PageRenderTime 129ms CodeModel.GetById 2ms app.highlight 113ms RepoModel.GetById 1ms app.codeStats 1ms

/fs/gfs2/dir.c

https://bitbucket.org/bradfa/linux
C | 2019 lines | 1438 code | 289 blank | 292 comment | 255 complexity | bdd5fe6236b5869a27bd00b74b992392 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_diskflags & 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
  64#include "gfs2.h"
  65#include "incore.h"
  66#include "dir.h"
  67#include "glock.h"
  68#include "inode.h"
  69#include "meta_io.h"
  70#include "quota.h"
  71#include "rgrp.h"
  72#include "trans.h"
  73#include "bmap.h"
  74#include "util.h"
  75
  76#define IS_LEAF     1 /* Hashed (leaf) directory */
  77#define IS_DINODE   2 /* Linear (stuffed dinode block) directory */
  78
  79#define MAX_RA_BLOCKS 32 /* max read-ahead blocks */
  80
  81#define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
  82#define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
  83
  84struct qstr gfs2_qdot __read_mostly;
  85struct qstr gfs2_qdotdot __read_mostly;
  86
  87typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
  88			    const struct qstr *name, void *opaque);
  89
  90int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
  91			    struct buffer_head **bhp)
  92{
  93	struct buffer_head *bh;
  94
  95	bh = gfs2_meta_new(ip->i_gl, block);
  96	gfs2_trans_add_meta(ip->i_gl, bh);
  97	gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
  98	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
  99	*bhp = bh;
 100	return 0;
 101}
 102
 103static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
 104					struct buffer_head **bhp)
 105{
 106	struct buffer_head *bh;
 107	int error;
 108
 109	error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh);
 110	if (error)
 111		return error;
 112	if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
 113		brelse(bh);
 114		return -EIO;
 115	}
 116	*bhp = bh;
 117	return 0;
 118}
 119
 120static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
 121				  unsigned int offset, unsigned int size)
 122{
 123	struct buffer_head *dibh;
 124	int error;
 125
 126	error = gfs2_meta_inode_buffer(ip, &dibh);
 127	if (error)
 128		return error;
 129
 130	gfs2_trans_add_meta(ip->i_gl, dibh);
 131	memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
 132	if (ip->i_inode.i_size < offset + size)
 133		i_size_write(&ip->i_inode, offset + size);
 134	ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 135	gfs2_dinode_out(ip, dibh->b_data);
 136
 137	brelse(dibh);
 138
 139	return size;
 140}
 141
 142
 143
 144/**
 145 * gfs2_dir_write_data - Write directory information to the inode
 146 * @ip: The GFS2 inode
 147 * @buf: The buffer containing information to be written
 148 * @offset: The file offset to start writing at
 149 * @size: The amount of data to write
 150 *
 151 * Returns: The number of bytes correctly written or error code
 152 */
 153static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
 154			       u64 offset, unsigned int size)
 155{
 156	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 157	struct buffer_head *dibh;
 158	u64 lblock, dblock;
 159	u32 extlen = 0;
 160	unsigned int o;
 161	int copied = 0;
 162	int error = 0;
 163	int new = 0;
 164
 165	if (!size)
 166		return 0;
 167
 168	if (gfs2_is_stuffed(ip) &&
 169	    offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
 170		return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
 171					      size);
 172
 173	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
 174		return -EINVAL;
 175
 176	if (gfs2_is_stuffed(ip)) {
 177		error = gfs2_unstuff_dinode(ip, NULL);
 178		if (error)
 179			return error;
 180	}
 181
 182	lblock = offset;
 183	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
 184
 185	while (copied < size) {
 186		unsigned int amount;
 187		struct buffer_head *bh;
 188
 189		amount = size - copied;
 190		if (amount > sdp->sd_sb.sb_bsize - o)
 191			amount = sdp->sd_sb.sb_bsize - o;
 192
 193		if (!extlen) {
 194			new = 1;
 195			error = gfs2_extent_map(&ip->i_inode, lblock, &new,
 196						&dblock, &extlen);
 197			if (error)
 198				goto fail;
 199			error = -EIO;
 200			if (gfs2_assert_withdraw(sdp, dblock))
 201				goto fail;
 202		}
 203
 204		if (amount == sdp->sd_jbsize || new)
 205			error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
 206		else
 207			error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
 208
 209		if (error)
 210			goto fail;
 211
 212		gfs2_trans_add_meta(ip->i_gl, bh);
 213		memcpy(bh->b_data + o, buf, amount);
 214		brelse(bh);
 215
 216		buf += amount;
 217		copied += amount;
 218		lblock++;
 219		dblock++;
 220		extlen--;
 221
 222		o = sizeof(struct gfs2_meta_header);
 223	}
 224
 225out:
 226	error = gfs2_meta_inode_buffer(ip, &dibh);
 227	if (error)
 228		return error;
 229
 230	if (ip->i_inode.i_size < offset + copied)
 231		i_size_write(&ip->i_inode, offset + copied);
 232	ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 233
 234	gfs2_trans_add_meta(ip->i_gl, dibh);
 235	gfs2_dinode_out(ip, dibh->b_data);
 236	brelse(dibh);
 237
 238	return copied;
 239fail:
 240	if (copied)
 241		goto out;
 242	return error;
 243}
 244
 245static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf,
 246				 unsigned int size)
 247{
 248	struct buffer_head *dibh;
 249	int error;
 250
 251	error = gfs2_meta_inode_buffer(ip, &dibh);
 252	if (!error) {
 253		memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), 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 * @size: Amount of data to transfer
 266 *
 267 * Returns: The amount of data actually copied or the error
 268 */
 269static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf,
 270			      unsigned int size)
 271{
 272	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 273	u64 lblock, dblock;
 274	u32 extlen = 0;
 275	unsigned int o;
 276	int copied = 0;
 277	int error = 0;
 278
 279	if (gfs2_is_stuffed(ip))
 280		return gfs2_dir_read_stuffed(ip, buf, size);
 281
 282	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
 283		return -EINVAL;
 284
 285	lblock = 0;
 286	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
 287
 288	while (copied < size) {
 289		unsigned int amount;
 290		struct buffer_head *bh;
 291		int new;
 292
 293		amount = size - copied;
 294		if (amount > sdp->sd_sb.sb_bsize - o)
 295			amount = sdp->sd_sb.sb_bsize - o;
 296
 297		if (!extlen) {
 298			new = 0;
 299			error = gfs2_extent_map(&ip->i_inode, lblock, &new,
 300						&dblock, &extlen);
 301			if (error || !dblock)
 302				goto fail;
 303			BUG_ON(extlen < 1);
 304			bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
 305		} else {
 306			error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh);
 307			if (error)
 308				goto fail;
 309		}
 310		error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
 311		if (error) {
 312			brelse(bh);
 313			goto fail;
 314		}
 315		dblock++;
 316		extlen--;
 317		memcpy(buf, bh->b_data + o, amount);
 318		brelse(bh);
 319		buf += (amount/sizeof(__be64));
 320		copied += amount;
 321		lblock++;
 322		o = sizeof(struct gfs2_meta_header);
 323	}
 324
 325	return copied;
 326fail:
 327	return (copied) ? copied : error;
 328}
 329
 330/**
 331 * gfs2_dir_get_hash_table - Get pointer to the dir hash table
 332 * @ip: The inode in question
 333 *
 334 * Returns: The hash table or an error
 335 */
 336
 337static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip)
 338{
 339	struct inode *inode = &ip->i_inode;
 340	int ret;
 341	u32 hsize;
 342	__be64 *hc;
 343
 344	BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH));
 345
 346	hc = ip->i_hash_cache;
 347	if (hc)
 348		return hc;
 349
 350	hsize = 1 << ip->i_depth;
 351	hsize *= sizeof(__be64);
 352	if (hsize != i_size_read(&ip->i_inode)) {
 353		gfs2_consist_inode(ip);
 354		return ERR_PTR(-EIO);
 355	}
 356
 357	hc = kmalloc(hsize, GFP_NOFS);
 358	ret = -ENOMEM;
 359	if (hc == NULL)
 360		return ERR_PTR(-ENOMEM);
 361
 362	ret = gfs2_dir_read_data(ip, hc, hsize);
 363	if (ret < 0) {
 364		kfree(hc);
 365		return ERR_PTR(ret);
 366	}
 367
 368	spin_lock(&inode->i_lock);
 369	if (ip->i_hash_cache)
 370		kfree(hc);
 371	else
 372		ip->i_hash_cache = hc;
 373	spin_unlock(&inode->i_lock);
 374
 375	return ip->i_hash_cache;
 376}
 377
 378/**
 379 * gfs2_dir_hash_inval - Invalidate dir hash
 380 * @ip: The directory inode
 381 *
 382 * Must be called with an exclusive glock, or during glock invalidation.
