ialloc.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/ext4/ialloc.c
   4 *
   5 * Copyright (C) 1992, 1993, 1994, 1995
   6 * Remy Card (card@masi.ibp.fr)
   7 * Laboratoire MASI - Institut Blaise Pascal
   8 * Universite Pierre et Marie Curie (Paris VI)
   9 *
  10 *  BSD ufs-inspired inode and directory allocation by
  11 *  Stephen Tweedie (sct@redhat.com), 1993
  12 *  Big-endian to little-endian byte-swapping/bitmaps by
  13 *        David S. Miller (davem@caip.rutgers.edu), 1995
  14 */
  15
  16#include <linux/time.h>
  17#include <linux/fs.h>
  18#include <linux/stat.h>
  19#include <linux/string.h>
  20#include <linux/quotaops.h>
  21#include <linux/buffer_head.h>
  22#include <linux/random.h>
  23#include <linux/bitops.h>
  24#include <linux/blkdev.h>
  25#include <linux/cred.h>
  26
  27#include <asm/byteorder.h>
  28
  29#include "ext4.h"
  30#include "ext4_jbd2.h"
  31#include "xattr.h"
  32#include "acl.h"
  33
  34#include <trace/events/ext4.h>
  35
  36/*
  37 * ialloc.c contains the inodes allocation and deallocation routines
  38 */
  39
  40/*
  41 * The free inodes are managed by bitmaps.  A file system contains several
  42 * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
  43 * block for inodes, N blocks for the inode table and data blocks.
  44 *
  45 * The file system contains group descriptors which are located after the
  46 * super block.  Each descriptor contains the number of the bitmap block and
  47 * the free blocks count in the block.
  48 */
  49
  50/*
  51 * To avoid calling the atomic setbit hundreds or thousands of times, we only
  52 * need to use it within a single byte (to ensure we get endianness right).
  53 * We can use memset for the rest of the bitmap as there are no other users.
  54 */
  55void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
  56{
  57	int i;
  58
  59	if (start_bit >= end_bit)
  60		return;
  61
  62	ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
  63	for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
  64		ext4_set_bit(i, bitmap);
  65	if (i < end_bit)
  66		memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
  67}
  68
  69void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
  70{
  71	if (uptodate) {
  72		set_buffer_uptodate(bh);
  73		set_bitmap_uptodate(bh);
  74	}
  75	unlock_buffer(bh);
  76	put_bh(bh);
  77}
  78
  79static int ext4_validate_inode_bitmap(struct super_block *sb,
  80				      struct ext4_group_desc *desc,
  81				      ext4_group_t block_group,
  82				      struct buffer_head *bh)
  83{
  84	ext4_fsblk_t	blk;
  85	struct ext4_group_info *grp;
  86
  87	if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
  88		return 0;
  89
  90	if (buffer_verified(bh))
  91		return 0;
  92
  93	grp = ext4_get_group_info(sb, block_group);
  94	if (!grp || EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
  95		return -EFSCORRUPTED;
  96
  97	ext4_lock_group(sb, block_group);
  98	if (buffer_verified(bh))
  99		goto verified;
 100	blk = ext4_inode_bitmap(sb, desc);
 101	if (!ext4_inode_bitmap_csum_verify(sb, desc, bh) ||
 102	    ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) {
 103		ext4_unlock_group(sb, block_group);
 104		ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
 105			   "inode_bitmap = %llu", block_group, blk);
 106		ext4_mark_group_bitmap_corrupted(sb, block_group,
 107					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 108		return -EFSBADCRC;
 109	}
 110	set_buffer_verified(bh);
 111verified:
 112	ext4_unlock_group(sb, block_group);
 113	return 0;
 114}
 115
 116/*
 117 * Read the inode allocation bitmap for a given block_group, reading
 118 * into the specified slot in the superblock's bitmap cache.
 119 *
 120 * Return buffer_head of bitmap on success, or an ERR_PTR on error.
 121 */
 122static struct buffer_head *
 123ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 124{
 125	struct ext4_group_desc *desc;
 126	struct ext4_sb_info *sbi = EXT4_SB(sb);
 127	struct buffer_head *bh = NULL;
 128	ext4_fsblk_t bitmap_blk;
 129	int err;
 130
 131	desc = ext4_get_group_desc(sb, block_group, NULL);
 132	if (!desc)
 133		return ERR_PTR(-EFSCORRUPTED);
 134
 135	bitmap_blk = ext4_inode_bitmap(sb, desc);
 136	if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 137	    (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
 138		ext4_error(sb, "Invalid inode bitmap blk %llu in "
 139			   "block_group %u", bitmap_blk, block_group);
 140		ext4_mark_group_bitmap_corrupted(sb, block_group,
 141					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 142		return ERR_PTR(-EFSCORRUPTED);
 143	}
 144	bh = sb_getblk(sb, bitmap_blk);
 145	if (unlikely(!bh)) {
 146		ext4_warning(sb, "Cannot read inode bitmap - "
 147			     "block_group = %u, inode_bitmap = %llu",
 148			     block_group, bitmap_blk);
 149		return ERR_PTR(-ENOMEM);
 150	}
 151	if (bitmap_uptodate(bh))
 152		goto verify;
 153
 154	lock_buffer(bh);
 155	if (bitmap_uptodate(bh)) {
 156		unlock_buffer(bh);
 157		goto verify;
 158	}
 159
 160	ext4_lock_group(sb, block_group);
 161	if (ext4_has_group_desc_csum(sb) &&
 162	    (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
 163		if (block_group == 0) {
 164			ext4_unlock_group(sb, block_group);
 165			unlock_buffer(bh);
 166			ext4_error(sb, "Inode bitmap for bg 0 marked "
 167				   "uninitialized");
 168			err = -EFSCORRUPTED;
 169			goto out;
 170		}
 171		memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
 172		ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 173				     sb->s_blocksize * 8, bh->b_data);
 174		set_bitmap_uptodate(bh);
 175		set_buffer_uptodate(bh);
 176		set_buffer_verified(bh);
 177		ext4_unlock_group(sb, block_group);
 178		unlock_buffer(bh);
 179		return bh;
 180	}
 181	ext4_unlock_group(sb, block_group);
 182
 183	if (buffer_uptodate(bh)) {
 184		/*
 185		 * if not uninit if bh is uptodate,
 186		 * bitmap is also uptodate
 187		 */
 188		set_bitmap_uptodate(bh);
 189		unlock_buffer(bh);
 190		goto verify;
 191	}
 192	/*
 193	 * submit the buffer_head for reading
 194	 */
 195	trace_ext4_load_inode_bitmap(sb, block_group);
 196	ext4_read_bh(bh, REQ_META | REQ_PRIO,
 197		     ext4_end_bitmap_read,
 198		     ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_EIO));
 199	if (!buffer_uptodate(bh)) {
 200		put_bh(bh);
 201		ext4_error_err(sb, EIO, "Cannot read inode bitmap - "
 202			       "block_group = %u, inode_bitmap = %llu",
 203			       block_group, bitmap_blk);
 204		ext4_mark_group_bitmap_corrupted(sb, block_group,
 205				EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 206		return ERR_PTR(-EIO);
 207	}
 208
 209verify:
 210	err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
 211	if (err)
 212		goto out;
 213	return bh;
 214out:
 215	put_bh(bh);
 216	return ERR_PTR(err);
 217}
 218
 219/*
 220 * NOTE! When we get the inode, we're the only people
 221 * that have access to it, and as such there are no
 222 * race conditions we have to worry about. The inode
 223 * is not on the hash-lists, and it cannot be reached
 224 * through the filesystem because the directory entry
 225 * has been deleted earlier.
