PageRenderTime 120ms CodeModel.GetById 2ms app.highlight 105ms RepoModel.GetById 1ms app.codeStats 0ms

/linux-2.6.21.x/fs/gfs2/dir.c

https://bitbucket.org/altlc/wive-rtnl-ralink-rt305x-routers-firmware-amod
C | 1964 lines | 1390 code | 291 blank | 283 comment | 248 complexity | b6a93a940adf8203d692e551d05dd20a MD5 | raw file
   1/*
   2 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3 * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   4 *
   5 * This copyrighted material is made available to anyone wishing to use,
   6 * modify, copy, or redistribute it subject to the terms and conditions
   7 * of the GNU General Public License version 2.
   8 */
   9
  10/*
  11 * Implements Extendible Hashing as described in:
  12 *   "Extendible Hashing" by Fagin, et al in
  13 *     __ACM Trans. on Database Systems__, Sept 1979.
  14 *
  15 *
  16 * Here's the layout of dirents which is essentially the same as that of ext2
  17 * within a single block. The field de_name_len is the number of bytes
  18 * actually required for the name (no null terminator). The field de_rec_len
  19 * is the number of bytes allocated to the dirent. The offset of the next
  20 * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
  21 * deleted, the preceding dirent inherits its allocated space, ie
  22 * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
  23 * by adding de_rec_len to the current dirent, this essentially causes the
  24 * deleted dirent to get jumped over when iterating through all the dirents.
  25 *
  26 * When deleting the first dirent in a block, there is no previous dirent so
  27 * the field de_ino is set to zero to designate it as deleted. When allocating
  28 * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
  29 * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
  30 * dirent is allocated. Otherwise it must go through all the 'used' dirents
  31 * searching for one in which the amount of total space minus the amount of
  32 * used space will provide enough space for the new dirent.
  33 *
  34 * There are two types of blocks in which dirents reside. In a stuffed dinode,
  35 * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
  36 * the block.  In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
  37 * beginning of the leaf block. The dirents reside in leaves when
  38 *
  39 * dip->i_di.di_flags & GFS2_DIF_EXHASH is true
  40 *
  41 * Otherwise, the dirents are "linear", within a single stuffed dinode block.
  42 *
  43 * When the dirents are in leaves, the actual contents of the directory file are
  44 * used as an array of 64-bit block pointers pointing to the leaf blocks. The
  45 * dirents are NOT in the directory file itself. There can be more than one
  46 * block pointer in the array that points to the same leaf. In fact, when a
  47 * directory is first converted from linear to exhash, all of the pointers
  48 * point to the same leaf.
  49 *
  50 * When a leaf is completely full, the size of the hash table can be
  51 * doubled unless it is already at the maximum size which is hard coded into
  52 * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
  53 * but never before the maximum hash table size has been reached.
  54 */
  55
  56#include <linux/slab.h>
  57#include <linux/spinlock.h>
  58#include <linux/buffer_head.h>
  59#include <linux/sort.h>
  60#include <linux/gfs2_ondisk.h>
  61#include <linux/crc32.h>
  62#include <linux/vmalloc.h>
  63#include <linux/lm_interface.h>
  64
  65#include "gfs2.h"
  66#include "incore.h"
  67#include "dir.h"
  68#include "glock.h"
  69#include "inode.h"
  70#include "meta_io.h"
  71#include "quota.h"
  72#include "rgrp.h"
  73#include "trans.h"
  74#include "bmap.h"
  75#include "util.h"
  76
  77#define IS_LEAF     1 /* Hashed (leaf) directory */
  78#define IS_DINODE   2 /* Linear (stuffed dinode block) directory */
  79
  80#define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
  81#define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
  82
  83typedef int (*leaf_call_t) (struct gfs2_inode *dip, u32 index, u32 len,
  84			    u64 leaf_no, void *data);
  85typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
  86			    const struct qstr *name, void *opaque);
  87
  88
  89int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
  90			    struct buffer_head **bhp)
  91{
  92	struct buffer_head *bh;
  93
  94	bh = gfs2_meta_new(ip->i_gl, block);
  95	gfs2_trans_add_bh(ip->i_gl, bh, 1);
  96	gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
  97	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
  98	*bhp = bh;
  99	return 0;
 100}
 101
 102static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
 103					struct buffer_head **bhp)
 104{
 105	struct buffer_head *bh;
 106	int error;
 107
 108	error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh);
 109	if (error)
 110		return error;
 111	if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
 112		brelse(bh);
 113		return -EIO;
 114	}
 115	*bhp = bh;
 116	return 0;
 117}
 118
 119static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
 120				  unsigned int offset, unsigned int size)
 121{
 122	struct buffer_head *dibh;
 123	int error;
 124
 125	error = gfs2_meta_inode_buffer(ip, &dibh);
 126	if (error)
 127		return error;
 128
 129	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 130	memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
 131	if (ip->i_di.di_size < offset + size)
 132		ip->i_di.di_size = offset + size;
 133	ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME_SEC;
 134	gfs2_dinode_out(ip, dibh->b_data);
 135
 136	brelse(dibh);
 137
 138	return size;
 139}
 140
 141
 142
 143/**
 144 * gfs2_dir_write_data - Write directory information to the inode
 145 * @ip: The GFS2 inode
 146 * @buf: The buffer containing information to be written
 147 * @offset: The file offset to start writing at
 148 * @size: The amount of data to write
 149 *
 150 * Returns: The number of bytes correctly written or error code
 151 */
 152static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
 153			       u64 offset, unsigned int size)
 154{
 155	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 156	struct buffer_head *dibh;
 157	u64 lblock, dblock;
 158	u32 extlen = 0;
 159	unsigned int o;
 160	int copied = 0;
 161	int error = 0;
 162
 163	if (!size)
 164		return 0;
 165
 166	if (gfs2_is_stuffed(ip) &&
 167	    offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
 168		return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
 169					      size);
 170
 171	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
 172		return -EINVAL;
 173
 174	if (gfs2_is_stuffed(ip)) {
 175		error = gfs2_unstuff_dinode(ip, NULL);
 176		if (error)
 177			return error;
 178	}
 179
 180	lblock = offset;
 181	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
 182
 183	while (copied < size) {
 184		unsigned int amount;
 185		struct buffer_head *bh;
 186		int new = 0;
 187
 188		amount = size - copied;
 189		if (amount > sdp->sd_sb.sb_bsize - o)
 190			amount = sdp->sd_sb.sb_bsize - o;
 191
 192		if (!extlen) {
 193			new = 1;
 194			error = gfs2_extent_map(&ip->i_inode, lblock, &new,
