[PATCH 07/16] GFS2: Directory handling

[PATCH 07/16] GFS2: Directory handling

[Steven Whitehouse]

[PATCH 07/16] GFS2: Directory handling

Directory handling code for GFS2.  GFS2 uses extensible hashing in the
directory code which makes lookups fast. Due to the way in which the hash
tables are split during directory expansion, the directory contents are
sorted during readdir and it would be nice to be able to eliminate this
operation. Suggestions are welcome, but note that we need to maintain
backward compatibility with GFS1's on disk format.


Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>


 fs/gfs2/dir.c | 1976 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 fs/gfs2/dir.h |   73 ++
 2 files changed, 2049 insertions(+)

--- /dev/null
+++ b/fs/gfs2/dir.c
@@ -0,0 +1,1976 @@
+/*
+ * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
+ * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ */
+
+/*
+ * Implements Extendible Hashing as described in:
+ *   "Extendible Hashing" by Fagin, et al in
+ *     __ACM Trans. on Database Systems__, Sept 1979.
+ *
+ *
+ * Here's the layout of dirents which is essentially the same as that of ext2
+ * within a single block. The field de_name_len is the number of bytes
+ * actually required for the name (no null terminator). The field de_rec_len
+ * is the number of bytes allocated to the dirent. The offset of the next
+ * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
+ * deleted, the preceding dirent inherits its allocated space, ie
+ * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
+ * by adding de_rec_len to the current dirent, this essentially causes the
+ * deleted dirent to get jumped over when iterating through all the dirents.
+ *
+ * When deleting the first dirent in a block, there is no previous dirent so
+ * the field de_ino is set to zero to designate it as deleted. When allocating
+ * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
+ * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
+ * dirent is allocated. Otherwise it must go through all the 'used' dirents
+ * searching for one in which the amount of total space minus the amount of
+ * used space will provide enough space for the new dirent.
+ *
+ * There are two types of blocks in which dirents reside. In a stuffed dinode,
+ * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
+ * the block.  In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
+ * beginning of the leaf block. The dirents reside in leaves when
+ *
+ * dip->i_di.di_flags & GFS2_DIF_EXHASH is true
+ *
+ * Otherwise, the dirents are "linear", within a single stuffed dinode block.
+ *
+ * When the dirents are in leaves, the actual contents of the directory file are
+ * used as an array of 64-bit block pointers pointing to the leaf blocks. The
+ * dirents are NOT in the directory file itself. There can be more than one
+ * block pointer in the array that points to the same leaf. In fact, when a
+ * directory is first converted from linear to exhash, all of the pointers
+ * point to the same leaf.
+ *
+ * When a leaf is completely full, the size of the hash table can be
+ * doubled unless it is already at the maximum size which is hard coded into
+ * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
+ * but never before the maximum hash table size has been reached.
+ */
+
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/buffer_head.h>
+#include <linux/sort.h>
+#include <linux/gfs2_ondisk.h>
+#include <linux/crc32.h>
+#include <linux/vmalloc.h>
+
+#include "gfs2.h"
+#include "lm_interface.h"
+#include "incore.h"
+#include "dir.h"
+#include "glock.h"
+#include "inode.h"
+#include "meta_io.h"
+#include "quota.h"
+#include "rgrp.h"
+#include "trans.h"
+#include "bmap.h"
+#include "util.h"
+
+#define IS_LEAF     1 /* Hashed (leaf) directory */
+#define IS_DINODE   2 /* Linear (stuffed dinode block) directory */
+
+#define gfs2_disk_hash2offset(h) (((uint64_t)(h)) >> 1)
+#define gfs2_dir_offset2hash(p) ((uint32_t)(((uint64_t)(p)) << 1))
+
+typedef int (*leaf_call_t) (struct gfs2_inode *dip,
+			    uint32_t index, uint32_t len, uint64_t leaf_no,
+			    void *data);
+
+
+int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, uint64_t block,
+			    struct buffer_head **bhp)
+{
+	struct buffer_head *bh;
+
+	bh = gfs2_meta_new(ip->i_gl, block);
+	gfs2_trans_add_bh(ip->i_gl, bh, 1);
+	gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
+	gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
+	*bhp = bh;
+	return 0;
+}
+
+static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, uint64_t block,
+					struct buffer_head **bhp)
+{
+	struct buffer_head *bh;
+	int error;
+
+	error = gfs2_meta_read(ip->i_gl, block, DIO_START | DIO_WAIT, &bh);
+	if (error)
+		return error;
+	if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
+		brelse(bh);
+		return -EIO;
+	}
+	*bhp = bh;
+	return 0;
+}
+
+static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
+				  unsigned int offset, unsigned int size)
+                               
+{
+	struct buffer_head *dibh;
+	int error;
+
+	error = gfs2_meta_inode_buffer(ip, &dibh);
+	if (error)
+		return error;
+
+	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
+	memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
+	if (ip->i_di.di_size < offset + size)
+		ip->i_di.di_size = offset + size;
+	ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
+	gfs2_dinode_out(&ip->i_di, dibh->b_data);
+
+	brelse(dibh);
+
+	return size;
+}
+
+
+
+/**
+ * gfs2_dir_write_data - Write directory information to the inode
+ * @ip: The GFS2 inode
+ * @buf: The buffer containing information to be written
+ * @offset: The file offset to start writing at
+ * @size: The amount of data to write
+ *
+ * Returns: The number of bytes correctly written or error code
+ */
+static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
+			       uint64_t offset, unsigned int size)
+{
+	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
+	struct buffer_head *dibh;
+	uint64_t lblock, dblock;
+	uint32_t extlen = 0;
+	unsigned int o;
+	int copied = 0;
+	int error = 0;
+
+	if (!size)
+		return 0;
+
+	if (gfs2_is_stuffed(ip) &&
+	    offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
+		return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
+					      size);
+
+	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
+		return -EINVAL;
+
+	if (gfs2_is_stuffed(ip)) {
+		error = gfs2_unstuff_dinode(ip, NULL);
+		if (error)
+			return error;
+	}
+
+	lblock = offset;
+	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
+
+	while (copied < size) {
+		unsigned int amount;
+		struct buffer_head *bh;
+		int new;
+
+		amount = size - copied;
+		if (amount > sdp->sd_sb.sb_bsize - o)
+			amount = sdp->sd_sb.sb_bsize - o;
+
+		if (!extlen) {
+			new = 1;
+			error = gfs2_extent_map(&ip->i_inode, lblock, &new,
+						&dblock, &extlen);
+			if (error)
+				goto fail;
+			error = -EIO;
+			if (gfs2_assert_withdraw(sdp, dblock))
+				goto fail;
+		}
+
+		if (amount == sdp->sd_jbsize || new)
+			error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
+		else
+			error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
+
+		if (error)
+			goto fail;
+
+		gfs2_trans_add_bh(ip->i_gl, bh, 1);
+		memcpy(bh->b_data + o, buf, amount);
+		brelse(bh);
+		if (error)
+			goto fail;
+
+		buf += amount;
+		copied += amount;
+		lblock++;
+		dblock++;
+		extlen--;
+
+		o = sizeof(struct gfs2_meta_header);
+	}
+
+out:
+	error = gfs2_meta_inode_buffer(ip, &dibh);
+	if (error)
+		return error;
+
+	if (ip->i_di.di_size < offset + copied)
+		ip->i_di.di_size = offset + copied;
+	ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
+
+	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
+	gfs2_dinode_out(&ip->i_di, dibh->b_data);
+	brelse(dibh);
+
+	return copied;
+fail:
+	if (copied)
+		goto out;
+	return error;
+}
+
+static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf,
+				 unsigned int offset, unsigned int size)
+{
+	struct buffer_head *dibh;
+	int error;
+
+	error = gfs2_meta_inode_buffer(ip, &dibh);
+	if (!error) {
+		offset += sizeof(struct gfs2_dinode);
+		memcpy(buf, dibh->b_data + offset, size);
+		brelse(dibh);
+	}
+
+	return (error) ? error : size;
+}
+
+
+/**
+ * gfs2_dir_read_data - Read a data from a directory inode
+ * @ip: The GFS2 Inode
+ * @buf: The buffer to place result into
+ * @offset: File offset to begin jdata_readng from
+ * @size: Amount of data to transfer
+ *
+ * Returns: The amount of data actually copied or the error
+ */
+static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf,
+			      uint64_t offset, unsigned int size)
+{
+	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
+	uint64_t lblock, dblock;
+	uint32_t extlen = 0;
+	unsigned int o;
+	int copied = 0;
+	int error = 0;
+
+	if (offset >= ip->i_di.di_size)
+		return 0;
+
+	if ((offset + size) > ip->i_di.di_size)
+		size = ip->i_di.di_size - offset;
+
+	if (!size)
+		return 0;
+
+	if (gfs2_is_stuffed(ip))
+		return gfs2_dir_read_stuffed(ip, buf, (unsigned int)offset,
+					     size);
+
+	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
+		return -EINVAL;
+
+	lblock = offset;
+	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
+
+	while (copied < size) {
+		unsigned int amount;
+		struct buffer_head *bh;
+		int new;
+
+		amount = size - copied;
+		if (amount > sdp->sd_sb.sb_bsize - o)
+			amount = sdp->sd_sb.sb_bsize - o;
+
+		if (!extlen) {
+			new = 0;
+			error = gfs2_extent_map(&ip->i_inode, lblock, &new,
+						&dblock, &extlen);
+			if (error)
+				goto fail;
+		}
+
+		if (extlen > 1)
+			gfs2_meta_ra(ip->i_gl, dblock, extlen);
+
+		if (dblock) {
+			if (new)
+				error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
+			else
+				error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
+			if (error)
+				goto fail;
+			dblock++;
+			extlen--;
+		} else
+			bh = NULL;
+
+		memcpy(buf, bh->b_data + o, amount);
+		brelse(bh);
+		if (error)
+			goto fail;
+
+		buf += amount;
+		copied += amount;
+		lblock++;
+
+		o = sizeof(struct gfs2_meta_header);
+	}
+
+	return copied;
+fail:
+	return (copied) ? copied : error;
+}
+
+typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
+			    const struct qstr *name,
+			    void *opaque);
+
+static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
+				     const struct qstr *name, int ret)
+{