 226 *
 227 * HOWEVER: we must make sure that we get no aliases,
 228 * which means that we have to call "clear_inode()"
 229 * _before_ we mark the inode not in use in the inode
 230 * bitmaps. Otherwise a newly created file might use
 231 * the same inode number (not actually the same pointer
 232 * though), and then we'd have two inodes sharing the
 233 * same inode number and space on the harddisk.
 234 */
 235void ext4_free_inode(handle_t *handle, struct inode *inode)
 236{
 237	struct super_block *sb = inode->i_sb;
 238	int is_directory;
 239	unsigned long ino;
 240	struct buffer_head *bitmap_bh = NULL;
 241	struct buffer_head *bh2;
 242	ext4_group_t block_group;
 243	unsigned long bit;
 244	struct ext4_group_desc *gdp;
 245	struct ext4_super_block *es;
 246	struct ext4_sb_info *sbi;
 247	int fatal = 0, err, count, cleared;
 248	struct ext4_group_info *grp;
 249
 250	if (!sb) {
 251		printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 252		       "nonexistent device\n", __func__, __LINE__);
 253		return;
 254	}
 255	if (icount_read(inode) > 1) {
 256		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 257			 __func__, __LINE__, inode->i_ino,
 258			 icount_read(inode));
 259		return;
 260	}
 261	if (inode->i_nlink) {
 262		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 263			 __func__, __LINE__, inode->i_ino, inode->i_nlink);
 264		return;
 265	}
 266	sbi = EXT4_SB(sb);
 267
 268	ino = inode->i_ino;
 269	ext4_debug("freeing inode %lu\n", ino);
 270	trace_ext4_free_inode(inode);
 271
 272	dquot_initialize(inode);
 273	dquot_free_inode(inode);
 274
 275	is_directory = S_ISDIR(inode->i_mode);
 276
 277	/* Do this BEFORE marking the inode not in use or returning an error */
 278	ext4_clear_inode(inode);
 279
 280	es = sbi->s_es;
 281	if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 282		ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 283		goto error_return;
 284	}
 285	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 286	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 287	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 288	/* Don't bother if the inode bitmap is corrupt. */
 289	if (IS_ERR(bitmap_bh)) {
 290		fatal = PTR_ERR(bitmap_bh);
 291		bitmap_bh = NULL;
 292		goto error_return;
 293	}
 294	if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
 295		grp = ext4_get_group_info(sb, block_group);
 296		if (!grp || unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
 297			fatal = -EFSCORRUPTED;
 298			goto error_return;
 299		}
 300	}
 301
 302	BUFFER_TRACE(bitmap_bh, "get_write_access");
 303	fatal = ext4_journal_get_write_access(handle, sb, bitmap_bh,
 304					      EXT4_JTR_NONE);
 305	if (fatal)
 306		goto error_return;
 307
 308	fatal = -ESRCH;
 309	gdp = ext4_get_group_desc(sb, block_group, &bh2);
 310	if (gdp) {
 311		BUFFER_TRACE(bh2, "get_write_access");
 312		fatal = ext4_journal_get_write_access(handle, sb, bh2,
 313						      EXT4_JTR_NONE);
 314	}
 315	ext4_lock_group(sb, block_group);
 316	cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 317	if (fatal || !cleared) {
 318		ext4_unlock_group(sb, block_group);
 319		goto out;
 320	}
 321
 322	count = ext4_free_inodes_count(sb, gdp) + 1;
 323	ext4_free_inodes_set(sb, gdp, count);
 324	if (is_directory) {
 325		count = ext4_used_dirs_count(sb, gdp) - 1;
 326		ext4_used_dirs_set(sb, gdp, count);
 327		if (percpu_counter_initialized(&sbi->s_dirs_counter))
 328			percpu_counter_dec(&sbi->s_dirs_counter);
 329	}
 330	ext4_inode_bitmap_csum_set(sb, gdp, bitmap_bh);
 331	ext4_group_desc_csum_set(sb, block_group, gdp);
 332	ext4_unlock_group(sb, block_group);
 333
 334	if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
 335		percpu_counter_inc(&sbi->s_freeinodes_counter);
 336	if (sbi->s_log_groups_per_flex) {
 337		struct flex_groups *fg;
 338
 339		fg = sbi_array_rcu_deref(sbi, s_flex_groups,
 340					 ext4_flex_group(sbi, block_group));
 341		atomic_inc(&fg->free_inodes);
 342		if (is_directory)
 343			atomic_dec(&fg->used_dirs);
 344	}
 345	BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 346	fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 347out:
 348	if (cleared) {
 349		BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 350		err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 351		if (!fatal)
 352			fatal = err;
 353	} else {
 354		ext4_error(sb, "bit already cleared for inode %lu", ino);
 355		ext4_mark_group_bitmap_corrupted(sb, block_group,
 356					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 357	}
 358
 359error_return:
 360	brelse(bitmap_bh);
 361	ext4_std_error(sb, fatal);
 362}
 363
 364struct orlov_stats {
 365	__u64 free_clusters;
 366	__u32 free_inodes;
 367	__u32 used_dirs;
 368};
 369
 370/*
 371 * Helper function for Orlov's allocator; returns critical information
 372 * for a particular block group or flex_bg.  If flex_size is 1, then g
 373 * is a block group number; otherwise it is flex_bg number.
 374 */
 375static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 376			    int flex_size, struct orlov_stats *stats)
 377{
 378	struct ext4_group_desc *desc;
 379
 380	if (flex_size > 1) {
 381		struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
 382							     s_flex_groups, g);
 383		stats->free_inodes = atomic_read(&fg->free_inodes);
 384		stats->free_clusters = atomic64_read(&fg->free_clusters);
 385		stats->used_dirs = atomic_read(&fg->used_dirs);
 386		return;
 387	}
 388
 389	desc = ext4_get_group_desc(sb, g, NULL);
 390	if (desc) {
 391		stats->free_inodes = ext4_free_inodes_count(sb, desc);
 392		stats->free_clusters = ext4_free_group_clusters(sb, desc);
 393		stats->used_dirs = ext4_used_dirs_count(sb, desc);
 394	} else {
 395		stats->free_inodes = 0;
 396		stats->free_clusters = 0;
 397		stats->used_dirs = 0;
 398	}
 399}
 400
 401/*
 402 * Orlov's allocator for directories.
 403 *
 404 * We always try to spread first-level directories.
 405 *
 406 * If there are blockgroups with both free inodes and free clusters counts
 407 * not worse than average we return one with smallest directory count.