 195						&dblock, &extlen);
 196			if (error)
 197				goto fail;
 198			error = -EIO;
 199			if (gfs2_assert_withdraw(sdp, dblock))
 200				goto fail;
 201		}
 202
 203		if (amount == sdp->sd_jbsize || new)
 204			error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
 205		else
 206			error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
 207
 208		if (error)
 209			goto fail;
 210
 211		gfs2_trans_add_bh(ip->i_gl, bh, 1);
 212		memcpy(bh->b_data + o, buf, amount);
 213		brelse(bh);
 214
 215		buf += amount;
 216		copied += amount;
 217		lblock++;
 218		dblock++;
 219		extlen--;
 220
 221		o = sizeof(struct gfs2_meta_header);
 222	}
 223
 224out:
 225	error = gfs2_meta_inode_buffer(ip, &dibh);
 226	if (error)
 227		return error;
 228
 229	if (ip->i_di.di_size < offset + copied)
 230		ip->i_di.di_size = offset + copied;
 231	ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME_SEC;
 232
 233	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 234	gfs2_dinode_out(ip, dibh->b_data);
 235	brelse(dibh);
 236
 237	return copied;
 238fail:
 239	if (copied)
 240		goto out;
 241	return error;
 242}
 243
 244static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf,
 245				 u64 offset, unsigned int size)
 246{
 247	struct buffer_head *dibh;
 248	int error;
 249
 250	error = gfs2_meta_inode_buffer(ip, &dibh);
 251	if (!error) {
 252		offset += sizeof(struct gfs2_dinode);
 253		memcpy(buf, dibh->b_data + offset, size);
 254		brelse(dibh);
 255	}
 256
 257	return (error) ? error : size;
 258}
 259
 260
 261/**
 262 * gfs2_dir_read_data - Read a data from a directory inode
 263 * @ip: The GFS2 Inode
 264 * @buf: The buffer to place result into
 265 * @offset: File offset to begin jdata_readng from
 266 * @size: Amount of data to transfer
 267 *
 268 * Returns: The amount of data actually copied or the error
 269 */
 270static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf, u64 offset,
 271			      unsigned int size, unsigned ra)
 272{
 273	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 274	u64 lblock, dblock;
 275	u32 extlen = 0;
 276	unsigned int o;
 277	int copied = 0;
 278	int error = 0;
 279
 280	if (offset >= ip->i_di.di_size)
 281		return 0;
 282
 283	if (offset + size > ip->i_di.di_size)
 284		size = ip->i_di.di_size - offset;
 285
 286	if (!size)
 287		return 0;
 288
 289	if (gfs2_is_stuffed(ip))
 290		return gfs2_dir_read_stuffed(ip, buf, offset, size);
 291
 292	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
 293		return -EINVAL;
 294
 295	lblock = offset;
 296	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
 297
 298	while (copied < size) {
 299		unsigned int amount;
 300		struct buffer_head *bh;
 301		int new;
 302
 303		amount = size - copied;
 304		if (amount > sdp->sd_sb.sb_bsize - o)
 305			amount = sdp->sd_sb.sb_bsize - o;
 306
 307		if (!extlen) {
 308			new = 0;
 309			error = gfs2_extent_map(&ip->i_inode, lblock, &new,
 310						&dblock, &extlen);
 311			if (error || !dblock)
 312				goto fail;
 313			BUG_ON(extlen < 1);
 314			if (!ra)
 315				extlen = 1;
 316			bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
 317		} else {
 318			error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh);
 319			if (error)
 320				goto fail;
 321		}
 322		error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
 323		if (error) {
 324			brelse(bh);
 325			goto fail;
 326		}
 327		dblock++;
 328		extlen--;
 329		memcpy(buf, bh->b_data + o, amount);
 330		brelse(bh);
 331		buf += amount;
 332		copied += amount;
 333		lblock++;
 334		o = sizeof(struct gfs2_meta_header);
 335	}
 336
 337	return copied;
 338fail:
 339	return (copied) ? copied : error;
 340}
 341
 342static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
 343{
 344	return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
 345}
 346
 347static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
 348				     const struct qstr *name, int ret)
 349{
 350	if (!gfs2_dirent_sentinel(dent) &&
 351	    be32_to_cpu(dent->de_hash) == name->hash &&
 352	    be16_to_cpu(dent->de_name_len) == name->len &&
 353	    memcmp(dent+1, name->name, name->len) == 0)
 354		return ret;
 355	return 0;
 356}
 357
 358static int gfs2_dirent_find(const struct gfs2_dirent *dent,
 359			    const struct qstr *name,
 360			    void *opaque)
 361{
 362	return __gfs2_dirent_find(dent, name, 1);
 363}
 364
 365static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
 366			    const struct qstr *name,
 367			    void *opaque)
 368{
 369	return __gfs2_dirent_find(dent, name, 2);
 370}
 371
 372/*
 373 * name->name holds ptr to start of block.
 374 * name->len holds size of block.
 375 */
 376static int gfs2_dirent_last(const struct gfs2_dirent *dent,
 377			    const struct qstr *name,
 378			    void *opaque)
 379{
 380	const char *start = name->name;
 381	const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
 382	if (name->len == (end - start))
 383		return 1;
 384	return 0;
 385}
 386
 387static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
 388				  const struct qstr *name,
 389				  void *opaque)
 390{
 391	unsigned required = GFS2_DIRENT_SIZE(name->len);
 392	unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
 393	unsigned totlen = be16_to_cpu(dent->de_rec_len);
 394
 395	if (gfs2_dirent_sentinel(dent))
 396		actual = GFS2_DIRENT_SIZE(0);
 397	if (totlen - actual >= required)
 398		return 1;
 399	return 0;
 400}
 401
 402struct dirent_gather {
 403	const struct gfs2_dirent **pdent;
 404	unsigned offset;
 405};
 406
 407static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
 408			      const struct qstr *name,
 409			      void *opaque)
 410{
 411	struct dirent_gather *g = opaque;
 412	if (!gfs2_dirent_sentinel(dent)) {
 413		g->pdent[g->offset++] = dent;
 414	}
 415	return 0;
 416}
 417
 418/*
 419 * Other possible things to check:
 420 * - Inode located within filesystem size (and on valid block)
 421 * - Valid directory entry type
 422 * Not sure how heavy-weight we want to make this... could also check
 423 * hash is correct for example, but that would take a lot of extra time.
 424 * For now the most important thing is to check that the various sizes
 425 * are correct.
 426 */
 427static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
 428			     unsigned int size, unsigned int len, int first)
 429{
 430	const char *msg = "gfs2_dirent too small";
 431	if (unlikely(size < sizeof(struct gfs2_dirent)))
 432		goto error;
 433	msg = "gfs2_dirent misaligned";
 434	if (unlikely(offset & 0x7))
 435		goto error;
 436	msg = "gfs2_dirent points beyond end of block";
 437	if (unlikely(offset + size > len))
 438		goto error;
 439	msg = "zero inode number";
 440	if (unlikely(!first && gfs2_dirent_sentinel(dent)))
 441		goto error;
 442	msg = "name length is greater than space in dirent";
 443	if (!gfs2_dirent_sentinel(dent) &&
 444	    unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
 445		     size))
 446		goto error;
 447	return 0;
 448error:
 449	printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg,
 450	       first ? "first in block" : "not first in block");
 451	return -EIO;
 452}
 453
