1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * (C) 1997 Linus Torvalds
   4 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
   5 */
   6#include <linux/export.h>
   7#include <linux/fs.h>
 
   8#include <linux/mm.h>
   9#include <linux/backing-dev.h>
  10#include <linux/hash.h>
  11#include <linux/swap.h>
  12#include <linux/security.h>
  13#include <linux/cdev.h>
  14#include <linux/memblock.h>
  15#include <linux/fsnotify.h>
  16#include <linux/mount.h>
  17#include <linux/posix_acl.h>
  18#include <linux/prefetch.h>
  19#include <linux/buffer_head.h> /* for inode_has_buffers */
  20#include <linux/ratelimit.h>
  21#include <linux/list_lru.h>
  22#include <linux/iversion.h>
  23#include <trace/events/writeback.h>
  24#include "internal.h"
  25
  26/*
  27 * Inode locking rules:
  28 *
  29 * inode->i_lock protects:
  30 *   inode->i_state, inode->i_hash, __iget(), inode->i_io_list
  31 * Inode LRU list locks protect:
  32 *   inode->i_sb->s_inode_lru, inode->i_lru
  33 * inode->i_sb->s_inode_list_lock protects:
  34 *   inode->i_sb->s_inodes, inode->i_sb_list
  35 * bdi->wb.list_lock protects:
  36 *   bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
  37 * inode_hash_lock protects:
  38 *   inode_hashtable, inode->i_hash
  39 *
  40 * Lock ordering:
  41 *
  42 * inode->i_sb->s_inode_list_lock
  43 *   inode->i_lock
  44 *     Inode LRU list locks
  45 *
  46 * bdi->wb.list_lock
  47 *   inode->i_lock
  48 *
  49 * inode_hash_lock
  50 *   inode->i_sb->s_inode_list_lock
  51 *   inode->i_lock
  52 *
  53 * iunique_lock
  54 *   inode_hash_lock
  55 */
  56
  57static unsigned int i_hash_mask __read_mostly;
  58static unsigned int i_hash_shift __read_mostly;
  59static struct hlist_head *inode_hashtable __read_mostly;
  60static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
  61
  62/*
  63 * Empty aops. Can be used for the cases where the user does not
  64 * define any of the address_space operations.
  65 */
  66const struct address_space_operations empty_aops = {
  67};
  68EXPORT_SYMBOL(empty_aops);
  69
  70static DEFINE_PER_CPU(unsigned long, nr_inodes);
  71static DEFINE_PER_CPU(unsigned long, nr_unused);
  72
  73static struct kmem_cache *inode_cachep __read_mostly;
  74
  75static long get_nr_inodes(void)
  76{
  77	int i;
  78	long sum = 0;
  79	for_each_possible_cpu(i)
  80		sum += per_cpu(nr_inodes, i);
  81	return sum < 0 ? 0 : sum;
  82}
  83
  84static inline long get_nr_inodes_unused(void)
  85{
  86	int i;
  87	long sum = 0;
  88	for_each_possible_cpu(i)
  89		sum += per_cpu(nr_unused, i);
  90	return sum < 0 ? 0 : sum;
  91}
  92
  93long get_nr_dirty_inodes(void)
  94{
  95	/* not actually dirty inodes, but a wild approximation */
  96	long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
  97	return nr_dirty > 0 ? nr_dirty : 0;
  98}
  99
 100/*
 101 * Handle nr_inode sysctl
 102 */
 103#ifdef CONFIG_SYSCTL
 104/*
 105 * Statistics gathering..
 106 */
 107static struct inodes_stat_t inodes_stat;
 108
 109static int proc_nr_inodes(struct ctl_table *table, int write, void *buffer,
 110			  size_t *lenp, loff_t *ppos)
 111{
 112	inodes_stat.nr_inodes = get_nr_inodes();
 113	inodes_stat.nr_unused = get_nr_inodes_unused();
 114	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 115}
 116
 117static struct ctl_table inodes_sysctls[] = {
 118	{
 119		.procname	= "inode-nr",
 120		.data		= &inodes_stat,
 121		.maxlen		= 2*sizeof(long),
 122		.mode		= 0444,
 123		.proc_handler	= proc_nr_inodes,
 124	},
 125	{
 126		.procname	= "inode-state",
 127		.data		= &inodes_stat,
 128		.maxlen		= 7*sizeof(long),
 129		.mode		= 0444,
 130		.proc_handler	= proc_nr_inodes,
 131	},
 132	{ }
 133};
 134
 135static int __init init_fs_inode_sysctls(void)
 136{
 137	register_sysctl_init("fs", inodes_sysctls);
 138	return 0;
 139}
 140early_initcall(init_fs_inode_sysctls);
 141#endif
 142
 143static int no_open(struct inode *inode, struct file *file)
 144{
 145	return -ENXIO;
 146}
 147
 148/**
 149 * inode_init_always - perform inode structure initialisation
 150 * @sb: superblock inode belongs to
 151 * @inode: inode to initialise
 152 *
 153 * These are initializations that need to be done on every inode
 154 * allocation as the fields are not initialised by slab allocation.
 155 */
 156int inode_init_always(struct super_block *sb, struct inode *inode)
 157{
 158	static const struct inode_operations empty_iops;
 159	static const struct file_operations no_open_fops = {.open = no_open};
 160	struct address_space *const mapping = &inode->i_data;
 161
 162	inode->i_sb = sb;
 163	inode->i_blkbits = sb->s_blocksize_bits;
 164	inode->i_flags = 0;
 165	atomic64_set(&inode->i_sequence, 0);
 166	atomic_set(&inode->i_count, 1);
 167	inode->i_op = &empty_iops;
 168	inode->i_fop = &no_open_fops;
 169	inode->i_ino = 0;
 170	inode->__i_nlink = 1;
 171	inode->i_opflags = 0;
 172	if (sb->s_xattr)
 173		inode->i_opflags |= IOP_XATTR;
 174	i_uid_write(inode, 0);
 175	i_gid_write(inode, 0);
 176	atomic_set(&inode->i_writecount, 0);
 177	inode->i_size = 0;
 178	inode->i_write_hint = WRITE_LIFE_NOT_SET;
 179	inode->i_blocks = 0;
 180	inode->i_bytes = 0;
 181	inode->i_generation = 0;
 182	inode->i_pipe = NULL;
 183	inode->i_cdev = NULL;
 184	inode->i_link = NULL;
 185	inode->i_dir_seq = 0;
 186	inode->i_rdev = 0;
 187	inode->dirtied_when = 0;
 188
 189#ifdef CONFIG_CGROUP_WRITEBACK
 190	inode->i_wb_frn_winner = 0;
 191	inode->i_wb_frn_avg_time = 0;
 192	inode->i_wb_frn_history = 0;
 193#endif
 194
 195	spin_lock_init(&inode->i_lock);
 196	lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
 197
 198	init_rwsem(&inode->i_rwsem);
 199	lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
 200
 201	atomic_set(&inode->i_dio_count, 0);
 202
 203	mapping->a_ops = &empty_aops;
 204	mapping->host = inode;
 205	mapping->flags = 0;
 206	mapping->wb_err = 0;
 207	atomic_set(&mapping->i_mmap_writable, 0);
 208#ifdef CONFIG_READ_ONLY_THP_FOR_FS
 209	atomic_set(&mapping->nr_thps, 0);
 210#endif
 211	mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
 212	mapping->private_data = NULL;
 213	mapping->writeback_index = 0;
 214	init_rwsem(&mapping->invalidate_lock);
 215	lockdep_set_class_and_name(&mapping->invalidate_lock,
 216				   &sb->s_type->invalidate_lock_key,
 217				   "mapping.invalidate_lock");
 
