Linux Audio

Check our new training course

Loading...
v4.6
 
   1/*
   2 *  linux/fs/ext4/ialloc.c
   3 *
   4 * Copyright (C) 1992, 1993, 1994, 1995
   5 * Remy Card (card@masi.ibp.fr)
   6 * Laboratoire MASI - Institut Blaise Pascal
   7 * Universite Pierre et Marie Curie (Paris VI)
   8 *
   9 *  BSD ufs-inspired inode and directory allocation by
  10 *  Stephen Tweedie (sct@redhat.com), 1993
  11 *  Big-endian to little-endian byte-swapping/bitmaps by
  12 *        David S. Miller (davem@caip.rutgers.edu), 1995
  13 */
  14
  15#include <linux/time.h>
  16#include <linux/fs.h>
  17#include <linux/stat.h>
  18#include <linux/string.h>
  19#include <linux/quotaops.h>
  20#include <linux/buffer_head.h>
  21#include <linux/random.h>
  22#include <linux/bitops.h>
  23#include <linux/blkdev.h>
 
 
  24#include <asm/byteorder.h>
  25
  26#include "ext4.h"
  27#include "ext4_jbd2.h"
  28#include "xattr.h"
  29#include "acl.h"
  30
  31#include <trace/events/ext4.h>
  32
  33/*
  34 * ialloc.c contains the inodes allocation and deallocation routines
  35 */
  36
  37/*
  38 * The free inodes are managed by bitmaps.  A file system contains several
  39 * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
  40 * block for inodes, N blocks for the inode table and data blocks.
  41 *
  42 * The file system contains group descriptors which are located after the
  43 * super block.  Each descriptor contains the number of the bitmap block and
  44 * the free blocks count in the block.
  45 */
  46
  47/*
  48 * To avoid calling the atomic setbit hundreds or thousands of times, we only
  49 * need to use it within a single byte (to ensure we get endianness right).
  50 * We can use memset for the rest of the bitmap as there are no other users.
  51 */
  52void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
  53{
  54	int i;
  55
  56	if (start_bit >= end_bit)
  57		return;
  58
  59	ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
  60	for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
  61		ext4_set_bit(i, bitmap);
  62	if (i < end_bit)
  63		memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
  64}
  65
  66/* Initializes an uninitialized inode bitmap */
  67static int ext4_init_inode_bitmap(struct super_block *sb,
  68				       struct buffer_head *bh,
  69				       ext4_group_t block_group,
  70				       struct ext4_group_desc *gdp)
  71{
  72	struct ext4_group_info *grp;
  73	struct ext4_sb_info *sbi = EXT4_SB(sb);
  74	J_ASSERT_BH(bh, buffer_locked(bh));
  75
  76	/* If checksum is bad mark all blocks and inodes use to prevent
  77	 * allocation, essentially implementing a per-group read-only flag. */
  78	if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
  79		grp = ext4_get_group_info(sb, block_group);
  80		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
  81			percpu_counter_sub(&sbi->s_freeclusters_counter,
  82					   grp->bb_free);
  83		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
  84		if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
  85			int count;
  86			count = ext4_free_inodes_count(sb, gdp);
  87			percpu_counter_sub(&sbi->s_freeinodes_counter,
  88					   count);
  89		}
  90		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
  91		return -EFSBADCRC;
  92	}
  93
  94	memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
  95	ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
  96			bh->b_data);
  97	ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
  98				   EXT4_INODES_PER_GROUP(sb) / 8);
  99	ext4_group_desc_csum_set(sb, block_group, gdp);
 100
 101	return 0;
 102}
 103
 104void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
 105{
 106	if (uptodate) {
 107		set_buffer_uptodate(bh);
 108		set_bitmap_uptodate(bh);
 109	}
 110	unlock_buffer(bh);
 111	put_bh(bh);
 112}
 113
 114static int ext4_validate_inode_bitmap(struct super_block *sb,
 115				      struct ext4_group_desc *desc,
 116				      ext4_group_t block_group,
 117				      struct buffer_head *bh)
 118{
 119	ext4_fsblk_t	blk;
 120	struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
 121	struct ext4_sb_info *sbi = EXT4_SB(sb);
 122
 123	if (buffer_verified(bh))
 124		return 0;
 125	if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
 126		return -EFSCORRUPTED;
 127
 128	ext4_lock_group(sb, block_group);
 
 
 129	blk = ext4_inode_bitmap(sb, desc);
 130	if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
 131					   EXT4_INODES_PER_GROUP(sb) / 8)) {
 132		ext4_unlock_group(sb, block_group);
 133		ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
 134			   "inode_bitmap = %llu", block_group, blk);
 135		grp = ext4_get_group_info(sb, block_group);
 136		if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 137			int count;
 138			count = ext4_free_inodes_count(sb, desc);
 139			percpu_counter_sub(&sbi->s_freeinodes_counter,
 140					   count);
 141		}
 142		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 143		return -EFSBADCRC;
 144	}
 145	set_buffer_verified(bh);
 
 146	ext4_unlock_group(sb, block_group);
 147	return 0;
 148}
 149
 150/*
 151 * Read the inode allocation bitmap for a given block_group, reading
 152 * into the specified slot in the superblock's bitmap cache.
 153 *
 154 * Return buffer_head of bitmap on success or NULL.
 155 */
 156static struct buffer_head *
 157ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 158{
 159	struct ext4_group_desc *desc;
 
 160	struct buffer_head *bh = NULL;
 161	ext4_fsblk_t bitmap_blk;
 162	int err;
 163
 164	desc = ext4_get_group_desc(sb, block_group, NULL);
 165	if (!desc)
 166		return ERR_PTR(-EFSCORRUPTED);
 167
 168	bitmap_blk = ext4_inode_bitmap(sb, desc);
 
 
 
 
 
 
 
 
 169	bh = sb_getblk(sb, bitmap_blk);
 170	if (unlikely(!bh)) {
 171		ext4_error(sb, "Cannot read inode bitmap - "
 172			    "block_group = %u, inode_bitmap = %llu",
 173			    block_group, bitmap_blk);
 174		return ERR_PTR(-EIO);
 175	}
 176	if (bitmap_uptodate(bh))
 177		goto verify;
 178
 179	lock_buffer(bh);
 180	if (bitmap_uptodate(bh)) {
 181		unlock_buffer(bh);
 182		goto verify;
 183	}
 184
 185	ext4_lock_group(sb, block_group);
 186	if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 187		err = ext4_init_inode_bitmap(sb, bh, block_group, desc);
 
 
 
 
 
 
 
 
 
 
 
 188		set_bitmap_uptodate(bh);
 189		set_buffer_uptodate(bh);
 190		set_buffer_verified(bh);
 191		ext4_unlock_group(sb, block_group);
 192		unlock_buffer(bh);
 193		if (err) {
 194			ext4_error(sb, "Failed to init inode bitmap for group "
 195				   "%u: %d", block_group, err);
 196			goto out;
 197		}
 198		return bh;
 199	}
 200	ext4_unlock_group(sb, block_group);
 201
 202	if (buffer_uptodate(bh)) {
 203		/*
 204		 * if not uninit if bh is uptodate,
 205		 * bitmap is also uptodate
 206		 */
 207		set_bitmap_uptodate(bh);
 208		unlock_buffer(bh);
 209		goto verify;
 210	}
 211	/*
 212	 * submit the buffer_head for reading
 213	 */
 214	trace_ext4_load_inode_bitmap(sb, block_group);
 215	bh->b_end_io = ext4_end_bitmap_read;
 216	get_bh(bh);
 217	submit_bh(READ | REQ_META | REQ_PRIO, bh);
 218	wait_on_buffer(bh);
 219	if (!buffer_uptodate(bh)) {
 220		put_bh(bh);
 221		ext4_error(sb, "Cannot read inode bitmap - "
 222			   "block_group = %u, inode_bitmap = %llu",
 223			   block_group, bitmap_blk);
 