 408 * Otherwise we simply return a random group.
 409 *
 410 * For the rest rules look so:
 411 *
 412 * It's OK to put directory into a group unless
 413 * it has too many directories already (max_dirs) or
 414 * it has too few free inodes left (min_inodes) or
 415 * it has too few free clusters left (min_clusters) or
 416 * Parent's group is preferred, if it doesn't satisfy these
 417 * conditions we search cyclically through the rest. If none
 418 * of the groups look good we just look for a group with more
 419 * free inodes than average (starting at parent's group).
 420 */
 421
 422static int find_group_orlov(struct super_block *sb, struct inode *parent,
 423			    ext4_group_t *group, umode_t mode,
 424			    const struct qstr *qstr)
 425{
 426	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 427	struct ext4_sb_info *sbi = EXT4_SB(sb);
 428	ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 429	int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 430	unsigned int freei, avefreei, grp_free;
 431	ext4_fsblk_t freec, avefreec;
 432	unsigned int ndirs;
 433	int max_dirs, min_inodes;
 434	ext4_grpblk_t min_clusters;
 435	ext4_group_t i, grp, g, ngroups;
 436	struct ext4_group_desc *desc;
 437	struct orlov_stats stats;
 438	int flex_size = ext4_flex_bg_size(sbi);
 439	struct dx_hash_info hinfo;
 440
 441	ngroups = real_ngroups;
 442	if (flex_size > 1) {
 443		ngroups = (real_ngroups + flex_size - 1) >>
 444			sbi->s_log_groups_per_flex;
 445		parent_group >>= sbi->s_log_groups_per_flex;
 446	}
 447
 448	freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 449	avefreei = freei / ngroups;
 450	freec = percpu_counter_read_positive(&sbi->s_freeclusters_counter);
 451	avefreec = freec;
 452	do_div(avefreec, ngroups);
 453	ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 454
 455	if (S_ISDIR(mode) &&
 456	    ((parent == d_inode(sb->s_root)) ||
 457	     (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 458		int best_ndir = inodes_per_group;
 459		int ret = -1;
 460
 461		if (qstr) {
 462			hinfo.hash_version = DX_HASH_HALF_MD4;
 463			hinfo.seed = sbi->s_hash_seed;
 464			ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
 465			parent_group = hinfo.hash % ngroups;
 466		} else
 467			parent_group = get_random_u32_below(ngroups);
 468		for (i = 0; i < ngroups; i++) {
 469			g = (parent_group + i) % ngroups;
 470			get_orlov_stats(sb, g, flex_size, &stats);
 471			if (!stats.free_inodes)
 472				continue;
 473			if (stats.used_dirs >= best_ndir)
 474				continue;
 475			if (stats.free_inodes < avefreei)
 476				continue;
 477			if (stats.free_clusters < avefreec)
 478				continue;
 479			grp = g;
 480			ret = 0;
 481			best_ndir = stats.used_dirs;
 482		}
 483		if (ret)
 484			goto fallback;
 485	found_flex_bg:
 486		if (flex_size == 1) {
 487			*group = grp;
 488			return 0;
 489		}
 490
 491		/*
 492		 * We pack inodes at the beginning of the flexgroup's
 493		 * inode tables.  Block allocation decisions will do
 494		 * something similar, although regular files will
 495		 * start at 2nd block group of the flexgroup.  See
 496		 * ext4_ext_find_goal() and ext4_find_near().
 497		 */
 498		grp *= flex_size;
 499		for (i = 0; i < flex_size; i++) {
 500			if (grp+i >= real_ngroups)
 501				break;
 502			desc = ext4_get_group_desc(sb, grp+i, NULL);
 503			if (desc && ext4_free_inodes_count(sb, desc)) {
 504				*group = grp+i;
 505				return 0;
 506			}
 507		}
 508		goto fallback;
 509	}
 510
 511	max_dirs = ndirs / ngroups + inodes_per_group*flex_size / 16;
 512	min_inodes = avefreei - inodes_per_group*flex_size / 4;
 513	if (min_inodes < 1)
 514		min_inodes = 1;
 515	min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 516	if (min_clusters < 0)
 517		min_clusters = 0;
 518
 519	/*
 520	 * Start looking in the flex group where we last allocated an
 521	 * inode for this parent directory
 522	 */
 523	if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 524		parent_group = EXT4_I(parent)->i_last_alloc_group;
 525		if (flex_size > 1)
 526			parent_group >>= sbi->s_log_groups_per_flex;
 527	}
 528
 529	for (i = 0; i < ngroups; i++) {
 530		grp = (parent_group + i) % ngroups;
 531		get_orlov_stats(sb, grp, flex_size, &stats);
 532		if (stats.used_dirs >= max_dirs)
 533			continue;
 534		if (stats.free_inodes < min_inodes)
 535			continue;
 536		if (stats.free_clusters < min_clusters)
 537			continue;
 538		goto found_flex_bg;
 539	}
 540
 541fallback:
 542	ngroups = real_ngroups;
 543	avefreei = freei / ngroups;
 544fallback_retry:
 545	parent_group = EXT4_I(parent)->i_block_group;
 546	for (i = 0; i < ngroups; i++) {
 547		grp = (parent_group + i) % ngroups;
 548		desc = ext4_get_group_desc(sb, grp, NULL);
 549		if (desc) {
 550			grp_free = ext4_free_inodes_count(sb, desc);
 551			if (grp_free && grp_free >= avefreei) {
 552				*group = grp;
 553				return 0;
 554			}
 555		}
 556	}
 557
 558	if (avefreei) {
 559		/*
 560		 * The free-inodes counter is approximate, and for really small
 561		 * filesystems the above test can fail to find any blockgroups
 562		 */
 563		avefreei = 0;
 564		goto fallback_retry;
 565	}
 566
 567	return -1;
 568}
 569
 570static int find_group_other(struct super_block *sb, struct inode *parent,
 571			    ext4_group_t *group, umode_t mode)
 572{
 573	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 574	ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 575	struct ext4_group_desc *desc;
 576	int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 577
 578	/*
 579	 * Try to place the inode is the same flex group as its
 580	 * parent.  If we can't find space, use the Orlov algorithm to
 581	 * find another flex group, and store that information in the
 582	 * parent directory's inode information so that use that flex
 583	 * group for future allocations.
 584	 */
 585	if (flex_size > 1) {
 586		int retry = 0;
 587
 588	try_again:
 589		parent_group &= ~(flex_size-1);
 590		last = parent_group + flex_size;
 591		if (last > ngroups)
 592			last = ngroups;
 593		for  (i = parent_group; i < last; i++) {
 594			desc = ext4_get_group_desc(sb, i, NULL);
 595			if (desc && ext4_free_inodes_count(sb, desc)) {
 596				*group = i;
 597				return 0;
 598			}
 599		}
 600		if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 601			retry = 1;
 602			parent_group = EXT4_I(parent)->i_last_alloc_group;
 603			goto try_again;
 604		}
 605		/*
 606		 * If this didn't work, use the Orlov search algorithm
 607		 * to find a new flex group; we pass in the mode to
 608		 * avoid the topdir algorithms.
