Linux Audio

Check our new training course

Loading...
v6.13.7
   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/filelock.h>
   9#include <linux/mm.h>
  10#include <linux/backing-dev.h>
  11#include <linux/hash.h>
  12#include <linux/swap.h>
  13#include <linux/security.h>
  14#include <linux/cdev.h>
  15#include <linux/memblock.h>
  16#include <linux/fsnotify.h>
  17#include <linux/mount.h>
  18#include <linux/posix_acl.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 <linux/rw_hint.h>
  24#include <linux/seq_file.h>
  25#include <linux/debugfs.h>
  26#include <trace/events/writeback.h>
  27#define CREATE_TRACE_POINTS
  28#include <trace/events/timestamp.h>
  29
  30#include "internal.h"
  31
  32/*
  33 * Inode locking rules:
  34 *
  35 * inode->i_lock protects:
  36 *   inode->i_state, inode->i_hash, __iget(), inode->i_io_list
  37 * Inode LRU list locks protect:
  38 *   inode->i_sb->s_inode_lru, inode->i_lru
  39 * inode->i_sb->s_inode_list_lock protects:
  40 *   inode->i_sb->s_inodes, inode->i_sb_list
  41 * bdi->wb.list_lock protects:
  42 *   bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
  43 * inode_hash_lock protects:
  44 *   inode_hashtable, inode->i_hash
  45 *
  46 * Lock ordering:
  47 *
  48 * inode->i_sb->s_inode_list_lock
  49 *   inode->i_lock
  50 *     Inode LRU list locks
  51 *
  52 * bdi->wb.list_lock
  53 *   inode->i_lock
  54 *
  55 * inode_hash_lock
  56 *   inode->i_sb->s_inode_list_lock
  57 *   inode->i_lock
  58 *
  59 * iunique_lock
  60 *   inode_hash_lock
  61 */
  62
  63static unsigned int i_hash_mask __ro_after_init;
  64static unsigned int i_hash_shift __ro_after_init;
  65static struct hlist_head *inode_hashtable __ro_after_init;
  66static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
  67
 
 
  68/*
  69 * Empty aops. Can be used for the cases where the user does not
  70 * define any of the address_space operations.
  71 */
  72const struct address_space_operations empty_aops = {
  73};
  74EXPORT_SYMBOL(empty_aops);
  75
 
 
 
 
 
  76static DEFINE_PER_CPU(unsigned long, nr_inodes);
  77static DEFINE_PER_CPU(unsigned long, nr_unused);
  78
  79static struct kmem_cache *inode_cachep __ro_after_init;
  80
  81static long get_nr_inodes(void)
  82{
  83	int i;
  84	long sum = 0;
  85	for_each_possible_cpu(i)
  86		sum += per_cpu(nr_inodes, i);
  87	return sum < 0 ? 0 : sum;
  88}
  89
  90static inline long get_nr_inodes_unused(void)
  91{
  92	int i;
  93	long sum = 0;
  94	for_each_possible_cpu(i)
  95		sum += per_cpu(nr_unused, i);
  96	return sum < 0 ? 0 : sum;
  97}
  98
  99long get_nr_dirty_inodes(void)
 100{
 101	/* not actually dirty inodes, but a wild approximation */
 102	long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
 103	return nr_dirty > 0 ? nr_dirty : 0;
 104}
 105
 106#ifdef CONFIG_DEBUG_FS
 107static DEFINE_PER_CPU(long, mg_ctime_updates);
 108static DEFINE_PER_CPU(long, mg_fine_stamps);
 109static DEFINE_PER_CPU(long, mg_ctime_swaps);
 110
 111static unsigned long get_mg_ctime_updates(void)
 112{
 113	unsigned long sum = 0;
 114	int i;
 115
 116	for_each_possible_cpu(i)
 117		sum += data_race(per_cpu(mg_ctime_updates, i));
 118	return sum;
 119}
 120
 121static unsigned long get_mg_fine_stamps(void)
 122{
 123	unsigned long sum = 0;
 124	int i;
 125
 126	for_each_possible_cpu(i)
 127		sum += data_race(per_cpu(mg_fine_stamps, i));
 128	return sum;
 129}
 130
 131static unsigned long get_mg_ctime_swaps(void)
 132{
 133	unsigned long sum = 0;
 134	int i;
 135
 136	for_each_possible_cpu(i)
 137		sum += data_race(per_cpu(mg_ctime_swaps, i));
 138	return sum;
 139}
 140
 141#define mgtime_counter_inc(__var)	this_cpu_inc(__var)
 142
 143static int mgts_show(struct seq_file *s, void *p)
 144{
 145	unsigned long ctime_updates = get_mg_ctime_updates();
 146	unsigned long ctime_swaps = get_mg_ctime_swaps();
 147	unsigned long fine_stamps = get_mg_fine_stamps();
 148	unsigned long floor_swaps = timekeeping_get_mg_floor_swaps();
 149
 150	seq_printf(s, "%lu %lu %lu %lu\n",
 151		   ctime_updates, ctime_swaps, fine_stamps, floor_swaps);
 152	return 0;
 153}
 154
 155DEFINE_SHOW_ATTRIBUTE(mgts);
 156
 157static int __init mg_debugfs_init(void)
 158{
 159	debugfs_create_file("multigrain_timestamps", S_IFREG | S_IRUGO, NULL, NULL, &mgts_fops);
 160	return 0;
 161}
 162late_initcall(mg_debugfs_init);
 163
 164#else /* ! CONFIG_DEBUG_FS */
 165
 166#define mgtime_counter_inc(__var)	do { } while (0)
 167
 168#endif /* CONFIG_DEBUG_FS */
 169
 170/*
 171 * Handle nr_inode sysctl
 172 */
 173#ifdef CONFIG_SYSCTL
 174/*
 175 * Statistics gathering..
 176 */
 177static struct inodes_stat_t inodes_stat;
 178
 179static int proc_nr_inodes(const struct ctl_table *table, int write, void *buffer,
 180			  size_t *lenp, loff_t *ppos)
 181{
 182	inodes_stat.nr_inodes = get_nr_inodes();
 183	inodes_stat.nr_unused = get_nr_inodes_unused();
 184	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 185}
 186
 187static struct ctl_table inodes_sysctls[] = {
 188	{
 189		.procname	= "inode-nr",
 190		.data		= &inodes_stat,
 191		.maxlen		= 2*sizeof(long),
 192		.mode		= 0444,
 193		.proc_handler	= proc_nr_inodes,
 194	},
 195	{
 196		.procname	= "inode-state",
 197		.data		= &inodes_stat,
 198		.maxlen		= 7*sizeof(long),
 199		.mode		= 0444,
 200		.proc_handler	= proc_nr_inodes,
 201	},
 202};
 203
 204static int __init init_fs_inode_sysctls(void)
 205{
 206	register_sysctl_init("fs", inodes_sysctls);
 207	return 0;
 208}
 209early_initcall(init_fs_inode_sysctls);
 210#endif
 211
 212static int no_open(struct inode *inode, struct file *file)
 213{
 214	return -ENXIO;
 215}
 216
 217/**
 218 * inode_init_always_gfp - perform inode structure initialisation
 219 * @sb: superblock inode belongs to
 220 * @inode: inode to initialise
 221 * @gfp: allocation flags
 222 *
 223 * These are initializations that need to be done on every inode
 224 * allocation as the fields are not initialised by slab allocation.
 225 * If there are additional allocations required @gfp is used.
 226 */
 227int inode_init_always_gfp(struct super_block *sb, struct inode *inode, gfp_t gfp)
 228{
 229	static const struct inode_operations empty_iops;
 230	static const struct file_operations no_open_fops = {.open = no_open};
 231	struct address_space *const mapping = &inode->i_data;
 232
 233	inode->i_sb = sb;
 234	inode->i_blkbits = sb->s_blocksize_bits;
 235	inode->i_flags = 0;
 236	inode->i_state = 0;
 237	atomic64_set(&inode->i_sequence, 0);
 238	atomic_set(&inode->i_count, 1);
 239	inode->i_op = &empty_iops;
 240	inode->i_fop = &no_open_fops;
 241	inode->i_ino = 0;
 242	inode->__i_nlink = 1;
 243	inode->i_opflags = 0;
 244	if (sb->s_xattr)
 245		inode->i_opflags |= IOP_XATTR;
 246	if (sb->s_type->fs_flags & FS_MGTIME)
 247		inode->i_opflags |= IOP_MGTIME;
 248	i_uid_write(inode, 0);
 249	i_gid_write(inode, 0);
 250	atomic_set(&inode->i_writecount, 0);
 251	inode->i_size = 0;
 252	inode->i_write_hint = WRITE_LIFE_NOT_SET;
 253	inode->i_blocks = 0;
 254	inode->i_bytes = 0;
 255	inode->i_generation = 0;
 
 
 
 256	inode->i_pipe = NULL;
 
 257	inode->i_cdev = NULL;
 258	inode->i_link = NULL;
 259	inode->i_dir_seq = 0;
 260	inode->i_rdev = 0;
 261	inode->dirtied_when = 0;
 262
 263#ifdef CONFIG_CGROUP_WRITEBACK
 264	inode->i_wb_frn_winner = 0;
 265	inode->i_wb_frn_avg_time = 0;
 266	inode->i_wb_frn_history = 0;
 267#endif
 268
 269	spin_lock_init(&inode->i_lock);
 270	lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
 271
 272	init_rwsem(&inode->i_rwsem);
 273	lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
 274
 275	atomic_set(&inode->i_dio_count, 0);
 276
 277	mapping->a_ops = &empty_aops;
 278	mapping->host = inode;
 279	mapping->flags = 0;
 280	mapping->wb_err = 0;
 281	atomic_set(&mapping->i_mmap_writable, 0);
 282#ifdef CONFIG_READ_ONLY_THP_FOR_FS
 283	atomic_set(&mapping->nr_thps, 0);
 284#endif
 285	mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
 286	mapping->i_private_data = NULL;
 
 287	mapping->writeback_index = 0;
 288	init_rwsem(&mapping->invalidate_lock);
 289	lockdep_set_class_and_name(&mapping->invalidate_lock,
 290				   &sb->s_type->invalidate_lock_key,
 291				   "mapping.invalidate_lock");
 292	if (sb->s_iflags & SB_I_STABLE_WRITES)
 293		mapping_set_stable_writes(mapping);
 
 
 
 
 
 
 294	inode->i_private = NULL;
 295	inode->i_mapping = mapping;
 296	INIT_HLIST_HEAD(&inode->i_dentry);	/* buggered by rcu freeing */
 297#ifdef CONFIG_FS_POSIX_ACL
 298	inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
 299#endif
 300
 301#ifdef CONFIG_FSNOTIFY
 302	inode->i_fsnotify_mask = 0;
 303#endif
 304	inode->i_flctx = NULL;
 305
 306	if (unlikely(security_inode_alloc(inode, gfp)))
 307		return -ENOMEM;
 308
 309	this_cpu_inc(nr_inodes);
 310
 311	return 0;
 
 
 312}
 313EXPORT_SYMBOL(inode_init_always_gfp);
 314
 315void free_inode_nonrcu(struct inode *inode)
 316{
 317	kmem_cache_free(inode_cachep, inode);
 318}
 319EXPORT_SYMBOL(free_inode_nonrcu);
 320
 321static void i_callback(struct rcu_head *head)
 322{
 323	struct inode *inode = container_of(head, struct inode, i_rcu);
 324	if (inode->free_inode)
 325		inode->free_inode(inode);
 326	else
 327		free_inode_nonrcu(inode);
 328}
 329
 330static struct inode *alloc_inode(struct super_block *sb)
 331{
 332	const struct super_operations *ops = sb->s_op;
 333	struct inode *inode;
 334
 335	if (ops->alloc_inode)
 336		inode = ops->alloc_inode(sb);
 337	else
 338		inode = alloc_inode_sb(sb, inode_cachep, GFP_KERNEL);
 339
 340	if (!inode)
 341		return NULL;
 342
 343	if (unlikely(inode_init_always(sb, inode))) {
 344		if (ops->destroy_inode) {
 345			ops->destroy_inode(inode);
 346			if (!ops->free_inode)
 347				return NULL;
 348		}
 349		inode->free_inode = ops->free_inode;
 350		i_callback(&inode->i_rcu);
 351		return NULL;
 352	}
 353
 354	return inode;
 355}
 356
 
 
 
 
 
 
 357void __destroy_inode(struct inode *inode)
 358{
 359	BUG_ON(inode_has_buffers(inode));
 360	inode_detach_wb(inode);
 361	security_inode_free(inode);
 362	fsnotify_inode_delete(inode);
 363	locks_free_lock_context(inode);
 364	if (!inode->i_nlink) {
 365		WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
 366		atomic_long_dec(&inode->i_sb->s_remove_count);
 367	}
 368
 369#ifdef CONFIG_FS_POSIX_ACL
 370	if (inode->i_acl && !is_uncached_acl(inode->i_acl))
 371		posix_acl_release(inode->i_acl);
 372	if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
 373		posix_acl_release(inode->i_default_acl);
 374#endif
 375	this_cpu_dec(nr_inodes);
 376}
 377EXPORT_SYMBOL(__destroy_inode);
 378
 379static void destroy_inode(struct inode *inode)
 380{
 381	const struct super_operations *ops = inode->i_sb->s_op;
 
 
 382
 
 
 383	BUG_ON(!list_empty(&inode->i_lru));
 384	__destroy_inode(inode);
 385	if (ops->destroy_inode) {
 386		ops->destroy_inode(inode);
 387		if (!ops->free_inode)
 388			return;
 389	}
 390	inode->free_inode = ops->free_inode;
 391	call_rcu(&inode->i_rcu, i_callback);
 392}
 393
 394/**
 395 * drop_nlink - directly drop an inode's link count
 396 * @inode: inode
 397 *
 398 * This is a low-level filesystem helper to replace any
 399 * direct filesystem manipulation of i_nlink.  In cases
 400 * where we are attempting to track writes to the
 401 * filesystem, a decrement to zero means an imminent
 402 * write when the file is truncated and actually unlinked
 403 * on the filesystem.
 404 */
 405void drop_nlink(struct inode *inode)
 406{
 407	WARN_ON(inode->i_nlink == 0);
 408	inode->__i_nlink--;
 409	if (!inode->i_nlink)
 410		atomic_long_inc(&inode->i_sb->s_remove_count);
 411}
 412EXPORT_SYMBOL(drop_nlink);
 413
 414/**
 415 * clear_nlink - directly zero an inode's link count
 416 * @inode: inode
 417 *
 418 * This is a low-level filesystem helper to replace any
 419 * direct filesystem manipulation of i_nlink.  See
 420 * drop_nlink() for why we care about i_nlink hitting zero.
 421 */
 422void clear_nlink(struct inode *inode)
 423{
 424	if (inode->i_nlink) {
 425		inode->__i_nlink = 0;
 426		atomic_long_inc(&inode->i_sb->s_remove_count);
 427	}
 428}
 429EXPORT_SYMBOL(clear_nlink);
 430
 431/**
 432 * set_nlink - directly set an inode's link count
 433 * @inode: inode
 434 * @nlink: new nlink (should be non-zero)
 435 *
 436 * This is a low-level filesystem helper to replace any
 437 * direct filesystem manipulation of i_nlink.
 438 */
 439void set_nlink(struct inode *inode, unsigned int nlink)
 440{
 441	if (!nlink) {
 442		clear_nlink(inode);
 443	} else {
 444		/* Yes, some filesystems do change nlink from zero to one */
 445		if (inode->i_nlink == 0)
 446			atomic_long_dec(&inode->i_sb->s_remove_count);
 447
 448		inode->__i_nlink = nlink;
 449	}
 450}
 451EXPORT_SYMBOL(set_nlink);
 452
 453/**
 454 * inc_nlink - directly increment an inode's link count
 455 * @inode: inode
 456 *
 457 * This is a low-level filesystem helper to replace any
 458 * direct filesystem manipulation of i_nlink.  Currently,
 459 * it is only here for parity with dec_nlink().
 460 */
 461void inc_nlink(struct inode *inode)
 462{
 463	if (unlikely(inode->i_nlink == 0)) {
 464		WARN_ON(!(inode->i_state & I_LINKABLE));
 465		atomic_long_dec(&inode->i_sb->s_remove_count);
 466	}
 467
 468	inode->__i_nlink++;
 469}
 470EXPORT_SYMBOL(inc_nlink);
 471
 472static void __address_space_init_once(struct address_space *mapping)
 473{
 474	xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
 475	init_rwsem(&mapping->i_mmap_rwsem);
 476	INIT_LIST_HEAD(&mapping->i_private_list);
 477	spin_lock_init(&mapping->i_private_lock);
 478	mapping->i_mmap = RB_ROOT_CACHED;
 479}
 480
 481void address_space_init_once(struct address_space *mapping)
 482{
 483	memset(mapping, 0, sizeof(*mapping));
 484	__address_space_init_once(mapping);
 
 
 
 
 
 
 485}
 486EXPORT_SYMBOL(address_space_init_once);
 487
 488/*
 489 * These are initializations that only need to be done
 490 * once, because the fields are idempotent across use
 491 * of the inode, so let the slab aware of that.
 492 */
 493void inode_init_once(struct inode *inode)
 494{
 495	memset(inode, 0, sizeof(*inode));
 496	INIT_HLIST_NODE(&inode->i_hash);
 497	INIT_LIST_HEAD(&inode->i_devices);
 498	INIT_LIST_HEAD(&inode->i_io_list);
 499	INIT_LIST_HEAD(&inode->i_wb_list);
 500	INIT_LIST_HEAD(&inode->i_lru);
 501	INIT_LIST_HEAD(&inode->i_sb_list);
 502	__address_space_init_once(&inode->i_data);
 503	i_size_ordered_init(inode);
 
 
 
 504}
 505EXPORT_SYMBOL(inode_init_once);
 506
 507static void init_once(void *foo)
 508{
 509	struct inode *inode = (struct inode *) foo;
 510
 511	inode_init_once(inode);
 512}
 513
 514/*
 
 
 
 
 
 
 
 
 515 * get additional reference to inode; caller must already hold one.
 516 */
 517void ihold(struct inode *inode)
 518{
 519	WARN_ON(atomic_inc_return(&inode->i_count) < 2);
 520}
 521EXPORT_SYMBOL(ihold);
 522
 523static void __inode_add_lru(struct inode *inode, bool rotate)
 524{
 525	if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
 526		return;
 527	if (atomic_read(&inode->i_count))
 528		return;
 529	if (!(inode->i_sb->s_flags & SB_ACTIVE))
 530		return;
 531	if (!mapping_shrinkable(&inode->i_data))
 532		return;
 533
 534	if (list_lru_add_obj(&inode->i_sb->s_inode_lru, &inode->i_lru))
 535		this_cpu_inc(nr_unused);
 536	else if (rotate)
 537		inode->i_state |= I_REFERENCED;
 538}
 539
 540struct wait_queue_head *inode_bit_waitqueue(struct wait_bit_queue_entry *wqe,
 541					    struct inode *inode, u32 bit)
 542{
 543        void *bit_address;
 544
 545        bit_address = inode_state_wait_address(inode, bit);
 546        init_wait_var_entry(wqe, bit_address, 0);
 547        return __var_waitqueue(bit_address);
 548}
 549EXPORT_SYMBOL(inode_bit_waitqueue);
 550
 551/*
 552 * Add inode to LRU if needed (inode is unused and clean).
 553 *
 554 * Needs inode->i_lock held.
 555 */
 556void inode_add_lru(struct inode *inode)
 557{
 558	__inode_add_lru(inode, false);
 