+	if (dent->de_inum.no_addr != 0 &&
+	    be32_to_cpu(dent->de_hash) == name->hash &&
+	    be16_to_cpu(dent->de_name_len) == name->len &&
+	    memcmp((char *)(dent+1), name->name, name->len) == 0)
+		return ret;
+	return 0;
+}
+
+static int gfs2_dirent_find(const struct gfs2_dirent *dent,
+			    const struct qstr *name,
+			    void *opaque)
+{
+	return __gfs2_dirent_find(dent, name, 1);
+}
+
+static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
+			    const struct qstr *name,
+			    void *opaque)
+{
+	return __gfs2_dirent_find(dent, name, 2);
+}
+
+/*
+ * name->name holds ptr to start of block.
+ * name->len holds size of block.
+ */
+static int gfs2_dirent_last(const struct gfs2_dirent *dent,
+			    const struct qstr *name,
+			    void *opaque)
+{
+	const char *start = name->name;
+	const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
+	if (name->len == (end - start))
+		return 1;
+	return 0;
+}
+
+static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
+				  const struct qstr *name,
+				  void *opaque)
+{
+	unsigned required = GFS2_DIRENT_SIZE(name->len);
+	unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
+	unsigned totlen = be16_to_cpu(dent->de_rec_len);
+
+	if (!dent->de_inum.no_addr)
+		actual = GFS2_DIRENT_SIZE(0);
+	if ((totlen - actual) >= required)
+		return 1;
+	return 0;
+}
+
+struct dirent_gather {
+	const struct gfs2_dirent **pdent;
+	unsigned offset;
+};
+
+static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
+			      const struct qstr *name,
+			      void *opaque)
+{
+	struct dirent_gather *g = opaque;
+	if (dent->de_inum.no_addr) {
+		g->pdent[g->offset++] = dent;
+	}
+	return 0;
+}
+
+/*
+ * Other possible things to check:
+ * - Inode located within filesystem size (and on valid block)
+ * - Valid directory entry type
+ * Not sure how heavy-weight we want to make this... could also check
+ * hash is correct for example, but that would take a lot of extra time.
+ * For now the most important thing is to check that the various sizes
+ * are correct.
+ */
+static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
+			     unsigned int size, unsigned int len, int first)
+{
+	const char *msg = "gfs2_dirent too small";
+	if (unlikely(size < sizeof(struct gfs2_dirent)))
+		goto error;
+	msg = "gfs2_dirent misaligned";
+	if (unlikely(offset & 0x7))
+		goto error;
+	msg = "gfs2_dirent points beyond end of block";
+	if (unlikely(offset + size > len))
+		goto error;
+	msg = "zero inode number";
+	if (unlikely(!first && !dent->de_inum.no_addr))
+		goto error;
+	msg = "name length is greater than space in dirent";
+	if (dent->de_inum.no_addr &&
+	    unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
+		     size))
+		goto error;
+	return 0;
+error:
+	printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg,
+	       first ? "first in block" : "not first in block");
+	return -EIO;
+}
+
+static int gfs2_dirent_offset(const void *buf)
+{
+	const struct gfs2_meta_header *h = buf;
+	int offset;
+
+	BUG_ON(buf == NULL);
+
+	switch(be32_to_cpu(h->mh_type)) {
+	case GFS2_METATYPE_LF:
+		offset = sizeof(struct gfs2_leaf);
+		break;
+	case GFS2_METATYPE_DI:
+		offset = sizeof(struct gfs2_dinode);
+		break;
+	default:
+		goto wrong_type;
+	}
+	return offset;
+wrong_type:
+	printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n",
+	       be32_to_cpu(h->mh_type));
+	return -1;
+}
+
+static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode,
+					    void *buf,
+					    unsigned int len, gfs2_dscan_t scan,
+					    const struct qstr *name,
+					    void *opaque)
+{
+	struct gfs2_dirent *dent, *prev;
+	unsigned offset;
+	unsigned size;
+	int ret = 0;
+
+	ret = gfs2_dirent_offset(buf);
+	if (ret < 0)
+		goto consist_inode;
+
+	offset = ret;
+	prev = NULL;
+	dent = (struct gfs2_dirent *)(buf + offset);
+	size = be16_to_cpu(dent->de_rec_len);
+	if (gfs2_check_dirent(dent, offset, size, len, 1))
+		goto consist_inode;
+	do {
+		ret = scan(dent, name, opaque);
+		if (ret)
+			break;
+		offset += size;
+		if (offset == len)
+			break;
+		prev = dent;
+		dent = (struct gfs2_dirent *)(buf + offset);
+		size = be16_to_cpu(dent->de_rec_len);
+		if (gfs2_check_dirent(dent, offset, size, len, 0))
+			goto consist_inode;
+	} while(1);
+
+	switch(ret) {
+	case 0:
+		return NULL;
+	case 1:
+		return dent;
+	case 2:
+		return prev ? prev : dent;
+	default:
+		BUG_ON(ret > 0);
+		return ERR_PTR(ret);
+	}
+
+consist_inode:
+	gfs2_consist_inode(GFS2_I(inode));
+	return ERR_PTR(-EIO);
+}
+
+
+/**
+ * dirent_first - Return the first dirent
+ * @dip: the directory
+ * @bh: The buffer
+ * @dent: Pointer to list of dirents
+ *
+ * return first dirent whether bh points to leaf or stuffed dinode
+ *
+ * Returns: IS_LEAF, IS_DINODE, or -errno
+ */
+
+static int dirent_first(struct gfs2_inode *dip, struct buffer_head *bh,
+			struct gfs2_dirent **dent)
+{
+	struct gfs2_meta_header *h = (struct gfs2_meta_header *)bh->b_data;
+
+	if (be32_to_cpu(h->mh_type) == GFS2_METATYPE_LF) {
+		if (gfs2_meta_check(GFS2_SB(&dip->i_inode), bh))
+			return -EIO;
+		*dent = (struct gfs2_dirent *)(bh->b_data +
+					       sizeof(struct gfs2_leaf));
+		return IS_LEAF;
+	} else {
+		if (gfs2_metatype_check(GFS2_SB(&dip->i_inode), bh, GFS2_METATYPE_DI))
+			return -EIO;
+		*dent = (struct gfs2_dirent *)(bh->b_data +
+					       sizeof(struct gfs2_dinode));
+		return IS_DINODE;
+	}
+}
+
+/**
+ * dirent_next - Next dirent
+ * @dip: the directory
+ * @bh: The buffer
+ * @dent: Pointer to list of dirents
+ *
+ * Returns: 0 on success, error code otherwise
+ */
+
+static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
+		       struct gfs2_dirent **dent)
+{
+	struct gfs2_dirent *tmp, *cur;
+	char *bh_end;
+	uint16_t cur_rec_len;
+
+	cur = *dent;
+	bh_end = bh->b_data + bh->b_size;
+	cur_rec_len = be16_to_cpu(cur->de_rec_len);
+
+	if ((char *)cur + cur_rec_len >= bh_end) {
+		if ((char *)cur + cur_rec_len > bh_end) {
+			gfs2_consist_inode(dip);
+			return -EIO;
+		}
+		return -ENOENT;
+	}
+
+	tmp = (struct gfs2_dirent *)((char *)cur + cur_rec_len);
+
+	if ((char *)tmp + be16_to_cpu(tmp->de_rec_len) > bh_end) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+	if (cur_rec_len == 0) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+        /* Only the first dent could ever have de_inum.no_addr == 0 */
+	if (!tmp->de_inum.no_addr) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+	*dent = tmp;
+
+	return 0;
+}
+
+/**
+ * dirent_del - Delete a dirent
+ * @dip: The GFS2 inode
+ * @bh: The buffer
+ * @prev: The previous dirent
+ * @cur: The current dirent
+ *
+ */
+
+static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
+		       struct gfs2_dirent *prev, struct gfs2_dirent *cur)
+{
+	uint16_t cur_rec_len, prev_rec_len;
+
+	if (!cur->de_inum.no_addr) {
+		gfs2_consist_inode(dip);
+		return;
+	}
+
+	gfs2_trans_add_bh(dip->i_gl, bh, 1);
+
+	/* If there is no prev entry, this is the first entry in the block.
+	   The de_rec_len is already as big as it needs to be.  Just zero
+	   out the inode number and return.  */
+
+	if (!prev) {
+		cur->de_inum.no_addr = 0;	/* No endianess worries */
+		return;
+	}
+
+	/*  Combine this dentry with the previous one.  */
+
+	prev_rec_len = be16_to_cpu(prev->de_rec_len);
+	cur_rec_len = be16_to_cpu(cur->de_rec_len);
+
+	if ((char *)prev + prev_rec_len != (char *)cur)
+		gfs2_consist_inode(dip);
+	if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
+		gfs2_consist_inode(dip);
+
+	prev_rec_len += cur_rec_len;
+	prev->de_rec_len = cpu_to_be16(prev_rec_len);
+}
+
+/*
+ * Takes a dent from which to grab space as an argument. Returns the
+ * newly created dent.
+ */
+static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
+					    struct gfs2_dirent *dent,
+					    const struct qstr *name,
+					    struct buffer_head *bh)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_dirent *ndent;
+	unsigned offset = 0, totlen;
+
+	if (dent->de_inum.no_addr)
+		offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
+	totlen = be16_to_cpu(dent->de_rec_len);
+	BUG_ON(offset + name->len > totlen);
+	gfs2_trans_add_bh(ip->i_gl, bh, 1);
+	ndent = (struct gfs2_dirent *)((char *)dent + offset);
+	dent->de_rec_len = cpu_to_be16(offset);
+	gfs2_qstr2dirent(name, totlen - offset, ndent);
+	return ndent;
+}
+
+static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
+					     struct buffer_head *bh,
+					     const struct qstr *name)
+{
+	struct gfs2_dirent *dent;
+	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, 
+				gfs2_dirent_find_space, name, NULL);
+	if (!dent || IS_ERR(dent))
+		return dent;
+	return gfs2_init_dirent(inode, dent, name, bh);
+}
+
+static int get_leaf(struct gfs2_inode *dip, uint64_t leaf_no,
+		    struct buffer_head **bhp)
+{
+	int error;
+
+	error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_START | DIO_WAIT, bhp);
+	if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) {
+		/* printk(KERN_INFO "block num=%llu\n", leaf_no); */
+		error = -EIO;
+	}
+
+	return error;
+}
+
+/**
+ * get_leaf_nr - Get a leaf number associated with the index
+ * @dip: The GFS2 inode
+ * @index:
+ * @leaf_out:
+ *
+ * Returns: 0 on success, error code otherwise
+ */
+
+static int get_leaf_nr(struct gfs2_inode *dip, uint32_t index,
+		       uint64_t *leaf_out)
+{
+	uint64_t leaf_no;
+	int error;
+
+	error = gfs2_dir_read_data(dip, (char *)&leaf_no,
+				    index * sizeof(uint64_t),
+				    sizeof(uint64_t));
+	if (error != sizeof(uint64_t))
+		return (error < 0) ? error : -EIO;
+
+	*leaf_out = be64_to_cpu(leaf_no);
+
+	return 0;
+}
+
+static int get_first_leaf(struct gfs2_inode *dip, uint32_t index,
+			  struct buffer_head **bh_out)
+{
+	uint64_t leaf_no;
+	int error;
+
+	error = get_leaf_nr(dip, index, &leaf_no);
+	if (!error)
+		error = get_leaf(dip, leaf_no, bh_out);
+
+	return error;
+}
+
+static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
+					      const struct qstr *name,
+					      gfs2_dscan_t scan,
+					      struct buffer_head **pbh)
+{
+	struct buffer_head *bh;
+	struct gfs2_dirent *dent;
+	struct gfs2_inode *ip = GFS2_I(inode);
+	int error;
+
+	if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
+		struct gfs2_leaf *leaf;
+		unsigned hsize = 1 << ip->i_di.di_depth;
+		unsigned index;
+		u64 ln;
+		if (hsize * sizeof(u64) != ip->i_di.di_size) {
+			gfs2_consist_inode(ip);
