Linux Audio

Check our new training course

Loading...
v3.1
   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/jbd2.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 <asm/byteorder.h>
  26
  27#include "ext4.h"
  28#include "ext4_jbd2.h"
  29#include "xattr.h"
  30#include "acl.h"
  31
  32#include <trace/events/ext4.h>
  33
  34/*
  35 * ialloc.c contains the inodes allocation and deallocation routines
  36 */
  37
  38/*
  39 * The free inodes are managed by bitmaps.  A file system contains several
  40 * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
  41 * block for inodes, N blocks for the inode table and data blocks.
  42 *
  43 * The file system contains group descriptors which are located after the
  44 * super block.  Each descriptor contains the number of the bitmap block and
  45 * the free blocks count in the block.
  46 */
  47
  48/*
  49 * To avoid calling the atomic setbit hundreds or thousands of times, we only
  50 * need to use it within a single byte (to ensure we get endianness right).
  51 * We can use memset for the rest of the bitmap as there are no other users.
  52 */
  53void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
  54{
  55	int i;
  56
  57	if (start_bit >= end_bit)
  58		return;
  59
  60	ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
  61	for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
  62		ext4_set_bit(i, bitmap);
  63	if (i < end_bit)
  64		memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
  65}
  66
  67/* Initializes an uninitialized inode bitmap */
  68static unsigned ext4_init_inode_bitmap(struct super_block *sb,
  69				       struct buffer_head *bh,
  70				       ext4_group_t block_group,
  71				       struct ext4_group_desc *gdp)
  72{
 
  73	struct ext4_sb_info *sbi = EXT4_SB(sb);
  74
  75	J_ASSERT_BH(bh, buffer_locked(bh));
  76
  77	/* If checksum is bad mark all blocks and inodes use to prevent
  78	 * allocation, essentially implementing a per-group read-only flag. */
  79	if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
  80		ext4_error(sb, "Checksum bad for group %u", block_group);
  81		ext4_free_blks_set(sb, gdp, 0);
  82		ext4_free_inodes_set(sb, gdp, 0);
  83		ext4_itable_unused_set(sb, gdp, 0);
  84		memset(bh->b_data, 0xff, sb->s_blocksize);
  85		return 0;
 
 
 
 
 
 
 
  86	}
  87
  88	memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
  89	ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
  90			bh->b_data);
 
 
 
  91
  92	return EXT4_INODES_PER_GROUP(sb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  93}
  94
  95/*
  96 * Read the inode allocation bitmap for a given block_group, reading
  97 * into the specified slot in the superblock's bitmap cache.
  98 *
  99 * Return buffer_head of bitmap on success or NULL.
 100 */
 101static struct buffer_head *
 102ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 103{
 104	struct ext4_group_desc *desc;
 105	struct buffer_head *bh = NULL;
 106	ext4_fsblk_t bitmap_blk;
 
 107
 108	desc = ext4_get_group_desc(sb, block_group, NULL);
 109	if (!desc)
 110		return NULL;
 111
 112	bitmap_blk = ext4_inode_bitmap(sb, desc);
 113	bh = sb_getblk(sb, bitmap_blk);
 114	if (unlikely(!bh)) {
 115		ext4_error(sb, "Cannot read inode bitmap - "
 116			    "block_group = %u, inode_bitmap = %llu",
 117			    block_group, bitmap_blk);
 118		return NULL;
 119	}
 120	if (bitmap_uptodate(bh))
 121		return bh;
 122
 123	lock_buffer(bh);
 124	if (bitmap_uptodate(bh)) {
 125		unlock_buffer(bh);
 126		return bh;
 127	}
 128
 129	ext4_lock_group(sb, block_group);
 130	if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 131		ext4_init_inode_bitmap(sb, bh, block_group, desc);
 132		set_bitmap_uptodate(bh);
 133		set_buffer_uptodate(bh);
 
 134		ext4_unlock_group(sb, block_group);
 135		unlock_buffer(bh);
 
 
 
 
 
 136		return bh;
 137	}
 138	ext4_unlock_group(sb, block_group);
 139
 140	if (buffer_uptodate(bh)) {
 141		/*
 142		 * if not uninit if bh is uptodate,
 143		 * bitmap is also uptodate
 144		 */
 145		set_bitmap_uptodate(bh);
 146		unlock_buffer(bh);
 147		return bh;
 148	}
 149	/*
 150	 * submit the buffer_head for read. We can
 151	 * safely mark the bitmap as uptodate now.
 152	 * We do it here so the bitmap uptodate bit
 153	 * get set with buffer lock held.
 154	 */
 155	trace_ext4_load_inode_bitmap(sb, block_group);
 156	set_bitmap_uptodate(bh);
 157	if (bh_submit_read(bh) < 0) {
 
 
 
 158		put_bh(bh);
 159		ext4_error(sb, "Cannot read inode bitmap - "
 160			    "block_group = %u, inode_bitmap = %llu",
 161			    block_group, bitmap_blk);
 162		return NULL;
 163	}
 
 
 
 
 
 164	return bh;
 
 
 
 165}
 166
 167/*
 168 * NOTE! When we get the inode, we're the only people
 169 * that have access to it, and as such there are no
 170 * race conditions we have to worry about. The inode
 171 * is not on the hash-lists, and it cannot be reached
 172 * through the filesystem because the directory entry
 173 * has been deleted earlier.
 174 *
 175 * HOWEVER: we must make sure that we get no aliases,
 176 * which means that we have to call "clear_inode()"
 177 * _before_ we mark the inode not in use in the inode
 178 * bitmaps. Otherwise a newly created file might use
 179 * the same inode number (not actually the same pointer
 180 * though), and then we'd have two inodes sharing the
 181 * same inode number and space on the harddisk.
 182 */
 183void ext4_free_inode(handle_t *handle, struct inode *inode)
 184{
 185	struct super_block *sb = inode->i_sb;
 186	int is_directory;
 187	unsigned long ino;
 188	struct buffer_head *bitmap_bh = NULL;
 189	struct buffer_head *bh2;
 190	ext4_group_t block_group;
 191	unsigned long bit;
 192	struct ext4_group_desc *gdp;
 193	struct ext4_super_block *es;
 194	struct ext4_sb_info *sbi;
 195	int fatal = 0, err, count, cleared;
 