 454static int gfs2_dirent_offset(const void *buf)
 455{
 456	const struct gfs2_meta_header *h = buf;
 457	int offset;
 458
 459	BUG_ON(buf == NULL);
 460
 461	switch(be32_to_cpu(h->mh_type)) {
 462	case GFS2_METATYPE_LF:
 463		offset = sizeof(struct gfs2_leaf);
 464		break;
 465	case GFS2_METATYPE_DI:
 466		offset = sizeof(struct gfs2_dinode);
 467		break;
 468	default:
 469		goto wrong_type;
 470	}
 471	return offset;
 472wrong_type:
 473	printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n",
 474	       be32_to_cpu(h->mh_type));
 475	return -1;
 476}
 477
 478static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
 479					    unsigned int len, gfs2_dscan_t scan,
 480					    const struct qstr *name,
 481					    void *opaque)
 482{
 483	struct gfs2_dirent *dent, *prev;
 484	unsigned offset;
 485	unsigned size;
 486	int ret = 0;
 487
 488	ret = gfs2_dirent_offset(buf);
 489	if (ret < 0)
 490		goto consist_inode;
 491
 492	offset = ret;
 493	prev = NULL;
 494	dent = buf + offset;
 495	size = be16_to_cpu(dent->de_rec_len);
 496	if (gfs2_check_dirent(dent, offset, size, len, 1))
 497		goto consist_inode;
 498	do {
 499		ret = scan(dent, name, opaque);
 500		if (ret)
 501			break;
 502		offset += size;
 503		if (offset == len)
 504			break;
 505		prev = dent;
 506		dent = buf + offset;
 507		size = be16_to_cpu(dent->de_rec_len);
 508		if (gfs2_check_dirent(dent, offset, size, len, 0))
 509			goto consist_inode;
 510	} while(1);
 511
 512	switch(ret) {
 513	case 0:
 514		return NULL;
 515	case 1:
 516		return dent;
 517	case 2:
 518		return prev ? prev : dent;
 519	default:
 520		BUG_ON(ret > 0);
 521		return ERR_PTR(ret);
 522	}
 523
 524consist_inode:
 525	gfs2_consist_inode(GFS2_I(inode));
 526	return ERR_PTR(-EIO);
 527}
 528
 529
 530/**
 531 * dirent_first - Return the first dirent
 532 * @dip: the directory
 533 * @bh: The buffer
 534 * @dent: Pointer to list of dirents
 535 *
 536 * return first dirent whether bh points to leaf or stuffed dinode
 537 *
 538 * Returns: IS_LEAF, IS_DINODE, or -errno
 539 */
 540
 541static int dirent_first(struct gfs2_inode *dip, struct buffer_head *bh,
 542			struct gfs2_dirent **dent)
 543{
 544	struct gfs2_meta_header *h = (struct gfs2_meta_header *)bh->b_data;
 545
 546	if (be32_to_cpu(h->mh_type) == GFS2_METATYPE_LF) {
 547		if (gfs2_meta_check(GFS2_SB(&dip->i_inode), bh))
 548			return -EIO;
 549		*dent = (struct gfs2_dirent *)(bh->b_data +
 550					       sizeof(struct gfs2_leaf));
 551		return IS_LEAF;
 552	} else {
 553		if (gfs2_metatype_check(GFS2_SB(&dip->i_inode), bh, GFS2_METATYPE_DI))
 554			return -EIO;
 555		*dent = (struct gfs2_dirent *)(bh->b_data +
 556					       sizeof(struct gfs2_dinode));
 557		return IS_DINODE;
 558	}
 559}
 560
 561static int dirent_check_reclen(struct gfs2_inode *dip,
 562			       const struct gfs2_dirent *d, const void *end_p)
 563{
 564	const void *ptr = d;
 565	u16 rec_len = be16_to_cpu(d->de_rec_len);
 566
 567	if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
 568		goto broken;
 569	ptr += rec_len;
 570	if (ptr < end_p)
 571		return rec_len;
 572	if (ptr == end_p)
 573		return -ENOENT;
 574broken:
 575	gfs2_consist_inode(dip);
 576	return -EIO;
 577}
 578
 579/**
 580 * dirent_next - Next dirent
 581 * @dip: the directory
 582 * @bh: The buffer
 583 * @dent: Pointer to list of dirents
 584 *
 585 * Returns: 0 on success, error code otherwise
 586 */
 587
 588static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
 589		       struct gfs2_dirent **dent)
 590{
 591	struct gfs2_dirent *cur = *dent, *tmp;
 592	char *bh_end = bh->b_data + bh->b_size;
 593	int ret;
 594
 595	ret = dirent_check_reclen(dip, cur, bh_end);
 596	if (ret < 0)
 597		return ret;
 598
 599	tmp = (void *)cur + ret;
 600	ret = dirent_check_reclen(dip, tmp, bh_end);
 601	if (ret == -EIO)
 602		return ret;
 603
 604        /* Only the first dent could ever have de_inum.no_addr == 0 */
 605	if (gfs2_dirent_sentinel(tmp)) {
 606		gfs2_consist_inode(dip);
 607		return -EIO;
 608	}
 609
 610	*dent = tmp;
 611	return 0;
 612}
 613
 614/**
 615 * dirent_del - Delete a dirent
 616 * @dip: The GFS2 inode
 617 * @bh: The buffer
 618 * @prev: The previous dirent
 619 * @cur: The current dirent
 620 *
 621 */
 622
 623static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
 624		       struct gfs2_dirent *prev, struct gfs2_dirent *cur)
 625{
 626	u16 cur_rec_len, prev_rec_len;
 627
 628	if (gfs2_dirent_sentinel(cur)) {
 629		gfs2_consist_inode(dip);
 630		return;
 631	}
 632
 633	gfs2_trans_add_bh(dip->i_gl, bh, 1);
 634
 635	/* If there is no prev entry, this is the first entry in the block.
 636	   The de_rec_len is already as big as it needs to be.  Just zero
 637	   out the inode number and return.  */
 638
 639	if (!prev) {
 640		cur->de_inum.no_addr = 0;
 641		cur->de_inum.no_formal_ino = 0;
 642		return;
 643	}
 644
 645	/*  Combine this dentry with the previous one.  */
 646
 647	prev_rec_len = be16_to_cpu(prev->de_rec_len);
 648	cur_rec_len = be16_to_cpu(cur->de_rec_len);
 649
 650	if ((char *)prev + prev_rec_len != (char *)cur)
 651		gfs2_consist_inode(dip);
 652	if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
 653		gfs2_consist_inode(dip);
 654
 655	prev_rec_len += cur_rec_len;
 656	prev->de_rec_len = cpu_to_be16(prev_rec_len);
 657}
 658
 659/*
 660 * Takes a dent from which to grab space as an argument. Returns the
 661 * newly created dent.
 662 */
 663static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
 664					    struct gfs2_dirent *dent,
 665					    const struct qstr *name,
 666					    struct buffer_head *bh)
 667{
 668	struct gfs2_inode *ip = GFS2_I(inode);
 669	struct gfs2_dirent *ndent;
 670	unsigned offset = 0, totlen;
 671
 672	if (!gfs2_dirent_sentinel(dent))
 673		offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
 674	totlen = be16_to_cpu(dent->de_rec_len);
 675	BUG_ON(offset + name->len > totlen);
 676	gfs2_trans_add_bh(ip->i_gl, bh, 1);
 677	ndent = (struct gfs2_dirent *)((char *)dent + offset);
 678	dent->de_rec_len = cpu_to_be16(offset);
 679	gfs2_qstr2dirent(name, totlen - offset, ndent);
 680	return ndent;
 681}
 682
 683static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
 684					     struct buffer_head *bh,
 685					     const struct qstr *name)
 686{
 687	struct gfs2_dirent *dent;
 688	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
 689				gfs2_dirent_find_space, name, NULL);
 690	if (!dent || IS_ERR(dent))
 691		return dent;
 692	return gfs2_init_dirent(inode, dent, name, bh);
 693}
 694
 695static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
 696		    struct buffer_head **bhp)
 697{
 698	int error;
 699
 700	error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp);
 701	if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
 702		/* printk(KERN_INFO "block num=%llu\n", leaf_no); */
 703		error = -EIO;
 704	}
 705
 706	return error;
 707}
 708
 709/**
 710 * get_leaf_nr - Get a leaf number associated with the index
 711 * @dip: The GFS2 inode
 712 * @index:
 713 * @leaf_out:
 714 *
 715 * Returns: 0 on success, error code otherwise
 716 */
 717
 718static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