+			return ERR_PTR(-EIO);
+		}
+		
+		index = name->hash >> (32 - ip->i_di.di_depth);
+		error = get_first_leaf(ip, index, &bh);
+		if (error)
+			return ERR_PTR(error);
+		do {
+			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
+						scan, name, NULL);
+			if (dent)
+				goto got_dent;
+			leaf = (struct gfs2_leaf *)bh->b_data;
+			ln = be64_to_cpu(leaf->lf_next);
+			brelse(bh);
+			if (!ln)
+				break;
+			
+			error = get_leaf(ip, ln, &bh);
+		} while(!error);
+
+		return error ? ERR_PTR(error) : NULL;
+	}
+
+	
+	error = gfs2_meta_inode_buffer(ip, &bh);
+	if (error)
+		return ERR_PTR(error);
+	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
+got_dent:
+	if (unlikely(dent == NULL || IS_ERR(dent))) {
+		brelse(bh);
+		bh = NULL;
+	}
+	*pbh = bh;
+	return dent;
+}
+
+static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	u64 bn = gfs2_alloc_meta(ip);
+	struct buffer_head *bh = gfs2_meta_new(ip->i_gl, bn);
+	struct gfs2_leaf *leaf;
+	struct gfs2_dirent *dent;
+	struct qstr name = { .name = "", .len = 0, .hash = 0 };
+	if (!bh)
+		return NULL;
+	
+	gfs2_trans_add_bh(ip->i_gl, bh, 1);
+	gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
+	leaf = (struct gfs2_leaf *)bh->b_data;
+	leaf->lf_depth = cpu_to_be16(depth);
+	leaf->lf_entries = cpu_to_be16(0);
+	leaf->lf_dirent_format = cpu_to_be16(GFS2_FORMAT_DE);
+	leaf->lf_next = cpu_to_be64(0);
+	memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved));
+	dent = (struct gfs2_dirent *)(leaf+1);
+	gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
+	*pbh = bh;
+	return leaf;
+}
+
+/**
+ * dir_make_exhash - Convert a stuffed directory into an ExHash directory
+ * @dip: The GFS2 inode
+ *
+ * Returns: 0 on success, error code otherwise
+ */
+
+static int dir_make_exhash(struct inode *inode)
+{
+	struct gfs2_inode *dip = GFS2_I(inode);
+	struct gfs2_sbd *sdp = GFS2_SB(inode);
+	struct gfs2_dirent *dent;
+	struct qstr args;
+	struct buffer_head *bh, *dibh;
+	struct gfs2_leaf *leaf;
+	int y;
+	uint32_t x;
+	uint64_t *lp, bn;
+	int error;
+
+	error = gfs2_meta_inode_buffer(dip, &dibh);
+	if (error)
+		return error;
+
+	/*  Turn over a new leaf  */
+
+	leaf = new_leaf(inode, &bh, 0);
+	if (!leaf)
+		return -ENOSPC;
+	bn = bh->b_blocknr;
+
+	gfs2_assert(sdp, dip->i_di.di_entries < (1 << 16));
+	leaf->lf_entries = cpu_to_be16(dip->i_di.di_entries);
+
+	/*  Copy dirents  */
+
+	gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
+			     sizeof(struct gfs2_dinode));
+
+	/*  Find last entry  */
+
+	x = 0;
+	args.len = bh->b_size - sizeof(struct gfs2_dinode) +
+		   sizeof(struct gfs2_leaf);
+	args.name = bh->b_data;
+	dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
+				gfs2_dirent_last, &args, NULL);
+	if (!dent) {
+		brelse(bh);
+		brelse(dibh);
+		return -EIO;
+	}
+	if (IS_ERR(dent)) {
+		brelse(bh);
+		brelse(dibh);
+		return PTR_ERR(dent);
+	}
+
+	/*  Adjust the last dirent's record length
+	   (Remember that dent still points to the last entry.)  */
+
+	dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
+		sizeof(struct gfs2_dinode) -
+		sizeof(struct gfs2_leaf));
+
+	brelse(bh);
+
+	/*  We're done with the new leaf block, now setup the new
+	    hash table.  */
+
+	gfs2_trans_add_bh(dip->i_gl, dibh, 1);
+	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
+
+	lp = (uint64_t *)(dibh->b_data + sizeof(struct gfs2_dinode));
+
+	for (x = sdp->sd_hash_ptrs; x--; lp++)
+		*lp = cpu_to_be64(bn);
+
+	dip->i_di.di_size = sdp->sd_sb.sb_bsize / 2;
+	dip->i_di.di_blocks++;
+	dip->i_di.di_flags |= GFS2_DIF_EXHASH;
+	dip->i_di.di_payload_format = 0;
+
+	for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
+	dip->i_di.di_depth = y;
+
+	gfs2_dinode_out(&dip->i_di, dibh->b_data);
+
+	brelse(dibh);
+
+	return 0;
+}
+
+/**
+ * dir_split_leaf - Split a leaf block into two
+ * @dip: The GFS2 inode
+ * @index:
+ * @leaf_no:
+ *
+ * Returns: 0 on success, error code on failure
+ */
+
+static int dir_split_leaf(struct inode *inode, const struct qstr *name)
+{
+	struct gfs2_inode *dip = GFS2_I(inode);
+	struct buffer_head *nbh, *obh, *dibh;
+	struct gfs2_leaf *nleaf, *oleaf;
+	struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
+	uint32_t start, len, half_len, divider;
+	uint64_t bn, *lp, leaf_no;
+	uint32_t index;
+	int x, moved = 0;
+	int error;
+
+	index = name->hash >> (32 - dip->i_di.di_depth);
+	error = get_leaf_nr(dip, index, &leaf_no);
+	if (error)
+		return error;
+
+	/*  Get the old leaf block  */
+	error = get_leaf(dip, leaf_no, &obh);
+	if (error)
+		return error;
+
+	oleaf = (struct gfs2_leaf *)obh->b_data;
+	if (dip->i_di.di_depth == be16_to_cpu(oleaf->lf_depth)) {
+		brelse(obh);
+		return 1; /* can't split */
+	}
+
+	gfs2_trans_add_bh(dip->i_gl, obh, 1);
+
+	nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
+	if (!nleaf) {
+		brelse(obh);
+		return -ENOSPC;
+	}
+	bn = nbh->b_blocknr;
+
+	/*  Compute the start and len of leaf pointers in the hash table.  */
+	len = 1 << (dip->i_di.di_depth - be16_to_cpu(oleaf->lf_depth));
+	half_len = len >> 1;
+	if (!half_len) {
+		printk(KERN_WARNING "di_depth %u lf_depth %u index %u\n", dip->i_di.di_depth, be16_to_cpu(oleaf->lf_depth), index);
+		gfs2_consist_inode(dip);
+		error = -EIO;
+		goto fail_brelse;
+	}
+
+	start = (index & ~(len - 1));
+
+	/* Change the pointers.
+	   Don't bother distinguishing stuffed from non-stuffed.
+	   This code is complicated enough already. */
+	lp = kmalloc(half_len * sizeof(uint64_t), GFP_NOFS | __GFP_NOFAIL);
+	/*  Change the pointers  */
+	for (x = 0; x < half_len; x++)
+		lp[x] = cpu_to_be64(bn);
+
+	error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(uint64_t),
+				    half_len * sizeof(uint64_t));
+	if (error != half_len * sizeof(uint64_t)) {
+		if (error >= 0)
+			error = -EIO;
+		goto fail_lpfree;
+	}
+
+	kfree(lp);
+
+	/*  Compute the divider  */
+	divider = (start + half_len) << (32 - dip->i_di.di_depth);
+
+	/*  Copy the entries  */
+	dirent_first(dip, obh, &dent);
+
+	do {
+		next = dent;
+		if (dirent_next(dip, obh, &next))
+			next = NULL;
+
+		if (dent->de_inum.no_addr &&
+		    be32_to_cpu(dent->de_hash) < divider) {
+			struct qstr str;
+			str.name = (char*)(dent+1);
+			str.len = be16_to_cpu(dent->de_name_len);
+			str.hash = be32_to_cpu(dent->de_hash);
+			new = gfs2_dirent_alloc(inode, nbh, &str);
+			if (IS_ERR(new)) {
+				error = PTR_ERR(new);
+				break;
+			}
+
+			new->de_inum = dent->de_inum; /* No endian worries */
+			new->de_type = dent->de_type; /* No endian worries */
+			nleaf->lf_entries = cpu_to_be16(be16_to_cpu(nleaf->lf_entries)+1);
+
+			dirent_del(dip, obh, prev, dent);
+
+			if (!oleaf->lf_entries)
+				gfs2_consist_inode(dip);
+			oleaf->lf_entries = cpu_to_be16(be16_to_cpu(oleaf->lf_entries)-1);
+
+			if (!prev)
+				prev = dent;
+
+			moved = 1;
+		} else {
+			prev = dent;
+		}
+		dent = next;
+	} while (dent);
+
+	oleaf->lf_depth = nleaf->lf_depth;
+
+	error = gfs2_meta_inode_buffer(dip, &dibh);
+	if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
+		dip->i_di.di_blocks++;
+		gfs2_dinode_out(&dip->i_di, dibh->b_data);
+		brelse(dibh);
+	}
+
+	brelse(obh);
+	brelse(nbh);
+
+	return error;
+
+fail_lpfree:
+	kfree(lp);
+
+fail_brelse:
+	brelse(obh);
+	brelse(nbh);
+	return error;
+}
+
+/**
+ * dir_double_exhash - Double size of ExHash table
+ * @dip: The GFS2 dinode
+ *
+ * Returns: 0 on success, error code on failure
+ */
+
+static int dir_double_exhash(struct gfs2_inode *dip)
+{
+	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
+	struct buffer_head *dibh;
+	uint32_t hsize;
+	uint64_t *buf;
+	uint64_t *from, *to;
+	uint64_t block;
+	int x;
+	int error = 0;
+
+	hsize = 1 << dip->i_di.di_depth;
+	if (hsize * sizeof(uint64_t) != dip->i_di.di_size) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+	/*  Allocate both the "from" and "to" buffers in one big chunk  */
+
+	buf = kcalloc(3, sdp->sd_hash_bsize, GFP_KERNEL | __GFP_NOFAIL);
+
+	for (block = dip->i_di.di_size >> sdp->sd_hash_bsize_shift; block--;) {
+		error = gfs2_dir_read_data(dip, (char *)buf,
+					    block * sdp->sd_hash_bsize,
+					    sdp->sd_hash_bsize);
+		if (error != sdp->sd_hash_bsize) {
+			if (error >= 0)
+				error = -EIO;
+			goto fail;
+		}
+
+		from = buf;
+		to = (uint64_t *)((char *)buf + sdp->sd_hash_bsize);
+
+		for (x = sdp->sd_hash_ptrs; x--; from++) {
+			*to++ = *from;	/*  No endianess worries  */
+			*to++ = *from;
+		}
+
+		error = gfs2_dir_write_data(dip,
+					     (char *)buf + sdp->sd_hash_bsize,
+					     block * sdp->sd_sb.sb_bsize,
+					     sdp->sd_sb.sb_bsize);
+		if (error != sdp->sd_sb.sb_bsize) {
+			if (error >= 0)
+				error = -EIO;
+			goto fail;
+		}
+	}
+
+	kfree(buf);
+
+	error = gfs2_meta_inode_buffer(dip, &dibh);
+	if (!gfs2_assert_withdraw(sdp, !error)) {
+		dip->i_di.di_depth++;
+		gfs2_dinode_out(&dip->i_di, dibh->b_data);
+		brelse(dibh);
+	}
+
+	return error;
+
+ fail:
+	kfree(buf);
+
+	return error;
+}
+
+/**
+ * compare_dents - compare directory entries by hash value
+ * @a: first dent
+ * @b: second dent
+ *
+ * When comparing the hash entries of @a to @b:
+ *   gt: returns 1
+ *   lt: returns -1
+ *   eq: returns 0
+ */
+
+static int compare_dents(const void *a, const void *b)
+{
+	struct gfs2_dirent *dent_a, *dent_b;
+	uint32_t hash_a, hash_b;
+	int ret = 0;
+
+	dent_a = *(struct gfs2_dirent **)a;
+	hash_a = be32_to_cpu(dent_a->de_hash);
+
+	dent_b = *(struct gfs2_dirent **)b;
+	hash_b = be32_to_cpu(dent_b->de_hash);
+
+	if (hash_a > hash_b)
+		ret = 1;
+	else if (hash_a < hash_b)
+		ret = -1;
+	else {
+		unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
+		unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
+
+		if (len_a > len_b)
+			ret = 1;
+		else if (len_a < len_b)
+			ret = -1;
+		else
+			ret = memcmp((char *)(dent_a + 1),
+				     (char *)(dent_b + 1),
+				     len_a);
+	}
+
+	return ret;
+}
+
+/**
+ * do_filldir_main - read out directory entries
+ * @dip: The GFS2 inode
+ * @offset: The offset in the file to read from
+ * @opaque: opaque data to pass to filldir
+ * @filldir: The function to pass entries to
+ * @darr: an array of struct gfs2_dirent pointers to read
+ * @entries: the number of entries in darr
+ * @copied: pointer to int that's non-zero if a entry has been copied out
+ *
+ * Jump through some hoops to make sure that if there are hash collsions,
+ * they are read out at the beginning of a buffer.  We want to minimize
