dir.c

   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * dir.c
   4 *
   5 * Creates, reads, walks and deletes directory-nodes
   6 *
   7 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   8 *
   9 *  Portions of this code from linux/fs/ext3/dir.c
  10 *
  11 *  Copyright (C) 1992, 1993, 1994, 1995
  12 *  Remy Card (card@masi.ibp.fr)
  13 *  Laboratoire MASI - Institut Blaise pascal
  14 *  Universite Pierre et Marie Curie (Paris VI)
  15 *
  16 *   from
  17 *
  18 *   linux/fs/minix/dir.c
  19 *
  20 *   Copyright (C) 1991, 1992 Linus Torvalds
  21 */
  22
  23#include <linux/fs.h>
  24#include <linux/types.h>
  25#include <linux/slab.h>
  26#include <linux/highmem.h>
  27#include <linux/quotaops.h>
  28#include <linux/sort.h>
  29#include <linux/iversion.h>
  30
  31#include <cluster/masklog.h>
  32
  33#include "ocfs2.h"
  34
  35#include "alloc.h"
  36#include "blockcheck.h"
  37#include "dir.h"
  38#include "dlmglue.h"
  39#include "extent_map.h"
  40#include "file.h"
  41#include "inode.h"
  42#include "journal.h"
  43#include "namei.h"
  44#include "suballoc.h"
  45#include "super.h"
  46#include "sysfile.h"
  47#include "uptodate.h"
  48#include "ocfs2_trace.h"
  49
  50#include "buffer_head_io.h"
  51
  52#define NAMEI_RA_CHUNKS  2
  53#define NAMEI_RA_BLOCKS  4
  54#define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  55
  56static int ocfs2_do_extend_dir(struct super_block *sb,
  57			       handle_t *handle,
  58			       struct inode *dir,
  59			       struct buffer_head *parent_fe_bh,
  60			       struct ocfs2_alloc_context *data_ac,
  61			       struct ocfs2_alloc_context *meta_ac,
  62			       struct buffer_head **new_bh);
  63static int ocfs2_dir_indexed(struct inode *inode);
  64
  65/*
  66 * These are distinct checks because future versions of the file system will
  67 * want to have a trailing dirent structure independent of indexing.
  68 */
  69static int ocfs2_supports_dir_trailer(struct inode *dir)
  70{
  71	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
  72
  73	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
  74		return 0;
  75
  76	return ocfs2_meta_ecc(osb) || ocfs2_dir_indexed(dir);
  77}
  78
  79/*
  80 * "new' here refers to the point at which we're creating a new
  81 * directory via "mkdir()", but also when we're expanding an inline
  82 * directory. In either case, we don't yet have the indexing bit set
  83 * on the directory, so the standard checks will fail in when metaecc
  84 * is turned off. Only directory-initialization type functions should
  85 * use this then. Everything else wants ocfs2_supports_dir_trailer()
  86 */
  87static int ocfs2_new_dir_wants_trailer(struct inode *dir)
  88{
  89	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
  90
  91	return ocfs2_meta_ecc(osb) ||
  92		ocfs2_supports_indexed_dirs(osb);
  93}
  94
  95static inline unsigned int ocfs2_dir_trailer_blk_off(struct super_block *sb)
  96{
  97	return sb->s_blocksize - sizeof(struct ocfs2_dir_block_trailer);
  98}
  99
 100#define ocfs2_trailer_from_bh(_bh, _sb) ((struct ocfs2_dir_block_trailer *) ((_bh)->b_data + ocfs2_dir_trailer_blk_off((_sb))))
 101
 102/* XXX ocfs2_block_dqtrailer() is similar but not quite - can we make
 103 * them more consistent? */
 104struct ocfs2_dir_block_trailer *ocfs2_dir_trailer_from_size(int blocksize,
 105							    void *data)
 106{
 107	char *p = data;
 108
 109	p += blocksize - sizeof(struct ocfs2_dir_block_trailer);
 110	return (struct ocfs2_dir_block_trailer *)p;
 111}
 112
 113/*
 114 * XXX: This is executed once on every dirent. We should consider optimizing
 115 * it.
 116 */
 117static int ocfs2_skip_dir_trailer(struct inode *dir,
 118				  struct ocfs2_dir_entry *de,
 119				  unsigned long offset,
 120				  unsigned long blklen)
 121{
 122	unsigned long toff = blklen - sizeof(struct ocfs2_dir_block_trailer);
 123
 124	if (!ocfs2_supports_dir_trailer(dir))
 125		return 0;
 126
 127	if (offset != toff)
 128		return 0;
 129
 130	return 1;
 131}
 132
 133static void ocfs2_init_dir_trailer(struct inode *inode,
 134				   struct buffer_head *bh, u16 rec_len)
 135{
 136	struct ocfs2_dir_block_trailer *trailer;
 137
 138	trailer = ocfs2_trailer_from_bh(bh, inode->i_sb);
 139	strscpy(trailer->db_signature, OCFS2_DIR_TRAILER_SIGNATURE);
 140	trailer->db_compat_rec_len =
 141			cpu_to_le16(sizeof(struct ocfs2_dir_block_trailer));
 142	trailer->db_parent_dinode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 143	trailer->db_blkno = cpu_to_le64(bh->b_blocknr);
 144	trailer->db_free_rec_len = cpu_to_le16(rec_len);
 145}
 146/*
 147 * Link an unindexed block with a dir trailer structure into the index free
 148 * list. This function will modify dirdata_bh, but assumes you've already
 149 * passed it to the journal.
 150 */
 151static int ocfs2_dx_dir_link_trailer(struct inode *dir, handle_t *handle,
 152				     struct buffer_head *dx_root_bh,
 153				     struct buffer_head *dirdata_bh)
 154{
 155	int ret;
 156	struct ocfs2_dx_root_block *dx_root;
 157	struct ocfs2_dir_block_trailer *trailer;
 158
 159	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
 160				      OCFS2_JOURNAL_ACCESS_WRITE);
 161	if (ret) {
 162		mlog_errno(ret);
 163		goto out;
 164	}
 165	trailer = ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
 166	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
 167
 168	trailer->db_free_next = dx_root->dr_free_blk;
 169	dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
 170
 171	ocfs2_journal_dirty(handle, dx_root_bh);
 172
 173out:
 174	return ret;
 175}
 176
 177static int ocfs2_free_list_at_root(struct ocfs2_dir_lookup_result *res)
 178{
 179	return res->dl_prev_leaf_bh == NULL;
 180}
 181
 182void ocfs2_free_dir_lookup_result(struct ocfs2_dir_lookup_result *res)
 183{
 184	brelse(res->dl_dx_root_bh);
 185	brelse(res->dl_leaf_bh);
 186	brelse(res->dl_dx_leaf_bh);
 187	brelse(res->dl_prev_leaf_bh);
 188}
 189
 190static int ocfs2_dir_indexed(struct inode *inode)
 191{
 192	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INDEXED_DIR_FL)
 193		return 1;
 194	return 0;
 195}
 196
 197static inline int ocfs2_dx_root_inline(struct ocfs2_dx_root_block *dx_root)
 198{
 199	return dx_root->dr_flags & OCFS2_DX_FLAG_INLINE;
 200}
 201
 202/*
 203 * Hashing code adapted from ext3
 204 */
 205#define DELTA 0x9E3779B9
 206
 207static void TEA_transform(__u32 buf[4], __u32 const in[])
 208{
 209	__u32	sum = 0;
 210	__u32	b0 = buf[0], b1 = buf[1];
 211	__u32	a = in[0], b = in[1], c = in[2], d = in[3];
 212	int	n = 16;
 213
 214	do {
 215		sum += DELTA;
 216		b0 += ((b1 << 4)+a) ^ (b1+sum) ^ ((b1 >> 5)+b);
 217		b1 += ((b0 << 4)+c) ^ (b0+sum) ^ ((b0 >> 5)+d);
 218	} while (--n);
 219
 220	buf[0] += b0;
 221	buf[1] += b1;
 222}
 223
 224static void str2hashbuf(const char *msg, int len, __u32 *buf, int num)
 225{
 226	__u32	pad, val;
 227	int	i;
 228
 229	pad = (__u32)len | ((__u32)len << 8);
 230	pad |= pad << 16;
 231
 232	val = pad;
 233	if (len > num*4)
 234		len = num * 4;
 235	for (i = 0; i < len; i++) {
 236		if ((i % 4) == 0)
 237			val = pad;
 238		val = msg[i] + (val << 8);
 239		if ((i % 4) == 3) {
 240			*buf++ = val;
 241			val = pad;
 242			num--;
 243		}
 244	}
 245	if (--num >= 0)
 246		*buf++ = val;
 247	while (--num >= 0)
 248		*buf++ = pad;
 249}
 250
 251static void ocfs2_dx_dir_name_hash(struct inode *dir, const char *name, int len,
 252				   struct ocfs2_dx_hinfo *hinfo)
 253{
 254	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 255	const char	*p;
 256	__u32		in[8], buf[4];
 257
 258	/*
 259	 * XXX: Is this really necessary, if the index is never looked
 260	 * at by readdir? Is a hash value of '0' a bad idea?
 261	 */
 262	if ((len == 1 && !strncmp(".", name, 1)) ||
 263	    (len == 2 && !strncmp("..", name, 2))) {
 264		buf[0] = buf[1] = 0;
 265		goto out;
 266	}
 267
 268#ifdef OCFS2_DEBUG_DX_DIRS
 269	/*
 270	 * This makes it very easy to debug indexing problems. We
 271	 * should never allow this to be selected without hand editing
 272	 * this file though.
 273	 */
 274	buf[0] = buf[1] = len;
 275	goto out;
 276#endif
 277
 278	memcpy(buf, osb->osb_dx_seed, sizeof(buf));
 279
 280	p = name;
 281	while (len > 0) {
 282		str2hashbuf(p, len, in, 4);
 283		TEA_transform(buf, in);
 284		len -= 16;
 285		p += 16;
 286	}
 287
 288out:
 289	hinfo->major_hash = buf[0];
 290	hinfo->minor_hash = buf[1];
 291}
 292
 293/*
 294 * bh passed here can be an inode block or a dir data block, depending
 295 * on the inode inline data flag.
 296 */
 297static int ocfs2_check_dir_entry(struct inode *dir,
 298				 struct ocfs2_dir_entry *de,
 299				 struct buffer_head *bh,
 300				 char *buf,
 301				 unsigned int size,
 302				 unsigned long offset)
 303{
 304	const char *error_msg = NULL;
 305	unsigned long next_offset;
 306	int rlen;
 307
 308	if (offset > size - OCFS2_DIR_REC_LEN(1)) {
 309		/* Dirent is (maybe partially) beyond the buffer
 310		 * boundaries so touching 'de' members is unsafe.
 311		 */
 312		mlog(ML_ERROR, "directory entry (#%llu: offset=%lu) "
 313		     "too close to end or out-of-bounds",
 314		     (unsigned long long)OCFS2_I(dir)->ip_blkno, offset);
 315		return 0;
 316	}
 317
 318	rlen = le16_to_cpu(de->rec_len);
 319	next_offset = ((char *) de - buf) + rlen;
 320
 321	if (unlikely(rlen < OCFS2_DIR_REC_LEN(1)))
 322		error_msg = "rec_len is smaller than minimal";
 323	else if (unlikely(rlen % 4 != 0))
 324		error_msg = "rec_len % 4 != 0";
 325	else if (unlikely(rlen < OCFS2_DIR_REC_LEN(de->name_len)))
 326		error_msg = "rec_len is too small for name_len";
 327	else if (unlikely(next_offset > size))
 328		error_msg = "directory entry overrun";
 329	else if (unlikely(next_offset > size - OCFS2_DIR_REC_LEN(1)) &&
 330		 next_offset != size)
 331		error_msg = "directory entry too close to end";
 332
 333	if (unlikely(error_msg != NULL))
 334		mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
 335		     "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
 336		     (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
 337		     offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
 338		     de->name_len);
 339
 340	return error_msg == NULL ? 1 : 0;
 341}
 342
 343static inline int ocfs2_match(int len,
 344			      const char * const name,
 345			      struct ocfs2_dir_entry *de)
 346{
 347	if (len != de->name_len)
 348		return 0;
 349	if (!de->inode)
 350		return 0;
 351	return !memcmp(name, de->name, len);
 352}
 353
 354/*
 355 * Returns 0 if not found, -1 on failure, and 1 on success
 356 */
 357static inline int ocfs2_search_dirblock(struct buffer_head *bh,
 358					struct inode *dir,
 359					const char *name, int namelen,
 360					unsigned long offset,
 361					char *first_de,
 362					unsigned int bytes,
 363					struct ocfs2_dir_entry **res_dir)
 364{
 365	struct ocfs2_dir_entry *de;
 366	char *dlimit, *de_buf;
 367	int de_len;
 368	int ret = 0;
 369
 370	de_buf = first_de;
 371	dlimit = de_buf + bytes;
 372
 373	while (de_buf < dlimit - OCFS2_DIR_MEMBER_LEN) {
 374		/* this code is executed quadratically often */
 375		/* do minimal checking `by hand' */
 376
 377		de = (struct ocfs2_dir_entry *) de_buf;
 378
 379		if (de->name + namelen <= dlimit &&
 380		    ocfs2_match(namelen, name, de)) {
 381			/* found a match - just to be sure, do a full check */
 382			if (!ocfs2_check_dir_entry(dir, de, bh, first_de,
 383						   bytes, offset)) {
 384				ret = -1;
 385				goto bail;
 386			}
 387			*res_dir = de;
 388			ret = 1;
 389			goto bail;
 390		}
 391
 392		/* prevent looping on a bad block */
 393		de_len = le16_to_cpu(de->rec_len);
 394		if (de_len <= 0) {
 395			ret = -1;
 396			goto bail;
 397		}
 398
 399		de_buf += de_len;
 400		offset += de_len;
 401	}
 402
 403bail:
 404	trace_ocfs2_search_dirblock(ret);
 405	return ret;
 406}
 407
 408static struct buffer_head *ocfs2_find_entry_id(const char *name,
 409					       int namelen,
 410					       struct inode *dir,
 411					       struct ocfs2_dir_entry **res_dir)
 412{
 413	int ret, found;
 414	struct buffer_head *di_bh = NULL;
 415	struct ocfs2_dinode *di;
 416	struct ocfs2_inline_data *data;
 417
 418	ret = ocfs2_read_inode_block(dir, &di_bh);
 419	if (ret) {
 420		mlog_errno(ret);
 421		goto out;
 422	}
 423
 424	di = (struct ocfs2_dinode *)di_bh->b_data;
 425	data = &di->id2.i_data;
 426
 427	found = ocfs2_search_dirblock(di_bh, dir, name, namelen, 0,
 428				      data->id_data, i_size_read(dir), res_dir);
 429	if (found == 1)
 430		return di_bh;
 431
 432	brelse(di_bh);
 433out:
 434	return NULL;
 435}
 436
 437static int ocfs2_validate_dir_block(struct super_block *sb,
 438				    struct buffer_head *bh)
 439{
 440	int rc;
 441	struct ocfs2_dir_block_trailer *trailer =
 442		ocfs2_trailer_from_bh(bh, sb);
 443
 444
 445	/*
 446	 * We don't validate dirents here, that's handled
 447	 * in-place when the code walks them.
 448	 */
 449	trace_ocfs2_validate_dir_block((unsigned long long)bh->b_blocknr);
 450
 451	BUG_ON(!buffer_uptodate(bh));
 452
 453	/*
 454	 * If the ecc fails, we return the error but otherwise
 455	 * leave the filesystem running.  We know any error is
 456	 * local to this block.
 457	 *
 458	 * Note that we are safe to call this even if the directory
 459	 * doesn't have a trailer.  Filesystems without metaecc will do
 460	 * nothing, and filesystems with it will have one.
 461	 */
 462	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &trailer->db_check);
 463	if (rc)
 464		mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
 465		     (unsigned long long)bh->b_blocknr);
 466
 467	return rc;
 468}
 469
 470/*
 471 * Validate a directory trailer.
 472 *
 473 * We check the trailer here rather than in ocfs2_validate_dir_block()
 474 * because that function doesn't have the inode to test.
 475 */
 476static int ocfs2_check_dir_trailer(struct inode *dir, struct buffer_head *bh)
 477{
 478	int rc = 0;
 479	struct ocfs2_dir_block_trailer *trailer;
 480
 481	trailer = ocfs2_trailer_from_bh(bh, dir->i_sb);
 482	if (!OCFS2_IS_VALID_DIR_TRAILER(trailer)) {
 483		rc = ocfs2_error(dir->i_sb,
 484				 "Invalid dirblock #%llu: signature = %.*s\n",
 485				 (unsigned long long)bh->b_blocknr, 7,
 486				 trailer->db_signature);
 487		goto out;
 488	}
 489	if (le64_to_cpu(trailer->db_blkno) != bh->b_blocknr) {
 490		rc = ocfs2_error(dir->i_sb,
 491				 "Directory block #%llu has an invalid db_blkno of %llu\n",
 492				 (unsigned long long)bh->b_blocknr,
 493				 (unsigned long long)le64_to_cpu(trailer->db_blkno));
 494		goto out;
 495	}
 496	if (le64_to_cpu(trailer->db_parent_dinode) !=
 497	    OCFS2_I(dir)->ip_blkno) {
 498		rc = ocfs2_error(dir->i_sb,
 499				 "Directory block #%llu on dinode #%llu has an invalid parent_dinode of %llu\n",
 500				 (unsigned long long)bh->b_blocknr,
 501				 (unsigned long long)OCFS2_I(dir)->ip_blkno,
 502				 (unsigned long long)le64_to_cpu(trailer->db_blkno));
 503		goto out;
 504	}
 505out:
 506	return rc;
 507}
 508
 509/*
 510 * This function forces all errors to -EIO for consistency with its
 511 * predecessor, ocfs2_bread().  We haven't audited what returning the
 512 * real error codes would do to callers.  We log the real codes with
 513 * mlog_errno() before we squash them.
 514 */
 515static int ocfs2_read_dir_block(struct inode *inode, u64 v_block,
 516				struct buffer_head **bh, int flags)
 517{
 518	int rc = 0;
 519	struct buffer_head *tmp = *bh;
 520
 521	rc = ocfs2_read_virt_blocks(inode, v_block, 1, &tmp, flags,
 522				    ocfs2_validate_dir_block);
 523	if (rc) {
 524		mlog_errno(rc);
 525		goto out;
 526	}
 527
 528	if (!(flags & OCFS2_BH_READAHEAD) &&
 529	    ocfs2_supports_dir_trailer(inode)) {
 530		rc = ocfs2_check_dir_trailer(inode, tmp);
 531		if (rc) {
 532			if (!*bh)
 533				brelse(tmp);
 534			mlog_errno(rc);
 535			goto out;
 536		}
 537	}
 538
 539	/* If ocfs2_read_virt_blocks() got us a new bh, pass it up. */
 540	if (!*bh)
 541		*bh = tmp;
 542
 543out:
 544	return rc ? -EIO : 0;
 545}
 546
 547/*
 548 * Read the block at 'phys' which belongs to this directory
 549 * inode. This function does no virtual->physical block translation -
 550 * what's passed in is assumed to be a valid directory block.