 
 559}
 560
 561static void inode_lru_list_del(struct inode *inode)
 562{
 563	if (list_lru_del_obj(&inode->i_sb->s_inode_lru, &inode->i_lru))
 564		this_cpu_dec(nr_unused);
 565}
 566
 567static void inode_pin_lru_isolating(struct inode *inode)
 568{
 569	lockdep_assert_held(&inode->i_lock);
 570	WARN_ON(inode->i_state & (I_LRU_ISOLATING | I_FREEING | I_WILL_FREE));
 571	inode->i_state |= I_LRU_ISOLATING;
 572}
 573
 574static void inode_unpin_lru_isolating(struct inode *inode)
 575{
 576	spin_lock(&inode->i_lock);
 577	WARN_ON(!(inode->i_state & I_LRU_ISOLATING));
 578	inode->i_state &= ~I_LRU_ISOLATING;
 579	/* Called with inode->i_lock which ensures memory ordering. */
 580	inode_wake_up_bit(inode, __I_LRU_ISOLATING);
 581	spin_unlock(&inode->i_lock);
 582}
 583
 584static void inode_wait_for_lru_isolating(struct inode *inode)
 585{
 586	struct wait_bit_queue_entry wqe;
 587	struct wait_queue_head *wq_head;
 588
 589	lockdep_assert_held(&inode->i_lock);
 590	if (!(inode->i_state & I_LRU_ISOLATING))
 591		return;
 592
 593	wq_head = inode_bit_waitqueue(&wqe, inode, __I_LRU_ISOLATING);
 594	for (;;) {
 595		prepare_to_wait_event(wq_head, &wqe.wq_entry, TASK_UNINTERRUPTIBLE);
 596		/*
 597		 * Checking I_LRU_ISOLATING with inode->i_lock guarantees
 598		 * memory ordering.
 599		 */
 600		if (!(inode->i_state & I_LRU_ISOLATING))
 601			break;
 602		spin_unlock(&inode->i_lock);
 603		schedule();
 604		spin_lock(&inode->i_lock);
 605	}
 606	finish_wait(wq_head, &wqe.wq_entry);
 607	WARN_ON(inode->i_state & I_LRU_ISOLATING);
 608}
 609
 610/**
 611 * inode_sb_list_add - add inode to the superblock list of inodes
 612 * @inode: inode to add
 613 */
 614void inode_sb_list_add(struct inode *inode)
 615{
 616	spin_lock(&inode->i_sb->s_inode_list_lock);
 617	list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
 618	spin_unlock(&inode->i_sb->s_inode_list_lock);
 619}
 620EXPORT_SYMBOL_GPL(inode_sb_list_add);
 621
 622static inline void inode_sb_list_del(struct inode *inode)
 623{
 624	if (!list_empty(&inode->i_sb_list)) {
 625		spin_lock(&inode->i_sb->s_inode_list_lock);
 626		list_del_init(&inode->i_sb_list);
 627		spin_unlock(&inode->i_sb->s_inode_list_lock);
 628	}
 629}
 630
 631static unsigned long hash(struct super_block *sb, unsigned long hashval)
 632{
 633	unsigned long tmp;
 634
 635	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
 636			L1_CACHE_BYTES;
 637	tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
 638	return tmp & i_hash_mask;
 639}
 640
 641/**
 642 *	__insert_inode_hash - hash an inode
 643 *	@inode: unhashed inode
 644 *	@hashval: unsigned long value used to locate this object in the
 645 *		inode_hashtable.
 646 *
 647 *	Add an inode to the inode hash for this superblock.
 648 */
 649void __insert_inode_hash(struct inode *inode, unsigned long hashval)
 650{
 651	struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
 652
 653	spin_lock(&inode_hash_lock);
 654	spin_lock(&inode->i_lock);
 655	hlist_add_head_rcu(&inode->i_hash, b);
 656	spin_unlock(&inode->i_lock);
 657	spin_unlock(&inode_hash_lock);
 658}
 659EXPORT_SYMBOL(__insert_inode_hash);
 660
 661/**
 662 *	__remove_inode_hash - remove an inode from the hash
 663 *	@inode: inode to unhash
 664 *
 665 *	Remove an inode from the superblock.
 666 */
 667void __remove_inode_hash(struct inode *inode)
 668{
 669	spin_lock(&inode_hash_lock);
 670	spin_lock(&inode->i_lock);
 671	hlist_del_init_rcu(&inode->i_hash);
 672	spin_unlock(&inode->i_lock);
 673	spin_unlock(&inode_hash_lock);
 674}
 675EXPORT_SYMBOL(__remove_inode_hash);
 676
 677void dump_mapping(const struct address_space *mapping)
 678{
 679	struct inode *host;
 680	const struct address_space_operations *a_ops;
 681	struct hlist_node *dentry_first;
 682	struct dentry *dentry_ptr;
 683	struct dentry dentry;
 684	char fname[64] = {};
 685	unsigned long ino;
 686
 687	/*
 688	 * If mapping is an invalid pointer, we don't want to crash
 689	 * accessing it, so probe everything depending on it carefully.
 690	 */
 691	if (get_kernel_nofault(host, &mapping->host) ||
 692	    get_kernel_nofault(a_ops, &mapping->a_ops)) {
 693		pr_warn("invalid mapping:%px\n", mapping);
 694		return;
 695	}
 696
 697	if (!host) {
 698		pr_warn("aops:%ps\n", a_ops);
 699		return;
 700	}
 701
 702	if (get_kernel_nofault(dentry_first, &host->i_dentry.first) ||
 703	    get_kernel_nofault(ino, &host->i_ino)) {
 704		pr_warn("aops:%ps invalid inode:%px\n", a_ops, host);
 705		return;
 706	}
 707
 708	if (!dentry_first) {
 709		pr_warn("aops:%ps ino:%lx\n", a_ops, ino);
 710		return;
 711	}
 712
 713	dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
 714	if (get_kernel_nofault(dentry, dentry_ptr) ||
 715	    !dentry.d_parent || !dentry.d_name.name) {
 716		pr_warn("aops:%ps ino:%lx invalid dentry:%px\n",
 717				a_ops, ino, dentry_ptr);
 718		return;
 719	}
 720
 721	if (strncpy_from_kernel_nofault(fname, dentry.d_name.name, 63) < 0)
 722		strscpy(fname, "<invalid>");
 723	/*
 724	 * Even if strncpy_from_kernel_nofault() succeeded,
 725	 * the fname could be unreliable
 726	 */
 727	pr_warn("aops:%ps ino:%lx dentry name(?):\"%s\"\n",
 728		a_ops, ino, fname);
 729}
 730
 731void clear_inode(struct inode *inode)
 732{
 
 733	/*
 734	 * We have to cycle the i_pages lock here because reclaim can be in the
 735	 * process of removing the last page (in __filemap_remove_folio())
 736	 * and we must not free the mapping under it.
 737	 */
 738	xa_lock_irq(&inode->i_data.i_pages);
 739	BUG_ON(inode->i_data.nrpages);
 740	/*
 741	 * Almost always, mapping_empty(&inode->i_data) here; but there are
 742	 * two known and long-standing ways in which nodes may get left behind
 743	 * (when deep radix-tree node allocation failed partway; or when THP
 744	 * collapse_file() failed). Until those two known cases are cleaned up,
 745	 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
 746	 * nor even WARN_ON(!mapping_empty).
 747	 */
 748	xa_unlock_irq(&inode->i_data.i_pages);
 749	BUG_ON(!list_empty(&inode->i_data.i_private_list));
 750	BUG_ON(!(inode->i_state & I_FREEING));
 751	BUG_ON(inode->i_state & I_CLEAR);
 752	BUG_ON(!list_empty(&inode->i_wb_list));
 753	/* don't need i_lock here, no concurrent mods to i_state */
 754	inode->i_state = I_FREEING | I_CLEAR;
 755}
 756EXPORT_SYMBOL(clear_inode);
 757
 758/*
 759 * Free the inode passed in, removing it from the lists it is still connected
 760 * to. We remove any pages still attached to the inode and wait for any IO that
 761 * is still in progress before finally destroying the inode.
 762 *
 763 * An inode must already be marked I_FREEING so that we avoid the inode being
 764 * moved back onto lists if we race with other code that manipulates the lists
 765 * (e.g. writeback_single_inode). The caller is responsible for setting this.
 766 *
 767 * An inode must already be removed from the LRU list before being evicted from
 768 * the cache. This should occur atomically with setting the I_FREEING state
 769 * flag, so no inodes here should ever be on the LRU when being evicted.
 770 */
 771static void evict(struct inode *inode)
 772{
 773	const struct super_operations *op = inode->i_sb->s_op;
 774
 775	BUG_ON(!(inode->i_state & I_FREEING));
 776	BUG_ON(!list_empty(&inode->i_lru));
 777
 778	if (!list_empty(&inode->i_io_list))
 779		inode_io_list_del(inode);
 780
 781	inode_sb_list_del(inode);
 782
 783	spin_lock(&inode->i_lock);
 784	inode_wait_for_lru_isolating(inode);
 785
 786	/*
 787	 * Wait for flusher thread to be done with the inode so that filesystem
 788	 * does not start destroying it while writeback is still running. Since
 789	 * the inode has I_FREEING set, flusher thread won't start new work on
 790	 * the inode.  We just have to wait for running writeback to finish.
 791	 */
 792	inode_wait_for_writeback(inode);
 793	spin_unlock(&inode->i_lock);
 794
 795	if (op->evict_inode) {
 796		op->evict_inode(inode);
 797	} else {
 798		truncate_inode_pages_final(&inode->i_data);
 799		clear_inode(inode);
 800	}
 
 
 801	if (S_ISCHR(inode->i_mode) && inode->i_cdev)
 802		cd_forget(inode);
 803
 804	remove_inode_hash(inode);
 805
 806	/*
 807	 * Wake up waiters in __wait_on_freeing_inode().
 808	 *
 809	 * Lockless hash lookup may end up finding the inode before we removed
 810	 * it above, but only lock it *after* we are done with the wakeup below.
 811	 * In this case the potential waiter cannot safely block.
 812	 *
 813	 * The inode being unhashed after the call to remove_inode_hash() is
 814	 * used as an indicator whether blocking on it is safe.
 815	 */
 816	spin_lock(&inode->i_lock);
 817	/*
 818	 * Pairs with the barrier in prepare_to_wait_event() to make sure
 819	 * ___wait_var_event() either sees the bit cleared or
 820	 * waitqueue_active() check in wake_up_var() sees the waiter.
 821	 */
 822	smp_mb__after_spinlock();
 823	inode_wake_up_bit(inode, __I_NEW);
 824	BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
 825	spin_unlock(&inode->i_lock);
 826
 827	destroy_inode(inode);
 828}
 829
 830/*
 831 * dispose_list - dispose of the contents of a local list
 832 * @head: the head of the list to free
 833 *
 834 * Dispose-list gets a local list with local inodes in it, so it doesn't
 835 * need to worry about list corruption and SMP locks.
 836 */
 837static void dispose_list(struct list_head *head)
 838{
 839	while (!list_empty(head)) {
 840		struct inode *inode;
 841
 842		inode = list_first_entry(head, struct inode, i_lru);
 843		list_del_init(&inode->i_lru);
 844
 845		evict(inode);
 846		cond_resched();
 847	}
 848}
 849
 850/**
 851 * evict_inodes	- evict all evictable inodes for a superblock
 852 * @sb:		superblock to operate on
 853 *
 854 * Make sure that no inodes with zero refcount are retained.  This is
 855 * called by superblock shutdown after having SB_ACTIVE flag removed,
 856 * so any inode reaching zero refcount during or after that call will
 857 * be immediately evicted.
 858 */
 859void evict_inodes(struct super_block *sb)
 860{
 861	struct inode *inode, *next;
 862	LIST_HEAD(dispose);
 863
 864again:
 865	spin_lock(&sb->s_inode_list_lock);
 866	list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 867		if (atomic_read(&inode->i_count))
 868			continue;
 869
 870		spin_lock(&inode->i_lock);
 871		if (atomic_read(&inode->i_count)) {
 872			spin_unlock(&inode->i_lock);
 873			continue;
 874		}
 875		if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 876			spin_unlock(&inode->i_lock);
 877			continue;
 878		}
 879
 880		inode->i_state |= I_FREEING;
 881		inode_lru_list_del(inode);
 882		spin_unlock(&inode->i_lock);
 883		list_add(&inode->i_lru, &dispose);
 884
 885		/*
 886		 * We can have a ton of inodes to evict at unmount time given
 887		 * enough memory, check to see if we need to go to sleep for a
 888		 * bit so we don't livelock.
 889		 */
 890		if (need_resched()) {
 891			spin_unlock(&sb->s_inode_list_lock);
 892			cond_resched();
 893			dispose_list(&dispose);
 894			goto again;
 895		}
 896	}
 897	spin_unlock(&sb->s_inode_list_lock);
 898
 899	dispose_list(&dispose);
 900}
 901EXPORT_SYMBOL_GPL(evict_inodes);
 902
 903/**
 904 * invalidate_inodes	- attempt to free all inodes on a superblock
 905 * @sb:		superblock to operate on
 
 906 *
 907 * Attempts to free all inodes (including dirty inodes) for a given superblock.
 
 
 
 908 */
 909void invalidate_inodes(struct super_block *sb)
 910{
 
 911	struct inode *inode, *next;
 912	LIST_HEAD(dispose);
 913
 914again:
 915	spin_lock(&sb->s_inode_list_lock);
 916	list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 917		spin_lock(&inode->i_lock);
 918		if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 919			spin_unlock(&inode->i_lock);
 920			continue;
 921		}
 
 
 
 
 
 922		if (atomic_read(&inode->i_count)) {
 923			spin_unlock(&inode->i_lock);
 
 924			continue;
 925		}
 926
 927		inode->i_state |= I_FREEING;
 928		inode_lru_list_del(inode);
 929		spin_unlock(&inode->i_lock);
 930		list_add(&inode->i_lru, &dispose);
 931		if (need_resched()) {
 932			spin_unlock(&sb->s_inode_list_lock);
 933			cond_resched();
 934			dispose_list(&dispose);
 935			goto again;
 936		}
 937	}
 938	spin_unlock(&sb->s_inode_list_lock);
 939
 940	dispose_list(&dispose);
 
 
 941}
 942
 943/*
 944 * Isolate the inode from the LRU in preparation for freeing it.
 945 *
 
 
 
 
 946 * If the inode has the I_REFERENCED flag set, then it means that it has been
 947 * used recently - the flag is set in iput_final(). When we encounter such an
 948 * inode, clear the flag and move it to the back of the LRU so it gets another
 949 * pass through the LRU before it gets reclaimed. This is necessary because of
 950 * the fact we are doing lazy LRU updates to minimise lock contention so the
 951 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
 952 * with this flag set because they are the inodes that are out of order.
 953 */
 954static enum lru_status inode_lru_isolate(struct list_head *item,
 955		struct list_lru_one *lru, void *arg)
 956{
 957	struct list_head *freeable = arg;
 958	struct inode	*inode = container_of(item, struct inode, i_lru);
 959
 960	/*
 961	 * We are inverting the lru lock/inode->i_lock here, so use a
 962	 * trylock. If we fail to get the lock, just skip it.
 963	 */
 964	if (!spin_trylock(&inode->i_lock))
 965		return LRU_SKIP;
 966
 967	/*
 968	 * Inodes can get referenced, redirtied, or repopulated while
 969	 * they're already on the LRU, and this can make them
 970	 * unreclaimable for a while. Remove them lazily here; iput,
 971	 * sync, or the last page cache deletion will requeue them.
 972	 */
 973	if (atomic_read(&inode->i_count) ||
 974	    (inode->i_state & ~I_REFERENCED) ||
 975	    !mapping_shrinkable(&inode->i_data)) {
 976		list_lru_isolate(lru, &inode->i_lru);
 977		spin_unlock(&inode->i_lock);
 978		this_cpu_dec(nr_unused);
 979		return LRU_REMOVED;
 980	}
 981
 982	/* Recently referenced inodes get one more pass */
 983	if (inode->i_state & I_REFERENCED) {
 984		inode->i_state &= ~I_REFERENCED;
 985		spin_unlock(&inode->i_lock);
 986		return LRU_ROTATE;
 987	}
 988
 989	/*
 990	 * On highmem systems, mapping_shrinkable() permits dropping
 991	 * page cache in order to free up struct inodes: lowmem might
 992	 * be under pressure before the cache inside the highmem zone.
 993	 */
 994	if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
 995		inode_pin_lru_isolating(inode);
 996		spin_unlock(&inode->i_lock);
 997		spin_unlock(&lru->lock);
 998		if (remove_inode_buffers(inode)) {
 999			unsigned long reap;
1000			reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
1001			if (current_is_kswapd())
1002				__count_vm_events(KSWAPD_INODESTEAL, reap);
1003			else
1004				__count_vm_events(PGINODESTEAL, reap);
1005			mm_account_reclaimed_pages(reap);
 
1006		}
1007		inode_unpin_lru_isolating(inode);
 
1008		return LRU_RETRY;
1009	}
1010
1011	WARN_ON(inode->i_state & I_NEW);
1012	inode->i_state |= I_FREEING;
1013	list_lru_isolate_move(lru, &inode->i_lru, freeable);
1014	spin_unlock(&inode->i_lock);
1015
1016	this_cpu_dec(nr_unused);
1017	return LRU_REMOVED;
1018}
1019
1020/*
1021 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
1022 * This is called from the superblock shrinker function with a number of inodes
1023 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
1024 * then are freed outside inode_lock by dispose_list().
1025 */
1026long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
 
1027{
1028	LIST_HEAD(freeable);
1029	long freed;
1030
1031	freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
1032				     inode_lru_isolate, &freeable);
1033	dispose_list(&freeable);
1034	return freed;
1035}
1036
1037static void __wait_on_freeing_inode(struct inode *inode, bool is_inode_hash_locked);
1038/*
1039 * Called with the inode lock held.
1040 */
1041static struct inode *find_inode(struct super_block *sb,
1042				struct hlist_head *head,
1043				int (*test)(struct inode *, void *),
1044				void *data, bool is_inode_hash_locked)
1045{
1046	struct inode *inode = NULL;
1047
1048	if (is_inode_hash_locked)
1049		lockdep_assert_held(&inode_hash_lock);
1050	else
1051		lockdep_assert_not_held(&inode_hash_lock);
1052
1053	rcu_read_lock();
1054repeat:
1055	hlist_for_each_entry_rcu(inode, head, i_hash) {
1056		if (inode->i_sb != sb)
1057			continue;
1058		if (!test(inode, data))
1059			continue;
1060		spin_lock(&inode->i_lock);
1061		if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1062			__wait_on_freeing_inode(inode, is_inode_hash_locked);
1063			goto repeat;
1064		}
1065		if (unlikely(inode->i_state & I_CREATING)) {
1066			spin_unlock(&inode->i_lock);
1067			rcu_read_unlock();
1068			return ERR_PTR(-ESTALE);
1069		}
1070		__iget(inode);
1071		spin_unlock(&inode->i_lock);
1072		rcu_read_unlock();
1073		return inode;
1074	}
1075	rcu_read_unlock();
1076	return NULL;
1077}
1078
1079/*
1080 * find_inode_fast is the fast path version of find_inode, see the comment at
1081 * iget_locked for details.
1082 */
1083static struct inode *find_inode_fast(struct super_block *sb,
1084				struct hlist_head *head, unsigned long ino,
1085				bool is_inode_hash_locked)
1086{
1087	struct inode *inode = NULL;
1088
1089	if (is_inode_hash_locked)
1090		lockdep_assert_held(&inode_hash_lock);
1091	else
1092		lockdep_assert_not_held(&inode_hash_lock);
1093
1094	rcu_read_lock();
1095repeat:
1096	hlist_for_each_entry_rcu(inode, head, i_hash) {
1097		if (inode->i_ino != ino)
1098			continue;
1099		if (inode->i_sb != sb)
1100			continue;
1101		spin_lock(&inode->i_lock);
1102		if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1103			__wait_on_freeing_inode(inode, is_inode_hash_locked);
1104			goto repeat;
1105		}
1106		if (unlikely(inode->i_state & I_CREATING)) {
1107			spin_unlock(&inode->i_lock);
1108			rcu_read_unlock();
1109			return ERR_PTR(-ESTALE);
1110		}
1111		__iget(inode);
1112		spin_unlock(&inode->i_lock);
1113		rcu_read_unlock();
1114		return inode;
1115	}
1116	rcu_read_unlock();
1117	return NULL;
1118}
1119
1120/*
1121 * Each cpu owns a range of LAST_INO_BATCH numbers.
1122 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
1123 * to renew the exhausted range.
1124 *
1125 * This does not significantly increase overflow rate because every CPU can
1126 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
1127 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
1128 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
1129 * overflow rate by 2x, which does not seem too significant.
1130 *
1131 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1132 * error if st_ino won't fit in target struct field. Use 32bit counter
1133 * here to attempt to avoid that.
1134 */
1135#define LAST_INO_BATCH 1024
1136static DEFINE_PER_CPU(unsigned int, last_ino);
1137
1138unsigned int get_next_ino(void)
1139{
1140	unsigned int *p = &get_cpu_var(last_ino);
1141	unsigned int res = *p;
1142
1143#ifdef CONFIG_SMP
1144	if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
1145		static atomic_t shared_last_ino;
1146		int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
1147
1148		res = next - LAST_INO_BATCH;
1149	}
1150#endif
1151
1152	res++;
1153	/* get_next_ino should not provide a 0 inode number */
1154	if (unlikely(!res))
1155		res++;
1156	*p = res;
1157	put_cpu_var(last_ino);
1158	return res;
1159}
1160EXPORT_SYMBOL(get_next_ino);
1161
1162/**
1163 *	new_inode_pseudo 	- obtain an inode
1164 *	@sb: superblock
1165 *
1166 *	Allocates a new inode for given superblock.
1167 *	Inode wont be chained in superblock s_inodes list
1168 *	This means :
1169 *	- fs can't be unmount
1170 *	- quotas, fsnotify, writeback can't work
1171 */
1172struct inode *new_inode_pseudo(struct super_block *sb)
1173{
1174	return alloc_inode(sb);
 