 
 218	inode->i_private = NULL;
 219	inode->i_mapping = mapping;
 220	INIT_HLIST_HEAD(&inode->i_dentry);	/* buggered by rcu freeing */
 221#ifdef CONFIG_FS_POSIX_ACL
 222	inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
 223#endif
 224
 225#ifdef CONFIG_FSNOTIFY
 226	inode->i_fsnotify_mask = 0;
 227#endif
 228	inode->i_flctx = NULL;
 229
 230	if (unlikely(security_inode_alloc(inode)))
 231		return -ENOMEM;
 232	this_cpu_inc(nr_inodes);
 233
 234	return 0;
 235}
 236EXPORT_SYMBOL(inode_init_always);
 237
 238void free_inode_nonrcu(struct inode *inode)
 239{
 240	kmem_cache_free(inode_cachep, inode);
 241}
 242EXPORT_SYMBOL(free_inode_nonrcu);
 243
 244static void i_callback(struct rcu_head *head)
 245{
 246	struct inode *inode = container_of(head, struct inode, i_rcu);
 247	if (inode->free_inode)
 248		inode->free_inode(inode);
 249	else
 250		free_inode_nonrcu(inode);
 251}
 252
 253static struct inode *alloc_inode(struct super_block *sb)
 254{
 255	const struct super_operations *ops = sb->s_op;
 256	struct inode *inode;
 257
 258	if (ops->alloc_inode)
 259		inode = ops->alloc_inode(sb);
 260	else
 261		inode = alloc_inode_sb(sb, inode_cachep, GFP_KERNEL);
 262
 263	if (!inode)
 264		return NULL;
 265
 266	if (unlikely(inode_init_always(sb, inode))) {
 267		if (ops->destroy_inode) {
 268			ops->destroy_inode(inode);
 269			if (!ops->free_inode)
 270				return NULL;
 271		}
 272		inode->free_inode = ops->free_inode;
 273		i_callback(&inode->i_rcu);
 274		return NULL;
 275	}
 276
 277	return inode;
 278}
 279
 280void __destroy_inode(struct inode *inode)
 281{
 282	BUG_ON(inode_has_buffers(inode));
 283	inode_detach_wb(inode);
 284	security_inode_free(inode);
 285	fsnotify_inode_delete(inode);
 286	locks_free_lock_context(inode);
 287	if (!inode->i_nlink) {
 288		WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
 289		atomic_long_dec(&inode->i_sb->s_remove_count);
 290	}
 291
 292#ifdef CONFIG_FS_POSIX_ACL
 293	if (inode->i_acl && !is_uncached_acl(inode->i_acl))
 294		posix_acl_release(inode->i_acl);
 295	if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
 296		posix_acl_release(inode->i_default_acl);
 297#endif
 298	this_cpu_dec(nr_inodes);
 299}
 300EXPORT_SYMBOL(__destroy_inode);
 301
 302static void destroy_inode(struct inode *inode)
 303{
 304	const struct super_operations *ops = inode->i_sb->s_op;
 305
 306	BUG_ON(!list_empty(&inode->i_lru));
 307	__destroy_inode(inode);
 308	if (ops->destroy_inode) {
 309		ops->destroy_inode(inode);
 310		if (!ops->free_inode)
 311			return;
 312	}
 313	inode->free_inode = ops->free_inode;
 314	call_rcu(&inode->i_rcu, i_callback);
 315}
 316
 317/**
 318 * drop_nlink - directly drop an inode's link count
 319 * @inode: inode
 320 *
 321 * This is a low-level filesystem helper to replace any
 322 * direct filesystem manipulation of i_nlink.  In cases
 323 * where we are attempting to track writes to the
 324 * filesystem, a decrement to zero means an imminent
 325 * write when the file is truncated and actually unlinked
 326 * on the filesystem.
 327 */
 328void drop_nlink(struct inode *inode)
 329{
 330	WARN_ON(inode->i_nlink == 0);
 331	inode->__i_nlink--;
 332	if (!inode->i_nlink)
 333		atomic_long_inc(&inode->i_sb->s_remove_count);
 334}
 335EXPORT_SYMBOL(drop_nlink);
 336
 337/**
 338 * clear_nlink - directly zero an inode's link count
 339 * @inode: inode
 340 *
 341 * This is a low-level filesystem helper to replace any
 342 * direct filesystem manipulation of i_nlink.  See
 343 * drop_nlink() for why we care about i_nlink hitting zero.
 344 */
 345void clear_nlink(struct inode *inode)
 346{
 347	if (inode->i_nlink) {
 348		inode->__i_nlink = 0;
 349		atomic_long_inc(&inode->i_sb->s_remove_count);
 350	}
 351}
 352EXPORT_SYMBOL(clear_nlink);
 353
 354/**
 355 * set_nlink - directly set an inode's link count
 356 * @inode: inode
 357 * @nlink: new nlink (should be non-zero)
 358 *
 359 * This is a low-level filesystem helper to replace any
 360 * direct filesystem manipulation of i_nlink.
 361 */
 362void set_nlink(struct inode *inode, unsigned int nlink)
 363{
 364	if (!nlink) {
 365		clear_nlink(inode);
 366	} else {
 367		/* Yes, some filesystems do change nlink from zero to one */
 368		if (inode->i_nlink == 0)
 369			atomic_long_dec(&inode->i_sb->s_remove_count);
 370
 371		inode->__i_nlink = nlink;
 372	}
 373}
 374EXPORT_SYMBOL(set_nlink);
 375
 376/**
 377 * inc_nlink - directly increment an inode's link count
 378 * @inode: inode
 379 *
 380 * This is a low-level filesystem helper to replace any
 381 * direct filesystem manipulation of i_nlink.  Currently,
 382 * it is only here for parity with dec_nlink().
 383 */
 384void inc_nlink(struct inode *inode)
 385{
 386	if (unlikely(inode->i_nlink == 0)) {
 387		WARN_ON(!(inode->i_state & I_LINKABLE));
 388		atomic_long_dec(&inode->i_sb->s_remove_count);
 389	}
 390
 391	inode->__i_nlink++;
 392}
 393EXPORT_SYMBOL(inc_nlink);
 394
 395static void __address_space_init_once(struct address_space *mapping)
 396{
 397	xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
 398	init_rwsem(&mapping->i_mmap_rwsem);
 399	INIT_LIST_HEAD(&mapping->private_list);
 400	spin_lock_init(&mapping->private_lock);
 401	mapping->i_mmap = RB_ROOT_CACHED;
 402}
 403
 404void address_space_init_once(struct address_space *mapping)
 405{
 406	memset(mapping, 0, sizeof(*mapping));
 407	__address_space_init_once(mapping);
 408}
 409EXPORT_SYMBOL(address_space_init_once);
 410
 411/*
 412 * These are initializations that only need to be done
 413 * once, because the fields are idempotent across use
 414 * of the inode, so let the slab aware of that.
 415 */
 416void inode_init_once(struct inode *inode)
 417{
 418	memset(inode, 0, sizeof(*inode));
 419	INIT_HLIST_NODE(&inode->i_hash);
 420	INIT_LIST_HEAD(&inode->i_devices);
 421	INIT_LIST_HEAD(&inode->i_io_list);
 422	INIT_LIST_HEAD(&inode->i_wb_list);
 423	INIT_LIST_HEAD(&inode->i_lru);
 424	INIT_LIST_HEAD(&inode->i_sb_list);
 425	__address_space_init_once(&inode->i_data);
 426	i_size_ordered_init(inode);
 427}
 428EXPORT_SYMBOL(inode_init_once);
 429
 430static void init_once(void *foo)
 431{
 432	struct inode *inode = (struct inode *) foo;
 433
 434	inode_init_once(inode);
 435}
 436
 437/*
 438 * inode->i_lock must be held
 439 */
 440void __iget(struct inode *inode)
 441{
 442	atomic_inc(&inode->i_count);
 443}
 444
 445/*
 446 * get additional reference to inode; caller must already hold one.
 447 */
 448void ihold(struct inode *inode)
 449{
 450	WARN_ON(atomic_inc_return(&inode->i_count) < 2);
 451}
 452EXPORT_SYMBOL(ihold);
 453
 454static void __inode_add_lru(struct inode *inode, bool rotate)
 455{
 456	if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
 457		return;
 458	if (atomic_read(&inode->i_count))
 459		return;
 460	if (!(inode->i_sb->s_flags & SB_ACTIVE))
 461		return;
 462	if (!mapping_shrinkable(&inode->i_data))
 463		return;
 464
 465	if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
 466		this_cpu_inc(nr_unused);
 467	else if (rotate)
 468		inode->i_state |= I_REFERENCED;
 469}
 470
 471/*
 472 * Add inode to LRU if needed (inode is unused and clean).
 473 *
 474 * Needs inode->i_lock held.
 475 */
 476void inode_add_lru(struct inode *inode)
 477{
 478	__inode_add_lru(inode, false);
 479}
 480
 481static void inode_lru_list_del(struct inode *inode)
 482{
 483	if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
 484		this_cpu_dec(nr_unused);
 485}
 486
 487/**
 488 * inode_sb_list_add - add inode to the superblock list of inodes
 489 * @inode: inode to add
 490 */
 491void inode_sb_list_add(struct inode *inode)
 492{
 493	spin_lock(&inode->i_sb->s_inode_list_lock);
 494	list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
 495	spin_unlock(&inode->i_sb->s_inode_list_lock);
 496}
 497EXPORT_SYMBOL_GPL(inode_sb_list_add);
 498
 499static inline void inode_sb_list_del(struct inode *inode)
 500{
 501	if (!list_empty(&inode->i_sb_list)) {
 502		spin_lock(&inode->i_sb->s_inode_list_lock);
 503		list_del_init(&inode->i_sb_list);
 504		spin_unlock(&inode->i_sb->s_inode_list_lock);
 505	}
 506}
 507
 508static unsigned long hash(struct super_block *sb, unsigned long hashval)
 509{
 510	unsigned long tmp;
 511
 512	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
 513			L1_CACHE_BYTES;
 514	tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
 515	return tmp & i_hash_mask;
 516}
 517
 518/**
 519 *	__insert_inode_hash - hash an inode
 520 *	@inode: unhashed inode
 521 *	@hashval: unsigned long value used to locate this object in the
 522 *		inode_hashtable.
 523 *
 524 *	Add an inode to the inode hash for this superblock.
 525 */
 526void __insert_inode_hash(struct inode *inode, unsigned long hashval)
 527{
 528	struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
 529
 530	spin_lock(&inode_hash_lock);
 531	spin_lock(&inode->i_lock);
 532	hlist_add_head_rcu(&inode->i_hash, b);
 533	spin_unlock(&inode->i_lock);
 534	spin_unlock(&inode_hash_lock);
 535}
 536EXPORT_SYMBOL(__insert_inode_hash);
 537
 538/**
 539 *	__remove_inode_hash - remove an inode from the hash
 540 *	@inode: inode to unhash
 541 *
 542 *	Remove an inode from the superblock.
 543 */
 544void __remove_inode_hash(struct inode *inode)
 545{
 546	spin_lock(&inode_hash_lock);
 547	spin_lock(&inode->i_lock);
 548	hlist_del_init_rcu(&inode->i_hash);
 549	spin_unlock(&inode->i_lock);
 550	spin_unlock(&inode_hash_lock);
 551}
 552EXPORT_SYMBOL(__remove_inode_hash);
 553
 554void dump_mapping(const struct address_space *mapping)
 555{
 556	struct inode *host;
 557	const struct address_space_operations *a_ops;
 558	struct hlist_node *dentry_first;
 559	struct dentry *dentry_ptr;
 560	struct dentry dentry;
 561	unsigned long ino;
 562
 563	/*
 564	 * If mapping is an invalid pointer, we don't want to crash
 565	 * accessing it, so probe everything depending on it carefully.
 566	 */
 567	if (get_kernel_nofault(host, &mapping->host) ||
 568	    get_kernel_nofault(a_ops, &mapping->a_ops)) {
 569		pr_warn("invalid mapping:%px\n", mapping);
 570		return;
 571	}
 572
 573	if (!host) {
 574		pr_warn("aops:%ps\n", a_ops);
 575		return;
 576	}
 577
 578	if (get_kernel_nofault(dentry_first, &host->i_dentry.first) ||
 579	    get_kernel_nofault(ino, &host->i_ino)) {
 580		pr_warn("aops:%ps invalid inode:%px\n", a_ops, host);
 581		return;
 582	}
 583
 584	if (!dentry_first) {
 585		pr_warn("aops:%ps ino:%lx\n", a_ops, ino);
 586		return;
 587	}
 588
 589	dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
 590	if (get_kernel_nofault(dentry, dentry_ptr)) {
 591		pr_warn("aops:%ps ino:%lx invalid dentry:%px\n",
 592				a_ops, ino, dentry_ptr);
 593		return;
 594	}
 595
 596	/*
 597	 * if dentry is corrupted, the %pd handler may still crash,
 598	 * but it's unlikely that we reach here with a corrupt mapping
 599	 */
 600	pr_warn("aops:%ps ino:%lx dentry name:\"%pd\"\n", a_ops, ino, &dentry);
 601}
 602
 603void clear_inode(struct inode *inode)
 604{
 605	/*
 606	 * We have to cycle the i_pages lock here because reclaim can be in the
 607	 * process of removing the last page (in __filemap_remove_folio())
 608	 * and we must not free the mapping under it.
 609	 */
 610	xa_lock_irq(&inode->i_data.i_pages);
 611	BUG_ON(inode->i_data.nrpages);
 612	/*
 613	 * Almost always, mapping_empty(&inode->i_data) here; but there are
 614	 * two known and long-standing ways in which nodes may get left behind
 615	 * (when deep radix-tree node allocation failed partway; or when THP
 616	 * collapse_file() failed). Until those two known cases are cleaned up,
 617	 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
 618	 * nor even WARN_ON(!mapping_empty).
 619	 */
 620	xa_unlock_irq(&inode->i_data.i_pages);
 621	BUG_ON(!list_empty(&inode->i_data.private_list));
 622	BUG_ON(!(inode->i_state & I_FREEING));
 623	BUG_ON(inode->i_state & I_CLEAR);
 624	BUG_ON(!list_empty(&inode->i_wb_list));
 625	/* don't need i_lock here, no concurrent mods to i_state */
 626	inode->i_state = I_FREEING | I_CLEAR;
 627}
 628EXPORT_SYMBOL(clear_inode);
 629
 630/*
 631 * Free the inode passed in, removing it from the lists it is still connected
 632 * to. We remove any pages still attached to the inode and wait for any IO that
 633 * is still in progress before finally destroying the inode.
 634 *
 635 * An inode must already be marked I_FREEING so that we avoid the inode being
 636 * moved back onto lists if we race with other code that manipulates the lists
 637 * (e.g. writeback_single_inode). The caller is responsible for setting this.
 638 *
 639 * An inode must already be removed from the LRU list before being evicted from
 640 * the cache. This should occur atomically with setting the I_FREEING state
 641 * flag, so no inodes here should ever be on the LRU when being evicted.
 642 */
 643static void evict(struct inode *inode)
 644{
 645	const struct super_operations *op = inode->i_sb->s_op;
 646
 647	BUG_ON(!(inode->i_state & I_FREEING));
 648	BUG_ON(!list_empty(&inode->i_lru));
 649
 650	if (!list_empty(&inode->i_io_list))
 651		inode_io_list_del(inode);
 652
 653	inode_sb_list_del(inode);
 654
 655	/*
 656	 * Wait for flusher thread to be done with the inode so that filesystem
 657	 * does not start destroying it while writeback is still running. Since
 658	 * the inode has I_FREEING set, flusher thread won't start new work on
 659	 * the inode.  We just have to wait for running writeback to finish.
 660	 */
 661	inode_wait_for_writeback(inode);
 662
 663	if (op->evict_inode) {
 664		op->evict_inode(inode);
 665	} else {
 666		truncate_inode_pages_final(&inode->i_data);
 667		clear_inode(inode);
 668	}
 669	if (S_ISCHR(inode->i_mode) && inode->i_cdev)