 196
 197	if (atomic_read(&inode->i_count) > 1) {
 198		printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
 199		       atomic_read(&inode->i_count));
 200		return;
 201	}
 202	if (inode->i_nlink) {
 203		printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
 204		       inode->i_nlink);
 
 205		return;
 206	}
 207	if (!sb) {
 208		printk(KERN_ERR "ext4_free_inode: inode on "
 209		       "nonexistent device\n");
 210		return;
 211	}
 212	sbi = EXT4_SB(sb);
 213
 214	ino = inode->i_ino;
 215	ext4_debug("freeing inode %lu\n", ino);
 216	trace_ext4_free_inode(inode);
 217
 218	/*
 219	 * Note: we must free any quota before locking the superblock,
 220	 * as writing the quota to disk may need the lock as well.
 221	 */
 222	dquot_initialize(inode);
 223	ext4_xattr_delete_inode(handle, inode);
 224	dquot_free_inode(inode);
 225	dquot_drop(inode);
 226
 227	is_directory = S_ISDIR(inode->i_mode);
 228
 229	/* Do this BEFORE marking the inode not in use or returning an error */
 230	ext4_clear_inode(inode);
 231
 232	es = EXT4_SB(sb)->s_es;
 233	if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 234		ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 235		goto error_return;
 236	}
 237	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 238	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 239	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 240	if (!bitmap_bh)
 
 
 
 
 241		goto error_return;
 
 
 
 
 
 242
 243	BUFFER_TRACE(bitmap_bh, "get_write_access");
 244	fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 245	if (fatal)
 246		goto error_return;
 247
 248	fatal = -ESRCH;
 249	gdp = ext4_get_group_desc(sb, block_group, &bh2);
 250	if (gdp) {
 251		BUFFER_TRACE(bh2, "get_write_access");
 252		fatal = ext4_journal_get_write_access(handle, bh2);
 253	}
 254	ext4_lock_group(sb, block_group);
 255	cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
 256	if (fatal || !cleared) {
 257		ext4_unlock_group(sb, block_group);
 258		goto out;
 259	}
 260
 261	count = ext4_free_inodes_count(sb, gdp) + 1;
 262	ext4_free_inodes_set(sb, gdp, count);
 263	if (is_directory) {
 264		count = ext4_used_dirs_count(sb, gdp) - 1;
 265		ext4_used_dirs_set(sb, gdp, count);
 266		percpu_counter_dec(&sbi->s_dirs_counter);
 267	}
 268	gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
 
 
 269	ext4_unlock_group(sb, block_group);
 270
 271	percpu_counter_inc(&sbi->s_freeinodes_counter);
 272	if (sbi->s_log_groups_per_flex) {
 273		ext4_group_t f = ext4_flex_group(sbi, block_group);
 274
 275		atomic_inc(&sbi->s_flex_groups[f].free_inodes);
 276		if (is_directory)
 277			atomic_dec(&sbi->s_flex_groups[f].used_dirs);
 278	}
 279	BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 280	fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 281out:
 282	if (cleared) {
 283		BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 284		err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 285		if (!fatal)
 286			fatal = err;
 287		ext4_mark_super_dirty(sb);
 288	} else
 289		ext4_error(sb, "bit already cleared for inode %lu", ino);
 
 
 
 
 
 
 