 383 */
 384void gfs2_dir_hash_inval(struct gfs2_inode *ip)
 385{
 386	__be64 *hc = ip->i_hash_cache;
 387	ip->i_hash_cache = NULL;
 388	kfree(hc);
 389}
 390
 391static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
 392{
 393	return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
 394}
 395
 396static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
 397				     const struct qstr *name, int ret)
 398{
 399	if (!gfs2_dirent_sentinel(dent) &&
 400	    be32_to_cpu(dent->de_hash) == name->hash &&
 401	    be16_to_cpu(dent->de_name_len) == name->len &&
 402	    memcmp(dent+1, name->name, name->len) == 0)
 403		return ret;
 404	return 0;
 405}
 406
 407static int gfs2_dirent_find(const struct gfs2_dirent *dent,
 408			    const struct qstr *name,
 409			    void *opaque)
 410{
 411	return __gfs2_dirent_find(dent, name, 1);
 412}
 413
 414static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
 415			    const struct qstr *name,
 416			    void *opaque)
 417{
 418	return __gfs2_dirent_find(dent, name, 2);
 419}
 420
 421/*
 422 * name->name holds ptr to start of block.
 423 * name->len holds size of block.
 424 */
 425static int gfs2_dirent_last(const struct gfs2_dirent *dent,
 426			    const struct qstr *name,
 427			    void *opaque)
 428{
 429	const char *start = name->name;
 430	const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
 431	if (name->len == (end - start))
 432		return 1;
 433	return 0;
 434}
 435
 436static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
 437				  const struct qstr *name,
 438				  void *opaque)
 439{
 440	unsigned required = GFS2_DIRENT_SIZE(name->len);
 441	unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
 442	unsigned totlen = be16_to_cpu(dent->de_rec_len);
 443
 444	if (gfs2_dirent_sentinel(dent))
 445		actual = 0;
 446	if (totlen - actual >= required)
 447		return 1;
 448	return 0;
 449}
 450
 451struct dirent_gather {
 452	const struct gfs2_dirent **pdent;
 453	unsigned offset;
 454};
 455
 456static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
 457			      const struct qstr *name,
 458			      void *opaque)
 459{
 460	struct dirent_gather *g = opaque;
 461	if (!gfs2_dirent_sentinel(dent)) {
 462		g->pdent[g->offset++] = dent;
 463	}
 464	return 0;
 465}
 466
 467/*
 468 * Other possible things to check:
 469 * - Inode located within filesystem size (and on valid block)
 470 * - Valid directory entry type
 471 * Not sure how heavy-weight we want to make this... could also check
 472 * hash is correct for example, but that would take a lot of extra time.
 473 * For now the most important thing is to check that the various sizes
 474 * are correct.
 475 */
 476static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
 477			     unsigned int size, unsigned int len, int first)
 478{
 479	const char *msg = "gfs2_dirent too small";
 480	if (unlikely(size < sizeof(struct gfs2_dirent)))
 481		goto error;
 482	msg = "gfs2_dirent misaligned";
 483	if (unlikely(offset & 0x7))
 484		goto error;
 485	msg = "gfs2_dirent points beyond end of block";
 486	if (unlikely(offset + size > len))
 487		goto error;
 488	msg = "zero inode number";
 489	if (unlikely(!first && gfs2_dirent_sentinel(dent)))
 490		goto error;
 491	msg = "name length is greater than space in dirent";
 492	if (!gfs2_dirent_sentinel(dent) &&
 493	    unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
 494		     size))
 495		goto error;
 496	return 0;
 497error:
 498	printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg,
 499	       first ? "first in block" : "not first in block");
 500	return -EIO;
 501}
 502
 503static int gfs2_dirent_offset(const void *buf)
 504{
 505	const struct gfs2_meta_header *h = buf;
 506	int offset;
 507
 508	BUG_ON(buf == NULL);
 509
 510	switch(be32_to_cpu(h->mh_type)) {
 511	case GFS2_METATYPE_LF:
 512		offset = sizeof(struct gfs2_leaf);
 513		break;
 514	case GFS2_METATYPE_DI:
 515		offset = sizeof(struct gfs2_dinode);
 516		break;
 517	default:
 518		goto wrong_type;
 519	}
 520	return offset;
 521wrong_type:
 522	printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n",
 523	       be32_to_cpu(h->mh_type));
 524	return -1;
 525}
 526
 527static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
 528					    unsigned int len, gfs2_dscan_t scan,
 529					    const struct qstr *name,
 530					    void *opaque)
 531{
 532	struct gfs2_dirent *dent, *prev;
 533	unsigned offset;
 534	unsigned size;
 535	int ret = 0;
 536
 537	ret = gfs2_dirent_offset(buf);
 538	if (ret < 0)
 539		goto consist_inode;
 540
 541	offset = ret;
 542	prev = NULL;
 543	dent = buf + offset;
 544	size = be16_to_cpu(dent->de_rec_len);
 545	if (gfs2_check_dirent(dent, offset, size, len, 1))
 546		goto consist_inode;
 547	do {
 548		ret = scan(dent, name, opaque);
 549		if (ret)
 550			break;
 551		offset += size;
 552		if (offset == len)
 553			break;
 554		prev = dent;
 555		dent = buf + offset;
 556		size = be16_to_cpu(dent->de_rec_len);
 557		if (gfs2_check_dirent(dent, offset, size, len, 0))
 558			goto consist_inode;
 559	} while(1);
 560
 561	switch(ret) {
 562	case 0:
 563		return NULL;
 564	case 1:
 565		return dent;
 566	case 2:
 567		return prev ? prev : dent;
 568	default:
 569		BUG_ON(ret > 0);
 570		return ERR_PTR(ret);
 571	}
 572
 573consist_inode:
 574	gfs2_consist_inode(GFS2_I(inode));
 575	return ERR_PTR(-EIO);
 576}
 577
 578static int dirent_check_reclen(struct gfs2_inode *dip,
 579			       const struct gfs2_dirent *d, const void *end_p)
 580{
 581	const void *ptr = d;
 582	u16 rec_len = be16_to_cpu(d->de_rec_len);
 583
 584	if (unlikely(rec_len < sizeof(struct gfs2_dirent)))
 585		goto broken;
 586	ptr += rec_len;
 587	if (ptr < end_p)
 588		return rec_len;
 589	if (ptr == end_p)
 590		return -ENOENT;
 591broken:
 592	gfs2_consist_inode(dip);
 593	return -EIO;
 594}
 595
 596/**
 597 * dirent_next - Next dirent
 598 * @dip: the directory
 599 * @bh: The buffer
 600 * @dent: Pointer to list of dirents
 601 *
 602 * Returns: 0 on success, error code otherwise
 603 */
 604
 605static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
 606		       struct gfs2_dirent **dent)
 607{
 608	struct gfs2_dirent *cur = *dent, *tmp;
 609	char *bh_end = bh->b_data + bh->b_size;
 610	int ret;
 611
 612	ret = dirent_check_reclen(dip, cur, bh_end);
 613	if (ret < 0)
 614		return ret;
 615
 616	tmp = (void *)cur + ret;
 617	ret = dirent_check_reclen(dip, tmp, bh_end);
 618	if (ret == -EIO)
 619		return ret;
 620
 621        /* Only the first dent could ever have de_inum.no_addr == 0 */
 622	if (gfs2_dirent_sentinel(tmp)) {
 623		gfs2_consist_inode(dip);
 624		return -EIO;
 625	}
 626
 627	*dent = tmp;
 628	return 0;
 629}
 630
 631/**
 632 * dirent_del - Delete a dirent
 633 * @dip: The GFS2 inode
 634 * @bh: The buffer
 635 * @prev: The previous dirent
 636 * @cur: The current dirent
 637 *
 638 */
 639
 640static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
 641		       struct gfs2_dirent *prev, struct gfs2_dirent *cur)
 642{
 643	u16 cur_rec_len, prev_rec_len;
 644
 645	if (gfs2_dirent_sentinel(cur)) {
 646		gfs2_consist_inode(dip);
 647		return;
 648	}
 649
 650	gfs2_trans_add_meta(dip->i_gl, bh);
 651
 652	/* If there is no prev entry, this is the first entry in the block.