 670		cd_forget(inode);
 671
 672	remove_inode_hash(inode);
 673
 674	spin_lock(&inode->i_lock);
 675	wake_up_bit(&inode->i_state, __I_NEW);
 676	BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
 677	spin_unlock(&inode->i_lock);
 678
 679	destroy_inode(inode);
 680}
 681
 682/*
 683 * dispose_list - dispose of the contents of a local list
 684 * @head: the head of the list to free
 685 *
 686 * Dispose-list gets a local list with local inodes in it, so it doesn't
 687 * need to worry about list corruption and SMP locks.
 688 */
 689static void dispose_list(struct list_head *head)
 690{
 691	while (!list_empty(head)) {
 692		struct inode *inode;
 693
 694		inode = list_first_entry(head, struct inode, i_lru);
 695		list_del_init(&inode->i_lru);
 696
 697		evict(inode);
 698		cond_resched();
 699	}
 700}
 701
 702/**
 703 * evict_inodes	- evict all evictable inodes for a superblock
 704 * @sb:		superblock to operate on
 705 *
 706 * Make sure that no inodes with zero refcount are retained.  This is
 707 * called by superblock shutdown after having SB_ACTIVE flag removed,
 708 * so any inode reaching zero refcount during or after that call will
 709 * be immediately evicted.
 710 */
 711void evict_inodes(struct super_block *sb)
 712{
 713	struct inode *inode, *next;
 714	LIST_HEAD(dispose);
 715
 716again:
 717	spin_lock(&sb->s_inode_list_lock);
 718	list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 719		if (atomic_read(&inode->i_count))
 720			continue;
 721
 722		spin_lock(&inode->i_lock);
 723		if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 724			spin_unlock(&inode->i_lock);
 725			continue;
 726		}
 727
 728		inode->i_state |= I_FREEING;
 729		inode_lru_list_del(inode);
 730		spin_unlock(&inode->i_lock);
 731		list_add(&inode->i_lru, &dispose);
 732
 733		/*
 734		 * We can have a ton of inodes to evict at unmount time given
 735		 * enough memory, check to see if we need to go to sleep for a
 736		 * bit so we don't livelock.
 737		 */
 738		if (need_resched()) {
 739			spin_unlock(&sb->s_inode_list_lock);
 740			cond_resched();
 741			dispose_list(&dispose);
 742			goto again;
 743		}
 744	}
 745	spin_unlock(&sb->s_inode_list_lock);
 746
 747	dispose_list(&dispose);
 748}
 749EXPORT_SYMBOL_GPL(evict_inodes);
 750
 751/**
 752 * invalidate_inodes	- attempt to free all inodes on a superblock
 753 * @sb:		superblock to operate on
 754 * @kill_dirty: flag to guide handling of dirty inodes
 755 *
 756 * Attempts to free all inodes for a given superblock.  If there were any
 757 * busy inodes return a non-zero value, else zero.
 758 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
 759 * them as busy.
 760 */
 761int invalidate_inodes(struct super_block *sb, bool kill_dirty)
 762{
 763	int busy = 0;
 764	struct inode *inode, *next;
 765	LIST_HEAD(dispose);
 766
 767again:
 768	spin_lock(&sb->s_inode_list_lock);
 769	list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 770		spin_lock(&inode->i_lock);
 771		if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 772			spin_unlock(&inode->i_lock);
 773			continue;
 774		}
 775		if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
 776			spin_unlock(&inode->i_lock);
 777			busy = 1;
 778			continue;
 779		}
 780		if (atomic_read(&inode->i_count)) {
 781			spin_unlock(&inode->i_lock);
 782			busy = 1;
 783			continue;
 784		}
 785
 786		inode->i_state |= I_FREEING;
 787		inode_lru_list_del(inode);
 788		spin_unlock(&inode->i_lock);
 789		list_add(&inode->i_lru, &dispose);
 790		if (need_resched()) {
 791			spin_unlock(&sb->s_inode_list_lock);
 792			cond_resched();
 793			dispose_list(&dispose);
 794			goto again;
 795		}
 796	}
 797	spin_unlock(&sb->s_inode_list_lock);
 798
 799	dispose_list(&dispose);
 800
 801	return busy;
 802}
 803
 804/*
 805 * Isolate the inode from the LRU in preparation for freeing it.
 806 *
 807 * If the inode has the I_REFERENCED flag set, then it means that it has been
 808 * used recently - the flag is set in iput_final(). When we encounter such an
 809 * inode, clear the flag and move it to the back of the LRU so it gets another
 810 * pass through the LRU before it gets reclaimed. This is necessary because of
 811 * the fact we are doing lazy LRU updates to minimise lock contention so the
 812 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
 813 * with this flag set because they are the inodes that are out of order.
 814 */
 815static enum lru_status inode_lru_isolate(struct list_head *item,
 816		struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
 817{
 818	struct list_head *freeable = arg;
 819	struct inode	*inode = container_of(item, struct inode, i_lru);
 820
 821	/*
 822	 * We are inverting the lru lock/inode->i_lock here, so use a
 823	 * trylock. If we fail to get the lock, just skip it.
 824	 */
 825	if (!spin_trylock(&inode->i_lock))
 826		return LRU_SKIP;
 827
 828	/*
 829	 * Inodes can get referenced, redirtied, or repopulated while
 830	 * they're already on the LRU, and this can make them
 831	 * unreclaimable for a while. Remove them lazily here; iput,
 832	 * sync, or the last page cache deletion will requeue them.
 833	 */
 834	if (atomic_read(&inode->i_count) ||
 835	    (inode->i_state & ~I_REFERENCED) ||
 836	    !mapping_shrinkable(&inode->i_data)) {
 837		list_lru_isolate(lru, &inode->i_lru);
 838		spin_unlock(&inode->i_lock);
 839		this_cpu_dec(nr_unused);
 840		return LRU_REMOVED;
 841	}
 842
 843	/* Recently referenced inodes get one more pass */
 844	if (inode->i_state & I_REFERENCED) {
 845		inode->i_state &= ~I_REFERENCED;
 846		spin_unlock(&inode->i_lock);
 847		return LRU_ROTATE;
 848	}
 849
 850	/*
 851	 * On highmem systems, mapping_shrinkable() permits dropping
 852	 * page cache in order to free up struct inodes: lowmem might
 853	 * be under pressure before the cache inside the highmem zone.
 854	 */
 855	if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
 856		__iget(inode);
 857		spin_unlock(&inode->i_lock);
 858		spin_unlock(lru_lock);
 859		if (remove_inode_buffers(inode)) {
 860			unsigned long reap;
 861			reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
 862			if (current_is_kswapd())
 863				__count_vm_events(KSWAPD_INODESTEAL, reap);
 864			else
 865				__count_vm_events(PGINODESTEAL, reap);
 866			if (current->reclaim_state)
 867				current->reclaim_state->reclaimed_slab += reap;
 868		}
 869		iput(inode);
 870		spin_lock(lru_lock);
 871		return LRU_RETRY;
 872	}
 873
 874	WARN_ON(inode->i_state & I_NEW);
 875	inode->i_state |= I_FREEING;
 876	list_lru_isolate_move(lru, &inode->i_lru, freeable);
 877	spin_unlock(&inode->i_lock);
 878
 879	this_cpu_dec(nr_unused);
 880	return LRU_REMOVED;
 881}
 882
 883/*
 884 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
 885 * This is called from the superblock shrinker function with a number of inodes
 886 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
 887 * then are freed outside inode_lock by dispose_list().
 888 */
 889long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
 890{
 891	LIST_HEAD(freeable);
 892	long freed;
 893
 894	freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
 895				     inode_lru_isolate, &freeable);
 896	dispose_list(&freeable);
 897	return freed;
 898}
 899
 900static void __wait_on_freeing_inode(struct inode *inode);
 901/*
 902 * Called with the inode lock held.
 903 */
 904static struct inode *find_inode(struct super_block *sb,
 905				struct hlist_head *head,
 906				int (*test)(struct inode *, void *),
 907				void *data)
 908{
 909	struct inode *inode = NULL;
 910
 911repeat:
 912	hlist_for_each_entry(inode, head, i_hash) {
 913		if (inode->i_sb != sb)
 914			continue;
 915		if (!test(inode, data))
 916			continue;
 917		spin_lock(&inode->i_lock);
 918		if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 919			__wait_on_freeing_inode(inode);
 920			goto repeat;
 921		}
 922		if (unlikely(inode->i_state & I_CREATING)) {
 923			spin_unlock(&inode->i_lock);
 924			return ERR_PTR(-ESTALE);
 925		}
 926		__iget(inode);
 927		spin_unlock(&inode->i_lock);
 928		return inode;
 929	}
 930	return NULL;
 931}
 932
 933/*
 934 * find_inode_fast is the fast path version of find_inode, see the comment at
 935 * iget_locked for details.
 936 */
 937static struct inode *find_inode_fast(struct super_block *sb,
 938				struct hlist_head *head, unsigned long ino)
 939{
 940	struct inode *inode = NULL;
 941
 942repeat:
 943	hlist_for_each_entry(inode, head, i_hash) {
 944		if (inode->i_ino != ino)
 945			continue;
 946		if (inode->i_sb != sb)
 947			continue;
 948		spin_lock(&inode->i_lock);
 949		if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 950			__wait_on_freeing_inode(inode);
 951			goto repeat;
 952		}
 953		if (unlikely(inode->i_state & I_CREATING)) {
 954			spin_unlock(&inode->i_lock);
 955			return ERR_PTR(-ESTALE);
 956		}
 957		__iget(inode);
 958		spin_unlock(&inode->i_lock);
 959		return inode;
 960	}
 961	return NULL;
 962}
 963
 964/*
 965 * Each cpu owns a range of LAST_INO_BATCH numbers.
 966 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
 967 * to renew the exhausted range.
 968 *
 969 * This does not significantly increase overflow rate because every CPU can
 970 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
 971 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
 972 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
 973 * overflow rate by 2x, which does not seem too significant.
 974 *
 975 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
 976 * error if st_ino won't fit in target struct field. Use 32bit counter
 977 * here to attempt to avoid that.
 978 */
 979#define LAST_INO_BATCH 1024
 980static DEFINE_PER_CPU(unsigned int, last_ino);
 981
 982unsigned int get_next_ino(void)
 983{
 984	unsigned int *p = &get_cpu_var(last_ino);
 985	unsigned int res = *p;
 986
 987#ifdef CONFIG_SMP
 988	if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
 989		static atomic_t shared_last_ino;
 990		int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
 991
 992		res = next - LAST_INO_BATCH;
 993	}
 994#endif
 995
 996	res++;
 997	/* get_next_ino should not provide a 0 inode number */
 998	if (unlikely(!res))
 999		res++;
1000	*p = res;
1001	put_cpu_var(last_ino);
1002	return res;
1003}
1004EXPORT_SYMBOL(get_next_ino);
1005
1006/**
1007 *	new_inode_pseudo 	- obtain an inode
1008 *	@sb: superblock
1009 *
1010 *	Allocates a new inode for given superblock.
1011 *	Inode wont be chained in superblock s_inodes list
1012 *	This means :
1013 *	- fs can't be unmount
1014 *	- quotas, fsnotify, writeback can't work
1015 */
1016struct inode *new_inode_pseudo(struct super_block *sb)
1017{
1018	struct inode *inode = alloc_inode(sb);
1019
1020	if (inode) {
1021		spin_lock(&inode->i_lock);
1022		inode->i_state = 0;
1023		spin_unlock(&inode->i_lock);
1024	}
1025	return inode;
1026}
1027
1028/**
1029 *	new_inode 	- obtain an inode
1030 *	@sb: superblock
1031 *
1032 *	Allocates a new inode for given superblock. The default gfp_mask
1033 *	for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1034 *	If HIGHMEM pages are unsuitable or it is known that pages allocated
1035 *	for the page cache are not reclaimable or migratable,
1036 *	mapping_set_gfp_mask() must be called with suitable flags on the
1037 *	newly created inode's mapping
1038 *
1039 */
1040struct inode *new_inode(struct super_block *sb)
1041{
1042	struct inode *inode;
1043
1044	spin_lock_prefetch(&sb->s_inode_list_lock);
1045
1046	inode = new_inode_pseudo(sb);
1047	if (inode)
1048		inode_sb_list_add(inode);
1049	return inode;
1050}
1051EXPORT_SYMBOL(new_inode);
1052
1053#ifdef CONFIG_DEBUG_LOCK_ALLOC
1054void lockdep_annotate_inode_mutex_key(struct inode *inode)
1055{
1056	if (S_ISDIR(inode->i_mode)) {
1057		struct file_system_type *type = inode->i_sb->s_type;
1058
1059		/* Set new key only if filesystem hasn't already changed it */
1060		if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
1061			/*
1062			 * ensure nobody is actually holding i_mutex
1063			 */
1064			// mutex_destroy(&inode->i_mutex);
1065			init_rwsem(&inode->i_rwsem);
1066			lockdep_set_class(&inode->i_rwsem,
1067					  &type->i_mutex_dir_key);
1068		}
1069	}
1070}
1071EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1072#endif
1073
1074/**
1075 * unlock_new_inode - clear the I_NEW state and wake up any waiters
1076 * @inode:	new inode to unlock
1077 *
1078 * Called when the inode is fully initialised to clear the new state of the
1079 * inode and wake up anyone waiting for the inode to finish initialisation.
1080 */
1081void unlock_new_inode(struct inode *inode)
1082{
1083	lockdep_annotate_inode_mutex_key(inode);
1084	spin_lock(&inode->i_lock);
1085	WARN_ON(!(inode->i_state & I_NEW));
1086	inode->i_state &= ~I_NEW & ~I_CREATING;
1087	smp_mb();
1088	wake_up_bit(&inode->i_state, __I_NEW);
1089	spin_unlock(&inode->i_lock);
1090}
1091EXPORT_SYMBOL(unlock_new_inode);
1092
1093void discard_new_inode(struct inode *inode)
1094{
1095	lockdep_annotate_inode_mutex_key(inode);
1096	spin_lock(&inode->i_lock);
1097	WARN_ON(!(inode->i_state & I_NEW));
1098	inode->i_state &= ~I_NEW;
1099	smp_mb();
1100	wake_up_bit(&inode->i_state, __I_NEW);
1101	spin_unlock(&inode->i_lock);
1102	iput(inode);
1103}
1104EXPORT_SYMBOL(discard_new_inode);
1105
1106/**
1107 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1108 *
1109 * Lock any non-NULL argument that is not a directory.
1110 * Zero, one or two objects may be locked by this function.
1111 *
1112 * @inode1: first inode to lock
1113 * @inode2: second inode to lock
1114 */
1115void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1116{
 