 
 290
 291error_return:
 292	brelse(bitmap_bh);
 293	ext4_std_error(sb, fatal);
 294}
 295
 296/*
 297 * There are two policies for allocating an inode.  If the new inode is
 298 * a directory, then a forward search is made for a block group with both
 299 * free space and a low directory-to-inode ratio; if that fails, then of
 300 * the groups with above-average free space, that group with the fewest
 301 * directories already is chosen.
 302 *
 303 * For other inodes, search forward from the parent directory\'s block
 304 * group to find a free inode.
 305 */
 306static int find_group_dir(struct super_block *sb, struct inode *parent,
 307				ext4_group_t *best_group)
 308{
 309	ext4_group_t ngroups = ext4_get_groups_count(sb);
 310	unsigned int freei, avefreei;
 311	struct ext4_group_desc *desc, *best_desc = NULL;
 312	ext4_group_t group;
 313	int ret = -1;
 314
 315	freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
 316	avefreei = freei / ngroups;
 317
 318	for (group = 0; group < ngroups; group++) {
 319		desc = ext4_get_group_desc(sb, group, NULL);
 320		if (!desc || !ext4_free_inodes_count(sb, desc))
 321			continue;
 322		if (ext4_free_inodes_count(sb, desc) < avefreei)
 323			continue;
 324		if (!best_desc ||
 325		    (ext4_free_blks_count(sb, desc) >
 326		     ext4_free_blks_count(sb, best_desc))) {
 327			*best_group = group;
 328			best_desc = desc;
 329			ret = 0;
 330		}
 331	}
 332	return ret;
 333}
 334
 335#define free_block_ratio 10
 336
 337static int find_group_flex(struct super_block *sb, struct inode *parent,
 338			   ext4_group_t *best_group)
 339{
 340	struct ext4_sb_info *sbi = EXT4_SB(sb);
 341	struct ext4_group_desc *desc;
 342	struct flex_groups *flex_group = sbi->s_flex_groups;
 343	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 344	ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
 345	ext4_group_t ngroups = ext4_get_groups_count(sb);
 346	int flex_size = ext4_flex_bg_size(sbi);
 347	ext4_group_t best_flex = parent_fbg_group;
 348	int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
 349	int flexbg_free_blocks;
 350	int flex_freeb_ratio;
 351	ext4_group_t n_fbg_groups;
 352	ext4_group_t i;
 353
 354	n_fbg_groups = (ngroups + flex_size - 1) >>
 355		sbi->s_log_groups_per_flex;
 356
 357find_close_to_parent:
 358	flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
 359	flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
 360	if (atomic_read(&flex_group[best_flex].free_inodes) &&
 361	    flex_freeb_ratio > free_block_ratio)
 362		goto found_flexbg;
 363
 364	if (best_flex && best_flex == parent_fbg_group) {
 365		best_flex--;
 366		goto find_close_to_parent;
 367	}
 368
 369	for (i = 0; i < n_fbg_groups; i++) {
 370		if (i == parent_fbg_group || i == parent_fbg_group - 1)
 371			continue;
 372
 373		flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
 374		flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
 375
 376		if (flex_freeb_ratio > free_block_ratio &&
 377		    (atomic_read(&flex_group[i].free_inodes))) {
 378			best_flex = i;
 379			goto found_flexbg;
 380		}
 381
 382		if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
 383		    ((atomic_read(&flex_group[i].free_blocks) >
 384		      atomic_read(&flex_group[best_flex].free_blocks)) &&
 385		     atomic_read(&flex_group[i].free_inodes)))
 386			best_flex = i;
 387	}
 388
 389	if (!atomic_read(&flex_group[best_flex].free_inodes) ||
 390	    !atomic_read(&flex_group[best_flex].free_blocks))
 391		return -1;
 392
 393found_flexbg:
 394	for (i = best_flex * flex_size; i < ngroups &&
 395		     i < (best_flex + 1) * flex_size; i++) {
 396		desc = ext4_get_group_desc(sb, i, NULL);
 397		if (ext4_free_inodes_count(sb, desc)) {
 398			*best_group = i;
 399			goto out;
 400		}
 401	}
 402
 403	return -1;
 404out:
 405	return 0;
 406}
 407
 408struct orlov_stats {
 
 409	__u32 free_inodes;
 410	__u32 free_blocks;
 411	__u32 used_dirs;
 412};
 413
 414/*
 415 * Helper function for Orlov's allocator; returns critical information
 416 * for a particular block group or flex_bg.  If flex_size is 1, then g
 417 * is a block group number; otherwise it is flex_bg number.
 418 */
 419static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 420			    int flex_size, struct orlov_stats *stats)
 421{
 422	struct ext4_group_desc *desc;
 423	struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
 424
 425	if (flex_size > 1) {
 426		stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
 427		stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
 428		stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
 429		return;
 430	}
 431
 432	desc = ext4_get_group_desc(sb, g, NULL);
 433	if (desc) {
 434		stats->free_inodes = ext4_free_inodes_count(sb, desc);
 435		stats->free_blocks = ext4_free_blks_count(sb, desc);
 436		stats->used_dirs = ext4_used_dirs_count(sb, desc);
 437	} else {
 438		stats->free_inodes = 0;
 439		stats->free_blocks = 0;
 440		stats->used_dirs = 0;
 441	}
 442}
 443
 444/*
 445 * Orlov's allocator for directories.
 446 *
 447 * We always try to spread first-level directories.
 448 *
 449 * If there are blockgroups with both free inodes and free blocks counts
 450 * not worse than average we return one with smallest directory count.
 451 * Otherwise we simply return a random group.
 452 *
 453 * For the rest rules look so:
 454 *
 455 * It's OK to put directory into a group unless
 456 * it has too many directories already (max_dirs) or
 457 * it has too few free inodes left (min_inodes) or
 458 * it has too few free blocks left (min_blocks) or
 459 * Parent's group is preferred, if it doesn't satisfy these
 460 * conditions we search cyclically through the rest. If none
 461 * of the groups look good we just look for a group with more
 462 * free inodes than average (starting at parent's group).
 463 */
 464
 465static int find_group_orlov(struct super_block *sb, struct inode *parent,
 466			    ext4_group_t *group, int mode,
 467			    const struct qstr *qstr)
 468{
 469	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 470	struct ext4_sb_info *sbi = EXT4_SB(sb);
 471	ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 472	int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 473	unsigned int freei, avefreei;
 474	ext4_fsblk_t freeb, avefreeb;
 475	unsigned int ndirs;
 476	int max_dirs, min_inodes;
 477	ext4_grpblk_t min_blocks;
 478	ext4_group_t i, grp, g, ngroups;
 479	struct ext4_group_desc *desc;
 480	struct orlov_stats stats;
 481	int flex_size = ext4_flex_bg_size(sbi);
 482	struct dx_hash_info hinfo;
 483
 484	ngroups = real_ngroups;
 485	if (flex_size > 1) {
 486		ngroups = (real_ngroups + flex_size - 1) >>
 487			sbi->s_log_groups_per_flex;
 488		parent_group >>= sbi->s_log_groups_per_flex;
 489	}
 490
 491	freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 492	avefreei = freei / ngroups;
 493	freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
 494	avefreeb = freeb;
 495	do_div(avefreeb, ngroups);
 