 719		       u64 *leaf_out)
 720{
 721	__be64 leaf_no;
 722	int error;
 723
 724	error = gfs2_dir_read_data(dip, (char *)&leaf_no,
 725				    index * sizeof(__be64),
 726				    sizeof(__be64), 0);
 727	if (error != sizeof(u64))
 728		return (error < 0) ? error : -EIO;
 729
 730	*leaf_out = be64_to_cpu(leaf_no);
 731
 732	return 0;
 733}
 734
 735static int get_first_leaf(struct gfs2_inode *dip, u32 index,
 736			  struct buffer_head **bh_out)
 737{
 738	u64 leaf_no;
 739	int error;
 740
 741	error = get_leaf_nr(dip, index, &leaf_no);
 742	if (!error)
 743		error = get_leaf(dip, leaf_no, bh_out);
 744
 745	return error;
 746}
 747
 748static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
 749					      const struct qstr *name,
 750					      gfs2_dscan_t scan,
 751					      struct buffer_head **pbh)
 752{
 753	struct buffer_head *bh;
 754	struct gfs2_dirent *dent;
 755	struct gfs2_inode *ip = GFS2_I(inode);
 756	int error;
 757
 758	if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
 759		struct gfs2_leaf *leaf;
 760		unsigned hsize = 1 << ip->i_di.di_depth;
 761		unsigned index;
 762		u64 ln;
 763		if (hsize * sizeof(u64) != ip->i_di.di_size) {
 764			gfs2_consist_inode(ip);
 765			return ERR_PTR(-EIO);
 766		}
 767
 768		index = name->hash >> (32 - ip->i_di.di_depth);
 769		error = get_first_leaf(ip, index, &bh);
 770		if (error)
 771			return ERR_PTR(error);
 772		do {
 773			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
 774						scan, name, NULL);
 775			if (dent)
 776				goto got_dent;
 777			leaf = (struct gfs2_leaf *)bh->b_data;
 778			ln = be64_to_cpu(leaf->lf_next);
 779			brelse(bh);
 780			if (!ln)
 781				break;
 782
 783			error = get_leaf(ip, ln, &bh);
 784		} while(!error);
 785
 786		return error ? ERR_PTR(error) : NULL;
 787	}
 788
 789
 790	error = gfs2_meta_inode_buffer(ip, &bh);
 791	if (error)
 792		return ERR_PTR(error);
 793	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
 794got_dent:
 795	if (unlikely(dent == NULL || IS_ERR(dent))) {
 796		brelse(bh);
 797		bh = NULL;
 798	}
 799	*pbh = bh;
 800	return dent;
 801}
 802
 803static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
 804{
 805	struct gfs2_inode *ip = GFS2_I(inode);
 806	u64 bn = gfs2_alloc_meta(ip);
 807	struct buffer_head *bh = gfs2_meta_new(ip->i_gl, bn);
 808	struct gfs2_leaf *leaf;
 809	struct gfs2_dirent *dent;
 810	struct qstr name = { .name = "", .len = 0, .hash = 0 };
 811	if (!bh)
 812		return NULL;
 813
 814	gfs2_trans_add_bh(ip->i_gl, bh, 1);
 815	gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
 816	leaf = (struct gfs2_leaf *)bh->b_data;
 817	leaf->lf_depth = cpu_to_be16(depth);
 818	leaf->lf_entries = 0;
 819	leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
 820	leaf->lf_next = 0;
 821	memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved));
 822	dent = (struct gfs2_dirent *)(leaf+1);
 823	gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
 824	*pbh = bh;
 825	return leaf;
 826}
 827
 828/**
 829 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
 830 * @dip: The GFS2 inode
 831 *
 832 * Returns: 0 on success, error code otherwise
 833 */
 834
 835static int dir_make_exhash(struct inode *inode)
 836{
 837	struct gfs2_inode *dip = GFS2_I(inode);
 838	struct gfs2_sbd *sdp = GFS2_SB(inode);
 839	struct gfs2_dirent *dent;
 840	struct qstr args;
 841	struct buffer_head *bh, *dibh;
 842	struct gfs2_leaf *leaf;
 843	int y;
 844	u32 x;
 845	__be64 *lp;
 846	u64 bn;
 847	int error;
 848
 849	error = gfs2_meta_inode_buffer(dip, &dibh);
 850	if (error)
 851		return error;
 852
 853	/*  Turn over a new leaf  */
 854
 855	leaf = new_leaf(inode, &bh, 0);
 856	if (!leaf)
 857		return -ENOSPC;
 858	bn = bh->b_blocknr;
 859
 860	gfs2_assert(sdp, dip->i_di.di_entries < (1 << 16));
 861	leaf->lf_entries = cpu_to_be16(dip->i_di.di_entries);
 862
 863	/*  Copy dirents  */
 864
 865	gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
 866			     sizeof(struct gfs2_dinode));
 867
 868	/*  Find last entry  */
 869
 870	x = 0;
 871	args.len = bh->b_size - sizeof(struct gfs2_dinode) +
 872		   sizeof(struct gfs2_leaf);
 873	args.name = bh->b_data;
 874	dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
 875				gfs2_dirent_last, &args, NULL);
 876	if (!dent) {
 877		brelse(bh);
 878		brelse(dibh);
 879		return -EIO;
 880	}
 881	if (IS_ERR(dent)) {
 882		brelse(bh);
 883		brelse(dibh);
 884		return PTR_ERR(dent);
 885	}
 886
 887	/*  Adjust the last dirent's record length
 888	   (Remember that dent still points to the last entry.)  */
 889
 890	dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
 891		sizeof(struct gfs2_dinode) -
 892		sizeof(struct gfs2_leaf));
 893
 894	brelse(bh);
 895
 896	/*  We're done with the new leaf block, now setup the new
 897	    hash table.  */
 898
 899	gfs2_trans_add_bh(dip->i_gl, dibh, 1);
 900	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 901
 902	lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
 903
 904	for (x = sdp->sd_hash_ptrs; x--; lp++)
 905		*lp = cpu_to_be64(bn);
 906
 907	dip->i_di.di_size = sdp->sd_sb.sb_bsize / 2;
 908	dip->i_di.di_blocks++;
 909	gfs2_set_inode_blocks(&dip->i_inode);
 910	dip->i_di.di_flags |= GFS2_DIF_EXHASH;
 911
 912	for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
 913	dip->i_di.di_depth = y;
 914
 915	gfs2_dinode_out(dip, dibh->b_data);
 916
 917	brelse(dibh);
 918
 919	return 0;
 920}
 921
 922/**
 923 * dir_split_leaf - Split a leaf block into two
 924 * @dip: The GFS2 inode
 925 * @index:
 926 * @leaf_no:
 927 *
 928 * Returns: 0 on success, error code on failure
 929 */
 930
 931static int dir_split_leaf(struct inode *inode, const struct qstr *name)
 932{
 933	struct gfs2_inode *dip = GFS2_I(inode);
 934	struct buffer_head *nbh, *obh, *dibh;
 935	struct gfs2_leaf *nleaf, *oleaf;
 936	struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
 937	u32 start, len, half_len, divider;
 938	u64 bn, leaf_no;
 939	__be64 *lp;
 940	u32 index;
 941	int x, moved = 0;
 942	int error;
 943
 944	index = name->hash >> (32 - dip->i_di.di_depth);
 945	error = get_leaf_nr(dip, index, &leaf_no);
 946	if (error)
 947		return error;
 948
 949	/*  Get the old leaf block  */
 950	error = get_leaf(dip, leaf_no, &obh);
 951	if (error)
 952		return error;
 953
 954	oleaf = (struct gfs2_leaf *)obh->b_data;
 955	if (dip->i_di.di_depth == be16_to_cpu(oleaf->lf_depth)) {
 956		brelse(obh);
 957		return 1; /* can't split */
 958	}
 959
 960	gfs2_trans_add_bh(dip->i_gl, obh, 1);
 961
 962	nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
 963	if (!nleaf) {
 964		brelse(obh);
 965		return -ENOSPC;
 966	}
 967	bn = nbh->b_blocknr;
 968
 969	/*  Compute the start and len of leaf pointers in the hash table.  */
 970	len = 1 << (dip->i_di.di_depth - be16_to_cpu(oleaf->lf_depth));
 971	half_len = len >> 1;
 972	if (!half_len) {
 973		printk(KERN_WARNING "di_depth %u lf_depth %u index %u\n", dip->i_di.di_depth, be16_to_cpu(oleaf->lf_depth), index);
 974		gfs2_consist_inode(dip);
 975		error = -EIO;
 976		goto fail_brelse;
 977	}
 978
 979	start = (index & ~(len - 1));
 980
 981	/* Change the pointers.