 
 
 
 
 
 
 
1175}
1176
1177/**
1178 *	new_inode 	- obtain an inode
1179 *	@sb: superblock
1180 *
1181 *	Allocates a new inode for given superblock. The default gfp_mask
1182 *	for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1183 *	If HIGHMEM pages are unsuitable or it is known that pages allocated
1184 *	for the page cache are not reclaimable or migratable,
1185 *	mapping_set_gfp_mask() must be called with suitable flags on the
1186 *	newly created inode's mapping
1187 *
1188 */
1189struct inode *new_inode(struct super_block *sb)
1190{
1191	struct inode *inode;
1192
 
 
1193	inode = new_inode_pseudo(sb);
1194	if (inode)
1195		inode_sb_list_add(inode);
1196	return inode;
1197}
1198EXPORT_SYMBOL(new_inode);
1199
1200#ifdef CONFIG_DEBUG_LOCK_ALLOC
1201void lockdep_annotate_inode_mutex_key(struct inode *inode)
1202{
1203	if (S_ISDIR(inode->i_mode)) {
1204		struct file_system_type *type = inode->i_sb->s_type;
1205
1206		/* Set new key only if filesystem hasn't already changed it */
1207		if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
1208			/*
1209			 * ensure nobody is actually holding i_mutex
1210			 */
1211			// mutex_destroy(&inode->i_mutex);
1212			init_rwsem(&inode->i_rwsem);
1213			lockdep_set_class(&inode->i_rwsem,
1214					  &type->i_mutex_dir_key);
1215		}
1216	}
1217}
1218EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1219#endif
1220
1221/**
1222 * unlock_new_inode - clear the I_NEW state and wake up any waiters
1223 * @inode:	new inode to unlock
1224 *
1225 * Called when the inode is fully initialised to clear the new state of the
1226 * inode and wake up anyone waiting for the inode to finish initialisation.
1227 */
1228void unlock_new_inode(struct inode *inode)
1229{
1230	lockdep_annotate_inode_mutex_key(inode);
1231	spin_lock(&inode->i_lock);
1232	WARN_ON(!(inode->i_state & I_NEW));
1233	inode->i_state &= ~I_NEW & ~I_CREATING;
1234	/*
1235	 * Pairs with the barrier in prepare_to_wait_event() to make sure
1236	 * ___wait_var_event() either sees the bit cleared or
1237	 * waitqueue_active() check in wake_up_var() sees the waiter.
1238	 */
1239	smp_mb();
1240	inode_wake_up_bit(inode, __I_NEW);
1241	spin_unlock(&inode->i_lock);
1242}
1243EXPORT_SYMBOL(unlock_new_inode);
1244
1245void discard_new_inode(struct inode *inode)
1246{
1247	lockdep_annotate_inode_mutex_key(inode);
1248	spin_lock(&inode->i_lock);
1249	WARN_ON(!(inode->i_state & I_NEW));
1250	inode->i_state &= ~I_NEW;
1251	/*
1252	 * Pairs with the barrier in prepare_to_wait_event() to make sure
1253	 * ___wait_var_event() either sees the bit cleared or
1254	 * waitqueue_active() check in wake_up_var() sees the waiter.
1255	 */
1256	smp_mb();
1257	inode_wake_up_bit(inode, __I_NEW);
1258	spin_unlock(&inode->i_lock);
1259	iput(inode);
1260}
1261EXPORT_SYMBOL(discard_new_inode);
1262
1263/**
1264 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1265 *
1266 * Lock any non-NULL argument. Passed objects must not be directories.
1267 * Zero, one or two objects may be locked by this function.
1268 *
1269 * @inode1: first inode to lock
1270 * @inode2: second inode to lock
1271 */
1272void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1273{
1274	if (inode1)
1275		WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1276	if (inode2)
1277		WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1278	if (inode1 > inode2)
1279		swap(inode1, inode2);
1280	if (inode1)
1281		inode_lock(inode1);
1282	if (inode2 && inode2 != inode1)
1283		inode_lock_nested(inode2, I_MUTEX_NONDIR2);
 
1284}
1285EXPORT_SYMBOL(lock_two_nondirectories);
1286
1287/**
1288 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1289 * @inode1: first inode to unlock
1290 * @inode2: second inode to unlock
1291 */
1292void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1293{
1294	if (inode1) {
1295		WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1296		inode_unlock(inode1);
1297	}
1298	if (inode2 && inode2 != inode1) {
1299		WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1300		inode_unlock(inode2);
1301	}
1302}
1303EXPORT_SYMBOL(unlock_two_nondirectories);
1304
1305/**
1306 * inode_insert5 - obtain an inode from a mounted file system
1307 * @inode:	pre-allocated inode to use for insert to cache
1308 * @hashval:	hash value (usually inode number) to get
1309 * @test:	callback used for comparisons between inodes
1310 * @set:	callback used to initialize a new struct inode
1311 * @data:	opaque data pointer to pass to @test and @set
1312 *
1313 * Search for the inode specified by @hashval and @data in the inode cache,
1314 * and if present return it with an increased reference count. This is a
1315 * variant of iget5_locked() that doesn't allocate an inode.
1316 *
1317 * If the inode is not present in the cache, insert the pre-allocated inode and
1318 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1319 * to fill it in before unlocking it via unlock_new_inode().
1320 *
1321 * Note that both @test and @set are called with the inode_hash_lock held, so
1322 * they can't sleep.
1323 */
1324struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1325			    int (*test)(struct inode *, void *),
1326			    int (*set)(struct inode *, void *), void *data)
1327{
1328	struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1329	struct inode *old;
1330
1331again:
1332	spin_lock(&inode_hash_lock);
1333	old = find_inode(inode->i_sb, head, test, data, true);
1334	if (unlikely(old)) {
1335		/*
1336		 * Uhhuh, somebody else created the same inode under us.
1337		 * Use the old inode instead of the preallocated one.
1338		 */
1339		spin_unlock(&inode_hash_lock);
1340		if (IS_ERR(old))
1341			return NULL;
1342		wait_on_inode(old);
1343		if (unlikely(inode_unhashed(old))) {
1344			iput(old);
1345			goto again;
1346		}
1347		return old;
1348	}
1349
1350	if (set && unlikely(set(inode, data))) {
1351		inode = NULL;
1352		goto unlock;
1353	}
1354
1355	/*
1356	 * Return the locked inode with I_NEW set, the
1357	 * caller is responsible for filling in the contents
1358	 */
1359	spin_lock(&inode->i_lock);
1360	inode->i_state |= I_NEW;
1361	hlist_add_head_rcu(&inode->i_hash, head);
1362	spin_unlock(&inode->i_lock);
1363
1364	/*
1365	 * Add inode to the sb list if it's not already. It has I_NEW at this
1366	 * point, so it should be safe to test i_sb_list locklessly.
1367	 */
1368	if (list_empty(&inode->i_sb_list))
1369		inode_sb_list_add(inode);
1370unlock:
1371	spin_unlock(&inode_hash_lock);
1372
1373	return inode;
1374}
1375EXPORT_SYMBOL(inode_insert5);
1376
1377/**
1378 * iget5_locked - obtain an inode from a mounted file system
1379 * @sb:		super block of file system
1380 * @hashval:	hash value (usually inode number) to get
1381 * @test:	callback used for comparisons between inodes
1382 * @set:	callback used to initialize a new struct inode
1383 * @data:	opaque data pointer to pass to @test and @set
1384 *
1385 * Search for the inode specified by @hashval and @data in the inode cache,
1386 * and if present return it with an increased reference count. This is a
1387 * generalized version of iget_locked() for file systems where the inode
1388 * number is not sufficient for unique identification of an inode.
1389 *
1390 * If the inode is not present in the cache, allocate and insert a new inode
1391 * and return it locked, hashed, and with the I_NEW flag set. The file system
1392 * gets to fill it in before unlocking it via unlock_new_inode().
1393 *
1394 * Note that both @test and @set are called with the inode_hash_lock held, so
1395 * they can't sleep.
1396 */
1397struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1398		int (*test)(struct inode *, void *),
1399		int (*set)(struct inode *, void *), void *data)
1400{
1401	struct inode *inode = ilookup5(sb, hashval, test, data);
1402
1403	if (!inode) {
1404		struct inode *new = alloc_inode(sb);
1405
1406		if (new) {
1407			inode = inode_insert5(new, hashval, test, set, data);
1408			if (unlikely(inode != new))
1409				destroy_inode(new);
1410		}
1411	}
1412	return inode;
1413}
1414EXPORT_SYMBOL(iget5_locked);
1415
1416/**
1417 * iget5_locked_rcu - obtain an inode from a mounted file system
1418 * @sb:		super block of file system
1419 * @hashval:	hash value (usually inode number) to get
1420 * @test:	callback used for comparisons between inodes
1421 * @set:	callback used to initialize a new struct inode
1422 * @data:	opaque data pointer to pass to @test and @set
1423 *
1424 * This is equivalent to iget5_locked, except the @test callback must
1425 * tolerate the inode not being stable, including being mid-teardown.
1426 */
1427struct inode *iget5_locked_rcu(struct super_block *sb, unsigned long hashval,
1428		int (*test)(struct inode *, void *),
1429		int (*set)(struct inode *, void *), void *data)
1430{
1431	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1432	struct inode *inode, *new;
 
 
 
 
1433
1434again:
1435	inode = find_inode(sb, head, test, data, false);
1436	if (inode) {
1437		if (IS_ERR(inode))
1438			return NULL;
1439		wait_on_inode(inode);
1440		if (unlikely(inode_unhashed(inode))) {
1441			iput(inode);
1442			goto again;
1443		}
1444		return inode;
1445	}
1446
1447	new = alloc_inode(sb);
1448	if (new) {
1449		inode = inode_insert5(new, hashval, test, set, data);
1450		if (unlikely(inode != new))
1451			destroy_inode(new);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1452	}
1453	return inode;
 
 
 
 
 
1454}
1455EXPORT_SYMBOL_GPL(iget5_locked_rcu);
1456
1457/**
1458 * iget_locked - obtain an inode from a mounted file system
1459 * @sb:		super block of file system
1460 * @ino:	inode number to get
1461 *
1462 * Search for the inode specified by @ino in the inode cache and if present
1463 * return it with an increased reference count. This is for file systems
1464 * where the inode number is sufficient for unique identification of an inode.
1465 *
1466 * If the inode is not in cache, allocate a new inode and return it locked,
1467 * hashed, and with the I_NEW flag set.  The file system gets to fill it in
1468 * before unlocking it via unlock_new_inode().
1469 */
1470struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1471{
1472	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1473	struct inode *inode;
1474again:
1475	inode = find_inode_fast(sb, head, ino, false);
 
 
1476	if (inode) {
1477		if (IS_ERR(inode))
1478			return NULL;
1479		wait_on_inode(inode);
1480		if (unlikely(inode_unhashed(inode))) {
1481			iput(inode);
1482			goto again;
1483		}
1484		return inode;
1485	}
1486
1487	inode = alloc_inode(sb);
1488	if (inode) {
1489		struct inode *old;
1490
1491		spin_lock(&inode_hash_lock);
1492		/* We released the lock, so.. */
1493		old = find_inode_fast(sb, head, ino, true);
1494		if (!old) {
1495			inode->i_ino = ino;
1496			spin_lock(&inode->i_lock);
1497			inode->i_state = I_NEW;
1498			hlist_add_head_rcu(&inode->i_hash, head);
1499			spin_unlock(&inode->i_lock);
1500			inode_sb_list_add(inode);
1501			spin_unlock(&inode_hash_lock);
1502
1503			/* Return the locked inode with I_NEW set, the
1504			 * caller is responsible for filling in the contents
1505			 */
1506			return inode;
1507		}
1508
1509		/*
1510		 * Uhhuh, somebody else created the same inode under
1511		 * us. Use the old inode instead of the one we just
1512		 * allocated.
1513		 */
1514		spin_unlock(&inode_hash_lock);
1515		destroy_inode(inode);
1516		if (IS_ERR(old))
1517			return NULL;
1518		inode = old;
1519		wait_on_inode(inode);
1520		if (unlikely(inode_unhashed(inode))) {
1521			iput(inode);
1522			goto again;
1523		}
1524	}
1525	return inode;
1526}
1527EXPORT_SYMBOL(iget_locked);
1528
1529/*
1530 * search the inode cache for a matching inode number.
1531 * If we find one, then the inode number we are trying to
1532 * allocate is not unique and so we should not use it.
1533 *
1534 * Returns 1 if the inode number is unique, 0 if it is not.
1535 */
1536static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1537{
1538	struct hlist_head *b = inode_hashtable + hash(sb, ino);
1539	struct inode *inode;
1540
1541	hlist_for_each_entry_rcu(inode, b, i_hash) {
1542		if (inode->i_ino == ino && inode->i_sb == sb)
 
 
1543			return 0;
 
1544	}
 
 
1545	return 1;
1546}
1547
1548/**
1549 *	iunique - get a unique inode number
1550 *	@sb: superblock
1551 *	@max_reserved: highest reserved inode number
1552 *
1553 *	Obtain an inode number that is unique on the system for a given
1554 *	superblock. This is used by file systems that have no natural
1555 *	permanent inode numbering system. An inode number is returned that
1556 *	is higher than the reserved limit but unique.
1557 *
1558 *	BUGS:
1559 *	With a large number of inodes live on the file system this function
1560 *	currently becomes quite slow.
1561 */
1562ino_t iunique(struct super_block *sb, ino_t max_reserved)
1563{
1564	/*
1565	 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1566	 * error if st_ino won't fit in target struct field. Use 32bit counter
1567	 * here to attempt to avoid that.
1568	 */
1569	static DEFINE_SPINLOCK(iunique_lock);
1570	static unsigned int counter;
1571	ino_t res;
1572
1573	rcu_read_lock();
1574	spin_lock(&iunique_lock);
1575	do {
1576		if (counter <= max_reserved)
1577			counter = max_reserved + 1;
1578		res = counter++;
1579	} while (!test_inode_iunique(sb, res));
1580	spin_unlock(&iunique_lock);
1581	rcu_read_unlock();
1582
1583	return res;
1584}
1585EXPORT_SYMBOL(iunique);
1586
1587struct inode *igrab(struct inode *inode)
1588{
1589	spin_lock(&inode->i_lock);
1590	if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1591		__iget(inode);
1592		spin_unlock(&inode->i_lock);
1593	} else {
1594		spin_unlock(&inode->i_lock);
1595		/*
1596		 * Handle the case where s_op->clear_inode is not been
1597		 * called yet, and somebody is calling igrab
1598		 * while the inode is getting freed.
1599		 */
1600		inode = NULL;
1601	}
1602	return inode;
1603}
1604EXPORT_SYMBOL(igrab);
1605
1606/**
1607 * ilookup5_nowait - search for an inode in the inode cache
1608 * @sb:		super block of file system to search
1609 * @hashval:	hash value (usually inode number) to search for
1610 * @test:	callback used for comparisons between inodes
1611 * @data:	opaque data pointer to pass to @test
1612 *
1613 * Search for the inode specified by @hashval and @data in the inode cache.
1614 * If the inode is in the cache, the inode is returned with an incremented
1615 * reference count.
1616 *
1617 * Note: I_NEW is not waited upon so you have to be very careful what you do
1618 * with the returned inode.  You probably should be using ilookup5() instead.
1619 *
1620 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1621 */
1622struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1623		int (*test)(struct inode *, void *), void *data)
1624{
1625	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1626	struct inode *inode;
1627
1628	spin_lock(&inode_hash_lock);
1629	inode = find_inode(sb, head, test, data, true);
1630	spin_unlock(&inode_hash_lock);
1631
1632	return IS_ERR(inode) ? NULL : inode;
1633}
1634EXPORT_SYMBOL(ilookup5_nowait);
1635
1636/**
1637 * ilookup5 - search for an inode in the inode cache
1638 * @sb:		super block of file system to search
1639 * @hashval:	hash value (usually inode number) to search for
1640 * @test:	callback used for comparisons between inodes
1641 * @data:	opaque data pointer to pass to @test
1642 *
1643 * Search for the inode specified by @hashval and @data in the inode cache,
1644 * and if the inode is in the cache, return the inode with an incremented
1645 * reference count.  Waits on I_NEW before returning the inode.
1646 * returned with an incremented reference count.
1647 *
1648 * This is a generalized version of ilookup() for file systems where the
1649 * inode number is not sufficient for unique identification of an inode.
1650 *
1651 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1652 */
1653struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1654		int (*test)(struct inode *, void *), void *data)
1655{
1656	struct inode *inode;
1657again:
1658	inode = ilookup5_nowait(sb, hashval, test, data);
1659	if (inode) {
1660		wait_on_inode(inode);
1661		if (unlikely(inode_unhashed(inode))) {
1662			iput(inode);
1663			goto again;
1664		}
1665	}
1666	return inode;
1667}
1668EXPORT_SYMBOL(ilookup5);
1669
1670/**
1671 * ilookup - search for an inode in the inode cache
1672 * @sb:		super block of file system to search
1673 * @ino:	inode number to search for
1674 *
1675 * Search for the inode @ino in the inode cache, and if the inode is in the
1676 * cache, the inode is returned with an incremented reference count.
1677 */
1678struct inode *ilookup(struct super_block *sb, unsigned long ino)
1679{
1680	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1681	struct inode *inode;
1682again:
1683	inode = find_inode_fast(sb, head, ino, false);
1684
1685	if (inode) {
1686		if (IS_ERR(inode))
1687			return NULL;
1688		wait_on_inode(inode);
1689		if (unlikely(inode_unhashed(inode))) {
1690			iput(inode);
1691			goto again;
1692		}
1693	}
1694	return inode;
1695}
1696EXPORT_SYMBOL(ilookup);
1697
1698/**
1699 * find_inode_nowait - find an inode in the inode cache
1700 * @sb:		super block of file system to search
1701 * @hashval:	hash value (usually inode number) to search for
1702 * @match:	callback used for comparisons between inodes
1703 * @data:	opaque data pointer to pass to @match
1704 *
1705 * Search for the inode specified by @hashval and @data in the inode
1706 * cache, where the helper function @match will return 0 if the inode
1707 * does not match, 1 if the inode does match, and -1 if the search
1708 * should be stopped.  The @match function must be responsible for
1709 * taking the i_lock spin_lock and checking i_state for an inode being
1710 * freed or being initialized, and incrementing the reference count
1711 * before returning 1.  It also must not sleep, since it is called with
1712 * the inode_hash_lock spinlock held.
1713 *
1714 * This is a even more generalized version of ilookup5() when the
1715 * function must never block --- find_inode() can block in
1716 * __wait_on_freeing_inode() --- or when the caller can not increment
1717 * the reference count because the resulting iput() might cause an
1718 * inode eviction.  The tradeoff is that the @match funtion must be
1719 * very carefully implemented.
1720 */
1721struct inode *find_inode_nowait(struct super_block *sb,
1722				unsigned long hashval,
1723				int (*match)(struct inode *, unsigned long,
1724					     void *),
1725				void *data)
1726{
1727	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1728	struct inode *inode, *ret_inode = NULL;
1729	int mval;
1730
1731	spin_lock(&inode_hash_lock);
1732	hlist_for_each_entry(inode, head, i_hash) {
1733		if (inode->i_sb != sb)
1734			continue;
1735		mval = match(inode, hashval, data);
1736		if (mval == 0)
1737			continue;
1738		if (mval == 1)
1739			ret_inode = inode;
1740		goto out;
1741	}
1742out:
1743	spin_unlock(&inode_hash_lock);
1744	return ret_inode;
1745}
1746EXPORT_SYMBOL(find_inode_nowait);
1747
1748/**
1749 * find_inode_rcu - find an inode in the inode cache
1750 * @sb:		Super block of file system to search
1751 * @hashval:	Key to hash
1752 * @test:	Function to test match on an inode
1753 * @data:	Data for test function
1754 *
1755 * Search for the inode specified by @hashval and @data in the inode cache,
1756 * where the helper function @test will return 0 if the inode does not match
1757 * and 1 if it does.  The @test function must be responsible for taking the
1758 * i_lock spin_lock and checking i_state for an inode being freed or being
1759 * initialized.
1760 *
1761 * If successful, this will return the inode for which the @test function
1762 * returned 1 and NULL otherwise.
1763 *
1764 * The @test function is not permitted to take a ref on any inode presented.
1765 * It is also not permitted to sleep.
1766 *
1767 * The caller must hold the RCU read lock.
1768 */
1769struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1770			     int (*test)(struct inode *, void *), void *data)
1771{
1772	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1773	struct inode *inode;
1774
1775	RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1776			 "suspicious find_inode_rcu() usage");
1777
1778	hlist_for_each_entry_rcu(inode, head, i_hash) {
1779		if (inode->i_sb == sb &&
1780		    !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1781		    test(inode, data))
1782			return inode;
1783	}
1784	return NULL;
1785}
1786EXPORT_SYMBOL(find_inode_rcu);
1787
1788/**
1789 * find_inode_by_ino_rcu - Find an inode in the inode cache
1790 * @sb:		Super block of file system to search
1791 * @ino:	The inode number to match
1792 *
1793 * Search for the inode specified by @hashval and @data in the inode cache,
1794 * where the helper function @test will return 0 if the inode does not match
1795 * and 1 if it does.  The @test function must be responsible for taking the
1796 * i_lock spin_lock and checking i_state for an inode being freed or being
1797 * initialized.
1798 *
1799 * If successful, this will return the inode for which the @test function
1800 * returned 1 and NULL otherwise.
1801 *
1802 * The @test function is not permitted to take a ref on any inode presented.
1803 * It is also not permitted to sleep.
1804 *
1805 * The caller must hold the RCU read lock.
1806 */
1807struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1808				    unsigned long ino)
1809{
1810	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1811	struct inode *inode;
1812
1813	RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1814			 "suspicious find_inode_by_ino_rcu() usage");
1815
1816	hlist_for_each_entry_rcu(inode, head, i_hash) {
1817		if (inode->i_ino == ino &&
1818		    inode->i_sb == sb &&
1819		    !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1820		    return inode;
1821	}
1822	return NULL;
1823}
1824EXPORT_SYMBOL(find_inode_by_ino_rcu);
1825
1826int insert_inode_locked(struct inode *inode)
1827{
1828	struct super_block *sb = inode->i_sb;
1829	ino_t ino = inode->i_ino;
1830	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1831
1832	while (1) {
1833		struct inode *old = NULL;
1834		spin_lock(&inode_hash_lock);
1835		hlist_for_each_entry(old, head, i_hash) {
1836			if (old->i_ino != ino)
1837				continue;
1838			if (old->i_sb != sb)
1839				continue;
1840			spin_lock(&old->i_lock);
1841			if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1842				spin_unlock(&old->i_lock);
1843				continue;
1844			}
1845			break;
1846		}
1847		if (likely(!old)) {
1848			spin_lock(&inode->i_lock);
1849			inode->i_state |= I_NEW | I_CREATING;
1850			hlist_add_head_rcu(&inode->i_hash, head);
1851			spin_unlock(&inode->i_lock);
1852			spin_unlock(&inode_hash_lock);
1853			return 0;
1854		}
1855		if (unlikely(old->i_state & I_CREATING)) {
1856			spin_unlock(&old->i_lock);
1857			spin_unlock(&inode_hash_lock);
1858			return -EBUSY;
1859		}
1860		__iget(old);
1861		spin_unlock(&old->i_lock);
1862		spin_unlock(&inode_hash_lock);
1863		wait_on_inode(old);
1864		if (unlikely(!inode_unhashed(old))) {
1865			iput(old);
1866			return -EBUSY;
1867		}
1868		iput(old);
1869	}
1870}
1871EXPORT_SYMBOL(insert_inode_locked);
1872
1873int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1874		int (*test)(struct inode *, void *), void *data)
1875{
1876	struct inode *old;
 