 653	   The de_rec_len is already as big as it needs to be.  Just zero
 654	   out the inode number and return.  */
 655
 656	if (!prev) {
 657		cur->de_inum.no_addr = 0;
 658		cur->de_inum.no_formal_ino = 0;
 659		return;
 660	}
 661
 662	/*  Combine this dentry with the previous one.  */
 663
 664	prev_rec_len = be16_to_cpu(prev->de_rec_len);
 665	cur_rec_len = be16_to_cpu(cur->de_rec_len);
 666
 667	if ((char *)prev + prev_rec_len != (char *)cur)
 668		gfs2_consist_inode(dip);
 669	if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
 670		gfs2_consist_inode(dip);
 671
 672	prev_rec_len += cur_rec_len;
 673	prev->de_rec_len = cpu_to_be16(prev_rec_len);
 674}
 675
 676/*
 677 * Takes a dent from which to grab space as an argument. Returns the
 678 * newly created dent.
 679 */
 680static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
 681					    struct gfs2_dirent *dent,
 682					    const struct qstr *name,
 683					    struct buffer_head *bh)
 684{
 685	struct gfs2_inode *ip = GFS2_I(inode);
 686	struct gfs2_dirent *ndent;
 687	unsigned offset = 0, totlen;
 688
 689	if (!gfs2_dirent_sentinel(dent))
 690		offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
 691	totlen = be16_to_cpu(dent->de_rec_len);
 692	BUG_ON(offset + name->len > totlen);
 693	gfs2_trans_add_meta(ip->i_gl, bh);
 694	ndent = (struct gfs2_dirent *)((char *)dent + offset);
 695	dent->de_rec_len = cpu_to_be16(offset);
 696	gfs2_qstr2dirent(name, totlen - offset, ndent);
 697	return ndent;
 698}
 699
 700static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
 701					     struct buffer_head *bh,
 702					     const struct qstr *name)
 703{
 704	struct gfs2_dirent *dent;
 705	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
 706				gfs2_dirent_find_space, name, NULL);
 707	if (!dent || IS_ERR(dent))
 708		return dent;
 709	return gfs2_init_dirent(inode, dent, name, bh);
 710}
 711
 712static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
 713		    struct buffer_head **bhp)
 714{
 715	int error;
 716
 717	error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp);
 718	if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
 719		/* printk(KERN_INFO "block num=%llu\n", leaf_no); */
 720		error = -EIO;
 721	}
 722
 723	return error;
 724}
 725
 726/**
 727 * get_leaf_nr - Get a leaf number associated with the index
 728 * @dip: The GFS2 inode
 729 * @index:
 730 * @leaf_out:
 731 *
 732 * Returns: 0 on success, error code otherwise
 733 */
 734
 735static int get_leaf_nr(struct gfs2_inode *dip, u32 index,
 736		       u64 *leaf_out)
 737{
 738	__be64 *hash;
 739
 740	hash = gfs2_dir_get_hash_table(dip);
 741	if (IS_ERR(hash))
 742		return PTR_ERR(hash);
 743	*leaf_out = be64_to_cpu(*(hash + index));
 744	return 0;
 745}
 746
 747static int get_first_leaf(struct gfs2_inode *dip, u32 index,
 748			  struct buffer_head **bh_out)
 749{
 750	u64 leaf_no;
 751	int error;
 752
 753	error = get_leaf_nr(dip, index, &leaf_no);
 754	if (!error)
 755		error = get_leaf(dip, leaf_no, bh_out);
 756
 757	return error;
 758}
 759
 760static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
 761					      const struct qstr *name,
 762					      gfs2_dscan_t scan,
 763					      struct buffer_head **pbh)
 764{
 765	struct buffer_head *bh;
 766	struct gfs2_dirent *dent;
 767	struct gfs2_inode *ip = GFS2_I(inode);
 768	int error;
 769
 770	if (ip->i_diskflags & GFS2_DIF_EXHASH) {
 771		struct gfs2_leaf *leaf;
 772		unsigned hsize = 1 << ip->i_depth;
 773		unsigned index;
 774		u64 ln;
 775		if (hsize * sizeof(u64) != i_size_read(inode)) {
 776			gfs2_consist_inode(ip);
 777			return ERR_PTR(-EIO);
 778		}
 779
 780		index = name->hash >> (32 - ip->i_depth);
 781		error = get_first_leaf(ip, index, &bh);
 782		if (error)
 783			return ERR_PTR(error);
 784		do {
 785			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
 786						scan, name, NULL);
 787			if (dent)
 788				goto got_dent;
 789			leaf = (struct gfs2_leaf *)bh->b_data;
 790			ln = be64_to_cpu(leaf->lf_next);
 791			brelse(bh);
 792			if (!ln)
 793				break;
 794
 795			error = get_leaf(ip, ln, &bh);
 796		} while(!error);
 797
 798		return error ? ERR_PTR(error) : NULL;
 799	}
 800
 801
 802	error = gfs2_meta_inode_buffer(ip, &bh);
 803	if (error)
 804		return ERR_PTR(error);
 805	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
 806got_dent:
 807	if (unlikely(dent == NULL || IS_ERR(dent))) {
 808		brelse(bh);
 809		bh = NULL;
 810	}
 811	*pbh = bh;
 812	return dent;
 813}
 814
 815static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
 816{
 817	struct gfs2_inode *ip = GFS2_I(inode);
 818	unsigned int n = 1;
 819	u64 bn;
 820	int error;
 821	struct buffer_head *bh;
 822	struct gfs2_leaf *leaf;
 823	struct gfs2_dirent *dent;
 824	struct qstr name = { .name = "" };
 825
