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