 551 */
 552static int ocfs2_read_dir_block_direct(struct inode *dir, u64 phys,
 553				       struct buffer_head **bh)
 554{
 555	int ret;
 556	struct buffer_head *tmp = *bh;
 557
 558	ret = ocfs2_read_block(INODE_CACHE(dir), phys, &tmp,
 559			       ocfs2_validate_dir_block);
 560	if (ret) {
 561		mlog_errno(ret);
 562		goto out;
 563	}
 564
 565	if (ocfs2_supports_dir_trailer(dir)) {
 566		ret = ocfs2_check_dir_trailer(dir, tmp);
 567		if (ret) {
 568			if (!*bh)
 569				brelse(tmp);
 570			mlog_errno(ret);
 571			goto out;
 572		}
 573	}
 574
 575	if (!ret && !*bh)
 576		*bh = tmp;
 577out:
 578	return ret;
 579}
 580
 581static int ocfs2_validate_dx_root(struct super_block *sb,
 582				  struct buffer_head *bh)
 583{
 584	int ret;
 585	struct ocfs2_dx_root_block *dx_root;
 586
 587	BUG_ON(!buffer_uptodate(bh));
 588
 589	dx_root = (struct ocfs2_dx_root_block *) bh->b_data;
 590
 591	ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_root->dr_check);
 592	if (ret) {
 593		mlog(ML_ERROR,
 594		     "Checksum failed for dir index root block %llu\n",
 595		     (unsigned long long)bh->b_blocknr);
 596		return ret;
 597	}
 598
 599	if (!OCFS2_IS_VALID_DX_ROOT(dx_root)) {
 600		ret = ocfs2_error(sb,
 601				  "Dir Index Root # %llu has bad signature %.*s\n",
 602				  (unsigned long long)le64_to_cpu(dx_root->dr_blkno),
 603				  7, dx_root->dr_signature);
 604	}
 605
 606	return ret;
 607}
 608
 609static int ocfs2_read_dx_root(struct inode *dir, struct ocfs2_dinode *di,
 610			      struct buffer_head **dx_root_bh)
 611{
 612	int ret;
 613	u64 blkno = le64_to_cpu(di->i_dx_root);
 614	struct buffer_head *tmp = *dx_root_bh;
 615
 616	ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
 617			       ocfs2_validate_dx_root);
 618
 619	/* If ocfs2_read_block() got us a new bh, pass it up. */
 620	if (!ret && !*dx_root_bh)
 621		*dx_root_bh = tmp;
 622
 623	return ret;
 624}
 625
 626static int ocfs2_validate_dx_leaf(struct super_block *sb,
 627				  struct buffer_head *bh)
 628{
 629	int ret;
 630	struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)bh->b_data;
 631
 632	BUG_ON(!buffer_uptodate(bh));
 633
 634	ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_leaf->dl_check);
 635	if (ret) {
 636		mlog(ML_ERROR,
 637		     "Checksum failed for dir index leaf block %llu\n",
 638		     (unsigned long long)bh->b_blocknr);
 639		return ret;
 640	}
 641
 642	if (!OCFS2_IS_VALID_DX_LEAF(dx_leaf)) {
 643		ret = ocfs2_error(sb, "Dir Index Leaf has bad signature %.*s\n",
 644				  7, dx_leaf->dl_signature);
 645	}
 646
 647	return ret;
 648}
 649
 650static int ocfs2_read_dx_leaf(struct inode *dir, u64 blkno,
 651			      struct buffer_head **dx_leaf_bh)
 652{
 653	int ret;
 654	struct buffer_head *tmp = *dx_leaf_bh;
 655
 656	ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
 657			       ocfs2_validate_dx_leaf);
 658
 659	/* If ocfs2_read_block() got us a new bh, pass it up. */
 660	if (!ret && !*dx_leaf_bh)
 661		*dx_leaf_bh = tmp;
 662
 663	return ret;
 664}
 665
 666/*
 667 * Read a series of dx_leaf blocks. This expects all buffer_head
 668 * pointers to be NULL on function entry.
 669 */
 670static int ocfs2_read_dx_leaves(struct inode *dir, u64 start, int num,
 671				struct buffer_head **dx_leaf_bhs)
 672{
 673	int ret;
 674
 675	ret = ocfs2_read_blocks(INODE_CACHE(dir), start, num, dx_leaf_bhs, 0,
 676				ocfs2_validate_dx_leaf);
 677	if (ret)
 678		mlog_errno(ret);
 679
 680	return ret;
 681}
 682
 683static struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen,
 684					       struct inode *dir,
 685					       struct ocfs2_dir_entry **res_dir)
 686{
 687	struct super_block *sb;
 688	struct buffer_head *bh_use[NAMEI_RA_SIZE];
 689	struct buffer_head *bh, *ret = NULL;
 690	unsigned long start, block, b;
 691	int ra_max = 0;		/* Number of bh's in the readahead
 692				   buffer, bh_use[] */
 693	int ra_ptr = 0;		/* Current index into readahead
 694				   buffer */
 695	int num = 0;
 696	int nblocks, i;
 697
 698	sb = dir->i_sb;
 699
 700	nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
 701	start = OCFS2_I(dir)->ip_dir_start_lookup;
 702	if (start >= nblocks)
 703		start = 0;
 704	block = start;
 705
 706restart:
 707	do {
 708		/*
 709		 * We deal with the read-ahead logic here.
 710		 */
 711		if (ra_ptr >= ra_max) {
 712			/* Refill the readahead buffer */
 713			ra_ptr = 0;
 714			b = block;
 715			for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 716				/*
 717				 * Terminate if we reach the end of the
 718				 * directory and must wrap, or if our
 719				 * search has finished at this block.
 720				 */
 721				if (b >= nblocks || (num && block == start)) {
 722					bh_use[ra_max] = NULL;
 723					break;
 724				}
 725				num++;
 726
 727				bh = NULL;
 728				ocfs2_read_dir_block(dir, b++, &bh,
 729							   OCFS2_BH_READAHEAD);
 730				bh_use[ra_max] = bh;
 731			}
 732		}
 733		if ((bh = bh_use[ra_ptr++]) == NULL)
 734			goto next;
 735		if (ocfs2_read_dir_block(dir, block, &bh, 0)) {
 736			/* read error, skip block & hope for the best.
 737			 * ocfs2_read_dir_block() has released the bh. */
 738			mlog(ML_ERROR, "reading directory %llu, "
 739				    "offset %lu\n",
 740				    (unsigned long long)OCFS2_I(dir)->ip_blkno,
 741				    block);
 742			goto next;
 743		}
 744		i = ocfs2_search_dirblock(bh, dir, name, namelen,
 745					  block << sb->s_blocksize_bits,
 746					  bh->b_data, sb->s_blocksize,
 747					  res_dir);
 748		if (i == 1) {
 749			OCFS2_I(dir)->ip_dir_start_lookup = block;
 750			ret = bh;
 751			goto cleanup_and_exit;
 752		} else {
 753			brelse(bh);
 754			if (i < 0)
 755				goto cleanup_and_exit;
 756		}
 757	next:
 758		if (++block >= nblocks)
 759			block = 0;
 760	} while (block != start);
 761
 762	/*
 763	 * If the directory has grown while we were searching, then
 764	 * search the last part of the directory before giving up.
 765	 */
 766	block = nblocks;
 767	nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
 768	if (block < nblocks) {
 769		start = 0;
 770		goto restart;
 771	}
 772
 773cleanup_and_exit:
 774	/* Clean up the read-ahead blocks */
 775	for (; ra_ptr < ra_max; ra_ptr++)
 776		brelse(bh_use[ra_ptr]);
 777
 778	trace_ocfs2_find_entry_el(ret);
 779	return ret;
 780}
 781
 782static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
 783				   struct ocfs2_extent_list *el,
 784				   u32 major_hash,
 785				   u32 *ret_cpos,
 786				   u64 *ret_phys_blkno,
 787				   unsigned int *ret_clen)
 788{
 789	int ret = 0, i, found;
 790	struct buffer_head *eb_bh = NULL;
 791	struct ocfs2_extent_block *eb;
 792	struct ocfs2_extent_rec *rec = NULL;
 793
 794	if (le16_to_cpu(el->l_count) !=
 795	    ocfs2_extent_recs_per_dx_root(inode->i_sb)) {
 796		ret = ocfs2_error(inode->i_sb,
 797				  "Inode %lu has invalid extent list length %u\n",
 798				  inode->i_ino, le16_to_cpu(el->l_count));
 799		goto out;
 800	}
 801
 802	if (el->l_tree_depth) {
 803		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, major_hash,
 804				      &eb_bh);
 805		if (ret) {
 806			mlog_errno(ret);
 807			goto out;
 808		}
 809
 810		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
 811		el = &eb->h_list;
 812
 813		if (el->l_tree_depth) {
 814			ret = ocfs2_error(inode->i_sb,
 815					  "Inode %lu has non zero tree depth in btree tree block %llu\n",
 816					  inode->i_ino,
 817					  (unsigned long long)eb_bh->b_blocknr);
 818			goto out;
 819		}
 820	}
 821
 822	if (le16_to_cpu(el->l_next_free_rec) == 0) {
 823		ret = ocfs2_error(inode->i_sb,
 824				  "Inode %lu has empty extent list at depth %u\n",
 825				  inode->i_ino,
 826				  le16_to_cpu(el->l_tree_depth));
 827		goto out;
 828	}
 829
 830	found = 0;
 831	for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
 832		rec = &el->l_recs[i];
 833
 834		if (le32_to_cpu(rec->e_cpos) <= major_hash) {
 835			found = 1;
 836			break;
 837		}
 838	}
 839
 840	if (!found) {
 841		ret = ocfs2_error(inode->i_sb,
 842				  "Inode %lu has bad extent record (%u, %u, 0) in btree\n",
 843				  inode->i_ino,
 844				  le32_to_cpu(rec->e_cpos),
 845				  ocfs2_rec_clusters(el, rec));
 846		goto out;
 847	}
 848
 849	if (ret_phys_blkno)
 850		*ret_phys_blkno = le64_to_cpu(rec->e_blkno);
 851	if (ret_cpos)
 852		*ret_cpos = le32_to_cpu(rec->e_cpos);
 853	if (ret_clen)
 854		*ret_clen = le16_to_cpu(rec->e_leaf_clusters);
 855
 856out:
 857	brelse(eb_bh);
 858	return ret;
 859}
 860
 861/*
 862 * Returns the block index, from the start of the cluster which this
 863 * hash belongs too.
 864 */
 865static inline unsigned int __ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
 866						   u32 minor_hash)
 867{
 868	return minor_hash & osb->osb_dx_mask;
 869}
 870
 871static inline unsigned int ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
 872					  struct ocfs2_dx_hinfo *hinfo)
 873{
 874	return __ocfs2_dx_dir_hash_idx(osb, hinfo->minor_hash);
 875}
 876
 877static int ocfs2_dx_dir_lookup(struct inode *inode,
 878			       struct ocfs2_extent_list *el,
 879			       struct ocfs2_dx_hinfo *hinfo,
 880			       u32 *ret_cpos,
 881			       u64 *ret_phys_blkno)
 882{
 883	int ret = 0;
 884	unsigned int cend, clen;
 885	u32 cpos;
 886	u64 blkno;
 887	u32 name_hash = hinfo->major_hash;
 888
 889	ret = ocfs2_dx_dir_lookup_rec(inode, el, name_hash, &cpos, &blkno,
 890				      &clen);
 891	if (ret) {
 892		mlog_errno(ret);
 893		goto out;
 894	}
 895
 896	cend = cpos + clen;
 897	if (name_hash >= cend) {
 898		/* We want the last cluster */
 899		blkno += ocfs2_clusters_to_blocks(inode->i_sb, clen - 1);
 900		cpos += clen - 1;
 901	} else {
 902		blkno += ocfs2_clusters_to_blocks(inode->i_sb,
 903						  name_hash - cpos);
 904		cpos = name_hash;
 905	}
 906
 907	/*
 908	 * We now have the cluster which should hold our entry. To
 909	 * find the exact block from the start of the cluster to
 910	 * search, we take the lower bits of the hash.
 911	 */
 912	blkno += ocfs2_dx_dir_hash_idx(OCFS2_SB(inode->i_sb), hinfo);
 913
 914	if (ret_phys_blkno)
 915		*ret_phys_blkno = blkno;
 916	if (ret_cpos)
 917		*ret_cpos = cpos;
 918
 919out:
 920
 921	return ret;
 922}
 923
 924static int ocfs2_dx_dir_search(const char *name, int namelen,
 925			       struct inode *dir,
 926			       struct ocfs2_dx_root_block *dx_root,
 927			       struct ocfs2_dir_lookup_result *res)
 928{
 929	int ret, i, found;
 930	u64 phys;
 931	struct buffer_head *dx_leaf_bh = NULL;
 932	struct ocfs2_dx_leaf *dx_leaf;
 933	struct ocfs2_dx_entry *dx_entry = NULL;
 934	struct buffer_head *dir_ent_bh = NULL;
 935	struct ocfs2_dir_entry *dir_ent = NULL;
 936	struct ocfs2_dx_hinfo *hinfo = &res->dl_hinfo;
 937	struct ocfs2_extent_list *dr_el;
 938	struct ocfs2_dx_entry_list *entry_list;
 939
 940	ocfs2_dx_dir_name_hash(dir, name, namelen, &res->dl_hinfo);
 941
 942	if (ocfs2_dx_root_inline(dx_root)) {
 943		entry_list = &dx_root->dr_entries;
 944		goto search;
 945	}
 946
 947	dr_el = &dx_root->dr_list;
 948
 949	ret = ocfs2_dx_dir_lookup(dir, dr_el, hinfo, NULL, &phys);
 950	if (ret) {
 951		mlog_errno(ret);
 952		goto out;
 953	}
 954
 955	trace_ocfs2_dx_dir_search((unsigned long long)OCFS2_I(dir)->ip_blkno,
 956				  namelen, name, hinfo->major_hash,
 957				  hinfo->minor_hash, (unsigned long long)phys);
 958
 959	ret = ocfs2_read_dx_leaf(dir, phys, &dx_leaf_bh);
 960	if (ret) {
 961		mlog_errno(ret);
 962		goto out;
 963	}
 964
 965	dx_leaf = (struct ocfs2_dx_leaf *) dx_leaf_bh->b_data;
 966
 967	trace_ocfs2_dx_dir_search_leaf_info(
 968			le16_to_cpu(dx_leaf->dl_list.de_num_used),
 969			le16_to_cpu(dx_leaf->dl_list.de_count));
 970
 971	entry_list = &dx_leaf->dl_list;
 972
 973search:
 974	/*
 975	 * Empty leaf is legal, so no need to check for that.
 976	 */
 977	found = 0;
 978	for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
 979		dx_entry = &entry_list->de_entries[i];
 980
 981		if (hinfo->major_hash != le32_to_cpu(dx_entry->dx_major_hash)
 982		    || hinfo->minor_hash != le32_to_cpu(dx_entry->dx_minor_hash))
 983			continue;
 984
 985		/*
 986		 * Search unindexed leaf block now. We're not
 987		 * guaranteed to find anything.
 988		 */
 989		ret = ocfs2_read_dir_block_direct(dir,
 990					  le64_to_cpu(dx_entry->dx_dirent_blk),
 991					  &dir_ent_bh);
 992		if (ret) {
 993			mlog_errno(ret);
 994			goto out;
 995		}
 996
 997		/*
 998		 * XXX: We should check the unindexed block here,
 999		 * before using it.
1000		 */
1001
1002		found = ocfs2_search_dirblock(dir_ent_bh, dir, name, namelen,
1003					      0, dir_ent_bh->b_data,
1004					      dir->i_sb->s_blocksize, &dir_ent);
1005		if (found == 1)
1006			break;
1007
1008		if (found == -1) {
1009			/* This means we found a bad directory entry. */
1010			ret = -EIO;
1011			mlog_errno(ret);
1012			goto out;
1013		}
1014
1015		brelse(dir_ent_bh);
1016		dir_ent_bh = NULL;
1017	}
1018
1019	if (found <= 0) {
1020		ret = -ENOENT;
1021		goto out;
1022	}
1023
1024	res->dl_leaf_bh = dir_ent_bh;
1025	res->dl_entry = dir_ent;
1026	res->dl_dx_leaf_bh = dx_leaf_bh;
1027	res->dl_dx_entry = dx_entry;
1028
1029	ret = 0;
1030out:
1031	if (ret) {
1032		brelse(dx_leaf_bh);
1033		brelse(dir_ent_bh);
1034	}
1035	return ret;
1036}
1037
1038static int ocfs2_find_entry_dx(const char *name, int namelen,
1039			       struct inode *dir,
1040			       struct ocfs2_dir_lookup_result *lookup)
1041{
1042	int ret;
1043	struct buffer_head *di_bh = NULL;
1044	struct ocfs2_dinode *di;
1045	struct buffer_head *dx_root_bh = NULL;
1046	struct ocfs2_dx_root_block *dx_root;
1047
1048	ret = ocfs2_read_inode_block(dir, &di_bh);
1049	if (ret) {
1050		mlog_errno(ret);
1051		goto out;
1052	}
1053
1054	di = (struct ocfs2_dinode *)di_bh->b_data;
1055
1056	ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
1057	if (ret) {
1058		mlog_errno(ret);
1059		goto out;
1060	}
1061	dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
1062
1063	ret = ocfs2_dx_dir_search(name, namelen, dir, dx_root, lookup);
1064	if (ret) {
1065		if (ret != -ENOENT)
1066			mlog_errno(ret);
1067		goto out;
1068	}
1069
1070	lookup->dl_dx_root_bh = dx_root_bh;
1071	dx_root_bh = NULL;
1072out:
1073	brelse(di_bh);
1074	brelse(dx_root_bh);
1075	return ret;
1076}
1077
1078/*
1079 * Try to find an entry of the provided name within 'dir'.
1080 *
1081 * If nothing was found, -ENOENT is returned. Otherwise, zero is
1082 * returned and the struct 'res' will contain information useful to
1083 * other directory manipulation functions.
1084 *
1085 * Caller can NOT assume anything about the contents of the
1086 * buffer_heads - they are passed back only so that it can be passed
1087 * into any one of the manipulation functions (add entry, delete
1088 * entry, etc). As an example, bh in the extent directory case is a
1089 * data block, in the inline-data case it actually points to an inode,
1090 * in the indexed directory case, multiple buffers are involved.
1091 */
1092int ocfs2_find_entry(const char *name, int namelen,
1093		     struct inode *dir, struct ocfs2_dir_lookup_result *lookup)
1094{
1095	struct buffer_head *bh;
1096	struct ocfs2_dir_entry *res_dir = NULL;
1097	int ret = 0;
1098
1099	if (ocfs2_dir_indexed(dir))
1100		return ocfs2_find_entry_dx(name, namelen, dir, lookup);
1101
1102	if (unlikely(i_size_read(dir) <= 0)) {
1103		ret = -EFSCORRUPTED;
1104		mlog_errno(ret);
1105		goto out;
1106	}
1107	/*
1108	 * The unindexed dir code only uses part of the lookup
1109	 * structure, so there's no reason to push it down further
1110	 * than this.
1111	 */
1112	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1113		if (unlikely(i_size_read(dir) > dir->i_sb->s_blocksize)) {
1114			ret = -EFSCORRUPTED;
1115			mlog_errno(ret);
1116			goto out;
1117		}
1118		bh = ocfs2_find_entry_id(name, namelen, dir, &res_dir);
1119	} else {
1120		bh = ocfs2_find_entry_el(name, namelen, dir, &res_dir);
1121	}
1122
1123	if (bh == NULL)
1124		return -ENOENT;
1125
1126	lookup->dl_leaf_bh = bh;
1127	lookup->dl_entry = res_dir;
1128out:
1129	return ret;
1130}
1131
1132/*
1133 * Update inode number and type of a previously found directory entry.
1134 */
1135int ocfs2_update_entry(struct inode *dir, handle_t *handle,
1136		       struct ocfs2_dir_lookup_result *res,
1137		       struct inode *new_entry_inode)
1138{
1139	int ret;
1140	ocfs2_journal_access_func access = ocfs2_journal_access_db;
1141	struct ocfs2_dir_entry *de = res->dl_entry;
1142	struct buffer_head *de_bh = res->dl_leaf_bh;
1143
1144	/*
1145	 * The same code works fine for both inline-data and extent
1146	 * based directories, so no need to split this up.  The only
1147	 * difference is the journal_access function.