 609		 */
 610		*group = parent_group + flex_size;
 611		if (*group > ngroups)
 612			*group = 0;
 613		return find_group_orlov(sb, parent, group, mode, NULL);
 614	}
 615
 616	/*
 617	 * Try to place the inode in its parent directory
 618	 */
 619	*group = parent_group;
 620	desc = ext4_get_group_desc(sb, *group, NULL);
 621	if (desc && ext4_free_inodes_count(sb, desc) &&
 622	    ext4_free_group_clusters(sb, desc))
 623		return 0;
 624
 625	/*
 626	 * We're going to place this inode in a different blockgroup from its
 627	 * parent.  We want to cause files in a common directory to all land in
 628	 * the same blockgroup.  But we want files which are in a different
 629	 * directory which shares a blockgroup with our parent to land in a
 630	 * different blockgroup.
 631	 *
 632	 * So add our directory's i_ino into the starting point for the hash.
 633	 */
 634	*group = (*group + parent->i_ino) % ngroups;
 635
 636	/*
 637	 * Use a quadratic hash to find a group with a free inode and some free
 638	 * blocks.
 639	 */
 640	for (i = 1; i < ngroups; i <<= 1) {
 641		*group += i;
 642		if (*group >= ngroups)
 643			*group -= ngroups;
 644		desc = ext4_get_group_desc(sb, *group, NULL);
 645		if (desc && ext4_free_inodes_count(sb, desc) &&
 646		    ext4_free_group_clusters(sb, desc))
 647			return 0;
 648	}
 649
 650	/*
 651	 * That failed: try linear search for a free inode, even if that group
 652	 * has no free blocks.
 653	 */
 654	*group = parent_group;
 655	for (i = 0; i < ngroups; i++) {
 656		if (++*group >= ngroups)
 657			*group = 0;
 658		desc = ext4_get_group_desc(sb, *group, NULL);
 659		if (desc && ext4_free_inodes_count(sb, desc))
 660			return 0;
 661	}
 662
 663	return -1;
 664}
 665
 666/*
 667 * In no journal mode, if an inode has recently been deleted, we want
 668 * to avoid reusing it until we're reasonably sure the inode table
 669 * block has been written back to disk.  (Yes, these values are
 670 * somewhat arbitrary...)
 671 */
 672#define RECENTCY_MIN	60
 673#define RECENTCY_DIRTY	300
 674
 675static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 676{
 677	struct ext4_group_desc	*gdp;
 678	struct ext4_inode	*raw_inode;
 679	struct buffer_head	*bh;
 680	int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 681	int offset, ret = 0;
 682	int recentcy = RECENTCY_MIN;
 683	u32 dtime, now;
 684
 685	gdp = ext4_get_group_desc(sb, group, NULL);
 686	if (unlikely(!gdp))
 687		return 0;
 688
 689	/* Inode was never used in this filesystem? */
 690	if (ext4_has_group_desc_csum(sb) &&
 691	    (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT) ||
 692	     ino >= EXT4_INODES_PER_GROUP(sb) - ext4_itable_unused_count(sb, gdp)))
 693		return 0;
 694
 695	bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
 696		       (ino / inodes_per_block));
 697	if (!bh || !buffer_uptodate(bh))
 698		/*
 699		 * If the block is not in the buffer cache, then it
 700		 * must have been written out, or, most unlikely, is
 701		 * being migrated - false failure should be OK here.
 702		 */
 703		goto out;
 704
 705	offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 706	raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 707
 708	/* i_dtime is only 32 bits on disk, but we only care about relative
 709	 * times in the range of a few minutes (i.e. long enough to sync a
 710	 * recently-deleted inode to disk), so using the low 32 bits of the
 711	 * clock (a 68 year range) is enough, see time_before32() */
 712	dtime = le32_to_cpu(raw_inode->i_dtime);
 713	now = ktime_get_real_seconds();
 714	if (buffer_dirty(bh))
 715		recentcy += RECENTCY_DIRTY;
 716
 717	if (dtime && time_before32(dtime, now) &&
 718	    time_before32(now, dtime + recentcy))
 719		ret = 1;
 720out:
 721	brelse(bh);
 722	return ret;
 723}
 724
 725static int find_inode_bit(struct super_block *sb, ext4_group_t group,
 726			  struct buffer_head *bitmap, unsigned long *ino)
 727{
 728	bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
 729	unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
 730
 731next:
 732	*ino = ext4_find_next_zero_bit((unsigned long *)
 733				       bitmap->b_data,
 734				       EXT4_INODES_PER_GROUP(sb), *ino);
 735	if (*ino >= EXT4_INODES_PER_GROUP(sb))
 736		goto not_found;
 737
 738	if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
 739		recently_deleted_ino = *ino;
 740		*ino = *ino + 1;
 741		if (*ino < EXT4_INODES_PER_GROUP(sb))
 742			goto next;
 743		goto not_found;
 744	}
 745	return 1;
 746not_found:
 747	if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
 748		return 0;
 749	/*
 750	 * Not reusing recently deleted inodes is mostly a preference. We don't
 751	 * want to report ENOSPC or skew allocation patterns because of that.
 752	 * So return even recently deleted inode if we could find better in the
 753	 * given range.