 496	ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 497
 498	if (S_ISDIR(mode) &&
 499	    ((parent == sb->s_root->d_inode) ||
 500	     (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 501		int best_ndir = inodes_per_group;
 502		int ret = -1;
 503
 504		if (qstr) {
 505			hinfo.hash_version = DX_HASH_HALF_MD4;
 506			hinfo.seed = sbi->s_hash_seed;
 507			ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
 508			grp = hinfo.hash;
 509		} else
 510			get_random_bytes(&grp, sizeof(grp));
 511		parent_group = (unsigned)grp % ngroups;
 512		for (i = 0; i < ngroups; i++) {
 513			g = (parent_group + i) % ngroups;
 514			get_orlov_stats(sb, g, flex_size, &stats);
 515			if (!stats.free_inodes)
 516				continue;
 517			if (stats.used_dirs >= best_ndir)
 518				continue;
 519			if (stats.free_inodes < avefreei)
 520				continue;
 521			if (stats.free_blocks < avefreeb)
 522				continue;
 523			grp = g;
 524			ret = 0;
 525			best_ndir = stats.used_dirs;
 526		}
 527		if (ret)
 528			goto fallback;
 529	found_flex_bg:
 530		if (flex_size == 1) {
 531			*group = grp;
 532			return 0;
 533		}
 534
 535		/*
 536		 * We pack inodes at the beginning of the flexgroup's
 537		 * inode tables.  Block allocation decisions will do
 538		 * something similar, although regular files will
 539		 * start at 2nd block group of the flexgroup.  See
 540		 * ext4_ext_find_goal() and ext4_find_near().
 541		 */
 542		grp *= flex_size;
 543		for (i = 0; i < flex_size; i++) {
 544			if (grp+i >= real_ngroups)
 545				break;
 546			desc = ext4_get_group_desc(sb, grp+i, NULL);
 547			if (desc && ext4_free_inodes_count(sb, desc)) {
 548				*group = grp+i;
 549				return 0;
 550			}
 551		}
 552		goto fallback;
 553	}
 554
 555	max_dirs = ndirs / ngroups + inodes_per_group / 16;
 556	min_inodes = avefreei - inodes_per_group*flex_size / 4;
 557	if (min_inodes < 1)
 558		min_inodes = 1;
 559	min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
 560
 561	/*
 562	 * Start looking in the flex group where we last allocated an
 563	 * inode for this parent directory
 564	 */
 565	if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 566		parent_group = EXT4_I(parent)->i_last_alloc_group;
 567		if (flex_size > 1)
 568			parent_group >>= sbi->s_log_groups_per_flex;
 569	}
 570
 571	for (i = 0; i < ngroups; i++) {
 572		grp = (parent_group + i) % ngroups;
 573		get_orlov_stats(sb, grp, flex_size, &stats);
 574		if (stats.used_dirs >= max_dirs)
 575			continue;
 576		if (stats.free_inodes < min_inodes)
 577			continue;
 578		if (stats.free_blocks < min_blocks)
 579			continue;
 580		goto found_flex_bg;
 581	}
 582
 583fallback:
 584	ngroups = real_ngroups;
 585	avefreei = freei / ngroups;
 586fallback_retry:
 587	parent_group = EXT4_I(parent)->i_block_group;
 588	for (i = 0; i < ngroups; i++) {
 589		grp = (parent_group + i) % ngroups;
 590		desc = ext4_get_group_desc(sb, grp, NULL);
 591		if (desc && ext4_free_inodes_count(sb, desc) &&
 592		    ext4_free_inodes_count(sb, desc) >= avefreei) {
 593			*group = grp;
 594			return 0;
 