 826	error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
 827	if (error)
 828		return NULL;
 829	bh = gfs2_meta_new(ip->i_gl, bn);
 830	if (!bh)
 831		return NULL;
 832
 833	gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1);
 834	gfs2_trans_add_meta(ip->i_gl, bh);
 835	gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
 836	leaf = (struct gfs2_leaf *)bh->b_data;
 837	leaf->lf_depth = cpu_to_be16(depth);
 838	leaf->lf_entries = 0;
 839	leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
 840	leaf->lf_next = 0;
 841	memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved));
 842	dent = (struct gfs2_dirent *)(leaf+1);
 843	gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
 844	*pbh = bh;
 845	return leaf;
 846}
 847
 848/**
 849 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
 850 * @dip: The GFS2 inode
 851 *
 852 * Returns: 0 on success, error code otherwise
 853 */
 854
 855static int dir_make_exhash(struct inode *inode)
 856{
 857	struct gfs2_inode *dip = GFS2_I(inode);
 858	struct gfs2_sbd *sdp = GFS2_SB(inode);
 859	struct gfs2_dirent *dent;
 860	struct qstr args;
 861	struct buffer_head *bh, *dibh;
 862	struct gfs2_leaf *leaf;
 863	int y;
 864	u32 x;
 865	__be64 *lp;
 866	u64 bn;
 867	int error;
 868
 869	error = gfs2_meta_inode_buffer(dip, &dibh);
 870	if (error)
 871		return error;
 872
 873	/*  Turn over a new leaf  */
 874
 875	leaf = new_leaf(inode, &bh, 0);
 876	if (!leaf)
 877		return -ENOSPC;
 878	bn = bh->b_blocknr;
 879
 880	gfs2_assert(sdp, dip->i_entries < (1 << 16));
 881	leaf->lf_entries = cpu_to_be16(dip->i_entries);
 882
 883	/*  Copy dirents  */
 884
 885	gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
 886			     sizeof(struct gfs2_dinode));
 887
 888	/*  Find last entry  */
 889
 890	x = 0;
 891	args.len = bh->b_size - sizeof(struct gfs2_dinode) +
 892		   sizeof(struct gfs2_leaf);
 893	args.name = bh->b_data;
 894	dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
 895				gfs2_dirent_last, &args, NULL);
 896	if (!dent) {
 897		brelse(bh);
 898		brelse(dibh);
 899		return -EIO;
 900	}
 901	if (IS_ERR(dent)) {
 902		brelse(bh);
 903		brelse(dibh);
 904		return PTR_ERR(dent);
 905	}
 906
 907	/*  Adjust the last dirent's record length
 908	   (Remember that dent still points to the last entry.)  */
 909
 910	dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
 911		sizeof(struct gfs2_dinode) -
 912		sizeof(struct gfs2_leaf));
 913
 914	brelse(bh);
 915
 916	/*  We're done with the new leaf block, now setup the new
 917	    hash table.  */
 918
 919	gfs2_trans_add_meta(dip->i_gl, dibh);
 920	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
 921
 922	lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
 923
 924	for (x = sdp->sd_hash_ptrs; x--; lp++)
 925		*lp = cpu_to_be64(bn);
 926
 927	i_size_write(inode, sdp->sd_sb.sb_bsize / 2);
 928	gfs2_add_inode_blocks(&dip->i_inode, 1);
 929	dip->i_diskflags |= GFS2_DIF_EXHASH;
 930
 931	for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
 932	dip->i_depth = y;
 933
 934	gfs2_dinode_out(dip, dibh->b_data);
 935
 936	brelse(dibh);
 937
 938	return 0;
 939}
 940
 941/**
 942 * dir_split_leaf - Split a leaf block into two
 943 * @dip: The GFS2 inode
 944 * @index:
 945 * @leaf_no:
 946 *
 947 * Returns: 0 on success, error code on failure
 948 */
 949
 950static int dir_split_leaf(struct inode *inode, const struct qstr *name)
 951{
 952	struct gfs2_inode *dip = GFS2_I(inode);
 953	struct buffer_head *nbh, *obh, *dibh;
 954	struct gfs2_leaf *nleaf, *oleaf;
 955	struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
 956	u32 start, len, half_len, divider;
 957	u64 bn, leaf_no;
 958	__be64 *lp;
 959	u32 index;
 960	int x, moved = 0;
 961	int error;
 962
 963	index = name->hash >> (32 - dip->i_depth);
 964	error = get_leaf_nr(dip, index, &leaf_no);
 965	if (error)
 966		return error;
 967
 968	/*  Get the old leaf block  */
 969	error = get_leaf(dip, leaf_no, &obh);
 970	if (error)
 971		return error;
 972
 973	oleaf = (struct gfs2_leaf *)obh->b_data;
 974	if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
 975		brelse(obh);
 976		return 1; /* can't split */
 977	}
 978
 979	gfs2_trans_add_meta(dip->i_gl, obh);
 980
 981	nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
 982	if (!nleaf) {
 983		brelse(obh);
 984		return -ENOSPC;
 985	}
 986	bn = nbh->b_blocknr;
 987
 988	/*  Compute the start and len of leaf pointers in the hash table.  */
 989	len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth));
 990	half_len = len >> 1;
 991	if (!half_len) {
 992		printk(KERN_WARNING "i_depth %u lf_depth %u index %u\n", dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
 993		gfs2_consist_inode(dip);
 994		error = -EIO;
 995		goto fail_brelse;
 996	}
 997
 998	start = (index & ~(len - 1));
 999
1000	/* Change the pointers.