+ * the possibility that they will fall into different readdir buffers or
+ * that someone will want to seek to that location.
+ *
+ * Returns: errno, >0 on exception from filldir
+ */
+
+static int do_filldir_main(struct gfs2_inode *dip, uint64_t *offset,
+			   void *opaque, gfs2_filldir_t filldir,
+			   const struct gfs2_dirent **darr, uint32_t entries,
+			   int *copied)
+{
+	const struct gfs2_dirent *dent, *dent_next;
+	struct gfs2_inum inum;
+	uint64_t off, off_next;
+	unsigned int x, y;
+	int run = 0;
+	int error = 0;
+
+	sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
+
+	dent_next = darr[0];
+	off_next = be32_to_cpu(dent_next->de_hash);
+	off_next = gfs2_disk_hash2offset(off_next);
+
+	for (x = 0, y = 1; x < entries; x++, y++) {
+		dent = dent_next;
+		off = off_next;
+
+		if (y < entries) {
+			dent_next = darr[y];
+			off_next = be32_to_cpu(dent_next->de_hash);
+			off_next = gfs2_disk_hash2offset(off_next);
+
+			if (off < *offset)
+				continue;
+			*offset = off;
+
+			if (off_next == off) {
+				if (*copied && !run)
+					return 1;
+				run = 1;
+			} else
+				run = 0;
+		} else {
+			if (off < *offset)
+				continue;
+			*offset = off;
+		}
+
+		gfs2_inum_in(&inum, (char *)&dent->de_inum);
+
+		error = filldir(opaque, (char *)(dent + 1),
+				be16_to_cpu(dent->de_name_len),
+				off, &inum,
+				be16_to_cpu(dent->de_type));
+		if (error)
+			return 1;
+
+		*copied = 1;
+	}
+
+	/* Increment the *offset by one, so the next time we come into the
+	   do_filldir fxn, we get the next entry instead of the last one in the
+	   current leaf */
+
+	(*offset)++;
+
+	return 0;
+}
+
+static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque,
+			      gfs2_filldir_t filldir, int *copied,
+			      unsigned *depth, u64 leaf_no)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct buffer_head *bh;
+	struct gfs2_leaf *lf;
+	unsigned entries = 0;
+	unsigned leaves = 0;
+	const struct gfs2_dirent **darr, *dent;
+	struct dirent_gather g;
+	struct buffer_head **larr;
+	int leaf = 0;
+	int error, i;
+	u64 lfn = leaf_no;
+
+	do {
+		error = get_leaf(ip, lfn, &bh);
+		if (error)
+			goto out;
+		lf = (struct gfs2_leaf *)bh->b_data;
+		if (leaves == 0)
+			*depth = be16_to_cpu(lf->lf_depth);
+		entries += be16_to_cpu(lf->lf_entries);
+		leaves++;
+		lfn = be64_to_cpu(lf->lf_next);
+		brelse(bh);
+	} while(lfn);
+
+	if (!entries)
+		return 0;
+
+	error = -ENOMEM;
+	larr = vmalloc((leaves + entries) * sizeof(void*));
+	if (!larr)
+		goto out;
+	darr = (const struct gfs2_dirent **)(larr + leaves);
+	g.pdent = darr;
+	g.offset = 0;
+	lfn = leaf_no;
+
+	do {
+		error = get_leaf(ip, lfn, &bh);
+		if (error)
+			goto out_kfree;
+		lf = (struct gfs2_leaf *)bh->b_data;
+		lfn = be64_to_cpu(lf->lf_next);
+		if (lf->lf_entries) {
+			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
+						gfs2_dirent_gather, NULL, &g);
+			error = PTR_ERR(dent);
+			if (IS_ERR(dent)) {
+				goto out_kfree;
+			}
+			error = 0;
+			larr[leaf++] = bh;
+		} else {
+			brelse(bh);
+		}
+	} while(lfn);
+
+	error = do_filldir_main(ip, offset, opaque, filldir, darr,
+				entries, copied);
+out_kfree:
+	for(i = 0; i < leaf; i++)
+		brelse(larr[i]);
+	vfree(larr);
+out:
+	return error;
+}
+
+/**
+ * dir_e_read - Reads the entries from a directory into a filldir buffer
+ * @dip: dinode pointer
+ * @offset: the hash of the last entry read shifted to the right once
+ * @opaque: buffer for the filldir function to fill
+ * @filldir: points to the filldir function to use
+ *
+ * Returns: errno
+ */
+
+static int dir_e_read(struct inode *inode, uint64_t *offset, void *opaque,
+		      gfs2_filldir_t filldir)
+{
+	struct gfs2_inode *dip = GFS2_I(inode);
+	struct gfs2_sbd *sdp = GFS2_SB(inode);
+	uint32_t hsize, len = 0;
+	uint32_t ht_offset, lp_offset, ht_offset_cur = -1;
+	uint32_t hash, index;
+	uint64_t *lp;
+	int copied = 0;
+	int error = 0;
+	unsigned depth = 0;
+
+	hsize = 1 << dip->i_di.di_depth;
+	if (hsize * sizeof(uint64_t) != dip->i_di.di_size) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+	hash = gfs2_dir_offset2hash(*offset);
+	index = hash >> (32 - dip->i_di.di_depth);
+
+	lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
+	if (!lp)
+		return -ENOMEM;
+
+	while (index < hsize) {
+		lp_offset = index & (sdp->sd_hash_ptrs - 1);
+		ht_offset = index - lp_offset;
+
+		if (ht_offset_cur != ht_offset) {
+			error = gfs2_dir_read_data(dip, (char *)lp,
+						ht_offset * sizeof(uint64_t),
+						sdp->sd_hash_bsize);
+			if (error != sdp->sd_hash_bsize) {
+				if (error >= 0)
+					error = -EIO;
+				goto out;
+			}
+			ht_offset_cur = ht_offset;
+		}
+
+		error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
+					   &copied, &depth,
+					   be64_to_cpu(lp[lp_offset]));
+		if (error)
+			break;
+
+		len = 1 << (dip->i_di.di_depth - depth);
+		index = (index & ~(len - 1)) + len;
+	}
+
+out:
+	kfree(lp);
+	if (error > 0)
+		error = 0;
+	return error;
+}
+
+int gfs2_dir_read(struct inode *inode, uint64_t *offset, void *opaque,
+		  gfs2_filldir_t filldir)
+{
+	struct gfs2_inode *dip = GFS2_I(inode);
+	struct dirent_gather g;
+	const struct gfs2_dirent **darr, *dent;
+	struct buffer_head *dibh;
+	int copied = 0;
+	int error;
+
+	if (!dip->i_di.di_entries)
+		return 0;
+
+	if (dip->i_di.di_flags & GFS2_DIF_EXHASH)
+		return dir_e_read(inode, offset, opaque, filldir);
+
+	if (!gfs2_is_stuffed(dip)) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+	error = gfs2_meta_inode_buffer(dip, &dibh);
+	if (error)
+		return error;
+
+	error = -ENOMEM;
+	darr = kmalloc(dip->i_di.di_entries * sizeof(struct gfs2_dirent *),
+		       GFP_KERNEL);
+	if (darr) {
+		g.pdent = darr;
+		g.offset = 0;
+		dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
+					gfs2_dirent_gather, NULL, &g);
+		if (IS_ERR(dent)) {
+			error = PTR_ERR(dent);
+			goto out;
+		}
+		error = do_filldir_main(dip, offset, opaque, filldir, darr,
+					dip->i_di.di_entries, &copied);
+out:
+		kfree(darr);
+	}
+
+	if (error > 0)
+		error = 0;
+
+	brelse(dibh);
+
+	return error;
+}
+
+/**
+ * gfs2_dir_search - Search a directory
+ * @dip: The GFS2 inode
+ * @filename:
+ * @inode:
+ *
+ * This routine searches a directory for a file or another directory.
+ * Assumes a glock is held on dip.
+ *
+ * Returns: errno
+ */
+
+int gfs2_dir_search(struct inode *dir, const struct qstr *name,
+		    struct gfs2_inum *inum, unsigned int *type)
+{
+	struct buffer_head *bh;
+	struct gfs2_dirent *dent;
+
+	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
+	if (dent) {
+		if (IS_ERR(dent))
+			return PTR_ERR(dent);
+		if (inum)
+			gfs2_inum_in(inum, (char *)&dent->de_inum);
+		if (type)
+			*type = be16_to_cpu(dent->de_type);
+		brelse(bh);
+		return 0;
+	}
+	return -ENOENT;
+}
+
+static int dir_new_leaf(struct inode *inode, const struct qstr *name)
+{
+	struct buffer_head *bh, *obh;
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_leaf *leaf, *oleaf;
+	int error;
+	u32 index;
+	u64 bn;
+
+	index = name->hash >> (32 - ip->i_di.di_depth);
+	error = get_first_leaf(ip, index, &obh);
+	if (error)
+		return error;
+	do {
+		oleaf = (struct gfs2_leaf *)obh->b_data;
+		bn = be64_to_cpu(oleaf->lf_next);
+		if (!bn)
+			break;
+		brelse(obh);
+		error = get_leaf(ip, bn, &obh);
+		if (error)
+			return error;
+	} while(1);
+
+	gfs2_trans_add_bh(ip->i_gl, obh, 1);
+
+	leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
+	if (!leaf) {
+		brelse(obh);
+		return -ENOSPC;
+	}
+	oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
+	brelse(bh);
+	brelse(obh);
+
+	error = gfs2_meta_inode_buffer(ip, &bh);
+	if (error)
+		return error;
+	gfs2_trans_add_bh(ip->i_gl, bh, 1);
+	ip->i_di.di_blocks++;
+	gfs2_dinode_out(&ip->i_di, bh->b_data);
+	brelse(bh);
+	return 0;
+}
+
+/**
+ * gfs2_dir_add - Add new filename into directory
+ * @dip: The GFS2 inode
+ * @filename: The new name
+ * @inode: The inode number of the entry
+ * @type: The type of the entry
+ *
+ * Returns: 0 on success, error code on failure
+ */
+
+int gfs2_dir_add(struct inode *inode, const struct qstr *name,
+		 const struct gfs2_inum *inum, unsigned type)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct buffer_head *bh;
+	struct gfs2_dirent *dent;
+	struct gfs2_leaf *leaf;
+	int error;
+
+	while(1) {
+		dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space,
+					  &bh);
+		if (dent) {
+			if (IS_ERR(dent))
+				return PTR_ERR(dent);
+			dent = gfs2_init_dirent(inode, dent, name, bh);
+			gfs2_inum_out(inum, (char *)&dent->de_inum);
+			dent->de_type = cpu_to_be16(type);
+			if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
+				leaf = (struct gfs2_leaf *)bh->b_data;
+				leaf->lf_entries = cpu_to_be16(be16_to_cpu(leaf->lf_entries) + 1);
+			}
+			brelse(bh);
+			error = gfs2_meta_inode_buffer(ip, &bh);
+			if (error)
+				break;
+			gfs2_trans_add_bh(ip->i_gl, bh, 1);
+			ip->i_di.di_entries++;
+			ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
+			gfs2_dinode_out(&ip->i_di, bh->b_data);
+			brelse(bh);
+			error = 0;
+			break;
+		}
+		if (!(ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
+			error = dir_make_exhash(inode);
+			if (error)
+				break;
+			continue;
+		}
+		error = dir_split_leaf(inode, name);
+		if (error == 0)
+			continue;
+		if (error < 0)
+			break;
+		if (ip->i_di.di_depth < GFS2_DIR_MAX_DEPTH) {
+			error = dir_double_exhash(ip);
+			if (error)
+				break;
+			error = dir_split_leaf(inode, name);
+			if (error < 0)
+				break;
+			if (error == 0)
+				continue;
+		}
+		error = dir_new_leaf(inode, name);
+		if (!error)
+			continue;
+		error = -ENOSPC;
+		break;
+	}
+	return error;
+}
+
+
+/**
+ * gfs2_dir_del - Delete a directory entry
+ * @dip: The GFS2 inode
+ * @filename: The filename
+ *
+ * Returns: 0 on success, error code on failure
+ */
+
+int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name)
+{
+	struct gfs2_dirent *dent, *prev = NULL;
+	struct buffer_head *bh;
+	int error;
+
+	/* Returns _either_ the entry (if its first in block) or the
+	   previous entry otherwise */
+	dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
+	if (!dent) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+	if (IS_ERR(dent)) {
+		gfs2_consist_inode(dip);
+		return PTR_ERR(dent);
+	}
+	/* If not first in block, adjust pointers accordingly */
+	if (gfs2_dirent_find(dent, name, NULL) == 0) {
+		prev = dent;