 
 
 
1117	if (inode1 > inode2)
1118		swap(inode1, inode2);
1119
1120	if (inode1 && !S_ISDIR(inode1->i_mode))
1121		inode_lock(inode1);
1122	if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1123		inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1124}
1125EXPORT_SYMBOL(lock_two_nondirectories);
1126
1127/**
1128 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1129 * @inode1: first inode to unlock
1130 * @inode2: second inode to unlock
1131 */
1132void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1133{
1134	if (inode1 && !S_ISDIR(inode1->i_mode))
 
1135		inode_unlock(inode1);
1136	if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
 
 
1137		inode_unlock(inode2);
 
1138}
1139EXPORT_SYMBOL(unlock_two_nondirectories);
1140
1141/**
1142 * inode_insert5 - obtain an inode from a mounted file system
1143 * @inode:	pre-allocated inode to use for insert to cache
1144 * @hashval:	hash value (usually inode number) to get
1145 * @test:	callback used for comparisons between inodes
1146 * @set:	callback used to initialize a new struct inode
1147 * @data:	opaque data pointer to pass to @test and @set
1148 *
1149 * Search for the inode specified by @hashval and @data in the inode cache,
1150 * and if present it is return it with an increased reference count. This is
1151 * a variant of iget5_locked() for callers that don't want to fail on memory
1152 * allocation of inode.
1153 *
1154 * If the inode is not in cache, insert the pre-allocated inode to cache and
1155 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1156 * to fill it in before unlocking it via unlock_new_inode().
1157 *
1158 * Note both @test and @set are called with the inode_hash_lock held, so can't
1159 * sleep.
1160 */
1161struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1162			    int (*test)(struct inode *, void *),
1163			    int (*set)(struct inode *, void *), void *data)
1164{
1165	struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1166	struct inode *old;
1167
1168again:
1169	spin_lock(&inode_hash_lock);
1170	old = find_inode(inode->i_sb, head, test, data);
1171	if (unlikely(old)) {
1172		/*
1173		 * Uhhuh, somebody else created the same inode under us.
1174		 * Use the old inode instead of the preallocated one.
1175		 */
1176		spin_unlock(&inode_hash_lock);
1177		if (IS_ERR(old))
1178			return NULL;
1179		wait_on_inode(old);
1180		if (unlikely(inode_unhashed(old))) {
1181			iput(old);
1182			goto again;
1183		}
1184		return old;
1185	}
1186
1187	if (set && unlikely(set(inode, data))) {
1188		inode = NULL;
1189		goto unlock;
1190	}
1191
1192	/*
1193	 * Return the locked inode with I_NEW set, the
1194	 * caller is responsible for filling in the contents
1195	 */
1196	spin_lock(&inode->i_lock);
1197	inode->i_state |= I_NEW;
1198	hlist_add_head_rcu(&inode->i_hash, head);
1199	spin_unlock(&inode->i_lock);
1200
1201	/*
1202	 * Add inode to the sb list if it's not already. It has I_NEW at this
1203	 * point, so it should be safe to test i_sb_list locklessly.
1204	 */
1205	if (list_empty(&inode->i_sb_list))
1206		inode_sb_list_add(inode);
1207unlock:
1208	spin_unlock(&inode_hash_lock);
1209
1210	return inode;
1211}
1212EXPORT_SYMBOL(inode_insert5);
1213
1214/**
1215 * iget5_locked - obtain an inode from a mounted file system
1216 * @sb:		super block of file system
1217 * @hashval:	hash value (usually inode number) to get
1218 * @test:	callback used for comparisons between inodes
1219 * @set:	callback used to initialize a new struct inode
1220 * @data:	opaque data pointer to pass to @test and @set
1221 *
1222 * Search for the inode specified by @hashval and @data in the inode cache,
1223 * and if present it is return it with an increased reference count. This is
1224 * a generalized version of iget_locked() for file systems where the inode
1225 * number is not sufficient for unique identification of an inode.
1226 *
1227 * If the inode is not in cache, allocate a new inode and return it locked,
1228 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1229 * before unlocking it via unlock_new_inode().
1230 *
1231 * Note both @test and @set are called with the inode_hash_lock held, so can't
1232 * sleep.
1233 */
1234struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1235		int (*test)(struct inode *, void *),
1236		int (*set)(struct inode *, void *), void *data)
1237{
1238	struct inode *inode = ilookup5(sb, hashval, test, data);
1239
1240	if (!inode) {
1241		struct inode *new = alloc_inode(sb);
1242
1243		if (new) {
1244			new->i_state = 0;
1245			inode = inode_insert5(new, hashval, test, set, data);
1246			if (unlikely(inode != new))
1247				destroy_inode(new);
1248		}
1249	}
1250	return inode;
1251}
1252EXPORT_SYMBOL(iget5_locked);
1253
1254/**
1255 * iget_locked - obtain an inode from a mounted file system
1256 * @sb:		super block of file system
1257 * @ino:	inode number to get
1258 *
1259 * Search for the inode specified by @ino in the inode cache and if present
1260 * return it with an increased reference count. This is for file systems
1261 * where the inode number is sufficient for unique identification of an inode.
1262 *
1263 * If the inode is not in cache, allocate a new inode and return it locked,
1264 * hashed, and with the I_NEW flag set.  The file system gets to fill it in
1265 * before unlocking it via unlock_new_inode().
1266 */
1267struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1268{
1269	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1270	struct inode *inode;
1271again:
1272	spin_lock(&inode_hash_lock);
1273	inode = find_inode_fast(sb, head, ino);
1274	spin_unlock(&inode_hash_lock);
1275	if (inode) {
1276		if (IS_ERR(inode))
1277			return NULL;
1278		wait_on_inode(inode);
1279		if (unlikely(inode_unhashed(inode))) {
1280			iput(inode);
1281			goto again;
1282		}
1283		return inode;
1284	}
1285
1286	inode = alloc_inode(sb);
1287	if (inode) {
1288		struct inode *old;
1289
1290		spin_lock(&inode_hash_lock);
1291		/* We released the lock, so.. */
1292		old = find_inode_fast(sb, head, ino);
1293		if (!old) {
1294			inode->i_ino = ino;
1295			spin_lock(&inode->i_lock);
1296			inode->i_state = I_NEW;
1297			hlist_add_head_rcu(&inode->i_hash, head);
1298			spin_unlock(&inode->i_lock);
1299			inode_sb_list_add(inode);
1300			spin_unlock(&inode_hash_lock);
1301
1302			/* Return the locked inode with I_NEW set, the
1303			 * caller is responsible for filling in the contents
1304			 */
1305			return inode;
1306		}
1307
1308		/*
1309		 * Uhhuh, somebody else created the same inode under
1310		 * us. Use the old inode instead of the one we just
1311		 * allocated.
1312		 */
1313		spin_unlock(&inode_hash_lock);
1314		destroy_inode(inode);
1315		if (IS_ERR(old))
1316			return NULL;
1317		inode = old;
1318		wait_on_inode(inode);
1319		if (unlikely(inode_unhashed(inode))) {
1320			iput(inode);
1321			goto again;
1322		}
1323	}
1324	return inode;
1325}
1326EXPORT_SYMBOL(iget_locked);
1327
1328/*
1329 * search the inode cache for a matching inode number.
1330 * If we find one, then the inode number we are trying to
1331 * allocate is not unique and so we should not use it.
1332 *
1333 * Returns 1 if the inode number is unique, 0 if it is not.
1334 */
1335static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1336{
1337	struct hlist_head *b = inode_hashtable + hash(sb, ino);
1338	struct inode *inode;
1339
1340	hlist_for_each_entry_rcu(inode, b, i_hash) {
1341		if (inode->i_ino == ino && inode->i_sb == sb)
1342			return 0;
1343	}
1344	return 1;
1345}
1346
1347/**
1348 *	iunique - get a unique inode number
1349 *	@sb: superblock
1350 *	@max_reserved: highest reserved inode number
1351 *
1352 *	Obtain an inode number that is unique on the system for a given
1353 *	superblock. This is used by file systems that have no natural
1354 *	permanent inode numbering system. An inode number is returned that
1355 *	is higher than the reserved limit but unique.
1356 *
1357 *	BUGS:
1358 *	With a large number of inodes live on the file system this function
1359 *	currently becomes quite slow.
1360 */
1361ino_t iunique(struct super_block *sb, ino_t max_reserved)
1362{
1363	/*
1364	 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1365	 * error if st_ino won't fit in target struct field. Use 32bit counter
1366	 * here to attempt to avoid that.
1367	 */
1368	static DEFINE_SPINLOCK(iunique_lock);
1369	static unsigned int counter;
1370	ino_t res;
1371
1372	rcu_read_lock();
1373	spin_lock(&iunique_lock);
1374	do {
1375		if (counter <= max_reserved)
1376			counter = max_reserved + 1;
1377		res = counter++;
1378	} while (!test_inode_iunique(sb, res));
1379	spin_unlock(&iunique_lock);
1380	rcu_read_unlock();
1381
1382	return res;
1383}
1384EXPORT_SYMBOL(iunique);
1385
1386struct inode *igrab(struct inode *inode)
1387{
1388	spin_lock(&inode->i_lock);
1389	if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1390		__iget(inode);
1391		spin_unlock(&inode->i_lock);
1392	} else {
1393		spin_unlock(&inode->i_lock);
1394		/*
1395		 * Handle the case where s_op->clear_inode is not been
1396		 * called yet, and somebody is calling igrab
1397		 * while the inode is getting freed.
1398		 */
1399		inode = NULL;
1400	}
1401	return inode;
1402}
1403EXPORT_SYMBOL(igrab);
1404
1405/**
1406 * ilookup5_nowait - search for an inode in the inode cache
1407 * @sb:		super block of file system to search
1408 * @hashval:	hash value (usually inode number) to search for
1409 * @test:	callback used for comparisons between inodes
1410 * @data:	opaque data pointer to pass to @test
1411 *
1412 * Search for the inode specified by @hashval and @data in the inode cache.
1413 * If the inode is in the cache, the inode is returned with an incremented
1414 * reference count.
1415 *
1416 * Note: I_NEW is not waited upon so you have to be very careful what you do
1417 * with the returned inode.  You probably should be using ilookup5() instead.
1418 *
1419 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1420 */
1421struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1422		int (*test)(struct inode *, void *), void *data)
1423{
1424	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1425	struct inode *inode;
1426
1427	spin_lock(&inode_hash_lock);
1428	inode = find_inode(sb, head, test, data);
1429	spin_unlock(&inode_hash_lock);
1430
1431	return IS_ERR(inode) ? NULL : inode;
1432}
1433EXPORT_SYMBOL(ilookup5_nowait);
1434
1435/**
1436 * ilookup5 - search for an inode in the inode cache
1437 * @sb:		super block of file system to search
1438 * @hashval:	hash value (usually inode number) to search for
1439 * @test:	callback used for comparisons between inodes
1440 * @data:	opaque data pointer to pass to @test
1441 *
1442 * Search for the inode specified by @hashval and @data in the inode cache,
1443 * and if the inode is in the cache, return the inode with an incremented
1444 * reference count.  Waits on I_NEW before returning the inode.
1445 * returned with an incremented reference count.
1446 *
1447 * This is a generalized version of ilookup() for file systems where the
1448 * inode number is not sufficient for unique identification of an inode.
1449 *
1450 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1451 */
1452struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1453		int (*test)(struct inode *, void *), void *data)