 
 224		return ERR_PTR(-EIO);
 225	}
 226
 227verify:
 228	err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
 229	if (err)
 230		goto out;
 231	return bh;
 232out:
 233	put_bh(bh);
 234	return ERR_PTR(err);
 235}
 236
 237/*
 238 * NOTE! When we get the inode, we're the only people
 239 * that have access to it, and as such there are no
 240 * race conditions we have to worry about. The inode
 241 * is not on the hash-lists, and it cannot be reached
 242 * through the filesystem because the directory entry
 243 * has been deleted earlier.
 244 *
 245 * HOWEVER: we must make sure that we get no aliases,
 246 * which means that we have to call "clear_inode()"
 247 * _before_ we mark the inode not in use in the inode
 248 * bitmaps. Otherwise a newly created file might use
 249 * the same inode number (not actually the same pointer
 250 * though), and then we'd have two inodes sharing the
 251 * same inode number and space on the harddisk.
 252 */
 253void ext4_free_inode(handle_t *handle, struct inode *inode)
 254{
 255	struct super_block *sb = inode->i_sb;
 256	int is_directory;
 257	unsigned long ino;
 258	struct buffer_head *bitmap_bh = NULL;
 259	struct buffer_head *bh2;
 260	ext4_group_t block_group;
 261	unsigned long bit;
 262	struct ext4_group_desc *gdp;
 263	struct ext4_super_block *es;
 264	struct ext4_sb_info *sbi;
 265	int fatal = 0, err, count, cleared;
 266	struct ext4_group_info *grp;
 267
 268	if (!sb) {
 269		printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 270		       "nonexistent device\n", __func__, __LINE__);
 271		return;
 272	}
 273	if (atomic_read(&inode->i_count) > 1) {
 274		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 275			 __func__, __LINE__, inode->i_ino,
 276			 atomic_read(&inode->i_count));
 277		return;
 278	}
 279	if (inode->i_nlink) {
 280		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 281			 __func__, __LINE__, inode->i_ino, inode->i_nlink);
 282		return;
 283	}
 284	sbi = EXT4_SB(sb);
 285
 286	ino = inode->i_ino;
 287	ext4_debug("freeing inode %lu\n", ino);
 288	trace_ext4_free_inode(inode);
 289
 290	/*
 291	 * Note: we must free any quota before locking the superblock,
 292	 * as writing the quota to disk may need the lock as well.
 293	 */
 294	dquot_initialize(inode);
 295	ext4_xattr_delete_inode(handle, inode);
 296	dquot_free_inode(inode);
 297	dquot_drop(inode);
 298
 299	is_directory = S_ISDIR(inode->i_mode);
 300
 301	/* Do this BEFORE marking the inode not in use or returning an error */
 302	ext4_clear_inode(inode);
 303
 304	es = EXT4_SB(sb)->s_es;
 305	if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 306		ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 307		goto error_return;
 308	}
 309	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 310	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 311	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 312	/* Don't bother if the inode bitmap is corrupt. */
 313	grp = ext4_get_group_info(sb, block_group);
 314	if (IS_ERR(bitmap_bh)) {
 315		fatal = PTR_ERR(bitmap_bh);
 316		bitmap_bh = NULL;
 317		goto error_return;
 318	}
 319	if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
 320		fatal = -EFSCORRUPTED;
 321		goto error_return;
 322	}
 323
 324	BUFFER_TRACE(bitmap_bh, "get_write_access");
 325	fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 326	if (fatal)
 327		goto error_return;
 328
 329	fatal = -ESRCH;
 330	gdp = ext4_get_group_desc(sb, block_group, &bh2);
 331	if (gdp) {
 332		BUFFER_TRACE(bh2, "get_write_access");
 333		fatal = ext4_journal_get_write_access(handle, bh2);
 334	}
 335	ext4_lock_group(sb, block_group);
 336	cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 337	if (fatal || !cleared) {
 338		ext4_unlock_group(sb, block_group);
 339		goto out;
 340	}
 341
 342	count = ext4_free_inodes_count(sb, gdp) + 1;
 343	ext4_free_inodes_set(sb, gdp, count);
 344	if (is_directory) {
 345		count = ext4_used_dirs_count(sb, gdp) - 1;
 346		ext4_used_dirs_set(sb, gdp, count);
 347		percpu_counter_dec(&sbi->s_dirs_counter);
 348	}
 349	ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
 350				   EXT4_INODES_PER_GROUP(sb) / 8);
 351	ext4_group_desc_csum_set(sb, block_group, gdp);
 352	ext4_unlock_group(sb, block_group);
 353
 354	percpu_counter_inc(&sbi->s_freeinodes_counter);
 355	if (sbi->s_log_groups_per_flex) {
 356		ext4_group_t f = ext4_flex_group(sbi, block_group);
 357
 358		atomic_inc(&sbi->s_flex_groups[f].free_inodes);
 359		if (is_directory)
 360			atomic_dec(&sbi->s_flex_groups[f].used_dirs);
 361	}
 362	BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 363	fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 364out:
 365	if (cleared) {
 366		BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 367		err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 368		if (!fatal)
 369			fatal = err;
 370	} else {
 371		ext4_error(sb, "bit already cleared for inode %lu", ino);
 372		if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 373			int count;
 374			count = ext4_free_inodes_count(sb, gdp);
 375			percpu_counter_sub(&sbi->s_freeinodes_counter,
 376					   count);
 377		}
 378		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 379	}
 380
 381error_return:
 382	brelse(bitmap_bh);
 383	ext4_std_error(sb, fatal);
 384}
 385
 386struct orlov_stats {
 387	__u64 free_clusters;
 388	__u32 free_inodes;
 389	__u32 used_dirs;
 390};
 391
 392/*
 393 * Helper function for Orlov's allocator; returns critical information
 394 * for a particular block group or flex_bg.  If flex_size is 1, then g
 395 * is a block group number; otherwise it is flex_bg number.
 396 */
 397static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 398			    int flex_size, struct orlov_stats *stats)
 399{
 400	struct ext4_group_desc *desc;
 401	struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
 402
 403	if (flex_size > 1) {
 404		stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
 405		stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
 406		stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
 407		return;
 408	}
 409
 410	desc = ext4_get_group_desc(sb, g, NULL);
 411	if (desc) {
 412		stats->free_inodes = ext4_free_inodes_count(sb, desc);
 413		stats->free_clusters = ext4_free_group_clusters(sb, desc);
 414		stats->used_dirs = ext4_used_dirs_count(sb, desc);
 415	} else {
 416		stats->free_inodes = 0;
 417		stats->free_clusters = 0;
 418		stats->used_dirs = 0;
 419	}
 420}
 421
 422/*
 423 * Orlov's allocator for directories.
 424 *
 425 * We always try to spread first-level directories.
 426 *
 427 * If there are blockgroups with both free inodes and free blocks counts
 428 * not worse than average we return one with smallest directory count.
 429 * Otherwise we simply return a random group.
 430 *
 431 * For the rest rules look so:
 432 *
 433 * It's OK to put directory into a group unless
 434 * it has too many directories already (max_dirs) or
 435 * it has too few free inodes left (min_inodes) or
 436 * it has too few free blocks left (min_blocks) or
 437 * Parent's group is preferred, if it doesn't satisfy these
 438 * conditions we search cyclically through the rest. If none
 439 * of the groups look good we just look for a group with more
 440 * free inodes than average (starting at parent's group).
 441 */
 442
 443static int find_group_orlov(struct super_block *sb, struct inode *parent,
 444			    ext4_group_t *group, umode_t mode,
 445			    const struct qstr *qstr)
 446{
 447	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 448	struct ext4_sb_info *sbi = EXT4_SB(sb);
 449	ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 450	int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 451	unsigned int freei, avefreei, grp_free;
 452	ext4_fsblk_t freeb, avefreec;
 453	unsigned int ndirs;
 454	int max_dirs, min_inodes;
 455	ext4_grpblk_t min_clusters;
 456	ext4_group_t i, grp, g, ngroups;
 457	struct ext4_group_desc *desc;
 458	struct orlov_stats stats;
 459	int flex_size = ext4_flex_bg_size(sbi);
 460	struct dx_hash_info hinfo;
 461
 462	ngroups = real_ngroups;
 463	if (flex_size > 1) {
 464		ngroups = (real_ngroups + flex_size - 1) >>
 465			sbi->s_log_groups_per_flex;
 466		parent_group >>= sbi->s_log_groups_per_flex;
 467	}
 468
 469	freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 470	avefreei = freei / ngroups;
 471	freeb = EXT4_C2B(sbi,
 472		percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 473	avefreec = freeb;
 474	do_div(avefreec, ngroups);
 475	ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 476
 477	if (S_ISDIR(mode) &&
 478	    ((parent == d_inode(sb->s_root)) ||
 479	     (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 480		int best_ndir = inodes_per_group;
 481		int ret = -1;
 482
 483		if (qstr) {
 484			hinfo.hash_version = DX_HASH_HALF_MD4;
 485			hinfo.seed = sbi->s_hash_seed;
 486			ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
 487			grp = hinfo.hash;
 488		} else
 489			grp = prandom_u32();
 490		parent_group = (unsigned)grp % ngroups;
 491		for (i = 0; i < ngroups; i++) {
 492			g = (parent_group + i) % ngroups;
 493			get_orlov_stats(sb, g, flex_size, &stats);
 494			if (!stats.free_inodes)
 495				continue;
 496			if (stats.used_dirs >= best_ndir)
 497				continue;
 498			if (stats.free_inodes < avefreei)
 499				continue;
 500			if (stats.free_clusters < avefreec)
 501				continue;
 502			grp = g;
 503			ret = 0;
 504			best_ndir = stats.used_dirs;
 505		}
 506		if (ret)
 507			goto fallback;
 508	found_flex_bg:
 509		if (flex_size == 1) {
 510			*group = grp;
 511			return 0;
 512		}
 513
 514		/*
 515		 * We pack inodes at the beginning of the flexgroup's
 516		 * inode tables.  Block allocation decisions will do
 517		 * something similar, although regular files will
 518		 * start at 2nd block group of the flexgroup.  See
 519		 * ext4_ext_find_goal() and ext4_find_near().
 520		 */
 521		grp *= flex_size;
 522		for (i = 0; i < flex_size; i++) {
 523			if (grp+i >= real_ngroups)
 524				break;
 525			desc = ext4_get_group_desc(sb, grp+i, NULL);
 526			if (desc && ext4_free_inodes_count(sb, desc)) {
 527				*group = grp+i;
 528				return 0;
 529			}
 530		}
 531		goto fallback;
 532	}
 533
 534	max_dirs = ndirs / ngroups + inodes_per_group / 16;
 535	min_inodes = avefreei - inodes_per_group*flex_size / 4;
 536	if (min_inodes < 1)
 537		min_inodes = 1;
 538	min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 539
 540	/*
 541	 * Start looking in the flex group where we last allocated an
 542	 * inode for this parent directory
 543	 */
 544	if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 545		parent_group = EXT4_I(parent)->i_last_alloc_group;
 546		if (flex_size > 1)
 547			parent_group >>= sbi->s_log_groups_per_flex;
 548	}
 549
 550	for (i = 0; i < ngroups; i++) {
 551		grp = (parent_group + i) % ngroups;
 552		get_orlov_stats(sb, grp, flex_size, &stats);
 553		if (stats.used_dirs >= max_dirs)
 554			continue;
 555		if (stats.free_inodes < min_inodes)
 556			continue;
 557		if (stats.free_clusters < min_clusters)
 558			continue;
 559		goto found_flex_bg;
 560	}
 561
 562fallback:
 563	ngroups = real_ngroups;
 564	avefreei = freei / ngroups;
 565fallback_retry:
 566	parent_group = EXT4_I(parent)->i_block_group;
 567	for (i = 0; i < ngroups; i++) {
 568		grp = (parent_group + i) % ngroups;
 569		desc = ext4_get_group_desc(sb, grp, NULL);
 570		if (desc) {
 571			grp_free = ext4_free_inodes_count(sb, desc);
 572			if (grp_free && grp_free >= avefreei) {
 573				*group = grp;
 574				return 0;
 575			}
 576		}
 577	}
 578
 579	if (avefreei) {
 580		/*
 581		 * The free-inodes counter is approximate, and for really small
 582		 * filesystems the above test can fail to find any blockgroups
 583		 */
 584		avefreei = 0;
 585		goto fallback_retry;
 586	}
 587
 588	return -1;
 589}
 590
 591static int find_group_other(struct super_block *sb, struct inode *parent,
 592			    ext4_group_t *group, umode_t mode)
 593{
 594	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 595	ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 596	struct ext4_group_desc *desc;
 597	int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 598
 599	/*
 600	 * Try to place the inode is the same flex group as its
 601	 * parent.  If we can't find space, use the Orlov algorithm to
 602	 * find another flex group, and store that information in the
 603	 * parent directory's inode information so that use that flex
 604	 * group for future allocations.
 605	 */
 606	if (flex_size > 1) {
 607		int retry = 0;
 608
 609	try_again:
 610		parent_group &= ~(flex_size-1);
 611		last = parent_group + flex_size;
 612		if (last > ngroups)
 613			last = ngroups;
 614		for  (i = parent_group; i < last; i++) {
 615			desc = ext4_get_group_desc(sb, i, NULL);
 616			if (desc && ext4_free_inodes_count(sb, desc)) {
 617				*group = i;
 618				return 0;
 619			}
 620		}
 621		if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 622			retry = 1;
 623			parent_group = EXT4_I(parent)->i_last_alloc_group;
 624			goto try_again;
 625		}
 626		/*
 627		 * If this didn't work, use the Orlov search algorithm
 628		 * to find a new flex group; we pass in the mode to
 629		 * avoid the topdir algorithms.
 630		 */
 631		*group = parent_group + flex_size;
 632		if (*group > ngroups)
 633			*group = 0;
 634		return find_group_orlov(sb, parent, group, mode, NULL);
 635	}
 636
 637	/*
 638	 * Try to place the inode in its parent directory
 639	 */
 640	*group = parent_group;
 641	desc = ext4_get_group_desc(sb, *group, NULL);
 642	if (desc && ext4_free_inodes_count(sb, desc) &&
 643	    ext4_free_group_clusters(sb, desc))
 644		return 0;
 645
 646	/*
 647	 * We're going to place this inode in a different blockgroup from its
 648	 * parent.  We want to cause files in a common directory to all land in
 649	 * the same blockgroup.  But we want files which are in a different
 650	 * directory which shares a blockgroup with our parent to land in a
 651	 * different blockgroup.
 652	 *
 653	 * So add our directory's i_ino into the starting point for the hash.
 654	 */
 655	*group = (*group + parent->i_ino) % ngroups;
 656
 657	/*
 658	 * Use a quadratic hash to find a group with a free inode and some free
 659	 * blocks.
 660	 */
 661	for (i = 1; i < ngroups; i <<= 1) {
 662		*group += i;
 663		if (*group >= ngroups)
 664			*group -= ngroups;
 665		desc = ext4_get_group_desc(sb, *group, NULL);
 666		if (desc && ext4_free_inodes_count(sb, desc) &&
 667		    ext4_free_group_clusters(sb, desc))
 668			return 0;
 669	}
 670
 671	/*
 672	 * That failed: try linear search for a free inode, even if that group
 673	 * has no free blocks.
 674	 */
 675	*group = parent_group;
 676	for (i = 0; i < ngroups; i++) {
 677		if (++*group >= ngroups)
 678			*group = 0;
 679		desc = ext4_get_group_desc(sb, *group, NULL);
 680		if (desc && ext4_free_inodes_count(sb, desc))
 681			return 0;
 682	}
 683
 684	return -1;
 685}
 686
 687/*
 688 * In no journal mode, if an inode has recently been deleted, we want
 689 * to avoid reusing it until we're reasonably sure the inode table
 690 * block has been written back to disk.  (Yes, these values are
 691 * somewhat arbitrary...)
 692 */
 693#define RECENTCY_MIN	5
 694#define RECENTCY_DIRTY	30
 695
 696static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 697{
 698	struct ext4_group_desc	*gdp;
 699	struct ext4_inode	*raw_inode;
 700	struct buffer_head	*bh;
 701	unsigned long		dtime, now;
 702	int	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 703	int	offset, ret = 0, recentcy = RECENTCY_MIN;
 