 982	   Don't bother distinguishing stuffed from non-stuffed.
 983	   This code is complicated enough already. */
 984	lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS | __GFP_NOFAIL);
 985	/*  Change the pointers  */
 986	for (x = 0; x < half_len; x++)
 987		lp[x] = cpu_to_be64(bn);
 988
 989	error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
 990				    half_len * sizeof(u64));
 991	if (error != half_len * sizeof(u64)) {
 992		if (error >= 0)
 993			error = -EIO;
 994		goto fail_lpfree;
 995	}
 996
 997	kfree(lp);
 998
 999	/*  Compute the divider  */
1000	divider = (start + half_len) << (32 - dip->i_di.di_depth);
1001
1002	/*  Copy the entries  */
1003	dirent_first(dip, obh, &dent);
1004
1005	do {
1006		next = dent;
1007		if (dirent_next(dip, obh, &next))
1008			next = NULL;
1009
1010		if (!gfs2_dirent_sentinel(dent) &&
1011		    be32_to_cpu(dent->de_hash) < divider) {
1012			struct qstr str;
1013			str.name = (char*)(dent+1);
1014			str.len = be16_to_cpu(dent->de_name_len);
1015			str.hash = be32_to_cpu(dent->de_hash);
1016			new = gfs2_dirent_alloc(inode, nbh, &str);
1017			if (IS_ERR(new)) {
1018				error = PTR_ERR(new);
1019				break;
1020			}
1021
1022			new->de_inum = dent->de_inum; /* No endian worries */
1023			new->de_type = dent->de_type; /* No endian worries */
1024			nleaf->lf_entries = cpu_to_be16(be16_to_cpu(nleaf->lf_entries)+1);
1025
1026			dirent_del(dip, obh, prev, dent);
1027
1028			if (!oleaf->lf_entries)
1029				gfs2_consist_inode(dip);
1030			oleaf->lf_entries = cpu_to_be16(be16_to_cpu(oleaf->lf_entries)-1);
1031
1032			if (!prev)
1033				prev = dent;
1034
1035			moved = 1;
1036		} else {
1037			prev = dent;
1038		}
1039		dent = next;
1040	} while (dent);
1041
1042	oleaf->lf_depth = nleaf->lf_depth;
1043
1044	error = gfs2_meta_inode_buffer(dip, &dibh);
1045	if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1046		dip->i_di.di_blocks++;
1047		gfs2_set_inode_blocks(&dip->i_inode);
1048		gfs2_dinode_out(dip, dibh->b_data);
1049		brelse(dibh);
1050	}
1051
1052	brelse(obh);
1053	brelse(nbh);
1054
1055	return error;
1056
1057fail_lpfree:
1058	kfree(lp);
1059
1060fail_brelse:
1061	brelse(obh);
1062	brelse(nbh);
1063	return error;
1064}
1065
1066/**
1067 * dir_double_exhash - Double size of ExHash table
1068 * @dip: The GFS2 dinode
1069 *
1070 * Returns: 0 on success, error code on failure
1071 */
1072
1073static int dir_double_exhash(struct gfs2_inode *dip)
1074{
1075	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1076	struct buffer_head *dibh;
1077	u32 hsize;
1078	u64 *buf;
1079	u64 *from, *to;
1080	u64 block;
1081	int x;
1082	int error = 0;
1083
1084	hsize = 1 << dip->i_di.di_depth;
1085	if (hsize * sizeof(u64) != dip->i_di.di_size) {
1086		gfs2_consist_inode(dip);
1087		return -EIO;
1088	}
1089
1090	/*  Allocate both the "from" and "to" buffers in one big chunk  */
1091
1092	buf = kcalloc(3, sdp->sd_hash_bsize, GFP_KERNEL | __GFP_NOFAIL);
1093
1094	for (block = dip->i_di.di_size >> sdp->sd_hash_bsize_shift; block--;) {
1095		error = gfs2_dir_read_data(dip, (char *)buf,
1096					    block * sdp->sd_hash_bsize,
1097					    sdp->sd_hash_bsize, 1);
1098		if (error != sdp->sd_hash_bsize) {
1099			if (error >= 0)
1100				error = -EIO;
1101			goto fail;
1102		}
1103
1104		from = buf;
1105		to = (u64 *)((char *)buf + sdp->sd_hash_bsize);
1106
1107		for (x = sdp->sd_hash_ptrs; x--; from++) {
1108			*to++ = *from;	/*  No endianess worries  */
1109			*to++ = *from;
1110		}
1111
1112		error = gfs2_dir_write_data(dip,
1113					     (char *)buf + sdp->sd_hash_bsize,
1114					     block * sdp->sd_sb.sb_bsize,
1115					     sdp->sd_sb.sb_bsize);
1116		if (error != sdp->sd_sb.sb_bsize) {
1117			if (error >= 0)
1118				error = -EIO;
1119			goto fail;
1120		}
1121	}
1122
1123	kfree(buf);
1124
1125	error = gfs2_meta_inode_buffer(dip, &dibh);
1126	if (!gfs2_assert_withdraw(sdp, !error)) {
1127		dip->i_di.di_depth++;
1128		gfs2_dinode_out(dip, dibh->b_data);
1129		brelse(dibh);
1130	}
1131
1132	return error;
1133
1134fail:
1135	kfree(buf);
1136	return error;
1137}
1138
1139/**
1140 * compare_dents - compare directory entries by hash value
1141 * @a: first dent
1142 * @b: second dent
1143 *
1144 * When comparing the hash entries of @a to @b:
1145 *   gt: returns 1
1146 *   lt: returns -1
1147 *   eq: returns 0
1148 */
1149
1150static int compare_dents(const void *a, const void *b)
1151{
1152	const struct gfs2_dirent *dent_a, *dent_b;
1153	u32 hash_a, hash_b;
1154	int ret = 0;
1155
1156	dent_a = *(const struct gfs2_dirent **)a;
1157	hash_a = be32_to_cpu(dent_a->de_hash);
1158
1159	dent_b = *(const struct gfs2_dirent **)b;
1160	hash_b = be32_to_cpu(dent_b->de_hash);
1161
1162	if (hash_a > hash_b)
1163		ret = 1;
1164	else if (hash_a < hash_b)
1165		ret = -1;
1166	else {
1167		unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1168		unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1169
1170		if (len_a > len_b)
1171			ret = 1;
1172		else if (len_a < len_b)
1173			ret = -1;
1174		else
1175			ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1176	}
1177
1178	return ret;
1179}
1180
1181/**
1182 * do_filldir_main - read out directory entries
1183 * @dip: The GFS2 inode
1184 * @offset: The offset in the file to read from
1185 * @opaque: opaque data to pass to filldir
1186 * @filldir: The function to pass entries to
1187 * @darr: an array of struct gfs2_dirent pointers to read
1188 * @entries: the number of entries in darr
1189 * @copied: pointer to int that's non-zero if a entry has been copied out
1190 *
1191 * Jump through some hoops to make sure that if there are hash collsions,
1192 * they are read out at the beginning of a buffer.  We want to minimize
1193 * the possibility that they will fall into different readdir buffers or
1194 * that someone will want to seek to that location.