1148	 */
1149
1150	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1151		access = ocfs2_journal_access_di;
1152
1153	ret = access(handle, INODE_CACHE(dir), de_bh,
1154		     OCFS2_JOURNAL_ACCESS_WRITE);
1155	if (ret) {
1156		mlog_errno(ret);
1157		goto out;
1158	}
1159
1160	de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno);
1161	ocfs2_set_de_type(de, new_entry_inode->i_mode);
1162
1163	ocfs2_journal_dirty(handle, de_bh);
1164
1165out:
1166	return ret;
1167}
1168
1169/*
1170 * __ocfs2_delete_entry deletes a directory entry by merging it with the
1171 * previous entry
1172 */
1173static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir,
1174				struct ocfs2_dir_entry *de_del,
1175				struct buffer_head *bh, char *first_de,
1176				unsigned int bytes)
1177{
1178	struct ocfs2_dir_entry *de, *pde;
1179	int i, status = -ENOENT;
1180	ocfs2_journal_access_func access = ocfs2_journal_access_db;
1181
1182	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1183		access = ocfs2_journal_access_di;
1184
1185	i = 0;
1186	pde = NULL;
1187	de = (struct ocfs2_dir_entry *) first_de;
1188	while (i < bytes) {
1189		if (!ocfs2_check_dir_entry(dir, de, bh, first_de, bytes, i)) {
1190			status = -EIO;
1191			mlog_errno(status);
1192			goto bail;
1193		}
1194		if (de == de_del)  {
1195			status = access(handle, INODE_CACHE(dir), bh,
1196					OCFS2_JOURNAL_ACCESS_WRITE);
1197			if (status < 0) {
1198				status = -EIO;
1199				mlog_errno(status);
1200				goto bail;
1201			}
1202			if (pde)
1203				le16_add_cpu(&pde->rec_len,
1204						le16_to_cpu(de->rec_len));
1205			de->inode = 0;
1206			inode_inc_iversion(dir);
1207			ocfs2_journal_dirty(handle, bh);
1208			goto bail;
1209		}
1210		i += le16_to_cpu(de->rec_len);
1211		pde = de;
1212		de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1213	}
1214bail:
1215	return status;
1216}
1217
1218static unsigned int ocfs2_figure_dirent_hole(struct ocfs2_dir_entry *de)
1219{
1220	unsigned int hole;
1221
1222	if (le64_to_cpu(de->inode) == 0)
1223		hole = le16_to_cpu(de->rec_len);
1224	else
1225		hole = le16_to_cpu(de->rec_len) -
1226			OCFS2_DIR_REC_LEN(de->name_len);
1227
1228	return hole;
1229}
1230
1231static int ocfs2_find_max_rec_len(struct super_block *sb,
1232				  struct buffer_head *dirblock_bh)
1233{
1234	int size, this_hole, largest_hole = 0;
1235	char *trailer, *de_buf, *limit, *start = dirblock_bh->b_data;
1236	struct ocfs2_dir_entry *de;
1237
1238	trailer = (char *)ocfs2_trailer_from_bh(dirblock_bh, sb);
1239	size = ocfs2_dir_trailer_blk_off(sb);
1240	limit = start + size;
1241	de_buf = start;
1242	de = (struct ocfs2_dir_entry *)de_buf;
1243	do {
1244		if (de_buf != trailer) {
1245			this_hole = ocfs2_figure_dirent_hole(de);
1246			if (this_hole > largest_hole)
1247				largest_hole = this_hole;
1248		}
1249
1250		de_buf += le16_to_cpu(de->rec_len);
1251		de = (struct ocfs2_dir_entry *)de_buf;
1252	} while (de_buf < limit);
1253
1254	if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
1255		return largest_hole;
1256	return 0;
1257}
1258
1259static void ocfs2_dx_list_remove_entry(struct ocfs2_dx_entry_list *entry_list,
1260				       int index)
1261{
1262	int num_used = le16_to_cpu(entry_list->de_num_used);
1263
1264	if (num_used == 1 || index == (num_used - 1))
1265		goto clear;
1266
1267	memmove(&entry_list->de_entries[index],
1268		&entry_list->de_entries[index + 1],
1269		(num_used - index - 1)*sizeof(struct ocfs2_dx_entry));
1270clear:
1271	num_used--;
1272	memset(&entry_list->de_entries[num_used], 0,
1273	       sizeof(struct ocfs2_dx_entry));
1274	entry_list->de_num_used = cpu_to_le16(num_used);
1275}
1276
1277static int ocfs2_delete_entry_dx(handle_t *handle, struct inode *dir,
1278				 struct ocfs2_dir_lookup_result *lookup)
1279{
1280	int ret, index, max_rec_len, add_to_free_list = 0;
1281	struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
1282	struct buffer_head *leaf_bh = lookup->dl_leaf_bh;
1283	struct ocfs2_dx_leaf *dx_leaf;
1284	struct ocfs2_dx_entry *dx_entry = lookup->dl_dx_entry;
1285	struct ocfs2_dir_block_trailer *trailer;
1286	struct ocfs2_dx_root_block *dx_root;
1287	struct ocfs2_dx_entry_list *entry_list;
1288
1289	/*
1290	 * This function gets a bit messy because we might have to
1291	 * modify the root block, regardless of whether the indexed
1292	 * entries are stored inline.
1293	 */
1294
1295	/*
1296	 * *Only* set 'entry_list' here, based on where we're looking
1297	 * for the indexed entries. Later, we might still want to
1298	 * journal both blocks, based on free list state.
1299	 */
1300	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
1301	if (ocfs2_dx_root_inline(dx_root)) {
1302		entry_list = &dx_root->dr_entries;
1303	} else {
1304		dx_leaf = (struct ocfs2_dx_leaf *) lookup->dl_dx_leaf_bh->b_data;
1305		entry_list = &dx_leaf->dl_list;
1306	}
1307
1308	/* Neither of these are a disk corruption - that should have
1309	 * been caught by lookup, before we got here. */
1310	BUG_ON(le16_to_cpu(entry_list->de_count) <= 0);
1311	BUG_ON(le16_to_cpu(entry_list->de_num_used) <= 0);
1312
1313	index = (char *)dx_entry - (char *)entry_list->de_entries;
1314	index /= sizeof(*dx_entry);
1315
1316	if (index >= le16_to_cpu(entry_list->de_num_used)) {
1317		mlog(ML_ERROR, "Dir %llu: Bad dx_entry ptr idx %d, (%p, %p)\n",
1318		     (unsigned long long)OCFS2_I(dir)->ip_blkno, index,
1319		     entry_list, dx_entry);
1320		return -EIO;
1321	}
1322
1323	/*
1324	 * We know that removal of this dirent will leave enough room
1325	 * for a new one, so add this block to the free list if it
1326	 * isn't already there.
1327	 */
1328	trailer = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
1329	if (trailer->db_free_rec_len == 0)
1330		add_to_free_list = 1;
1331
1332	/*
1333	 * Add the block holding our index into the journal before
1334	 * removing the unindexed entry. If we get an error return
1335	 * from __ocfs2_delete_entry(), then it hasn't removed the
1336	 * entry yet. Likewise, successful return means we *must*
1337	 * remove the indexed entry.
1338	 *
1339	 * We're also careful to journal the root tree block here as
1340	 * the entry count needs to be updated. Also, we might be
1341	 * adding to the start of the free list.
1342	 */
1343	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
1344				      OCFS2_JOURNAL_ACCESS_WRITE);
1345	if (ret) {
1346		mlog_errno(ret);
1347		goto out;
1348	}
1349
1350	if (!ocfs2_dx_root_inline(dx_root)) {
1351		ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
1352					      lookup->dl_dx_leaf_bh,
1353					      OCFS2_JOURNAL_ACCESS_WRITE);
1354		if (ret) {
1355			mlog_errno(ret);
1356			goto out;
1357		}
1358	}
1359
1360	trace_ocfs2_delete_entry_dx((unsigned long long)OCFS2_I(dir)->ip_blkno,
1361				    index);
1362
1363	ret = __ocfs2_delete_entry(handle, dir, lookup->dl_entry,
1364				   leaf_bh, leaf_bh->b_data, leaf_bh->b_size);
1365	if (ret) {
1366		mlog_errno(ret);
1367		goto out;
1368	}
1369
1370	max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, leaf_bh);
1371	trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1372	if (add_to_free_list) {
1373		trailer->db_free_next = dx_root->dr_free_blk;
1374		dx_root->dr_free_blk = cpu_to_le64(leaf_bh->b_blocknr);
1375		ocfs2_journal_dirty(handle, dx_root_bh);
1376	}
1377
1378	/* leaf_bh was journal_accessed for us in __ocfs2_delete_entry */
1379	ocfs2_journal_dirty(handle, leaf_bh);
1380
1381	le32_add_cpu(&dx_root->dr_num_entries, -1);
1382	ocfs2_journal_dirty(handle, dx_root_bh);
1383
1384	ocfs2_dx_list_remove_entry(entry_list, index);
1385
1386	if (!ocfs2_dx_root_inline(dx_root))
1387		ocfs2_journal_dirty(handle, lookup->dl_dx_leaf_bh);
1388
1389out:
1390	return ret;
1391}
1392
1393static inline int ocfs2_delete_entry_id(handle_t *handle,
1394					struct inode *dir,
1395					struct ocfs2_dir_entry *de_del,
1396					struct buffer_head *bh)
1397{
1398	int ret;
1399	struct buffer_head *di_bh = NULL;
1400	struct ocfs2_dinode *di;
1401	struct ocfs2_inline_data *data;
1402
1403	ret = ocfs2_read_inode_block(dir, &di_bh);
1404	if (ret) {
1405		mlog_errno(ret);
1406		goto out;
1407	}
1408
1409	di = (struct ocfs2_dinode *)di_bh->b_data;
1410	data = &di->id2.i_data;
1411
1412	ret = __ocfs2_delete_entry(handle, dir, de_del, bh, data->id_data,
1413				   i_size_read(dir));
1414
1415	brelse(di_bh);
1416out:
1417	return ret;
1418}
1419
1420static inline int ocfs2_delete_entry_el(handle_t *handle,
1421					struct inode *dir,
1422					struct ocfs2_dir_entry *de_del,
1423					struct buffer_head *bh)
1424{
1425	return __ocfs2_delete_entry(handle, dir, de_del, bh, bh->b_data,
1426				    bh->b_size);
1427}
1428
1429/*
1430 * Delete a directory entry. Hide the details of directory
1431 * implementation from the caller.
1432 */
1433int ocfs2_delete_entry(handle_t *handle,
1434		       struct inode *dir,
1435		       struct ocfs2_dir_lookup_result *res)
1436{
1437	if (ocfs2_dir_indexed(dir))
1438		return ocfs2_delete_entry_dx(handle, dir, res);
1439
1440	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1441		return ocfs2_delete_entry_id(handle, dir, res->dl_entry,
1442					     res->dl_leaf_bh);
1443
1444	return ocfs2_delete_entry_el(handle, dir, res->dl_entry,
1445				     res->dl_leaf_bh);
1446}
1447
1448/*
1449 * Check whether 'de' has enough room to hold an entry of
1450 * 'new_rec_len' bytes.
1451 */
1452static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de,
1453					 unsigned int new_rec_len)
1454{
1455	unsigned int de_really_used;
1456
1457	/* Check whether this is an empty record with enough space */
1458	if (le64_to_cpu(de->inode) == 0 &&
1459	    le16_to_cpu(de->rec_len) >= new_rec_len)
1460		return 1;
1461
1462	/*
1463	 * Record might have free space at the end which we can
1464	 * use.
1465	 */
1466	de_really_used = OCFS2_DIR_REC_LEN(de->name_len);
1467	if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len))
1468	    return 1;
1469
1470	return 0;
1471}
1472
1473static void ocfs2_dx_dir_leaf_insert_tail(struct ocfs2_dx_leaf *dx_leaf,
1474					  struct ocfs2_dx_entry *dx_new_entry)
1475{
1476	int i;
1477
1478	i = le16_to_cpu(dx_leaf->dl_list.de_num_used);
1479	dx_leaf->dl_list.de_entries[i] = *dx_new_entry;
1480
1481	le16_add_cpu(&dx_leaf->dl_list.de_num_used, 1);
1482}
1483
1484static void ocfs2_dx_entry_list_insert(struct ocfs2_dx_entry_list *entry_list,
1485				       struct ocfs2_dx_hinfo *hinfo,
1486				       u64 dirent_blk)
1487{
1488	int i;
1489	struct ocfs2_dx_entry *dx_entry;
1490
1491	i = le16_to_cpu(entry_list->de_num_used);
1492	dx_entry = &entry_list->de_entries[i];
1493
1494	memset(dx_entry, 0, sizeof(*dx_entry));
1495	dx_entry->dx_major_hash = cpu_to_le32(hinfo->major_hash);
1496	dx_entry->dx_minor_hash = cpu_to_le32(hinfo->minor_hash);
1497	dx_entry->dx_dirent_blk = cpu_to_le64(dirent_blk);
1498
1499	le16_add_cpu(&entry_list->de_num_used, 1);
1500}
1501
1502static int __ocfs2_dx_dir_leaf_insert(struct inode *dir, handle_t *handle,
1503				      struct ocfs2_dx_hinfo *hinfo,
1504				      u64 dirent_blk,
1505				      struct buffer_head *dx_leaf_bh)
1506{
1507	int ret;
1508	struct ocfs2_dx_leaf *dx_leaf;
1509
1510	ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
1511				      OCFS2_JOURNAL_ACCESS_WRITE);
1512	if (ret) {
1513		mlog_errno(ret);
1514		goto out;
1515	}
1516
1517	dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
1518	ocfs2_dx_entry_list_insert(&dx_leaf->dl_list, hinfo, dirent_blk);
1519	ocfs2_journal_dirty(handle, dx_leaf_bh);
1520
1521out:
1522	return ret;
1523}
1524
1525static void ocfs2_dx_inline_root_insert(struct inode *dir, handle_t *handle,
1526					struct ocfs2_dx_hinfo *hinfo,
1527					u64 dirent_blk,
1528					struct ocfs2_dx_root_block *dx_root)
1529{
1530	ocfs2_dx_entry_list_insert(&dx_root->dr_entries, hinfo, dirent_blk);
1531}
1532
1533static int ocfs2_dx_dir_insert(struct inode *dir, handle_t *handle,
1534			       struct ocfs2_dir_lookup_result *lookup)
1535{
1536	int ret = 0;
1537	struct ocfs2_dx_root_block *dx_root;
1538	struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
1539
1540	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
1541				      OCFS2_JOURNAL_ACCESS_WRITE);
1542	if (ret) {
1543		mlog_errno(ret);
1544		goto out;
1545	}
1546
1547	dx_root = (struct ocfs2_dx_root_block *)lookup->dl_dx_root_bh->b_data;
1548	if (ocfs2_dx_root_inline(dx_root)) {
1549		ocfs2_dx_inline_root_insert(dir, handle,
1550					    &lookup->dl_hinfo,
1551					    lookup->dl_leaf_bh->b_blocknr,
1552					    dx_root);
1553	} else {
1554		ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &lookup->dl_hinfo,
1555						 lookup->dl_leaf_bh->b_blocknr,
1556						 lookup->dl_dx_leaf_bh);
1557		if (ret)
1558			goto out;
1559	}
1560
1561	le32_add_cpu(&dx_root->dr_num_entries, 1);
1562	ocfs2_journal_dirty(handle, dx_root_bh);
1563
1564out:
1565	return ret;
1566}
1567
1568static void ocfs2_remove_block_from_free_list(struct inode *dir,
1569				       handle_t *handle,
1570				       struct ocfs2_dir_lookup_result *lookup)
1571{
1572	struct ocfs2_dir_block_trailer *trailer, *prev;
1573	struct ocfs2_dx_root_block *dx_root;
1574	struct buffer_head *bh;
1575
1576	trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1577
1578	if (ocfs2_free_list_at_root(lookup)) {
1579		bh = lookup->dl_dx_root_bh;
1580		dx_root = (struct ocfs2_dx_root_block *)bh->b_data;
1581		dx_root->dr_free_blk = trailer->db_free_next;
1582	} else {
1583		bh = lookup->dl_prev_leaf_bh;
1584		prev = ocfs2_trailer_from_bh(bh, dir->i_sb);
1585		prev->db_free_next = trailer->db_free_next;
1586	}
1587
1588	trailer->db_free_rec_len = cpu_to_le16(0);
1589	trailer->db_free_next = cpu_to_le64(0);
1590
1591	ocfs2_journal_dirty(handle, bh);
1592	ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1593}
1594
1595/*
1596 * This expects that a journal write has been reserved on
1597 * lookup->dl_prev_leaf_bh or lookup->dl_dx_root_bh
1598 */
1599static void ocfs2_recalc_free_list(struct inode *dir, handle_t *handle,
1600				   struct ocfs2_dir_lookup_result *lookup)
1601{
1602	int max_rec_len;
1603	struct ocfs2_dir_block_trailer *trailer;
1604
1605	/* Walk dl_leaf_bh to figure out what the new free rec_len is. */
1606	max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, lookup->dl_leaf_bh);
1607	if (max_rec_len) {
1608		/*
1609		 * There's still room in this block, so no need to remove it
1610		 * from the free list. In this case, we just want to update
1611		 * the rec len accounting.
1612		 */
1613		trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
1614		trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
1615		ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
1616	} else {
1617		ocfs2_remove_block_from_free_list(dir, handle, lookup);
1618	}
1619}
1620
1621/* we don't always have a dentry for what we want to add, so people
1622 * like orphan dir can call this instead.
1623 *
1624 * The lookup context must have been filled from
1625 * ocfs2_prepare_dir_for_insert.
1626 */
1627int __ocfs2_add_entry(handle_t *handle,
1628		      struct inode *dir,
1629		      const char *name, int namelen,
1630		      struct inode *inode, u64 blkno,
1631		      struct buffer_head *parent_fe_bh,
1632		      struct ocfs2_dir_lookup_result *lookup)
1633{
1634	unsigned long offset;
1635	unsigned short rec_len;
1636	struct ocfs2_dir_entry *de, *de1;
1637	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_fe_bh->b_data;
1638	struct super_block *sb = dir->i_sb;
1639	int retval;
1640	unsigned int size = sb->s_blocksize;
1641	struct buffer_head *insert_bh = lookup->dl_leaf_bh;
1642	char *data_start = insert_bh->b_data;
1643
1644	if (ocfs2_dir_indexed(dir)) {
1645		struct buffer_head *bh;
1646
1647		/*
1648		 * An indexed dir may require that we update the free space
1649		 * list. Reserve a write to the previous node in the list so
1650		 * that we don't fail later.
1651		 *
1652		 * XXX: This can be either a dx_root_block, or an unindexed
1653		 * directory tree leaf block.