 704
 705	gdp = ext4_get_group_desc(sb, group, NULL);
 706	if (unlikely(!gdp))
 707		return 0;
 708
 709	bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
 710		       (ino / inodes_per_block));
 711	if (unlikely(!bh) || !buffer_uptodate(bh))
 712		/*
 713		 * If the block is not in the buffer cache, then it
 714		 * must have been written out.
 715		 */
 716		goto out;
 717
 718	offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 719	raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 
 
 
 
 
 720	dtime = le32_to_cpu(raw_inode->i_dtime);
 721	now = get_seconds();
 722	if (buffer_dirty(bh))
 723		recentcy += RECENTCY_DIRTY;
 724
 725	if (dtime && (dtime < now) && (now < dtime + recentcy))
 
 726		ret = 1;
 727out:
 728	brelse(bh);
 729	return ret;
 730}
 731
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 732/*
 733 * There are two policies for allocating an inode.  If the new inode is
 734 * a directory, then a forward search is made for a block group with both
 735 * free space and a low directory-to-inode ratio; if that fails, then of
 736 * the groups with above-average free space, that group with the fewest
 737 * directories already is chosen.
 738 *
 739 * For other inodes, search forward from the parent directory's block
 740 * group to find a free inode.
 741 */
 742struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 743			       umode_t mode, const struct qstr *qstr,
 744			       __u32 goal, uid_t *owner, int handle_type,
 745			       unsigned int line_no, int nblocks)
 
 746{
 747	struct super_block *sb;
 748	struct buffer_head *inode_bitmap_bh = NULL;
 749	struct buffer_head *group_desc_bh;
 750	ext4_group_t ngroups, group = 0;
 751	unsigned long ino = 0;
 752	struct inode *inode;
 753	struct ext4_group_desc *gdp = NULL;
 754	struct ext4_inode_info *ei;
 755	struct ext4_sb_info *sbi;
 756	int ret2, err;
 757	struct inode *ret;
 758	ext4_group_t i;
 759	ext4_group_t flex_group;
 760	struct ext4_group_info *grp;
 761	int encrypt = 0;
 762
 763	/* Cannot create files in a deleted directory */
 764	if (!dir || !dir->i_nlink)
 765		return ERR_PTR(-EPERM);
 766
 767	if ((ext4_encrypted_inode(dir) ||
 768	     DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) &&
 769	    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
 770		err = ext4_get_encryption_info(dir);
 
 
 
 
 
 
 771		if (err)
 772			return ERR_PTR(err);
 773		if (ext4_encryption_info(dir) == NULL)
 774			return ERR_PTR(-EPERM);
 775		if (!handle)
 776			nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb);
 777		encrypt = 1;
 778	}
 779
 780	sb = dir->i_sb;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 781	ngroups = ext4_get_groups_count(sb);
 782	trace_ext4_request_inode(dir, mode);
 783	inode = new_inode(sb);
 784	if (!inode)
 785		return ERR_PTR(-ENOMEM);
 786	ei = EXT4_I(inode);
 787	sbi = EXT4_SB(sb);
 788
 789	/*
 790	 * Initialize owners and quota early so that we don't have to account
 791	 * for quota initialization worst case in standard inode creating
 792	 * transaction
 793	 */
 794	if (owner) {
 795		inode->i_mode = mode;
 796		i_uid_write(inode, owner[0]);
 797		i_gid_write(inode, owner[1]);
 798	} else if (test_opt(sb, GRPID)) {
 799		inode->i_mode = mode;
 800		inode->i_uid = current_fsuid();
 801		inode->i_gid = dir->i_gid;
 802	} else
 803		inode_init_owner(inode, dir, mode);
 804
 805	if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_PROJECT) &&
 806	    ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
 807		ei->i_projid = EXT4_I(dir)->i_projid;
 808	else
 809		ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
 810
 811	err = dquot_initialize(inode);
 812	if (err)
 813		goto out;
 814
 815	if (!goal)
 816		goal = sbi->s_inode_goal;
 817
 818	if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 819		group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 820		ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 821		ret2 = 0;
 822		goto got_group;
 823	}
 824
 825	if (S_ISDIR(mode))
 826		ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 827	else
 828		ret2 = find_group_other(sb, dir, &group, mode);
 829
 830got_group:
 831	EXT4_I(dir)->i_last_alloc_group = group;
 832	err = -ENOSPC;
 833	if (ret2 == -1)
 834		goto out;
 835
 836	/*
 837	 * Normally we will only go through one pass of this loop,
 838	 * unless we get unlucky and it turns out the group we selected
 839	 * had its last inode grabbed by someone else.
 840	 */
 841	for (i = 0; i < ngroups; i++, ino = 0) {
 842		err = -EIO;
 843
 844		gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 845		if (!gdp)
 846			goto out;
 847
 848		/*
 849		 * Check free inodes count before loading bitmap.
 850		 */
 851		if (ext4_free_inodes_count(sb, gdp) == 0) {
 852			if (++group == ngroups)
 853				group = 0;
 854			continue;
 855		}
 856
 857		grp = ext4_get_group_info(sb, group);
 858		/* Skip groups with already-known suspicious inode tables */
 859		if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 860			if (++group == ngroups)
 861				group = 0;
 862			continue;
 863		}
 864
 865		brelse(inode_bitmap_bh);
 866		inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 867		/* Skip groups with suspicious inode tables */
 868		if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
 869		    IS_ERR(inode_bitmap_bh)) {
 870			inode_bitmap_bh = NULL;
 871			if (++group == ngroups)
 872				group = 0;
 873			continue;
 874		}
 875
 876repeat_in_this_group:
 877		ino = ext4_find_next_zero_bit((unsigned long *)
 878					      inode_bitmap_bh->b_data,
 879					      EXT4_INODES_PER_GROUP(sb), ino);
 880		if (ino >= EXT4_INODES_PER_GROUP(sb))
 881			goto next_group;
 882		if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
 
 883			ext4_error(sb, "reserved inode found cleared - "
 884				   "inode=%lu", ino + 1);
 885			continue;
 886		}
 887		if ((EXT4_SB(sb)->s_journal == NULL) &&
 888		    recently_deleted(sb, group, ino)) {
 889			ino++;
 890			goto next_inode;
 891		}
 
 892		if (!handle) {
 893			BUG_ON(nblocks <= 0);
 894			handle = __ext4_journal_start_sb(dir->i_sb, line_no,
 895							 handle_type, nblocks,
 896							 0);
 897			if (IS_ERR(handle)) {
 898				err = PTR_ERR(handle);
 899				ext4_std_error(sb, err);
 900				goto out;
 901			}
 902		}
 903		BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 904		err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
 905		if (err) {
 906			ext4_std_error(sb, err);
 907			goto out;
 908		}
 909		ext4_lock_group(sb, group);
 910		ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
 
 
 
 
 
 
 
 
 
 
 
 
 911		ext4_unlock_group(sb, group);
 912		ino++;		/* the inode bitmap is zero-based */
 913		if (!ret2)
 914			goto got; /* we grabbed the inode! */
 915next_inode:
 916		if (ino < EXT4_INODES_PER_GROUP(sb))
 917			goto repeat_in_this_group;
 918next_group:
 919		if (++group == ngroups)
 920			group = 0;
 921	}
 922	err = -ENOSPC;
 923	goto out;
 924
 925got:
 926	BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 927	err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
 928	if (err) {
 929		ext4_std_error(sb, err);
 930		goto out;
 931	}
 932
 933	BUFFER_TRACE(group_desc_bh, "get_write_access");
 934	err = ext4_journal_get_write_access(handle, group_desc_bh);
 935	if (err) {
 936		ext4_std_error(sb, err);
 937		goto out;
 938	}
 939
 940	/* We may have to initialize the block bitmap if it isn't already */
 941	if (ext4_has_group_desc_csum(sb) &&
 942	    gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 943		struct buffer_head *block_bitmap_bh;
 944
 945		block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 946		if (IS_ERR(block_bitmap_bh)) {
 947			err = PTR_ERR(block_bitmap_bh);
 948			goto out;
 949		}
 950		BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
 951		err = ext4_journal_get_write_access(handle, block_bitmap_bh);
 952		if (err) {
 953			brelse(block_bitmap_bh);
 954			ext4_std_error(sb, err);
 955			goto out;
 956		}
 957
 958		BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
 959		err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
 960
 961		/* recheck and clear flag under lock if we still need to */
 962		ext4_lock_group(sb, group);
 963		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 
 964			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
 965			ext4_free_group_clusters_set(sb, gdp,
 966				ext4_free_clusters_after_init(sb, group, gdp));
 967			ext4_block_bitmap_csum_set(sb, group, gdp,
 968						   block_bitmap_bh);
 969			ext4_group_desc_csum_set(sb, group, gdp);
 970		}
 971		ext4_unlock_group(sb, group);
 972		brelse(block_bitmap_bh);
 973
 974		if (err) {
 975			ext4_std_error(sb, err);
 976			goto out;
 977		}
 978	}
 979
 980	/* Update the relevant bg descriptor fields */
 981	if (ext4_has_group_desc_csum(sb)) {
 982		int free;
 983		struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 984
 985		down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
 986		ext4_lock_group(sb, group); /* while we modify the bg desc */
 987		free = EXT4_INODES_PER_GROUP(sb) -
 988			ext4_itable_unused_count(sb, gdp);
 989		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 990			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
 991			free = 0;
 992		}
 993		/*
 994		 * Check the relative inode number against the last used
 995		 * relative inode number in this group. if it is greater
 996		 * we need to update the bg_itable_unused count
 997		 */
 998		if (ino > free)
 999			ext4_itable_unused_set(sb, gdp,
1000					(EXT4_INODES_PER_GROUP(sb) - ino));
1001		up_read(&grp->alloc_sem);
1002	} else {
1003		ext4_lock_group(sb, group);
1004	}
1005
1006	ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1007	if (S_ISDIR(mode)) {
1008		ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1009		if (sbi->s_log_groups_per_flex) {
1010			ext4_group_t f = ext4_flex_group(sbi, group);
1011
1012			atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1013		}
1014	}
1015	if (ext4_has_group_desc_csum(sb)) {
1016		ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1017					   EXT4_INODES_PER_GROUP(sb) / 8);
1018		ext4_group_desc_csum_set(sb, group, gdp);
1019	}
1020	ext4_unlock_group(sb, group);
1021
1022	BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1023	err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1024	if (err) {
1025		ext4_std_error(sb, err);
1026		goto out;
1027	}
1028
1029	percpu_counter_dec(&sbi->s_freeinodes_counter);
1030	if (S_ISDIR(mode))
1031		percpu_counter_inc(&sbi->s_dirs_counter);
1032
1033	if (sbi->s_log_groups_per_flex) {
1034		flex_group = ext4_flex_group(sbi, group);
1035		atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1036	}
1037
1038	inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1039	/* This is the optimal IO size (for stat), not the fs block size */
1040	inode->i_blocks = 0;
1041	inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1042						       ext4_current_time(inode);
1043
1044	memset(ei->i_data, 0, sizeof(ei->i_data));
1045	ei->i_dir_start_lookup = 0;
1046	ei->i_disksize = 0;
1047
1048	/* Don't inherit extent flag from directory, amongst others. */
1049	ei->i_flags =
1050		ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
 