1454{
1455	struct inode *inode;
1456again:
1457	inode = ilookup5_nowait(sb, hashval, test, data);
1458	if (inode) {
1459		wait_on_inode(inode);
1460		if (unlikely(inode_unhashed(inode))) {
1461			iput(inode);
1462			goto again;
1463		}
1464	}
1465	return inode;
1466}
1467EXPORT_SYMBOL(ilookup5);
1468
1469/**
1470 * ilookup - search for an inode in the inode cache
1471 * @sb:		super block of file system to search
1472 * @ino:	inode number to search for
1473 *
1474 * Search for the inode @ino in the inode cache, and if the inode is in the
1475 * cache, the inode is returned with an incremented reference count.
1476 */
1477struct inode *ilookup(struct super_block *sb, unsigned long ino)
1478{
1479	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1480	struct inode *inode;
1481again:
1482	spin_lock(&inode_hash_lock);
1483	inode = find_inode_fast(sb, head, ino);
1484	spin_unlock(&inode_hash_lock);
1485
1486	if (inode) {
1487		if (IS_ERR(inode))
1488			return NULL;
1489		wait_on_inode(inode);
1490		if (unlikely(inode_unhashed(inode))) {
1491			iput(inode);
1492			goto again;
1493		}
1494	}
1495	return inode;
1496}
1497EXPORT_SYMBOL(ilookup);
1498
1499/**
1500 * find_inode_nowait - find an inode in the inode cache
1501 * @sb:		super block of file system to search
1502 * @hashval:	hash value (usually inode number) to search for
1503 * @match:	callback used for comparisons between inodes
1504 * @data:	opaque data pointer to pass to @match
1505 *
1506 * Search for the inode specified by @hashval and @data in the inode
1507 * cache, where the helper function @match will return 0 if the inode
1508 * does not match, 1 if the inode does match, and -1 if the search
1509 * should be stopped.  The @match function must be responsible for
1510 * taking the i_lock spin_lock and checking i_state for an inode being
1511 * freed or being initialized, and incrementing the reference count
1512 * before returning 1.  It also must not sleep, since it is called with
1513 * the inode_hash_lock spinlock held.
1514 *
1515 * This is a even more generalized version of ilookup5() when the
1516 * function must never block --- find_inode() can block in
1517 * __wait_on_freeing_inode() --- or when the caller can not increment
1518 * the reference count because the resulting iput() might cause an
1519 * inode eviction.  The tradeoff is that the @match funtion must be
1520 * very carefully implemented.
1521 */
1522struct inode *find_inode_nowait(struct super_block *sb,
1523				unsigned long hashval,
1524				int (*match)(struct inode *, unsigned long,
1525					     void *),
1526				void *data)
1527{
1528	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1529	struct inode *inode, *ret_inode = NULL;
1530	int mval;
1531
1532	spin_lock(&inode_hash_lock);
1533	hlist_for_each_entry(inode, head, i_hash) {
1534		if (inode->i_sb != sb)
1535			continue;
1536		mval = match(inode, hashval, data);
1537		if (mval == 0)
1538			continue;
1539		if (mval == 1)
1540			ret_inode = inode;
1541		goto out;
1542	}
1543out:
1544	spin_unlock(&inode_hash_lock);
1545	return ret_inode;
1546}
1547EXPORT_SYMBOL(find_inode_nowait);
1548
1549/**
1550 * find_inode_rcu - find an inode in the inode cache
1551 * @sb:		Super block of file system to search
1552 * @hashval:	Key to hash
1553 * @test:	Function to test match on an inode
1554 * @data:	Data for test function
1555 *
1556 * Search for the inode specified by @hashval and @data in the inode cache,
1557 * where the helper function @test will return 0 if the inode does not match
1558 * and 1 if it does.  The @test function must be responsible for taking the
1559 * i_lock spin_lock and checking i_state for an inode being freed or being
1560 * initialized.
1561 *
1562 * If successful, this will return the inode for which the @test function
1563 * returned 1 and NULL otherwise.
1564 *
1565 * The @test function is not permitted to take a ref on any inode presented.
1566 * It is also not permitted to sleep.
1567 *
1568 * The caller must hold the RCU read lock.
1569 */
1570struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1571			     int (*test)(struct inode *, void *), void *data)
1572{
1573	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1574	struct inode *inode;
1575
1576	RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1577			 "suspicious find_inode_rcu() usage");
1578
1579	hlist_for_each_entry_rcu(inode, head, i_hash) {
1580		if (inode->i_sb == sb &&
1581		    !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1582		    test(inode, data))
1583			return inode;
1584	}
1585	return NULL;
1586}
1587EXPORT_SYMBOL(find_inode_rcu);
1588
1589/**
1590 * find_inode_by_ino_rcu - Find an inode in the inode cache
1591 * @sb:		Super block of file system to search
1592 * @ino:	The inode number to match
1593 *
1594 * Search for the inode specified by @hashval and @data in the inode cache,
1595 * where the helper function @test will return 0 if the inode does not match
1596 * and 1 if it does.  The @test function must be responsible for taking the
1597 * i_lock spin_lock and checking i_state for an inode being freed or being
1598 * initialized.
1599 *
1600 * If successful, this will return the inode for which the @test function
1601 * returned 1 and NULL otherwise.
1602 *
1603 * The @test function is not permitted to take a ref on any inode presented.
1604 * It is also not permitted to sleep.
1605 *
1606 * The caller must hold the RCU read lock.
1607 */
1608struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1609				    unsigned long ino)
1610{
1611	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1612	struct inode *inode;
1613
1614	RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1615			 "suspicious find_inode_by_ino_rcu() usage");
1616
1617	hlist_for_each_entry_rcu(inode, head, i_hash) {
1618		if (inode->i_ino == ino &&
1619		    inode->i_sb == sb &&
1620		    !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1621		    return inode;
1622	}
1623	return NULL;
1624}
1625EXPORT_SYMBOL(find_inode_by_ino_rcu);
1626
1627int insert_inode_locked(struct inode *inode)
1628{
1629	struct super_block *sb = inode->i_sb;
1630	ino_t ino = inode->i_ino;
1631	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1632
1633	while (1) {
1634		struct inode *old = NULL;
1635		spin_lock(&inode_hash_lock);
1636		hlist_for_each_entry(old, head, i_hash) {
1637			if (old->i_ino != ino)
1638				continue;
1639			if (old->i_sb != sb)
1640				continue;
1641			spin_lock(&old->i_lock);
1642			if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1643				spin_unlock(&old->i_lock);
1644				continue;
1645			}
1646			break;
1647		}
1648		if (likely(!old)) {
1649			spin_lock(&inode->i_lock);
1650			inode->i_state |= I_NEW | I_CREATING;
1651			hlist_add_head_rcu(&inode->i_hash, head);
1652			spin_unlock(&inode->i_lock);
1653			spin_unlock(&inode_hash_lock);
1654			return 0;
1655		}
1656		if (unlikely(old->i_state & I_CREATING)) {
1657			spin_unlock(&old->i_lock);
1658			spin_unlock(&inode_hash_lock);
1659			return -EBUSY;
1660		}
1661		__iget(old);
1662		spin_unlock(&old->i_lock);
1663		spin_unlock(&inode_hash_lock);
1664		wait_on_inode(old);
1665		if (unlikely(!inode_unhashed(old))) {
1666			iput(old);
1667			return -EBUSY;
1668		}
1669		iput(old);
1670	}
1671}
1672EXPORT_SYMBOL(insert_inode_locked);
1673
1674int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1675		int (*test)(struct inode *, void *), void *data)
1676{
1677	struct inode *old;
1678
1679	inode->i_state |= I_CREATING;
1680	old = inode_insert5(inode, hashval, test, NULL, data);
1681
1682	if (old != inode) {
1683		iput(old);
1684		return -EBUSY;
1685	}
1686	return 0;
1687}
1688EXPORT_SYMBOL(insert_inode_locked4);
1689
1690
1691int generic_delete_inode(struct inode *inode)
1692{
1693	return 1;
1694}
1695EXPORT_SYMBOL(generic_delete_inode);
1696
1697/*
1698 * Called when we're dropping the last reference
1699 * to an inode.
1700 *
1701 * Call the FS "drop_inode()" function, defaulting to
1702 * the legacy UNIX filesystem behaviour.  If it tells
1703 * us to evict inode, do so.  Otherwise, retain inode
1704 * in cache if fs is alive, sync and evict if fs is
1705 * shutting down.
1706 */
1707static void iput_final(struct inode *inode)
1708{
1709	struct super_block *sb = inode->i_sb;
1710	const struct super_operations *op = inode->i_sb->s_op;
1711	unsigned long state;
1712	int drop;
1713
1714	WARN_ON(inode->i_state & I_NEW);
1715
1716	if (op->drop_inode)
1717		drop = op->drop_inode(inode);
1718	else
1719		drop = generic_drop_inode(inode);
1720
1721	if (!drop &&
1722	    !(inode->i_state & I_DONTCACHE) &&
1723	    (sb->s_flags & SB_ACTIVE)) {
1724		__inode_add_lru(inode, true);
1725		spin_unlock(&inode->i_lock);
1726		return;
1727	}
1728
1729	state = inode->i_state;
1730	if (!drop) {
1731		WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1732		spin_unlock(&inode->i_lock);
1733
1734		write_inode_now(inode, 1);
1735
1736		spin_lock(&inode->i_lock);
1737		state = inode->i_state;
1738		WARN_ON(state & I_NEW);
1739		state &= ~I_WILL_FREE;
1740	}
1741
1742	WRITE_ONCE(inode->i_state, state | I_FREEING);
1743	if (!list_empty(&inode->i_lru))
1744		inode_lru_list_del(inode);
1745	spin_unlock(&inode->i_lock);
1746
1747	evict(inode);
1748}
1749
1750/**
1751 *	iput	- put an inode
1752 *	@inode: inode to put
1753 *
1754 *	Puts an inode, dropping its usage count. If the inode use count hits
1755 *	zero, the inode is then freed and may also be destroyed.
1756 *
1757 *	Consequently, iput() can sleep.
1758 */
1759void iput(struct inode *inode)
1760{
1761	if (!inode)
1762		return;
1763	BUG_ON(inode->i_state & I_CLEAR);
1764retry:
1765	if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1766		if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1767			atomic_inc(&inode->i_count);
1768			spin_unlock(&inode->i_lock);
1769			trace_writeback_lazytime_iput(inode);
1770			mark_inode_dirty_sync(inode);
1771			goto retry;
1772		}
1773		iput_final(inode);
1774	}
1775}
1776EXPORT_SYMBOL(iput);
1777
1778#ifdef CONFIG_BLOCK
1779/**
1780 *	bmap	- find a block number in a file
1781 *	@inode:  inode owning the block number being requested
1782 *	@block: pointer containing the block to find
1783 *
1784 *	Replaces the value in ``*block`` with the block number on the device holding
1785 *	corresponding to the requested block number in the file.
1786 *	That is, asked for block 4 of inode 1 the function will replace the
1787 *	4 in ``*block``, with disk block relative to the disk start that holds that
1788 *	block of the file.
1789 *
1790 *	Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1791 *	hole, returns 0 and ``*block`` is also set to 0.
1792 */
1793int bmap(struct inode *inode, sector_t *block)
1794{
1795	if (!inode->i_mapping->a_ops->bmap)
1796		return -EINVAL;
1797
1798	*block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1799	return 0;
1800}
1801EXPORT_SYMBOL(bmap);
1802#endif
1803
1804/*
1805 * With relative atime, only update atime if the previous atime is
1806 * earlier than either the ctime or mtime or if at least a day has
1807 * passed since the last atime update.
1808 */
1809static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1810			     struct timespec64 now)
1811{
 