1654		 */
1655		if (ocfs2_free_list_at_root(lookup)) {
1656			bh = lookup->dl_dx_root_bh;
1657			retval = ocfs2_journal_access_dr(handle,
1658						 INODE_CACHE(dir), bh,
1659						 OCFS2_JOURNAL_ACCESS_WRITE);
1660		} else {
1661			bh = lookup->dl_prev_leaf_bh;
1662			retval = ocfs2_journal_access_db(handle,
1663						 INODE_CACHE(dir), bh,
1664						 OCFS2_JOURNAL_ACCESS_WRITE);
1665		}
1666		if (retval) {
1667			mlog_errno(retval);
1668			return retval;
1669		}
1670	} else if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1671		data_start = di->id2.i_data.id_data;
1672		size = i_size_read(dir);
1673
1674		BUG_ON(insert_bh != parent_fe_bh);
1675	}
1676
1677	rec_len = OCFS2_DIR_REC_LEN(namelen);
1678	offset = 0;
1679	de = (struct ocfs2_dir_entry *) data_start;
1680	while (1) {
1681		BUG_ON((char *)de >= (size + data_start));
1682
1683		/* These checks should've already been passed by the
1684		 * prepare function, but I guess we can leave them
1685		 * here anyway. */
1686		if (!ocfs2_check_dir_entry(dir, de, insert_bh, data_start,
1687					   size, offset)) {
1688			retval = -ENOENT;
1689			goto bail;
1690		}
1691		if (ocfs2_match(namelen, name, de)) {
1692			retval = -EEXIST;
1693			goto bail;
1694		}
1695
1696		/* We're guaranteed that we should have space, so we
1697		 * can't possibly have hit the trailer...right? */
1698		mlog_bug_on_msg(ocfs2_skip_dir_trailer(dir, de, offset, size),
1699				"Hit dir trailer trying to insert %.*s "
1700			        "(namelen %d) into directory %llu.  "
1701				"offset is %lu, trailer offset is %d\n",
1702				namelen, name, namelen,
1703				(unsigned long long)parent_fe_bh->b_blocknr,
1704				offset, ocfs2_dir_trailer_blk_off(dir->i_sb));
1705
1706		if (ocfs2_dirent_would_fit(de, rec_len)) {
1707			inode_set_mtime_to_ts(dir,
1708					      inode_set_ctime_current(dir));
1709			retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1710			if (retval < 0) {
1711				mlog_errno(retval);
1712				goto bail;
1713			}
1714
1715			if (insert_bh == parent_fe_bh)
1716				retval = ocfs2_journal_access_di(handle,
1717								 INODE_CACHE(dir),
1718								 insert_bh,
1719								 OCFS2_JOURNAL_ACCESS_WRITE);
1720			else {
1721				retval = ocfs2_journal_access_db(handle,
1722								 INODE_CACHE(dir),
1723								 insert_bh,
1724					      OCFS2_JOURNAL_ACCESS_WRITE);
1725
1726				if (!retval && ocfs2_dir_indexed(dir))
1727					retval = ocfs2_dx_dir_insert(dir,
1728								handle,
1729								lookup);
1730			}
1731
1732			if (retval) {
1733				mlog_errno(retval);
1734				goto bail;
1735			}
1736
1737			/* By now the buffer is marked for journaling */
1738			offset += le16_to_cpu(de->rec_len);
1739			if (le64_to_cpu(de->inode)) {
1740				de1 = (struct ocfs2_dir_entry *)((char *) de +
1741					OCFS2_DIR_REC_LEN(de->name_len));
1742				de1->rec_len =
1743					cpu_to_le16(le16_to_cpu(de->rec_len) -
1744					OCFS2_DIR_REC_LEN(de->name_len));
1745				de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1746				de = de1;
1747			}
1748			de->file_type = FT_UNKNOWN;
1749			if (blkno) {
1750				de->inode = cpu_to_le64(blkno);
1751				ocfs2_set_de_type(de, inode->i_mode);
1752			} else
1753				de->inode = 0;
1754			de->name_len = namelen;
1755			memcpy(de->name, name, namelen);
1756
1757			if (ocfs2_dir_indexed(dir))
1758				ocfs2_recalc_free_list(dir, handle, lookup);
1759
1760			inode_inc_iversion(dir);
1761			ocfs2_journal_dirty(handle, insert_bh);
1762			retval = 0;
1763			goto bail;
1764		}
1765
1766		offset += le16_to_cpu(de->rec_len);
1767		de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1768	}
1769
1770	/* when you think about it, the assert above should prevent us
1771	 * from ever getting here. */
1772	retval = -ENOSPC;
1773bail:
1774	if (retval)
1775		mlog_errno(retval);
1776
1777	return retval;
1778}
1779
1780static int ocfs2_dir_foreach_blk_id(struct inode *inode,
1781				    u64 *f_version,
1782				    struct dir_context *ctx)
1783{
1784	int ret, i;
1785	unsigned long offset = ctx->pos;
1786	struct buffer_head *di_bh = NULL;
1787	struct ocfs2_dinode *di;
1788	struct ocfs2_inline_data *data;
1789	struct ocfs2_dir_entry *de;
1790
1791	ret = ocfs2_read_inode_block(inode, &di_bh);
1792	if (ret) {
1793		mlog(ML_ERROR, "Unable to read inode block for dir %llu\n",
1794		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1795		goto out;
1796	}
1797
1798	di = (struct ocfs2_dinode *)di_bh->b_data;
1799	data = &di->id2.i_data;
1800
1801	while (ctx->pos < i_size_read(inode)) {
1802		/* If the dir block has changed since the last call to
1803		 * readdir(2), then we might be pointing to an invalid
1804		 * dirent right now.  Scan from the start of the block
1805		 * to make sure. */
1806		if (!inode_eq_iversion(inode, *f_version)) {
1807			for (i = 0; i < i_size_read(inode) && i < offset; ) {
1808				de = (struct ocfs2_dir_entry *)
1809					(data->id_data + i);
1810				/* It's too expensive to do a full
1811				 * dirent test each time round this
1812				 * loop, but we do have to test at
1813				 * least that it is non-zero.  A
1814				 * failure will be detected in the
1815				 * dirent test below. */
1816				if (le16_to_cpu(de->rec_len) <
1817				    OCFS2_DIR_REC_LEN(1))
1818					break;
1819				i += le16_to_cpu(de->rec_len);
1820			}
1821			ctx->pos = offset = i;
1822			*f_version = inode_query_iversion(inode);
1823		}
1824
1825		de = (struct ocfs2_dir_entry *) (data->id_data + ctx->pos);
1826		if (!ocfs2_check_dir_entry(inode, de, di_bh, (char *)data->id_data,
1827					   i_size_read(inode), ctx->pos)) {
1828			/* On error, skip the f_pos to the end. */
1829			ctx->pos = i_size_read(inode);
1830			break;
1831		}
1832		offset += le16_to_cpu(de->rec_len);
1833		if (le64_to_cpu(de->inode)) {
1834			if (!dir_emit(ctx, de->name, de->name_len,
1835				      le64_to_cpu(de->inode),
1836				      fs_ftype_to_dtype(de->file_type)))
1837				goto out;
1838		}
1839		ctx->pos += le16_to_cpu(de->rec_len);
1840	}
1841out:
1842	brelse(di_bh);
1843	return 0;
1844}
1845
1846/*
1847 * NOTE: This function can be called against unindexed directories,
1848 * and indexed ones.
1849 */
1850static int ocfs2_dir_foreach_blk_el(struct inode *inode,
1851				    u64 *f_version,
1852				    struct dir_context *ctx,
1853				    bool persist)
1854{
1855	unsigned long offset, blk, last_ra_blk = 0;
1856	int i;
1857	struct buffer_head * bh, * tmp;
1858	struct ocfs2_dir_entry * de;
1859	struct super_block * sb = inode->i_sb;
1860	unsigned int ra_sectors = 16;
1861	int stored = 0;
1862
1863	bh = NULL;
1864
1865	offset = ctx->pos & (sb->s_blocksize - 1);
1866
1867	while (ctx->pos < i_size_read(inode)) {
1868		blk = ctx->pos >> sb->s_blocksize_bits;
1869		if (ocfs2_read_dir_block(inode, blk, &bh, 0)) {
1870			/* Skip the corrupt dirblock and keep trying */
1871			ctx->pos += sb->s_blocksize - offset;
1872			continue;
1873		}
1874
1875		/* The idea here is to begin with 8k read-ahead and to stay
1876		 * 4k ahead of our current position.
1877		 *
1878		 * TODO: Use the pagecache for this. We just need to
1879		 * make sure it's cluster-safe... */
1880		if (!last_ra_blk
1881		    || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
1882			for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
1883			     i > 0; i--) {
1884				tmp = NULL;
1885				if (!ocfs2_read_dir_block(inode, ++blk, &tmp,
1886							  OCFS2_BH_READAHEAD))
1887					brelse(tmp);
1888			}
1889			last_ra_blk = blk;
1890			ra_sectors = 8;
1891		}
1892
1893		/* If the dir block has changed since the last call to
1894		 * readdir(2), then we might be pointing to an invalid
1895		 * dirent right now.  Scan from the start of the block
1896		 * to make sure. */
1897		if (!inode_eq_iversion(inode, *f_version)) {
1898			for (i = 0; i < sb->s_blocksize && i < offset; ) {
1899				de = (struct ocfs2_dir_entry *) (bh->b_data + i);
1900				/* It's too expensive to do a full
1901				 * dirent test each time round this
1902				 * loop, but we do have to test at
1903				 * least that it is non-zero.  A
1904				 * failure will be detected in the
1905				 * dirent test below. */
1906				if (le16_to_cpu(de->rec_len) <
1907				    OCFS2_DIR_REC_LEN(1))
1908					break;
1909				i += le16_to_cpu(de->rec_len);
1910			}
1911			offset = i;
1912			ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
1913				| offset;
1914			*f_version = inode_query_iversion(inode);
1915		}
1916
1917		while (ctx->pos < i_size_read(inode)
1918		       && offset < sb->s_blocksize) {
1919			de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
1920			if (!ocfs2_check_dir_entry(inode, de, bh, bh->b_data,
1921						   sb->s_blocksize, offset)) {
1922				/* On error, skip the f_pos to the
1923				   next block. */
1924				ctx->pos = (ctx->pos | (sb->s_blocksize - 1)) + 1;
1925				break;
1926			}
1927			if (le64_to_cpu(de->inode)) {
1928				if (!dir_emit(ctx, de->name,
1929						de->name_len,
1930						le64_to_cpu(de->inode),
1931					fs_ftype_to_dtype(de->file_type))) {
1932					brelse(bh);
1933					return 0;
1934				}
1935				stored++;
1936			}
1937			offset += le16_to_cpu(de->rec_len);
1938			ctx->pos += le16_to_cpu(de->rec_len);
1939		}
1940		offset = 0;
1941		brelse(bh);
1942		bh = NULL;
1943		if (!persist && stored)
1944			break;
1945	}
1946	return 0;
1947}
1948
1949static int ocfs2_dir_foreach_blk(struct inode *inode, u64 *f_version,
1950				 struct dir_context *ctx,
1951				 bool persist)
1952{
1953	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1954		return ocfs2_dir_foreach_blk_id(inode, f_version, ctx);
1955	return ocfs2_dir_foreach_blk_el(inode, f_version, ctx, persist);
1956}
1957
1958/*
1959 * This is intended to be called from inside other kernel functions,
1960 * so we fake some arguments.
1961 */
1962int ocfs2_dir_foreach(struct inode *inode, struct dir_context *ctx)
1963{
1964	u64 version = inode_query_iversion(inode);
1965	ocfs2_dir_foreach_blk(inode, &version, ctx, true);
1966	return 0;
1967}
1968
1969/*
1970 * ocfs2_readdir()
1971 *
1972 */
1973int ocfs2_readdir(struct file *file, struct dir_context *ctx)
1974{
1975	int error = 0;
1976	struct inode *inode = file_inode(file);
1977	struct ocfs2_file_private *fp = file->private_data;
1978	int lock_level = 0;
1979
1980	trace_ocfs2_readdir((unsigned long long)OCFS2_I(inode)->ip_blkno);
1981
1982	error = ocfs2_inode_lock_atime(inode, file->f_path.mnt, &lock_level, 1);
1983	if (lock_level && error >= 0) {
1984		/* We release EX lock which used to update atime
1985		 * and get PR lock again to reduce contention
1986		 * on commonly accessed directories. */
1987		ocfs2_inode_unlock(inode, 1);
1988		lock_level = 0;
1989		error = ocfs2_inode_lock(inode, NULL, 0);
1990	}
1991	if (error < 0) {
1992		if (error != -ENOENT)
1993			mlog_errno(error);
1994		/* we haven't got any yet, so propagate the error. */
1995		goto bail_nolock;
1996	}
1997
1998	error = ocfs2_dir_foreach_blk(inode, &fp->cookie, ctx, false);
1999
2000	ocfs2_inode_unlock(inode, lock_level);
2001	if (error)
2002		mlog_errno(error);
2003
2004bail_nolock:
2005
2006	return error;
2007}
2008
2009/*
2010 * NOTE: this should always be called with parent dir i_rwsem taken.
2011 */
2012int ocfs2_find_files_on_disk(const char *name,
2013			     int namelen,
2014			     u64 *blkno,
2015			     struct inode *inode,
2016			     struct ocfs2_dir_lookup_result *lookup)
2017{
2018	int status = -ENOENT;
2019
2020	trace_ocfs2_find_files_on_disk(namelen, name, blkno,
2021				(unsigned long long)OCFS2_I(inode)->ip_blkno);
2022
2023	status = ocfs2_find_entry(name, namelen, inode, lookup);
2024	if (status)
2025		goto leave;
2026
2027	*blkno = le64_to_cpu(lookup->dl_entry->inode);
2028
2029	status = 0;
2030leave:
2031
2032	return status;
2033}
2034
2035/*
2036 * Convenience function for callers which just want the block number
2037 * mapped to a name and don't require the full dirent info, etc.
2038 */
2039int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
2040			       int namelen, u64 *blkno)
2041{
2042	int ret;
2043	struct ocfs2_dir_lookup_result lookup = { NULL, };
2044
2045	ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &lookup);
2046	ocfs2_free_dir_lookup_result(&lookup);
2047
2048	return ret;
2049}
2050
2051/* Check for a name within a directory.
2052 *
2053 * Return 0 if the name does not exist
2054 * Return -EEXIST if the directory contains the name
2055 * Return -EFSCORRUPTED if found corruption
2056 *
2057 * Callers should have i_rwsem + a cluster lock on dir
2058 */
2059int ocfs2_check_dir_for_entry(struct inode *dir,
2060			      const char *name,
2061			      int namelen)
2062{
2063	int ret = 0;
2064	struct ocfs2_dir_lookup_result lookup = { NULL, };
2065
2066	trace_ocfs2_check_dir_for_entry(
2067		(unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
2068
2069	ret = ocfs2_find_entry(name, namelen, dir, &lookup);
2070	if (ret == 0) {
2071		ret = -EEXIST;
2072		mlog_errno(ret);
2073	} else if (ret == -ENOENT) {
2074		ret = 0;
2075	}
2076
2077	ocfs2_free_dir_lookup_result(&lookup);
2078
2079	return ret;
2080}
2081
2082struct ocfs2_empty_dir_priv {
2083	struct dir_context ctx;
2084	unsigned seen_dot;
2085	unsigned seen_dot_dot;
2086	unsigned seen_other;
2087	unsigned dx_dir;
2088};
2089static bool ocfs2_empty_dir_filldir(struct dir_context *ctx, const char *name,
2090				   int name_len, loff_t pos, u64 ino,
2091				   unsigned type)
2092{
2093	struct ocfs2_empty_dir_priv *p =
2094		container_of(ctx, struct ocfs2_empty_dir_priv, ctx);
2095
2096	/*
2097	 * Check the positions of "." and ".." records to be sure
2098	 * they're in the correct place.
2099	 *
2100	 * Indexed directories don't need to proceed past the first
2101	 * two entries, so we end the scan after seeing '..'. Despite
2102	 * that, we allow the scan to proceed In the event that we
2103	 * have a corrupted indexed directory (no dot or dot dot
2104	 * entries). This allows us to double check for existing
2105	 * entries which might not have been found in the index.
2106	 */
2107	if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
2108		p->seen_dot = 1;
2109		return true;
2110	}
2111
2112	if (name_len == 2 && !strncmp("..", name, 2) &&
2113	    pos == OCFS2_DIR_REC_LEN(1)) {
2114		p->seen_dot_dot = 1;
2115
2116		if (p->dx_dir && p->seen_dot)
2117			return false;
2118
2119		return true;
2120	}
2121
2122	p->seen_other = 1;
2123	return false;
2124}
2125
2126static int ocfs2_empty_dir_dx(struct inode *inode,
2127			      struct ocfs2_empty_dir_priv *priv)
2128{
2129	int ret;
2130	struct buffer_head *di_bh = NULL;
2131	struct buffer_head *dx_root_bh = NULL;
2132	struct ocfs2_dinode *di;
2133	struct ocfs2_dx_root_block *dx_root;
2134
2135	priv->dx_dir = 1;
2136
2137	ret = ocfs2_read_inode_block(inode, &di_bh);
2138	if (ret) {
2139		mlog_errno(ret);
2140		goto out;
2141	}
2142	di = (struct ocfs2_dinode *)di_bh->b_data;
2143
2144	ret = ocfs2_read_dx_root(inode, di, &dx_root_bh);
2145	if (ret) {
2146		mlog_errno(ret);
2147		goto out;
2148	}
2149	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2150
2151	if (le32_to_cpu(dx_root->dr_num_entries) != 2)
2152		priv->seen_other = 1;
2153
2154out:
2155	brelse(di_bh);
2156	brelse(dx_root_bh);
2157	return ret;
2158}
2159
2160/*
2161 * routine to check that the specified directory is empty (for rmdir)
2162 *
2163 * Returns 1 if dir is empty, zero otherwise.
2164 *
2165 * XXX: This is a performance problem for unindexed directories.
2166 */
2167int ocfs2_empty_dir(struct inode *inode)
2168{
2169	int ret;
2170	struct ocfs2_empty_dir_priv priv = {
2171		.ctx.actor = ocfs2_empty_dir_filldir,
2172	};
2173
2174	if (ocfs2_dir_indexed(inode)) {
2175		ret = ocfs2_empty_dir_dx(inode, &priv);
2176		if (ret)
2177			mlog_errno(ret);
2178		/*
2179		 * We still run ocfs2_dir_foreach to get the checks
2180		 * for "." and "..".
2181		 */
2182	}
2183
2184	ret = ocfs2_dir_foreach(inode, &priv.ctx);
2185	if (ret)
2186		mlog_errno(ret);
2187
2188	if (!priv.seen_dot || !priv.seen_dot_dot) {
2189		mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
2190		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
2191		/*
2192		 * XXX: Is it really safe to allow an unlink to continue?
2193		 */
2194		return 1;
2195	}
2196
2197	return !priv.seen_other;
2198}
2199
2200/*
2201 * Fills "." and ".." dirents in a new directory block. Returns dirent for
2202 * "..", which might be used during creation of a directory with a trailing
2203 * header. It is otherwise safe to ignore the return code.
2204 */
2205static struct ocfs2_dir_entry *ocfs2_fill_initial_dirents(struct inode *inode,
2206							  struct inode *parent,
2207							  char *start,
2208							  unsigned int size)
2209{
2210	struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start;
2211
2212	de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
2213	de->name_len = 1;
2214	de->rec_len =
2215		cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
2216	strscpy(de->name, ".");
2217	ocfs2_set_de_type(de, S_IFDIR);
2218
2219	de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
2220	de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
2221	de->rec_len = cpu_to_le16(size - OCFS2_DIR_REC_LEN(1));
2222	de->name_len = 2;
2223	strscpy(de->name, "..");
2224	ocfs2_set_de_type(de, S_IFDIR);
2225
2226	return de;
2227}
2228
2229/*
2230 * This works together with code in ocfs2_mknod_locked() which sets
2231 * the inline-data flag and initializes the inline-data section.
2232 */
2233static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb,
2234				 handle_t *handle,
2235				 struct inode *parent,
2236				 struct inode *inode,
2237				 struct buffer_head *di_bh)
2238{
2239	int ret;
2240	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2241	struct ocfs2_inline_data *data = &di->id2.i_data;
2242	unsigned int size = le16_to_cpu(data->id_count);
2243
2244	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2245				      OCFS2_JOURNAL_ACCESS_WRITE);
2246	if (ret) {
2247		mlog_errno(ret);
2248		goto out;
2249	}
2250
2251	ocfs2_fill_initial_dirents(inode, parent, data->id_data, size);
2252	ocfs2_journal_dirty(handle, di_bh);
2253
2254	i_size_write(inode, size);
2255	set_nlink(inode, 2);
2256	inode->i_blocks = ocfs2_inode_sector_count(inode);
2257
2258	ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2259	if (ret < 0)
2260		mlog_errno(ret);
2261
2262out:
2263	return ret;
2264}
2265
2266static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
2267				 handle_t *handle,
2268				 struct inode *parent,
2269				 struct inode *inode,
2270				 struct buffer_head *fe_bh,
2271				 struct ocfs2_alloc_context *data_ac,
2272				 struct buffer_head **ret_new_bh)
2273{
2274	int status;
2275	unsigned int size = osb->sb->s_blocksize;
2276	struct buffer_head *new_bh = NULL;
2277	struct ocfs2_dir_entry *de;
2278
2279	if (ocfs2_new_dir_wants_trailer(inode))
2280		size = ocfs2_dir_trailer_blk_off(parent->i_sb);
2281
2282	status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
2283				     data_ac, NULL, &new_bh);
2284	if (status < 0) {
2285		mlog_errno(status);
2286		goto bail;
2287	}
2288
2289	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2290
2291	status = ocfs2_journal_access_db(handle, INODE_CACHE(inode), new_bh,
2292					 OCFS2_JOURNAL_ACCESS_CREATE);
2293	if (status < 0) {
2294		mlog_errno(status);
2295		goto bail;
2296	}
2297	memset(new_bh->b_data, 0, osb->sb->s_blocksize);
2298
2299	de = ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data, size);
2300	if (ocfs2_new_dir_wants_trailer(inode)) {
2301		int size = le16_to_cpu(de->rec_len);
2302
2303		/*
2304		 * Figure out the size of the hole left over after
2305		 * insertion of '.' and '..'. The trailer wants this
2306		 * information.