 
 595		}
 596	}
 597
 598	if (avefreei) {
 599		/*
 600		 * The free-inodes counter is approximate, and for really small
 601		 * filesystems the above test can fail to find any blockgroups
 602		 */
 603		avefreei = 0;
 604		goto fallback_retry;
 605	}
 606
 607	return -1;
 608}
 609
 610static int find_group_other(struct super_block *sb, struct inode *parent,
 611			    ext4_group_t *group, int mode)
 612{
 613	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 614	ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 615	struct ext4_group_desc *desc;
 616	int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 617
 618	/*
 619	 * Try to place the inode is the same flex group as its
 620	 * parent.  If we can't find space, use the Orlov algorithm to
 621	 * find another flex group, and store that information in the
 622	 * parent directory's inode information so that use that flex
 623	 * group for future allocations.
 624	 */
 625	if (flex_size > 1) {
 626		int retry = 0;
 627
 628	try_again:
 629		parent_group &= ~(flex_size-1);
 630		last = parent_group + flex_size;
 631		if (last > ngroups)
 632			last = ngroups;
 633		for  (i = parent_group; i < last; i++) {
 634			desc = ext4_get_group_desc(sb, i, NULL);
 635			if (desc && ext4_free_inodes_count(sb, desc)) {
 636				*group = i;
 637				return 0;
 638			}
 639		}
 640		if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 641			retry = 1;
 642			parent_group = EXT4_I(parent)->i_last_alloc_group;
 643			goto try_again;
 644		}
 645		/*
 646		 * If this didn't work, use the Orlov search algorithm
 647		 * to find a new flex group; we pass in the mode to
 648		 * avoid the topdir algorithms.
 649		 */
 650		*group = parent_group + flex_size;
 651		if (*group > ngroups)
 652			*group = 0;
 653		return find_group_orlov(sb, parent, group, mode, NULL);
 654	}
 655
 656	/*
 657	 * Try to place the inode in its parent directory
 658	 */
 659	*group = parent_group;
 660	desc = ext4_get_group_desc(sb, *group, NULL);
 661	if (desc && ext4_free_inodes_count(sb, desc) &&
 662			ext4_free_blks_count(sb, desc))
 663		return 0;
 664
 665	/*
 666	 * We're going to place this inode in a different blockgroup from its
 667	 * parent.  We want to cause files in a common directory to all land in
 668	 * the same blockgroup.  But we want files which are in a different
 669	 * directory which shares a blockgroup with our parent to land in a
 670	 * different blockgroup.
 671	 *
 672	 * So add our directory's i_ino into the starting point for the hash.
 673	 */
 674	*group = (*group + parent->i_ino) % ngroups;
 675
 676	/*
 677	 * Use a quadratic hash to find a group with a free inode and some free
 678	 * blocks.
 679	 */
 680	for (i = 1; i < ngroups; i <<= 1) {
 681		*group += i;
 682		if (*group >= ngroups)
 683			*group -= ngroups;
 684		desc = ext4_get_group_desc(sb, *group, NULL);
 685		if (desc && ext4_free_inodes_count(sb, desc) &&
 686				ext4_free_blks_count(sb, desc))
 687			return 0;
 688	}
 689
 690	/*
 691	 * That failed: try linear search for a free inode, even if that group
 692	 * has no free blocks.
 693	 */
 694	*group = parent_group;
 695	for (i = 0; i < ngroups; i++) {
 696		if (++*group >= ngroups)
 697			*group = 0;
 698		desc = ext4_get_group_desc(sb, *group, NULL);
 699		if (desc && ext4_free_inodes_count(sb, desc))
 700			return 0;
 701	}
 702
 703	return -1;
 704}
 705
 706/*
 707 * claim the inode from the inode bitmap. If the group
 708 * is uninit we need to take the groups's ext4_group_lock
 709 * and clear the uninit flag. The inode bitmap update
 710 * and group desc uninit flag clear should be done
 711 * after holding ext4_group_lock so that ext4_read_inode_bitmap
 712 * doesn't race with the ext4_claim_inode
 713 */
 714static int ext4_claim_inode(struct super_block *sb,
 715			struct buffer_head *inode_bitmap_bh,
 716			unsigned long ino, ext4_group_t group, int mode)
 717{
 718	int free = 0, retval = 0, count;
 719	struct ext4_sb_info *sbi = EXT4_SB(sb);
 720	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 721	struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 722
 723	/*
 724	 * We have to be sure that new inode allocation does not race with
 725	 * inode table initialization, because otherwise we may end up
 726	 * allocating and writing new inode right before sb_issue_zeroout
 727	 * takes place and overwriting our new inode with zeroes. So we
 728	 * take alloc_sem to prevent it.
 729	 */
 730	down_read(&grp->alloc_sem);
 731	ext4_lock_group(sb, group);
 732	if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
 733		/* not a free inode */
 734		retval = 1;
 735		goto err_ret;
 736	}
 737	ino++;
 738	if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
 739			ino > EXT4_INODES_PER_GROUP(sb)) {
 740		ext4_unlock_group(sb, group);
 741		up_read(&grp->alloc_sem);
 742		ext4_error(sb, "reserved inode or inode > inodes count - "
 743			   "block_group = %u, inode=%lu", group,
 744			   ino + group * EXT4_INODES_PER_GROUP(sb));
 745		return 1;
 746	}
 747	/* If we didn't allocate from within the initialized part of the inode
 748	 * table then we need to initialize up to this inode. */
 749	if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
 750
 751		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 752			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
 753			/* When marking the block group with
 754			 * ~EXT4_BG_INODE_UNINIT we don't want to depend
 755			 * on the value of bg_itable_unused even though
 756			 * mke2fs could have initialized the same for us.
 757			 * Instead we calculated the value below
 758			 */
 759
 760			free = 0;
 761		} else {
 762			free = EXT4_INODES_PER_GROUP(sb) -
 763				ext4_itable_unused_count(sb, gdp);
 764		}
 765
 
 
 
 766		/*
 767		 * Check the relative inode number against the last used
 768		 * relative inode number in this group. if it is greater
 769		 * we need to  update the bg_itable_unused count
 770		 *
 771		 */
 772		if (ino > free)
 773			ext4_itable_unused_set(sb, gdp,
 774					(EXT4_INODES_PER_GROUP(sb) - ino));
 775	}
 776	count = ext4_free_inodes_count(sb, gdp) - 1;
 777	ext4_free_inodes_set(sb, gdp, count);
 778	if (S_ISDIR(mode)) {
 779		count = ext4_used_dirs_count(sb, gdp) + 1;
 780		ext4_used_dirs_set(sb, gdp, count);
 781		if (sbi->s_log_groups_per_flex) {
 782			ext4_group_t f = ext4_flex_group(sbi, group);
 783
 784			atomic_inc(&sbi->s_flex_groups[f].used_dirs);
 785		}
 786	}
 787	gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
 788err_ret:
 789	ext4_unlock_group(sb, group);
 790	up_read(&grp->alloc_sem);
 791	return retval;
 
 
 
 
 792}
 793
 794/*
 795 * There are two policies for allocating an inode.  If the new inode is
 796 * a directory, then a forward search is made for a block group with both
 797 * free space and a low directory-to-inode ratio; if that fails, then of
 798 * the groups with above-average free space, that group with the fewest
 799 * directories already is chosen.
 800 *
 801 * For other inodes, search forward from the parent directory's block
 802 * group to find a free inode.
 803 */
 804struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
 805			     const struct qstr *qstr, __u32 goal)
 
 
 806{
 807	struct super_block *sb;
 808	struct buffer_head *inode_bitmap_bh = NULL;
 809	struct buffer_head *group_desc_bh;
 810	ext4_group_t ngroups, group = 0;
 811	unsigned long ino = 0;
 812	struct inode *inode;
 813	struct ext4_group_desc *gdp = NULL;
 814	struct ext4_inode_info *ei;
 815	struct ext4_sb_info *sbi;
 816	int ret2, err = 0;
 817	struct inode *ret;
 818	ext4_group_t i;
 819	int free = 0;
 820	static int once = 1;
 821	ext4_group_t flex_group;
 