+		dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
+	}
+
+	dirent_del(dip, bh, prev, dent);
+	if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
+		struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
+		u16 entries = be16_to_cpu(leaf->lf_entries);
+		if (!entries)
+			gfs2_consist_inode(dip);
+		leaf->lf_entries = cpu_to_be16(--entries);
+	}
+	brelse(bh);
+
+	error = gfs2_meta_inode_buffer(dip, &bh);
+	if (error)
+		return error;
+
+	if (!dip->i_di.di_entries)
+		gfs2_consist_inode(dip);
+	gfs2_trans_add_bh(dip->i_gl, bh, 1);
+	dip->i_di.di_entries--;
+	dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
+	gfs2_dinode_out(&dip->i_di, bh->b_data);
+	brelse(bh);
+	mark_inode_dirty(&dip->i_inode);
+
+	return error;
+}
+
+/**
+ * gfs2_dir_mvino - Change inode number of directory entry
+ * @dip: The GFS2 inode
+ * @filename:
+ * @new_inode:
+ *
+ * This routine changes the inode number of a directory entry.  It's used
+ * by rename to change ".." when a directory is moved.
+ * Assumes a glock is held on dvp.
+ *
+ * Returns: errno
+ */
+
+int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
+		   struct gfs2_inum *inum, unsigned int new_type)
+{
+	struct buffer_head *bh;
+	struct gfs2_dirent *dent;
+	int error;
+
+	dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
+	if (!dent) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+	if (IS_ERR(dent))
+		return PTR_ERR(dent);
+
+	gfs2_trans_add_bh(dip->i_gl, bh, 1);
+	gfs2_inum_out(inum, (char *)&dent->de_inum);
+	dent->de_type = cpu_to_be16(new_type);
+
+	if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
+		brelse(bh);
+		error = gfs2_meta_inode_buffer(dip, &bh);
+		if (error)
+			return error;
+		gfs2_trans_add_bh(dip->i_gl, bh, 1);
+	}
+
+	dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
+	gfs2_dinode_out(&dip->i_di, bh->b_data);
+	brelse(bh);
+	return 0;
+}
+
+/**
+ * foreach_leaf - call a function for each leaf in a directory
+ * @dip: the directory
+ * @lc: the function to call for each each
+ * @data: private data to pass to it
+ *
+ * Returns: errno
+ */
+
+static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data)
+{
+	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
+	struct buffer_head *bh;
+	struct gfs2_leaf *leaf;
+	uint32_t hsize, len;
+	uint32_t ht_offset, lp_offset, ht_offset_cur = -1;
+	uint32_t index = 0;
+	uint64_t *lp;
+	uint64_t leaf_no;
+	int error = 0;
+
+	hsize = 1 << dip->i_di.di_depth;
+	if (hsize * sizeof(uint64_t) != dip->i_di.di_size) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+	lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
+	if (!lp)
+		return -ENOMEM;
+
+	while (index < hsize) {
+		lp_offset = index & (sdp->sd_hash_ptrs - 1);
+		ht_offset = index - lp_offset;
+
+		if (ht_offset_cur != ht_offset) {
+			error = gfs2_dir_read_data(dip, (char *)lp,
+						ht_offset * sizeof(uint64_t),
+						sdp->sd_hash_bsize);
+			if (error != sdp->sd_hash_bsize) {
+				if (error >= 0)
+					error = -EIO;
+				goto out;
+			}
+			ht_offset_cur = ht_offset;
+		}
+
+		leaf_no = be64_to_cpu(lp[lp_offset]);
+		if (leaf_no) {
+			error = get_leaf(dip, leaf_no, &bh);
+			if (error)
+				goto out;
+			leaf = (struct gfs2_leaf *)bh->b_data;
+			len = 1 << (dip->i_di.di_depth - be16_to_cpu(leaf->lf_depth));
+			brelse(bh);
+
+			error = lc(dip, index, len, leaf_no, data);
+			if (error)
+				goto out;
+
+			index = (index & ~(len - 1)) + len;
+		} else
+			index++;
+	}
+
+	if (index != hsize) {
+		gfs2_consist_inode(dip);
+		error = -EIO;
+	}
+
+out:
+	kfree(lp);
+
+	return error;
+}
+
+/**
+ * leaf_dealloc - Deallocate a directory leaf
+ * @dip: the directory
+ * @index: the hash table offset in the directory
+ * @len: the number of pointers to this leaf
+ * @leaf_no: the leaf number
+ * @data: not used
+ *
+ * Returns: errno
+ */
+
+static int leaf_dealloc(struct gfs2_inode *dip, uint32_t index, uint32_t len,
+			uint64_t leaf_no, void *data)
+{
+	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
+	struct gfs2_leaf *tmp_leaf;
+	struct gfs2_rgrp_list rlist;
+	struct buffer_head *bh, *dibh;
+	uint64_t blk, nblk;
+	unsigned int rg_blocks = 0, l_blocks = 0;
+	char *ht;
+	unsigned int x, size = len * sizeof(uint64_t);
+	int error;
+
+	memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
+
+	ht = kzalloc(size, GFP_KERNEL);
+	if (!ht)
+		return -ENOMEM;
+
+	gfs2_alloc_get(dip);
+
+	error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
+	if (error)
+		goto out;
+
+	error = gfs2_rindex_hold(sdp, &dip->i_alloc.al_ri_gh);
+	if (error)
+		goto out_qs;
+
+	/*  Count the number of leaves  */
+
+	for (blk = leaf_no; blk; blk = nblk) {
+		error = get_leaf(dip, blk, &bh);
+		if (error)
+			goto out_rlist;
+		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
+		nblk = be64_to_cpu(tmp_leaf->lf_next);
+		brelse(bh);
+
+		gfs2_rlist_add(sdp, &rlist, blk);
+		l_blocks++;
+	}
+
+	gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, 0);
+
+	for (x = 0; x < rlist.rl_rgrps; x++) {
+		struct gfs2_rgrpd *rgd;
+		rgd = rlist.rl_ghs[x].gh_gl->gl_object;
+		rg_blocks += rgd->rd_ri.ri_length;
+	}
+
+	error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
+	if (error)
+		goto out_rlist;
+
+	error = gfs2_trans_begin(sdp,
+			rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
+			RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
+	if (error)
+		goto out_rg_gunlock;
+
+	for (blk = leaf_no; blk; blk = nblk) {
+		error = get_leaf(dip, blk, &bh);
+		if (error)
+			goto out_end_trans;
+		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
+		nblk = be64_to_cpu(tmp_leaf->lf_next);
+		brelse(bh);
+
+		gfs2_free_meta(dip, blk, 1);
+
+		if (!dip->i_di.di_blocks)
+			gfs2_consist_inode(dip);
+		dip->i_di.di_blocks--;
+	}
+
+	error = gfs2_dir_write_data(dip, ht, index * sizeof(uint64_t), size);
+	if (error != size) {
+		if (error >= 0)
+			error = -EIO;
+		goto out_end_trans;
+	}
+
+	error = gfs2_meta_inode_buffer(dip, &dibh);
+	if (error)
+		goto out_end_trans;
+
+	gfs2_trans_add_bh(dip->i_gl, dibh, 1);
+	gfs2_dinode_out(&dip->i_di, dibh->b_data);
+	brelse(dibh);
+
+ out_end_trans:
+	gfs2_trans_end(sdp);
+
+ out_rg_gunlock:
+	gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
+
+ out_rlist:
+	gfs2_rlist_free(&rlist);
+	gfs2_glock_dq_uninit(&dip->i_alloc.al_ri_gh);
+
+ out_qs:
+	gfs2_quota_unhold(dip);
+
+ out:
+	gfs2_alloc_put(dip);
+	kfree(ht);
+
+	return error;
+}
+
+/**
+ * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
+ * @dip: the directory
+ *
+ * Dealloc all on-disk directory leaves to FREEMETA state
+ * Change on-disk inode type to "regular file"
+ *
+ * Returns: errno
+ */
+
+int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
+{
+	struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
+	struct buffer_head *bh;
+	int error;
+
+	/* Dealloc on-disk leaves to FREEMETA state */
+	error = foreach_leaf(dip, leaf_dealloc, NULL);
+	if (error)
+		return error;
+
+	/* Make this a regular file in case we crash.
+	   (We don't want to free these blocks a second time.)  */
+
+	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
+	if (error)
+		return error;
+
+	error = gfs2_meta_inode_buffer(dip, &bh);
+	if (!error) {
+		gfs2_trans_add_bh(dip->i_gl, bh, 1);
+		((struct gfs2_dinode *)bh->b_data)->di_mode =
+						cpu_to_be32(S_IFREG);
+		brelse(bh);
+	}
+
+	gfs2_trans_end(sdp);
+
+	return error;
+}
+
+/**
+ * gfs2_diradd_alloc_required - find if adding entry will require an allocation
+ * @ip: the file being written to
+ * @filname: the filename that's going to be added
+ *
+ * Returns: 1 if alloc required, 0 if not, -ve on error
+ */
+
+int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name)
+{
+	struct gfs2_dirent *dent;
+	struct buffer_head *bh;
+
+	dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
+	if (!dent) {
+		return 1;
+	}
+	if (IS_ERR(dent))
+		return PTR_ERR(dent);
+	brelse(bh);
+	return 0;
+}
+
--- /dev/null
+++ b/fs/gfs2/dir.h
@@ -0,0 +1,73 @@
+/*
+ * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
+ * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ */
+
+#ifndef __DIR_DOT_H__
+#define __DIR_DOT_H__
+
+/**
+ * gfs2_filldir_t - Report a directory entry to the caller of gfs2_dir_read()
+ * @opaque: opaque data used by the function
+ * @name: the name of the directory entry
+ * @length: the length of the name
+ * @offset: the entry's offset in the directory
+ * @inum: the inode number the entry points to
+ * @type: the type of inode the entry points to
+ *
+ * Returns: 0 on success, 1 if buffer full
+ */
+
+typedef int (*gfs2_filldir_t) (void *opaque,
+			      const char *name, unsigned int length,
+			      uint64_t offset,
+			      struct gfs2_inum *inum, unsigned int type);
+
+int gfs2_dir_search(struct inode *dir, const struct qstr *filename,
+		    struct gfs2_inum *inum, unsigned int *type);
+int gfs2_dir_add(struct inode *inode, const struct qstr *filename,
+		 const struct gfs2_inum *inum, unsigned int type);
+int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *filename);
+int gfs2_dir_read(struct inode *inode, uint64_t * offset, void *opaque,
+		  gfs2_filldir_t filldir);
+int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
+		   struct gfs2_inum *new_inum, unsigned int new_type);
+
+int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip);
+
+int gfs2_diradd_alloc_required(struct inode *dir,
+			       const struct qstr *filename);
+int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, uint64_t block,
+			    struct buffer_head **bhp);
+
+static inline uint32_t gfs2_disk_hash(const char *data, int len)
+{
+        return crc32_le(0xFFFFFFFF, data, len) ^ 0xFFFFFFFF;
+}
+
+
+static inline void gfs2_str2qstr(struct qstr *name, const char *fname)
+{
+	name->name = fname;
+	name->len = strlen(fname);
+	name->hash = gfs2_disk_hash(name->name, name->len);
+}
+
+/* N.B. This probably ought to take inum & type as args as well */
+static inline void gfs2_qstr2dirent(const struct qstr *name, u16 reclen, struct gfs2_dirent *dent)
+{
+	dent->de_inum.no_addr = cpu_to_be64(0);
+	dent->de_inum.no_formal_ino = cpu_to_be64(0);
+	dent->de_hash = cpu_to_be32(name->hash);
+	dent->de_rec_len = cpu_to_be16(reclen);
+	dent->de_name_len = cpu_to_be16(name->len);
+	dent->de_type = cpu_to_be16(0);
+	memset(dent->__pad, 0, sizeof(dent->__pad));
+	memcpy((char*)(dent+1), name->name, name->len);
+}
+
+#endif /* __DIR_DOT_H__ */