1812
1813	if (!(mnt->mnt_flags & MNT_RELATIME))
1814		return 1;
1815	/*
1816	 * Is mtime younger than atime? If yes, update atime:
1817	 */
1818	if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1819		return 1;
 
 
1820	/*
1821	 * Is ctime younger than atime? If yes, update atime:
1822	 */
1823	if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1824		return 1;
 
1825
1826	/*
1827	 * Is the previous atime value older than a day? If yes,
1828	 * update atime:
1829	 */
1830	if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1831		return 1;
1832	/*
1833	 * Good, we can skip the atime update:
1834	 */
1835	return 0;
1836}
1837
1838int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1839{
1840	int dirty_flags = 0;
 
1841
1842	if (flags & (S_ATIME | S_CTIME | S_MTIME)) {
1843		if (flags & S_ATIME)
1844			inode->i_atime = *time;
1845		if (flags & S_CTIME)
1846			inode->i_ctime = *time;
1847		if (flags & S_MTIME)
1848			inode->i_mtime = *time;
1849
1850		if (inode->i_sb->s_flags & SB_LAZYTIME)
1851			dirty_flags |= I_DIRTY_TIME;
1852		else
1853			dirty_flags |= I_DIRTY_SYNC;
 
 
 
1854	}
1855
1856	if ((flags & S_VERSION) && inode_maybe_inc_iversion(inode, false))
1857		dirty_flags |= I_DIRTY_SYNC;
 
 
 
 
 
 
 
 
 
1858
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1859	__mark_inode_dirty(inode, dirty_flags);
1860	return 0;
1861}
1862EXPORT_SYMBOL(generic_update_time);
1863
1864/*
1865 * This does the actual work of updating an inodes time or version.  Must have
1866 * had called mnt_want_write() before calling this.
1867 */
1868int inode_update_time(struct inode *inode, struct timespec64 *time, int flags)
1869{
1870	if (inode->i_op->update_time)
1871		return inode->i_op->update_time(inode, time, flags);
1872	return generic_update_time(inode, time, flags);
 