2307		 */
2308		size -= OCFS2_DIR_REC_LEN(2);
2309		size -= sizeof(struct ocfs2_dir_block_trailer);
2310
2311		ocfs2_init_dir_trailer(inode, new_bh, size);
2312	}
2313
2314	ocfs2_journal_dirty(handle, new_bh);
2315
2316	i_size_write(inode, inode->i_sb->s_blocksize);
2317	set_nlink(inode, 2);
2318	inode->i_blocks = ocfs2_inode_sector_count(inode);
2319	status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
2320	if (status < 0) {
2321		mlog_errno(status);
2322		goto bail;
2323	}
2324
2325	status = 0;
2326	if (ret_new_bh) {
2327		*ret_new_bh = new_bh;
2328		new_bh = NULL;
2329	}
2330bail:
2331	brelse(new_bh);
2332
2333	return status;
2334}
2335
2336static int ocfs2_dx_dir_attach_index(struct ocfs2_super *osb,
2337				     handle_t *handle, struct inode *dir,
2338				     struct buffer_head *di_bh,
2339				     struct buffer_head *dirdata_bh,
2340				     struct ocfs2_alloc_context *meta_ac,
2341				     int dx_inline, u32 num_entries,
2342				     struct buffer_head **ret_dx_root_bh)
2343{
2344	int ret;
2345	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
2346	u16 dr_suballoc_bit;
2347	u64 suballoc_loc, dr_blkno;
2348	unsigned int num_bits;
2349	struct buffer_head *dx_root_bh = NULL;
2350	struct ocfs2_dx_root_block *dx_root;
2351	struct ocfs2_dir_block_trailer *trailer =
2352		ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
2353
2354	ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
2355				   &dr_suballoc_bit, &num_bits, &dr_blkno);
2356	if (ret) {
2357		mlog_errno(ret);
2358		goto out;
2359	}
2360
2361	trace_ocfs2_dx_dir_attach_index(
2362				(unsigned long long)OCFS2_I(dir)->ip_blkno,
2363				(unsigned long long)dr_blkno);
2364
2365	dx_root_bh = sb_getblk(osb->sb, dr_blkno);
2366	if (dx_root_bh == NULL) {
2367		ret = -ENOMEM;
2368		goto out;
2369	}
2370	ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dx_root_bh);
2371
2372	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
2373				      OCFS2_JOURNAL_ACCESS_CREATE);
2374	if (ret < 0) {
2375		mlog_errno(ret);
2376		goto out;
2377	}
2378
2379	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2380	memset(dx_root, 0, osb->sb->s_blocksize);
2381	strscpy(dx_root->dr_signature, OCFS2_DX_ROOT_SIGNATURE);
2382	dx_root->dr_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
2383	dx_root->dr_suballoc_loc = cpu_to_le64(suballoc_loc);
2384	dx_root->dr_suballoc_bit = cpu_to_le16(dr_suballoc_bit);
2385	dx_root->dr_fs_generation = cpu_to_le32(osb->fs_generation);
2386	dx_root->dr_blkno = cpu_to_le64(dr_blkno);
2387	dx_root->dr_dir_blkno = cpu_to_le64(OCFS2_I(dir)->ip_blkno);
2388	dx_root->dr_num_entries = cpu_to_le32(num_entries);
2389	if (le16_to_cpu(trailer->db_free_rec_len))
2390		dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
2391	else
2392		dx_root->dr_free_blk = cpu_to_le64(0);
2393
2394	if (dx_inline) {
2395		dx_root->dr_flags |= OCFS2_DX_FLAG_INLINE;
2396		dx_root->dr_entries.de_count =
2397			cpu_to_le16(ocfs2_dx_entries_per_root(osb->sb));
2398	} else {
2399		dx_root->dr_list.l_count =
2400			cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
2401	}
2402	ocfs2_journal_dirty(handle, dx_root_bh);
2403
2404	ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
2405				      OCFS2_JOURNAL_ACCESS_CREATE);
2406	if (ret) {
2407		mlog_errno(ret);
2408		goto out;
2409	}
2410
2411	di->i_dx_root = cpu_to_le64(dr_blkno);
2412
2413	spin_lock(&OCFS2_I(dir)->ip_lock);
2414	OCFS2_I(dir)->ip_dyn_features |= OCFS2_INDEXED_DIR_FL;
2415	di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
2416	spin_unlock(&OCFS2_I(dir)->ip_lock);
2417
2418	ocfs2_journal_dirty(handle, di_bh);
2419
2420	*ret_dx_root_bh = dx_root_bh;
2421	dx_root_bh = NULL;
2422
2423out:
2424	brelse(dx_root_bh);
2425	return ret;
2426}
2427
2428static int ocfs2_dx_dir_format_cluster(struct ocfs2_super *osb,
2429				       handle_t *handle, struct inode *dir,
2430				       struct buffer_head **dx_leaves,
2431				       int num_dx_leaves, u64 start_blk)
2432{
2433	int ret, i;
2434	struct ocfs2_dx_leaf *dx_leaf;
2435	struct buffer_head *bh;
2436
2437	for (i = 0; i < num_dx_leaves; i++) {
2438		bh = sb_getblk(osb->sb, start_blk + i);
2439		if (bh == NULL) {
2440			ret = -ENOMEM;
2441			goto out;
2442		}
2443		dx_leaves[i] = bh;
2444
2445		ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), bh);
2446
2447		ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), bh,
2448					      OCFS2_JOURNAL_ACCESS_CREATE);
2449		if (ret < 0) {
2450			mlog_errno(ret);
2451			goto out;
2452		}
2453
2454		dx_leaf = (struct ocfs2_dx_leaf *) bh->b_data;
2455
2456		memset(dx_leaf, 0, osb->sb->s_blocksize);
2457		strscpy(dx_leaf->dl_signature, OCFS2_DX_LEAF_SIGNATURE);
2458		dx_leaf->dl_fs_generation = cpu_to_le32(osb->fs_generation);
2459		dx_leaf->dl_blkno = cpu_to_le64(bh->b_blocknr);
2460		dx_leaf->dl_list.de_count =
2461			cpu_to_le16(ocfs2_dx_entries_per_leaf(osb->sb));
2462
2463		trace_ocfs2_dx_dir_format_cluster(
2464				(unsigned long long)OCFS2_I(dir)->ip_blkno,
2465				(unsigned long long)bh->b_blocknr,
2466				le16_to_cpu(dx_leaf->dl_list.de_count));
2467
2468		ocfs2_journal_dirty(handle, bh);
2469	}
2470
2471	ret = 0;
2472out:
2473	return ret;
2474}
2475
2476/*
2477 * Allocates and formats a new cluster for use in an indexed dir
2478 * leaf. This version will not do the extent insert, so that it can be
2479 * used by operations which need careful ordering.
2480 */
2481static int __ocfs2_dx_dir_new_cluster(struct inode *dir,
2482				      u32 cpos, handle_t *handle,
2483				      struct ocfs2_alloc_context *data_ac,
2484				      struct buffer_head **dx_leaves,
2485				      int num_dx_leaves, u64 *ret_phys_blkno)
2486{
2487	int ret;
2488	u32 phys, num;
2489	u64 phys_blkno;
2490	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2491
2492	/*
2493	 * XXX: For create, this should claim cluster for the index
2494	 * *before* the unindexed insert so that we have a better
2495	 * chance of contiguousness as the directory grows in number
2496	 * of entries.
2497	 */
2498	ret = __ocfs2_claim_clusters(handle, data_ac, 1, 1, &phys, &num);
2499	if (ret) {
2500		mlog_errno(ret);
2501		goto out;
2502	}
2503
2504	/*
2505	 * Format the new cluster first. That way, we're inserting
2506	 * valid data.
2507	 */
2508	phys_blkno = ocfs2_clusters_to_blocks(osb->sb, phys);
2509	ret = ocfs2_dx_dir_format_cluster(osb, handle, dir, dx_leaves,
2510					  num_dx_leaves, phys_blkno);
2511	if (ret) {
2512		mlog_errno(ret);
2513		goto out;
2514	}
2515
2516	*ret_phys_blkno = phys_blkno;
2517out:
2518	return ret;
2519}
2520
2521static int ocfs2_dx_dir_new_cluster(struct inode *dir,
2522				    struct ocfs2_extent_tree *et,
2523				    u32 cpos, handle_t *handle,
2524				    struct ocfs2_alloc_context *data_ac,
2525				    struct ocfs2_alloc_context *meta_ac,
2526				    struct buffer_head **dx_leaves,
2527				    int num_dx_leaves)
2528{
2529	int ret;
2530	u64 phys_blkno;
2531
2532	ret = __ocfs2_dx_dir_new_cluster(dir, cpos, handle, data_ac, dx_leaves,
2533					 num_dx_leaves, &phys_blkno);
2534	if (ret) {
2535		mlog_errno(ret);
2536		goto out;
2537	}
2538
2539	ret = ocfs2_insert_extent(handle, et, cpos, phys_blkno, 1, 0,
2540				  meta_ac);
2541	if (ret)
2542		mlog_errno(ret);
2543out:
2544	return ret;
2545}
2546
2547static struct buffer_head **ocfs2_dx_dir_kmalloc_leaves(struct super_block *sb,
2548							int *ret_num_leaves)
2549{
2550	int num_dx_leaves = ocfs2_clusters_to_blocks(sb, 1);
2551	struct buffer_head **dx_leaves;
2552
2553	dx_leaves = kcalloc(num_dx_leaves, sizeof(struct buffer_head *),
2554			    GFP_NOFS);
2555	if (dx_leaves && ret_num_leaves)
2556		*ret_num_leaves = num_dx_leaves;
2557
2558	return dx_leaves;
2559}
2560
2561static int ocfs2_fill_new_dir_dx(struct ocfs2_super *osb,
2562				 handle_t *handle,
2563				 struct inode *parent,
2564				 struct inode *inode,
2565				 struct buffer_head *di_bh,
2566				 struct ocfs2_alloc_context *data_ac,
2567				 struct ocfs2_alloc_context *meta_ac)
2568{
2569	int ret;
2570	struct buffer_head *leaf_bh = NULL;
2571	struct buffer_head *dx_root_bh = NULL;
2572	struct ocfs2_dx_hinfo hinfo;
2573	struct ocfs2_dx_root_block *dx_root;
2574	struct ocfs2_dx_entry_list *entry_list;
2575
2576	/*
2577	 * Our strategy is to create the directory as though it were
2578	 * unindexed, then add the index block. This works with very
2579	 * little complication since the state of a new directory is a
2580	 * very well known quantity.
2581	 *
2582	 * Essentially, we have two dirents ("." and ".."), in the 1st
2583	 * block which need indexing. These are easily inserted into
2584	 * the index block.
2585	 */
2586
2587	ret = ocfs2_fill_new_dir_el(osb, handle, parent, inode, di_bh,
2588				    data_ac, &leaf_bh);
2589	if (ret) {
2590		mlog_errno(ret);
2591		goto out;
2592	}
2593
2594	ret = ocfs2_dx_dir_attach_index(osb, handle, inode, di_bh, leaf_bh,
2595					meta_ac, 1, 2, &dx_root_bh);
2596	if (ret) {
2597		mlog_errno(ret);
2598		goto out;
2599	}
2600	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2601	entry_list = &dx_root->dr_entries;
2602
2603	/* Buffer has been journaled for us by ocfs2_dx_dir_attach_index */
2604	ocfs2_dx_dir_name_hash(inode, ".", 1, &hinfo);
2605	ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
2606
2607	ocfs2_dx_dir_name_hash(inode, "..", 2, &hinfo);
2608	ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
2609
2610out:
2611	brelse(dx_root_bh);
2612	brelse(leaf_bh);
2613	return ret;
2614}
2615
2616int ocfs2_fill_new_dir(struct ocfs2_super *osb,
2617		       handle_t *handle,
2618		       struct inode *parent,
2619		       struct inode *inode,
2620		       struct buffer_head *fe_bh,
2621		       struct ocfs2_alloc_context *data_ac,
2622		       struct ocfs2_alloc_context *meta_ac)
2623
2624{
2625	BUG_ON(!ocfs2_supports_inline_data(osb) && data_ac == NULL);
2626
2627	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2628		return ocfs2_fill_new_dir_id(osb, handle, parent, inode, fe_bh);
2629
2630	if (ocfs2_supports_indexed_dirs(osb))
2631		return ocfs2_fill_new_dir_dx(osb, handle, parent, inode, fe_bh,
2632					     data_ac, meta_ac);
2633
2634	return ocfs2_fill_new_dir_el(osb, handle, parent, inode, fe_bh,
2635				     data_ac, NULL);
2636}
2637
2638static int ocfs2_dx_dir_index_block(struct inode *dir,
2639				    handle_t *handle,
2640				    struct buffer_head **dx_leaves,
2641				    int num_dx_leaves,
2642				    u32 *num_dx_entries,
2643				    struct buffer_head *dirent_bh)
2644{
2645	int ret = 0, namelen, i;
2646	char *de_buf, *limit;
2647	struct ocfs2_dir_entry *de;
2648	struct buffer_head *dx_leaf_bh;
2649	struct ocfs2_dx_hinfo hinfo;
2650	u64 dirent_blk = dirent_bh->b_blocknr;
2651
2652	de_buf = dirent_bh->b_data;
2653	limit = de_buf + dir->i_sb->s_blocksize;
2654
2655	while (de_buf < limit) {
2656		de = (struct ocfs2_dir_entry *)de_buf;
2657
2658		namelen = de->name_len;
2659		if (!namelen || !de->inode)
2660			goto inc;
2661
2662		ocfs2_dx_dir_name_hash(dir, de->name, namelen, &hinfo);
2663
2664		i = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb), &hinfo);
2665		dx_leaf_bh = dx_leaves[i];
2666
2667		ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &hinfo,
2668						 dirent_blk, dx_leaf_bh);
2669		if (ret) {
2670			mlog_errno(ret);
2671			goto out;
2672		}
2673
2674		*num_dx_entries = *num_dx_entries + 1;
2675
2676inc:
2677		de_buf += le16_to_cpu(de->rec_len);
2678	}
2679
2680out:
2681	return ret;
2682}
2683
2684/*
2685 * XXX: This expects dx_root_bh to already be part of the transaction.
2686 */
2687static void ocfs2_dx_dir_index_root_block(struct inode *dir,
2688					 struct buffer_head *dx_root_bh,
2689					 struct buffer_head *dirent_bh)
2690{
2691	char *de_buf, *limit;
2692	struct ocfs2_dx_root_block *dx_root;
2693	struct ocfs2_dir_entry *de;
2694	struct ocfs2_dx_hinfo hinfo;
2695	u64 dirent_blk = dirent_bh->b_blocknr;
2696
2697	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
2698
2699	de_buf = dirent_bh->b_data;
2700	limit = de_buf + dir->i_sb->s_blocksize;
2701
2702	while (de_buf < limit) {
2703		de = (struct ocfs2_dir_entry *)de_buf;
2704
2705		if (!de->name_len || !de->inode)
2706			goto inc;
2707
2708		ocfs2_dx_dir_name_hash(dir, de->name, de->name_len, &hinfo);
2709
2710		trace_ocfs2_dx_dir_index_root_block(
2711				(unsigned long long)dir->i_ino,
2712				hinfo.major_hash, hinfo.minor_hash,
2713				de->name_len, de->name,
2714				le16_to_cpu(dx_root->dr_entries.de_num_used));
2715
2716		ocfs2_dx_entry_list_insert(&dx_root->dr_entries, &hinfo,
2717					   dirent_blk);
2718
2719		le32_add_cpu(&dx_root->dr_num_entries, 1);
2720inc:
2721		de_buf += le16_to_cpu(de->rec_len);
2722	}
2723}
2724
2725/*
2726 * Count the number of inline directory entries in di_bh and compare
2727 * them against the number of entries we can hold in an inline dx root
2728 * block.
2729 */
2730static int ocfs2_new_dx_should_be_inline(struct inode *dir,
2731					 struct buffer_head *di_bh)
2732{
2733	int dirent_count = 0;
2734	char *de_buf, *limit;
2735	struct ocfs2_dir_entry *de;
2736	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2737
2738	de_buf = di->id2.i_data.id_data;
2739	limit = de_buf + i_size_read(dir);
2740
2741	while (de_buf < limit) {
2742		de = (struct ocfs2_dir_entry *)de_buf;
2743
2744		if (de->name_len && de->inode)
2745			dirent_count++;
2746
2747		de_buf += le16_to_cpu(de->rec_len);
2748	}
2749
2750	/* We are careful to leave room for one extra record. */
2751	return dirent_count < ocfs2_dx_entries_per_root(dir->i_sb);
2752}
2753
2754/*
2755 * Expand rec_len of the rightmost dirent in a directory block so that it
2756 * contains the end of our valid space for dirents. We do this during
2757 * expansion from an inline directory to one with extents. The first dir block
2758 * in that case is taken from the inline data portion of the inode block.
2759 *
2760 * This will also return the largest amount of contiguous space for a dirent
2761 * in the block. That value is *not* necessarily the last dirent, even after
2762 * expansion. The directory indexing code wants this value for free space
2763 * accounting. We do this here since we're already walking the entire dir
2764 * block.
2765 *
2766 * We add the dir trailer if this filesystem wants it.
2767 */
2768static unsigned int ocfs2_expand_last_dirent(char *start, unsigned int old_size,
2769					     struct inode *dir)
2770{
2771	struct super_block *sb = dir->i_sb;
2772	struct ocfs2_dir_entry *de;
2773	struct ocfs2_dir_entry *prev_de;
2774	char *de_buf, *limit;
2775	unsigned int new_size = sb->s_blocksize;
2776	unsigned int bytes, this_hole;
2777	unsigned int largest_hole = 0;
2778
2779	if (ocfs2_new_dir_wants_trailer(dir))
2780		new_size = ocfs2_dir_trailer_blk_off(sb);
2781
2782	bytes = new_size - old_size;
2783
2784	limit = start + old_size;
2785	de_buf = start;
2786	de = (struct ocfs2_dir_entry *)de_buf;
2787	do {
2788		this_hole = ocfs2_figure_dirent_hole(de);
2789		if (this_hole > largest_hole)
2790			largest_hole = this_hole;
2791
2792		prev_de = de;
2793		de_buf += le16_to_cpu(de->rec_len);
2794		de = (struct ocfs2_dir_entry *)de_buf;
2795	} while (de_buf < limit);
2796
2797	le16_add_cpu(&prev_de->rec_len, bytes);
2798
2799	/* We need to double check this after modification of the final
2800	 * dirent. */
2801	this_hole = ocfs2_figure_dirent_hole(prev_de);
2802	if (this_hole > largest_hole)
2803		largest_hole = this_hole;
2804
2805	if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
2806		return largest_hole;
2807	return 0;
2808}
2809
2810/*
2811 * We allocate enough clusters to fulfill "blocks_wanted", but set
2812 * i_size to exactly one block. Ocfs2_extend_dir() will handle the
2813 * rest automatically for us.
2814 *
2815 * *first_block_bh is a pointer to the 1st data block allocated to the
2816 *  directory.