 754	 */
 755	*ino = recently_deleted_ino;
 756	return 1;
 757}
 758
 759int ext4_mark_inode_used(struct super_block *sb, int ino)
 760{
 761	unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
 762	struct buffer_head *inode_bitmap_bh = NULL, *group_desc_bh = NULL;
 763	struct ext4_group_desc *gdp;
 764	ext4_group_t group;
 765	int bit;
 766	int err;
 767
 768	if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
 769		return -EFSCORRUPTED;
 770
 771	group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 772	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 773	inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 774	if (IS_ERR(inode_bitmap_bh))
 775		return PTR_ERR(inode_bitmap_bh);
 776
 777	if (ext4_test_bit(bit, inode_bitmap_bh->b_data)) {
 778		err = 0;
 779		goto out;
 780	}
 781
 782	gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 783	if (!gdp) {
 784		err = -EINVAL;
 785		goto out;
 786	}
 787
 788	ext4_set_bit(bit, inode_bitmap_bh->b_data);
 789
 790	BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 791	err = ext4_handle_dirty_metadata(NULL, NULL, inode_bitmap_bh);
 792	if (err) {
 793		ext4_std_error(sb, err);
 794		goto out;
 795	}
 796	err = sync_dirty_buffer(inode_bitmap_bh);
 797	if (err) {
 798		ext4_std_error(sb, err);
 799		goto out;
 800	}
 801
 802	/* We may have to initialize the block bitmap if it isn't already */
 803	if (ext4_has_group_desc_csum(sb) &&
 804	    gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 805		struct buffer_head *block_bitmap_bh;
 806
 807		block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 808		if (IS_ERR(block_bitmap_bh)) {
 809			err = PTR_ERR(block_bitmap_bh);
 810			goto out;
 811		}
 812
 813		BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
 814		err = ext4_handle_dirty_metadata(NULL, NULL, block_bitmap_bh);
 815		sync_dirty_buffer(block_bitmap_bh);
 816
 817		/* recheck and clear flag under lock if we still need to */
 818		ext4_lock_group(sb, group);
 819		if (ext4_has_group_desc_csum(sb) &&
 820		    (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
 821			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
 822			ext4_free_group_clusters_set(sb, gdp,
 823				ext4_free_clusters_after_init(sb, group, gdp));
 824			ext4_block_bitmap_csum_set(sb, gdp, block_bitmap_bh);
 825			ext4_group_desc_csum_set(sb, group, gdp);
 826		}
 827		ext4_unlock_group(sb, group);
 828		brelse(block_bitmap_bh);
 829
 830		if (err) {
 831			ext4_std_error(sb, err);
 832			goto out;
 833		}
 834	}
 835
 836	/* Update the relevant bg descriptor fields */
 837	if (ext4_has_group_desc_csum(sb)) {
 838		int free;
 839
 840		ext4_lock_group(sb, group); /* while we modify the bg desc */
 841		free = EXT4_INODES_PER_GROUP(sb) -
 842			ext4_itable_unused_count(sb, gdp);
 843		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 844			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
 845			free = 0;
 846		}
 847
 848		/*
 849		 * Check the relative inode number against the last used
 850		 * relative inode number in this group. if it is greater
 851		 * we need to update the bg_itable_unused count
 852		 */
 853		if (bit >= free)
 854			ext4_itable_unused_set(sb, gdp,
 855					(EXT4_INODES_PER_GROUP(sb) - bit - 1));
 856	} else {
 857		ext4_lock_group(sb, group);
 858	}
 859
 860	ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
 861	if (ext4_has_group_desc_csum(sb)) {
 862		ext4_inode_bitmap_csum_set(sb, gdp, inode_bitmap_bh);
 863		ext4_group_desc_csum_set(sb, group, gdp);
 864	}
 865
 866	ext4_unlock_group(sb, group);
 867	err = ext4_handle_dirty_metadata(NULL, NULL, group_desc_bh);
 868	sync_dirty_buffer(group_desc_bh);
 869out:
 870	brelse(inode_bitmap_bh);
 871	return err;
 872}
 873
 874static int ext4_xattr_credits_for_new_inode(struct inode *dir, mode_t mode,
 875					    bool encrypt)
 876{
 877	struct super_block *sb = dir->i_sb;
 878	int nblocks = 0;
 879#ifdef CONFIG_EXT4_FS_POSIX_ACL
 880	struct posix_acl *p = get_inode_acl(dir, ACL_TYPE_DEFAULT);
 881
 882	if (IS_ERR(p))
 883		return PTR_ERR(p);
 884	if (p) {
 885		int acl_size = p->a_count * sizeof(ext4_acl_entry);
 886
 887		nblocks += (S_ISDIR(mode) ? 2 : 1) *
 888			__ext4_xattr_set_credits(sb, NULL /* inode */,
 889						 NULL /* block_bh */, acl_size,
 890						 true /* is_create */);
 891		posix_acl_release(p);
 892	}
 893#endif
 894
 895#ifdef CONFIG_SECURITY
 896	{
 897		int num_security_xattrs = 1;
 898
 899#ifdef CONFIG_INTEGRITY
 900		num_security_xattrs++;
 901#endif
 902		/*
 903		 * We assume that security xattrs are never more than 1k.
 904		 * In practice they are under 128 bytes.
 905		 */
 906		nblocks += num_security_xattrs *
 907			__ext4_xattr_set_credits(sb, NULL /* inode */,
 908						 NULL /* block_bh */, 1024,
 909						 true /* is_create */);
 910	}
 911#endif
 912	if (encrypt)
 913		nblocks += __ext4_xattr_set_credits(sb,
 914						    NULL /* inode */,
 915						    NULL /* block_bh */,
 916						    FSCRYPT_SET_CONTEXT_MAX_SIZE,
 917						    true /* is_create */);
 918	return nblocks;
 919}
 920
 921/*
 922 * There are two policies for allocating an inode.  If the new inode is
 923 * a directory, then a forward search is made for a block group with both
 924 * free space and a low directory-to-inode ratio; if that fails, then of
 925 * the groups with above-average free space, that group with the fewest
 926 * directories already is chosen.
 927 *
 928 * For other inodes, search forward from the parent directory's block
 929 * group to find a free inode.
 930 */
 931struct inode *__ext4_new_inode(struct mnt_idmap *idmap,
 932			       handle_t *handle, struct inode *dir,
 933			       umode_t mode, const struct qstr *qstr,
 934			       __u32 goal, uid_t *owner, __u32 i_flags,
 935			       int handle_type, unsigned int line_no,
 936			       int nblocks)
 937{
 938	struct super_block *sb;
 939	struct buffer_head *inode_bitmap_bh = NULL;
 940	struct buffer_head *group_desc_bh;
 941	ext4_group_t ngroups, group = 0;
 942	unsigned long ino = 0;
 943	struct inode *inode;
 944	struct ext4_group_desc *gdp = NULL;
 945	struct ext4_inode_info *ei;
 946	struct ext4_sb_info *sbi;
 947	int ret2, err;
 948	struct inode *ret;
 949	ext4_group_t i;
 950	ext4_group_t flex_group;
 951	struct ext4_group_info *grp = NULL;
 952	bool encrypt = false;
 953
 954	/* Cannot create files in a deleted directory */
 955	if (!dir || !dir->i_nlink)
 956		return ERR_PTR(-EPERM);
 957
 958	sb = dir->i_sb;
 959	sbi = EXT4_SB(sb);
 960
 961	ret2 = ext4_emergency_state(sb);
 962	if (unlikely(ret2))
 963		return ERR_PTR(ret2);
 964
 965	ngroups = ext4_get_groups_count(sb);
 966	trace_ext4_request_inode(dir, mode);
 967	inode = new_inode(sb);
 968	if (!inode)
 969		return ERR_PTR(-ENOMEM);
 970	ei = EXT4_I(inode);
 971
 972	/*
 973	 * Initialize owners and quota early so that we don't have to account
 974	 * for quota initialization worst case in standard inode creating
 975	 * transaction
 976	 */
 977	if (owner) {
 978		inode->i_mode = mode;
 979		i_uid_write(inode, owner[0]);
 980		i_gid_write(inode, owner[1]);
 981	} else if (test_opt(sb, GRPID)) {
 982		inode->i_mode = mode;
 983		inode_fsuid_set(inode, idmap);
 984		inode->i_gid = dir->i_gid;
 985	} else
 986		inode_init_owner(idmap, inode, dir, mode);
 987
 988	if (ext4_has_feature_project(sb) &&
 989	    ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
 990		ei->i_projid = EXT4_I(dir)->i_projid;
 991	else
 992		ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
 993
 994	if (!(i_flags & EXT4_EA_INODE_FL)) {
 995		err = fscrypt_prepare_new_inode(dir, inode, &encrypt);
 996		if (err)
 997			goto out;
 998	}
 999
1000	err = dquot_initialize(inode);
1001	if (err)
1002		goto out;
1003
1004	if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
1005		ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt);
1006		if (ret2 < 0) {
1007			err = ret2;
1008			goto out;
1009		}
1010		nblocks += ret2;
1011	}
1012
1013	if (!goal)
1014		goal = sbi->s_inode_goal;
1015
1016	if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
1017		group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
1018		ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
1019		ret2 = 0;
1020		goto got_group;
1021	}
1022
1023	if (S_ISDIR(mode))
1024		ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
1025	else
1026		ret2 = find_group_other(sb, dir, &group, mode);
1027
1028got_group:
1029	EXT4_I(dir)->i_last_alloc_group = group;
1030	err = -ENOSPC;
1031	if (ret2 == -1)
1032		goto out;
1033
1034	/*
1035	 * Normally we will only go through one pass of this loop,
1036	 * unless we get unlucky and it turns out the group we selected
1037	 * had its last inode grabbed by someone else.