1873}
1874EXPORT_SYMBOL(inode_update_time);
1875
1876/**
1877 *	atime_needs_update	-	update the access time
1878 *	@path: the &struct path to update
1879 *	@inode: inode to update
1880 *
1881 *	Update the accessed time on an inode and mark it for writeback.
1882 *	This function automatically handles read only file systems and media,
1883 *	as well as the "noatime" flag and inode specific "noatime" markers.
1884 */
1885bool atime_needs_update(const struct path *path, struct inode *inode)
1886{
1887	struct vfsmount *mnt = path->mnt;
1888	struct timespec64 now;
1889
1890	if (inode->i_flags & S_NOATIME)
1891		return false;
1892
1893	/* Atime updates will likely cause i_uid and i_gid to be written
1894	 * back improprely if their true value is unknown to the vfs.
1895	 */
1896	if (HAS_UNMAPPED_ID(mnt_user_ns(mnt), inode))
1897		return false;
1898
1899	if (IS_NOATIME(inode))
1900		return false;
1901	if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1902		return false;
1903
1904	if (mnt->mnt_flags & MNT_NOATIME)
1905		return false;
1906	if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1907		return false;
1908
1909	now = current_time(inode);
1910
1911	if (!relatime_need_update(mnt, inode, now))
1912		return false;
1913
1914	if (timespec64_equal(&inode->i_atime, &now))
 
1915		return false;
1916
1917	return true;
1918}
1919
1920void touch_atime(const struct path *path)
1921{
1922	struct vfsmount *mnt = path->mnt;
1923	struct inode *inode = d_inode(path->dentry);
1924	struct timespec64 now;
1925
1926	if (!atime_needs_update(path, inode))
1927		return;
1928
1929	if (!sb_start_write_trylock(inode->i_sb))
1930		return;
1931
1932	if (__mnt_want_write(mnt) != 0)
1933		goto skip_update;
1934	/*
1935	 * File systems can error out when updating inodes if they need to
1936	 * allocate new space to modify an inode (such is the case for
1937	 * Btrfs), but since we touch atime while walking down the path we
1938	 * really don't care if we failed to update the atime of the file,
1939	 * so just ignore the return value.
1940	 * We may also fail on filesystems that have the ability to make parts
1941	 * of the fs read only, e.g. subvolumes in Btrfs.
1942	 */
1943	now = current_time(inode);
1944	inode_update_time(inode, &now, S_ATIME);
1945	__mnt_drop_write(mnt);
1946skip_update:
1947	sb_end_write(inode->i_sb);
1948}
1949EXPORT_SYMBOL(touch_atime);
1950
1951/*
1952 * Return mask of changes for notify_change() that need to be done as a
1953 * response to write or truncate. Return 0 if nothing has to be changed.
1954 * Negative value on error (change should be denied).
1955 */
1956int dentry_needs_remove_privs(struct user_namespace *mnt_userns,
1957			      struct dentry *dentry)
1958{
1959	struct inode *inode = d_inode(dentry);
1960	int mask = 0;
1961	int ret;
1962
1963	if (IS_NOSEC(inode))
1964		return 0;
1965
1966	mask = setattr_should_drop_suidgid(mnt_userns, inode);
1967	ret = security_inode_need_killpriv(dentry);
1968	if (ret < 0)
1969		return ret;
1970	if (ret)
1971		mask |= ATTR_KILL_PRIV;
1972	return mask;
1973}
1974
1975static int __remove_privs(struct user_namespace *mnt_userns,
1976			  struct dentry *dentry, int kill)
1977{
1978	struct iattr newattrs;
1979
1980	newattrs.ia_valid = ATTR_FORCE | kill;
1981	/*
1982	 * Note we call this on write, so notify_change will not
1983	 * encounter any conflicting delegations:
1984	 */
1985	return notify_change(mnt_userns, dentry, &newattrs, NULL);
1986}
1987
1988static int __file_remove_privs(struct file *file, unsigned int flags)
1989{
1990	struct dentry *dentry = file_dentry(file);
1991	struct inode *inode = file_inode(file);
1992	int error = 0;
1993	int kill;
1994
1995	if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1996		return 0;
1997
1998	kill = dentry_needs_remove_privs(file_mnt_user_ns(file), dentry);
1999	if (kill < 0)
2000		return kill;
2001
2002	if (kill) {
2003		if (flags & IOCB_NOWAIT)
2004			return -EAGAIN;
2005
2006		error = __remove_privs(file_mnt_user_ns(file), dentry, kill);
2007	}
2008
2009	if (!error)
2010		inode_has_no_xattr(inode);
2011	return error;
2012}
2013
2014/**
2015 * file_remove_privs - remove special file privileges (suid, capabilities)
2016 * @file: file to remove privileges from
2017 *
2018 * When file is modified by a write or truncation ensure that special
2019 * file privileges are removed.
2020 *
2021 * Return: 0 on success, negative errno on failure.
2022 */
2023int file_remove_privs(struct file *file)
2024{
2025	return __file_remove_privs(file, 0);
2026}
2027EXPORT_SYMBOL(file_remove_privs);
2028
2029static int inode_needs_update_time(struct inode *inode, struct timespec64 *now)
2030{
2031	int sync_it = 0;
 
 
2032
2033	/* First try to exhaust all avenues to not sync */
2034	if (IS_NOCMTIME(inode))
2035		return 0;
2036
2037	if (!timespec64_equal(&inode->i_mtime, now))
 
2038		sync_it = S_MTIME;
2039
2040	if (!timespec64_equal(&inode->i_ctime, now))
 
2041		sync_it |= S_CTIME;
2042
2043	if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2044		sync_it |= S_VERSION;
2045
2046	return sync_it;
2047}
2048
2049static int __file_update_time(struct file *file, struct timespec64 *now,
2050			int sync_mode)
2051{
2052	int ret = 0;
2053	struct inode *inode = file_inode(file);
2054
2055	/* try to update time settings */
2056	if (!__mnt_want_write_file(file)) {
2057		ret = inode_update_time(inode, now, sync_mode);
2058		__mnt_drop_write_file(file);
2059	}
2060
2061	return ret;
2062}
2063
2064/**
2065 * file_update_time - update mtime and ctime time
2066 * @file: file accessed
2067 *
2068 * Update the mtime and ctime members of an inode and mark the inode for
2069 * writeback. Note that this function is meant exclusively for usage in
2070 * the file write path of filesystems, and filesystems may choose to
2071 * explicitly ignore updates via this function with the _NOCMTIME inode
2072 * flag, e.g. for network filesystem where these imestamps are handled
2073 * by the server. This can return an error for file systems who need to
2074 * allocate space in order to update an inode.
2075 *
2076 * Return: 0 on success, negative errno on failure.
2077 */
2078int file_update_time(struct file *file)
2079{
2080	int ret;
2081	struct inode *inode = file_inode(file);
2082	struct timespec64 now = current_time(inode);
2083
2084	ret = inode_needs_update_time(inode, &now);
2085	if (ret <= 0)
2086		return ret;
2087
2088	return __file_update_time(file, &now, ret);
2089}
2090EXPORT_SYMBOL(file_update_time);
2091
2092/**
2093 * file_modified_flags - handle mandated vfs changes when modifying a file
2094 * @file: file that was modified
2095 * @flags: kiocb flags
2096 *
2097 * When file has been modified ensure that special
2098 * file privileges are removed and time settings are updated.
2099 *
2100 * If IOCB_NOWAIT is set, special file privileges will not be removed and
2101 * time settings will not be updated. It will return -EAGAIN.
2102 *
2103 * Context: Caller must hold the file's inode lock.
2104 *
2105 * Return: 0 on success, negative errno on failure.
2106 */
2107static int file_modified_flags(struct file *file, int flags)
2108{
2109	int ret;
2110	struct inode *inode = file_inode(file);
2111	struct timespec64 now = current_time(inode);
2112
2113	/*
2114	 * Clear the security bits if the process is not being run by root.
2115	 * This keeps people from modifying setuid and setgid binaries.
2116	 */
2117	ret = __file_remove_privs(file, flags);
2118	if (ret)
2119		return ret;
2120
2121	if (unlikely(file->f_mode & FMODE_NOCMTIME))
2122		return 0;
2123
2124	ret = inode_needs_update_time(inode, &now);
2125	if (ret <= 0)
2126		return ret;
2127	if (flags & IOCB_NOWAIT)
2128		return -EAGAIN;
2129
2130	return __file_update_time(file, &now, ret);
2131}
2132
2133/**
2134 * file_modified - handle mandated vfs changes when modifying a file
2135 * @file: file that was modified
2136 *
2137 * When file has been modified ensure that special
2138 * file privileges are removed and time settings are updated.
2139 *
2140 * Context: Caller must hold the file's inode lock.
2141 *
2142 * Return: 0 on success, negative errno on failure.
2143 */
2144int file_modified(struct file *file)
2145{
2146	return file_modified_flags(file, 0);
2147}
2148EXPORT_SYMBOL(file_modified);
2149
2150/**
2151 * kiocb_modified - handle mandated vfs changes when modifying a file
2152 * @iocb: iocb that was modified
2153 *
2154 * When file has been modified ensure that special
2155 * file privileges are removed and time settings are updated.
2156 *
2157 * Context: Caller must hold the file's inode lock.
2158 *
2159 * Return: 0 on success, negative errno on failure.
2160 */
2161int kiocb_modified(struct kiocb *iocb)
2162{
2163	return file_modified_flags(iocb->ki_filp, iocb->ki_flags);
2164}
2165EXPORT_SYMBOL_GPL(kiocb_modified);
2166
2167int inode_needs_sync(struct inode *inode)
2168{
2169	if (IS_SYNC(inode))
2170		return 1;
2171	if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2172		return 1;
2173	return 0;
2174}
2175EXPORT_SYMBOL(inode_needs_sync);
2176
2177/*
2178 * If we try to find an inode in the inode hash while it is being
2179 * deleted, we have to wait until the filesystem completes its
2180 * deletion before reporting that it isn't found.  This function waits
2181 * until the deletion _might_ have completed.  Callers are responsible
2182 * to recheck inode state.
2183 *
2184 * It doesn't matter if I_NEW is not set initially, a call to
2185 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2186 * will DTRT.
2187 */
2188static void __wait_on_freeing_inode(struct inode *inode)
2189{
2190	wait_queue_head_t *wq;
2191	DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2192	wq = bit_waitqueue(&inode->i_state, __I_NEW);
2193	prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2194	spin_unlock(&inode->i_lock);
2195	spin_unlock(&inode_hash_lock);
2196	schedule();
2197	finish_wait(wq, &wait.wq_entry);
2198	spin_lock(&inode_hash_lock);
2199}
2200
2201static __initdata unsigned long ihash_entries;
2202static int __init set_ihash_entries(char *str)
2203{
2204	if (!str)
2205		return 0;
2206	ihash_entries = simple_strtoul(str, &str, 0);
2207	return 1;
2208}
2209__setup("ihash_entries=", set_ihash_entries);
2210
2211/*
2212 * Initialize the waitqueues and inode hash table.
2213 */
2214void __init inode_init_early(void)
2215{
2216	/* If hashes are distributed across NUMA nodes, defer
2217	 * hash allocation until vmalloc space is available.
2218	 */
2219	if (hashdist)
2220		return;
2221
2222	inode_hashtable =
2223		alloc_large_system_hash("Inode-cache",
2224					sizeof(struct hlist_head),
2225					ihash_entries,
2226					14,
2227					HASH_EARLY | HASH_ZERO,
2228					&i_hash_shift,
2229					&i_hash_mask,
2230					0,
2231					0);
2232}
2233
2234void __init inode_init(void)
2235{
2236	/* inode slab cache */
2237	inode_cachep = kmem_cache_create("inode_cache",
2238					 sizeof(struct inode),
2239					 0,
2240					 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2241					 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2242					 init_once);
2243
2244	/* Hash may have been set up in inode_init_early */
2245	if (!hashdist)
2246		return;
2247
2248	inode_hashtable =
2249		alloc_large_system_hash("Inode-cache",
2250					sizeof(struct hlist_head),
2251					ihash_entries,
2252					14,
2253					HASH_ZERO,
2254					&i_hash_shift,
2255					&i_hash_mask,
2256					0,
2257					0);
2258}
2259
2260void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2261{
2262	inode->i_mode = mode;
2263	if (S_ISCHR(mode)) {
2264		inode->i_fop = &def_chr_fops;
2265		inode->i_rdev = rdev;
2266	} else if (S_ISBLK(mode)) {
2267		inode->i_fop = &def_blk_fops;
 