Re: [PATCH 07/16] GFS2: Directory handling

[Jan Engelhardt]

>+
>+	return copied;
>+fail:
>+	return (copied) ? copied : error;
>+}
>+
>+typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
>+			    const struct qstr *name,
>+			    void *opaque);

Collect all the typedefs above all functions (applies to all .c files).

>+static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
>+				     const struct qstr *name, int ret)
>+{
>+	if (dent->de_inum.no_addr != 0 &&
>+	    be32_to_cpu(dent->de_hash) == name->hash &&
>+	    be16_to_cpu(dent->de_name_len) == name->len &&
>+	    memcmp((char *)(dent+1), name->name, name->len) == 0)

Nocast.

>+static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode,
>+					    void *buf,
>+					    unsigned int len, gfs2_dscan_t scan,
>+					    const struct qstr *name,
>+					    void *opaque)
>+{
>+	struct gfs2_dirent *dent, *prev;
>+	unsigned offset;
>+	unsigned size;
>+	int ret = 0;
>+
>+	ret = gfs2_dirent_offset(buf);
>+	if (ret < 0)
>+		goto consist_inode;
>+
>+	offset = ret;
>+	prev = NULL;
>+	dent = (struct gfs2_dirent *)(buf + offset);

Nocast.

>+		dent = (struct gfs2_dirent *)(buf + offset);


>+	if ((char *)cur + cur_rec_len >= bh_end) {
>+		if ((char *)cur + cur_rec_len > bh_end) {
>+			gfs2_consist_inode(dip);
>+			return -EIO;
>+		}
>+		return -ENOENT;
>+	}

if((char *)cur + cur_rec_len > bh_end) {
	gfs2_consist_inode(dip);
	return -EIO;
} else if((char *)cur + cur_rec_len == bh_end)
	return -ENOENT;

>+	tmp = (struct gfs2_dirent *)((char *)cur + cur_rec_len);
>+
>+	if ((char *)tmp + be16_to_cpu(tmp->de_rec_len) > bh_end) {

Aah, this makes my eyes hurt. Though, it is probably a task not too short to
think of something that would do without casts.

>+	leaf->lf_depth = cpu_to_be16(depth);
>+	leaf->lf_entries = cpu_to_be16(0);

0 is said to be portable across all CPUs, therefore

	leaf->lf_entries = 0;

should suffice.