1038	 */
1039	for (i = 0; i < ngroups; i++, ino = 0) {
1040		err = -EIO;
1041
1042		gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1043		if (!gdp)
1044			goto out;
1045
1046		/*
1047		 * Check free inodes count before loading bitmap.
1048		 */
1049		if (ext4_free_inodes_count(sb, gdp) == 0)
1050			goto next_group;
1051
1052		if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1053			grp = ext4_get_group_info(sb, group);
1054			/*
1055			 * Skip groups with already-known suspicious inode
1056			 * tables
1057			 */
1058			if (!grp || EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1059				goto next_group;
1060		}
1061
1062		brelse(inode_bitmap_bh);
1063		inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
1064		/* Skip groups with suspicious inode tables */
1065		if (IS_ERR(inode_bitmap_bh)) {
1066			inode_bitmap_bh = NULL;
1067			goto next_group;
1068		}
1069		if (!(sbi->s_mount_state & EXT4_FC_REPLAY) &&
1070		    EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1071			goto next_group;
1072
1073		ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1074		if (!ret2)
1075			goto next_group;
1076
1077		if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
1078			ext4_error(sb, "reserved inode found cleared - "
1079				   "inode=%lu", ino + 1);
1080			ext4_mark_group_bitmap_corrupted(sb, group,
1081					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1082			goto next_group;
1083		}
1084
1085		if ((!(sbi->s_mount_state & EXT4_FC_REPLAY)) && !handle) {
1086			BUG_ON(nblocks <= 0);
1087			handle = __ext4_journal_start_sb(NULL, dir->i_sb,
1088				 line_no, handle_type, nblocks, 0,
1089				 ext4_trans_default_revoke_credits(sb));
1090			if (IS_ERR(handle)) {
1091				err = PTR_ERR(handle);
1092				ext4_std_error(sb, err);
1093				goto out;
1094			}
1095		}
1096		BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
1097		err = ext4_journal_get_write_access(handle, sb, inode_bitmap_bh,
1098						    EXT4_JTR_NONE);
1099		if (err) {
1100			ext4_std_error(sb, err);
1101			goto out;
1102		}
1103		ext4_lock_group(sb, group);
1104		ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
1105		if (ret2) {
1106			/* Someone already took the bit. Repeat the search
1107			 * with lock held.
1108			 */
1109			ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1110			if (ret2) {
1111				ext4_set_bit(ino, inode_bitmap_bh->b_data);
1112				ret2 = 0;
1113			} else {
1114				ret2 = 1; /* we didn't grab the inode */
1115			}
1116		}
1117		ext4_unlock_group(sb, group);
1118		ino++;		/* the inode bitmap is zero-based */
1119		if (!ret2)
1120			goto got; /* we grabbed the inode! */
1121
1122next_group:
1123		if (++group == ngroups)
1124			group = 0;
1125	}
1126	err = -ENOSPC;
1127	goto out;
1128
1129got:
1130	BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
1131	err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
1132	if (err) {
1133		ext4_std_error(sb, err);
1134		goto out;
1135	}
1136
1137	BUFFER_TRACE(group_desc_bh, "get_write_access");
1138	err = ext4_journal_get_write_access(handle, sb, group_desc_bh,
1139					    EXT4_JTR_NONE);
1140	if (err) {
1141		ext4_std_error(sb, err);
1142		goto out;
1143	}
1144
1145	/* We may have to initialize the block bitmap if it isn't already */
1146	if (ext4_has_group_desc_csum(sb) &&
1147	    gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1148		struct buffer_head *block_bitmap_bh;
1149
1150		block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1151		if (IS_ERR(block_bitmap_bh)) {
1152			err = PTR_ERR(block_bitmap_bh);
1153			goto out;
1154		}
1155		BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1156		err = ext4_journal_get_write_access(handle, sb, block_bitmap_bh,
1157						    EXT4_JTR_NONE);
1158		if (err) {
1159			brelse(block_bitmap_bh);
1160			ext4_std_error(sb, err);
1161			goto out;
1162		}
1163
1164		BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1165		err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1166
1167		/* recheck and clear flag under lock if we still need to */
1168		ext4_lock_group(sb, group);
1169		if (ext4_has_group_desc_csum(sb) &&
1170		    (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1171			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1172			ext4_free_group_clusters_set(sb, gdp,
1173				ext4_free_clusters_after_init(sb, group, gdp));
1174			ext4_block_bitmap_csum_set(sb, gdp, block_bitmap_bh);
1175			ext4_group_desc_csum_set(sb, group, gdp);
1176		}
1177		ext4_unlock_group(sb, group);
1178		brelse(block_bitmap_bh);
1179
1180		if (err) {
1181			ext4_std_error(sb, err);
1182			goto out;
1183		}
1184	}
1185
1186	/* Update the relevant bg descriptor fields */
1187	if (ext4_has_group_desc_csum(sb)) {
1188		int free;
1189		struct ext4_group_info *grp = NULL;
1190
1191		if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1192			grp = ext4_get_group_info(sb, group);
1193			if (!grp) {
1194				err = -EFSCORRUPTED;
1195				goto out;
1196			}
1197			down_read(&grp->alloc_sem); /*
1198						     * protect vs itable
1199						     * lazyinit
1200						     */
1201		}
1202		ext4_lock_group(sb, group); /* while we modify the bg desc */
1203		free = EXT4_INODES_PER_GROUP(sb) -
1204			ext4_itable_unused_count(sb, gdp);
1205		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1206			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1207			free = 0;
1208		}
1209		/*
1210		 * Check the relative inode number against the last used
1211		 * relative inode number in this group. if it is greater
1212		 * we need to update the bg_itable_unused count
1213		 */
1214		if (ino > free)
1215			ext4_itable_unused_set(sb, gdp,
1216					(EXT4_INODES_PER_GROUP(sb) - ino));
1217		if (!(sbi->s_mount_state & EXT4_FC_REPLAY))
1218			up_read(&grp->alloc_sem);
1219	} else {
1220		ext4_lock_group(sb, group);
1221	}
1222
1223	ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1224	if (S_ISDIR(mode)) {
1225		ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1226		if (sbi->s_log_groups_per_flex) {
1227			ext4_group_t f = ext4_flex_group(sbi, group);
1228
1229			atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1230							f)->used_dirs);
1231		}
1232	}
1233	if (ext4_has_group_desc_csum(sb)) {
1234		ext4_inode_bitmap_csum_set(sb, gdp, inode_bitmap_bh);
1235		ext4_group_desc_csum_set(sb, group, gdp);
1236	}
1237	ext4_unlock_group(sb, group);
1238
1239	BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1240	err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1241	if (err) {
1242		ext4_std_error(sb, err);
1243		goto out;
1244	}
1245
1246	percpu_counter_dec(&sbi->s_freeinodes_counter);
1247	if (S_ISDIR(mode))
1248		percpu_counter_inc(&sbi->s_dirs_counter);
1249
1250	if (sbi->s_log_groups_per_flex) {
1251		flex_group = ext4_flex_group(sbi, group);
1252		atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1253						flex_group)->free_inodes);
1254	}
1255
1256	inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1257	/* This is the optimal IO size (for stat), not the fs block size */
1258	inode->i_blocks = 0;
1259	simple_inode_init_ts(inode);
1260	ei->i_crtime = inode_get_mtime(inode);
1261
1262	memset(ei->i_data, 0, sizeof(ei->i_data));
1263	ei->i_dir_start_lookup = 0;
1264	ei->i_disksize = 0;
1265
1266	/* Don't inherit extent flag from directory, amongst others. */
1267	ei->i_flags =
1268		ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1269	ei->i_flags |= i_flags;
1270	ei->i_file_acl = 0;
1271	ei->i_dtime = 0;
1272	ei->i_block_group = group;
1273	ei->i_last_alloc_group = ~0;
1274
1275	ext4_set_inode_flags(inode, true);
1276	if (IS_DIRSYNC(inode))
1277		ext4_handle_sync(handle);
1278	if (insert_inode_locked(inode) < 0) {
1279		/*
1280		 * Likely a bitmap corruption causing inode to be allocated
1281		 * twice.