1195 *
1196 * Returns: errno, >0 on exception from filldir
1197 */
1198
1199static int do_filldir_main(struct gfs2_inode *dip, u64 *offset,
1200			   void *opaque, filldir_t filldir,
1201			   const struct gfs2_dirent **darr, u32 entries,
1202			   int *copied)
1203{
1204	const struct gfs2_dirent *dent, *dent_next;
1205	u64 off, off_next;
1206	unsigned int x, y;
1207	int run = 0;
1208	int error = 0;
1209
1210	sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
1211
1212	dent_next = darr[0];
1213	off_next = be32_to_cpu(dent_next->de_hash);
1214	off_next = gfs2_disk_hash2offset(off_next);
1215
1216	for (x = 0, y = 1; x < entries; x++, y++) {
1217		dent = dent_next;
1218		off = off_next;
1219
1220		if (y < entries) {
1221			dent_next = darr[y];
1222			off_next = be32_to_cpu(dent_next->de_hash);
1223			off_next = gfs2_disk_hash2offset(off_next);
1224
1225			if (off < *offset)
1226				continue;
1227			*offset = off;
1228
1229			if (off_next == off) {
1230				if (*copied && !run)
1231					return 1;
1232				run = 1;
1233			} else
1234				run = 0;
1235		} else {
1236			if (off < *offset)
1237				continue;
1238			*offset = off;
1239		}
1240
1241		error = filldir(opaque, (const char *)(dent + 1),
1242				be16_to_cpu(dent->de_name_len),
1243				off, be64_to_cpu(dent->de_inum.no_addr),
1244				be16_to_cpu(dent->de_type));
1245		if (error)
1246			return 1;
1247
1248		*copied = 1;
1249	}
1250
1251	/* Increment the *offset by one, so the next time we come into the
1252	   do_filldir fxn, we get the next entry instead of the last one in the
1253	   current leaf */
1254
1255	(*offset)++;
1256
1257	return 0;
1258}
1259
1260static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque,
1261			      filldir_t filldir, int *copied, unsigned *depth,
1262			      u64 leaf_no)
1263{
1264	struct gfs2_inode *ip = GFS2_I(inode);
1265	struct buffer_head *bh;
1266	struct gfs2_leaf *lf;
1267	unsigned entries = 0;
1268	unsigned leaves = 0;
1269	const struct gfs2_dirent **darr, *dent;
1270	struct dirent_gather g;
1271	struct buffer_head **larr;
1272	int leaf = 0;
1273	int error, i;
1274	u64 lfn = leaf_no;
1275
1276	do {
1277		error = get_leaf(ip, lfn, &bh);
1278		if (error)
1279			goto out;
1280		lf = (struct gfs2_leaf *)bh->b_data;
1281		if (leaves == 0)
1282			*depth = be16_to_cpu(lf->lf_depth);
1283		entries += be16_to_cpu(lf->lf_entries);
1284		leaves++;
1285		lfn = be64_to_cpu(lf->lf_next);
1286		brelse(bh);
1287	} while(lfn);
1288
1289	if (!entries)
1290		return 0;
1291
1292	error = -ENOMEM;
1293	larr = vmalloc((leaves + entries) * sizeof(void *));
1294	if (!larr)
1295		goto out;
1296	darr = (const struct gfs2_dirent **)(larr + leaves);
1297	g.pdent = darr;
1298	g.offset = 0;
1299	lfn = leaf_no;
1300
1301	do {
1302		error = get_leaf(ip, lfn, &bh);
1303		if (error)
1304			goto out_kfree;
1305		lf = (struct gfs2_leaf *)bh->b_data;
1306		lfn = be64_to_cpu(lf->lf_next);
1307		if (lf->lf_entries) {
1308			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1309						gfs2_dirent_gather, NULL, &g);
1310			error = PTR_ERR(dent);
1311			if (IS_ERR(dent)) {
1312				goto out_kfree;
1313			}
1314			error = 0;
1315			larr[leaf++] = bh;
1316		} else {
1317			brelse(bh);
1318		}
1319	} while(lfn);
1320
1321	error = do_filldir_main(ip, offset, opaque, filldir, darr,
1322				entries, copied);
1323out_kfree:
1324	for(i = 0; i < leaf; i++)
1325		brelse(larr[i]);
1326	vfree(larr);
1327out:
1328	return error;
1329}
1330
1331/**
1332 * dir_e_read - Reads the entries from a directory into a filldir buffer
1333 * @dip: dinode pointer
1334 * @offset: the hash of the last entry read shifted to the right once
1335 * @opaque: buffer for the filldir function to fill
1336 * @filldir: points to the filldir function to use
1337 *
1338 * Returns: errno
1339 */
1340
1341static int dir_e_read(struct inode *inode, u64 *offset, void *opaque,
1342		      filldir_t filldir)
1343{
1344	struct gfs2_inode *dip = GFS2_I(inode);
1345	struct gfs2_sbd *sdp = GFS2_SB(inode);
1346	u32 hsize, len = 0;
1347	u32 ht_offset, lp_offset, ht_offset_cur = -1;
1348	u32 hash, index;
1349	__be64 *lp;
1350	int copied = 0;
1351	int error = 0;
1352	unsigned depth = 0;
1353
1354	hsize = 1 << dip->i_di.di_depth;
1355	if (hsize * sizeof(u64) != dip->i_di.di_size) {
1356		gfs2_consist_inode(dip);
1357		return -EIO;
1358	}
1359
1360	hash = gfs2_dir_offset2hash(*offset);
1361	index = hash >> (32 - dip->i_di.di_depth);
1362
1363	lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
1364	if (!lp)
1365		return -ENOMEM;
1366
1367	while (index < hsize) {
1368		lp_offset = index & (sdp->sd_hash_ptrs - 1);
1369		ht_offset = index - lp_offset;
1370
1371		if (ht_offset_cur != ht_offset) {
1372			error = gfs2_dir_read_data(dip, (char *)lp,
1373						ht_offset * sizeof(__be64),
1374						sdp->sd_hash_bsize, 1);
1375			if (error != sdp->sd_hash_bsize) {
1376				if (error >= 0)
1377					error = -EIO;
1378				goto out;
1379			}
1380			ht_offset_cur = ht_offset;
1381		}
1382
1383		error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
1384					   &copied, &depth,
1385					   be64_to_cpu(lp[lp_offset]));
1386		if (error)
1387			break;
1388
1389		len = 1 << (dip->i_di.di_depth - depth);
1390		index = (index & ~(len - 1)) + len;
1391	}
1392
1393out:
1394	kfree(lp);
1395	if (error > 0)
1396		error = 0;
1397	return error;
1398}
1399
1400int gfs2_dir_read(struct inode *inode, u64 *offset, void *opaque,
1401		  filldir_t filldir)
1402{
1403	struct gfs2_inode *dip = GFS2_I(inode);
1404	struct dirent_gather g;
1405	const struct gfs2_dirent **darr, *dent;
1406	struct buffer_head *dibh;
1407	int copied = 0;
1408	int error;
1409
1410	if (!dip->i_di.di_entries)
1411		return 0;
1412
1413	if (dip->i_di.di_flags & GFS2_DIF_EXHASH)
1414		return dir_e_read(inode, offset, opaque, filldir);
1415
1416	if (!gfs2_is_stuffed(dip)) {
1417		gfs2_consist_inode(dip);
1418		return -EIO;
1419	}
1420
1421	error = gfs2_meta_inode_buffer(dip, &dibh);
1422	if (error)
1423		return error;
1424
1425	error = -ENOMEM;
1426	darr = kmalloc(dip->i_di.di_entries * sizeof(struct gfs2_dirent *),
1427		       GFP_KERNEL);
1428	if (darr) {
1429		g.pdent = darr;
1430		g.offset = 0;
1431		dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1432					gfs2_dirent_gather, NULL, &g);
1433		if (IS_ERR(dent)) {
1434			error = PTR_ERR(dent);
1435			goto out;
1436		}
1437		error = do_filldir_main(dip, offset, opaque, filldir, darr,
1438					dip->i_di.di_entries, &copied);
1439out:
1440		kfree(darr);
1441	}
1442
1443	if (error > 0)
1444		error = 0;
1445
1446	brelse(dibh);
1447
1448	return error;
1449}
1450
1451/**
1452 * gfs2_dir_search - Search a directory
1453 * @dip: The GFS2 inode
1454 * @filename:
1455 * @inode:
1456 *
1457 * This routine searches a directory for a file or another directory.