1001	   Don't bother distinguishing stuffed from non-stuffed.
1002	   This code is complicated enough already. */
1003	lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS);
1004	if (!lp) {
1005		error = -ENOMEM;
1006		goto fail_brelse;
1007	}
1008
1009	/*  Change the pointers  */
1010	for (x = 0; x < half_len; x++)
1011		lp[x] = cpu_to_be64(bn);
1012
1013	gfs2_dir_hash_inval(dip);
1014
1015	error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
1016				    half_len * sizeof(u64));
1017	if (error != half_len * sizeof(u64)) {
1018		if (error >= 0)
1019			error = -EIO;
1020		goto fail_lpfree;
1021	}
1022
1023	kfree(lp);
1024
1025	/*  Compute the divider  */
1026	divider = (start + half_len) << (32 - dip->i_depth);
1027
1028	/*  Copy the entries  */
1029	dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf));
1030
1031	do {
1032		next = dent;
1033		if (dirent_next(dip, obh, &next))
1034			next = NULL;
1035
1036		if (!gfs2_dirent_sentinel(dent) &&
1037		    be32_to_cpu(dent->de_hash) < divider) {
1038			struct qstr str;
1039			str.name = (char*)(dent+1);
1040			str.len = be16_to_cpu(dent->de_name_len);
1041			str.hash = be32_to_cpu(dent->de_hash);
1042			new = gfs2_dirent_alloc(inode, nbh, &str);
1043			if (IS_ERR(new)) {
1044				error = PTR_ERR(new);
1045				break;
1046			}
1047
1048			new->de_inum = dent->de_inum; /* No endian worries */
1049			new->de_type = dent->de_type; /* No endian worries */
1050			be16_add_cpu(&nleaf->lf_entries, 1);
1051
1052			dirent_del(dip, obh, prev, dent);
1053
1054			if (!oleaf->lf_entries)
1055				gfs2_consist_inode(dip);
1056			be16_add_cpu(&oleaf->lf_entries, -1);
1057
1058			if (!prev)
1059				prev = dent;
1060
1061			moved = 1;
1062		} else {
1063			prev = dent;
1064		}
1065		dent = next;
1066	} while (dent);
1067
1068	oleaf->lf_depth = nleaf->lf_depth;
1069
1070	error = gfs2_meta_inode_buffer(dip, &dibh);
1071	if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
1072		gfs2_trans_add_meta(dip->i_gl, dibh);
1073		gfs2_add_inode_blocks(&dip->i_inode, 1);
1074		gfs2_dinode_out(dip, dibh->b_data);
1075		brelse(dibh);
1076	}
1077
1078	brelse(obh);
1079	brelse(nbh);
1080
1081	return error;
1082
1083fail_lpfree:
1084	kfree(lp);
1085
1086fail_brelse:
1087	brelse(obh);
1088	brelse(nbh);
1089	return error;
1090}
1091
1092/**
1093 * dir_double_exhash - Double size of ExHash table
1094 * @dip: The GFS2 dinode
1095 *
1096 * Returns: 0 on success, error code on failure
1097 */
1098
1099static int dir_double_exhash(struct gfs2_inode *dip)
1100{
1101	struct buffer_head *dibh;
1102	u32 hsize;
1103	u32 hsize_bytes;
1104	__be64 *hc;
1105	__be64 *hc2, *h;
1106	int x;
1107	int error = 0;
1108
1109	hsize = 1 << dip->i_depth;
1110	hsize_bytes = hsize * sizeof(__be64);
1111
1112	hc = gfs2_dir_get_hash_table(dip);
1113	if (IS_ERR(hc))
1114		return PTR_ERR(hc);
1115
1116	h = hc2 = kmalloc(hsize_bytes * 2, GFP_NOFS);
1117	if (!hc2)
1118		return -ENOMEM;
1119
1120	error = gfs2_meta_inode_buffer(dip, &dibh);
1121	if (error)
1122		goto out_kfree;
1123
1124	for (x = 0; x < hsize; x++) {
1125		*h++ = *hc;
1126		*h++ = *hc;
1127		hc++;
1128	}
1129
1130	error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2);
1131	if (error != (hsize_bytes * 2))
1132		goto fail;
1133
1134	gfs2_dir_hash_inval(dip);
1135	dip->i_hash_cache = hc2;
1136	dip->i_depth++;
1137	gfs2_dinode_out(dip, dibh->b_data);
1138	brelse(dibh);
1139	return 0;
1140
1141fail:
1142	/* Replace original hash table & size */
1143	gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes);
1144	i_size_write(&dip->i_inode, hsize_bytes);
1145	gfs2_dinode_out(dip, dibh->b_data);
1146	brelse(dibh);
1147out_kfree:
1148	kfree(hc2);
1149	return error;
1150}
1151
1152/**
1153 * compare_dents - compare directory entries by hash value
1154 * @a: first dent
1155 * @b: second dent
1156 *
1157 * When comparing the hash entries of @a to @b:
1158 *   gt: returns 1
1159 *   lt: returns -1
1160 *   eq: returns 0
1161 */
1162
1163static int compare_dents(const void *a, const void *b)
1164{
1165	const struct gfs2_dirent *dent_a, *dent_b;
1166	u32 hash_a, hash_b;
1167	int ret = 0;
1168
1169	dent_a = *(const struct gfs2_dirent **)a;
1170	hash_a = be32_to_cpu(dent_a->de_hash);
1171
1172	dent_b = *(const struct gfs2_dirent **)b;
1173	hash_b = be32_to_cpu(dent_b->de_hash);
1174
1175	if (hash_a > hash_b)
1176		ret = 1;
1177	else if (hash_a < hash_b)
1178		ret = -1;
1179	else {
1180		unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1181		unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1182
1183		if (len_a > len_b)
1184			ret = 1;
1185		else if (len_a < len_b)
1186			ret = -1;
1187		else
1188			ret = memcmp(dent_a + 1, dent_b + 1, len_a);
1189	}
1190
1191	return ret;
1192}
1193
1194/**
1195 * do_filldir_main - read out directory entries
1196 * @dip: The GFS2 inode
1197 * @offset: The offset in the file to read from
1198 * @opaque: opaque data to pass to filldir
1199 * @filldir: The function to pass entries to
1200 * @darr: an array of struct gfs2_dirent pointers to read
1201 * @entries: the number of entries in darr
1202 * @copied: pointer to int that's non-zero if a entry has been copied out
1203 *
1204 * Jump through some hoops to make sure that if there are hash collsions,
1205 * they are read out at the beginning of a buffer.  We want to minimize
1206 * the possibility that they will fall into different readdir buffers or
1207 * that someone will want to seek to that location.
1208 *
1209 * Returns: errno, >0 on exception from filldir
1210 */
1211
1212static int do_filldir_main(struct gfs2_inode *dip, u64 *offset,
1213			   void *opaque, filldir_t filldir,
1214			   const struct gfs2_dirent **darr, u32 entries,
1215			   int *copied)
1216{
1217	const struct gfs2_dirent *dent, *dent_next;
1218	u64 off, off_next;
1219	unsigned int x, y;
1220	int run = 0;
1221	int error = 0;
1222
1223	sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
1224
1225	dent_next = darr[0];
1226	off_next = be32_to_cpu(dent_next->de_hash);
1227	off_next = gfs2_disk_hash2offset(off_next);
1228
1229	for (x = 0, y = 1; x < entries; x++, y++) {
1230		dent = dent_next;
1231		off = off_next;
1232
1233		if (y < entries) {
1234			dent_next = darr[y];
1235			off_next = be32_to_cpu(dent_next->de_hash);
1236			off_next = gfs2_disk_hash2offset(off_next);
1237
1238			if (off < *offset)
1239				continue;
1240			*offset = off;
1241
1242			if (off_next == off) {
1243				if (*copied && !run)
1244					return 1;
1245				run = 1;
1246			} else
1247				run = 0;
1248		} else {
1249			if (off < *offset)
1250				continue;
1251			*offset = off;
1252		}
1253
1254		error = filldir(opaque, (const char *)(dent + 1),
1255				be16_to_cpu(dent->de_name_len),
1256				off, be64_to_cpu(dent->de_inum.no_addr),
1257				be16_to_cpu(dent->de_type));
1258		if (error)
1259			return 1;
1260
1261		*copied = 1;
1262	}
1263
1264	/* Increment the *offset by one, so the next time we come into the
1265	   do_filldir fxn, we get the next entry instead of the last one in the
1266	   current leaf */
1267
1268	(*offset)++;
1269
1270	return 0;
1271}
1272
1273static void *gfs2_alloc_sort_buffer(unsigned size)
1274{
1275	void *ptr = NULL;
1276
1277	if (size < KMALLOC_MAX_SIZE)
1278		ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
1279	if (!ptr)
1280		ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL);
1281	return ptr;
1282}
1283
1284static void gfs2_free_sort_buffer(void *ptr)
1285{
1286	if (is_vmalloc_addr(ptr))
1287		vfree(ptr);
1288	else
1289		kfree(ptr);
1290}
1291
1292static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque,
1293			      filldir_t filldir, int *copied, unsigned *depth,
1294			      u64 leaf_no)
1295{
1296	struct gfs2_inode *ip = GFS2_I(inode);
1297	struct gfs2_sbd *sdp = GFS2_SB(inode);
1298	struct buffer_head *bh;
1299	struct gfs2_leaf *lf;
1300	unsigned entries = 0, entries2 = 0;
1301	unsigned leaves = 0;
1302	const struct gfs2_dirent **darr, *dent;
1303	struct dirent_gather g;
1304	struct buffer_head **larr;
1305	int leaf = 0;
1306	int error, i;
1307	u64 lfn = leaf_no;
1308
1309	do {
1310		error = get_leaf(ip, lfn, &bh);
1311		if (error)
1312			goto out;
1313		lf = (struct gfs2_leaf *)bh->b_data;
1314		if (leaves == 0)
1315			*depth = be16_to_cpu(lf->lf_depth);
1316		entries += be16_to_cpu(lf->lf_entries);
1317		leaves++;
1318		lfn = be64_to_cpu(lf->lf_next);
1319		brelse(bh);
1320	} while(lfn);
1321
1322	if (!entries)
1323		return 0;
1324
1325	error = -ENOMEM;
1326	/*
1327	 * The extra 99 entries are not normally used, but are a buffer
1328	 * zone in case the number of entries in the leaf is corrupt.
