121		goto out_unlock;
2122
2123	/*
2124	 * All freezers share a single active reference.
2125	 * So just unlock in case there are any left.
2126	 */
2127	if (freeze_dec(sb, who))
2128		goto out_unlock;
2129
2130	if (sb_rdonly(sb)) {
2131		sb->s_writers.frozen = SB_UNFROZEN;
2132		wake_up_var(&sb->s_writers.frozen);
2133		goto out_deactivate;
2134	}
2135
2136	lockdep_sb_freeze_acquire(sb);
2137
2138	if (sb->s_op->unfreeze_fs) {
2139		error = sb->s_op->unfreeze_fs(sb);
2140		if (error) {
2141			pr_err("VFS: Filesystem thaw failed\n");
2142			freeze_inc(sb, who);
2143			lockdep_sb_freeze_release(sb);
2144			goto out_unlock;
 
2145		}
2146	}
2147
2148	sb->s_writers.frozen = SB_UNFROZEN;
2149	wake_up_var(&sb->s_writers.frozen);
2150	sb_freeze_unlock(sb, SB_FREEZE_FS);
2151out_deactivate:
2152	deactivate_locked_super(sb);
2153	return 0;
2154
2155out_unlock:
2156	super_unlock_excl(sb);
2157	return error;
2158}
2159
2160/**
2161 * thaw_super -- unlock filesystem
2162 * @sb: the super to thaw
2163 * @who: context that wants to freeze
2164 *
2165 * Unlocks the filesystem and marks it writeable again after freeze_super()
2166 * if there are no remaining freezes on the filesystem.
2167 *
2168 * @who should be:
2169 * * %FREEZE_HOLDER_USERSPACE if userspace wants to thaw the fs;
2170 * * %FREEZE_HOLDER_KERNEL if the kernel wants to thaw the fs.
2171 * * %FREEZE_MAY_NEST whether nesting freeze and thaw requests is allowed
2172 *
2173 * A filesystem may hold multiple devices and thus a filesystems may
2174 * have been frozen through the block layer via multiple block devices.
2175 * The filesystem remains frozen until all block devices are unfrozen.
2176 */
2177int thaw_super(struct super_block *sb, enum freeze_holder who)
2178{
2179	if (!super_lock_excl(sb)) {
2180		WARN_ON_ONCE("Dying superblock while thawing!");
2181		return -EINVAL;
2182	}
2183	return thaw_super_locked(sb, who);
2184}
2185EXPORT_SYMBOL(thaw_super);
2186
2187/*
2188 * Create workqueue for deferred direct IO completions. We allocate the
2189 * workqueue when it's first needed. This avoids creating workqueue for
2190 * filesystems that don't need it and also allows us to create the workqueue
2191 * late enough so the we can include s_id in the name of the workqueue.
2192 */
2193int sb_init_dio_done_wq(struct super_block *sb)
2194{
2195	struct workqueue_struct *old;
2196	struct workqueue_struct *wq = alloc_workqueue("dio/%s",
2197						      WQ_MEM_RECLAIM, 0,
2198						      sb->s_id);
2199	if (!wq)
2200		return -ENOMEM;
2201	/*
2202	 * This has to be atomic as more DIOs can race to create the workqueue
2203	 */
2204	old = cmpxchg(&sb->s_dio_done_wq, NULL, wq);
2205	/* Someone created workqueue before us? Free ours... */
2206	if (old)
2207		destroy_workqueue(wq);
2208	return 0;
2209}
2210EXPORT_SYMBOL_GPL(sb_init_dio_done_wq);