 
 822
 823	/* Cannot create files in a deleted directory */
 824	if (!dir || !dir->i_nlink)
 825		return ERR_PTR(-EPERM);
 826
 
 
 
 
 
 
 
 
 
 
 
 
 
 827	sb = dir->i_sb;
 828	ngroups = ext4_get_groups_count(sb);
 829	trace_ext4_request_inode(dir, mode);
 830	inode = new_inode(sb);
 831	if (!inode)
 832		return ERR_PTR(-ENOMEM);
 833	ei = EXT4_I(inode);
 834	sbi = EXT4_SB(sb);
 835
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 836	if (!goal)
 837		goal = sbi->s_inode_goal;
 838
 839	if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 840		group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 841		ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 842		ret2 = 0;
 843		goto got_group;
 844	}
 845
 846	if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
 847		ret2 = find_group_flex(sb, dir, &group);
 848		if (ret2 == -1) {
 849			ret2 = find_group_other(sb, dir, &group, mode);
 850			if (ret2 == 0 && once) {
 851				once = 0;
 852				printk(KERN_NOTICE "ext4: find_group_flex "
 853				       "failed, fallback succeeded dir %lu\n",
 854				       dir->i_ino);
 855			}
 856		}
 857		goto got_group;
 858	}
 859
 860	if (S_ISDIR(mode)) {
 861		if (test_opt(sb, OLDALLOC))
 862			ret2 = find_group_dir(sb, dir, &group);
 863		else
 864			ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 865	} else
 866		ret2 = find_group_other(sb, dir, &group, mode);
 867
 868got_group:
 869	EXT4_I(dir)->i_last_alloc_group = group;
 870	err = -ENOSPC;
 871	if (ret2 == -1)
 872		goto out;
 873
 
 
 
 
 
 874	for (i = 0; i < ngroups; i++, ino = 0) {
 875		err = -EIO;
 876
 877		gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 878		if (!gdp)
 879			goto fail;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 880
 881		brelse(inode_bitmap_bh);
 882		inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 883		if (!inode_bitmap_bh)
 884			goto fail;
 
 
 
 
 
 
 885
 886repeat_in_this_group:
 887		ino = ext4_find_next_zero_bit((unsigned long *)
 888					      inode_bitmap_bh->b_data,
 889					      EXT4_INODES_PER_GROUP(sb), ino);
 890
 891		if (ino < EXT4_INODES_PER_GROUP(sb)) {
 892
 893			BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 894			err = ext4_journal_get_write_access(handle,
 895							    inode_bitmap_bh);
 896			if (err)
 897				goto fail;
 898
 899			BUFFER_TRACE(group_desc_bh, "get_write_access");
 900			err = ext4_journal_get_write_access(handle,
 901								group_desc_bh);
 902			if (err)
 903				goto fail;
 904			if (!ext4_claim_inode(sb, inode_bitmap_bh,
 905						ino, group, mode)) {
 906				/* we won it */
 907				BUFFER_TRACE(inode_bitmap_bh,
 908					"call ext4_handle_dirty_metadata");
 909				err = ext4_handle_dirty_metadata(handle,
 910								 NULL,
 911							inode_bitmap_bh);
 912				if (err)
 913					goto fail;
 914				/* zero bit is inode number 1*/
 915				ino++;
 916				goto got;
 917			}
 918			/* we lost it */
 919			ext4_handle_release_buffer(handle, inode_bitmap_bh);
 920			ext4_handle_release_buffer(handle, group_desc_bh);
 921
 922			if (++ino < EXT4_INODES_PER_GROUP(sb))
 923				goto repeat_in_this_group;
 924		}
 925
 926		/*
 927		 * This case is possible in concurrent environment.  It is very
 928		 * rare.  We cannot repeat the find_group_xxx() call because
 929		 * that will simply return the same blockgroup, because the
 930		 * group descriptor metadata has not yet been updated.
 931		 * So we just go onto the next blockgroup.
 932		 */
 
 
 
 
 
 
 
 
 933		if (++group == ngroups)
 934			group = 0;
 935	}
 936	err = -ENOSPC;
 937	goto out;
 938
 939got:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 940	/* We may have to initialize the block bitmap if it isn't already */
 941	if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
 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		BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
 947		err = ext4_journal_get_write_access(handle, block_bitmap_bh);
 948		if (err) {
 949			brelse(block_bitmap_bh);
 950			goto fail;
 
 951		}
 952
 953		free = 0;
 954		ext4_lock_group(sb, group);
 
 955		/* recheck and clear flag under lock if we still need to */
 
 956		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 957			free = ext4_free_blocks_after_init(sb, group, gdp);
 958			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
 959			ext4_free_blks_set(sb, gdp, free);
 960			gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
 961								gdp);
 
 
 962		}
 963		ext4_unlock_group(sb, group);
 
 964
 965		/* Don't need to dirty bitmap block if we didn't change it */
 966		if (free) {
 967			BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
 968			err = ext4_handle_dirty_metadata(handle,
 969							NULL, block_bitmap_bh);
 970		}
 
 971
 972		brelse(block_bitmap_bh);
 973		if (err)
 974			goto fail;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 975	}
 
 
 
 
 
 
 
 976	BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
 977	err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
 978	if (err)
 979		goto fail;
 