1329	 * 99 is the maximum number of entries that can fit in a single
1330	 * leaf block.
1331	 */
1332	larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
1333	if (!larr)
1334		goto out;
1335	darr = (const struct gfs2_dirent **)(larr + leaves);
1336	g.pdent = darr;
1337	g.offset = 0;
1338	lfn = leaf_no;
1339
1340	do {
1341		error = get_leaf(ip, lfn, &bh);
1342		if (error)
1343			goto out_free;
1344		lf = (struct gfs2_leaf *)bh->b_data;
1345		lfn = be64_to_cpu(lf->lf_next);
1346		if (lf->lf_entries) {
1347			entries2 += be16_to_cpu(lf->lf_entries);
1348			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1349						gfs2_dirent_gather, NULL, &g);
1350			error = PTR_ERR(dent);
1351			if (IS_ERR(dent))
1352				goto out_free;
1353			if (entries2 != g.offset) {
1354				fs_warn(sdp, "Number of entries corrupt in dir "
1355						"leaf %llu, entries2 (%u) != "
1356						"g.offset (%u)\n",
1357					(unsigned long long)bh->b_blocknr,
1358					entries2, g.offset);
1359					
1360				error = -EIO;
1361				goto out_free;
1362			}
1363			error = 0;
1364			larr[leaf++] = bh;
1365		} else {
1366			brelse(bh);
1367		}
1368	} while(lfn);
1369
1370	BUG_ON(entries2 != entries);
1371	error = do_filldir_main(ip, offset, opaque, filldir, darr,
1372				entries, copied);
1373out_free:
1374	for(i = 0; i < leaf; i++)
1375		brelse(larr[i]);
1376	gfs2_free_sort_buffer(larr);
1377out:
1378	return error;
1379}
1380
1381/**
1382 * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks.
1383 *
1384 * Note: we can't calculate each index like dir_e_read can because we don't
1385 * have the leaf, and therefore we don't have the depth, and therefore we
1386 * don't have the length. So we have to just read enough ahead to make up
1387 * for the loss of information.
1388 */
1389static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index,
1390			       struct file_ra_state *f_ra)
1391{
1392	struct gfs2_inode *ip = GFS2_I(inode);
1393	struct gfs2_glock *gl = ip->i_gl;
1394	struct buffer_head *bh;
1395	u64 blocknr = 0, last;
1396	unsigned count;
1397
1398	/* First check if we've already read-ahead for the whole range. */
1399	if (index + MAX_RA_BLOCKS < f_ra->start)
1400		return;
1401
1402	f_ra->start = max((pgoff_t)index, f_ra->start);
1403	for (count = 0; count < MAX_RA_BLOCKS; count++) {
1404		if (f_ra->start >= hsize) /* if exceeded the hash table */
1405			break;
1406
1407		last = blocknr;
1408		blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]);
1409		f_ra->start++;
1410		if (blocknr == last)
1411			continue;
1412
1413		bh = gfs2_getbuf(gl, blocknr, 1);
1414		if (trylock_buffer(bh)) {
1415			if (buffer_uptodate(bh)) {
1416				unlock_buffer(bh);
1417				brelse(bh);
1418				continue;
1419			}
1420			bh->b_end_io = end_buffer_read_sync;
1421			submit_bh(READA | REQ_META, bh);
1422			continue;
1423		}
1424		brelse(bh);
1425	}
1426}
1427
1428/**
1429 * dir_e_read - Reads the entries from a directory into a filldir buffer
1430 * @dip: dinode pointer
1431 * @offset: the hash of the last entry read shifted to the right once
1432 * @opaque: buffer for the filldir function to fill
1433 * @filldir: points to the filldir function to use
1434 *
1435 * Returns: errno
1436 */
1437
1438static int dir_e_read(struct inode *inode, u64 *offset, void *opaque,
1439		      filldir_t filldir, struct file_ra_state *f_ra)
1440{
1441	struct gfs2_inode *dip = GFS2_I(inode);
1442	u32 hsize, len = 0;
1443	u32 hash, index;
1444	__be64 *lp;
1445	int copied = 0;
1446	int error = 0;
1447	unsigned depth = 0;
1448
1449	hsize = 1 << dip->i_depth;
1450	hash = gfs2_dir_offset2hash(*offset);
1451	index = hash >> (32 - dip->i_depth);
1452
1453	if (dip->i_hash_cache == NULL)
1454		f_ra->start = 0;
1455	lp = gfs2_dir_get_hash_table(dip);
1456	if (IS_ERR(lp))
1457		return PTR_ERR(lp);
1458
1459	gfs2_dir_readahead(inode, hsize, index, f_ra);
1460
1461	while (index < hsize) {
1462		error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
1463					   &copied, &depth,
1464					   be64_to_cpu(lp[index]));
1465		if (error)
1466			break;
1467
1468		len = 1 << (dip->i_depth - depth);
1469		index = (index & ~(len - 1)) + len;
1470	}
1471
1472	if (error > 0)
1473		error = 0;
1474	return error;
1475}
1476
1477int gfs2_dir_read(struct inode *inode, u64 *offset, void *opaque,
1478		  filldir_t filldir, struct file_ra_state *f_ra)
1479{
1480	struct gfs2_inode *dip = GFS2_I(inode);
1481	struct gfs2_sbd *sdp = GFS2_SB(inode);
1482	struct dirent_gather g;
1483	const struct gfs2_dirent **darr, *dent;
1484	struct buffer_head *dibh;
1485	int copied = 0;
1486	int error;
1487
1488	if (!dip->i_entries)
1489		return 0;
1490
1491	if (dip->i_diskflags & GFS2_DIF_EXHASH)
1492		return dir_e_read(inode, offset, opaque, filldir, f_ra);
1493
1494	if (!gfs2_is_stuffed(dip)) {
1495		gfs2_consist_inode(dip);
1496		return -EIO;
1497	}
1498
1499	error = gfs2_meta_inode_buffer(dip, &dibh);
1500	if (error)
1501		return error;
1502
1503	error = -ENOMEM;
1504	/* 96 is max number of dirents which can be stuffed into an inode */
1505	darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS);
1506	if (darr) {
1507		g.pdent = darr;
1508		g.offset = 0;
1509		dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1510					gfs2_dirent_gather, NULL, &g);
1511		if (IS_ERR(dent)) {
1512			error = PTR_ERR(dent);
1513			goto out;
1514		}
1515		if (dip->i_entries != g.offset) {
1516			fs_warn(sdp, "Number of entries corrupt in dir %llu, "
1517				"ip->i_entries (%u) != g.offset (%u)\n",
1518				(unsigned long long)dip->i_no_addr,
1519				dip->i_entries,
1520				g.offset);
1521			error = -EIO;
1522			goto out;
1523		}
1524		error = do_filldir_main(dip, offset, opaque, filldir, darr,
1525					dip->i_entries, &copied);
1526out:
1527		kfree(darr);
1528	}
1529
1530	if (error > 0)
1531		error = 0;
1532
1533	brelse(dibh);
1534
1535	return error;
1536}
1537
1538/**
1539 * gfs2_dir_search - Search a directory
1540 * @dip: The GFS2 inode
1541 * @filename:
1542 * @inode:
1543 *
1544 * This routine searches a directory for a file or another directory.