>+	leaf->lf_next = cpu_to_be64(0);


>+		for (x = sdp->sd_hash_ptrs; x--; from++) {
>+			*to++ = *from;	/*  No endianess worries  */
>+			*to++ = *from;
>+		}

Add /* Hakuna Matata */ and there will never be worries :-)

>+static int compare_dents(const void *a, const void *b)
>+{
>+	struct gfs2_dirent *dent_a, *dent_b;
>+	uint32_t hash_a, hash_b;
>+	int ret = 0;
>+
>+	dent_a = *(struct gfs2_dirent **)a;

But in this function you can have it const! Or not?

>+	hash_a = be32_to_cpu(dent_a->de_hash);
>+
>+	dent_b = *(struct gfs2_dirent **)b;
>+	hash_b = be32_to_cpu(dent_b->de_hash);
>+
>+	if (hash_a > hash_b)
>+		ret = 1;
>+	else if (hash_a < hash_b)
>+		ret = -1;
>+	else {
>+		unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
>+		unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
>+
>+		if (len_a > len_b)
>+			ret = 1;
>+		else if (len_a < len_b)
>+			ret = -1;
>+		else
>+			ret = memcmp((char *)(dent_a + 1),
>+				     (char *)(dent_b + 1),
>+				     len_a);

Nocast.

>+		error = filldir(opaque, (char *)(dent + 1),

If you case, then cast it directly, in this case, filldir takes "const char *"
as 2nd type.

>+	larr = vmalloc((leaves + entries) * sizeof(void*));
                                                       ^
Space, to go in line with all the other casts.

>+static inline uint32_t gfs2_disk_hash(const char *data, int len)
>+{
>+        return crc32_le(0xFFFFFFFF, data, len) ^ 0xFFFFFFFF;
>+}

Mind using (uint32_t)-1 or (uint32_t)~0 for that?

>+	memcpy((char*)(dent+1), name->name, name->len);

Last but not least, nocast.


Otherwise just stuff already mentioned.

Jan Engelhardt
-- 

-- 
VGER BF report: H 0.0146386

Re: [PATCH 07/16] GFS2: Directory handling

[Ingo Molnar]

* Steven Whitehouse <swhiteho@redhat.com> wrote:

> > >+static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
> > >+				     const struct qstr *name, int ret)
> > >+{
> > >+	if (dent->de_inum.no_addr != 0 &&
> > >+	    be32_to_cpu(dent->de_hash) == name->hash &&
> > >+	    be16_to_cpu(dent->de_name_len) == name->len &&
> > >+	    memcmp((char *)(dent+1), name->name, name->len) == 0)
> > 
> > Nocast.
> > 
> ok

actually, sizeof(*dent) != 1, so how can a non-casted memcmp be correct 
here?

> > >+	if ((char *)cur + cur_rec_len >= bh_end) {
> > >+		if ((char *)cur + cur_rec_len > bh_end) {
> > >+			gfs2_consist_inode(dip);
> > >+			return -EIO;
> > >+		}
> > >+		return -ENOENT;
> > >+	}
> > 
> > if((char *)cur + cur_rec_len > bh_end) {
> > 	gfs2_consist_inode(dip);
> > 	return -EIO;
> > } else if((char *)cur + cur_rec_len == bh_end)
> > 	return -ENOENT;
> > 
> ok

this one is not OK! Firstly, Jan, and i mentioned this before, please 
stop using 'if(', it is highly inconsistent and against basic taste. We 
only use this construct for function calls (and macros), not for C 
statements.

Secondly, whenever we have curly braces in the first block, we tend to 
do it in the second block too, for easier parsing. I.e.:

	if ((char *)cur + cur_rec_len > bh_end) {
		gfs2_consist_inode(dip);
		return -EIO;
	} else {
		if ((char *)cur + cur_rec_len == bh_end)
			return -ENOENT;
	}

Thirdly, the original code was quite fine as-is! What's the point of 
introducing random perturbations like this? It is an open invitation for 
the introduction of bugs... So unless there is a clear style reason to 
do a change, i'd suggest to not touch the code.

	Ingo

Re: [PATCH 07/16] GFS2: Directory handling

[Steven Whitehouse]

Hi,

On Mon, 2006-09-04 at 13:35 +0200, Jan Engelhardt wrote:
> >+
> >+	return copied;
> >+fail:
> >+	return (copied) ? copied : error;
> >+}
> >+
> >+typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
> >+			    const struct qstr *name,
> >+			    void *opaque);
> 
> Collect all the typedefs above all functions (applies to all .c files).
> 
ok

> >+static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
> >+				     const struct qstr *name, int ret)
> >+{
> >+	if (dent->de_inum.no_addr != 0 &&
> >+	    be32_to_cpu(dent->de_hash) == name->hash &&
> >+	    be16_to_cpu(dent->de_name_len) == name->len &&
> >+	    memcmp((char *)(dent+1), name->name, name->len) == 0)
> 
> Nocast.
> 
ok

> >+static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode,
[lines snipped]
> >+	offset = ret;
> >+	prev = NULL;
> >+	dent = (struct gfs2_dirent *)(buf + offset);
> 
> Nocast.
> 
ok
> >+		dent = (struct gfs2_dirent *)(buf + offset);
> 
> 
> >+	if ((char *)cur + cur_rec_len >= bh_end) {
> >+		if ((char *)cur + cur_rec_len > bh_end) {
> >+			gfs2_consist_inode(dip);
> >+			return -EIO;
> >+		}
> >+		return -ENOENT;
> >+	}
> 
> if((char *)cur + cur_rec_len > bh_end) {
> 	gfs2_consist_inode(dip);
> 	return -EIO;
> } else if((char *)cur + cur_rec_len == bh_end)
> 	return -ENOENT;
> 
ok

> >+	tmp = (struct gfs2_dirent *)((char *)cur + cur_rec_len);
> >+
> >+	if ((char *)tmp + be16_to_cpu(tmp->de_rec_len) > bh_end) {
> 
> Aah, this makes my eyes hurt. Though, it is probably a task not too short to
> think of something that would do without casts.
> 
Well I've had a crack at this. I've got it down to one cast. Let me know
if you think thats ok now... I managed to factor out some common code at
the same time.

> >+	leaf->lf_depth = cpu_to_be16(depth);
> >+	leaf->lf_entries = cpu_to_be16(0);
> 
> 0 is said to be portable across all CPUs, therefore
> 
> 	leaf->lf_entries = 0;
> 
> should suffice.
> 
> >+	leaf->lf_next = cpu_to_be64(0);
> 
ok, both now fixed.

> 
> >+		for (x = sdp->sd_hash_ptrs; x--; from++) {
> >+			*to++ = *from;	/*  No endianess worries  */
> >+			*to++ = *from;
> >+		}
> 
> Add /* Hakuna Matata */ and there will never be worries :-)
> 
:-) I had to use google to understand the reference :-)

> >+static int compare_dents(const void *a, const void *b)
> >+{
> >+	struct gfs2_dirent *dent_a, *dent_b;
> >+	uint32_t hash_a, hash_b;
> >+	int ret = 0;
> >+
> >+	dent_a = *(struct gfs2_dirent **)a;
> 
> But in this function you can have it const! Or not?
> 
Yes, and now all converted to const
[lines snipped]
> 		ret = memcmp((char *)(dent_a + 1),
> >+				     (char *)(dent_b + 1),
> >+				     len_a);
> 
> Nocast.
> 
ok

> >+		error = filldir(opaque, (char *)(dent + 1),
> 
> If you case, then cast it directly, in this case, filldir takes "const char *"
> as 2nd type.
> 
ok

> >+	larr = vmalloc((leaves + entries) * sizeof(void*));
>                                                        ^
> Space, to go in line with all the other casts.
> 
fixed

> >+static inline uint32_t gfs2_disk_hash(const char *data, int len)
> >+{
> >+        return crc32_le(0xFFFFFFFF, data, len) ^ 0xFFFFFFFF;
> >+}
> 
> Mind using (uint32_t)-1 or (uint32_t)~0 for that?
> 
~0 is the variant I've chosen.

> >+	memcpy((char*)(dent+1), name->name, name->len);
> 
> Last but not least, nocast.
> 
ok.

The patch is:
http://www.kernel.org/git/?p=linux/kernel/git/steve/gfs2-2.6.git;a=commitdiff;h=2bdbc5d73961c040fdc9b30d985fab3047d697a0

Thanks for doing all this review work, btw. I'll be working on your more
recent emails today,

Steve.

Re: [PATCH 07/16] GFS2: Directory handling

[Andreas Schwab]

Ingo Molnar <mingo@elte.hu> writes:

> * Steven Whitehouse <swhiteho@redhat.com> wrote:
>
>> > >+static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
>> > >+				     const struct qstr *name, int ret)
>> > >+{
>> > >+	if (dent->de_inum.no_addr != 0 &&
>> > >+	    be32_to_cpu(dent->de_hash) == name->hash &&
>> > >+	    be16_to_cpu(dent->de_name_len) == name->len &&
>> > >+	    memcmp((char *)(dent+1), name->name, name->len) == 0)
>> > 
>> > Nocast.
>> > 
>> ok
>
> actually, sizeof(*dent) != 1, so how can a non-casted memcmp be correct 
> here?

How can the cast change anything?

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
PGP key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [PATCH 07/16] GFS2: Directory handling

[Jan Engelhardt]

>> > >+static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
>> > >+				     const struct qstr *name, int ret)
>> > >+{
>> > >+	if (dent->de_inum.no_addr != 0 &&
>> > >+	    be32_to_cpu(dent->de_hash) == name->hash &&
>> > >+	    be16_to_cpu(dent->de_name_len) == name->len &&
>> > >+	    memcmp((char *)(dent+1), name->name, name->len) == 0)
>> > 
>> > Nocast.
>> > 
>> ok
>
>actually, sizeof(*dent) != 1, so how can a non-casted memcmp be correct 
>here?

There is an implicit reinterpret_cast<> from char* to void* when using 
memcmp, because void* is what memcpy takes as first argument. So, in effect 
we are doing

	memcmp((void *)(char *)(dent + 1), ...)

and that cast is, in this case, redundant, meaning we can do

	memcmp((void *)(dent + 1), ...)

But since conversion from and to void* is implicit in C, we can do

	memcmp(dent + 1, ...)