1877
1878	inode->i_state |= I_CREATING;
1879	old = inode_insert5(inode, hashval, test, NULL, data);
1880
1881	if (old != inode) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1882		iput(old);
1883		return -EBUSY;
1884	}
1885	return 0;
1886}
1887EXPORT_SYMBOL(insert_inode_locked4);
1888
1889
1890int generic_delete_inode(struct inode *inode)
1891{
1892	return 1;
1893}
1894EXPORT_SYMBOL(generic_delete_inode);
1895
1896/*
1897 * Called when we're dropping the last reference
1898 * to an inode.
1899 *
1900 * Call the FS "drop_inode()" function, defaulting to
1901 * the legacy UNIX filesystem behaviour.  If it tells
1902 * us to evict inode, do so.  Otherwise, retain inode
1903 * in cache if fs is alive, sync and evict if fs is
1904 * shutting down.
1905 */
1906static void iput_final(struct inode *inode)
1907{
1908	struct super_block *sb = inode->i_sb;
1909	const struct super_operations *op = inode->i_sb->s_op;
1910	unsigned long state;
1911	int drop;
1912
1913	WARN_ON(inode->i_state & I_NEW);
1914
1915	if (op->drop_inode)
1916		drop = op->drop_inode(inode);
1917	else
1918		drop = generic_drop_inode(inode);
1919
1920	if (!drop &&
1921	    !(inode->i_state & I_DONTCACHE) &&
1922	    (sb->s_flags & SB_ACTIVE)) {
1923		__inode_add_lru(inode, true);
1924		spin_unlock(&inode->i_lock);
1925		return;
1926	}
1927
1928	state = inode->i_state;
1929	if (!drop) {
1930		WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1931		spin_unlock(&inode->i_lock);
1932
1933		write_inode_now(inode, 1);
1934
1935		spin_lock(&inode->i_lock);
1936		state = inode->i_state;
1937		WARN_ON(state & I_NEW);
1938		state &= ~I_WILL_FREE;
1939	}
1940
1941	WRITE_ONCE(inode->i_state, state | I_FREEING);
1942	if (!list_empty(&inode->i_lru))
1943		inode_lru_list_del(inode);
1944	spin_unlock(&inode->i_lock);
1945
1946	evict(inode);
1947}
1948
1949/**
1950 *	iput	- put an inode
1951 *	@inode: inode to put
1952 *
1953 *	Puts an inode, dropping its usage count. If the inode use count hits
1954 *	zero, the inode is then freed and may also be destroyed.
1955 *
1956 *	Consequently, iput() can sleep.
1957 */
1958void iput(struct inode *inode)
1959{
1960	if (!inode)
1961		return;
1962	BUG_ON(inode->i_state & I_CLEAR);
1963retry:
1964	if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1965		if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1966			atomic_inc(&inode->i_count);
1967			spin_unlock(&inode->i_lock);
1968			trace_writeback_lazytime_iput(inode);
1969			mark_inode_dirty_sync(inode);
1970			goto retry;
1971		}
1972		iput_final(inode);
1973	}
1974}
1975EXPORT_SYMBOL(iput);
1976
1977#ifdef CONFIG_BLOCK
1978/**
1979 *	bmap	- find a block number in a file
1980 *	@inode:  inode owning the block number being requested
1981 *	@block: pointer containing the block to find
1982 *
1983 *	Replaces the value in ``*block`` with the block number on the device holding
1984 *	corresponding to the requested block number in the file.
1985 *	That is, asked for block 4 of inode 1 the function will replace the
1986 *	4 in ``*block``, with disk block relative to the disk start that holds that
1987 *	block of the file.
1988 *
1989 *	Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1990 *	hole, returns 0 and ``*block`` is also set to 0.
1991 */
1992int bmap(struct inode *inode, sector_t *block)
1993{
1994	if (!inode->i_mapping->a_ops->bmap)
1995		return -EINVAL;
1996
1997	*block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1998	return 0;
 
 
1999}
2000EXPORT_SYMBOL(bmap);
2001#endif
2002
2003/*
2004 * With relative atime, only update atime if the previous atime is
2005 * earlier than or equal to either the ctime or mtime,
2006 * or if at least a day has passed since the last atime update.
2007 */
2008static bool relatime_need_update(struct vfsmount *mnt, struct inode *inode,
2009			     struct timespec64 now)
2010{
2011	struct timespec64 atime, mtime, ctime;
2012
2013	if (!(mnt->mnt_flags & MNT_RELATIME))
2014		return true;
2015	/*
2016	 * Is mtime younger than or equal to atime? If yes, update atime:
2017	 */
2018	atime = inode_get_atime(inode);
2019	mtime = inode_get_mtime(inode);
2020	if (timespec64_compare(&mtime, &atime) >= 0)
2021		return true;
2022	/*
2023	 * Is ctime younger than or equal to atime? If yes, update atime:
2024	 */
2025	ctime = inode_get_ctime(inode);
2026	if (timespec64_compare(&ctime, &atime) >= 0)
2027		return true;
2028
2029	/*
2030	 * Is the previous atime value older than a day? If yes,
2031	 * update atime:
2032	 */
2033	if ((long)(now.tv_sec - atime.tv_sec) >= 24*60*60)
2034		return true;
2035	/*
2036	 * Good, we can skip the atime update:
2037	 */
2038	return false;
2039}
2040
2041/**
2042 * inode_update_timestamps - update the timestamps on the inode
2043 * @inode: inode to be updated
2044 * @flags: S_* flags that needed to be updated
2045 *
2046 * The update_time function is called when an inode's timestamps need to be
2047 * updated for a read or write operation. This function handles updating the
2048 * actual timestamps. It's up to the caller to ensure that the inode is marked
2049 * dirty appropriately.
2050 *
2051 * In the case where any of S_MTIME, S_CTIME, or S_VERSION need to be updated,
2052 * attempt to update all three of them. S_ATIME updates can be handled
2053 * independently of the rest.
2054 *
2055 * Returns a set of S_* flags indicating which values changed.
2056 */
2057int inode_update_timestamps(struct inode *inode, int flags)
2058{
2059	int updated = 0;
2060	struct timespec64 now;
2061
2062	if (flags & (S_MTIME|S_CTIME|S_VERSION)) {
2063		struct timespec64 ctime = inode_get_ctime(inode);
2064		struct timespec64 mtime = inode_get_mtime(inode);
2065
2066		now = inode_set_ctime_current(inode);
2067		if (!timespec64_equal(&now, &ctime))
2068			updated |= S_CTIME;
2069		if (!timespec64_equal(&now, &mtime)) {
2070			inode_set_mtime_to_ts(inode, now);
2071			updated |= S_MTIME;
2072		}
2073		if (IS_I_VERSION(inode) && inode_maybe_inc_iversion(inode, updated))
2074			updated |= S_VERSION;
2075	} else {
2076		now = current_time(inode);
2077	}
2078
2079	if (flags & S_ATIME) {
2080		struct timespec64 atime = inode_get_atime(inode);
2081
2082		if (!timespec64_equal(&now, &atime)) {
2083			inode_set_atime_to_ts(inode, now);
2084			updated |= S_ATIME;
2085		}
2086	}
2087	return updated;
2088}
2089EXPORT_SYMBOL(inode_update_timestamps);
2090
2091/**
2092 * generic_update_time - update the timestamps on the inode
2093 * @inode: inode to be updated
2094 * @flags: S_* flags that needed to be updated
2095 *
2096 * The update_time function is called when an inode's timestamps need to be
2097 * updated for a read or write operation. In the case where any of S_MTIME, S_CTIME,
2098 * or S_VERSION need to be updated we attempt to update all three of them. S_ATIME
2099 * updates can be handled done independently of the rest.
2100 *
2101 * Returns a S_* mask indicating which fields were updated.
2102 */
2103int generic_update_time(struct inode *inode, int flags)
2104{
2105	int updated = inode_update_timestamps(inode, flags);
2106	int dirty_flags = 0;
2107
2108	if (updated & (S_ATIME|S_MTIME|S_CTIME))
2109		dirty_flags = inode->i_sb->s_flags & SB_LAZYTIME ? I_DIRTY_TIME : I_DIRTY_SYNC;
2110	if (updated & S_VERSION)
2111		dirty_flags |= I_DIRTY_SYNC;
2112	__mark_inode_dirty(inode, dirty_flags);
2113	return updated;
2114}
2115EXPORT_SYMBOL(generic_update_time);
2116
2117/*
2118 * This does the actual work of updating an inodes time or version.  Must have
2119 * had called mnt_want_write() before calling this.
2120 */
2121int inode_update_time(struct inode *inode, int flags)
2122{
2123	if (inode->i_op->update_time)
2124		return inode->i_op->update_time(inode, flags);
2125	generic_update_time(inode, flags);
 
 
 
 
 
 
 
 
 
2126	return 0;
2127}
2128EXPORT_SYMBOL(inode_update_time);
2129
2130/**
2131 *	atime_needs_update	-	update the access time
2132 *	@path: the &struct path to update
2133 *	@inode: inode to update
2134 *
2135 *	Update the accessed time on an inode and mark it for writeback.
2136 *	This function automatically handles read only file systems and media,
2137 *	as well as the "noatime" flag and inode specific "noatime" markers.
2138 */
2139bool atime_needs_update(const struct path *path, struct inode *inode)
2140{
2141	struct vfsmount *mnt = path->mnt;
2142	struct timespec64 now, atime;
 
2143
2144	if (inode->i_flags & S_NOATIME)
2145		return false;
2146
2147	/* Atime updates will likely cause i_uid and i_gid to be written
2148	 * back improprely if their true value is unknown to the vfs.
2149	 */
2150	if (HAS_UNMAPPED_ID(mnt_idmap(mnt), inode))
2151		return false;
2152
2153	if (IS_NOATIME(inode))
2154		return false;
2155	if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
2156		return false;
2157
2158	if (mnt->mnt_flags & MNT_NOATIME)
2159		return false;
2160	if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
2161		return false;
2162
2163	now = current_time(inode);
2164
2165	if (!relatime_need_update(mnt, inode, now))
2166		return false;
2167
2168	atime = inode_get_atime(inode);
2169	if (timespec64_equal(&atime, &now))
2170		return false;
2171
2172	return true;
2173}
2174
2175void touch_atime(const struct path *path)
2176{
2177	struct vfsmount *mnt = path->mnt;
2178	struct inode *inode = d_inode(path->dentry);
2179
2180	if (!atime_needs_update(path, inode))
2181		return;
2182
2183	if (!sb_start_write_trylock(inode->i_sb))
2184		return;
2185
2186	if (mnt_get_write_access(mnt) != 0)
2187		goto skip_update;
2188	/*
2189	 * File systems can error out when updating inodes if they need to
2190	 * allocate new space to modify an inode (such is the case for
2191	 * Btrfs), but since we touch atime while walking down the path we
2192	 * really don't care if we failed to update the atime of the file,
2193	 * so just ignore the return value.
2194	 * We may also fail on filesystems that have the ability to make parts
2195	 * of the fs read only, e.g. subvolumes in Btrfs.
2196	 */
2197	inode_update_time(inode, S_ATIME);
2198	mnt_put_write_access(mnt);
2199skip_update:
2200	sb_end_write(inode->i_sb);
2201}
2202EXPORT_SYMBOL(touch_atime);
2203
2204/*
2205 * Return mask of changes for notify_change() that need to be done as a
2206 * response to write or truncate. Return 0 if nothing has to be changed.
2207 * Negative value on error (change should be denied).
 
2208 */
2209int dentry_needs_remove_privs(struct mnt_idmap *idmap,
2210			      struct dentry *dentry)
2211{
2212	struct inode *inode = d_inode(dentry);
2213	int mask = 0;
2214	int ret;
2215
2216	if (IS_NOSEC(inode))
2217		return 0;
 
 
 
 
 
 
 
 
 
 
 
2218
2219	mask = setattr_should_drop_suidgid(idmap, inode);
2220	ret = security_inode_need_killpriv(dentry);
2221	if (ret < 0)
2222		return ret;
2223	if (ret)
2224		mask |= ATTR_KILL_PRIV;
2225	return mask;
2226}
 
2227
2228static int __remove_privs(struct mnt_idmap *idmap,
2229			  struct dentry *dentry, int kill)
2230{
2231	struct iattr newattrs;
2232
2233	newattrs.ia_valid = ATTR_FORCE | kill;
2234	/*
2235	 * Note we call this on write, so notify_change will not
2236	 * encounter any conflicting delegations:
2237	 */
2238	return notify_change(idmap, dentry, &newattrs, NULL);
2239}
2240
2241int file_remove_privs_flags(struct file *file, unsigned int flags)
2242{
2243	struct dentry *dentry = file_dentry(file);
2244	struct inode *inode = file_inode(file);
 
 
2245	int error = 0;
2246	int kill;
2247
2248	if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
 
2249		return 0;
2250
2251	kill = dentry_needs_remove_privs(file_mnt_idmap(file), dentry);
2252	if (kill < 0)
2253		return kill;
2254
2255	if (kill) {
2256		if (flags & IOCB_NOWAIT)
2257			return -EAGAIN;
 
 
 
 
 
2258
2259		error = __remove_privs(file_mnt_idmap(file), dentry, kill);
2260	}
2261
2262	if (!error)
2263		inode_has_no_xattr(inode);
2264	return error;
2265}
2266EXPORT_SYMBOL_GPL(file_remove_privs_flags);
2267
2268/**
2269 * file_remove_privs - remove special file privileges (suid, capabilities)
2270 * @file: file to remove privileges from
2271 *
2272 * When file is modified by a write or truncation ensure that special
2273 * file privileges are removed.
2274 *
2275 * Return: 0 on success, negative errno on failure.
2276 */
2277int file_remove_privs(struct file *file)
2278{
2279	return file_remove_privs_flags(file, 0);
2280}
2281EXPORT_SYMBOL(file_remove_privs);
2282
2283/**
2284 * current_time - Return FS time (possibly fine-grained)
2285 * @inode: inode.
2286 *
2287 * Return the current time truncated to the time granularity supported by
2288 * the fs, as suitable for a ctime/mtime change. If the ctime is flagged
2289 * as having been QUERIED, get a fine-grained timestamp, but don't update
2290 * the floor.
2291 *
2292 * For a multigrain inode, this is effectively an estimate of the timestamp
2293 * that a file would receive. An actual update must go through
2294 * inode_set_ctime_current().
2295 */
2296struct timespec64 current_time(struct inode *inode)
2297{
2298	struct timespec64 now;
2299	u32 cns;
2300
2301	ktime_get_coarse_real_ts64_mg(&now);
2302
2303	if (!is_mgtime(inode))
2304		goto out;
2305
2306	/* If nothing has queried it, then coarse time is fine */
2307	cns = smp_load_acquire(&inode->i_ctime_nsec);
2308	if (cns & I_CTIME_QUERIED) {
2309		/*
2310		 * If there is no apparent change, then get a fine-grained
2311		 * timestamp.
2312		 */
2313		if (now.tv_nsec == (cns & ~I_CTIME_QUERIED))
2314			ktime_get_real_ts64(&now);
2315	}
2316out:
2317	return timestamp_truncate(now, inode);
2318}
2319EXPORT_SYMBOL(current_time);
2320
2321static int inode_needs_update_time(struct inode *inode)
2322{
2323	struct timespec64 now, ts;
 
2324	int sync_it = 0;
 
2325
2326	/* First try to exhaust all avenues to not sync */
2327	if (IS_NOCMTIME(inode))
2328		return 0;
2329
2330	now = current_time(inode);
2331
2332	ts = inode_get_mtime(inode);
2333	if (!timespec64_equal(&ts, &now))
2334		sync_it |= S_MTIME;
2335
2336	ts = inode_get_ctime(inode);
2337	if (!timespec64_equal(&ts, &now))
2338		sync_it |= S_CTIME;
2339
2340	if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2341		sync_it |= S_VERSION;
2342
2343	return sync_it;
2344}
2345
2346static int __file_update_time(struct file *file, int sync_mode)
2347{
2348	int ret = 0;
2349	struct inode *inode = file_inode(file);
2350
2351	/* try to update time settings */
2352	if (!mnt_get_write_access_file(file)) {
2353		ret = inode_update_time(inode, sync_mode);
2354		mnt_put_write_access_file(file);
2355	}
2356
2357	return ret;
2358}
2359
2360/**
2361 * file_update_time - update mtime and ctime time
2362 * @file: file accessed
2363 *
2364 * Update the mtime and ctime members of an inode and mark the inode for
2365 * writeback. Note that this function is meant exclusively for usage in
2366 * the file write path of filesystems, and filesystems may choose to
2367 * explicitly ignore updates via this function with the _NOCMTIME inode
2368 * flag, e.g. for network filesystem where these imestamps are handled
2369 * by the server. This can return an error for file systems who need to
2370 * allocate space in order to update an inode.
2371 *
2372 * Return: 0 on success, negative errno on failure.
2373 */
2374int file_update_time(struct file *file)
2375{
2376	int ret;
2377	struct inode *inode = file_inode(file);
2378
2379	ret = inode_needs_update_time(inode);
2380	if (ret <= 0)
2381		return ret;
2382
2383	return __file_update_time(file, ret);
2384}
2385EXPORT_SYMBOL(file_update_time);
2386
2387/**
2388 * file_modified_flags - handle mandated vfs changes when modifying a file
2389 * @file: file that was modified
2390 * @flags: kiocb flags
2391 *
2392 * When file has been modified ensure that special
2393 * file privileges are removed and time settings are updated.
2394 *
2395 * If IOCB_NOWAIT is set, special file privileges will not be removed and
2396 * time settings will not be updated. It will return -EAGAIN.
2397 *
2398 * Context: Caller must hold the file's inode lock.
2399 *
2400 * Return: 0 on success, negative errno on failure.
2401 */
2402static int file_modified_flags(struct file *file, int flags)
2403{
2404	int ret;
2405	struct inode *inode = file_inode(file);
2406
2407	/*
2408	 * Clear the security bits if the process is not being run by root.
2409	 * This keeps people from modifying setuid and setgid binaries.
2410	 */
2411	ret = file_remove_privs_flags(file, flags);
2412	if (ret)
2413		return ret;
2414
2415	if (unlikely(file->f_mode & FMODE_NOCMTIME))
2416		return 0;
2417
2418	ret = inode_needs_update_time(inode);
2419	if (ret <= 0)
2420		return ret;
2421	if (flags & IOCB_NOWAIT)
2422		return -EAGAIN;
2423
2424	return __file_update_time(file, ret);
2425}
2426
2427/**
2428 * file_modified - handle mandated vfs changes when modifying a file
2429 * @file: file that was modified
2430 *
2431 * When file has been modified ensure that special
2432 * file privileges are removed and time settings are updated.
2433 *
2434 * Context: Caller must hold the file's inode lock.
2435 *
2436 * Return: 0 on success, negative errno on failure.
2437 */
2438int file_modified(struct file *file)
2439{
2440	return file_modified_flags(file, 0);
2441}
2442EXPORT_SYMBOL(file_modified);
2443
2444/**
2445 * kiocb_modified - handle mandated vfs changes when modifying a file
2446 * @iocb: iocb that was modified
2447 *
2448 * When file has been modified ensure that special
2449 * file privileges are removed and time settings are updated.
2450 *
2451 * Context: Caller must hold the file's inode lock.
2452 *
2453 * Return: 0 on success, negative errno on failure.
2454 */
2455int kiocb_modified(struct kiocb *iocb)
2456{
2457	return file_modified_flags(iocb->ki_filp, iocb->ki_flags);
2458}
2459EXPORT_SYMBOL_GPL(kiocb_modified);
2460
2461int inode_needs_sync(struct inode *inode)
2462{
2463	if (IS_SYNC(inode))
2464		return 1;
2465	if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2466		return 1;
2467	return 0;
2468}
2469EXPORT_SYMBOL(inode_needs_sync);
2470
 