2817 */
2818static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
2819				   unsigned int blocks_wanted,
2820				   struct ocfs2_dir_lookup_result *lookup,
2821				   struct buffer_head **first_block_bh)
2822{
2823	u32 alloc, dx_alloc, bit_off, len, num_dx_entries = 0;
2824	struct super_block *sb = dir->i_sb;
2825	int ret, i, num_dx_leaves = 0, dx_inline = 0,
2826		credits = ocfs2_inline_to_extents_credits(sb);
2827	u64 dx_insert_blkno, blkno,
2828		bytes = blocks_wanted << sb->s_blocksize_bits;
2829	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2830	struct ocfs2_inode_info *oi = OCFS2_I(dir);
2831	struct ocfs2_alloc_context *data_ac = NULL;
2832	struct ocfs2_alloc_context *meta_ac = NULL;
2833	struct buffer_head *dirdata_bh = NULL;
2834	struct buffer_head *dx_root_bh = NULL;
2835	struct buffer_head **dx_leaves = NULL;
2836	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2837	handle_t *handle;
2838	struct ocfs2_extent_tree et;
2839	struct ocfs2_extent_tree dx_et;
2840	int did_quota = 0, bytes_allocated = 0;
2841
2842	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir), di_bh);
2843
2844	alloc = ocfs2_clusters_for_bytes(sb, bytes);
2845	dx_alloc = 0;
2846
2847	down_write(&oi->ip_alloc_sem);
2848
2849	if (ocfs2_supports_indexed_dirs(osb)) {
2850		credits += ocfs2_add_dir_index_credits(sb);
2851
2852		dx_inline = ocfs2_new_dx_should_be_inline(dir, di_bh);
2853		if (!dx_inline) {
2854			/* Add one more cluster for an index leaf */
2855			dx_alloc++;
2856			dx_leaves = ocfs2_dx_dir_kmalloc_leaves(sb,
2857								&num_dx_leaves);
2858			if (!dx_leaves) {
2859				ret = -ENOMEM;
2860				mlog_errno(ret);
2861				goto out;
2862			}
2863		}
2864
2865		/* This gets us the dx_root */
2866		ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
2867		if (ret) {
2868			mlog_errno(ret);
2869			goto out;
2870		}
2871	}
2872
2873	/*
2874	 * We should never need more than 2 clusters for the unindexed
2875	 * tree - maximum dirent size is far less than one block. In
2876	 * fact, the only time we'd need more than one cluster is if
2877	 * blocksize == clustersize and the dirent won't fit in the
2878	 * extra space that the expansion to a single block gives. As
2879	 * of today, that only happens on 4k/4k file systems.
2880	 */
2881	BUG_ON(alloc > 2);
2882
2883	ret = ocfs2_reserve_clusters(osb, alloc + dx_alloc, &data_ac);
2884	if (ret) {
2885		mlog_errno(ret);
2886		goto out;
2887	}
2888
2889	/*
2890	 * Prepare for worst case allocation scenario of two separate
2891	 * extents in the unindexed tree.
2892	 */
2893	if (alloc == 2)
2894		credits += OCFS2_SUBALLOC_ALLOC;
2895
2896	handle = ocfs2_start_trans(osb, credits);
2897	if (IS_ERR(handle)) {
2898		ret = PTR_ERR(handle);
2899		mlog_errno(ret);
2900		goto out;
2901	}
2902
2903	ret = dquot_alloc_space_nodirty(dir,
2904		ocfs2_clusters_to_bytes(osb->sb, alloc + dx_alloc));
2905	if (ret)
2906		goto out_commit;
2907	did_quota = 1;
2908
2909	if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
2910		/*
2911		 * Allocate our index cluster first, to maximize the
2912		 * possibility that unindexed leaves grow
2913		 * contiguously.
2914		 */
2915		ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac,
2916						 dx_leaves, num_dx_leaves,
2917						 &dx_insert_blkno);
2918		if (ret) {
2919			mlog_errno(ret);
2920			goto out_commit;
2921		}
2922		bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
2923	}
2924
2925	/*
2926	 * Try to claim as many clusters as the bitmap can give though
2927	 * if we only get one now, that's enough to continue. The rest
2928	 * will be claimed after the conversion to extents.
2929	 */
2930	if (ocfs2_dir_resv_allowed(osb))
2931		data_ac->ac_resv = &oi->ip_la_data_resv;
2932	ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off, &len);
2933	if (ret) {
2934		mlog_errno(ret);
2935		goto out_commit;
2936	}
2937	bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
2938
2939	/*
2940	 * Operations are carefully ordered so that we set up the new
2941	 * data block first. The conversion from inline data to
2942	 * extents follows.
2943	 */
2944	blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
2945	dirdata_bh = sb_getblk(sb, blkno);
2946	if (!dirdata_bh) {
2947		ret = -ENOMEM;
2948		mlog_errno(ret);
2949		goto out_commit;
2950	}
2951
2952	ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dirdata_bh);
2953
2954	ret = ocfs2_journal_access_db(handle, INODE_CACHE(dir), dirdata_bh,
2955				      OCFS2_JOURNAL_ACCESS_CREATE);
2956	if (ret) {
2957		mlog_errno(ret);
2958		goto out_commit;
2959	}
2960
2961	memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir));
2962	memset(dirdata_bh->b_data + i_size_read(dir), 0,
2963	       sb->s_blocksize - i_size_read(dir));
2964	i = ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir), dir);
2965	if (ocfs2_new_dir_wants_trailer(dir)) {
2966		/*
2967		 * Prepare the dir trailer up front. It will otherwise look
2968		 * like a valid dirent. Even if inserting the index fails
2969		 * (unlikely), then all we'll have done is given first dir
2970		 * block a small amount of fragmentation.
2971		 */
2972		ocfs2_init_dir_trailer(dir, dirdata_bh, i);
2973	}
2974
2975	ocfs2_update_inode_fsync_trans(handle, dir, 1);
2976	ocfs2_journal_dirty(handle, dirdata_bh);
2977
2978	if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
2979		/*
2980		 * Dx dirs with an external cluster need to do this up
2981		 * front. Inline dx root's get handled later, after
2982		 * we've allocated our root block. We get passed back
2983		 * a total number of items so that dr_num_entries can
2984		 * be correctly set once the dx_root has been
2985		 * allocated.
2986		 */
2987		ret = ocfs2_dx_dir_index_block(dir, handle, dx_leaves,
2988					       num_dx_leaves, &num_dx_entries,
2989					       dirdata_bh);
2990		if (ret) {
2991			mlog_errno(ret);
2992			goto out_commit;
2993		}
2994	}
2995
2996	/*
2997	 * Set extent, i_size, etc on the directory. After this, the
2998	 * inode should contain the same exact dirents as before and
2999	 * be fully accessible from system calls.
3000	 *
3001	 * We let the later dirent insert modify c/mtime - to the user
3002	 * the data hasn't changed.
3003	 */
3004	ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
3005				      OCFS2_JOURNAL_ACCESS_CREATE);
3006	if (ret) {
3007		mlog_errno(ret);
3008		goto out_commit;
3009	}
3010
3011	spin_lock(&oi->ip_lock);
3012	oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
3013	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
3014	spin_unlock(&oi->ip_lock);
3015
3016	ocfs2_dinode_new_extent_list(dir, di);
3017
3018	i_size_write(dir, sb->s_blocksize);
3019	inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3020
3021	di->i_size = cpu_to_le64(sb->s_blocksize);
3022	di->i_ctime = di->i_mtime = cpu_to_le64(inode_get_ctime_sec(dir));
3023	di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode_get_ctime_nsec(dir));
3024	ocfs2_update_inode_fsync_trans(handle, dir, 1);
3025
3026	/*
3027	 * This should never fail as our extent list is empty and all
3028	 * related blocks have been journaled already.
3029	 */
3030	ret = ocfs2_insert_extent(handle, &et, 0, blkno, len,
3031				  0, NULL);
3032	if (ret) {
3033		mlog_errno(ret);
3034		goto out_commit;
3035	}
3036
3037	/*
3038	 * Set i_blocks after the extent insert for the most up to
3039	 * date ip_clusters value.
3040	 */
3041	dir->i_blocks = ocfs2_inode_sector_count(dir);
3042
3043	ocfs2_journal_dirty(handle, di_bh);
3044
3045	if (ocfs2_supports_indexed_dirs(osb)) {
3046		ret = ocfs2_dx_dir_attach_index(osb, handle, dir, di_bh,
3047						dirdata_bh, meta_ac, dx_inline,
3048						num_dx_entries, &dx_root_bh);
3049		if (ret) {
3050			mlog_errno(ret);
3051			goto out_commit;
3052		}
3053
3054		if (dx_inline) {
3055			ocfs2_dx_dir_index_root_block(dir, dx_root_bh,
3056						      dirdata_bh);
3057		} else {
3058			ocfs2_init_dx_root_extent_tree(&dx_et,
3059						       INODE_CACHE(dir),
3060						       dx_root_bh);
3061			ret = ocfs2_insert_extent(handle, &dx_et, 0,
3062						  dx_insert_blkno, 1, 0, NULL);
3063			if (ret)
3064				mlog_errno(ret);
3065		}
3066	}
3067
3068	/*
3069	 * We asked for two clusters, but only got one in the 1st
3070	 * pass. Claim the 2nd cluster as a separate extent.
3071	 */
3072	if (alloc > len) {
3073		ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
3074					   &len);
3075		if (ret) {
3076			mlog_errno(ret);
3077			goto out_commit;
3078		}
3079		blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
3080
3081		ret = ocfs2_insert_extent(handle, &et, 1,
3082					  blkno, len, 0, NULL);
3083		if (ret) {
3084			mlog_errno(ret);
3085			goto out_commit;
3086		}
3087		bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
3088	}
3089
3090	*first_block_bh = dirdata_bh;
3091	dirdata_bh = NULL;
3092	if (ocfs2_supports_indexed_dirs(osb)) {
3093		unsigned int off;
3094
3095		if (!dx_inline) {
3096			/*
3097			 * We need to return the correct block within the
3098			 * cluster which should hold our entry.
3099			 */
3100			off = ocfs2_dx_dir_hash_idx(osb,
3101						    &lookup->dl_hinfo);
3102			get_bh(dx_leaves[off]);
3103			lookup->dl_dx_leaf_bh = dx_leaves[off];
3104		}
3105		lookup->dl_dx_root_bh = dx_root_bh;
3106		dx_root_bh = NULL;
3107	}
3108
3109out_commit:
3110	if (ret < 0 && did_quota)
3111		dquot_free_space_nodirty(dir, bytes_allocated);
3112
3113	ocfs2_commit_trans(osb, handle);
3114
3115out:
3116	up_write(&oi->ip_alloc_sem);
3117	if (data_ac)
3118		ocfs2_free_alloc_context(data_ac);
3119	if (meta_ac)
3120		ocfs2_free_alloc_context(meta_ac);
3121
3122	if (dx_leaves) {
3123		for (i = 0; i < num_dx_leaves; i++)
3124			brelse(dx_leaves[i]);
3125		kfree(dx_leaves);
3126	}
3127
3128	brelse(dirdata_bh);
3129	brelse(dx_root_bh);
3130
3131	return ret;
3132}
3133
3134/* returns a bh of the 1st new block in the allocation. */
3135static int ocfs2_do_extend_dir(struct super_block *sb,
3136			       handle_t *handle,
3137			       struct inode *dir,
3138			       struct buffer_head *parent_fe_bh,
3139			       struct ocfs2_alloc_context *data_ac,
3140			       struct ocfs2_alloc_context *meta_ac,
3141			       struct buffer_head **new_bh)
3142{
3143	int status;
3144	int extend, did_quota = 0;
3145	u64 p_blkno, v_blkno;
3146
3147	spin_lock(&OCFS2_I(dir)->ip_lock);
3148	extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
3149	spin_unlock(&OCFS2_I(dir)->ip_lock);
3150
3151	if (extend) {
3152		u32 offset = OCFS2_I(dir)->ip_clusters;
3153
3154		status = dquot_alloc_space_nodirty(dir,
3155					ocfs2_clusters_to_bytes(sb, 1));
3156		if (status)
3157			goto bail;
3158		did_quota = 1;
3159
3160		status = ocfs2_add_inode_data(OCFS2_SB(sb), dir, &offset,
3161					      1, 0, parent_fe_bh, handle,
3162					      data_ac, meta_ac, NULL);
3163		BUG_ON(status == -EAGAIN);
3164		if (status < 0) {
3165			mlog_errno(status);
3166			goto bail;
3167		}
3168	}
3169
3170	v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
3171	status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
3172	if (status < 0) {
3173		mlog_errno(status);
3174		goto bail;
3175	}
3176
3177	*new_bh = sb_getblk(sb, p_blkno);
3178	if (!*new_bh) {
3179		status = -ENOMEM;
3180		mlog_errno(status);
3181		goto bail;
3182	}
3183	status = 0;
3184bail:
3185	if (did_quota && status < 0)
3186		dquot_free_space_nodirty(dir, ocfs2_clusters_to_bytes(sb, 1));
3187	return status;
3188}
3189
3190/*
3191 * Assumes you already have a cluster lock on the directory.
3192 *
3193 * 'blocks_wanted' is only used if we have an inline directory which
3194 * is to be turned into an extent based one. The size of the dirent to
3195 * insert might be larger than the space gained by growing to just one
3196 * block, so we may have to grow the inode by two blocks in that case.
3197 *
3198 * If the directory is already indexed, dx_root_bh must be provided.
3199 */
3200static int ocfs2_extend_dir(struct ocfs2_super *osb,
3201			    struct inode *dir,
3202			    struct buffer_head *parent_fe_bh,
3203			    unsigned int blocks_wanted,
3204			    struct ocfs2_dir_lookup_result *lookup,
3205			    struct buffer_head **new_de_bh)
3206{
3207	int status = 0;
3208	int credits, num_free_extents, drop_alloc_sem = 0;
3209	loff_t dir_i_size;
3210	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
3211	struct ocfs2_extent_list *el = &fe->id2.i_list;
3212	struct ocfs2_alloc_context *data_ac = NULL;
3213	struct ocfs2_alloc_context *meta_ac = NULL;
3214	handle_t *handle = NULL;
3215	struct buffer_head *new_bh = NULL;
3216	struct ocfs2_dir_entry * de;
3217	struct super_block *sb = osb->sb;
3218	struct ocfs2_extent_tree et;
3219	struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
3220
3221	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
3222		/*
3223		 * This would be a code error as an inline directory should
3224		 * never have an index root.
3225		 */
3226		BUG_ON(dx_root_bh);
3227
3228		status = ocfs2_expand_inline_dir(dir, parent_fe_bh,
3229						 blocks_wanted, lookup,
3230						 &new_bh);
3231		if (status) {
3232			mlog_errno(status);
3233			goto bail;
3234		}
3235
3236		/* Expansion from inline to an indexed directory will
3237		 * have given us this. */
3238		dx_root_bh = lookup->dl_dx_root_bh;
3239
3240		if (blocks_wanted == 1) {
3241			/*
3242			 * If the new dirent will fit inside the space
3243			 * created by pushing out to one block, then
3244			 * we can complete the operation
3245			 * here. Otherwise we have to expand i_size
3246			 * and format the 2nd block below.
3247			 */
3248			BUG_ON(new_bh == NULL);
3249			goto bail_bh;
3250		}
3251
3252		/*
3253		 * Get rid of 'new_bh' - we want to format the 2nd
3254		 * data block and return that instead.
3255		 */
3256		brelse(new_bh);
3257		new_bh = NULL;
3258
3259		down_write(&OCFS2_I(dir)->ip_alloc_sem);
3260		drop_alloc_sem = 1;
3261		dir_i_size = i_size_read(dir);
3262		credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3263		goto do_extend;
3264	}
3265
3266	down_write(&OCFS2_I(dir)->ip_alloc_sem);
3267	drop_alloc_sem = 1;
3268	dir_i_size = i_size_read(dir);
3269	trace_ocfs2_extend_dir((unsigned long long)OCFS2_I(dir)->ip_blkno,
3270			       dir_i_size);
3271
3272	/* dir->i_size is always block aligned. */
3273	spin_lock(&OCFS2_I(dir)->ip_lock);
3274	if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
3275		spin_unlock(&OCFS2_I(dir)->ip_lock);
3276		ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir),
3277					      parent_fe_bh);
3278		num_free_extents = ocfs2_num_free_extents(&et);
3279		if (num_free_extents < 0) {
3280			status = num_free_extents;
3281			mlog_errno(status);
3282			goto bail;
3283		}
3284
3285		if (!num_free_extents) {
3286			status = ocfs2_reserve_new_metadata(osb, el, &meta_ac);
3287			if (status < 0) {
3288				if (status != -ENOSPC)
3289					mlog_errno(status);
3290				goto bail;
3291			}
3292		}
3293
3294		status = ocfs2_reserve_clusters(osb, 1, &data_ac);
3295		if (status < 0) {
3296			if (status != -ENOSPC)
3297				mlog_errno(status);
3298			goto bail;
3299		}
3300
3301		if (ocfs2_dir_resv_allowed(osb))
3302			data_ac->ac_resv = &OCFS2_I(dir)->ip_la_data_resv;
3303
3304		credits = ocfs2_calc_extend_credits(sb, el);
3305	} else {
3306		spin_unlock(&OCFS2_I(dir)->ip_lock);
3307		credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
3308	}
3309
3310do_extend:
3311	if (ocfs2_dir_indexed(dir))
3312		credits++; /* For attaching the new dirent block to the
3313			    * dx_root */
3314
3315	handle = ocfs2_start_trans(osb, credits);
3316	if (IS_ERR(handle)) {
3317		status = PTR_ERR(handle);
3318		handle = NULL;
3319		mlog_errno(status);
3320		goto bail;
3321	}
3322
3323	status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
3324				     data_ac, meta_ac, &new_bh);
3325	if (status < 0) {
3326		mlog_errno(status);
3327		goto bail;
3328	}
3329
3330	ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), new_bh);
3331
3332	status = ocfs2_journal_access_db(handle, INODE_CACHE(dir), new_bh,
3333					 OCFS2_JOURNAL_ACCESS_CREATE);
3334	if (status < 0) {
3335		mlog_errno(status);
3336		goto bail;
3337	}
3338	memset(new_bh->b_data, 0, sb->s_blocksize);
3339
3340	de = (struct ocfs2_dir_entry *) new_bh->b_data;
3341	de->inode = 0;
3342	if (ocfs2_supports_dir_trailer(dir)) {
3343		de->rec_len = cpu_to_le16(ocfs2_dir_trailer_blk_off(sb));
3344
3345		ocfs2_init_dir_trailer(dir, new_bh, le16_to_cpu(de->rec_len));
3346
3347		if (ocfs2_dir_indexed(dir)) {
3348			status = ocfs2_dx_dir_link_trailer(dir, handle,
3349							   dx_root_bh, new_bh);
3350			if (status) {
3351				mlog_errno(status);
3352				goto bail;
3353			}
3354		}
3355	} else {
3356		de->rec_len = cpu_to_le16(sb->s_blocksize);
3357	}
3358	ocfs2_update_inode_fsync_trans(handle, dir, 1);
3359	ocfs2_journal_dirty(handle, new_bh);
3360
3361	dir_i_size += dir->i_sb->s_blocksize;
3362	i_size_write(dir, dir_i_size);
3363	dir->i_blocks = ocfs2_inode_sector_count(dir);
3364	status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
3365	if (status < 0) {
3366		mlog_errno(status);
3367		goto bail;
3368	}
3369
3370bail_bh:
3371	*new_de_bh = new_bh;
3372	get_bh(*new_de_bh);
3373bail:
3374	if (handle)
3375		ocfs2_commit_trans(osb, handle);
3376	if (drop_alloc_sem)
3377		up_write(&OCFS2_I(dir)->ip_alloc_sem);
3378
3379	if (data_ac)
3380		ocfs2_free_alloc_context(data_ac);
3381	if (meta_ac)
3382		ocfs2_free_alloc_context(meta_ac);
3383
3384	brelse(new_bh);
3385
3386	return status;
3387}
3388
3389static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
3390				   const char *name, int namelen,
3391				   struct buffer_head **ret_de_bh,
3392				   unsigned int *blocks_wanted)
3393{
3394	int ret;
3395	struct super_block *sb = dir->i_sb;
3396	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3397	struct ocfs2_dir_entry *de, *last_de = NULL;
3398	char *first_de, *de_buf, *limit;
3399	unsigned long offset = 0;
3400	unsigned int rec_len, new_rec_len, free_space;
3401
3402	/*
3403	 * This calculates how many free bytes we'd have in block zero, should
3404	 * this function force expansion to an extent tree.