1051	ei->i_file_acl = 0;
1052	ei->i_dtime = 0;
1053	ei->i_block_group = group;
1054	ei->i_last_alloc_group = ~0;
1055
1056	ext4_set_inode_flags(inode);
1057	if (IS_DIRSYNC(inode))
1058		ext4_handle_sync(handle);
1059	if (insert_inode_locked(inode) < 0) {
1060		/*
1061		 * Likely a bitmap corruption causing inode to be allocated
1062		 * twice.
1063		 */
1064		err = -EIO;
1065		ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1066			   inode->i_ino);
 
 
1067		goto out;
1068	}
1069	spin_lock(&sbi->s_next_gen_lock);
1070	inode->i_generation = sbi->s_next_generation++;
1071	spin_unlock(&sbi->s_next_gen_lock);
1072
1073	/* Precompute checksum seed for inode metadata */
1074	if (ext4_has_metadata_csum(sb)) {
1075		__u32 csum;
1076		__le32 inum = cpu_to_le32(inode->i_ino);
1077		__le32 gen = cpu_to_le32(inode->i_generation);
1078		csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1079				   sizeof(inum));
1080		ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1081					      sizeof(gen));
1082	}
1083
1084	ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1085	ext4_set_inode_state(inode, EXT4_STATE_NEW);
1086
1087	ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1088	ei->i_inline_off = 0;
1089	if (ext4_has_feature_inline_data(sb))
1090		ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1091	ret = inode;
1092	err = dquot_alloc_inode(inode);
1093	if (err)
1094		goto fail_drop;
1095
1096	err = ext4_init_acl(handle, inode, dir);
1097	if (err)
1098		goto fail_free_drop;
 
 
 
 
 
 
 
1099
1100	err = ext4_init_security(handle, inode, dir, qstr);
1101	if (err)
1102		goto fail_free_drop;
 
 
 
 
 
 
1103
1104	if (ext4_has_feature_extents(sb)) {
1105		/* set extent flag only for directory, file and normal symlink*/
1106		if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1107			ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1108			ext4_ext_tree_init(handle, inode);
1109		}
1110	}
1111
1112	if (ext4_handle_valid(handle)) {
1113		ei->i_sync_tid = handle->h_transaction->t_tid;
1114		ei->i_datasync_tid = handle->h_transaction->t_tid;
1115	}
1116
1117	if (encrypt) {
1118		err = ext4_inherit_context(dir, inode);
1119		if (err)
1120			goto fail_free_drop;
1121	}
1122
1123	err = ext4_mark_inode_dirty(handle, inode);
1124	if (err) {
1125		ext4_std_error(sb, err);
1126		goto fail_free_drop;
1127	}
1128
1129	ext4_debug("allocating inode %lu\n", inode->i_ino);
1130	trace_ext4_allocate_inode(inode, dir, mode);
1131	brelse(inode_bitmap_bh);
1132	return ret;
1133
1134fail_free_drop:
1135	dquot_free_inode(inode);
1136fail_drop:
1137	clear_nlink(inode);
1138	unlock_new_inode(inode);
1139out:
1140	dquot_drop(inode);
1141	inode->i_flags |= S_NOQUOTA;
1142	iput(inode);
1143	brelse(inode_bitmap_bh);
1144	return ERR_PTR(err);
1145}
1146
1147/* Verify that we are loading a valid orphan from disk */
1148struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1149{
1150	unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1151	ext4_group_t block_group;
1152	int bit;
1153	struct buffer_head *bitmap_bh;
1154	struct inode *inode = NULL;
1155	long err = -EIO;
1156
1157	/* Error cases - e2fsck has already cleaned up for us */
1158	if (ino > max_ino) {
1159		ext4_warning(sb, "bad orphan ino %lu!  e2fsck was run?", ino);
1160		err = -EFSCORRUPTED;
1161		goto error;
1162	}
1163
1164	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1165	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1166	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1167	if (IS_ERR(bitmap_bh)) {
1168		err = PTR_ERR(bitmap_bh);
1169		ext4_warning(sb, "inode bitmap error %ld for orphan %lu",
1170			     ino, err);
1171		goto error;
1172	}
1173
1174	/* Having the inode bit set should be a 100% indicator that this
1175	 * is a valid orphan (no e2fsck run on fs).  Orphans also include
1176	 * inodes that were being truncated, so we can't check i_nlink==0.
1177	 */
1178	if (!ext4_test_bit(bit, bitmap_bh->b_data))
1179		goto bad_orphan;
1180
1181	inode = ext4_iget(sb, ino);
1182	if (IS_ERR(inode))
1183		goto iget_failed;
 
 
 
 
1184
1185	/*
1186	 * If the orphans has i_nlinks > 0 then it should be able to be
1187	 * truncated, otherwise it won't be removed from the orphan list
1188	 * during processing and an infinite loop will result.
 
1189	 */
1190	if (inode->i_nlink && !ext4_can_truncate(inode))
 
1191		goto bad_orphan;
1192
1193	if (NEXT_ORPHAN(inode) > max_ino)
1194		goto bad_orphan;
1195	brelse(bitmap_bh);
1196	return inode;
1197
1198iget_failed:
1199	err = PTR_ERR(inode);
1200	inode = NULL;
1201bad_orphan:
1202	ext4_warning(sb, "bad orphan inode %lu!  e2fsck was run?", ino);
1203	printk(KERN_WARNING "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1204	       bit, (unsigned long long)bitmap_bh->b_blocknr,
1205	       ext4_test_bit(bit, bitmap_bh->b_data));
1206	printk(KERN_WARNING "inode=%p\n", inode);
1207	if (inode) {
1208		printk(KERN_WARNING "is_bad_inode(inode)=%d\n",
1209		       is_bad_inode(inode));
1210		printk(KERN_WARNING "NEXT_ORPHAN(inode)=%u\n",
1211		       NEXT_ORPHAN(inode));
1212		printk(KERN_WARNING "max_ino=%lu\n", max_ino);
1213		printk(KERN_WARNING "i_nlink=%u\n", inode->i_nlink);
1214		/* Avoid freeing blocks if we got a bad deleted inode */
1215		if (inode->i_nlink == 0)
1216			inode->i_blocks = 0;
1217		iput(inode);
1218	}
1219	brelse(bitmap_bh);
1220error:
1221	return ERR_PTR(err);
1222}
1223
1224unsigned long ext4_count_free_inodes(struct super_block *sb)
1225{
1226	unsigned long desc_count;
1227	struct ext4_group_desc *gdp;
1228	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1229#ifdef EXT4FS_DEBUG
1230	struct ext4_super_block *es;
1231	unsigned long bitmap_count, x;
1232	struct buffer_head *bitmap_bh = NULL;
1233
1234	es = EXT4_SB(sb)->s_es;
1235	desc_count = 0;
1236	bitmap_count = 0;
1237	gdp = NULL;
1238	for (i = 0; i < ngroups; i++) {
1239		gdp = ext4_get_group_desc(sb, i, NULL);
1240		if (!gdp)
1241			continue;
1242		desc_count += ext4_free_inodes_count(sb, gdp);
1243		brelse(bitmap_bh);
1244		bitmap_bh = ext4_read_inode_bitmap(sb, i);
1245		if (IS_ERR(bitmap_bh)) {
1246			bitmap_bh = NULL;
1247			continue;
1248		}
1249
1250		x = ext4_count_free(bitmap_bh->b_data,
1251				    EXT4_INODES_PER_GROUP(sb) / 8);
1252		printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1253			(unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1254		bitmap_count += x;
1255	}
1256	brelse(bitmap_bh);
1257	printk(KERN_DEBUG "ext4_count_free_inodes: "
1258	       "stored = %u, computed = %lu, %lu\n",
1259	       le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1260	return desc_count;
1261#else
1262	desc_count = 0;
1263	for (i = 0; i < ngroups; i++) {
1264		gdp = ext4_get_group_desc(sb, i, NULL);
1265		if (!gdp)
1266			continue;
1267		desc_count += ext4_free_inodes_count(sb, gdp);
1268		cond_resched();
1269	}
1270	return desc_count;
1271#endif
1272}
1273
1274/* Called at mount-time, super-block is locked */
1275unsigned long ext4_count_dirs(struct super_block * sb)
1276{
1277	unsigned long count = 0;
1278	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1279
1280	for (i = 0; i < ngroups; i++) {
1281		struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1282		if (!gdp)
1283			continue;
1284		count += ext4_used_dirs_count(sb, gdp);
1285	}
1286	return count;
1287}
1288
1289/*
1290 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1291 * inode table. Must be called without any spinlock held. The only place
1292 * where it is called from on active part of filesystem is ext4lazyinit
1293 * thread, so we do not need any special locks, however we have to prevent
1294 * inode allocation from the current group, so we take alloc_sem lock, to
1295 * block ext4_new_inode() until we are finished.
1296 */
1297int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1298				 int barrier)
1299{
1300	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1301	struct ext4_sb_info *sbi = EXT4_SB(sb);
1302	struct ext4_group_desc *gdp = NULL;
1303	struct buffer_head *group_desc_bh;
1304	handle_t *handle;
1305	ext4_fsblk_t blk;
1306	int num, ret = 0, used_blks = 0;
1307
1308	/* This should not happen, but just to be sure check this */
1309	if (sb->s_flags & MS_RDONLY) {
1310		ret = 1;
1311		goto out;
1312	}
1313
1314	gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1315	if (!gdp)
1316		goto out;
1317
1318	/*
1319	 * We do not need to lock this, because we are the only one
1320	 * handling this flag.
1321	 */
1322	if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1323		goto out;
1324
1325	handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1326	if (IS_ERR(handle)) {
1327		ret = PTR_ERR(handle);
1328		goto out;
1329	}
1330
1331	down_write(&grp->alloc_sem);
1332	/*
1333	 * If inode bitmap was already initialized there may be some
1334	 * used inodes so we need to skip blocks with used inodes in
1335	 * inode table.
1336	 */
1337	if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1338		used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1339			    ext4_itable_unused_count(sb, gdp)),
1340			    sbi->s_inodes_per_block);
1341
1342	if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
 
 
 
1343		ext4_error(sb, "Something is wrong with group %u: "
1344			   "used itable blocks: %d; "
1345			   "itable unused count: %u",
1346			   group, used_blks,
1347			   ext4_itable_unused_count(sb, gdp));
1348		ret = 1;
1349		goto err_out;
1350	}
1351
1352	blk = ext4_inode_table(sb, gdp) + used_blks;
1353	num = sbi->s_itb_per_group - used_blks;
1354
1355	BUFFER_TRACE(group_desc_bh, "get_write_access");
1356	ret = ext4_journal_get_write_access(handle,
1357					    group_desc_bh);
1358	if (ret)
1359		goto err_out;
1360
1361	/*
1362	 * Skip zeroout if the inode table is full. But we set the ZEROED
1363	 * flag anyway, because obviously, when it is full it does not need
1364	 * further zeroing.
1365	 */
1366	if (unlikely(num == 0))
1367		goto skip_zeroout;
1368
1369	ext4_debug("going to zero out inode table in group %d\n",
1370		   group);
1371	ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1372	if (ret < 0)
1373		goto err_out;
1374	if (barrier)
1375		blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1376
1377skip_zeroout:
1378	ext4_lock_group(sb, group);
1379	gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1380	ext4_group_desc_csum_set(sb, group, gdp);
1381	ext4_unlock_group(sb, group);
1382
1383	BUFFER_TRACE(group_desc_bh,
1384		     "call ext4_handle_dirty_metadata");
1385	ret = ext4_handle_dirty_metadata(handle, NULL,
1386					 group_desc_bh);
1387
1388err_out:
1389	up_write(&grp->alloc_sem);
1390	ext4_journal_stop(handle);
1391out:
1392	return ret;
1393}
v5.4
   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 = ext4_get_group_info(sb, block_group);
 