 
 
 
 
 
 
2471/*
2472 * If we try to find an inode in the inode hash while it is being
2473 * deleted, we have to wait until the filesystem completes its
2474 * deletion before reporting that it isn't found.  This function waits
2475 * until the deletion _might_ have completed.  Callers are responsible
2476 * to recheck inode state.
2477 *
2478 * It doesn't matter if I_NEW is not set initially, a call to
2479 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2480 * will DTRT.
2481 */
2482static void __wait_on_freeing_inode(struct inode *inode, bool is_inode_hash_locked)
2483{
2484	struct wait_bit_queue_entry wqe;
2485	struct wait_queue_head *wq_head;
2486
2487	/*
2488	 * Handle racing against evict(), see that routine for more details.
2489	 */
2490	if (unlikely(inode_unhashed(inode))) {
2491		WARN_ON(is_inode_hash_locked);
2492		spin_unlock(&inode->i_lock);
2493		return;
2494	}
2495
2496	wq_head = inode_bit_waitqueue(&wqe, inode, __I_NEW);
2497	prepare_to_wait_event(wq_head, &wqe.wq_entry, TASK_UNINTERRUPTIBLE);
2498	spin_unlock(&inode->i_lock);
2499	rcu_read_unlock();
2500	if (is_inode_hash_locked)
2501		spin_unlock(&inode_hash_lock);
2502	schedule();
2503	finish_wait(wq_head, &wqe.wq_entry);
2504	if (is_inode_hash_locked)
2505		spin_lock(&inode_hash_lock);
2506	rcu_read_lock();
2507}
2508
2509static __initdata unsigned long ihash_entries;
2510static int __init set_ihash_entries(char *str)
2511{
2512	if (!str)
2513		return 0;
2514	ihash_entries = simple_strtoul(str, &str, 0);
2515	return 1;
2516}
2517__setup("ihash_entries=", set_ihash_entries);
2518
2519/*
2520 * Initialize the waitqueues and inode hash table.
2521 */
2522void __init inode_init_early(void)
2523{
 
 
2524	/* If hashes are distributed across NUMA nodes, defer
2525	 * hash allocation until vmalloc space is available.
2526	 */
2527	if (hashdist)
2528		return;
2529
2530	inode_hashtable =
2531		alloc_large_system_hash("Inode-cache",
2532					sizeof(struct hlist_head),
2533					ihash_entries,
2534					14,
2535					HASH_EARLY | HASH_ZERO,
2536					&i_hash_shift,
2537					&i_hash_mask,
2538					0,
2539					0);
 
 
 
2540}
2541
2542void __init inode_init(void)
2543{
 
 
2544	/* inode slab cache */
2545	inode_cachep = kmem_cache_create("inode_cache",
2546					 sizeof(struct inode),
2547					 0,
2548					 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2549					 SLAB_ACCOUNT),
2550					 init_once);
2551
2552	/* Hash may have been set up in inode_init_early */
2553	if (!hashdist)
2554		return;
2555
2556	inode_hashtable =
2557		alloc_large_system_hash("Inode-cache",
2558					sizeof(struct hlist_head),
2559					ihash_entries,
2560					14,
2561					HASH_ZERO,
2562					&i_hash_shift,
2563					&i_hash_mask,
2564					0,
2565					0);
 
 
 
2566}
2567
2568void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2569{
2570	inode->i_mode = mode;
2571	if (S_ISCHR(mode)) {
2572		inode->i_fop = &def_chr_fops;
2573		inode->i_rdev = rdev;
2574	} else if (S_ISBLK(mode)) {
2575		if (IS_ENABLED(CONFIG_BLOCK))
2576			inode->i_fop = &def_blk_fops;
2577		inode->i_rdev = rdev;
2578	} else if (S_ISFIFO(mode))
2579		inode->i_fop = &pipefifo_fops;
2580	else if (S_ISSOCK(mode))
2581		;	/* leave it no_open_fops */
2582	else
2583		printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2584				  " inode %s:%lu\n", mode, inode->i_sb->s_id,
2585				  inode->i_ino);
2586}
2587EXPORT_SYMBOL(init_special_inode);
2588
2589/**
2590 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2591 * @idmap: idmap of the mount the inode was created from
2592 * @inode: New inode
2593 * @dir: Directory inode
2594 * @mode: mode of the new inode
2595 *
2596 * If the inode has been created through an idmapped mount the idmap of
2597 * the vfsmount must be passed through @idmap. This function will then take
2598 * care to map the inode according to @idmap before checking permissions
2599 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2600 * checking is to be performed on the raw inode simply pass @nop_mnt_idmap.
2601 */
2602void inode_init_owner(struct mnt_idmap *idmap, struct inode *inode,
2603		      const struct inode *dir, umode_t mode)
2604{
2605	inode_fsuid_set(inode, idmap);
2606	if (dir && dir->i_mode & S_ISGID) {
2607		inode->i_gid = dir->i_gid;
2608
2609		/* Directories are special, and always inherit S_ISGID */
2610		if (S_ISDIR(mode))
2611			mode |= S_ISGID;
2612	} else
2613		inode_fsgid_set(inode, idmap);
2614	inode->i_mode = mode;
2615}
2616EXPORT_SYMBOL(inode_init_owner);
2617
2618/**
2619 * inode_owner_or_capable - check current task permissions to inode
2620 * @idmap: idmap of the mount the inode was found from
2621 * @inode: inode being checked
2622 *
2623 * Return true if current either has CAP_FOWNER in a namespace with the
2624 * inode owner uid mapped, or owns the file.
2625 *
2626 * If the inode has been found through an idmapped mount the idmap of
2627 * the vfsmount must be passed through @idmap. This function will then take
2628 * care to map the inode according to @idmap before checking permissions.
2629 * On non-idmapped mounts or if permission checking is to be performed on the
2630 * raw inode simply pass @nop_mnt_idmap.
2631 */
2632bool inode_owner_or_capable(struct mnt_idmap *idmap,
2633			    const struct inode *inode)
2634{
2635	vfsuid_t vfsuid;
2636	struct user_namespace *ns;
2637
2638	vfsuid = i_uid_into_vfsuid(idmap, inode);
2639	if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
2640		return true;
2641
2642	ns = current_user_ns();
2643	if (vfsuid_has_mapping(ns, vfsuid) && ns_capable(ns, CAP_FOWNER))
2644		return true;
2645	return false;
2646}
2647EXPORT_SYMBOL(inode_owner_or_capable);
2648
2649/*
2650 * Direct i/o helper functions
2651 */
2652bool inode_dio_finished(const struct inode *inode)
2653{
2654	return atomic_read(&inode->i_dio_count) == 0;
 
 
 
 
 
 
 
 
2655}
2656EXPORT_SYMBOL(inode_dio_finished);
2657
2658/**
2659 * inode_dio_wait - wait for outstanding DIO requests to finish
2660 * @inode: inode to wait for
2661 *
2662 * Waits for all pending direct I/O requests to finish so that we can
2663 * proceed with a truncate or equivalent operation.
2664 *
2665 * Must be called under a lock that serializes taking new references
2666 * to i_dio_count, usually by inode->i_mutex.
2667 */
2668void inode_dio_wait(struct inode *inode)
2669{
2670	wait_var_event(&inode->i_dio_count, inode_dio_finished(inode));
 
2671}
2672EXPORT_SYMBOL(inode_dio_wait);
2673
2674void inode_dio_wait_interruptible(struct inode *inode)
 
 
 
 
 
 
 
2675{
2676	wait_var_event_interruptible(&inode->i_dio_count,
2677				     inode_dio_finished(inode));
2678}
2679EXPORT_SYMBOL(inode_dio_wait_interruptible);
2680
2681/*
2682 * inode_set_flags - atomically set some inode flags
2683 *
2684 * Note: the caller should be holding i_mutex, or else be sure that
2685 * they have exclusive access to the inode structure (i.e., while the
2686 * inode is being instantiated).  The reason for the cmpxchg() loop
2687 * --- which wouldn't be necessary if all code paths which modify
2688 * i_flags actually followed this rule, is that there is at least one
2689 * code path which doesn't today so we use cmpxchg() out of an abundance
2690 * of caution.
 
2691 *
2692 * In the long run, i_mutex is overkill, and we should probably look
2693 * at using the i_lock spinlock to protect i_flags, and then make sure
2694 * it is so documented in include/linux/fs.h and that all code follows
2695 * the locking convention!!
2696 */
2697void inode_set_flags(struct inode *inode, unsigned int flags,
2698		     unsigned int mask)
2699{
 
 
2700	WARN_ON_ONCE(flags & ~mask);
2701	set_mask_bits(&inode->i_flags, mask, flags);
 
 
 
 
2702}
2703EXPORT_SYMBOL(inode_set_flags);
2704
2705void inode_nohighmem(struct inode *inode)
2706{
2707	mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2708}
2709EXPORT_SYMBOL(inode_nohighmem);
2710
2711struct timespec64 inode_set_ctime_to_ts(struct inode *inode, struct timespec64 ts)
2712{
2713	trace_inode_set_ctime_to_ts(inode, &ts);
2714	set_normalized_timespec64(&ts, ts.tv_sec, ts.tv_nsec);
2715	inode->i_ctime_sec = ts.tv_sec;
2716	inode->i_ctime_nsec = ts.tv_nsec;
2717	return ts;
2718}
2719EXPORT_SYMBOL(inode_set_ctime_to_ts);
2720
2721/**
2722 * timestamp_truncate - Truncate timespec to a granularity
2723 * @t: Timespec
2724 * @inode: inode being updated
2725 *
2726 * Truncate a timespec to the granularity supported by the fs
2727 * containing the inode. Always rounds down. gran must
2728 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2729 */
2730struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2731{
2732	struct super_block *sb = inode->i_sb;
2733	unsigned int gran = sb->s_time_gran;
2734
2735	t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2736	if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2737		t.tv_nsec = 0;
2738
2739	/* Avoid division in the common cases 1 ns and 1 s. */
2740	if (gran == 1)
2741		; /* nothing */
2742	else if (gran == NSEC_PER_SEC)
2743		t.tv_nsec = 0;
2744	else if (gran > 1 && gran < NSEC_PER_SEC)
2745		t.tv_nsec -= t.tv_nsec % gran;
2746	else
2747		WARN(1, "invalid file time granularity: %u", gran);
2748	return t;
2749}
2750EXPORT_SYMBOL(timestamp_truncate);
2751
2752/**
2753 * inode_set_ctime_current - set the ctime to current_time
2754 * @inode: inode
2755 *
2756 * Set the inode's ctime to the current value for the inode. Returns the
2757 * current value that was assigned. If this is not a multigrain inode, then we
2758 * set it to the later of the coarse time and floor value.
2759 *
2760 * If it is multigrain, then we first see if the coarse-grained timestamp is
2761 * distinct from what is already there. If so, then use that. Otherwise, get a
2762 * fine-grained timestamp.
2763 *
2764 * After that, try to swap the new value into i_ctime_nsec. Accept the
2765 * resulting ctime, regardless of the outcome of the swap. If it has
2766 * already been replaced, then that timestamp is later than the earlier
2767 * unacceptable one, and is thus acceptable.
2768 */
2769struct timespec64 inode_set_ctime_current(struct inode *inode)
2770{
2771	struct timespec64 now;
2772	u32 cns, cur;
2773
2774	ktime_get_coarse_real_ts64_mg(&now);
2775	now = timestamp_truncate(now, inode);
2776
2777	/* Just return that if this is not a multigrain fs */
2778	if (!is_mgtime(inode)) {
2779		inode_set_ctime_to_ts(inode, now);
2780		goto out;
2781	}
2782
2783	/*
2784	 * A fine-grained time is only needed if someone has queried
2785	 * for timestamps, and the current coarse grained time isn't
2786	 * later than what's already there.
2787	 */
2788	cns = smp_load_acquire(&inode->i_ctime_nsec);
2789	if (cns & I_CTIME_QUERIED) {
2790		struct timespec64 ctime = { .tv_sec = inode->i_ctime_sec,
2791					    .tv_nsec = cns & ~I_CTIME_QUERIED };
2792
2793		if (timespec64_compare(&now, &ctime) <= 0) {
2794			ktime_get_real_ts64_mg(&now);
2795			now = timestamp_truncate(now, inode);
2796			mgtime_counter_inc(mg_fine_stamps);
2797		}
2798	}
2799	mgtime_counter_inc(mg_ctime_updates);
2800
2801	/* No need to cmpxchg if it's exactly the same */
2802	if (cns == now.tv_nsec && inode->i_ctime_sec == now.tv_sec) {
2803		trace_ctime_xchg_skip(inode, &now);
2804		goto out;
2805	}
2806	cur = cns;
2807retry:
2808	/* Try to swap the nsec value into place. */
2809	if (try_cmpxchg(&inode->i_ctime_nsec, &cur, now.tv_nsec)) {
2810		/* If swap occurred, then we're (mostly) done */
2811		inode->i_ctime_sec = now.tv_sec;
2812		trace_ctime_ns_xchg(inode, cns, now.tv_nsec, cur);
2813		mgtime_counter_inc(mg_ctime_swaps);
2814	} else {
2815		/*
2816		 * Was the change due to someone marking the old ctime QUERIED?
2817		 * If so then retry the swap. This can only happen once since
2818		 * the only way to clear I_CTIME_QUERIED is to stamp the inode
2819		 * with a new ctime.
2820		 */
2821		if (!(cns & I_CTIME_QUERIED) && (cns | I_CTIME_QUERIED) == cur) {
2822			cns = cur;
2823			goto retry;
2824		}
2825		/* Otherwise, keep the existing ctime */
2826		now.tv_sec = inode->i_ctime_sec;
2827		now.tv_nsec = cur & ~I_CTIME_QUERIED;
2828	}
2829out:
2830	return now;
2831}
2832EXPORT_SYMBOL(inode_set_ctime_current);
2833
2834/**
2835 * inode_set_ctime_deleg - try to update the ctime on a delegated inode
2836 * @inode: inode to update
2837 * @update: timespec64 to set the ctime
2838 *
2839 * Attempt to atomically update the ctime on behalf of a delegation holder.
2840 *
2841 * The nfs server can call back the holder of a delegation to get updated
2842 * inode attributes, including the mtime. When updating the mtime, update
2843 * the ctime to a value at least equal to that.
2844 *
2845 * This can race with concurrent updates to the inode, in which
2846 * case the update is skipped.
2847 *
2848 * Note that this works even when multigrain timestamps are not enabled,
2849 * so it is used in either case.
2850 */
2851struct timespec64 inode_set_ctime_deleg(struct inode *inode, struct timespec64 update)
2852{
2853	struct timespec64 now, cur_ts;
2854	u32 cur, old;
2855
2856	/* pairs with try_cmpxchg below */
2857	cur = smp_load_acquire(&inode->i_ctime_nsec);
2858	cur_ts.tv_nsec = cur & ~I_CTIME_QUERIED;
2859	cur_ts.tv_sec = inode->i_ctime_sec;
2860
2861	/* If the update is older than the existing value, skip it. */
2862	if (timespec64_compare(&update, &cur_ts) <= 0)
2863		return cur_ts;
2864
2865	ktime_get_coarse_real_ts64_mg(&now);
2866
2867	/* Clamp the update to "now" if it's in the future */
2868	if (timespec64_compare(&update, &now) > 0)
2869		update = now;
2870
2871	update = timestamp_truncate(update, inode);
2872
2873	/* No need to update if the values are already the same */
2874	if (timespec64_equal(&update, &cur_ts))
2875		return cur_ts;
2876
2877	/*
2878	 * Try to swap the nsec value into place. If it fails, that means
2879	 * it raced with an update due to a write or similar activity. That
2880	 * stamp takes precedence, so just skip the update.
2881	 */
2882retry:
2883	old = cur;
2884	if (try_cmpxchg(&inode->i_ctime_nsec, &cur, update.tv_nsec)) {
2885		inode->i_ctime_sec = update.tv_sec;
2886		mgtime_counter_inc(mg_ctime_swaps);
2887		return update;
2888	}
2889
2890	/*
2891	 * Was the change due to another task marking the old ctime QUERIED?
2892	 *
2893	 * If so, then retry the swap. This can only happen once since
2894	 * the only way to clear I_CTIME_QUERIED is to stamp the inode
2895	 * with a new ctime.
2896	 */
2897	if (!(old & I_CTIME_QUERIED) && (cur == (old | I_CTIME_QUERIED)))
2898		goto retry;
2899
2900	/* Otherwise, it was a new timestamp. */
2901	cur_ts.tv_sec = inode->i_ctime_sec;
2902	cur_ts.tv_nsec = cur & ~I_CTIME_QUERIED;
2903	return cur_ts;
2904}
2905EXPORT_SYMBOL(inode_set_ctime_deleg);
2906
2907/**
2908 * in_group_or_capable - check whether caller is CAP_FSETID privileged
2909 * @idmap:	idmap of the mount @inode was found from
2910 * @inode:	inode to check
2911 * @vfsgid:	the new/current vfsgid of @inode
2912 *
2913 * Check whether @vfsgid is in the caller's group list or if the caller is
2914 * privileged with CAP_FSETID over @inode. This can be used to determine
2915 * whether the setgid bit can be kept or must be dropped.
2916 *
2917 * Return: true if the caller is sufficiently privileged, false if not.
2918 */
2919bool in_group_or_capable(struct mnt_idmap *idmap,
2920			 const struct inode *inode, vfsgid_t vfsgid)
2921{
2922	if (vfsgid_in_group_p(vfsgid))
2923		return true;
2924	if (capable_wrt_inode_uidgid(idmap, inode, CAP_FSETID))
2925		return true;
2926	return false;
2927}
2928EXPORT_SYMBOL(in_group_or_capable);
2929
2930/**
2931 * mode_strip_sgid - handle the sgid bit for non-directories
2932 * @idmap: idmap of the mount the inode was created from
2933 * @dir: parent directory inode
2934 * @mode: mode of the file to be created in @dir
2935 *
2936 * If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2937 * raised and @dir has the S_ISGID bit raised ensure that the caller is
2938 * either in the group of the parent directory or they have CAP_FSETID
2939 * in their user namespace and are privileged over the parent directory.
2940 * In all other cases, strip the S_ISGID bit from @mode.
2941 *
2942 * Return: the new mode to use for the file
2943 */
2944umode_t mode_strip_sgid(struct mnt_idmap *idmap,
2945			const struct inode *dir, umode_t mode)
2946{
2947	if ((mode & (S_ISGID | S_IXGRP)) != (S_ISGID | S_IXGRP))
2948		return mode;
2949	if (S_ISDIR(mode) || !dir || !(dir->i_mode & S_ISGID))
2950		return mode;
2951	if (in_group_or_capable(idmap, dir, i_gid_into_vfsgid(idmap, dir)))
2952		return mode;
2953	return mode & ~S_ISGID;
2954}
2955EXPORT_SYMBOL(mode_strip_sgid);
v3.15
 
   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 "internal.h"
  22
  23/*
  24 * Inode locking rules:
  25 *
  26 * inode->i_lock protects:
  27 *   inode->i_state, inode->i_hash, __iget()
  28 * Inode LRU list locks protect:
  29 *   inode->i_sb->s_inode_lru, inode->i_lru
  30 * inode_sb_list_lock protects:
  31 *   sb->s_inodes, inode->i_sb_list
  32 * bdi->wb.list_lock protects:
  33 *   bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
  34 * inode_hash_lock protects:
  35 *   inode_hashtable, inode->i_hash
  36 *
  37 * Lock ordering:
  38 *
  39 * inode_sb_list_lock
  40 *   inode->i_lock
  41 *     Inode LRU list locks
  42 *
  43 * bdi->wb.list_lock
  44 *   inode->i_lock
  45 *
  46 * inode_hash_lock
  47 *   inode_sb_list_lock
  48 *   inode->i_lock
  49 *
  50 * iunique_lock
  51 *   inode_hash_lock
  52 */
  53
  54static unsigned int i_hash_mask __read_mostly;
  55static unsigned int i_hash_shift __read_mostly;
  56static struct hlist_head *inode_hashtable __read_mostly;
  57static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
  58
  59__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_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(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
 
 
 
 
 
 117/**
 118 * inode_init_always - perform inode structure intialisation
 119 * @sb: superblock inode belongs to
 120 * @inode: inode to initialise
 
 121 *
 122 * These are initializations that need to be done on every inode
 123 * allocation as the fields are not initialised by slab allocation.
 