 
 980
 981	percpu_counter_dec(&sbi->s_freeinodes_counter);
 982	if (S_ISDIR(mode))
 983		percpu_counter_inc(&sbi->s_dirs_counter);
 984	ext4_mark_super_dirty(sb);
 985
 986	if (sbi->s_log_groups_per_flex) {
 987		flex_group = ext4_flex_group(sbi, group);
 988		atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
 989	}
 990
 991	if (test_opt(sb, GRPID)) {
 992		inode->i_mode = mode;
 993		inode->i_uid = current_fsuid();
 994		inode->i_gid = dir->i_gid;
 995	} else
 996		inode_init_owner(inode, dir, mode);
 997
 998	inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
 999	/* This is the optimal IO size (for stat), not the fs block size */
1000	inode->i_blocks = 0;
1001	inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1002						       ext4_current_time(inode);
1003
1004	memset(ei->i_data, 0, sizeof(ei->i_data));
1005	ei->i_dir_start_lookup = 0;
1006	ei->i_disksize = 0;
1007
1008	/*
1009	 * Don't inherit extent flag from directory, amongst others. We set
1010	 * extent flag on newly created directory and file only if -o extent
1011	 * mount option is specified
1012	 */
1013	ei->i_flags =
1014		ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1015	ei->i_file_acl = 0;
1016	ei->i_dtime = 0;
1017	ei->i_block_group = group;
1018	ei->i_last_alloc_group = ~0;
1019
1020	ext4_set_inode_flags(inode);
1021	if (IS_DIRSYNC(inode))
1022		ext4_handle_sync(handle);
1023	if (insert_inode_locked(inode) < 0) {
1024		err = -EINVAL;
1025		goto fail_drop;
 
 
 
 
 
 
1026	}
1027	spin_lock(&sbi->s_next_gen_lock);
1028	inode->i_generation = sbi->s_next_generation++;
1029	spin_unlock(&sbi->s_next_gen_lock);
1030
 
 
 
 
 
 
 
 
 
 
 
1031	ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1032	ext4_set_inode_state(inode, EXT4_STATE_NEW);
1033
1034	ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1035
 
 
1036	ret = inode;
1037	dquot_initialize(inode);
1038	err = dquot_alloc_inode(inode);
1039	if (err)
1040		goto fail_drop;
1041
1042	err = ext4_init_acl(handle, inode, dir);
1043	if (err)
1044		goto fail_free_drop;
1045
1046	err = ext4_init_security(handle, inode, dir, qstr);
1047	if (err)
1048		goto fail_free_drop;
1049
1050	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1051		/* set extent flag only for directory, file and normal symlink*/
1052		if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1053			ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1054			ext4_ext_tree_init(handle, inode);
1055		}
1056	}
1057
1058	if (ext4_handle_valid(handle)) {
1059		ei->i_sync_tid = handle->h_transaction->t_tid;
1060		ei->i_datasync_tid = handle->h_transaction->t_tid;
1061	}
1062
 
 
 
 
 
 
1063	err = ext4_mark_inode_dirty(handle, inode);
1064	if (err) {
1065		ext4_std_error(sb, err);
1066		goto fail_free_drop;
1067	}
1068
1069	ext4_debug("allocating inode %lu\n", inode->i_ino);
1070	trace_ext4_allocate_inode(inode, dir, mode);
1071	goto really_out;
1072fail:
1073	ext4_std_error(sb, err);
1074out:
1075	iput(inode);
1076	ret = ERR_PTR(err);
1077really_out:
1078	brelse(inode_bitmap_bh);
1079	return ret;
1080
1081fail_free_drop:
1082	dquot_free_inode(inode);
1083
1084fail_drop:
 
 
 
1085	dquot_drop(inode);
1086	inode->i_flags |= S_NOQUOTA;
1087	inode->i_nlink = 0;
1088	unlock_new_inode(inode);
1089	iput(inode);
1090	brelse(inode_bitmap_bh);
1091	return ERR_PTR(err);
1092}
1093
1094/* Verify that we are loading a valid orphan from disk */
1095struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1096{
1097	unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1098	ext4_group_t block_group;
1099	int bit;
1100	struct buffer_head *bitmap_bh;
1101	struct inode *inode = NULL;
1102	long err = -EIO;
1103
1104	/* Error cases - e2fsck has already cleaned up for us */
1105	if (ino > max_ino) {
1106		ext4_warning(sb, "bad orphan ino %lu!  e2fsck was run?", ino);
 
1107		goto error;
1108	}
1109
1110	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1111	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1112	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1113	if (!bitmap_bh) {
1114		ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
 
 
1115		goto error;
1116	}
1117
1118	/* Having the inode bit set should be a 100% indicator that this
1119	 * is a valid orphan (no e2fsck run on fs).  Orphans also include
1120	 * inodes that were being truncated, so we can't check i_nlink==0.
1121	 */
1122	if (!ext4_test_bit(bit, bitmap_bh->b_data))
1123		goto bad_orphan;
1124
1125	inode = ext4_iget(sb, ino);
1126	if (IS_ERR(inode))
1127		goto iget_failed;
1128
1129	/*
1130	 * If the orphans has i_nlinks > 0 then it should be able to be
1131	 * truncated, otherwise it won't be removed from the orphan list
1132	 * during processing and an infinite loop will result.
1133	 */
1134	if (inode->i_nlink && !ext4_can_truncate(inode))
1135		goto bad_orphan;
1136
1137	if (NEXT_ORPHAN(inode) > max_ino)
1138		goto bad_orphan;
1139	brelse(bitmap_bh);
1140	return inode;
1141
1142iget_failed:
1143	err = PTR_ERR(inode);
1144	inode = NULL;
1145bad_orphan:
1146	ext4_warning(sb, "bad orphan inode %lu!  e2fsck was run?", ino);
1147	printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1148	       bit, (unsigned long long)bitmap_bh->b_blocknr,
1149	       ext4_test_bit(bit, bitmap_bh->b_data));
1150	printk(KERN_NOTICE "inode=%p\n", inode);
1151	if (inode) {
1152		printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1153		       is_bad_inode(inode));
1154		printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1155		       NEXT_ORPHAN(inode));
1156		printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1157		printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1158		/* Avoid freeing blocks if we got a bad deleted inode */
1159		if (inode->i_nlink == 0)
1160			inode->i_blocks = 0;
1161		iput(inode);
1162	}
1163	brelse(bitmap_bh);
1164error:
1165	return ERR_PTR(err);
1166}
1167
1168unsigned long ext4_count_free_inodes(struct super_block *sb)
1169{
1170	unsigned long desc_count;
1171	struct ext4_group_desc *gdp;
1172	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1173#ifdef EXT4FS_DEBUG
1174	struct ext4_super_block *es;
1175	unsigned long bitmap_count, x;
1176	struct buffer_head *bitmap_bh = NULL;
1177
1178	es = EXT4_SB(sb)->s_es;
1179	desc_count = 0;
1180	bitmap_count = 0;
1181	gdp = NULL;
1182	for (i = 0; i < ngroups; i++) {
1183		gdp = ext4_get_group_desc(sb, i, NULL);
1184		if (!gdp)
1185			continue;
1186		desc_count += ext4_free_inodes_count(sb, gdp);
1187		brelse(bitmap_bh);
1188		bitmap_bh = ext4_read_inode_bitmap(sb, i);
1189		if (!bitmap_bh)
 