  86
  87	if (buffer_verified(bh))
  88		return 0;
  89	if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
  90		return -EFSCORRUPTED;
  91
  92	ext4_lock_group(sb, block_group);
  93	if (buffer_verified(bh))
  94		goto verified;
  95	blk = ext4_inode_bitmap(sb, desc);
  96	if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
  97					   EXT4_INODES_PER_GROUP(sb) / 8)) {
  98		ext4_unlock_group(sb, block_group);
  99		ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
 100			   "inode_bitmap = %llu", block_group, blk);
 101		ext4_mark_group_bitmap_corrupted(sb, block_group,
 102					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 
 
 
 
 
 
 103		return -EFSBADCRC;
 104	}
 105	set_buffer_verified(bh);
 106verified:
 107	ext4_unlock_group(sb, block_group);
 108	return 0;
 109}
 110
 111/*
 112 * Read the inode allocation bitmap for a given block_group, reading
 113 * into the specified slot in the superblock's bitmap cache.
 114 *
 115 * Return buffer_head of bitmap on success or NULL.
 116 */
 117static struct buffer_head *
 118ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 119{
 120	struct ext4_group_desc *desc;
 121	struct ext4_sb_info *sbi = EXT4_SB(sb);
 122	struct buffer_head *bh = NULL;
 123	ext4_fsblk_t bitmap_blk;
 124	int err;
 125
 126	desc = ext4_get_group_desc(sb, block_group, NULL);
 127	if (!desc)
 128		return ERR_PTR(-EFSCORRUPTED);
 129
 130	bitmap_blk = ext4_inode_bitmap(sb, desc);
 131	if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 132	    (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
 133		ext4_error(sb, "Invalid inode bitmap blk %llu in "
 134			   "block_group %u", bitmap_blk, block_group);
 135		ext4_mark_group_bitmap_corrupted(sb, block_group,
 136					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 137		return ERR_PTR(-EFSCORRUPTED);
 138	}
 139	bh = sb_getblk(sb, bitmap_blk);
 140	if (unlikely(!bh)) {
 141		ext4_warning(sb, "Cannot read inode bitmap - "
 142			     "block_group = %u, inode_bitmap = %llu",
 143			     block_group, bitmap_blk);
 144		return ERR_PTR(-ENOMEM);
 145	}
 146	if (bitmap_uptodate(bh))
 147		goto verify;
 148
 149	lock_buffer(bh);
 150	if (bitmap_uptodate(bh)) {
 151		unlock_buffer(bh);
 152		goto verify;
 153	}
 154
 155	ext4_lock_group(sb, block_group);
 156	if (ext4_has_group_desc_csum(sb) &&
 157	    (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
 158		if (block_group == 0) {
 159			ext4_unlock_group(sb, block_group);
 160			unlock_buffer(bh);
 161			ext4_error(sb, "Inode bitmap for bg 0 marked "
 162				   "uninitialized");
 163			err = -EFSCORRUPTED;
 164			goto out;
 165		}
 166		memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
 167		ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 168				     sb->s_blocksize * 8, bh->b_data);
 169		set_bitmap_uptodate(bh);
 170		set_buffer_uptodate(bh);
 171		set_buffer_verified(bh);
 172		ext4_unlock_group(sb, block_group);
 173		unlock_buffer(bh);
 
 
 
 
 
 174		return bh;
 175	}
 176	ext4_unlock_group(sb, block_group);
 177
 178	if (buffer_uptodate(bh)) {
 179		/*
 180		 * if not uninit if bh is uptodate,
 181		 * bitmap is also uptodate
 182		 */
 183		set_bitmap_uptodate(bh);
 184		unlock_buffer(bh);
 185		goto verify;
 186	}
 187	/*
 188	 * submit the buffer_head for reading
 189	 */
 190	trace_ext4_load_inode_bitmap(sb, block_group);
 191	bh->b_end_io = ext4_end_bitmap_read;
 192	get_bh(bh);
 193	submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
 194	wait_on_buffer(bh);
 195	if (!buffer_uptodate(bh)) {
 196		put_bh(bh);
 197		ext4_error(sb, "Cannot read inode bitmap - "
 198			   "block_group = %u, inode_bitmap = %llu",
 199			   block_group, bitmap_blk);
 200		ext4_mark_group_bitmap_corrupted(sb, block_group,
 201				EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 202		return ERR_PTR(-EIO);
 203	}
 204
 205verify:
 206	err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
 207	if (err)
 208		goto out;
 209	return bh;
 210out:
 211	put_bh(bh);
 212	return ERR_PTR(err);
 213}
 214
 215/*
 216 * NOTE! When we get the inode, we're the only people
 217 * that have access to it, and as such there are no
 218 * race conditions we have to worry about. The inode
 219 * is not on the hash-lists, and it cannot be reached
 220 * through the filesystem because the directory entry
 221 * has been deleted earlier.
 222 *
 223 * HOWEVER: we must make sure that we get no aliases,
 224 * which means that we have to call "clear_inode()"
 225 * _before_ we mark the inode not in use in the inode
 226 * bitmaps. Otherwise a newly created file might use
 227 * the same inode number (not actually the same pointer
 228 * though), and then we'd have two inodes sharing the
 229 * same inode number and space on the harddisk.
 230 */
 231void ext4_free_inode(handle_t *handle, struct inode *inode)
 232{
 233	struct super_block *sb = inode->i_sb;
 234	int is_directory;
 235	unsigned long ino;
 236	struct buffer_head *bitmap_bh = NULL;
 237	struct buffer_head *bh2;
 238	ext4_group_t block_group;
 239	unsigned long bit;
 240	struct ext4_group_desc *gdp;
 241	struct ext4_super_block *es;
 242	struct ext4_sb_info *sbi;
 243	int fatal = 0, err, count, cleared;
 244	struct ext4_group_info *grp;
 245
 246	if (!sb) {
 247		printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 248		       "nonexistent device\n", __func__, __LINE__);
 249		return;
 250	}
 251	if (atomic_read(&inode->i_count) > 1) {
 252		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 253			 __func__, __LINE__, inode->i_ino,
 254			 atomic_read(&inode->i_count));
 255		return;
 256	}
 257	if (inode->i_nlink) {
 258		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 259			 __func__, __LINE__, inode->i_ino, inode->i_nlink);
 260		return;
 261	}
 262	sbi = EXT4_SB(sb);
 263
 264	ino = inode->i_ino;
 265	ext4_debug("freeing inode %lu\n", ino);
 266	trace_ext4_free_inode(inode);
 267
 268	/*
 269	 * Note: we must free any quota before locking the superblock,
 270	 * as writing the quota to disk may need the lock as well.
 271	 */
 272	dquot_initialize(inode);
 
 273	dquot_free_inode(inode);
 274	dquot_drop(inode);
 275
 276	is_directory = S_ISDIR(inode->i_mode);
 277
 278	/* Do this BEFORE marking the inode not in use or returning an error */
 279	ext4_clear_inode(inode);
 280
 281	es = sbi->s_es;
 282	if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 283		ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 284		goto error_return;
 285	}
 286	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 287	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 288	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 289	/* Don't bother if the inode bitmap is corrupt. */
 290	grp = ext4_get_group_info(sb, block_group);
 291	if (IS_ERR(bitmap_bh)) {
 292		fatal = PTR_ERR(bitmap_bh);
 293		bitmap_bh = NULL;
 294		goto error_return;
 295	}
 296	if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
 297		fatal = -EFSCORRUPTED;
 298		goto error_return;
 299	}
 300
 301	BUFFER_TRACE(bitmap_bh, "get_write_access");
 302	fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 303	if (fatal)
 304		goto error_return;
 305
 306	fatal = -ESRCH;
 307	gdp = ext4_get_group_desc(sb, block_group, &bh2);
 308	if (gdp) {
 309		BUFFER_TRACE(bh2, "get_write_access");
 310		fatal = ext4_journal_get_write_access(handle, bh2);
 311	}
 312	ext4_lock_group(sb, block_group);
 313	cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 314	if (fatal || !cleared) {
 315		ext4_unlock_group(sb, block_group);
 316		goto out;
 317	}
 318
 319	count = ext4_free_inodes_count(sb, gdp) + 1;
 320	ext4_free_inodes_set(sb, gdp, count);
 321	if (is_directory) {
 322		count = ext4_used_dirs_count(sb, gdp) - 1;
 323		ext4_used_dirs_set(sb, gdp, count);
 324		percpu_counter_dec(&sbi->s_dirs_counter);
 325	}
 326	ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
 327				   EXT4_INODES_PER_GROUP(sb) / 8);
 328	ext4_group_desc_csum_set(sb, block_group, gdp);
 329	ext4_unlock_group(sb, block_group);
 330
 331	percpu_counter_inc(&sbi->s_freeinodes_counter);
 332	if (sbi->s_log_groups_per_flex) {
 333		ext4_group_t f = ext4_flex_group(sbi, block_group);
 334
 335		atomic_inc(&sbi->s_flex_groups[f].free_inodes);
 336		if (is_directory)
 337			atomic_dec(&sbi->s_flex_groups[f].used_dirs);
 338	}
 339	BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 340	fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 341out:
 342	if (cleared) {
 343		BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 344		err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 345		if (!fatal)
 346			fatal = err;
 347	} else {
 348		ext4_error(sb, "bit already cleared for inode %lu", ino);
 349		ext4_mark_group_bitmap_corrupted(sb, block_group,
 350					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 
 
 
 