 124 */
 125int inode_init_always(struct super_block *sb, struct inode *inode)
 126{
 127	static const struct inode_operations empty_iops;
 128	static const struct file_operations empty_fops;
 129	struct address_space *const mapping = &inode->i_data;
 130
 131	inode->i_sb = sb;
 132	inode->i_blkbits = sb->s_blocksize_bits;
 133	inode->i_flags = 0;
 
 
 134	atomic_set(&inode->i_count, 1);
 135	inode->i_op = &empty_iops;
 136	inode->i_fop = &empty_fops;
 
 137	inode->__i_nlink = 1;
 138	inode->i_opflags = 0;
 
 
 
 
 139	i_uid_write(inode, 0);
 140	i_gid_write(inode, 0);
 141	atomic_set(&inode->i_writecount, 0);
 142	inode->i_size = 0;
 
 143	inode->i_blocks = 0;
 144	inode->i_bytes = 0;
 145	inode->i_generation = 0;
 146#ifdef CONFIG_QUOTA
 147	memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
 148#endif
 149	inode->i_pipe = NULL;
 150	inode->i_bdev = NULL;
 151	inode->i_cdev = NULL;
 
 
 152	inode->i_rdev = 0;
 153	inode->dirtied_when = 0;
 154
 155	if (security_inode_alloc(inode))
 156		goto out;
 
 
 
 
 157	spin_lock_init(&inode->i_lock);
 158	lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
 159
 160	mutex_init(&inode->i_mutex);
 161	lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
 162
 163	atomic_set(&inode->i_dio_count, 0);
 164
 165	mapping->a_ops = &empty_aops;
 166	mapping->host = inode;
 167	mapping->flags = 0;
 
 
 
 
 
 168	mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
 169	mapping->private_data = NULL;
 170	mapping->backing_dev_info = &default_backing_dev_info;
 171	mapping->writeback_index = 0;
 172
 173	/*
 174	 * If the block_device provides a backing_dev_info for client
 175	 * inodes then use that.  Otherwise the inode share the bdev's
 176	 * backing_dev_info.
 177	 */
 178	if (sb->s_bdev) {
 179		struct backing_dev_info *bdi;
 180
 181		bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
 182		mapping->backing_dev_info = bdi;
 183	}
 184	inode->i_private = NULL;
 185	inode->i_mapping = mapping;
 186	INIT_HLIST_HEAD(&inode->i_dentry);	/* buggered by rcu freeing */
 187#ifdef CONFIG_FS_POSIX_ACL
 188	inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
 189#endif
 190
 191#ifdef CONFIG_FSNOTIFY
 192	inode->i_fsnotify_mask = 0;
 193#endif
 
 
 
 
 194
 195	this_cpu_inc(nr_inodes);
 196
 197	return 0;
 198out:
 199	return -ENOMEM;
 200}
 201EXPORT_SYMBOL(inode_init_always);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 202
 203static struct inode *alloc_inode(struct super_block *sb)
 204{
 
 205	struct inode *inode;
 206
 207	if (sb->s_op->alloc_inode)
 208		inode = sb->s_op->alloc_inode(sb);
 209	else
 210		inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
 211
 212	if (!inode)
 213		return NULL;
 214
 215	if (unlikely(inode_init_always(sb, inode))) {
 216		if (inode->i_sb->s_op->destroy_inode)
 217			inode->i_sb->s_op->destroy_inode(inode);
 218		else
 219			kmem_cache_free(inode_cachep, inode);
 
 
 
 220		return NULL;
 221	}
 222
 223	return inode;
 224}
 225
 226void free_inode_nonrcu(struct inode *inode)
 227{
 228	kmem_cache_free(inode_cachep, inode);
 229}
 230EXPORT_SYMBOL(free_inode_nonrcu);
 231
 232void __destroy_inode(struct inode *inode)
 233{
 234	BUG_ON(inode_has_buffers(inode));
 
 235	security_inode_free(inode);
 236	fsnotify_inode_delete(inode);
 
 237	if (!inode->i_nlink) {
 238		WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
 239		atomic_long_dec(&inode->i_sb->s_remove_count);
 240	}
 241
 242#ifdef CONFIG_FS_POSIX_ACL
 243	if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
 244		posix_acl_release(inode->i_acl);
 245	if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
 246		posix_acl_release(inode->i_default_acl);
 247#endif
 248	this_cpu_dec(nr_inodes);
 249}
 250EXPORT_SYMBOL(__destroy_inode);
 251
 252static void i_callback(struct rcu_head *head)
 253{
 254	struct inode *inode = container_of(head, struct inode, i_rcu);
 255	kmem_cache_free(inode_cachep, inode);
 256}
 257
 258static void destroy_inode(struct inode *inode)
 259{
 260	BUG_ON(!list_empty(&inode->i_lru));
 261	__destroy_inode(inode);
 262	if (inode->i_sb->s_op->destroy_inode)
 263		inode->i_sb->s_op->destroy_inode(inode);
 264	else
 265		call_rcu(&inode->i_rcu, i_callback);
 
 
 
 266}
 267
 268/**
 269 * drop_nlink - directly drop an inode's link count
 270 * @inode: inode
 271 *
 272 * This is a low-level filesystem helper to replace any
 273 * direct filesystem manipulation of i_nlink.  In cases
 274 * where we are attempting to track writes to the
 275 * filesystem, a decrement to zero means an imminent
 276 * write when the file is truncated and actually unlinked
 277 * on the filesystem.
 278 */
 279void drop_nlink(struct inode *inode)
 280{
 281	WARN_ON(inode->i_nlink == 0);
 282	inode->__i_nlink--;
 283	if (!inode->i_nlink)
 284		atomic_long_inc(&inode->i_sb->s_remove_count);
 285}
 286EXPORT_SYMBOL(drop_nlink);
 287
 288/**
 289 * clear_nlink - directly zero an inode's link count
 290 * @inode: inode
 291 *
 292 * This is a low-level filesystem helper to replace any
 293 * direct filesystem manipulation of i_nlink.  See
 294 * drop_nlink() for why we care about i_nlink hitting zero.
 295 */
 296void clear_nlink(struct inode *inode)
 297{
 298	if (inode->i_nlink) {
 299		inode->__i_nlink = 0;
 300		atomic_long_inc(&inode->i_sb->s_remove_count);
 301	}
 302}
 303EXPORT_SYMBOL(clear_nlink);
 304
 305/**
 306 * set_nlink - directly set an inode's link count
 307 * @inode: inode
 308 * @nlink: new nlink (should be non-zero)
 309 *
 310 * This is a low-level filesystem helper to replace any
 311 * direct filesystem manipulation of i_nlink.
 312 */
 313void set_nlink(struct inode *inode, unsigned int nlink)
 314{
 315	if (!nlink) {
 316		clear_nlink(inode);
 317	} else {
 318		/* Yes, some filesystems do change nlink from zero to one */
 319		if (inode->i_nlink == 0)
 320			atomic_long_dec(&inode->i_sb->s_remove_count);
 321
 322		inode->__i_nlink = nlink;
 323	}
 324}
 325EXPORT_SYMBOL(set_nlink);
 326
 327/**
 328 * inc_nlink - directly increment an inode's link count
 329 * @inode: inode
 330 *
 331 * This is a low-level filesystem helper to replace any
 332 * direct filesystem manipulation of i_nlink.  Currently,
 333 * it is only here for parity with dec_nlink().
 334 */
 335void inc_nlink(struct inode *inode)
 336{
 337	if (unlikely(inode->i_nlink == 0)) {
 338		WARN_ON(!(inode->i_state & I_LINKABLE));
 339		atomic_long_dec(&inode->i_sb->s_remove_count);
 340	}
 341
 342	inode->__i_nlink++;
 343}
 344EXPORT_SYMBOL(inc_nlink);
 345
 
 
 
 
 
 
 
 
 
 346void address_space_init_once(struct address_space *mapping)
 347{
 348	memset(mapping, 0, sizeof(*mapping));
 349	INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
 350	spin_lock_init(&mapping->tree_lock);
 351	mutex_init(&mapping->i_mmap_mutex);
 352	INIT_LIST_HEAD(&mapping->private_list);
 353	spin_lock_init(&mapping->private_lock);
 354	mapping->i_mmap = RB_ROOT;
 355	INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
 356}
 357EXPORT_SYMBOL(address_space_init_once);
 358
 359/*
 360 * These are initializations that only need to be done
 361 * once, because the fields are idempotent across use
 362 * of the inode, so let the slab aware of that.
 363 */
 364void inode_init_once(struct inode *inode)
 365{
 366	memset(inode, 0, sizeof(*inode));
 367	INIT_HLIST_NODE(&inode->i_hash);
 368	INIT_LIST_HEAD(&inode->i_devices);
 
 369	INIT_LIST_HEAD(&inode->i_wb_list);
 370	INIT_LIST_HEAD(&inode->i_lru);
 371	address_space_init_once(&inode->i_data);
 
 372	i_size_ordered_init(inode);
 373#ifdef CONFIG_FSNOTIFY
 374	INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
 375#endif
 376}
 377EXPORT_SYMBOL(inode_init_once);
 378
 379static void init_once(void *foo)
 380{
 381	struct inode *inode = (struct inode *) foo;
 382
 383	inode_init_once(inode);
 384}
 385
 386/*
 387 * inode->i_lock must be held
 388 */
 389void __iget(struct inode *inode)
 390{
 391	atomic_inc(&inode->i_count);
 392}
 393
 394/*
 395 * get additional reference to inode; caller must already hold one.
 396 */
 397void ihold(struct inode *inode)
 398{
 399	WARN_ON(atomic_inc_return(&inode->i_count) < 2);
 400}
 401EXPORT_SYMBOL(ihold);
 402
 403static void inode_lru_list_add(struct inode *inode)
 404{
 405	if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
 
 
 
 
 
 
 
 
 
 406		this_cpu_inc(nr_unused);
 
 
 407}
 408
 
 
 
 
 
 
 
 
 
 
 
 409/*
 410 * Add inode to LRU if needed (inode is unused and clean).
 411 *
 412 * Needs inode->i_lock held.
 413 */
 414void inode_add_lru(struct inode *inode)
 415{
 416	if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) &&
 417	    !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
 418		inode_lru_list_add(inode);
 419}
 420
 
 
 
 
 
 
 
 
 
 
 
 
 421
 422static void inode_lru_list_del(struct inode *inode)
 
 
 
 
 
 
 
 
 
 
 423{
 
 
 
 
 
 
 424
 425	if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
 426		this_cpu_dec(nr_unused);
 
 
 
 
 
 
 
 
 
 
 
 
 
 427}
 428
 429/**
 430 * inode_sb_list_add - add inode to the superblock list of inodes
 431 * @inode: inode to add
 432 */
 433void inode_sb_list_add(struct inode *inode)
 434{
 435	spin_lock(&inode_sb_list_lock);
 436	list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
 437	spin_unlock(&inode_sb_list_lock);
 438}
 439EXPORT_SYMBOL_GPL(inode_sb_list_add);
 440
 441static inline void inode_sb_list_del(struct inode *inode)
 442{
 443	if (!list_empty(&inode->i_sb_list)) {
 444		spin_lock(&inode_sb_list_lock);
 445		list_del_init(&inode->i_sb_list);
 446		spin_unlock(&inode_sb_list_lock);
 447	}
 448}
 449
 450static unsigned long hash(struct super_block *sb, unsigned long hashval)
 451{
 452	unsigned long tmp;
 453
 454	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
 455			L1_CACHE_BYTES;
 456	tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
 457	return tmp & i_hash_mask;
 458}
 459
 460/**
 461 *	__insert_inode_hash - hash an inode
 462 *	@inode: unhashed inode
 463 *	@hashval: unsigned long value used to locate this object in the
 464 *		inode_hashtable.
 465 *
 466 *	Add an inode to the inode hash for this superblock.
 467 */
 468void __insert_inode_hash(struct inode *inode, unsigned long hashval)
 469{
 470	struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
 471
 472	spin_lock(&inode_hash_lock);
 473	spin_lock(&inode->i_lock);
 474	hlist_add_head(&inode->i_hash, b);
 475	spin_unlock(&inode->i_lock);
 476	spin_unlock(&inode_hash_lock);
 477}
 478EXPORT_SYMBOL(__insert_inode_hash);
 479
 480/**
 481 *	__remove_inode_hash - remove an inode from the hash
 482 *	@inode: inode to unhash
 483 *
 484 *	Remove an inode from the superblock.
 485 */
 486void __remove_inode_hash(struct inode *inode)
 487{
 488	spin_lock(&inode_hash_lock);
 489	spin_lock(&inode->i_lock);
 490	hlist_del_init(&inode->i_hash);
 491	spin_unlock(&inode->i_lock);
 492	spin_unlock(&inode_hash_lock);
 493}
 494EXPORT_SYMBOL(__remove_inode_hash);
 495
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 496void clear_inode(struct inode *inode)
 497{
 498	might_sleep();
 499	/*
 500	 * We have to cycle tree_lock here because reclaim can be still in the
 501	 * process of removing the last page (in __delete_from_page_cache())
 502	 * and we must not free mapping under it.
 503	 */
 504	spin_lock_irq(&inode->i_data.tree_lock);
 505	BUG_ON(inode->i_data.nrpages);
 506	BUG_ON(inode->i_data.nrshadows);
 507	spin_unlock_irq(&inode->i_data.tree_lock);
 508	BUG_ON(!list_empty(&inode->i_data.private_list));
 
 
 
 
 
 
 
 509	BUG_ON(!(inode->i_state & I_FREEING));
 510	BUG_ON(inode->i_state & I_CLEAR);
 
 511	/* don't need i_lock here, no concurrent mods to i_state */
 512	inode->i_state = I_FREEING | I_CLEAR;
 513}
 514EXPORT_SYMBOL(clear_inode);
 515
 516/*
 517 * Free the inode passed in, removing it from the lists it is still connected
 518 * to. We remove any pages still attached to the inode and wait for any IO that
 519 * is still in progress before finally destroying the inode.
 520 *
 521 * An inode must already be marked I_FREEING so that we avoid the inode being
 522 * moved back onto lists if we race with other code that manipulates the lists
 523 * (e.g. writeback_single_inode). The caller is responsible for setting this.
 524 *
 525 * An inode must already be removed from the LRU list before being evicted from
 526 * the cache. This should occur atomically with setting the I_FREEING state
 527 * flag, so no inodes here should ever be on the LRU when being evicted.
 528 */
 529static void evict(struct inode *inode)
 530{
 531	const struct super_operations *op = inode->i_sb->s_op;
 532
 533	BUG_ON(!(inode->i_state & I_FREEING));
 534	BUG_ON(!list_empty(&inode->i_lru));
 535
 536	if (!list_empty(&inode->i_wb_list))
 537		inode_wb_list_del(inode);
 538
 539	inode_sb_list_del(inode);
 540
 
 
 
 541	/*
 542	 * Wait for flusher thread to be done with the inode so that filesystem
 543	 * does not start destroying it while writeback is still running. Since
 544	 * the inode has I_FREEING set, flusher thread won't start new work on
 545	 * the inode.  We just have to wait for running writeback to finish.
 546	 */
 547	inode_wait_for_writeback(inode);
 
 548
 549	if (op->evict_inode) {
 550		op->evict_inode(inode);
 551	} else {
 552		truncate_inode_pages_final(&inode->i_data);
 553		clear_inode(inode);
 554	}
 555	if (S_ISBLK(inode->i_mode) && inode->i_bdev)
 556		bd_forget(inode);
 557	if (S_ISCHR(inode->i_mode) && inode->i_cdev)
 558		cd_forget(inode);
 559
 560	remove_inode_hash(inode);
 561
 
 
 
 
 
 
 
 
 
 
 562	spin_lock(&inode->i_lock);
 563	wake_up_bit(&inode->i_state, __I_NEW);
 
 
 
 
 
 
 564	BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
 565	spin_unlock(&inode->i_lock);
 566
 567	destroy_inode(inode);
 568}
 569
 570/*
 571 * dispose_list - dispose of the contents of a local list
 572 * @head: the head of the list to free
 573 *
 574 * Dispose-list gets a local list with local inodes in it, so it doesn't
 575 * need to worry about list corruption and SMP locks.
 576 */
 577static void dispose_list(struct list_head *head)
 578{
 579	while (!list_empty(head)) {
 580		struct inode *inode;
 581
 582		inode = list_first_entry(head, struct inode, i_lru);
 583		list_del_init(&inode->i_lru);
 584
 585		evict(inode);
 
 586	}
 587}
 588
 589/**
 590 * evict_inodes	- evict all evictable inodes for a superblock
 591 * @sb:		superblock to operate on
 592 *
 593 * Make sure that no inodes with zero refcount are retained.  This is
 594 * called by superblock shutdown after having MS_ACTIVE flag removed,
 595 * so any inode reaching zero refcount during or after that call will
 596 * be immediately evicted.
 597 */
 598void evict_inodes(struct super_block *sb)
 599{
 600	struct inode *inode, *next;
 601	LIST_HEAD(dispose);
 602
 603	spin_lock(&inode_sb_list_lock);
 
 604	list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 605		if (atomic_read(&inode->i_count))
 606			continue;
 607
 608		spin_lock(&inode->i_lock);
 
 
 
 
 609		if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 610			spin_unlock(&inode->i_lock);
 611			continue;
 612		}
 613
 614		inode->i_state |= I_FREEING;
 615		inode_lru_list_del(inode);
 616		spin_unlock(&inode->i_lock);
 617		list_add(&inode->i_lru, &dispose);
 
 
 
 
 
 
 
 
 
 
 
 
 618	}
 619	spin_unlock(&inode_sb_list_lock);
 620
 621	dispose_list(&dispose);
 622}
 
 623
 624/**
 625 * invalidate_inodes	- attempt to free all inodes on a superblock
 626 * @sb:		superblock to operate on
 627 * @kill_dirty: flag to guide handling of dirty inodes
 628 *
 629 * Attempts to free all inodes for a given superblock.  If there were any
 630 * busy inodes return a non-zero value, else zero.
 631 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
 632 * them as busy.
 633 */
 634int invalidate_inodes(struct super_block *sb, bool kill_dirty)
 635{
 636	int busy = 0;
 637	struct inode *inode, *next;
 638	LIST_HEAD(dispose);
 639
 640	spin_lock(&inode_sb_list_lock);
 
 641	list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 642		spin_lock(&inode->i_lock);
 643		if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 644			spin_unlock(&inode->i_lock);
 645			continue;
 646		}
 647		if (inode->i_state & I_DIRTY && !kill_dirty) {
 648			spin_unlock(&inode->i_lock);
 649			busy = 1;
 650			continue;
 651		}
 652		if (atomic_read(&inode->i_count)) {
 653			spin_unlock(&inode->i_lock);
 654			busy = 1;
 655			continue;
 656		}
 657
 658		inode->i_state |= I_FREEING;
 659		inode_lru_list_del(inode);
 660		spin_unlock(&inode->i_lock);
 661		list_add(&inode->i_lru, &dispose);
 
 
 
 
 
 
 662	}
 663	spin_unlock(&inode_sb_list_lock);
 664
 665	dispose_list(&dispose);
 666
 667	return busy;
 668}
 669
 670/*
 671 * Isolate the inode from the LRU in preparation for freeing it.
 672 *
 673 * Any inodes which are pinned purely because of attached pagecache have their
 674 * pagecache removed.  If the inode has metadata buffers attached to
 675 * mapping->private_list then try to remove them.
 676 *
 677 * If the inode has the I_REFERENCED flag set, then it means that it has been
 678 * used recently - the flag is set in iput_final(). When we encounter such an
 679 * inode, clear the flag and move it to the back of the LRU so it gets another
 680 * pass through the LRU before it gets reclaimed. This is necessary because of
 681 * the fact we are doing lazy LRU updates to minimise lock contention so the
 682 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
 683 * with this flag set because they are the inodes that are out of order.
 684 */
 685static enum lru_status
 686inode_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
 687{
 688	struct list_head *freeable = arg;
 689	struct inode	*inode = container_of(item, struct inode, i_lru);
 690
 691	/*
 692	 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
 693	 * If we fail to get the lock, just skip it.
 694	 */
 695	if (!spin_trylock(&inode->i_lock))
 696		return LRU_SKIP;
 697
 698	/*
 699	 * Referenced or dirty inodes are still in use. Give them another pass
 700	 * through the LRU as we canot reclaim them now.
 