1190			continue;
 
1191
1192		x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
 
1193		printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1194			(unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1195		bitmap_count += x;
1196	}
1197	brelse(bitmap_bh);
1198	printk(KERN_DEBUG "ext4_count_free_inodes: "
1199	       "stored = %u, computed = %lu, %lu\n",
1200	       le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1201	return desc_count;
1202#else
1203	desc_count = 0;
1204	for (i = 0; i < ngroups; i++) {
1205		gdp = ext4_get_group_desc(sb, i, NULL);
1206		if (!gdp)
1207			continue;
1208		desc_count += ext4_free_inodes_count(sb, gdp);
1209		cond_resched();
1210	}
1211	return desc_count;
1212#endif
1213}
1214
1215/* Called at mount-time, super-block is locked */
1216unsigned long ext4_count_dirs(struct super_block * sb)
1217{
1218	unsigned long count = 0;
1219	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1220
1221	for (i = 0; i < ngroups; i++) {
1222		struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1223		if (!gdp)
1224			continue;
1225		count += ext4_used_dirs_count(sb, gdp);
1226	}
1227	return count;
1228}
1229
1230/*
1231 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1232 * inode table. Must be called without any spinlock held. The only place
1233 * where it is called from on active part of filesystem is ext4lazyinit
1234 * thread, so we do not need any special locks, however we have to prevent
1235 * inode allocation from the current group, so we take alloc_sem lock, to
1236 * block ext4_claim_inode until we are finished.
1237 */
1238extern int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1239				 int barrier)
1240{
1241	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1242	struct ext4_sb_info *sbi = EXT4_SB(sb);
1243	struct ext4_group_desc *gdp = NULL;
1244	struct buffer_head *group_desc_bh;
1245	handle_t *handle;
1246	ext4_fsblk_t blk;
1247	int num, ret = 0, used_blks = 0;
1248
1249	/* This should not happen, but just to be sure check this */
1250	if (sb->s_flags & MS_RDONLY) {
1251		ret = 1;
1252		goto out;
1253	}
1254
1255	gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1256	if (!gdp)
1257		goto out;
1258
1259	/*
1260	 * We do not need to lock this, because we are the only one
1261	 * handling this flag.
1262	 */
1263	if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1264		goto out;
1265
1266	handle = ext4_journal_start_sb(sb, 1);
1267	if (IS_ERR(handle)) {
1268		ret = PTR_ERR(handle);
1269		goto out;
1270	}
1271
1272	down_write(&grp->alloc_sem);
1273	/*
1274	 * If inode bitmap was already initialized there may be some
1275	 * used inodes so we need to skip blocks with used inodes in
1276	 * inode table.
1277	 */
1278	if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1279		used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1280			    ext4_itable_unused_count(sb, gdp)),
1281			    sbi->s_inodes_per_block);
1282
1283	if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1284		ext4_error(sb, "Something is wrong with group %u\n"
1285			   "Used itable blocks: %d"
1286			   "itable unused count: %u\n",
1287			   group, used_blks,
1288			   ext4_itable_unused_count(sb, gdp));
1289		ret = 1;
1290		goto err_out;
1291	}
1292
1293	blk = ext4_inode_table(sb, gdp) + used_blks;
1294	num = sbi->s_itb_per_group - used_blks;
1295
1296	BUFFER_TRACE(group_desc_bh, "get_write_access");
1297	ret = ext4_journal_get_write_access(handle,
1298					    group_desc_bh);
1299	if (ret)
1300		goto err_out;
1301
1302	/*
1303	 * Skip zeroout if the inode table is full. But we set the ZEROED
1304	 * flag anyway, because obviously, when it is full it does not need
1305	 * further zeroing.
1306	 */
1307	if (unlikely(num == 0))
1308		goto skip_zeroout;
1309
1310	ext4_debug("going to zero out inode table in group %d\n",
1311		   group);
1312	ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1313	if (ret < 0)
1314		goto err_out;
1315	if (barrier)
1316		blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1317
1318skip_zeroout:
1319	ext4_lock_group(sb, group);
1320	gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1321	gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1322	ext4_unlock_group(sb, group);
1323
1324	BUFFER_TRACE(group_desc_bh,
1325		     "call ext4_handle_dirty_metadata");
1326	ret = ext4_handle_dirty_metadata(handle, NULL,
1327					 group_desc_bh);
1328
1329err_out:
1330	up_write(&grp->alloc_sem);
1331	ext4_journal_stop(handle);
1332out:
1333	return ret;
1334}
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}