 
 351	}
 352
 353error_return:
 354	brelse(bitmap_bh);
 355	ext4_std_error(sb, fatal);
 356}
 357
 358struct orlov_stats {
 359	__u64 free_clusters;
 360	__u32 free_inodes;
 361	__u32 used_dirs;
 362};
 363
 364/*
 365 * Helper function for Orlov's allocator; returns critical information
 366 * for a particular block group or flex_bg.  If flex_size is 1, then g
 367 * is a block group number; otherwise it is flex_bg number.
 368 */
 369static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 370			    int flex_size, struct orlov_stats *stats)
 371{
 372	struct ext4_group_desc *desc;
 373	struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
 374
 375	if (flex_size > 1) {
 376		stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
 377		stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
 378		stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
 379		return;
 380	}
 381
 382	desc = ext4_get_group_desc(sb, g, NULL);
 383	if (desc) {
 384		stats->free_inodes = ext4_free_inodes_count(sb, desc);
 385		stats->free_clusters = ext4_free_group_clusters(sb, desc);
 386		stats->used_dirs = ext4_used_dirs_count(sb, desc);
 387	} else {
 388		stats->free_inodes = 0;
 389		stats->free_clusters = 0;
 390		stats->used_dirs = 0;
 391	}
 392}
 393
 394/*
 395 * Orlov's allocator for directories.
 396 *
 397 * We always try to spread first-level directories.
 398 *
 399 * If there are blockgroups with both free inodes and free blocks counts
 400 * not worse than average we return one with smallest directory count.
 401 * Otherwise we simply return a random group.
 402 *
 403 * For the rest rules look so:
 404 *
 405 * It's OK to put directory into a group unless
 406 * it has too many directories already (max_dirs) or
 407 * it has too few free inodes left (min_inodes) or
 408 * it has too few free blocks left (min_blocks) or
 409 * Parent's group is preferred, if it doesn't satisfy these
 410 * conditions we search cyclically through the rest. If none
 411 * of the groups look good we just look for a group with more
 412 * free inodes than average (starting at parent's group).
 413 */
 414
 415static int find_group_orlov(struct super_block *sb, struct inode *parent,
 416			    ext4_group_t *group, umode_t mode,
 417			    const struct qstr *qstr)
 418{
 419	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 420	struct ext4_sb_info *sbi = EXT4_SB(sb);
 421	ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 422	int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 423	unsigned int freei, avefreei, grp_free;
 424	ext4_fsblk_t freeb, avefreec;
 425	unsigned int ndirs;
 426	int max_dirs, min_inodes;
 427	ext4_grpblk_t min_clusters;
 428	ext4_group_t i, grp, g, ngroups;
 429	struct ext4_group_desc *desc;
 430	struct orlov_stats stats;
 431	int flex_size = ext4_flex_bg_size(sbi);
 432	struct dx_hash_info hinfo;
 433
 434	ngroups = real_ngroups;
 435	if (flex_size > 1) {
 436		ngroups = (real_ngroups + flex_size - 1) >>
 437			sbi->s_log_groups_per_flex;
 438		parent_group >>= sbi->s_log_groups_per_flex;
 439	}
 440
 441	freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 442	avefreei = freei / ngroups;
 443	freeb = EXT4_C2B(sbi,
 444		percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 445	avefreec = freeb;
 446	do_div(avefreec, ngroups);
 447	ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 448
 449	if (S_ISDIR(mode) &&
 450	    ((parent == d_inode(sb->s_root)) ||
 451	     (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 452		int best_ndir = inodes_per_group;
 453		int ret = -1;
 454
 455		if (qstr) {
 456			hinfo.hash_version = DX_HASH_HALF_MD4;
 457			hinfo.seed = sbi->s_hash_seed;
 458			ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
 459			grp = hinfo.hash;
 460		} else
 461			grp = prandom_u32();
 462		parent_group = (unsigned)grp % ngroups;
 463		for (i = 0; i < ngroups; i++) {
 464			g = (parent_group + i) % ngroups;
 465			get_orlov_stats(sb, g, flex_size, &stats);
 466			if (!stats.free_inodes)
 467				continue;
 468			if (stats.used_dirs >= best_ndir)
 469				continue;
 470			if (stats.free_inodes < avefreei)
 471				continue;
 472			if (stats.free_clusters < avefreec)
 473				continue;
 474			grp = g;
 475			ret = 0;
 476			best_ndir = stats.used_dirs;
 477		}
 478		if (ret)
 479			goto fallback;
 480	found_flex_bg:
 481		if (flex_size == 1) {
 482			*group = grp;
 483			return 0;
 484		}
 485
 486		/*
 487		 * We pack inodes at the beginning of the flexgroup's
 488		 * inode tables.  Block allocation decisions will do
 489		 * something similar, although regular files will
 490		 * start at 2nd block group of the flexgroup.  See
 491		 * ext4_ext_find_goal() and ext4_find_near().
 492		 */
 493		grp *= flex_size;
 494		for (i = 0; i < flex_size; i++) {
 495			if (grp+i >= real_ngroups)
 496				break;
 497			desc = ext4_get_group_desc(sb, grp+i, NULL);
 498			if (desc && ext4_free_inodes_count(sb, desc)) {
 499				*group = grp+i;
 500				return 0;
 501			}
 502		}
 503		goto fallback;
 504	}
 505
 506	max_dirs = ndirs / ngroups + inodes_per_group / 16;
 507	min_inodes = avefreei - inodes_per_group*flex_size / 4;
 508	if (min_inodes < 1)
 509		min_inodes = 1;
 510	min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 511
 512	/*
 513	 * Start looking in the flex group where we last allocated an
 514	 * inode for this parent directory
 515	 */
 516	if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 517		parent_group = EXT4_I(parent)->i_last_alloc_group;
 518		if (flex_size > 1)
 519			parent_group >>= sbi->s_log_groups_per_flex;
 520	}
 521
 522	for (i = 0; i < ngroups; i++) {
 523		grp = (parent_group + i) % ngroups;
 524		get_orlov_stats(sb, grp, flex_size, &stats);
 525		if (stats.used_dirs >= max_dirs)
 526			continue;
 527		if (stats.free_inodes < min_inodes)
 528			continue;
 529		if (stats.free_clusters < min_clusters)
 530			continue;
 531		goto found_flex_bg;
 532	}
 533
 534fallback:
 535	ngroups = real_ngroups;
 536	avefreei = freei / ngroups;
 537fallback_retry:
 538	parent_group = EXT4_I(parent)->i_block_group;
 539	for (i = 0; i < ngroups; i++) {
 540		grp = (parent_group + i) % ngroups;
 541		desc = ext4_get_group_desc(sb, grp, NULL);
 542		if (desc) {
 543			grp_free = ext4_free_inodes_count(sb, desc);
 544			if (grp_free && grp_free >= avefreei) {
 545				*group = grp;
 546				return 0;
 547			}
 548		}
 549	}
 550
 551	if (avefreei) {
 552		/*
 553		 * The free-inodes counter is approximate, and for really small
 554		 * filesystems the above test can fail to find any blockgroups
 555		 */
 556		avefreei = 0;
 557		goto fallback_retry;
 558	}
 559
 560	return -1;
 561}
 562
 563static int find_group_other(struct super_block *sb, struct inode *parent,
 564			    ext4_group_t *group, umode_t mode)
 565{
 566	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 567	ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 568	struct ext4_group_desc *desc;
 569	int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 570
 571	/*
 572	 * Try to place the inode is the same flex group as its
 573	 * parent.  If we can't find space, use the Orlov algorithm to
 574	 * find another flex group, and store that information in the
 575	 * parent directory's inode information so that use that flex
 576	 * group for future allocations.
 577	 */
 578	if (flex_size > 1) {
 579		int retry = 0;
 580
 581	try_again:
 582		parent_group &= ~(flex_size-1);
 583		last = parent_group + flex_size;
 584		if (last > ngroups)
 585			last = ngroups;
 586		for  (i = parent_group; i < last; i++) {
 587			desc = ext4_get_group_desc(sb, i, NULL);
 588			if (desc && ext4_free_inodes_count(sb, desc)) {
 589				*group = i;
 590				return 0;
 591			}
 592		}
 593		if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 594			retry = 1;
 595			parent_group = EXT4_I(parent)->i_last_alloc_group;
 596			goto try_again;
 597		}
 598		/*
 599		 * If this didn't work, use the Orlov search algorithm
 600		 * to find a new flex group; we pass in the mode to
 601		 * avoid the topdir algorithms.
 602		 */
 603		*group = parent_group + flex_size;
 604		if (*group > ngroups)
 605			*group = 0;
 606		return find_group_orlov(sb, parent, group, mode, NULL);
 607	}
 608
 609	/*
 610	 * Try to place the inode in its parent directory
 611	 */
 612	*group = parent_group;
 613	desc = ext4_get_group_desc(sb, *group, NULL);
 614	if (desc && ext4_free_inodes_count(sb, desc) &&
 615	    ext4_free_group_clusters(sb, desc))
 616		return 0;
 617
 618	/*
 619	 * We're going to place this inode in a different blockgroup from its
 620	 * parent.  We want to cause files in a common directory to all land in
 621	 * the same blockgroup.  But we want files which are in a different
 622	 * directory which shares a blockgroup with our parent to land in a
 623	 * different blockgroup.
 624	 *
 625	 * So add our directory's i_ino into the starting point for the hash.
 626	 */
 627	*group = (*group + parent->i_ino) % ngroups;
 628
 629	/*
 630	 * Use a quadratic hash to find a group with a free inode and some free
 631	 * blocks.
 632	 */
 633	for (i = 1; i < ngroups; i <<= 1) {
 634		*group += i;
 635		if (*group >= ngroups)
 636			*group -= ngroups;
 637		desc = ext4_get_group_desc(sb, *group, NULL);
 638		if (desc && ext4_free_inodes_count(sb, desc) &&
 639		    ext4_free_group_clusters(sb, desc))
 640			return 0;
 641	}
 642
 643	/*
 644	 * That failed: try linear search for a free inode, even if that group
 645	 * has no free blocks.
 646	 */
 647	*group = parent_group;
 648	for (i = 0; i < ngroups; i++) {
 649		if (++*group >= ngroups)
 650			*group = 0;
 651		desc = ext4_get_group_desc(sb, *group, NULL);
 652		if (desc && ext4_free_inodes_count(sb, desc))
 653			return 0;
 654	}
 655
 656	return -1;
 657}
 658
 659/*
 660 * In no journal mode, if an inode has recently been deleted, we want
 661 * to avoid reusing it until we're reasonably sure the inode table
 662 * block has been written back to disk.  (Yes, these values are
 663 * somewhat arbitrary...)
 664 */
 665#define RECENTCY_MIN	5
 666#define RECENTCY_DIRTY	300
 667
 668static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 669{
 670	struct ext4_group_desc	*gdp;
 671	struct ext4_inode	*raw_inode;
 672	struct buffer_head	*bh;
 673	int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 674	int offset, ret = 0;
 675	int recentcy = RECENTCY_MIN;
 676	u32 dtime, now;
 677
 678	gdp = ext4_get_group_desc(sb, group, NULL);
 679	if (unlikely(!gdp))
 680		return 0;
 681
 682	bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
 683		       (ino / inodes_per_block));
 684	if (!bh || !buffer_uptodate(bh))
 685		/*
 686		 * If the block is not in the buffer cache, then it
 687		 * must have been written out.
 688		 */
 689		goto out;
 690
 691	offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 692	raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 693
 694	/* i_dtime is only 32 bits on disk, but we only care about relative
 695	 * times in the range of a few minutes (i.e. long enough to sync a
 696	 * recently-deleted inode to disk), so using the low 32 bits of the
 697	 * clock (a 68 year range) is enough, see time_before32() */
 698	dtime = le32_to_cpu(raw_inode->i_dtime);
 699	now = ktime_get_real_seconds();
 700	if (buffer_dirty(bh))
 701		recentcy += RECENTCY_DIRTY;
 702
 703	if (dtime && time_before32(dtime, now) &&
 704	    time_before32(now, dtime + recentcy))
 705		ret = 1;
 706out:
 707	brelse(bh);
 708	return ret;
 709}
 710
 711static int find_inode_bit(struct super_block *sb, ext4_group_t group,
 712			  struct buffer_head *bitmap, unsigned long *ino)
 713{
 714next:
 715	*ino = ext4_find_next_zero_bit((unsigned long *)
 716				       bitmap->b_data,
 717				       EXT4_INODES_PER_GROUP(sb), *ino);
 718	if (*ino >= EXT4_INODES_PER_GROUP(sb))
 719		return 0;
 720
 721	if ((EXT4_SB(sb)->s_journal == NULL) &&
 722	    recently_deleted(sb, group, *ino)) {
 723		*ino = *ino + 1;
 724		if (*ino < EXT4_INODES_PER_GROUP(sb))
 725			goto next;
 726		return 0;
 727	}
 728
 729	return 1;
 730}
 731
 732/*
 733 * There are two policies for allocating an inode.  If the new inode is
 734 * a directory, then a forward search is made for a block group with both
 735 * free space and a low directory-to-inode ratio; if that fails, then of
 736 * the groups with above-average free space, that group with the fewest
 737 * directories already is chosen.
 738 *
 739 * For other inodes, search forward from the parent directory's block
 740 * group to find a free inode.
 741 */
 742struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 743			       umode_t mode, const struct qstr *qstr,
 744			       __u32 goal, uid_t *owner, __u32 i_flags,
 745			       int handle_type, unsigned int line_no,
 746			       int nblocks)
 747{
 748	struct super_block *sb;
 749	struct buffer_head *inode_bitmap_bh = NULL;
 750	struct buffer_head *group_desc_bh;
 751	ext4_group_t ngroups, group = 0;
 752	unsigned long ino = 0;
 753	struct inode *inode;
 754	struct ext4_group_desc *gdp = NULL;
 755	struct ext4_inode_info *ei;
 756	struct ext4_sb_info *sbi;
 757	int ret2, err;
 758	struct inode *ret;
 759	ext4_group_t i;
 760	ext4_group_t flex_group;
 761	struct ext4_group_info *grp;
 762	int encrypt = 0;
 763
 764	/* Cannot create files in a deleted directory */
 765	if (!dir || !dir->i_nlink)
 766		return ERR_PTR(-EPERM);
 767
 768	sb = dir->i_sb;
 769	sbi = EXT4_SB(sb);
 770
 771	if (unlikely(ext4_forced_shutdown(sbi)))
 772		return ERR_PTR(-EIO);
 773
 774	if ((IS_ENCRYPTED(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
 775	    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
 776	    !(i_flags & EXT4_EA_INODE_FL)) {
 777		err = fscrypt_get_encryption_info(dir);
 778		if (err)
 779			return ERR_PTR(err);
 780		if (!fscrypt_has_encryption_key(dir))
 781			return ERR_PTR(-ENOKEY);
 