 
 701	 */
 702	if (atomic_read(&inode->i_count) ||
 703	    (inode->i_state & ~I_REFERENCED)) {
 704		list_del_init(&inode->i_lru);
 
 705		spin_unlock(&inode->i_lock);
 706		this_cpu_dec(nr_unused);
 707		return LRU_REMOVED;
 708	}
 709
 710	/* recently referenced inodes get one more pass */
 711	if (inode->i_state & I_REFERENCED) {
 712		inode->i_state &= ~I_REFERENCED;
 713		spin_unlock(&inode->i_lock);
 714		return LRU_ROTATE;
 715	}
 716
 717	if (inode_has_buffers(inode) || inode->i_data.nrpages) {
 718		__iget(inode);
 
 
 
 
 
 719		spin_unlock(&inode->i_lock);
 720		spin_unlock(lru_lock);
 721		if (remove_inode_buffers(inode)) {
 722			unsigned long reap;
 723			reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
 724			if (current_is_kswapd())
 725				__count_vm_events(KSWAPD_INODESTEAL, reap);
 726			else
 727				__count_vm_events(PGINODESTEAL, reap);
 728			if (current->reclaim_state)
 729				current->reclaim_state->reclaimed_slab += reap;
 730		}
 731		iput(inode);
 732		spin_lock(lru_lock);
 733		return LRU_RETRY;
 734	}
 735
 736	WARN_ON(inode->i_state & I_NEW);
 737	inode->i_state |= I_FREEING;
 738	list_move(&inode->i_lru, freeable);
 739	spin_unlock(&inode->i_lock);
 740
 741	this_cpu_dec(nr_unused);
 742	return LRU_REMOVED;
 743}
 744
 745/*
 746 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
 747 * This is called from the superblock shrinker function with a number of inodes
 748 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
 749 * then are freed outside inode_lock by dispose_list().
 750 */
 751long prune_icache_sb(struct super_block *sb, unsigned long nr_to_scan,
 752		     int nid)
 753{
 754	LIST_HEAD(freeable);
 755	long freed;
 756
 757	freed = list_lru_walk_node(&sb->s_inode_lru, nid, inode_lru_isolate,
 758				       &freeable, &nr_to_scan);
 759	dispose_list(&freeable);
 760	return freed;
 761}
 762
 763static void __wait_on_freeing_inode(struct inode *inode);
 764/*
 765 * Called with the inode lock held.
 766 */
 767static struct inode *find_inode(struct super_block *sb,
 768				struct hlist_head *head,
 769				int (*test)(struct inode *, void *),
 770				void *data)
 771{
 772	struct inode *inode = NULL;
 773
 
 
 
 
 
 
 774repeat:
 775	hlist_for_each_entry(inode, head, i_hash) {
 776		if (inode->i_sb != sb)
 777			continue;
 778		if (!test(inode, data))
 779			continue;
 780		spin_lock(&inode->i_lock);
 781		if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 782			__wait_on_freeing_inode(inode);
 783			goto repeat;
 784		}
 
 
 
 
 
 785		__iget(inode);
 786		spin_unlock(&inode->i_lock);
 
 787		return inode;
 788	}
 
 789	return NULL;
 790}
 791
 792/*
 793 * find_inode_fast is the fast path version of find_inode, see the comment at
 794 * iget_locked for details.
 795 */
 796static struct inode *find_inode_fast(struct super_block *sb,
 797				struct hlist_head *head, unsigned long ino)
 
 798{
 799	struct inode *inode = NULL;
 800
 
 
 
 
 
 
 801repeat:
 802	hlist_for_each_entry(inode, head, i_hash) {
 803		if (inode->i_ino != ino)
 804			continue;
 805		if (inode->i_sb != sb)
 806			continue;
 807		spin_lock(&inode->i_lock);
 808		if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 809			__wait_on_freeing_inode(inode);
 810			goto repeat;
 811		}
 
 
 
 
 
 812		__iget(inode);
 813		spin_unlock(&inode->i_lock);
 
 814		return inode;
 815	}
 
 816	return NULL;
 817}
 818
 819/*
 820 * Each cpu owns a range of LAST_INO_BATCH numbers.
 821 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
 822 * to renew the exhausted range.
 823 *
 824 * This does not significantly increase overflow rate because every CPU can
 825 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
 826 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
 827 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
 828 * overflow rate by 2x, which does not seem too significant.
 829 *
 830 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
 831 * error if st_ino won't fit in target struct field. Use 32bit counter
 832 * here to attempt to avoid that.
 833 */
 834#define LAST_INO_BATCH 1024
 835static DEFINE_PER_CPU(unsigned int, last_ino);
 836
 837unsigned int get_next_ino(void)
 838{
 839	unsigned int *p = &get_cpu_var(last_ino);
 840	unsigned int res = *p;
 841
 842#ifdef CONFIG_SMP
 843	if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
 844		static atomic_t shared_last_ino;
 845		int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
 846
 847		res = next - LAST_INO_BATCH;
 848	}
 849#endif
 850
 851	*p = ++res;
 
 
 
 
 852	put_cpu_var(last_ino);
 853	return res;
 854}
 855EXPORT_SYMBOL(get_next_ino);
 856
 857/**
 858 *	new_inode_pseudo 	- obtain an inode
 859 *	@sb: superblock
 860 *
 861 *	Allocates a new inode for given superblock.
 862 *	Inode wont be chained in superblock s_inodes list
 863 *	This means :
 864 *	- fs can't be unmount
 865 *	- quotas, fsnotify, writeback can't work
 866 */
 867struct inode *new_inode_pseudo(struct super_block *sb)
 868{
 869	struct inode *inode = alloc_inode(sb);
 870
 871	if (inode) {
 872		spin_lock(&inode->i_lock);
 873		inode->i_state = 0;
 874		spin_unlock(&inode->i_lock);
 875		INIT_LIST_HEAD(&inode->i_sb_list);
 876	}
 877	return inode;
 878}
 879
 880/**
 881 *	new_inode 	- obtain an inode
 882 *	@sb: superblock
 883 *
 884 *	Allocates a new inode for given superblock. The default gfp_mask
 885 *	for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
 886 *	If HIGHMEM pages are unsuitable or it is known that pages allocated
 887 *	for the page cache are not reclaimable or migratable,
 888 *	mapping_set_gfp_mask() must be called with suitable flags on the
 889 *	newly created inode's mapping
 890 *
 891 */
 892struct inode *new_inode(struct super_block *sb)
 893{
 894	struct inode *inode;
 895
 896	spin_lock_prefetch(&inode_sb_list_lock);
 897
 898	inode = new_inode_pseudo(sb);
 899	if (inode)
 900		inode_sb_list_add(inode);
 901	return inode;
 902}
 903EXPORT_SYMBOL(new_inode);
 904
 905#ifdef CONFIG_DEBUG_LOCK_ALLOC
 906void lockdep_annotate_inode_mutex_key(struct inode *inode)
 907{
 908	if (S_ISDIR(inode->i_mode)) {
 909		struct file_system_type *type = inode->i_sb->s_type;
 910
 911		/* Set new key only if filesystem hasn't already changed it */
 912		if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
 913			/*
 914			 * ensure nobody is actually holding i_mutex
 915			 */
 916			mutex_destroy(&inode->i_mutex);
 917			mutex_init(&inode->i_mutex);
 918			lockdep_set_class(&inode->i_mutex,
 919					  &type->i_mutex_dir_key);
 920		}
 921	}
 922}
 923EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
 924#endif
 925
 926/**
 927 * unlock_new_inode - clear the I_NEW state and wake up any waiters
 928 * @inode:	new inode to unlock
 929 *
 930 * Called when the inode is fully initialised to clear the new state of the
 931 * inode and wake up anyone waiting for the inode to finish initialisation.
 932 */
 933void unlock_new_inode(struct inode *inode)
 934{
 935	lockdep_annotate_inode_mutex_key(inode);
 936	spin_lock(&inode->i_lock);
 937	WARN_ON(!(inode->i_state & I_NEW));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 938	inode->i_state &= ~I_NEW;
 
 
 
 
 
 939	smp_mb();
 940	wake_up_bit(&inode->i_state, __I_NEW);
 941	spin_unlock(&inode->i_lock);
 
 942}
 943EXPORT_SYMBOL(unlock_new_inode);
 944
 945/**
 946 * lock_two_nondirectories - take two i_mutexes on non-directory objects
 947 *
 948 * Lock any non-NULL argument that is not a directory.
 949 * Zero, one or two objects may be locked by this function.
 950 *
 951 * @inode1: first inode to lock
 952 * @inode2: second inode to lock
 953 */
 954void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
 955{
 
 
 
 
 956	if (inode1 > inode2)
 957		swap(inode1, inode2);
 958
 959	if (inode1 && !S_ISDIR(inode1->i_mode))
 960		mutex_lock(&inode1->i_mutex);
 961	if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
 962		mutex_lock_nested(&inode2->i_mutex, I_MUTEX_NONDIR2);
 963}
 964EXPORT_SYMBOL(lock_two_nondirectories);
 965
 966/**
 967 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
 968 * @inode1: first inode to unlock
 969 * @inode2: second inode to unlock
 970 */
 971void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
 972{
 973	if (inode1 && !S_ISDIR(inode1->i_mode))
 974		mutex_unlock(&inode1->i_mutex);
 975	if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
 976		mutex_unlock(&inode2->i_mutex);
 
 
 
 
 977}
 978EXPORT_SYMBOL(unlock_two_nondirectories);
 979
 980/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 981 * iget5_locked - obtain an inode from a mounted file system
 982 * @sb:		super block of file system
 983 * @hashval:	hash value (usually inode number) to get
 984 * @test:	callback used for comparisons between inodes
 985 * @set:	callback used to initialize a new struct inode
 986 * @data:	opaque data pointer to pass to @test and @set
 987 *
 988 * Search for the inode specified by @hashval and @data in the inode cache,
 989 * and if present it is return it with an increased reference count. This is
 990 * a generalized version of iget_locked() for file systems where the inode
 991 * number is not sufficient for unique identification of an inode.
 992 *
 993 * If the inode is not in cache, allocate a new inode and return it locked,
 994 * hashed, and with the I_NEW flag set. The file system gets to fill it in
 995 * before unlocking it via unlock_new_inode().
 996 *
 997 * Note both @test and @set are called with the inode_hash_lock held, so can't
 998 * sleep.
 999 */
1000struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1001		int (*test)(struct inode *, void *),
1002		int (*set)(struct inode *, void *), void *data)
1003{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1004	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1005	struct inode *inode;
1006
1007	spin_lock(&inode_hash_lock);
1008	inode = find_inode(sb, head, test, data);
1009	spin_unlock(&inode_hash_lock);
1010
 
 
1011	if (inode) {
 
 
1012		wait_on_inode(inode);
 
 
 
 
1013		return inode;
1014	}
1015
1016	inode = alloc_inode(sb);
1017	if (inode) {
1018		struct inode *old;
1019
1020		spin_lock(&inode_hash_lock);
1021		/* We released the lock, so.. */
1022		old = find_inode(sb, head, test, data);
1023		if (!old) {
1024			if (set(inode, data))
1025				goto set_failed;
1026
1027			spin_lock(&inode->i_lock);
1028			inode->i_state = I_NEW;
1029			hlist_add_head(&inode->i_hash, head);
1030			spin_unlock(&inode->i_lock);
1031			inode_sb_list_add(inode);
1032			spin_unlock(&inode_hash_lock);
1033
1034			/* Return the locked inode with I_NEW set, the
1035			 * caller is responsible for filling in the contents
1036			 */
1037			return inode;
1038		}
1039
1040		/*
1041		 * Uhhuh, somebody else created the same inode under
1042		 * us. Use the old inode instead of the one we just
1043		 * allocated.
1044		 */
1045		spin_unlock(&inode_hash_lock);
1046		destroy_inode(inode);
1047		inode = old;
1048		wait_on_inode(inode);
1049	}
1050	return inode;
1051
1052set_failed:
1053	spin_unlock(&inode_hash_lock);
1054	destroy_inode(inode);
1055	return NULL;
1056}
1057EXPORT_SYMBOL(iget5_locked);
1058
1059/**
1060 * iget_locked - obtain an inode from a mounted file system
1061 * @sb:		super block of file system
1062 * @ino:	inode number to get
1063 *
1064 * Search for the inode specified by @ino in the inode cache and if present
1065 * return it with an increased reference count. This is for file systems
1066 * where the inode number is sufficient for unique identification of an inode.
1067 *
1068 * If the inode is not in cache, allocate a new inode and return it locked,
1069 * hashed, and with the I_NEW flag set.  The file system gets to fill it in
1070 * before unlocking it via unlock_new_inode().
1071 */
1072struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1073{
1074	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1075	struct inode *inode;
1076
1077	spin_lock(&inode_hash_lock);
1078	inode = find_inode_fast(sb, head, ino);
1079	spin_unlock(&inode_hash_lock);
1080	if (inode) {
 
 
1081		wait_on_inode(inode);
 
 
 
 
1082		return inode;
1083	}
1084
1085	inode = alloc_inode(sb);
1086	if (inode) {
1087		struct inode *old;
1088
1089		spin_lock(&inode_hash_lock);
1090		/* We released the lock, so.. */
1091		old = find_inode_fast(sb, head, ino);
1092		if (!old) {
1093			inode->i_ino = ino;
1094			spin_lock(&inode->i_lock);
1095			inode->i_state = I_NEW;
1096			hlist_add_head(&inode->i_hash, head);
1097			spin_unlock(&inode->i_lock);
1098			inode_sb_list_add(inode);
1099			spin_unlock(&inode_hash_lock);
1100
1101			/* Return the locked inode with I_NEW set, the
1102			 * caller is responsible for filling in the contents
1103			 */
1104			return inode;
1105		}
1106
1107		/*
1108		 * Uhhuh, somebody else created the same inode under
1109		 * us. Use the old inode instead of the one we just
1110		 * allocated.
1111		 */
1112		spin_unlock(&inode_hash_lock);
1113		destroy_inode(inode);
 
 
1114		inode = old;
1115		wait_on_inode(inode);
 
 
 
 
1116	}
1117	return inode;
1118}
1119EXPORT_SYMBOL(iget_locked);
1120
1121/*
1122 * search the inode cache for a matching inode number.
1123 * If we find one, then the inode number we are trying to
1124 * allocate is not unique and so we should not use it.
1125 *
1126 * Returns 1 if the inode number is unique, 0 if it is not.
1127 */
1128static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1129{
1130	struct hlist_head *b = inode_hashtable + hash(sb, ino);
1131	struct inode *inode;
1132
1133	spin_lock(&inode_hash_lock);
1134	hlist_for_each_entry(inode, b, i_hash) {
1135		if (inode->i_ino == ino && inode->i_sb == sb) {
1136			spin_unlock(&inode_hash_lock);
1137			return 0;
1138		}
1139	}
1140	spin_unlock(&inode_hash_lock);
1141
1142	return 1;
1143}
1144
1145/**
1146 *	iunique - get a unique inode number
1147 *	@sb: superblock
1148 *	@max_reserved: highest reserved inode number
1149 *
1150 *	Obtain an inode number that is unique on the system for a given
1151 *	superblock. This is used by file systems that have no natural
1152 *	permanent inode numbering system. An inode number is returned that
1153 *	is higher than the reserved limit but unique.
1154 *
1155 *	BUGS:
1156 *	With a large number of inodes live on the file system this function
1157 *	currently becomes quite slow.
1158 */
1159ino_t iunique(struct super_block *sb, ino_t max_reserved)
1160{
1161	/*
1162	 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1163	 * error if st_ino won't fit in target struct field. Use 32bit counter
1164	 * here to attempt to avoid that.
1165	 */
1166	static DEFINE_SPINLOCK(iunique_lock);
1167	static unsigned int counter;
1168	ino_t res;
1169
 
1170	spin_lock(&iunique_lock);
1171	do {
1172		if (counter <= max_reserved)
1173			counter = max_reserved + 1;
1174		res = counter++;
1175	} while (!test_inode_iunique(sb, res));
1176	spin_unlock(&iunique_lock);
 
1177
1178	return res;
1179}
1180EXPORT_SYMBOL(iunique);
1181
1182struct inode *igrab(struct inode *inode)
1183{
1184	spin_lock(&inode->i_lock);
1185	if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1186		__iget(inode);
1187		spin_unlock(&inode->i_lock);
1188	} else {
1189		spin_unlock(&inode->i_lock);
1190		/*
1191		 * Handle the case where s_op->clear_inode is not been
1192		 * called yet, and somebody is calling igrab
1193		 * while the inode is getting freed.
1194		 */
1195		inode = NULL;
1196	}
1197	return inode;
1198}
1199EXPORT_SYMBOL(igrab);
1200
1201/**
1202 * ilookup5_nowait - search for an inode in the inode cache
1203 * @sb:		super block of file system to search
1204 * @hashval:	hash value (usually inode number) to search for
1205 * @test:	callback used for comparisons between inodes
1206 * @data:	opaque data pointer to pass to @test
1207 *
1208 * Search for the inode specified by @hashval and @data in the inode cache.
1209 * If the inode is in the cache, the inode is returned with an incremented
1210 * reference count.
1211 *
1212 * Note: I_NEW is not waited upon so you have to be very careful what you do
1213 * with the returned inode.  You probably should be using ilookup5() instead.
1214 *
1215 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1216 */
1217struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1218		int (*test)(struct inode *, void *), void *data)
1219{
1220	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1221	struct inode *inode;
1222
1223	spin_lock(&inode_hash_lock);
1224	inode = find_inode(sb, head, test, data);
1225	spin_unlock(&inode_hash_lock);
1226
1227	return inode;
1228}
1229EXPORT_SYMBOL(ilookup5_nowait);
1230
1231/**
1232 * ilookup5 - search for an inode in the inode cache
1233 * @sb:		super block of file system to search
1234 * @hashval:	hash value (usually inode number) to search for
1235 * @test:	callback used for comparisons between inodes
1236 * @data:	opaque data pointer to pass to @test
1237 *
1238 * Search for the inode specified by @hashval and @data in the inode cache,
1239 * and if the inode is in the cache, return the inode with an incremented
1240 * reference count.  Waits on I_NEW before returning the inode.
1241 * returned with an incremented reference count.
1242 *
1243 * This is a generalized version of ilookup() for file systems where the
1244 * inode number is not sufficient for unique identification of an inode.
1245 *
1246 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1247 */
1248struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1249		int (*test)(struct inode *, void *), void *data)
1250{
1251	struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1252
1253	if (inode)
 
1254		wait_on_inode(inode);
 
 
 
 
 
1255	return inode;
1256}
1257EXPORT_SYMBOL(ilookup5);
1258
1259/**
1260 * ilookup - search for an inode in the inode cache
1261 * @sb:		super block of file system to search
1262 * @ino:	inode number to search for
1263 *
1264 * Search for the inode @ino in the inode cache, and if the inode is in the
1265 * cache, the inode is returned with an incremented reference count.
1266 */
1267struct inode *ilookup(struct super_block *sb, unsigned long ino)
1268{
1269	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1270	struct inode *inode;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1271
1272	spin_lock(&inode_hash_lock);
1273	inode = find_inode_fast(sb, head, ino);
 
 
 
 
 
 
 
 
 
 
1274	spin_unlock(&inode_hash_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1275
1276	if (inode)
1277		wait_on_inode(inode);
1278	return inode;
 
 
 