3405	 */
3406	if (ocfs2_new_dir_wants_trailer(dir))
3407		free_space = ocfs2_dir_trailer_blk_off(sb) - i_size_read(dir);
3408	else
3409		free_space = dir->i_sb->s_blocksize - i_size_read(dir);
3410
3411	first_de = di->id2.i_data.id_data;
3412	de_buf = first_de;
3413	limit = de_buf + i_size_read(dir);
3414	rec_len = OCFS2_DIR_REC_LEN(namelen);
3415
3416	while (de_buf < limit) {
3417		de = (struct ocfs2_dir_entry *)de_buf;
3418
3419		if (!ocfs2_check_dir_entry(dir, de, di_bh, first_de,
3420					   i_size_read(dir), offset)) {
3421			ret = -ENOENT;
3422			goto out;
3423		}
3424		if (ocfs2_match(namelen, name, de)) {
3425			ret = -EEXIST;
3426			goto out;
3427		}
3428		/*
3429		 * No need to check for a trailing dirent record here as
3430		 * they're not used for inline dirs.
3431		 */
3432
3433		if (ocfs2_dirent_would_fit(de, rec_len)) {
3434			/* Ok, we found a spot. Return this bh and let
3435			 * the caller actually fill it in. */
3436			*ret_de_bh = di_bh;
3437			get_bh(*ret_de_bh);
3438			ret = 0;
3439			goto out;
3440		}
3441
3442		last_de = de;
3443		de_buf += le16_to_cpu(de->rec_len);
3444		offset += le16_to_cpu(de->rec_len);
3445	}
3446
3447	if (!last_de) {
3448		ret = ocfs2_error(sb, "Directory entry (#%llu: size=%lld) "
3449				  "is unexpectedly short",
3450				  (unsigned long long)OCFS2_I(dir)->ip_blkno,
3451				  i_size_read(dir));
3452		goto out;
3453	}
3454
3455	/*
3456	 * We're going to require expansion of the directory - figure
3457	 * out how many blocks we'll need so that a place for the
3458	 * dirent can be found.
3459	 */
3460	*blocks_wanted = 1;
3461	new_rec_len = le16_to_cpu(last_de->rec_len) + free_space;
3462	if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len)))
3463		*blocks_wanted = 2;
3464
3465	ret = -ENOSPC;
3466out:
3467	return ret;
3468}
3469
3470static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
3471				   int namelen, struct buffer_head **ret_de_bh)
3472{
3473	unsigned long offset;
3474	struct buffer_head *bh = NULL;
3475	unsigned short rec_len;
3476	struct ocfs2_dir_entry *de;
3477	struct super_block *sb = dir->i_sb;
3478	int status;
3479	int blocksize = dir->i_sb->s_blocksize;
3480
3481	status = ocfs2_read_dir_block(dir, 0, &bh, 0);
3482	if (status)
3483		goto bail;
3484
3485	rec_len = OCFS2_DIR_REC_LEN(namelen);
3486	offset = 0;
3487	de = (struct ocfs2_dir_entry *) bh->b_data;
3488	while (1) {
3489		if ((char *)de >= sb->s_blocksize + bh->b_data) {
3490			brelse(bh);
3491			bh = NULL;
3492
3493			if (i_size_read(dir) <= offset) {
3494				/*
3495				 * Caller will have to expand this
3496				 * directory.
3497				 */
3498				status = -ENOSPC;
3499				goto bail;
3500			}
3501			status = ocfs2_read_dir_block(dir,
3502					     offset >> sb->s_blocksize_bits,
3503					     &bh, 0);
3504			if (status)
3505				goto bail;
3506
3507			/* move to next block */
3508			de = (struct ocfs2_dir_entry *) bh->b_data;
3509		}
3510		if (!ocfs2_check_dir_entry(dir, de, bh, bh->b_data, blocksize,
3511					   offset)) {
3512			status = -ENOENT;
3513			goto bail;
3514		}
3515		if (ocfs2_match(namelen, name, de)) {
3516			status = -EEXIST;
3517			goto bail;
3518		}
3519
3520		if (ocfs2_skip_dir_trailer(dir, de, offset % blocksize,
3521					   blocksize))
3522			goto next;
3523
3524		if (ocfs2_dirent_would_fit(de, rec_len)) {
3525			/* Ok, we found a spot. Return this bh and let
3526			 * the caller actually fill it in. */
3527			*ret_de_bh = bh;
3528			get_bh(*ret_de_bh);
3529			status = 0;
3530			goto bail;
3531		}
3532next:
3533		offset += le16_to_cpu(de->rec_len);
3534		de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
3535	}
3536
3537bail:
3538	brelse(bh);
3539	if (status)
3540		mlog_errno(status);
3541
3542	return status;
3543}
3544
3545static int dx_leaf_sort_cmp(const void *a, const void *b)
3546{
3547	const struct ocfs2_dx_entry *entry1 = a;
3548	const struct ocfs2_dx_entry *entry2 = b;
3549	u32 major_hash1 = le32_to_cpu(entry1->dx_major_hash);
3550	u32 major_hash2 = le32_to_cpu(entry2->dx_major_hash);
3551	u32 minor_hash1 = le32_to_cpu(entry1->dx_minor_hash);
3552	u32 minor_hash2 = le32_to_cpu(entry2->dx_minor_hash);
3553
3554	if (major_hash1 > major_hash2)
3555		return 1;
3556	if (major_hash1 < major_hash2)
3557		return -1;
3558
3559	/*
3560	 * It is not strictly necessary to sort by minor
3561	 */
3562	if (minor_hash1 > minor_hash2)
3563		return 1;
3564	if (minor_hash1 < minor_hash2)
3565		return -1;
3566	return 0;
3567}
3568
3569static int ocfs2_dx_leaf_same_major(struct ocfs2_dx_leaf *dx_leaf)
3570{
3571	struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3572	int i, num = le16_to_cpu(dl_list->de_num_used);
3573
3574	for (i = 0; i < (num - 1); i++) {
3575		if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) !=
3576		    le32_to_cpu(dl_list->de_entries[i + 1].dx_major_hash))
3577			return 0;
3578	}
3579
3580	return 1;
3581}
3582
3583/*
3584 * Find the optimal value to split this leaf on. This expects the leaf
3585 * entries to be in sorted order.
3586 *
3587 * leaf_cpos is the cpos of the leaf we're splitting. insert_hash is
3588 * the hash we want to insert.
3589 *
3590 * This function is only concerned with the major hash - that which
3591 * determines which cluster an item belongs to.
3592 */
3593static int ocfs2_dx_dir_find_leaf_split(struct ocfs2_dx_leaf *dx_leaf,
3594					u32 leaf_cpos, u32 insert_hash,
3595					u32 *split_hash)
3596{
3597	struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
3598	int i, num_used = le16_to_cpu(dl_list->de_num_used);
3599	int allsame;
3600
3601	/*
3602	 * There's a couple rare, but nasty corner cases we have to
3603	 * check for here. All of them involve a leaf where all value
3604	 * have the same hash, which is what we look for first.
3605	 *
3606	 * Most of the time, all of the above is false, and we simply
3607	 * pick the median value for a split.
3608	 */
3609	allsame = ocfs2_dx_leaf_same_major(dx_leaf);
3610	if (allsame) {
3611		u32 val = le32_to_cpu(dl_list->de_entries[0].dx_major_hash);
3612
3613		if (val == insert_hash) {
3614			/*
3615			 * No matter where we would choose to split,
3616			 * the new entry would want to occupy the same
3617			 * block as these. Since there's no space left
3618			 * in their existing block, we know there
3619			 * won't be space after the split.
3620			 */
3621			return -ENOSPC;
3622		}
3623
3624		if (val == leaf_cpos) {
3625			/*
3626			 * Because val is the same as leaf_cpos (which
3627			 * is the smallest value this leaf can have),
3628			 * yet is not equal to insert_hash, then we
3629			 * know that insert_hash *must* be larger than
3630			 * val (and leaf_cpos). At least cpos+1 in value.
3631			 *
3632			 * We also know then, that there cannot be an
3633			 * adjacent extent (otherwise we'd be looking
3634			 * at it). Choosing this value gives us a
3635			 * chance to get some contiguousness.
3636			 */
3637			*split_hash = leaf_cpos + 1;
3638			return 0;
3639		}
3640
3641		if (val > insert_hash) {
3642			/*
3643			 * val can not be the same as insert hash, and
3644			 * also must be larger than leaf_cpos. Also,
3645			 * we know that there can't be a leaf between
3646			 * cpos and val, otherwise the entries with
3647			 * hash 'val' would be there.
3648			 */
3649			*split_hash = val;
3650			return 0;
3651		}
3652
3653		*split_hash = insert_hash;
3654		return 0;
3655	}
3656
3657	/*
3658	 * Since the records are sorted and the checks above
3659	 * guaranteed that not all records in this block are the same,
3660	 * we simple travel forward, from the median, and pick the 1st
3661	 * record whose value is larger than leaf_cpos.
3662	 */
3663	for (i = (num_used / 2); i < num_used; i++)
3664		if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) >
3665		    leaf_cpos)
3666			break;
3667
3668	BUG_ON(i == num_used); /* Should be impossible */
3669	*split_hash = le32_to_cpu(dl_list->de_entries[i].dx_major_hash);
3670	return 0;
3671}
3672
3673/*
3674 * Transfer all entries in orig_dx_leaves whose major hash is equal to or
3675 * larger than split_hash into new_dx_leaves. We use a temporary
3676 * buffer (tmp_dx_leaf) to make the changes to the original leaf blocks.
3677 *
3678 * Since the block offset inside a leaf (cluster) is a constant mask
3679 * of minor_hash, we can optimize - an item at block offset X within
3680 * the original cluster, will be at offset X within the new cluster.
3681 */
3682static void ocfs2_dx_dir_transfer_leaf(struct inode *dir, u32 split_hash,
3683				       handle_t *handle,
3684				       struct ocfs2_dx_leaf *tmp_dx_leaf,
3685				       struct buffer_head **orig_dx_leaves,
3686				       struct buffer_head **new_dx_leaves,
3687				       int num_dx_leaves)
3688{
3689	int i, j, num_used;
3690	u32 major_hash;
3691	struct ocfs2_dx_leaf *orig_dx_leaf, *new_dx_leaf;
3692	struct ocfs2_dx_entry_list *orig_list, *tmp_list;
3693	struct ocfs2_dx_entry *dx_entry;
3694
3695	tmp_list = &tmp_dx_leaf->dl_list;
3696
3697	for (i = 0; i < num_dx_leaves; i++) {
3698		orig_dx_leaf = (struct ocfs2_dx_leaf *) orig_dx_leaves[i]->b_data;
3699		orig_list = &orig_dx_leaf->dl_list;
3700		new_dx_leaf = (struct ocfs2_dx_leaf *) new_dx_leaves[i]->b_data;
3701
3702		num_used = le16_to_cpu(orig_list->de_num_used);
3703
3704		memcpy(tmp_dx_leaf, orig_dx_leaf, dir->i_sb->s_blocksize);
3705		tmp_list->de_num_used = cpu_to_le16(0);
3706		memset(&tmp_list->de_entries, 0, sizeof(*dx_entry)*num_used);
3707
3708		for (j = 0; j < num_used; j++) {
3709			dx_entry = &orig_list->de_entries[j];
3710			major_hash = le32_to_cpu(dx_entry->dx_major_hash);
3711			if (major_hash >= split_hash)
3712				ocfs2_dx_dir_leaf_insert_tail(new_dx_leaf,
3713							      dx_entry);
3714			else
3715				ocfs2_dx_dir_leaf_insert_tail(tmp_dx_leaf,
3716							      dx_entry);
3717		}
3718		memcpy(orig_dx_leaf, tmp_dx_leaf, dir->i_sb->s_blocksize);
3719
3720		ocfs2_journal_dirty(handle, orig_dx_leaves[i]);
3721		ocfs2_journal_dirty(handle, new_dx_leaves[i]);
3722	}
3723}
3724
3725static int ocfs2_dx_dir_rebalance_credits(struct ocfs2_super *osb,
3726					  struct ocfs2_dx_root_block *dx_root)
3727{
3728	int credits = ocfs2_clusters_to_blocks(osb->sb, 3);
3729
3730	credits += ocfs2_calc_extend_credits(osb->sb, &dx_root->dr_list);
3731	credits += ocfs2_quota_trans_credits(osb->sb);
3732	return credits;
3733}
3734
3735/*
3736 * Find the median value in dx_leaf_bh and allocate a new leaf to move
3737 * half our entries into.
3738 */
3739static int ocfs2_dx_dir_rebalance(struct ocfs2_super *osb, struct inode *dir,
3740				  struct buffer_head *dx_root_bh,
3741				  struct buffer_head *dx_leaf_bh,
3742				  struct ocfs2_dx_hinfo *hinfo, u32 leaf_cpos,
3743				  u64 leaf_blkno)
3744{
3745	struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
3746	int credits, ret, i, num_used, did_quota = 0;
3747	u32 cpos, split_hash, insert_hash = hinfo->major_hash;
3748	u64 orig_leaves_start;
3749	int num_dx_leaves;
3750	struct buffer_head **orig_dx_leaves = NULL;
3751	struct buffer_head **new_dx_leaves = NULL;
3752	struct ocfs2_alloc_context *data_ac = NULL, *meta_ac = NULL;
3753	struct ocfs2_extent_tree et;
3754	handle_t *handle = NULL;
3755	struct ocfs2_dx_root_block *dx_root;
3756	struct ocfs2_dx_leaf *tmp_dx_leaf = NULL;
3757
3758	trace_ocfs2_dx_dir_rebalance((unsigned long long)OCFS2_I(dir)->ip_blkno,
3759				     (unsigned long long)leaf_blkno,
3760				     insert_hash);
3761
3762	ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
3763
3764	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3765	/*
3766	 * XXX: This is a rather large limit. We should use a more
3767	 * realistic value.
3768	 */
3769	if (le32_to_cpu(dx_root->dr_clusters) == UINT_MAX)
3770		return -ENOSPC;
3771
3772	num_used = le16_to_cpu(dx_leaf->dl_list.de_num_used);
3773	if (num_used < le16_to_cpu(dx_leaf->dl_list.de_count)) {
3774		mlog(ML_ERROR, "DX Dir: %llu, Asked to rebalance empty leaf: "
3775		     "%llu, %d\n", (unsigned long long)OCFS2_I(dir)->ip_blkno,
3776		     (unsigned long long)leaf_blkno, num_used);
3777		ret = -EIO;
3778		goto out;
3779	}
3780
3781	orig_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
3782	if (!orig_dx_leaves) {
3783		ret = -ENOMEM;
3784		mlog_errno(ret);
3785		goto out;
3786	}
3787
3788	new_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, NULL);
3789	if (!new_dx_leaves) {
3790		ret = -ENOMEM;
3791		mlog_errno(ret);
3792		goto out;
3793	}
3794
3795	ret = ocfs2_lock_allocators(dir, &et, 1, 0, &data_ac, &meta_ac);
3796	if (ret) {
3797		if (ret != -ENOSPC)
3798			mlog_errno(ret);
3799		goto out;
3800	}
3801
3802	credits = ocfs2_dx_dir_rebalance_credits(osb, dx_root);
3803	handle = ocfs2_start_trans(osb, credits);
3804	if (IS_ERR(handle)) {
3805		ret = PTR_ERR(handle);
3806		handle = NULL;
3807		mlog_errno(ret);
3808		goto out;
3809	}
3810
3811	ret = dquot_alloc_space_nodirty(dir,
3812				       ocfs2_clusters_to_bytes(dir->i_sb, 1));
3813	if (ret)
3814		goto out_commit;
3815	did_quota = 1;
3816
3817	ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
3818				      OCFS2_JOURNAL_ACCESS_WRITE);
3819	if (ret) {
3820		mlog_errno(ret);
3821		goto out_commit;
3822	}
3823
3824	/*
3825	 * This block is changing anyway, so we can sort it in place.
3826	 */
3827	sort(dx_leaf->dl_list.de_entries, num_used,
3828	     sizeof(struct ocfs2_dx_entry), dx_leaf_sort_cmp,
3829	     NULL);
3830
3831	ocfs2_journal_dirty(handle, dx_leaf_bh);
3832
3833	ret = ocfs2_dx_dir_find_leaf_split(dx_leaf, leaf_cpos, insert_hash,
3834					   &split_hash);
3835	if (ret) {
3836		mlog_errno(ret);
3837		goto  out_commit;
3838	}
3839
3840	trace_ocfs2_dx_dir_rebalance_split(leaf_cpos, split_hash, insert_hash);
3841
3842	/*
3843	 * We have to carefully order operations here. There are items
3844	 * which want to be in the new cluster before insert, but in
3845	 * order to put those items in the new cluster, we alter the
3846	 * old cluster. A failure to insert gets nasty.
3847	 *
3848	 * So, start by reserving writes to the old
3849	 * cluster. ocfs2_dx_dir_new_cluster will reserve writes on
3850	 * the new cluster for us, before inserting it. The insert
3851	 * won't happen if there's an error before that. Once the
3852	 * insert is done then, we can transfer from one leaf into the
3853	 * other without fear of hitting any error.
3854	 */
3855
3856	/*
3857	 * The leaf transfer wants some scratch space so that we don't
3858	 * wind up doing a bunch of expensive memmove().
3859	 */
3860	tmp_dx_leaf = kmalloc(osb->sb->s_blocksize, GFP_NOFS);
3861	if (!tmp_dx_leaf) {
3862		ret = -ENOMEM;
3863		mlog_errno(ret);
3864		goto out_commit;
3865	}
3866
3867	orig_leaves_start = ocfs2_block_to_cluster_start(dir->i_sb, leaf_blkno);
3868	ret = ocfs2_read_dx_leaves(dir, orig_leaves_start, num_dx_leaves,
3869				   orig_dx_leaves);
3870	if (ret) {
3871		mlog_errno(ret);
3872		goto out_commit;
3873	}
3874
3875	cpos = split_hash;
3876	ret = ocfs2_dx_dir_new_cluster(dir, &et, cpos, handle,
3877				       data_ac, meta_ac, new_dx_leaves,
3878				       num_dx_leaves);
3879	if (ret) {
3880		mlog_errno(ret);
3881		goto out_commit;
3882	}
3883
3884	for (i = 0; i < num_dx_leaves; i++) {
3885		ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
3886					      orig_dx_leaves[i],
3887					      OCFS2_JOURNAL_ACCESS_WRITE);
3888		if (ret) {
3889			mlog_errno(ret);
3890			goto out_commit;
3891		}
3892
3893		ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
3894					      new_dx_leaves[i],
3895					      OCFS2_JOURNAL_ACCESS_WRITE);
3896		if (ret) {
3897			mlog_errno(ret);
3898			goto out_commit;
3899		}
3900	}
3901
3902	ocfs2_dx_dir_transfer_leaf(dir, split_hash, handle, tmp_dx_leaf,
3903				   orig_dx_leaves, new_dx_leaves, num_dx_leaves);
3904
3905out_commit:
3906	if (ret < 0 && did_quota)
3907		dquot_free_space_nodirty(dir,
3908				ocfs2_clusters_to_bytes(dir->i_sb, 1));
3909
3910	ocfs2_update_inode_fsync_trans(handle, dir, 1);
3911	ocfs2_commit_trans(osb, handle);
3912
3913out:
3914	if (orig_dx_leaves || new_dx_leaves) {
3915		for (i = 0; i < num_dx_leaves; i++) {
3916			if (orig_dx_leaves)
3917				brelse(orig_dx_leaves[i]);
3918			if (new_dx_leaves)
3919				brelse(new_dx_leaves[i]);
3920		}
3921		kfree(orig_dx_leaves);
3922		kfree(new_dx_leaves);
3923	}
3924
3925	if (meta_ac)
3926		ocfs2_free_alloc_context(meta_ac);
3927	if (data_ac)
3928		ocfs2_free_alloc_context(data_ac);
3929
3930	kfree(tmp_dx_leaf);
3931	return ret;
3932}
3933
3934static int ocfs2_find_dir_space_dx(struct ocfs2_super *osb, struct inode *dir,
3935				   struct buffer_head *di_bh,
3936				   struct buffer_head *dx_root_bh,
3937				   const char *name, int namelen,
3938				   struct ocfs2_dir_lookup_result *lookup)
3939{
3940	int ret, rebalanced = 0;
3941	struct ocfs2_dx_root_block *dx_root;
3942	struct buffer_head *dx_leaf_bh = NULL;
3943	struct ocfs2_dx_leaf *dx_leaf;
3944	u64 blkno;
3945	u32 leaf_cpos;
3946
3947	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
3948
3949restart_search:
3950	ret = ocfs2_dx_dir_lookup(dir, &dx_root->dr_list, &lookup->dl_hinfo,
3951				  &leaf_cpos, &blkno);
3952	if (ret) {
3953		mlog_errno(ret);
3954		goto out;
3955	}
3956
3957	ret = ocfs2_read_dx_leaf(dir, blkno, &dx_leaf_bh);
3958	if (ret) {
3959		mlog_errno(ret);
3960		goto out;
3961	}
3962
3963	dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
3964
3965	if (le16_to_cpu(dx_leaf->dl_list.de_num_used) >=
3966	    le16_to_cpu(dx_leaf->dl_list.de_count)) {
3967		if (rebalanced) {
3968			/*
3969			 * Rebalancing should have provided us with
3970			 * space in an appropriate leaf.