 
 782		encrypt = 1;
 783	}
 784
 785	if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
 786#ifdef CONFIG_EXT4_FS_POSIX_ACL
 787		struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
 788
 789		if (IS_ERR(p))
 790			return ERR_CAST(p);
 791		if (p) {
 792			int acl_size = p->a_count * sizeof(ext4_acl_entry);
 793
 794			nblocks += (S_ISDIR(mode) ? 2 : 1) *
 795				__ext4_xattr_set_credits(sb, NULL /* inode */,
 796					NULL /* block_bh */, acl_size,
 797					true /* is_create */);
 798			posix_acl_release(p);
 799		}
 800#endif
 801
 802#ifdef CONFIG_SECURITY
 803		{
 804			int num_security_xattrs = 1;
 805
 806#ifdef CONFIG_INTEGRITY
 807			num_security_xattrs++;
 808#endif
 809			/*
 810			 * We assume that security xattrs are never
 811			 * more than 1k.  In practice they are under
 812			 * 128 bytes.
 813			 */
 814			nblocks += num_security_xattrs *
 815				__ext4_xattr_set_credits(sb, NULL /* inode */,
 816					NULL /* block_bh */, 1024,
 817					true /* is_create */);
 818		}
 819#endif
 820		if (encrypt)
 821			nblocks += __ext4_xattr_set_credits(sb,
 822					NULL /* inode */, NULL /* block_bh */,
 823					FSCRYPT_SET_CONTEXT_MAX_SIZE,
 824					true /* is_create */);
 825	}
 826
 827	ngroups = ext4_get_groups_count(sb);
 828	trace_ext4_request_inode(dir, mode);
 829	inode = new_inode(sb);
 830	if (!inode)
 831		return ERR_PTR(-ENOMEM);
 832	ei = EXT4_I(inode);
 
 833
 834	/*
 835	 * Initialize owners and quota early so that we don't have to account
 836	 * for quota initialization worst case in standard inode creating
 837	 * transaction
 838	 */
 839	if (owner) {
 840		inode->i_mode = mode;
 841		i_uid_write(inode, owner[0]);
 842		i_gid_write(inode, owner[1]);
 843	} else if (test_opt(sb, GRPID)) {
 844		inode->i_mode = mode;
 845		inode->i_uid = current_fsuid();
 846		inode->i_gid = dir->i_gid;
 847	} else
 848		inode_init_owner(inode, dir, mode);
 849
 850	if (ext4_has_feature_project(sb) &&
 851	    ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
 852		ei->i_projid = EXT4_I(dir)->i_projid;
 853	else
 854		ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
 855
 856	err = dquot_initialize(inode);
 857	if (err)
 858		goto out;
 859
 860	if (!goal)
 861		goal = sbi->s_inode_goal;
 862
 863	if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 864		group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 865		ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 866		ret2 = 0;
 867		goto got_group;
 868	}
 869
 870	if (S_ISDIR(mode))
 871		ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 872	else
 873		ret2 = find_group_other(sb, dir, &group, mode);
 874
 875got_group:
 876	EXT4_I(dir)->i_last_alloc_group = group;
 877	err = -ENOSPC;
 878	if (ret2 == -1)
 879		goto out;
 880
 881	/*
 882	 * Normally we will only go through one pass of this loop,
 883	 * unless we get unlucky and it turns out the group we selected
 884	 * had its last inode grabbed by someone else.
 885	 */
 886	for (i = 0; i < ngroups; i++, ino = 0) {
 887		err = -EIO;
 888
 889		gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 890		if (!gdp)
 891			goto out;
 892
 893		/*
 894		 * Check free inodes count before loading bitmap.
 895		 */
 896		if (ext4_free_inodes_count(sb, gdp) == 0)
 897			goto next_group;
 
 
 
 898
 899		grp = ext4_get_group_info(sb, group);
 900		/* Skip groups with already-known suspicious inode tables */
 901		if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
 902			goto next_group;
 
 
 
 903
 904		brelse(inode_bitmap_bh);
 905		inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 906		/* Skip groups with suspicious inode tables */
 907		if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
 908		    IS_ERR(inode_bitmap_bh)) {
 909			inode_bitmap_bh = NULL;
 910			goto next_group;
 
 
 911		}
 912
 913repeat_in_this_group:
 914		ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 915		if (!ret2)
 
 
 916			goto next_group;
 917
 918		if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
 919			ext4_error(sb, "reserved inode found cleared - "
 920				   "inode=%lu", ino + 1);
 921			ext4_mark_group_bitmap_corrupted(sb, group,
 922					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 923			goto next_group;
 
 
 
 924		}
 925
 926		if (!handle) {
 927			BUG_ON(nblocks <= 0);
 928			handle = __ext4_journal_start_sb(dir->i_sb, line_no,
 929							 handle_type, nblocks,
 930							 0);
 931			if (IS_ERR(handle)) {
 932				err = PTR_ERR(handle);
 933				ext4_std_error(sb, err);
 934				goto out;
 935			}
 936		}
 937		BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 938		err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
 939		if (err) {
 940			ext4_std_error(sb, err);
 941			goto out;
 942		}
 943		ext4_lock_group(sb, group);
 944		ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
 945		if (ret2) {
 946			/* Someone already took the bit. Repeat the search
 947			 * with lock held.
 948			 */
 949			ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 950			if (ret2) {
 951				ext4_set_bit(ino, inode_bitmap_bh->b_data);
 952				ret2 = 0;
 953			} else {
 954				ret2 = 1; /* we didn't grab the inode */
 955			}
 956		}
 957		ext4_unlock_group(sb, group);
 958		ino++;		/* the inode bitmap is zero-based */
 959		if (!ret2)
 960			goto got; /* we grabbed the inode! */
 961
 962		if (ino < EXT4_INODES_PER_GROUP(sb))
 963			goto repeat_in_this_group;
 964next_group:
 965		if (++group == ngroups)
 966			group = 0;
 967	}
 968	err = -ENOSPC;
 969	goto out;
 970
 971got:
 972	BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 973	err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
 974	if (err) {
 975		ext4_std_error(sb, err);
 976		goto out;
 977	}
 978
 979	BUFFER_TRACE(group_desc_bh, "get_write_access");
 980	err = ext4_journal_get_write_access(handle, group_desc_bh);
 981	if (err) {
 982		ext4_std_error(sb, err);
 983		goto out;
 984	}
 985
 986	/* We may have to initialize the block bitmap if it isn't already */
 987	if (ext4_has_group_desc_csum(sb) &&
 988	    gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 989		struct buffer_head *block_bitmap_bh;
 990
 991		block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 992		if (IS_ERR(block_bitmap_bh)) {
 993			err = PTR_ERR(block_bitmap_bh);
 994			goto out;
 995		}
 996		BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
 997		err = ext4_journal_get_write_access(handle, block_bitmap_bh);
 998		if (err) {
 999			brelse(block_bitmap_bh);
1000			ext4_std_error(sb, err);
1001			goto out;
1002		}
1003
1004		BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1005		err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1006
1007		/* recheck and clear flag under lock if we still need to */
1008		ext4_lock_group(sb, group);
1009		if (ext4_has_group_desc_csum(sb) &&
1010		    (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1011			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1012			ext4_free_group_clusters_set(sb, gdp,
1013				ext4_free_clusters_after_init(sb, group, gdp));
1014			ext4_block_bitmap_csum_set(sb, group, gdp,
1015						   block_bitmap_bh);
1016			ext4_group_desc_csum_set(sb, group, gdp);
1017		}
1018		ext4_unlock_group(sb, group);
1019		brelse(block_bitmap_bh);
1020
1021		if (err) {
1022			ext4_std_error(sb, err);
1023			goto out;
1024		}
1025	}
1026
1027	/* Update the relevant bg descriptor fields */
1028	if (ext4_has_group_desc_csum(sb)) {
1029		int free;
1030		struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1031
1032		down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1033		ext4_lock_group(sb, group); /* while we modify the bg desc */
1034		free = EXT4_INODES_PER_GROUP(sb) -
1035			ext4_itable_unused_count(sb, gdp);
1036		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1037			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1038			free = 0;
1039		}
1040		/*
1041		 * Check the relative inode number against the last used
1042		 * relative inode number in this group. if it is greater
1043		 * we need to update the bg_itable_unused count
1044		 */
1045		if (ino > free)
1046			ext4_itable_unused_set(sb, gdp,
1047					(EXT4_INODES_PER_GROUP(sb) - ino));
1048		up_read(&grp->alloc_sem);
1049	} else {
1050		ext4_lock_group(sb, group);
1051	}
1052
1053	ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1054	if (S_ISDIR(mode)) {
1055		ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1056		if (sbi->s_log_groups_per_flex) {
1057			ext4_group_t f = ext4_flex_group(sbi, group);
1058
1059			atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1060		}
1061	}
1062	if (ext4_has_group_desc_csum(sb)) {
1063		ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1064					   EXT4_INODES_PER_GROUP(sb) / 8);
1065		ext4_group_desc_csum_set(sb, group, gdp);
1066	}
1067	ext4_unlock_group(sb, group);
1068
1069	BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1070	err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1071	if (err) {
1072		ext4_std_error(sb, err);
1073		goto out;
1074	}
1075
1076	percpu_counter_dec(&sbi->s_freeinodes_counter);
1077	if (S_ISDIR(mode))
1078		percpu_counter_inc(&sbi->s_dirs_counter);
1079
1080	if (sbi->s_log_groups_per_flex) {
1081		flex_group = ext4_flex_group(sbi, group);
1082		atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1083	}
1084
1085	inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1086	/* This is the optimal IO size (for stat), not the fs block size */
1087	inode->i_blocks = 0;
1088	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1089	ei->i_crtime = inode->i_mtime;
1090
1091	memset(ei->i_data, 0, sizeof(ei->i_data));
1092	ei->i_dir_start_lookup = 0;
1093	ei->i_disksize = 0;
1094
1095	/* Don't inherit extent flag from directory, amongst others. */
1096	ei->i_flags =
1097		ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1098	ei->i_flags |= i_flags;
1099	ei->i_file_acl = 0;
1100	ei->i_dtime = 0;
1101	ei->i_block_group = group;
1102	ei->i_last_alloc_group = ~0;
1103
1104	ext4_set_inode_flags(inode);
1105	if (IS_DIRSYNC(inode))
1106		ext4_handle_sync(handle);
1107	if (insert_inode_locked(inode) < 0) {
1108		/*
1109		 * Likely a bitmap corruption causing inode to be allocated
1110		 * twice.
1111		 */
1112		err = -EIO;
1113		ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1114			   inode->i_ino);
1115		ext4_mark_group_bitmap_corrupted(sb, group,
1116					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1117		goto out;
1118	}
1119	inode->i_generation = prandom_u32();
 