 
1279}
1280EXPORT_SYMBOL(ilookup);
1281
1282int insert_inode_locked(struct inode *inode)
1283{
1284	struct super_block *sb = inode->i_sb;
1285	ino_t ino = inode->i_ino;
1286	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1287
1288	while (1) {
1289		struct inode *old = NULL;
1290		spin_lock(&inode_hash_lock);
1291		hlist_for_each_entry(old, head, i_hash) {
1292			if (old->i_ino != ino)
1293				continue;
1294			if (old->i_sb != sb)
1295				continue;
1296			spin_lock(&old->i_lock);
1297			if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1298				spin_unlock(&old->i_lock);
1299				continue;
1300			}
1301			break;
1302		}
1303		if (likely(!old)) {
1304			spin_lock(&inode->i_lock);
1305			inode->i_state |= I_NEW;
1306			hlist_add_head(&inode->i_hash, head);
1307			spin_unlock(&inode->i_lock);
1308			spin_unlock(&inode_hash_lock);
1309			return 0;
1310		}
 
 
 
 
 
1311		__iget(old);
1312		spin_unlock(&old->i_lock);
1313		spin_unlock(&inode_hash_lock);
1314		wait_on_inode(old);
1315		if (unlikely(!inode_unhashed(old))) {
1316			iput(old);
1317			return -EBUSY;
1318		}
1319		iput(old);
1320	}
1321}
1322EXPORT_SYMBOL(insert_inode_locked);
1323
1324int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1325		int (*test)(struct inode *, void *), void *data)
1326{
1327	struct super_block *sb = inode->i_sb;
1328	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1329
1330	while (1) {
1331		struct inode *old = NULL;
1332
1333		spin_lock(&inode_hash_lock);
1334		hlist_for_each_entry(old, head, i_hash) {
1335			if (old->i_sb != sb)
1336				continue;
1337			if (!test(old, data))
1338				continue;
1339			spin_lock(&old->i_lock);
1340			if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1341				spin_unlock(&old->i_lock);
1342				continue;
1343			}
1344			break;
1345		}
1346		if (likely(!old)) {
1347			spin_lock(&inode->i_lock);
1348			inode->i_state |= I_NEW;
1349			hlist_add_head(&inode->i_hash, head);
1350			spin_unlock(&inode->i_lock);
1351			spin_unlock(&inode_hash_lock);
1352			return 0;
1353		}
1354		__iget(old);
1355		spin_unlock(&old->i_lock);
1356		spin_unlock(&inode_hash_lock);
1357		wait_on_inode(old);
1358		if (unlikely(!inode_unhashed(old))) {
1359			iput(old);
1360			return -EBUSY;
1361		}
1362		iput(old);
 
1363	}
 
1364}
1365EXPORT_SYMBOL(insert_inode_locked4);
1366
1367
1368int generic_delete_inode(struct inode *inode)
1369{
1370	return 1;
1371}
1372EXPORT_SYMBOL(generic_delete_inode);
1373
1374/*
1375 * Called when we're dropping the last reference
1376 * to an inode.
1377 *
1378 * Call the FS "drop_inode()" function, defaulting to
1379 * the legacy UNIX filesystem behaviour.  If it tells
1380 * us to evict inode, do so.  Otherwise, retain inode
1381 * in cache if fs is alive, sync and evict if fs is
1382 * shutting down.
1383 */
1384static void iput_final(struct inode *inode)
1385{
1386	struct super_block *sb = inode->i_sb;
1387	const struct super_operations *op = inode->i_sb->s_op;
 
1388	int drop;
1389
1390	WARN_ON(inode->i_state & I_NEW);
1391
1392	if (op->drop_inode)
1393		drop = op->drop_inode(inode);
1394	else
1395		drop = generic_drop_inode(inode);
1396
1397	if (!drop && (sb->s_flags & MS_ACTIVE)) {
1398		inode->i_state |= I_REFERENCED;
1399		inode_add_lru(inode);
 
1400		spin_unlock(&inode->i_lock);
1401		return;
1402	}
1403
 
1404	if (!drop) {
1405		inode->i_state |= I_WILL_FREE;
1406		spin_unlock(&inode->i_lock);
 
1407		write_inode_now(inode, 1);
 
1408		spin_lock(&inode->i_lock);
1409		WARN_ON(inode->i_state & I_NEW);
1410		inode->i_state &= ~I_WILL_FREE;
 
1411	}
1412
1413	inode->i_state |= I_FREEING;
1414	if (!list_empty(&inode->i_lru))
1415		inode_lru_list_del(inode);
1416	spin_unlock(&inode->i_lock);
1417
1418	evict(inode);
1419}
1420
1421/**
1422 *	iput	- put an inode
1423 *	@inode: inode to put
1424 *
1425 *	Puts an inode, dropping its usage count. If the inode use count hits
1426 *	zero, the inode is then freed and may also be destroyed.
1427 *
1428 *	Consequently, iput() can sleep.
1429 */
1430void iput(struct inode *inode)
1431{
1432	if (inode) {
1433		BUG_ON(inode->i_state & I_CLEAR);
1434
1435		if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1436			iput_final(inode);
 
 
 
 
 
 
 
 
1437	}
1438}
1439EXPORT_SYMBOL(iput);
1440
 
1441/**
1442 *	bmap	- find a block number in a file
1443 *	@inode: inode of file
1444 *	@block: block to find
 
 
 
 
 
 
1445 *
1446 *	Returns the block number on the device holding the inode that
1447 *	is the disk block number for the block of the file requested.
1448 *	That is, asked for block 4 of inode 1 the function will return the
1449 *	disk block relative to the disk start that holds that block of the
1450 *	file.
1451 */
1452sector_t bmap(struct inode *inode, sector_t block)
1453{
1454	sector_t res = 0;
1455	if (inode->i_mapping->a_ops->bmap)
1456		res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1457	return res;
1458}
1459EXPORT_SYMBOL(bmap);
 
1460
1461/*
1462 * With relative atime, only update atime if the previous atime is
1463 * earlier than either the ctime or mtime or if at least a day has
1464 * passed since the last atime update.
1465 */
1466static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1467			     struct timespec now)
1468{
 
1469
1470	if (!(mnt->mnt_flags & MNT_RELATIME))
1471		return 1;
1472	/*
1473	 * Is mtime younger than atime? If yes, update atime:
1474	 */
1475	if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1476		return 1;
 
 
1477	/*
1478	 * Is ctime younger than atime? If yes, update atime:
1479	 */
1480	if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1481		return 1;
 
1482
1483	/*
1484	 * Is the previous atime value older than a day? If yes,
1485	 * update atime:
1486	 */
1487	if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1488		return 1;
1489	/*
1490	 * Good, we can skip the atime update:
1491	 */
1492	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1493}
 
1494
1495/*
1496 * This does the actual work of updating an inodes time or version.  Must have
1497 * had called mnt_want_write() before calling this.
1498 */
1499static int update_time(struct inode *inode, struct timespec *time, int flags)
1500{
1501	if (inode->i_op->update_time)
1502		return inode->i_op->update_time(inode, time, flags);
1503
1504	if (flags & S_ATIME)
1505		inode->i_atime = *time;
1506	if (flags & S_VERSION)
1507		inode_inc_iversion(inode);
1508	if (flags & S_CTIME)
1509		inode->i_ctime = *time;
1510	if (flags & S_MTIME)
1511		inode->i_mtime = *time;
1512	mark_inode_dirty_sync(inode);
1513	return 0;
1514}
 
1515
1516/**
1517 *	touch_atime	-	update the access time
1518 *	@path: the &struct path to update
 
1519 *
1520 *	Update the accessed time on an inode and mark it for writeback.
1521 *	This function automatically handles read only file systems and media,
1522 *	as well as the "noatime" flag and inode specific "noatime" markers.
1523 */
1524void touch_atime(const struct path *path)
1525{
1526	struct vfsmount *mnt = path->mnt;
1527	struct inode *inode = path->dentry->d_inode;
1528	struct timespec now;
1529
1530	if (inode->i_flags & S_NOATIME)
1531		return;
 
 
 
 
 
 
 
1532	if (IS_NOATIME(inode))
1533		return;
1534	if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1535		return;
1536
1537	if (mnt->mnt_flags & MNT_NOATIME)
1538		return;
1539	if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1540		return;
1541
1542	now = current_fs_time(inode->i_sb);
1543
1544	if (!relatime_need_update(mnt, inode, now))
1545		return;
 
 
 
 
 
 
 
 
 
 
 
 
1546
1547	if (timespec_equal(&inode->i_atime, &now))
1548		return;
1549
1550	if (!sb_start_write_trylock(inode->i_sb))
1551		return;
1552
1553	if (__mnt_want_write(mnt))
1554		goto skip_update;
1555	/*
1556	 * File systems can error out when updating inodes if they need to
1557	 * allocate new space to modify an inode (such is the case for
1558	 * Btrfs), but since we touch atime while walking down the path we
1559	 * really don't care if we failed to update the atime of the file,
1560	 * so just ignore the return value.
1561	 * We may also fail on filesystems that have the ability to make parts
1562	 * of the fs read only, e.g. subvolumes in Btrfs.
1563	 */
1564	update_time(inode, &now, S_ATIME);
1565	__mnt_drop_write(mnt);
1566skip_update:
1567	sb_end_write(inode->i_sb);
1568}
1569EXPORT_SYMBOL(touch_atime);
1570
1571/*
1572 * The logic we want is
1573 *
1574 *	if suid or (sgid and xgrp)
1575 *		remove privs
1576 */
1577int should_remove_suid(struct dentry *dentry)
 
1578{
1579	umode_t mode = dentry->d_inode->i_mode;
1580	int kill = 0;
 
1581
1582	/* suid always must be killed */
1583	if (unlikely(mode & S_ISUID))
1584		kill = ATTR_KILL_SUID;
1585
1586	/*
1587	 * sgid without any exec bits is just a mandatory locking mark; leave
1588	 * it alone.  If some exec bits are set, it's a real sgid; kill it.
1589	 */
1590	if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1591		kill |= ATTR_KILL_SGID;
1592
1593	if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1594		return kill;
1595
1596	return 0;
 
 
 
 
 
 
1597}
1598EXPORT_SYMBOL(should_remove_suid);
1599
1600static int __remove_suid(struct dentry *dentry, int kill)
 
1601{
1602	struct iattr newattrs;
1603
1604	newattrs.ia_valid = ATTR_FORCE | kill;
1605	/*
1606	 * Note we call this on write, so notify_change will not
1607	 * encounter any conflicting delegations:
1608	 */
1609	return notify_change(dentry, &newattrs, NULL);
1610}
1611
1612int file_remove_suid(struct file *file)
1613{
1614	struct dentry *dentry = file->f_path.dentry;
1615	struct inode *inode = dentry->d_inode;
1616	int killsuid;
1617	int killpriv;
1618	int error = 0;
 
1619
1620	/* Fast path for nothing security related */
1621	if (IS_NOSEC(inode))
1622		return 0;
1623
1624	killsuid = should_remove_suid(dentry);
1625	killpriv = security_inode_need_killpriv(dentry);
 
1626
1627	if (killpriv < 0)
1628		return killpriv;
1629	if (killpriv)
1630		error = security_inode_killpriv(dentry);
1631	if (!error && killsuid)
1632		error = __remove_suid(dentry, killsuid);
1633	if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1634		inode->i_flags |= S_NOSEC;
1635
 
 
 
 
 
1636	return error;
1637}
1638EXPORT_SYMBOL(file_remove_suid);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1639
1640/**
1641 *	file_update_time	-	update mtime and ctime time
1642 *	@file: file accessed
1643 *
1644 *	Update the mtime and ctime members of an inode and mark the inode
1645 *	for writeback.  Note that this function is meant exclusively for
1646 *	usage in the file write path of filesystems, and filesystems may
1647 *	choose to explicitly ignore update via this function with the
1648 *	S_NOCMTIME inode flag, e.g. for network filesystem where these
1649 *	timestamps are handled by the server.  This can return an error for
1650 *	file systems who need to allocate space in order to update an inode.
 
1651 */
 
 
 
 
1652
1653int file_update_time(struct file *file)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1654{
1655	struct inode *inode = file_inode(file);
1656	struct timespec now;
1657	int sync_it = 0;
1658	int ret;
1659
1660	/* First try to exhaust all avenues to not sync */
1661	if (IS_NOCMTIME(inode))
1662		return 0;
1663
1664	now = current_fs_time(inode->i_sb);
1665	if (!timespec_equal(&inode->i_mtime, &now))
1666		sync_it = S_MTIME;
 
 
1667
1668	if (!timespec_equal(&inode->i_ctime, &now))
 
1669		sync_it |= S_CTIME;
1670
1671	if (IS_I_VERSION(inode))
1672		sync_it |= S_VERSION;
1673
1674	if (!sync_it)
1675		return 0;
1676
1677	/* Finally allowed to write? Takes lock. */
1678	if (__mnt_want_write_file(file))
1679		return 0;
 
1680
1681	ret = update_time(inode, &now, sync_it);
1682	__mnt_drop_write_file(file);
 
 
 
1683
1684	return ret;
1685}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1686EXPORT_SYMBOL(file_update_time);
1687
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1688int inode_needs_sync(struct inode *inode)
1689{
1690	if (IS_SYNC(inode))
1691		return 1;
1692	if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1693		return 1;
1694	return 0;
1695}
1696EXPORT_SYMBOL(inode_needs_sync);
1697
1698int inode_wait(void *word)
1699{
1700	schedule();
1701	return 0;
1702}
1703EXPORT_SYMBOL(inode_wait);
1704
1705/*
1706 * If we try to find an inode in the inode hash while it is being
1707 * deleted, we have to wait until the filesystem completes its
1708 * deletion before reporting that it isn't found.  This function waits
1709 * until the deletion _might_ have completed.  Callers are responsible
1710 * to recheck inode state.
1711 *
1712 * It doesn't matter if I_NEW is not set initially, a call to
1713 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1714 * will DTRT.
1715 */
1716static void __wait_on_freeing_inode(struct inode *inode)
1717{
1718	wait_queue_head_t *wq;
1719	DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1720	wq = bit_waitqueue(&inode->i_state, __I_NEW);
1721	prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
 
 
 
 
 
 
 
 
 
 
1722	spin_unlock(&inode->i_lock);
1723	spin_unlock(&inode_hash_lock);
 
 
1724	schedule();
1725	finish_wait(wq, &wait.wait);
1726	spin_lock(&inode_hash_lock);
 
 
1727}
1728
1729static __initdata unsigned long ihash_entries;
1730static int __init set_ihash_entries(char *str)
1731{
1732	if (!str)
1733		return 0;
1734	ihash_entries = simple_strtoul(str, &str, 0);
1735	return 1;
1736}
1737__setup("ihash_entries=", set_ihash_entries);
1738
1739/*
1740 * Initialize the waitqueues and inode hash table.
1741 */
1742void __init inode_init_early(void)
1743{
1744	unsigned int loop;
1745
1746	/* If hashes are distributed across NUMA nodes, defer
1747	 * hash allocation until vmalloc space is available.
1748	 */
1749	if (hashdist)
1750		return;
1751
1752	inode_hashtable =
1753		alloc_large_system_hash("Inode-cache",
1754					sizeof(struct hlist_head),
1755					ihash_entries,
1756					14,
1757					HASH_EARLY,
1758					&i_hash_shift,
1759					&i_hash_mask,
1760					0,
1761					0);
1762
1763	for (loop = 0; loop < (1U << i_hash_shift); loop++)
1764		INIT_HLIST_HEAD(&inode_hashtable[loop]);
1765}
1766
1767void __init inode_init(void)
1768{
1769	unsigned int loop;
1770
1771	/* inode slab cache */
1772	inode_cachep = kmem_cache_create("inode_cache",
1773					 sizeof(struct inode),
1774					 0,
1775					 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1776					 SLAB_MEM_SPREAD),
1777					 init_once);
1778
1779	/* Hash may have been set up in inode_init_early */
1780	if (!hashdist)
1781		return;
1782
1783	inode_hashtable =
1784		alloc_large_system_hash("Inode-cache",
1785					sizeof(struct hlist_head),
1786					ihash_entries,
1787					14,
1788					0,
1789					&i_hash_shift,
1790					&i_hash_mask,
1791					0,
1792					0);
1793
1794	for (loop = 0; loop < (1U << i_hash_shift); loop++)
1795		INIT_HLIST_HEAD(&inode_hashtable[loop]);
1796}
1797
1798void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1799{
1800	inode->i_mode = mode;
1801	if (S_ISCHR(mode)) {
1802		inode->i_fop = &def_chr_fops;
1803		inode->i_rdev = rdev;
1804	} else if (S_ISBLK(mode)) {
1805		inode->i_fop = &def_blk_fops;
 
1806		inode->i_rdev = rdev;
1807	} else if (S_ISFIFO(mode))
1808		inode->i_fop = &pipefifo_fops;
1809	else if (S_ISSOCK(mode))
1810		inode->i_fop = &bad_sock_fops;
1811	else
1812		printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1813				  " inode %s:%lu\n", mode, inode->i_sb->s_id,
1814				  inode->i_ino);
1815}
1816EXPORT_SYMBOL(init_special_inode);
1817
1818/**
1819 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
 
1820 * @inode: New inode
1821 * @dir: Directory inode
1822 * @mode: mode of the new inode
 
 
 
 
 
 
1823 */
1824void inode_init_owner(struct inode *inode, const struct inode *dir,
1825			umode_t mode)
1826{
1827	inode->i_uid = current_fsuid();
1828	if (dir && dir->i_mode & S_ISGID) {
1829		inode->i_gid = dir->i_gid;
 
 
1830		if (S_ISDIR(mode))
1831			mode |= S_ISGID;
1832	} else
1833		inode->i_gid = current_fsgid();
1834	inode->i_mode = mode;
1835}
1836EXPORT_SYMBOL(inode_init_owner);
1837
1838/**
1839 * inode_owner_or_capable - check current task permissions to inode
 
1840 * @inode: inode being checked
1841 *
1842 * Return true if current either has CAP_FOWNER to the inode, or
1843 * owns the file.
 
 
 
 
 
 
1844 */
1845bool inode_owner_or_capable(const struct inode *inode)
 
1846{
1847	if (uid_eq(current_fsuid(), inode->i_uid))
 
 
 
 
1848		return true;
1849	if (inode_capable(inode, CAP_FOWNER))
 
 
1850		return true;
1851	return false;
1852}
1853EXPORT_SYMBOL(inode_owner_or_capable);
1854
1855/*
1856 * Direct i/o helper functions
1857 */
1858static void __inode_dio_wait(struct inode *inode)
1859{
1860	wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1861	DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1862
1863	do {
1864		prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1865		if (atomic_read(&inode->i_dio_count))
1866			schedule();
1867	} while (atomic_read(&inode->i_dio_count));
1868	finish_wait(wq, &q.wait);
1869}
 
1870
1871/**
1872 * inode_dio_wait - wait for outstanding DIO requests to finish
1873 * @inode: inode to wait for
1874 *
1875 * Waits for all pending direct I/O requests to finish so that we can
1876 * proceed with a truncate or equivalent operation.
1877 *
1878 * Must be called under a lock that serializes taking new references
1879 * to i_dio_count, usually by inode->i_mutex.
1880 */
1881void inode_dio_wait(struct inode *inode)
1882{
1883	if (atomic_read(&inode->i_dio_count))
1884		__inode_dio_wait(inode);
1885}
1886EXPORT_SYMBOL(inode_dio_wait);
1887
1888/*
1889 * inode_dio_done - signal finish of a direct I/O requests
1890 * @inode: inode the direct I/O happens on
1891 *
1892 * This is called once we've finished processing a direct I/O request,
1893 * and is used to wake up callers waiting for direct I/O to be quiesced.
1894 */
1895void inode_dio_done(struct inode *inode)
1896{
1897	if (atomic_dec_and_test(&inode->i_dio_count))
1898		wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1899}
1900EXPORT_SYMBOL(inode_dio_done);
1901
1902/*
1903 * inode_set_flags - atomically set some inode flags
1904 *
1905 * Note: the caller should be holding i_mutex, or else be sure that
1906 * they have exclusive access to the inode structure (i.e., while the
1907 * inode is being instantiated).  The reason for the cmpxchg() loop
1908 * --- which wouldn't be necessary if all code paths which modify
1909 * i_flags actually followed this rule, is that there is at least one
1910 * code path which doesn't today --- for example,
1911 * __generic_file_aio_write() calls file_remove_suid() without holding
1912 * i_mutex --- so we use cmpxchg() out of an abundance of caution.
1913 *
1914 * In the long run, i_mutex is overkill, and we should probably look
1915 * at using the i_lock spinlock to protect i_flags, and then make sure
1916 * it is so documented in include/linux/fs.h and that all code follows
1917 * the locking convention!!
1918 */
1919void inode_set_flags(struct inode *inode, unsigned int flags,
1920		     unsigned int mask)
1921{
1922	unsigned int old_flags, new_flags;
1923
1924	WARN_ON_ONCE(flags & ~mask);
1925	do {
1926		old_flags = ACCESS_ONCE(inode->i_flags);
1927		new_flags = (old_flags & ~mask) | flags;
1928	} while (unlikely(cmpxchg(&inode->i_flags, old_flags,
1929				  new_flags) != old_flags));
1930}
1931EXPORT_SYMBOL(inode_set_flags);