1282		 */
1283		err = -EIO;
1284		ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1285			   inode->i_ino);
1286		ext4_mark_group_bitmap_corrupted(sb, group,
1287					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1288		goto out;
1289	}
1290	inode->i_generation = get_random_u32();
1291
1292	/* Precompute checksum seed for inode metadata */
1293	if (ext4_has_feature_metadata_csum(sb)) {
1294		__u32 csum;
1295		__le32 inum = cpu_to_le32(inode->i_ino);
1296		__le32 gen = cpu_to_le32(inode->i_generation);
1297		csum = ext4_chksum(sbi->s_csum_seed, (__u8 *)&inum,
1298				   sizeof(inum));
1299		ei->i_csum_seed = ext4_chksum(csum, (__u8 *)&gen, sizeof(gen));
1300	}
1301
1302	ext4_set_inode_state(inode, EXT4_STATE_NEW);
1303
1304	ei->i_extra_isize = sbi->s_want_extra_isize;
1305	ei->i_inline_off = 0;
1306	if (ext4_has_feature_inline_data(sb) &&
1307	    (!(ei->i_flags & (EXT4_DAX_FL|EXT4_EA_INODE_FL)) || S_ISDIR(mode)))
1308		ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1309	ret = inode;
1310	err = dquot_alloc_inode(inode);
1311	if (err)
1312		goto fail_drop;
1313
1314	/*
1315	 * Since the encryption xattr will always be unique, create it first so
1316	 * that it's less likely to end up in an external xattr block and
1317	 * prevent its deduplication.
1318	 */
1319	if (encrypt) {
1320		err = fscrypt_set_context(inode, handle);
1321		if (err)
1322			goto fail_free_drop;
1323	}
1324
1325	if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1326		err = ext4_init_acl(handle, inode, dir);
1327		if (err)
1328			goto fail_free_drop;
1329
1330		err = ext4_init_security(handle, inode, dir, qstr);
1331		if (err)
1332			goto fail_free_drop;
1333	}
1334
1335	if (ext4_has_feature_extents(sb)) {
1336		/* set extent flag only for directory, file and normal symlink*/
1337		if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1338			ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1339			ext4_ext_tree_init(handle, inode);
1340		}
1341	}
1342
1343	ext4_set_inode_mapping_order(inode);
1344
1345	ext4_update_inode_fsync_trans(handle, inode, 1);
1346
1347	err = ext4_mark_inode_dirty(handle, inode);
1348	if (err) {
1349		ext4_std_error(sb, err);
1350		goto fail_free_drop;
1351	}
1352
1353	ext4_debug("allocating inode %lu\n", inode->i_ino);
1354	trace_ext4_allocate_inode(inode, dir, mode);
1355	brelse(inode_bitmap_bh);
1356	return ret;
1357
1358fail_free_drop:
1359	dquot_free_inode(inode);
1360fail_drop:
1361	clear_nlink(inode);
1362	unlock_new_inode(inode);
1363out:
1364	dquot_drop(inode);
1365	inode->i_flags |= S_NOQUOTA;
1366	iput(inode);
1367	brelse(inode_bitmap_bh);
1368	return ERR_PTR(err);
1369}
1370
1371/* Verify that we are loading a valid orphan from disk */
1372struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1373{
1374	unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1375	ext4_group_t block_group;
1376	int bit;
1377	struct buffer_head *bitmap_bh = NULL;
1378	struct inode *inode = NULL;
1379	int err = -EFSCORRUPTED;
1380
1381	if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1382		goto bad_orphan;
1383
1384	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1385	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1386	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1387	if (IS_ERR(bitmap_bh))
1388		return ERR_CAST(bitmap_bh);
1389
1390	/* Having the inode bit set should be a 100% indicator that this
1391	 * is a valid orphan (no e2fsck run on fs).  Orphans also include
1392	 * inodes that were being truncated, so we can't check i_nlink==0.
1393	 */
1394	if (!ext4_test_bit(bit, bitmap_bh->b_data))
1395		goto bad_orphan;
1396
1397	inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1398	if (IS_ERR(inode)) {
1399		err = PTR_ERR(inode);
1400		ext4_error_err(sb, -err,
1401			       "couldn't read orphan inode %lu (err %d)",
1402			       ino, err);
1403		brelse(bitmap_bh);
1404		return inode;
1405	}
1406
1407	/*
1408	 * If the orphans has i_nlinks > 0 then it should be able to
1409	 * be truncated, otherwise it won't be removed from the orphan
1410	 * list during processing and an infinite loop will result.
1411	 * Similarly, it must not be a bad inode.