 
1120
1121	/* Precompute checksum seed for inode metadata */
1122	if (ext4_has_metadata_csum(sb)) {
1123		__u32 csum;
1124		__le32 inum = cpu_to_le32(inode->i_ino);
1125		__le32 gen = cpu_to_le32(inode->i_generation);
1126		csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1127				   sizeof(inum));
1128		ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1129					      sizeof(gen));
1130	}
1131
1132	ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1133	ext4_set_inode_state(inode, EXT4_STATE_NEW);
1134
1135	ei->i_extra_isize = sbi->s_want_extra_isize;
1136	ei->i_inline_off = 0;
1137	if (ext4_has_feature_inline_data(sb))
1138		ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1139	ret = inode;
1140	err = dquot_alloc_inode(inode);
1141	if (err)
1142		goto fail_drop;
1143
1144	/*
1145	 * Since the encryption xattr will always be unique, create it first so
1146	 * that it's less likely to end up in an external xattr block and
1147	 * prevent its deduplication.
1148	 */
1149	if (encrypt) {
1150		err = fscrypt_inherit_context(dir, inode, handle, true);
1151		if (err)
1152			goto fail_free_drop;
1153	}
1154
1155	if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1156		err = ext4_init_acl(handle, inode, dir);
1157		if (err)
1158			goto fail_free_drop;
1159
1160		err = ext4_init_security(handle, inode, dir, qstr);
1161		if (err)
1162			goto fail_free_drop;
1163	}
1164
1165	if (ext4_has_feature_extents(sb)) {
1166		/* set extent flag only for directory, file and normal symlink*/
1167		if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1168			ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1169			ext4_ext_tree_init(handle, inode);
1170		}
1171	}
1172
1173	if (ext4_handle_valid(handle)) {
1174		ei->i_sync_tid = handle->h_transaction->t_tid;
1175		ei->i_datasync_tid = handle->h_transaction->t_tid;
1176	}
1177
 
 
 
 
 
 
1178	err = ext4_mark_inode_dirty(handle, inode);
1179	if (err) {
1180		ext4_std_error(sb, err);
1181		goto fail_free_drop;
1182	}
1183
1184	ext4_debug("allocating inode %lu\n", inode->i_ino);
1185	trace_ext4_allocate_inode(inode, dir, mode);
1186	brelse(inode_bitmap_bh);
1187	return ret;
1188
1189fail_free_drop:
1190	dquot_free_inode(inode);
1191fail_drop:
1192	clear_nlink(inode);
1193	unlock_new_inode(inode);
1194out:
1195	dquot_drop(inode);
1196	inode->i_flags |= S_NOQUOTA;
1197	iput(inode);
1198	brelse(inode_bitmap_bh);
1199	return ERR_PTR(err);
1200}
1201
1202/* Verify that we are loading a valid orphan from disk */
1203struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1204{
1205	unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1206	ext4_group_t block_group;
1207	int bit;
1208	struct buffer_head *bitmap_bh = NULL;
1209	struct inode *inode = NULL;
1210	int err = -EFSCORRUPTED;
1211
1212	if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1213		goto bad_orphan;
 
 
 
 
1214
1215	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1216	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1217	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1218	if (IS_ERR(bitmap_bh))
1219		return ERR_CAST(bitmap_bh);
 
 
 
 
1220
1221	/* Having the inode bit set should be a 100% indicator that this
1222	 * is a valid orphan (no e2fsck run on fs).  Orphans also include
1223	 * inodes that were being truncated, so we can't check i_nlink==0.
1224	 */
1225	if (!ext4_test_bit(bit, bitmap_bh->b_data))
1226		goto bad_orphan;
1227
1228	inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1229	if (IS_ERR(inode)) {
1230		err = PTR_ERR(inode);
1231		ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1232			   ino, err);
1233		return inode;
1234	}
1235
1236	/*
1237	 * If the orphans has i_nlinks > 0 then it should be able to
1238	 * be truncated, otherwise it won't be removed from the orphan
1239	 * list during processing and an infinite loop will result.
1240	 * Similarly, it must not be a bad inode.
1241	 */
1242	if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1243	    is_bad_inode(inode))
1244		goto bad_orphan;
1245
1246	if (NEXT_ORPHAN(inode) > max_ino)
1247		goto bad_orphan;
1248	brelse(bitmap_bh);
1249	return inode;
1250
 
 
 
1251bad_orphan:
1252	ext4_error(sb, "bad orphan inode %lu", ino);
1253	if (bitmap_bh)
1254		printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1255		       bit, (unsigned long long)bitmap_bh->b_blocknr,
1256		       ext4_test_bit(bit, bitmap_bh->b_data));
1257	if (inode) {
1258		printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1259		       is_bad_inode(inode));
1260		printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1261		       NEXT_ORPHAN(inode));
1262		printk(KERN_ERR "max_ino=%lu\n", max_ino);
1263		printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1264		/* Avoid freeing blocks if we got a bad deleted inode */
1265		if (inode->i_nlink == 0)
1266			inode->i_blocks = 0;
1267		iput(inode);
1268	}
1269	brelse(bitmap_bh);
 
1270	return ERR_PTR(err);
1271}
1272
1273unsigned long ext4_count_free_inodes(struct super_block *sb)
1274{
1275	unsigned long desc_count;
1276	struct ext4_group_desc *gdp;
1277	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1278#ifdef EXT4FS_DEBUG
1279	struct ext4_super_block *es;
1280	unsigned long bitmap_count, x;
1281	struct buffer_head *bitmap_bh = NULL;
1282
1283	es = EXT4_SB(sb)->s_es;
1284	desc_count = 0;
1285	bitmap_count = 0;
1286	gdp = NULL;
1287	for (i = 0; i < ngroups; i++) {
1288		gdp = ext4_get_group_desc(sb, i, NULL);
1289		if (!gdp)
1290			continue;
1291		desc_count += ext4_free_inodes_count(sb, gdp);
1292		brelse(bitmap_bh);
1293		bitmap_bh = ext4_read_inode_bitmap(sb, i);
1294		if (IS_ERR(bitmap_bh)) {
1295			bitmap_bh = NULL;
1296			continue;
1297		}
1298
1299		x = ext4_count_free(bitmap_bh->b_data,
1300				    EXT4_INODES_PER_GROUP(sb) / 8);
1301		printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1302			(unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1303		bitmap_count += x;
1304	}
1305	brelse(bitmap_bh);
1306	printk(KERN_DEBUG "ext4_count_free_inodes: "
1307	       "stored = %u, computed = %lu, %lu\n",
1308	       le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1309	return desc_count;
1310#else
1311	desc_count = 0;
1312	for (i = 0; i < ngroups; i++) {
1313		gdp = ext4_get_group_desc(sb, i, NULL);
1314		if (!gdp)
1315			continue;
1316		desc_count += ext4_free_inodes_count(sb, gdp);
1317		cond_resched();
1318	}
1319	return desc_count;
1320#endif
1321}
1322
1323/* Called at mount-time, super-block is locked */
1324unsigned long ext4_count_dirs(struct super_block * sb)
1325{
1326	unsigned long count = 0;
1327	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1328
1329	for (i = 0; i < ngroups; i++) {
1330		struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1331		if (!gdp)
1332			continue;
1333		count += ext4_used_dirs_count(sb, gdp);
1334	}
1335	return count;
1336}
1337
1338/*
1339 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1340 * inode table. Must be called without any spinlock held. The only place
1341 * where it is called from on active part of filesystem is ext4lazyinit
1342 * thread, so we do not need any special locks, however we have to prevent
1343 * inode allocation from the current group, so we take alloc_sem lock, to
1344 * block ext4_new_inode() until we are finished.
1345 */
1346int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1347				 int barrier)
1348{
1349	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1350	struct ext4_sb_info *sbi = EXT4_SB(sb);
1351	struct ext4_group_desc *gdp = NULL;
1352	struct buffer_head *group_desc_bh;
1353	handle_t *handle;
1354	ext4_fsblk_t blk;
1355	int num, ret = 0, used_blks = 0;
1356
1357	/* This should not happen, but just to be sure check this */
1358	if (sb_rdonly(sb)) {
1359		ret = 1;
1360		goto out;
1361	}
1362
1363	gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1364	if (!gdp)
1365		goto out;
1366
1367	/*
1368	 * We do not need to lock this, because we are the only one
1369	 * handling this flag.
1370	 */
1371	if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1372		goto out;
1373
1374	handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1375	if (IS_ERR(handle)) {
1376		ret = PTR_ERR(handle);
1377		goto out;
1378	}
1379
1380	down_write(&grp->alloc_sem);
1381	/*
1382	 * If inode bitmap was already initialized there may be some
1383	 * used inodes so we need to skip blocks with used inodes in
1384	 * inode table.
1385	 */
1386	if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1387		used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1388			    ext4_itable_unused_count(sb, gdp)),
1389			    sbi->s_inodes_per_block);
1390
1391	if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1392	    ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1393			       ext4_itable_unused_count(sb, gdp)) <
1394			      EXT4_FIRST_INO(sb)))) {
1395		ext4_error(sb, "Something is wrong with group %u: "
1396			   "used itable blocks: %d; "
1397			   "itable unused count: %u",
1398			   group, used_blks,
1399			   ext4_itable_unused_count(sb, gdp));
1400		ret = 1;
1401		goto err_out;
1402	}
1403
1404	blk = ext4_inode_table(sb, gdp) + used_blks;
1405	num = sbi->s_itb_per_group - used_blks;
1406
1407	BUFFER_TRACE(group_desc_bh, "get_write_access");
1408	ret = ext4_journal_get_write_access(handle,
1409					    group_desc_bh);
1410	if (ret)
1411		goto err_out;
1412
1413	/*
1414	 * Skip zeroout if the inode table is full. But we set the ZEROED
1415	 * flag anyway, because obviously, when it is full it does not need
1416	 * further zeroing.
1417	 */
1418	if (unlikely(num == 0))
1419		goto skip_zeroout;
1420
1421	ext4_debug("going to zero out inode table in group %d\n",
1422		   group);
1423	ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1424	if (ret < 0)
1425		goto err_out;
1426	if (barrier)
1427		blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1428
1429skip_zeroout:
1430	ext4_lock_group(sb, group);
1431	gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1432	ext4_group_desc_csum_set(sb, group, gdp);
1433	ext4_unlock_group(sb, group);
1434
1435	BUFFER_TRACE(group_desc_bh,
1436		     "call ext4_handle_dirty_metadata");
1437	ret = ext4_handle_dirty_metadata(handle, NULL,
1438					 group_desc_bh);
1439
1440err_out:
1441	up_write(&grp->alloc_sem);
1442	ext4_journal_stop(handle);
1443out:
1444	return ret;
1445}