1458 * Assumes a glock is held on dip.
1459 *
1460 * Returns: errno
1461 */
1462
1463int gfs2_dir_search(struct inode *dir, const struct qstr *name,
1464		    struct gfs2_inum_host *inum, unsigned int *type)
1465{
1466	struct buffer_head *bh;
1467	struct gfs2_dirent *dent;
1468
1469	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1470	if (dent) {
1471		if (IS_ERR(dent))
1472			return PTR_ERR(dent);
1473		if (inum)
1474			gfs2_inum_in(inum, (char *)&dent->de_inum);
1475		if (type)
1476			*type = be16_to_cpu(dent->de_type);
1477		brelse(bh);
1478		return 0;
1479	}
1480	return -ENOENT;
1481}
1482
1483static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1484{
1485	struct buffer_head *bh, *obh;
1486	struct gfs2_inode *ip = GFS2_I(inode);
1487	struct gfs2_leaf *leaf, *oleaf;
1488	int error;
1489	u32 index;
1490	u64 bn;
1491
1492	index = name->hash >> (32 - ip->i_di.di_depth);
1493	error = get_first_leaf(ip, index, &obh);
1494	if (error)
1495		return error;
1496	do {
1497		oleaf = (struct gfs2_leaf *)obh->b_data;
1498		bn = be64_to_cpu(oleaf->lf_next);
1499		if (!bn)
1500			break;
1501		brelse(obh);
1502		error = get_leaf(ip, bn, &obh);
1503		if (error)
1504			return error;
1505	} while(1);
1506
1507	gfs2_trans_add_bh(ip->i_gl, obh, 1);
1508
1509	leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1510	if (!leaf) {
1511		brelse(obh);
1512		return -ENOSPC;
1513	}
1514	oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1515	brelse(bh);
1516	brelse(obh);
1517
1518	error = gfs2_meta_inode_buffer(ip, &bh);
1519	if (error)
1520		return error;
1521	gfs2_trans_add_bh(ip->i_gl, bh, 1);
1522	ip->i_di.di_blocks++;
1523	gfs2_set_inode_blocks(&ip->i_inode);
1524	gfs2_dinode_out(ip, bh->b_data);
1525	brelse(bh);
1526	return 0;
1527}
1528
1529/**
1530 * gfs2_dir_add - Add new filename into directory
1531 * @dip: The GFS2 inode
1532 * @filename: The new name
1533 * @inode: The inode number of the entry
1534 * @type: The type of the entry
1535 *
1536 * Returns: 0 on success, error code on failure
1537 */
1538
1539int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1540		 const struct gfs2_inum_host *inum, unsigned type)
1541{
1542	struct gfs2_inode *ip = GFS2_I(inode);
1543	struct buffer_head *bh;
1544	struct gfs2_dirent *dent;
1545	struct gfs2_leaf *leaf;
1546	int error;
1547
1548	while(1) {
1549		dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space,
1550					  &bh);
1551		if (dent) {
1552			if (IS_ERR(dent))
1553				return PTR_ERR(dent);
1554			dent = gfs2_init_dirent(inode, dent, name, bh);
1555			gfs2_inum_out(inum, (char *)&dent->de_inum);
1556			dent->de_type = cpu_to_be16(type);
1557			if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
1558				leaf = (struct gfs2_leaf *)bh->b_data;
1559				leaf->lf_entries = cpu_to_be16(be16_to_cpu(leaf->lf_entries) + 1);
1560			}
1561			brelse(bh);
1562			error = gfs2_meta_inode_buffer(ip, &bh);
1563			if (error)
1564				break;
1565			gfs2_trans_add_bh(ip->i_gl, bh, 1);
1566			ip->i_di.di_entries++;
1567			ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME_SEC;
1568			gfs2_dinode_out(ip, bh->b_data);
1569			brelse(bh);
1570			error = 0;
1571			break;
1572		}
1573		if (!(ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
1574			error = dir_make_exhash(inode);
1575			if (error)
1576				break;
1577			continue;
1578		}
1579		error = dir_split_leaf(inode, name);
1580		if (error == 0)
1581			continue;
1582		if (error < 0)
1583			break;
1584		if (ip->i_di.di_depth < GFS2_DIR_MAX_DEPTH) {
1585			error = dir_double_exhash(ip);
1586			if (error)
1587				break;
1588			error = dir_split_leaf(inode, name);
1589			if (error < 0)
1590				break;
1591			if (error == 0)
1592				continue;
1593		}
1594		error = dir_new_leaf(inode, name);
1595		if (!error)
1596			continue;
1597		error = -ENOSPC;
1598		break;
1599	}
1600	return error;
1601}
1602
1603
1604/**
1605 * gfs2_dir_del - Delete a directory entry
1606 * @dip: The GFS2 inode
1607 * @filename: The filename
1608 *
1609 * Returns: 0 on success, error code on failure
1610 */
1611
1612int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name)
1613{
1614	struct gfs2_dirent *dent, *prev = NULL;
1615	struct buffer_head *bh;
1616	int error;
1617
1618	/* Returns _either_ the entry (if its first in block) or the
1619	   previous entry otherwise */
1620	dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1621	if (!dent) {
1622		gfs2_consist_inode(dip);
1623		return -EIO;
1624	}
1625	if (IS_ERR(dent)) {
1626		gfs2_consist_inode(dip);
1627		return PTR_ERR(dent);
1628	}
1629	/* If not first in block, adjust pointers accordingly */
1630	if (gfs2_dirent_find(dent, name, NULL) == 0) {
1631		prev = dent;
1632		dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1633	}
1634
1635	dirent_del(dip, bh, prev, dent);
1636	if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
1637		struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1638		u16 entries = be16_to_cpu(leaf->lf_entries);
1639		if (!entries)
1640			gfs2_consist_inode(dip);
1641		leaf->lf_entries = cpu_to_be16(--entries);
1642	}
1643	brelse(bh);
1644
1645	error = gfs2_meta_inode_buffer(dip, &bh);
1646	if (error)
1647		return error;
1648
1649	if (!dip->i_di.di_entries)
1650		gfs2_consist_inode(dip);
1651	gfs2_trans_add_bh(dip->i_gl, bh, 1);
1652	dip->i_di.di_entries--;
1653	dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME_SEC;
1654	gfs2_dinode_out(dip, bh->b_data);
1655	brelse(bh);
1656	mark_inode_dirty(&dip->i_inode);
1657
1658	return error;
1659}
1660
1661/**
1662 * gfs2_dir_mvino - Change inode number of directory entry
1663 * @dip: The GFS2 inode
1664 * @filename:
1665 * @new_inode:
1666 *
1667 * This routine changes the inode number of a directory entry.  It's used
1668 * by rename to change ".." when a directory is moved.
1669 * Assumes a glock is held on dvp.