1545 * Assumes a glock is held on dip.
1546 *
1547 * Returns: errno
1548 */
1549
1550struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name)
1551{
1552	struct buffer_head *bh;
1553	struct gfs2_dirent *dent;
1554	struct inode *inode;
1555
1556	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1557	if (dent) {
1558		if (IS_ERR(dent))
1559			return ERR_CAST(dent);
1560		inode = gfs2_inode_lookup(dir->i_sb, 
1561				be16_to_cpu(dent->de_type),
1562				be64_to_cpu(dent->de_inum.no_addr),
1563				be64_to_cpu(dent->de_inum.no_formal_ino), 0);
1564		brelse(bh);
1565		return inode;
1566	}
1567	return ERR_PTR(-ENOENT);
1568}
1569
1570int gfs2_dir_check(struct inode *dir, const struct qstr *name,
1571		   const struct gfs2_inode *ip)
1572{
1573	struct buffer_head *bh;
1574	struct gfs2_dirent *dent;
1575	int ret = -ENOENT;
1576
1577	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1578	if (dent) {
1579		if (IS_ERR(dent))
1580			return PTR_ERR(dent);
1581		if (ip) {
1582			if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
1583				goto out;
1584			if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
1585			    ip->i_no_formal_ino)
1586				goto out;
1587			if (unlikely(IF2DT(ip->i_inode.i_mode) !=
1588			    be16_to_cpu(dent->de_type))) {
1589				gfs2_consist_inode(GFS2_I(dir));
1590				ret = -EIO;
1591				goto out;
1592			}
1593		}
1594		ret = 0;
1595out:
1596		brelse(bh);
1597	}
1598	return ret;
1599}
1600
1601static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1602{
1603	struct buffer_head *bh, *obh;
1604	struct gfs2_inode *ip = GFS2_I(inode);
1605	struct gfs2_leaf *leaf, *oleaf;
1606	int error;
1607	u32 index;
1608	u64 bn;
1609
1610	index = name->hash >> (32 - ip->i_depth);
1611	error = get_first_leaf(ip, index, &obh);
1612	if (error)
1613		return error;
1614	do {
1615		oleaf = (struct gfs2_leaf *)obh->b_data;
1616		bn = be64_to_cpu(oleaf->lf_next);
1617		if (!bn)
1618			break;
1619		brelse(obh);
1620		error = get_leaf(ip, bn, &obh);
1621		if (error)
1622			return error;
1623	} while(1);
1624
1625	gfs2_trans_add_meta(ip->i_gl, obh);
1626
1627	leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1628	if (!leaf) {
1629		brelse(obh);
1630		return -ENOSPC;
1631	}
1632	oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1633	brelse(bh);
1634	brelse(obh);
1635
1636	error = gfs2_meta_inode_buffer(ip, &bh);
1637	if (error)
1638		return error;
1639	gfs2_trans_add_meta(ip->i_gl, bh);
1640	gfs2_add_inode_blocks(&ip->i_inode, 1);
1641	gfs2_dinode_out(ip, bh->b_data);
1642	brelse(bh);
1643	return 0;
1644}
1645
1646/**
1647 * gfs2_dir_add - Add new filename into directory
1648 * @dip: The GFS2 inode
1649 * @filename: The new name
1650 * @inode: The inode number of the entry
1651 * @type: The type of the entry
1652 *
1653 * Returns: 0 on success, error code on failure
1654 */
1655
1656int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1657		 const struct gfs2_inode *nip)
1658{
1659	struct gfs2_inode *ip = GFS2_I(inode);
1660	struct buffer_head *bh;
1661	struct gfs2_dirent *dent;
1662	struct gfs2_leaf *leaf;
1663	int error;
1664
1665	while(1) {
1666		dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space,
1667					  &bh);
1668		if (dent) {
1669			if (IS_ERR(dent))
1670				return PTR_ERR(dent);
1671			dent = gfs2_init_dirent(inode, dent, name, bh);
1672			gfs2_inum_out(nip, dent);
1673			dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode));
1674			if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1675				leaf = (struct gfs2_leaf *)bh->b_data;
1676				be16_add_cpu(&leaf->lf_entries, 1);
1677			}
1678			brelse(bh);
1679			ip->i_entries++;
1680			ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
1681			if (S_ISDIR(nip->i_inode.i_mode))
1682				inc_nlink(&ip->i_inode);
1683			mark_inode_dirty(inode);
1684			error = 0;
1685			break;
1686		}
1687		if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1688			error = dir_make_exhash(inode);
1689			if (error)
1690				break;
1691			continue;
1692		}
1693		error = dir_split_leaf(inode, name);
1694		if (error == 0)
1695			continue;
1696		if (error < 0)
1697			break;
1698		if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
1699			error = dir_double_exhash(ip);
1700			if (error)
1701				break;
1702			error = dir_split_leaf(inode, name);
1703			if (error < 0)
1704				break;
1705			if (error == 0)
1706				continue;
1707		}
1708		error = dir_new_leaf(inode, name);
1709		if (!error)
1710			continue;
1711		error = -ENOSPC;
1712		break;
1713	}
1714	return error;
1715}
1716
1717
1718/**
1719 * gfs2_dir_del - Delete a directory entry
1720 * @dip: The GFS2 inode
1721 * @filename: The filename
1722 *
1723 * Returns: 0 on success, error code on failure
1724 */
1725
1726int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry)
1727{
1728	const struct qstr *name = &dentry->d_name;
1729	struct gfs2_dirent *dent, *prev = NULL;
1730	struct buffer_head *bh;
1731
1732	/* Returns _either_ the entry (if its first in block) or the
1733	   previous entry otherwise */
1734	dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
1735	if (!dent) {
1736		gfs2_consist_inode(dip);
1737		return -EIO;
1738	}
1739	if (IS_ERR(dent)) {
1740		gfs2_consist_inode(dip);
1741		return PTR_ERR(dent);
1742	}
1743	/* If not first in block, adjust pointers accordingly */
1744	if (gfs2_dirent_find(dent, name, NULL) == 0) {
1745		prev = dent;
1746		dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1747	}
1748
1749	dirent_del(dip, bh, prev, dent);
1750	if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1751		struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1752		u16 entries = be16_to_cpu(leaf->lf_entries);
1753		if (!entries)
1754			gfs2_consist_inode(dip);
1755		leaf->lf_entries = cpu_to_be16(--entries);
1756	}
1757	brelse(bh);
1758
1759	if (!dip->i_entries)
1760		gfs2_consist_inode(dip);
1761	dip->i_entries--;
1762	dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1763	if (S_ISDIR(dentry->d_inode->i_mode))
1764		drop_nlink(&dip->i_inode);
1765	mark_inode_dirty(&dip->i_inode);
1766
1767	return 0;
1768}
1769
1770/**
1771 * gfs2_dir_mvino - Change inode number of directory entry
1772 * @dip: The GFS2 inode
1773 * @filename:
1774 * @new_inode:
1775 *
1776 * This routine changes the inode number of a directory entry.  It's used
1777 * by rename to change ".." when a directory is moved.