1412	 */
1413	if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1414	    is_bad_inode(inode))
1415		goto bad_orphan;
1416
1417	if (NEXT_ORPHAN(inode) > max_ino)
1418		goto bad_orphan;
1419	brelse(bitmap_bh);
1420	return inode;
1421
1422bad_orphan:
1423	ext4_error(sb, "bad orphan inode %lu", ino);
1424	if (bitmap_bh)
1425		printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1426		       bit, (unsigned long long)bitmap_bh->b_blocknr,
1427		       ext4_test_bit(bit, bitmap_bh->b_data));
1428	if (inode) {
1429		printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1430		       is_bad_inode(inode));
1431		printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1432		       NEXT_ORPHAN(inode));
1433		printk(KERN_ERR "max_ino=%lu\n", max_ino);
1434		printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1435		/* Avoid freeing blocks if we got a bad deleted inode */
1436		if (inode->i_nlink == 0)
1437			inode->i_blocks = 0;
1438		iput(inode);
1439	}
1440	brelse(bitmap_bh);
1441	return ERR_PTR(err);
1442}
1443
1444unsigned long ext4_count_free_inodes(struct super_block *sb)
1445{
1446	unsigned long desc_count;
1447	struct ext4_group_desc *gdp;
1448	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1449#ifdef EXT4FS_DEBUG
1450	struct ext4_super_block *es;
1451	unsigned long bitmap_count, x;
1452	struct buffer_head *bitmap_bh = NULL;
1453
1454	es = EXT4_SB(sb)->s_es;
1455	desc_count = 0;
1456	bitmap_count = 0;
1457	gdp = NULL;
1458	for (i = 0; i < ngroups; i++) {
1459		gdp = ext4_get_group_desc(sb, i, NULL);
1460		if (!gdp)
1461			continue;
1462		desc_count += ext4_free_inodes_count(sb, gdp);
1463		brelse(bitmap_bh);
1464		bitmap_bh = ext4_read_inode_bitmap(sb, i);
1465		if (IS_ERR(bitmap_bh)) {
1466			bitmap_bh = NULL;
1467			continue;
1468		}
1469
1470		x = ext4_count_free(bitmap_bh->b_data,
1471				    EXT4_INODES_PER_GROUP(sb) / 8);
1472		printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1473			(unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1474		bitmap_count += x;
1475	}
1476	brelse(bitmap_bh);
1477	printk(KERN_DEBUG "ext4_count_free_inodes: "
1478	       "stored = %u, computed = %lu, %lu\n",
1479	       le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1480	return desc_count;
1481#else
1482	desc_count = 0;
1483	for (i = 0; i < ngroups; i++) {
1484		gdp = ext4_get_group_desc(sb, i, NULL);
1485		if (!gdp)
1486			continue;
1487		desc_count += ext4_free_inodes_count(sb, gdp);
1488		cond_resched();
1489	}
1490	return desc_count;
1491#endif
1492}
1493
1494/* Called at mount-time, super-block is locked */
1495unsigned long ext4_count_dirs(struct super_block * sb)
1496{
1497	unsigned long count = 0;
1498	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1499
1500	for (i = 0; i < ngroups; i++) {
1501		struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1502		if (!gdp)
1503			continue;
1504		count += ext4_used_dirs_count(sb, gdp);
1505	}
1506	return count;
1507}
1508
1509/*
1510 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1511 * inode table. Must be called without any spinlock held. The only place
1512 * where it is called from on active part of filesystem is ext4lazyinit
1513 * thread, so we do not need any special locks, however we have to prevent
1514 * inode allocation from the current group, so we take alloc_sem lock, to
1515 * block ext4_new_inode() until we are finished.
1516 */
1517int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1518				 int barrier)
1519{
1520	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1521	struct ext4_sb_info *sbi = EXT4_SB(sb);
1522	struct ext4_group_desc *gdp = NULL;
1523	struct buffer_head *group_desc_bh;
1524	handle_t *handle;
1525	ext4_fsblk_t blk;
1526	int num, ret = 0, used_blks = 0;
1527	unsigned long used_inos = 0;
1528
1529	gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1530	if (!gdp || !grp)
1531		goto out;
1532
1533	/*
1534	 * We do not need to lock this, because we are the only one
1535	 * handling this flag.
1536	 */
1537	if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1538		goto out;
1539
1540	handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1541	if (IS_ERR(handle)) {
1542		ret = PTR_ERR(handle);
1543		goto out;
1544	}
1545
1546	down_write(&grp->alloc_sem);
1547	/*
1548	 * If inode bitmap was already initialized there may be some
1549	 * used inodes so we need to skip blocks with used inodes in
1550	 * inode table.
1551	 */
1552	if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
1553		used_inos = EXT4_INODES_PER_GROUP(sb) -
1554			    ext4_itable_unused_count(sb, gdp);
1555		used_blks = DIV_ROUND_UP(used_inos, sbi->s_inodes_per_block);
1556
1557		/* Bogus inode unused count? */
1558		if (used_blks < 0 || used_blks > sbi->s_itb_per_group) {
1559			ext4_error(sb, "Something is wrong with group %u: "
1560				   "used itable blocks: %d; "
1561				   "itable unused count: %u",
1562				   group, used_blks,
1563				   ext4_itable_unused_count(sb, gdp));
1564			ret = 1;
1565			goto err_out;
1566		}
1567
1568		used_inos += group * EXT4_INODES_PER_GROUP(sb);
1569		/*
1570		 * Are there some uninitialized inodes in the inode table
1571		 * before the first normal inode?
1572		 */
1573		if ((used_blks != sbi->s_itb_per_group) &&
1574		     (used_inos < EXT4_FIRST_INO(sb))) {
1575			ext4_error(sb, "Something is wrong with group %u: "
1576				   "itable unused count: %u; "
1577				   "itables initialized count: %ld",
1578				   group, ext4_itable_unused_count(sb, gdp),
1579				   used_inos);
1580			ret = 1;
1581			goto err_out;
1582		}
1583	}
1584
1585	blk = ext4_inode_table(sb, gdp) + used_blks;
1586	num = sbi->s_itb_per_group - used_blks;
1587
1588	BUFFER_TRACE(group_desc_bh, "get_write_access");
1589	ret = ext4_journal_get_write_access(handle, sb, group_desc_bh,
1590					    EXT4_JTR_NONE);
1591	if (ret)
1592		goto err_out;
1593
1594	/*
1595	 * Skip zeroout if the inode table is full. But we set the ZEROED
1596	 * flag anyway, because obviously, when it is full it does not need
1597	 * further zeroing.
1598	 */
1599	if (unlikely(num == 0))
1600		goto skip_zeroout;
1601
1602	ext4_debug("going to zero out inode table in group %d\n",
1603		   group);
1604	ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1605	if (ret < 0)
1606		goto err_out;
1607	if (barrier)
1608		blkdev_issue_flush(sb->s_bdev);
1609
1610skip_zeroout:
1611	ext4_lock_group(sb, group);
1612	gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1613	ext4_group_desc_csum_set(sb, group, gdp);
1614	ext4_unlock_group(sb, group);
1615
1616	BUFFER_TRACE(group_desc_bh,
1617		     "call ext4_handle_dirty_metadata");
1618	ret = ext4_handle_dirty_metadata(handle, NULL,
1619					 group_desc_bh);
1620
1621err_out:
1622	up_write(&grp->alloc_sem);
1623	ext4_journal_stop(handle);
1624out:
1625	return ret;
1626}