1670 *
1671 * Returns: errno
1672 */
1673
1674int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1675		   struct gfs2_inum_host *inum, unsigned int new_type)
1676{
1677	struct buffer_head *bh;
1678	struct gfs2_dirent *dent;
1679	int error;
1680
1681	dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1682	if (!dent) {
1683		gfs2_consist_inode(dip);
1684		return -EIO;
1685	}
1686	if (IS_ERR(dent))
1687		return PTR_ERR(dent);
1688
1689	gfs2_trans_add_bh(dip->i_gl, bh, 1);
1690	gfs2_inum_out(inum, (char *)&dent->de_inum);
1691	dent->de_type = cpu_to_be16(new_type);
1692
1693	if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
1694		brelse(bh);
1695		error = gfs2_meta_inode_buffer(dip, &bh);
1696		if (error)
1697			return error;
1698		gfs2_trans_add_bh(dip->i_gl, bh, 1);
1699	}
1700
1701	dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME_SEC;
1702	gfs2_dinode_out(dip, bh->b_data);
1703	brelse(bh);
1704	return 0;
1705}
1706
1707/**
1708 * foreach_leaf - call a function for each leaf in a directory
1709 * @dip: the directory
1710 * @lc: the function to call for each each
1711 * @data: private data to pass to it
1712 *
1713 * Returns: errno
1714 */
1715
1716static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data)
1717{
1718	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1719	struct buffer_head *bh;
1720	struct gfs2_leaf *leaf;
1721	u32 hsize, len;
1722	u32 ht_offset, lp_offset, ht_offset_cur = -1;
1723	u32 index = 0;
1724	__be64 *lp;
1725	u64 leaf_no;
1726	int error = 0;
1727
1728	hsize = 1 << dip->i_di.di_depth;
1729	if (hsize * sizeof(u64) != dip->i_di.di_size) {
1730		gfs2_consist_inode(dip);
1731		return -EIO;
1732	}
1733
1734	lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
1735	if (!lp)
1736		return -ENOMEM;
1737
1738	while (index < hsize) {
1739		lp_offset = index & (sdp->sd_hash_ptrs - 1);
1740		ht_offset = index - lp_offset;
1741
1742		if (ht_offset_cur != ht_offset) {
1743			error = gfs2_dir_read_data(dip, (char *)lp,
1744						ht_offset * sizeof(__be64),
1745						sdp->sd_hash_bsize, 1);
1746			if (error != sdp->sd_hash_bsize) {
1747				if (error >= 0)
1748					error = -EIO;
1749				goto out;
1750			}
1751			ht_offset_cur = ht_offset;
1752		}
1753
1754		leaf_no = be64_to_cpu(lp[lp_offset]);
1755		if (leaf_no) {
1756			error = get_leaf(dip, leaf_no, &bh);
1757			if (error)
1758				goto out;
1759			leaf = (struct gfs2_leaf *)bh->b_data;
1760			len = 1 << (dip->i_di.di_depth - be16_to_cpu(leaf->lf_depth));
1761			brelse(bh);
1762
1763			error = lc(dip, index, len, leaf_no, data);
1764			if (error)
1765				goto out;
1766
1767			index = (index & ~(len - 1)) + len;
1768		} else
1769			index++;
1770	}
1771
1772	if (index != hsize) {
1773		gfs2_consist_inode(dip);
1774		error = -EIO;
1775	}
1776
1777out:
1778	kfree(lp);
1779
1780	return error;
1781}
1782
1783/**
1784 * leaf_dealloc - Deallocate a directory leaf
1785 * @dip: the directory
1786 * @index: the hash table offset in the directory
1787 * @len: the number of pointers to this leaf
1788 * @leaf_no: the leaf number
1789 * @data: not used
1790 *
1791 * Returns: errno
1792 */
1793
1794static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1795			u64 leaf_no, void *data)
1796{
1797	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1798	struct gfs2_leaf *tmp_leaf;
1799	struct gfs2_rgrp_list rlist;
1800	struct buffer_head *bh, *dibh;
1801	u64 blk, nblk;
1802	unsigned int rg_blocks = 0, l_blocks = 0;
1803	char *ht;
1804	unsigned int x, size = len * sizeof(u64);
1805	int error;
1806
1807	memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1808
1809	ht = kzalloc(size, GFP_KERNEL);
1810	if (!ht)
1811		return -ENOMEM;
1812
1813	gfs2_alloc_get(dip);
1814
1815	error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1816	if (error)
1817		goto out;
1818
1819	error = gfs2_rindex_hold(sdp, &dip->i_alloc.al_ri_gh);
1820	if (error)
1821		goto out_qs;
1822
1823	/*  Count the number of leaves  */
1824
1825	for (blk = leaf_no; blk; blk = nblk) {
1826		error = get_leaf(dip, blk, &bh);
1827		if (error)
1828			goto out_rlist;
1829		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1830		nblk = be64_to_cpu(tmp_leaf->lf_next);
1831		brelse(bh);
1832
1833		gfs2_rlist_add(sdp, &rlist, blk);
1834		l_blocks++;
1835	}
1836
1837	gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, 0);
1838
1839	for (x = 0; x < rlist.rl_rgrps; x++) {
1840		struct gfs2_rgrpd *rgd;
1841		rgd = rlist.rl_ghs[x].gh_gl->gl_object;
1842		rg_blocks += rgd->rd_ri.ri_length;
1843	}
1844
1845	error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
1846	if (error)
1847		goto out_rlist;
1848
1849	error = gfs2_trans_begin(sdp,
1850			rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
1851			RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
1852	if (error)
1853		goto out_rg_gunlock;
1854
1855	for (blk = leaf_no; blk; blk = nblk) {
1856		error = get_leaf(dip, blk, &bh);
1857		if (error)
1858			goto out_end_trans;
1859		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1860		nblk = be64_to_cpu(tmp_leaf->lf_next);
1861		brelse(bh);
1862
1863		gfs2_free_meta(dip, blk, 1);
1864
1865		if (!dip->i_di.di_blocks)
1866			gfs2_consist_inode(dip);
1867		dip->i_di.di_blocks--;
1868		gfs2_set_inode_blocks(&dip->i_inode);
1869	}
1870
1871	error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
1872	if (error != size) {
1873		if (error >= 0)
1874			error = -EIO;
1875		goto out_end_trans;
1876	}
1877
1878	error = gfs2_meta_inode_buffer(dip, &dibh);
1879	if (error)
1880		goto out_end_trans;
1881
1882	gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1883	gfs2_dinode_out(dip, dibh->b_data);
1884	brelse(dibh);
1885
1886out_end_trans:
1887	gfs2_trans_end(sdp);
1888out_rg_gunlock:
1889	gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
1890out_rlist:
1891	gfs2_rlist_free(&rlist);
1892	gfs2_glock_dq_uninit(&dip->i_alloc.al_ri_gh);
1893out_qs:
1894	gfs2_quota_unhold(dip);
1895out:
1896	gfs2_alloc_put(dip);
1897	kfree(ht);
1898	return error;
1899}
1900
1901/**
1902 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1903 * @dip: the directory
1904 *
1905 * Dealloc all on-disk directory leaves to FREEMETA state
1906 * Change on-disk inode type to "regular file"
1907 *
1908 * Returns: errno
1909 */
1910
1911int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
1912{
1913	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1914	struct buffer_head *bh;
1915	int error;
1916
1917	/* Dealloc on-disk leaves to FREEMETA state */
1918	error = foreach_leaf(dip, leaf_dealloc, NULL);
1919	if (error)
1920		return error;
1921
1922	/* Make this a regular file in case we crash.
1923	   (We don't want to free these blocks a second time.)  */
1924
1925	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1926	if (error)
1927		return error;
1928
1929	error = gfs2_meta_inode_buffer(dip, &bh);
1930	if (!error) {
1931		gfs2_trans_add_bh(dip->i_gl, bh, 1);
1932		((struct gfs2_dinode *)bh->b_data)->di_mode =
1933						cpu_to_be32(S_IFREG);
1934		brelse(bh);
1935	}
1936
1937	gfs2_trans_end(sdp);
1938
1939	return error;
1940}
1941
1942/**
1943 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
1944 * @ip: the file being written to
1945 * @filname: the filename that's going to be added
1946 *
1947 * Returns: 1 if alloc required, 0 if not, -ve on error
1948 */
1949
1950int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name)
1951{
1952	struct gfs2_dirent *dent;
1953	struct buffer_head *bh;
1954
1955	dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
1956	if (!dent) {
1957		return 1;
1958	}
1959	if (IS_ERR(dent))
1960		return PTR_ERR(dent);
1961	brelse(bh);
1962	return 0;
1963}
1964