1778 * Assumes a glock is held on dvp.
1779 *
1780 * Returns: errno
1781 */
1782
1783int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1784		   const struct gfs2_inode *nip, unsigned int new_type)
1785{
1786	struct buffer_head *bh;
1787	struct gfs2_dirent *dent;
1788	int error;
1789
1790	dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
1791	if (!dent) {
1792		gfs2_consist_inode(dip);
1793		return -EIO;
1794	}
1795	if (IS_ERR(dent))
1796		return PTR_ERR(dent);
1797
1798	gfs2_trans_add_meta(dip->i_gl, bh);
1799	gfs2_inum_out(nip, dent);
1800	dent->de_type = cpu_to_be16(new_type);
1801
1802	if (dip->i_diskflags & GFS2_DIF_EXHASH) {
1803		brelse(bh);
1804		error = gfs2_meta_inode_buffer(dip, &bh);
1805		if (error)
1806			return error;
1807		gfs2_trans_add_meta(dip->i_gl, bh);
1808	}
1809
1810	dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
1811	gfs2_dinode_out(dip, bh->b_data);
1812	brelse(bh);
1813	return 0;
1814}
1815
1816/**
1817 * leaf_dealloc - Deallocate a directory leaf
1818 * @dip: the directory
1819 * @index: the hash table offset in the directory
1820 * @len: the number of pointers to this leaf
1821 * @leaf_no: the leaf number
1822 * @leaf_bh: buffer_head for the starting leaf
1823 * last_dealloc: 1 if this is the final dealloc for the leaf, else 0
1824 *
1825 * Returns: errno
1826 */
1827
1828static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
1829			u64 leaf_no, struct buffer_head *leaf_bh,
1830			int last_dealloc)
1831{
1832	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
1833	struct gfs2_leaf *tmp_leaf;
1834	struct gfs2_rgrp_list rlist;
1835	struct buffer_head *bh, *dibh;
1836	u64 blk, nblk;
1837	unsigned int rg_blocks = 0, l_blocks = 0;
1838	char *ht;
1839	unsigned int x, size = len * sizeof(u64);
1840	int error;
1841
1842	error = gfs2_rindex_update(sdp);
1843	if (error)
1844		return error;
1845
1846	memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1847
1848	ht = kzalloc(size, GFP_NOFS);
1849	if (!ht)
1850		return -ENOMEM;
1851
1852	error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1853	if (error)
1854		goto out;
1855
1856	/*  Count the number of leaves  */
1857	bh = leaf_bh;
1858
1859	for (blk = leaf_no; blk; blk = nblk) {
1860		if (blk != leaf_no) {
1861			error = get_leaf(dip, blk, &bh);
1862			if (error)
1863				goto out_rlist;
1864		}
1865		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1866		nblk = be64_to_cpu(tmp_leaf->lf_next);
1867		if (blk != leaf_no)
1868			brelse(bh);
1869
1870		gfs2_rlist_add(dip, &rlist, blk);
1871		l_blocks++;
1872	}
1873
1874	gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
1875
1876	for (x = 0; x < rlist.rl_rgrps; x++) {
1877		struct gfs2_rgrpd *rgd;
1878		rgd = rlist.rl_ghs[x].gh_gl->gl_object;
1879		rg_blocks += rgd->rd_length;
1880	}
1881
1882	error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
1883	if (error)
1884		goto out_rlist;
1885
1886	error = gfs2_trans_begin(sdp,
1887			rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
1888			RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
1889	if (error)
1890		goto out_rg_gunlock;
1891
1892	bh = leaf_bh;
1893
1894	for (blk = leaf_no; blk; blk = nblk) {
1895		if (blk != leaf_no) {
1896			error = get_leaf(dip, blk, &bh);
1897			if (error)
1898				goto out_end_trans;
1899		}
1900		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1901		nblk = be64_to_cpu(tmp_leaf->lf_next);
1902		if (blk != leaf_no)
1903			brelse(bh);
1904
1905		gfs2_free_meta(dip, blk, 1);
1906		gfs2_add_inode_blocks(&dip->i_inode, -1);
1907	}
1908
1909	error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
1910	if (error != size) {
1911		if (error >= 0)
1912			error = -EIO;
1913		goto out_end_trans;
1914	}
1915
1916	error = gfs2_meta_inode_buffer(dip, &dibh);
1917	if (error)
1918		goto out_end_trans;
1919
1920	gfs2_trans_add_meta(dip->i_gl, dibh);
1921	/* On the last dealloc, make this a regular file in case we crash.
1922	   (We don't want to free these blocks a second time.)  */
1923	if (last_dealloc)
1924		dip->i_inode.i_mode = S_IFREG;
1925	gfs2_dinode_out(dip, dibh->b_data);
1926	brelse(dibh);
1927
1928out_end_trans:
1929	gfs2_trans_end(sdp);
1930out_rg_gunlock:
1931	gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
1932out_rlist:
1933	gfs2_rlist_free(&rlist);
1934	gfs2_quota_unhold(dip);
1935out:
1936	kfree(ht);
1937	return error;
1938}
1939
1940/**
1941 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1942 * @dip: the directory
1943 *
1944 * Dealloc all on-disk directory leaves to FREEMETA state
1945 * Change on-disk inode type to "regular file"
1946 *
1947 * Returns: errno
1948 */
1949
1950int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
1951{
1952	struct buffer_head *bh;
1953	struct gfs2_leaf *leaf;
1954	u32 hsize, len;
1955	u32 index = 0, next_index;
1956	__be64 *lp;
1957	u64 leaf_no;
1958	int error = 0, last;
1959
1960	hsize = 1 << dip->i_depth;
1961
1962	lp = gfs2_dir_get_hash_table(dip);
1963	if (IS_ERR(lp))
1964		return PTR_ERR(lp);
1965
1966	while (index < hsize) {
1967		leaf_no = be64_to_cpu(lp[index]);
1968		if (leaf_no) {
1969			error = get_leaf(dip, leaf_no, &bh);
1970			if (error)
1971				goto out;
1972			leaf = (struct gfs2_leaf *)bh->b_data;
1973			len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth));
1974
1975			next_index = (index & ~(len - 1)) + len;
1976			last = ((next_index >= hsize) ? 1 : 0);
1977			error = leaf_dealloc(dip, index, len, leaf_no, bh,
1978					     last);
1979			brelse(bh);
1980			if (error)
1981				goto out;
1982			index = next_index;
1983		} else
1984			index++;
1985	}
1986
1987	if (index != hsize) {
1988		gfs2_consist_inode(dip);
1989		error = -EIO;
1990	}
1991
1992out:
1993
1994	return error;
1995}
1996
1997/**
1998 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
1999 * @ip: the file being written to
2000 * @filname: the filename that's going to be added
2001 *
2002 * Returns: 1 if alloc required, 0 if not, -ve on error
2003 */
2004
2005int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name)
2006{
2007	struct gfs2_dirent *dent;
2008	struct buffer_head *bh;
2009
2010	dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
2011	if (!dent) {
2012		return 1;
2013	}
2014	if (IS_ERR(dent))
2015		return PTR_ERR(dent);
2016	brelse(bh);
2017	return 0;
2018}
2019