2268		inode->i_rdev = rdev;
2269	} else if (S_ISFIFO(mode))
2270		inode->i_fop = &pipefifo_fops;
2271	else if (S_ISSOCK(mode))
2272		;	/* leave it no_open_fops */
2273	else
2274		printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2275				  " inode %s:%lu\n", mode, inode->i_sb->s_id,
2276				  inode->i_ino);
2277}
2278EXPORT_SYMBOL(init_special_inode);
2279
2280/**
2281 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2282 * @mnt_userns:	User namespace of the mount the inode was created from
2283 * @inode: New inode
2284 * @dir: Directory inode
2285 * @mode: mode of the new inode
2286 *
2287 * If the inode has been created through an idmapped mount the user namespace of
2288 * the vfsmount must be passed through @mnt_userns. This function will then take
2289 * care to map the inode according to @mnt_userns before checking permissions
2290 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2291 * checking is to be performed on the raw inode simply passs init_user_ns.
2292 */
2293void inode_init_owner(struct user_namespace *mnt_userns, struct inode *inode,
2294		      const struct inode *dir, umode_t mode)
2295{
2296	inode_fsuid_set(inode, mnt_userns);
2297	if (dir && dir->i_mode & S_ISGID) {
2298		inode->i_gid = dir->i_gid;
2299
2300		/* Directories are special, and always inherit S_ISGID */
2301		if (S_ISDIR(mode))
2302			mode |= S_ISGID;
2303	} else
2304		inode_fsgid_set(inode, mnt_userns);
2305	inode->i_mode = mode;
2306}
2307EXPORT_SYMBOL(inode_init_owner);
2308
2309/**
2310 * inode_owner_or_capable - check current task permissions to inode
2311 * @mnt_userns:	user namespace of the mount the inode was found from
2312 * @inode: inode being checked
2313 *
2314 * Return true if current either has CAP_FOWNER in a namespace with the
2315 * inode owner uid mapped, or owns the file.
2316 *
2317 * If the inode has been found through an idmapped mount the user namespace of
2318 * the vfsmount must be passed through @mnt_userns. This function will then take
2319 * care to map the inode according to @mnt_userns before checking permissions.
2320 * On non-idmapped mounts or if permission checking is to be performed on the
2321 * raw inode simply passs init_user_ns.
2322 */
2323bool inode_owner_or_capable(struct user_namespace *mnt_userns,
2324			    const struct inode *inode)
2325{
2326	vfsuid_t vfsuid;
2327	struct user_namespace *ns;
2328
2329	vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
2330	if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
2331		return true;
2332
2333	ns = current_user_ns();
2334	if (vfsuid_has_mapping(ns, vfsuid) && ns_capable(ns, CAP_FOWNER))
2335		return true;
2336	return false;
2337}
2338EXPORT_SYMBOL(inode_owner_or_capable);
2339
2340/*
2341 * Direct i/o helper functions
2342 */
2343static void __inode_dio_wait(struct inode *inode)
2344{
2345	wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2346	DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2347
2348	do {
2349		prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2350		if (atomic_read(&inode->i_dio_count))
2351			schedule();
2352	} while (atomic_read(&inode->i_dio_count));
2353	finish_wait(wq, &q.wq_entry);
2354}
2355
2356/**
2357 * inode_dio_wait - wait for outstanding DIO requests to finish
2358 * @inode: inode to wait for
2359 *
2360 * Waits for all pending direct I/O requests to finish so that we can
2361 * proceed with a truncate or equivalent operation.
2362 *
2363 * Must be called under a lock that serializes taking new references
2364 * to i_dio_count, usually by inode->i_mutex.
2365 */
2366void inode_dio_wait(struct inode *inode)
2367{
2368	if (atomic_read(&inode->i_dio_count))
2369		__inode_dio_wait(inode);
2370}
2371EXPORT_SYMBOL(inode_dio_wait);
2372
2373/*
2374 * inode_set_flags - atomically set some inode flags
2375 *
2376 * Note: the caller should be holding i_mutex, or else be sure that
2377 * they have exclusive access to the inode structure (i.e., while the
2378 * inode is being instantiated).  The reason for the cmpxchg() loop
2379 * --- which wouldn't be necessary if all code paths which modify
2380 * i_flags actually followed this rule, is that there is at least one
2381 * code path which doesn't today so we use cmpxchg() out of an abundance
2382 * of caution.
2383 *
2384 * In the long run, i_mutex is overkill, and we should probably look
2385 * at using the i_lock spinlock to protect i_flags, and then make sure
2386 * it is so documented in include/linux/fs.h and that all code follows
2387 * the locking convention!!
2388 */
2389void inode_set_flags(struct inode *inode, unsigned int flags,
2390		     unsigned int mask)
2391{
2392	WARN_ON_ONCE(flags & ~mask);
2393	set_mask_bits(&inode->i_flags, mask, flags);
2394}
2395EXPORT_SYMBOL(inode_set_flags);
2396
2397void inode_nohighmem(struct inode *inode)
2398{
2399	mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2400}
2401EXPORT_SYMBOL(inode_nohighmem);
2402
2403/**
2404 * timestamp_truncate - Truncate timespec to a granularity
2405 * @t: Timespec
2406 * @inode: inode being updated
2407 *
2408 * Truncate a timespec to the granularity supported by the fs
2409 * containing the inode. Always rounds down. gran must
2410 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2411 */
2412struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2413{
2414	struct super_block *sb = inode->i_sb;
2415	unsigned int gran = sb->s_time_gran;
2416
2417	t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2418	if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2419		t.tv_nsec = 0;
2420
2421	/* Avoid division in the common cases 1 ns and 1 s. */
2422	if (gran == 1)
2423		; /* nothing */
2424	else if (gran == NSEC_PER_SEC)
2425		t.tv_nsec = 0;
2426	else if (gran > 1 && gran < NSEC_PER_SEC)
2427		t.tv_nsec -= t.tv_nsec % gran;
2428	else
2429		WARN(1, "invalid file time granularity: %u", gran);
2430	return t;
2431}
2432EXPORT_SYMBOL(timestamp_truncate);
2433
2434/**
2435 * current_time - Return FS time
2436 * @inode: inode.
2437 *
2438 * Return the current time truncated to the time granularity supported by
2439 * the fs.
2440 *
2441 * Note that inode and inode->sb cannot be NULL.
2442 * Otherwise, the function warns and returns time without truncation.
2443 */
2444struct timespec64 current_time(struct inode *inode)
2445{
2446	struct timespec64 now;
2447
2448	ktime_get_coarse_real_ts64(&now);
2449
2450	if (unlikely(!inode->i_sb)) {
2451		WARN(1, "current_time() called with uninitialized super_block in the inode");
2452		return now;
2453	}
2454
2455	return timestamp_truncate(now, inode);
2456}
2457EXPORT_SYMBOL(current_time);
2458
2459/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2460 * in_group_or_capable - check whether caller is CAP_FSETID privileged
2461 * @mnt_userns: user namespace of the mount @inode was found from
2462 * @inode:	inode to check
2463 * @vfsgid:	the new/current vfsgid of @inode
2464 *
2465 * Check wether @vfsgid is in the caller's group list or if the caller is
2466 * privileged with CAP_FSETID over @inode. This can be used to determine
2467 * whether the setgid bit can be kept or must be dropped.
2468 *
2469 * Return: true if the caller is sufficiently privileged, false if not.
2470 */
2471bool in_group_or_capable(struct user_namespace *mnt_userns,
2472			 const struct inode *inode, vfsgid_t vfsgid)
2473{
2474	if (vfsgid_in_group_p(vfsgid))
2475		return true;
2476	if (capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FSETID))
2477		return true;
2478	return false;
2479}
2480
2481/**
2482 * mode_strip_sgid - handle the sgid bit for non-directories
2483 * @mnt_userns: User namespace of the mount the inode was created from
2484 * @dir: parent directory inode
2485 * @mode: mode of the file to be created in @dir
2486 *
2487 * If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2488 * raised and @dir has the S_ISGID bit raised ensure that the caller is
2489 * either in the group of the parent directory or they have CAP_FSETID
2490 * in their user namespace and are privileged over the parent directory.
2491 * In all other cases, strip the S_ISGID bit from @mode.
2492 *
2493 * Return: the new mode to use for the file
2494 */
2495umode_t mode_strip_sgid(struct user_namespace *mnt_userns,
2496			const struct inode *dir, umode_t mode)
2497{
2498	if ((mode & (S_ISGID | S_IXGRP)) != (S_ISGID | S_IXGRP))
2499		return mode;
2500	if (S_ISDIR(mode) || !dir || !(dir->i_mode & S_ISGID))
2501		return mode;
2502	if (in_group_or_capable(mnt_userns, dir,
2503				i_gid_into_vfsgid(mnt_userns, dir)))
2504		return mode;
2505	return mode & ~S_ISGID;
2506}
2507EXPORT_SYMBOL(mode_strip_sgid);