3971			 *
3972			 * XXX: Is this an abnormal condition then?
3973			 * Should we print a message here?
3974			 */
3975			ret = -ENOSPC;
3976			goto out;
3977		}
3978
3979		ret = ocfs2_dx_dir_rebalance(osb, dir, dx_root_bh, dx_leaf_bh,
3980					     &lookup->dl_hinfo, leaf_cpos,
3981					     blkno);
3982		if (ret) {
3983			if (ret != -ENOSPC)
3984				mlog_errno(ret);
3985			goto out;
3986		}
3987
3988		/*
3989		 * Restart the lookup. The rebalance might have
3990		 * changed which block our item fits into. Mark our
3991		 * progress, so we only execute this once.
3992		 */
3993		brelse(dx_leaf_bh);
3994		dx_leaf_bh = NULL;
3995		rebalanced = 1;
3996		goto restart_search;
3997	}
3998
3999	lookup->dl_dx_leaf_bh = dx_leaf_bh;
4000	dx_leaf_bh = NULL;
4001
4002out:
4003	brelse(dx_leaf_bh);
4004	return ret;
4005}
4006
4007static int ocfs2_search_dx_free_list(struct inode *dir,
4008				     struct buffer_head *dx_root_bh,
4009				     int namelen,
4010				     struct ocfs2_dir_lookup_result *lookup)
4011{
4012	int ret = -ENOSPC;
4013	struct buffer_head *leaf_bh = NULL, *prev_leaf_bh = NULL;
4014	struct ocfs2_dir_block_trailer *db;
4015	u64 next_block;
4016	int rec_len = OCFS2_DIR_REC_LEN(namelen);
4017	struct ocfs2_dx_root_block *dx_root;
4018
4019	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4020	next_block = le64_to_cpu(dx_root->dr_free_blk);
4021
4022	while (next_block) {
4023		brelse(prev_leaf_bh);
4024		prev_leaf_bh = leaf_bh;
4025		leaf_bh = NULL;
4026
4027		ret = ocfs2_read_dir_block_direct(dir, next_block, &leaf_bh);
4028		if (ret) {
4029			mlog_errno(ret);
4030			goto out;
4031		}
4032
4033		db = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
4034		if (rec_len <= le16_to_cpu(db->db_free_rec_len)) {
4035			lookup->dl_leaf_bh = leaf_bh;
4036			lookup->dl_prev_leaf_bh = prev_leaf_bh;
4037			leaf_bh = NULL;
4038			prev_leaf_bh = NULL;
4039			break;
4040		}
4041
4042		next_block = le64_to_cpu(db->db_free_next);
4043	}
4044
4045	if (!next_block)
4046		ret = -ENOSPC;
4047
4048out:
4049
4050	brelse(leaf_bh);
4051	brelse(prev_leaf_bh);
4052	return ret;
4053}
4054
4055static int ocfs2_expand_inline_dx_root(struct inode *dir,
4056				       struct buffer_head *dx_root_bh)
4057{
4058	int ret, num_dx_leaves, i, j, did_quota = 0;
4059	struct buffer_head **dx_leaves = NULL;
4060	struct ocfs2_extent_tree et;
4061	u64 insert_blkno;
4062	struct ocfs2_alloc_context *data_ac = NULL;
4063	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4064	handle_t *handle = NULL;
4065	struct ocfs2_dx_root_block *dx_root;
4066	struct ocfs2_dx_entry_list *entry_list;
4067	struct ocfs2_dx_entry *dx_entry;
4068	struct ocfs2_dx_leaf *target_leaf;
4069
4070	ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
4071	if (ret) {
4072		mlog_errno(ret);
4073		goto out;
4074	}
4075
4076	dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
4077	if (!dx_leaves) {
4078		ret = -ENOMEM;
4079		mlog_errno(ret);
4080		goto out;
4081	}
4082
4083	handle = ocfs2_start_trans(osb, ocfs2_calc_dxi_expand_credits(osb->sb));
4084	if (IS_ERR(handle)) {
4085		ret = PTR_ERR(handle);
4086		mlog_errno(ret);
4087		goto out;
4088	}
4089
4090	ret = dquot_alloc_space_nodirty(dir,
4091				       ocfs2_clusters_to_bytes(osb->sb, 1));
4092	if (ret)
4093		goto out_commit;
4094	did_quota = 1;
4095
4096	/*
4097	 * We do this up front, before the allocation, so that a
4098	 * failure to add the dx_root_bh to the journal won't result
4099	 * us losing clusters.
4100	 */
4101	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
4102				      OCFS2_JOURNAL_ACCESS_WRITE);
4103	if (ret) {
4104		mlog_errno(ret);
4105		goto out_commit;
4106	}
4107
4108	ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac, dx_leaves,
4109					 num_dx_leaves, &insert_blkno);
4110	if (ret) {
4111		mlog_errno(ret);
4112		goto out_commit;
4113	}
4114
4115	/*
4116	 * Transfer the entries from our dx_root into the appropriate
4117	 * block
4118	 */
4119	dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4120	entry_list = &dx_root->dr_entries;
4121
4122	for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
4123		dx_entry = &entry_list->de_entries[i];
4124
4125		j = __ocfs2_dx_dir_hash_idx(osb,
4126					    le32_to_cpu(dx_entry->dx_minor_hash));
4127		target_leaf = (struct ocfs2_dx_leaf *)dx_leaves[j]->b_data;
4128
4129		ocfs2_dx_dir_leaf_insert_tail(target_leaf, dx_entry);
4130
4131		/* Each leaf has been passed to the journal already
4132		 * via __ocfs2_dx_dir_new_cluster() */
4133	}
4134
4135	dx_root->dr_flags &= ~OCFS2_DX_FLAG_INLINE;
4136
4137	dx_root->dr_list.l_tree_depth = 0;
4138	dx_root->dr_list.l_count =
4139		cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
4140	dx_root->dr_list.l_next_free_rec = 0;
4141	memset(&dx_root->dr_list.l_recs, 0,
4142	       osb->sb->s_blocksize -
4143	       (offsetof(struct ocfs2_dx_root_block, dr_list) +
4144		offsetof(struct ocfs2_extent_list, l_recs)));
4145
4146	/* This should never fail considering we start with an empty
4147	 * dx_root. */
4148	ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
4149	ret = ocfs2_insert_extent(handle, &et, 0, insert_blkno, 1, 0, NULL);
4150	if (ret)
4151		mlog_errno(ret);
4152	did_quota = 0;
4153
4154	ocfs2_update_inode_fsync_trans(handle, dir, 1);
4155	ocfs2_journal_dirty(handle, dx_root_bh);
4156
4157out_commit:
4158	if (ret < 0 && did_quota)
4159		dquot_free_space_nodirty(dir,
4160					  ocfs2_clusters_to_bytes(dir->i_sb, 1));
4161
4162	ocfs2_commit_trans(osb, handle);
4163
4164out:
4165	if (data_ac)
4166		ocfs2_free_alloc_context(data_ac);
4167
4168	if (dx_leaves) {
4169		for (i = 0; i < num_dx_leaves; i++)
4170			brelse(dx_leaves[i]);
4171		kfree(dx_leaves);
4172	}
4173	return ret;
4174}
4175
4176static int ocfs2_inline_dx_has_space(struct buffer_head *dx_root_bh)
4177{
4178	struct ocfs2_dx_root_block *dx_root;
4179	struct ocfs2_dx_entry_list *entry_list;
4180
4181	dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4182	entry_list = &dx_root->dr_entries;
4183
4184	if (le16_to_cpu(entry_list->de_num_used) >=
4185	    le16_to_cpu(entry_list->de_count))
4186		return -ENOSPC;
4187
4188	return 0;
4189}
4190
4191static int ocfs2_prepare_dx_dir_for_insert(struct inode *dir,
4192					   struct buffer_head *di_bh,
4193					   const char *name,
4194					   int namelen,
4195					   struct ocfs2_dir_lookup_result *lookup)
4196{
4197	int ret, free_dx_root = 1;
4198	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4199	struct buffer_head *dx_root_bh = NULL;
4200	struct buffer_head *leaf_bh = NULL;
4201	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4202	struct ocfs2_dx_root_block *dx_root;
4203
4204	ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4205	if (ret) {
4206		mlog_errno(ret);
4207		goto out;
4208	}
4209
4210	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4211	if (le32_to_cpu(dx_root->dr_num_entries) == OCFS2_DX_ENTRIES_MAX) {
4212		ret = -ENOSPC;
4213		mlog_errno(ret);
4214		goto out;
4215	}
4216
4217	if (ocfs2_dx_root_inline(dx_root)) {
4218		ret = ocfs2_inline_dx_has_space(dx_root_bh);
4219
4220		if (ret == 0)
4221			goto search_el;
4222
4223		/*
4224		 * We ran out of room in the root block. Expand it to
4225		 * an extent, then allow ocfs2_find_dir_space_dx to do
4226		 * the rest.
4227		 */
4228		ret = ocfs2_expand_inline_dx_root(dir, dx_root_bh);
4229		if (ret) {
4230			mlog_errno(ret);
4231			goto out;
4232		}
4233	}
4234
4235	/*
4236	 * Insert preparation for an indexed directory is split into two
4237	 * steps. The call to find_dir_space_dx reserves room in the index for
4238	 * an additional item. If we run out of space there, it's a real error
4239	 * we can't continue on.
4240	 */
4241	ret = ocfs2_find_dir_space_dx(osb, dir, di_bh, dx_root_bh, name,
4242				      namelen, lookup);
4243	if (ret) {
4244		mlog_errno(ret);
4245		goto out;
4246	}
4247
4248search_el:
4249	/*
4250	 * Next, we need to find space in the unindexed tree. This call
4251	 * searches using the free space linked list. If the unindexed tree
4252	 * lacks sufficient space, we'll expand it below. The expansion code
4253	 * is smart enough to add any new blocks to the free space list.
4254	 */
4255	ret = ocfs2_search_dx_free_list(dir, dx_root_bh, namelen, lookup);
4256	if (ret && ret != -ENOSPC) {
4257		mlog_errno(ret);
4258		goto out;
4259	}
4260
4261	/* Do this up here - ocfs2_extend_dir might need the dx_root */
4262	lookup->dl_dx_root_bh = dx_root_bh;
4263	free_dx_root = 0;
4264
4265	if (ret == -ENOSPC) {
4266		ret = ocfs2_extend_dir(osb, dir, di_bh, 1, lookup, &leaf_bh);
4267
4268		if (ret) {
4269			mlog_errno(ret);
4270			goto out;
4271		}
4272
4273		/*
4274		 * We make the assumption here that new leaf blocks are added
4275		 * to the front of our free list.
4276		 */
4277		lookup->dl_prev_leaf_bh = NULL;
4278		lookup->dl_leaf_bh = leaf_bh;
4279	}
4280
4281out:
4282	if (free_dx_root)
4283		brelse(dx_root_bh);
4284	return ret;
4285}
4286
4287/*
4288 * Get a directory ready for insert. Any directory allocation required
4289 * happens here. Success returns zero, and enough context in the dir
4290 * lookup result that ocfs2_add_entry() will be able complete the task
4291 * with minimal performance impact.
4292 */
4293int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
4294				 struct inode *dir,
4295				 struct buffer_head *parent_fe_bh,
4296				 const char *name,
4297				 int namelen,
4298				 struct ocfs2_dir_lookup_result *lookup)
4299{
4300	int ret;
4301	unsigned int blocks_wanted = 1;
4302	struct buffer_head *bh = NULL;
4303
4304	trace_ocfs2_prepare_dir_for_insert(
4305		(unsigned long long)OCFS2_I(dir)->ip_blkno, namelen);
4306
4307	/*
4308	 * Do this up front to reduce confusion.
4309	 *
4310	 * The directory might start inline, then be turned into an
4311	 * indexed one, in which case we'd need to hash deep inside
4312	 * ocfs2_find_dir_space_id(). Since
4313	 * ocfs2_prepare_dx_dir_for_insert() also needs this hash
4314	 * done, there seems no point in spreading out the calls. We
4315	 * can optimize away the case where the file system doesn't
4316	 * support indexing.
4317	 */
4318	if (ocfs2_supports_indexed_dirs(osb))
4319		ocfs2_dx_dir_name_hash(dir, name, namelen, &lookup->dl_hinfo);
4320
4321	if (ocfs2_dir_indexed(dir)) {
4322		ret = ocfs2_prepare_dx_dir_for_insert(dir, parent_fe_bh,
4323						      name, namelen, lookup);
4324		if (ret)
4325			mlog_errno(ret);
4326		goto out;
4327	}
4328
4329	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4330		ret = ocfs2_find_dir_space_id(dir, parent_fe_bh, name,
4331					      namelen, &bh, &blocks_wanted);
4332	} else
4333		ret = ocfs2_find_dir_space_el(dir, name, namelen, &bh);
4334
4335	if (ret && ret != -ENOSPC) {
4336		mlog_errno(ret);
4337		goto out;
4338	}
4339
4340	if (ret == -ENOSPC) {
4341		/*
4342		 * We have to expand the directory to add this name.
4343		 */
4344		BUG_ON(bh);
4345
4346		ret = ocfs2_extend_dir(osb, dir, parent_fe_bh, blocks_wanted,
4347				       lookup, &bh);
4348		if (ret) {
4349			if (ret != -ENOSPC)
4350				mlog_errno(ret);
4351			goto out;
4352		}
4353
4354		BUG_ON(!bh);
4355	}
4356
4357	lookup->dl_leaf_bh = bh;
4358	bh = NULL;
4359out:
4360	brelse(bh);
4361	return ret;
4362}
4363
4364static int ocfs2_dx_dir_remove_index(struct inode *dir,
4365				     struct buffer_head *di_bh,
4366				     struct buffer_head *dx_root_bh)
4367{
4368	int ret;
4369	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4370	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4371	struct ocfs2_dx_root_block *dx_root;
4372	struct inode *dx_alloc_inode = NULL;
4373	struct buffer_head *dx_alloc_bh = NULL;
4374	handle_t *handle;
4375	u64 blk;
4376	u16 bit;
4377	u64 bg_blkno;
4378
4379	dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
4380
4381	dx_alloc_inode = ocfs2_get_system_file_inode(osb,
4382					EXTENT_ALLOC_SYSTEM_INODE,
4383					le16_to_cpu(dx_root->dr_suballoc_slot));
4384	if (!dx_alloc_inode) {
4385		ret = -ENOMEM;
4386		mlog_errno(ret);
4387		goto out;
4388	}
4389	inode_lock(dx_alloc_inode);
4390
4391	ret = ocfs2_inode_lock(dx_alloc_inode, &dx_alloc_bh, 1);
4392	if (ret) {
4393		mlog_errno(ret);
4394		goto out_mutex;
4395	}
4396
4397	handle = ocfs2_start_trans(osb, OCFS2_DX_ROOT_REMOVE_CREDITS);
4398	if (IS_ERR(handle)) {
4399		ret = PTR_ERR(handle);
4400		mlog_errno(ret);
4401		goto out_unlock;
4402	}
4403
4404	ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
4405				      OCFS2_JOURNAL_ACCESS_WRITE);
4406	if (ret) {
4407		mlog_errno(ret);
4408		goto out_commit;
4409	}
4410
4411	spin_lock(&OCFS2_I(dir)->ip_lock);
4412	OCFS2_I(dir)->ip_dyn_features &= ~OCFS2_INDEXED_DIR_FL;
4413	di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
4414	spin_unlock(&OCFS2_I(dir)->ip_lock);
4415	di->i_dx_root = cpu_to_le64(0ULL);
4416	ocfs2_update_inode_fsync_trans(handle, dir, 1);
4417
4418	ocfs2_journal_dirty(handle, di_bh);
4419
4420	blk = le64_to_cpu(dx_root->dr_blkno);
4421	bit = le16_to_cpu(dx_root->dr_suballoc_bit);
4422	if (dx_root->dr_suballoc_loc)
4423		bg_blkno = le64_to_cpu(dx_root->dr_suballoc_loc);
4424	else
4425		bg_blkno = ocfs2_which_suballoc_group(blk, bit);
4426	ret = ocfs2_free_suballoc_bits(handle, dx_alloc_inode, dx_alloc_bh,
4427				       bit, bg_blkno, 1);
4428	if (ret)
4429		mlog_errno(ret);
4430
4431out_commit:
4432	ocfs2_commit_trans(osb, handle);
4433
4434out_unlock:
4435	ocfs2_inode_unlock(dx_alloc_inode, 1);
4436
4437out_mutex:
4438	inode_unlock(dx_alloc_inode);
4439	brelse(dx_alloc_bh);
4440out:
4441	iput(dx_alloc_inode);
4442	return ret;
4443}
4444
4445int ocfs2_dx_dir_truncate(struct inode *dir, struct buffer_head *di_bh)
4446{
4447	int ret;
4448	unsigned int clen;
4449	u32 major_hash = UINT_MAX, p_cpos, cpos;
4450	u64 blkno;
4451	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
4452	struct buffer_head *dx_root_bh = NULL;
4453	struct ocfs2_dx_root_block *dx_root;
4454	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
4455	struct ocfs2_cached_dealloc_ctxt dealloc;
4456	struct ocfs2_extent_tree et;
4457
4458	ocfs2_init_dealloc_ctxt(&dealloc);
4459
4460	if (!ocfs2_dir_indexed(dir))
4461		return 0;
4462
4463	ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
4464	if (ret) {
4465		mlog_errno(ret);
4466		goto out;
4467	}
4468	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
4469
4470	if (ocfs2_dx_root_inline(dx_root))
4471		goto remove_index;
4472
4473	ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
4474
4475	/* XXX: What if dr_clusters is too large? */
4476	while (le32_to_cpu(dx_root->dr_clusters)) {
4477		ret = ocfs2_dx_dir_lookup_rec(dir, &dx_root->dr_list,
4478					      major_hash, &cpos, &blkno, &clen);
4479		if (ret) {
4480			mlog_errno(ret);
4481			goto out;
4482		}
4483
4484		p_cpos = ocfs2_blocks_to_clusters(dir->i_sb, blkno);
4485
4486		ret = ocfs2_remove_btree_range(dir, &et, cpos, p_cpos, clen, 0,
4487					       &dealloc, 0, false);
4488		if (ret) {
4489			mlog_errno(ret);
4490			goto out;
4491		}
4492
4493		if (cpos == 0)
4494			break;
4495
4496		major_hash = cpos - 1;
4497	}
4498
4499remove_index:
4500	ret = ocfs2_dx_dir_remove_index(dir, di_bh, dx_root_bh);
4501	if (ret) {
4502		mlog_errno(ret);
4503		goto out;
4504	}
4505
4506	ocfs2_remove_from_cache(INODE_CACHE(dir), dx_root_bh);
4507out:
4508	ocfs2_schedule_truncate_log_flush(osb, 1);
4509	ocfs2_run_deallocs(osb, &dealloc);
4510
4511	brelse(dx_root_bh);
4512	return ret;
4513}