>> > >+	if ((char *)cur + cur_rec_len >= bh_end) {
>> > >+		if ((char *)cur + cur_rec_len > bh_end) {
>> > >+			gfs2_consist_inode(dip);
>> > >+			return -EIO;
>> > >+		}
>> > >+		return -ENOENT;
>> > >+	}
>> > 
>> > if((char *)cur + cur_rec_len > bh_end) {
>> > 	gfs2_consist_inode(dip);
>> > 	return -EIO;
>> > } else if((char *)cur + cur_rec_len == bh_end)
>> > 	return -ENOENT;
>> > 
>> ok
>
>this one is not OK! Firstly, Jan, and i mentioned this before, please 
>stop using 'if(', it is highly inconsistent and against basic taste. We 
>only use this construct for function calls (and macros), not for C 
>statements.

Now there is no rule in CodingStyle for this yet. Plus, I was wanting to show
how to reorder the construct, so change in whitespace between "if" and "(" is
outoftopic.

11:17 gwdg-wb04A:~/linux > grep -Pri '(if|for|while)\(' . | wc -l
24242

[To be honest, I also present the other number:]
11:17 gwdg-wb04A:~/linux > grep -Pri '(if|for|while) \(' . | wc -l
380895

Although a minority, it does not seem so uncommon.



>Secondly, whenever we have curly braces in the first block, we tend to 
>do it in the second block too, for easier parsing. I.e.:
>
>	if ((char *)cur + cur_rec_len > bh_end) {
>		gfs2_consist_inode(dip);
>		return -EIO;
>	} else {
>		if ((char *)cur + cur_rec_len == bh_end)
>			return -ENOENT;
>	}

I would very much like to do just this

		many insns;
	} else if(...) {
		single insn;
	}

but again, some people think no {} should be there because it's just a single
insn. Though I don't go harvesting in lkml.org archives right now to prove
this claim.



>Thirdly, the original code was quite fine as-is! What's the point of 
>introducing random perturbations like this? It is an open invitation for 
>the introduction of bugs... So unless there is a clear style reason to 
>do a change, i'd suggest to not touch the code.

Intent was reduction of indent. ""The answer to that is that if you need
more than 3 levels of indentation, you're screwed anyway, and should fix
your program."" so says CodingStyle, though the 3-level barrier was only
touched this time.

No offense!



Jan Engelhardt
-- 

Re: [PATCH 07/16] GFS2: Directory handling

[Ingo Molnar]

* Andreas Schwab <schwab@suse.de> wrote:

> >> > >+	if (dent->de_inum.no_addr != 0 &&
> >> > >+	    be32_to_cpu(dent->de_hash) == name->hash &&
> >> > >+	    be16_to_cpu(dent->de_name_len) == name->len &&
> >> > >+	    memcmp((char *)(dent+1), name->name, name->len) == 0)
> >> > 
> >> > Nocast.
> >> > 
> >> ok
> >
> > actually, sizeof(*dent) != 1, so how can a non-casted memcmp be correct 
> > here?
> 
> How can the cast change anything?

yeah - i for a minute thought that the '+' is outside the cast - but it 
is inside.

	Ingo

Re: [PATCH 07/16] GFS2: Directory handling

[Ingo Molnar]

* Jan Engelhardt <jengelh@linux01.gwdg.de> wrote:

> >> > >+	if ((char *)cur + cur_rec_len >= bh_end) {
> >> > >+		if ((char *)cur + cur_rec_len > bh_end) {
> >> > >+			gfs2_consist_inode(dip);
> >> > >+			return -EIO;
> >> > >+		}
> >> > >+		return -ENOENT;
> >> > >+	}
> >> > 
> >> > if((char *)cur + cur_rec_len > bh_end) {
> >> > 	gfs2_consist_inode(dip);
> >> > 	return -EIO;
> >> > } else if((char *)cur + cur_rec_len == bh_end)
> >> > 	return -ENOENT;
> >> > 
> >> ok
> >
> >this one is not OK! Firstly, Jan, and i mentioned this before, please 
> >stop using 'if(', it is highly inconsistent and against basic taste. We 
> >only use this construct for function calls (and macros), not for C 
> >statements.
> 
> Now there is no rule in CodingStyle for this yet. [...]

as i told you before, it is basic taste. Show me the CodingStyle rules 
that forbids this construct:

	if((((i+-+-+-+-+1-1))))
		var=0;;;;;;;

instead of:

	if (i)
		var = 0;

?

You will find no explicit rule in CodingStyle that explicitly forbids 
any aspect of this line, because CodingStyle mainly concentrates on the 
harder to follow rules. So please use common sense in combination with 
CodingStyle.

(And even if you dont agree with the taste, "if ( )" is done by 98% of 
core code, so for the sake of consistency it's the thing to follow.)

> 11:17 gwdg-wb04A:~/linux > grep -Pri '(if|for|while)\(' . | wc -l
> 24242
> 
> [To be honest, I also present the other number:]
> 11:17 gwdg-wb04A:~/linux > grep -Pri '(if|for|while) \(' . | wc -l
> 380895
> 
> Although a minority, it does not seem so uncommon.

What is your point?

ugly code is not at all uncommon, sadly - but i'm glad that it's only 6% 
in this particular case (and 1.9% for core kernel code).

> [...] Plus, I was wanting to show how to reorder the construct, so 
> change in whitespace between "if" and "(" is outoftopic.

it is not offtopic, because you _are_ nitpicking over style issues 
(which is OK and desirable, if not overdone), while still seeming to 
believe that "if(" is fine :-) The one who judges others is judged by 
higher standards. In other words, i'm trying to fix the guy who is 
fixing others ;-)

	Ingo

Re: [PATCH 07/16] GFS2: Directory handling

[Ingo Molnar]

* Jan Engelhardt <jengelh@linux01.gwdg.de> wrote:

> >Secondly, whenever we have curly braces in the first block, we tend to 
> >do it in the second block too, for easier parsing. I.e.:
> >
> >	if ((char *)cur + cur_rec_len > bh_end) {
> >		gfs2_consist_inode(dip);
> >		return -EIO;
> >	} else {
> >		if ((char *)cur + cur_rec_len == bh_end)
> >			return -ENOENT;
> >	}
> 
> I would very much like to do just this
> 
> 		many insns;
> 	} else if(...) {
> 		single insn;
> 	}

yeah, almost - what we want is:

		many insns;
	} else if (...) {
		single insn;
	}

;-)

> but again, some people think no {} should be there because it's just a 
> single insn. [...]

i used to be such a person but got converted: it makes visual sense to 
have symmetry of curly braces. So it's fine to have:

	if (...)
		single insn;
	else
		single insn;

but otherwise, if any of the blocks grows larger than 1 line, it should 
be curly braces all around.

	Ingo

Re: [PATCH 07/16] GFS2: Directory handling

[Steven Whitehouse]

On Tue, 2006-09-05 at 10:43 +0200, Ingo Molnar wrote:
> * Steven Whitehouse <swhiteho@redhat.com> wrote:
[some lines snipped]
> > > >+	if ((char *)cur + cur_rec_len >= bh_end) {
> > > >+		if ((char *)cur + cur_rec_len > bh_end) {
> > > >+			gfs2_consist_inode(dip);
> > > >+			return -EIO;
> > > >+		}
> > > >+		return -ENOENT;
> > > >+	}
> > > 
> > > if((char *)cur + cur_rec_len > bh_end) {
> > > 	gfs2_consist_inode(dip);
> > > 	return -EIO;
> > > } else if((char *)cur + cur_rec_len == bh_end)
> > > 	return -ENOENT;
> > > 
> > ok
> 
> this one is not OK! Firstly, Jan, and i mentioned this before, please 
> stop using 'if(', it is highly inconsistent and against basic taste. We 
> only use this construct for function calls (and macros), not for C 
> statements.

I did use use if "if (" form when I made this change rather than a
direct copy & paste from this. So it is ok I think,

Steve.

Re: [PATCH 07/16] GFS2: Directory handling

[Steven Whitehouse]

Hi,

On Fri, 2006-04-21 at 17:16 +0100, Christoph Hellwig wrote:
> > +/*
> > +* Implements Extendible Hashing as described in:
> > +*   "Extendible Hashing" by Fagin, et al in
> > +*     __ACM Trans. on Database Systems__, Sept 1979.
> > +*
> > +*
> 
> please follow the normal comment style, that is leave a space before the *
> for block comments so it lines up nicely with the * in the start tag.
> 
> > +#include <linux/sched.h>
> > +#include <linux/slab.h>
> > +#include <linux/spinlock.h>
> > +#include <linux/completion.h>
> 
> you don't seem to be using any completion in this file
> 
> > +#include <linux/buffer_head.h>
> > +#include <linux/sort.h>
> > +#include <linux/gfs2_ondisk.h>
> > +#include <linux/crc32.h>
> > +#include <linux/vmalloc.h>
> > +#include <asm/semaphore.h>
> 
> you're not using any semaphore in this file
> 
> > +int gfs2_dir_get_buffer(struct gfs2_inode *ip, uint64_t block, int new,
> > +		         struct buffer_head **bhp)
[function body cut for clarity]
> 
> the code is completely different for the new vs !new case, so there's no
> point in merging it to a single function.
> 
These points are now fixed in the git tree:
http://www.kernel.org/git/?p=linux/kernel/git/steve/gfs2-2.6.git;a=commitdiff;h=61e085a88cb59232eb8ff5b446d70491c7bf2c68

Thanks for the comments, please let me know if I missed anything,

Steve.

Re: [PATCH 07/16] GFS2: Directory handling

[Christoph Hellwig]

> +/*
> +* Implements Extendible Hashing as described in:
> +*   "Extendible Hashing" by Fagin, et al in
> +*     __ACM Trans. on Database Systems__, Sept 1979.
> +*
> +*

please follow the normal comment style, that is leave a space before the *
for block comments so it lines up nicely with the * in the start tag.

> +#include <linux/sched.h>
> +#include <linux/slab.h>
> +#include <linux/spinlock.h>
> +#include <linux/completion.h>

you don't seem to be using any completion in this file

> +#include <linux/buffer_head.h>
> +#include <linux/sort.h>
> +#include <linux/gfs2_ondisk.h>
> +#include <linux/crc32.h>
> +#include <linux/vmalloc.h>
> +#include <asm/semaphore.h>

you're not using any semaphore in this file

> +int gfs2_dir_get_buffer(struct gfs2_inode *ip, uint64_t block, int new,
> +		         struct buffer_head **bhp)
> +{
> +	struct buffer_head *bh;
> +	int error = 0;
> +
> +	if (new) {
> +		bh = gfs2_meta_new(ip->i_gl, block);
> +		gfs2_trans_add_bh(ip->i_gl, bh, 1);
> +		gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
> +		gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
> +	} else {
> +		error = gfs2_meta_read(ip->i_gl, block, DIO_START | DIO_WAIT,
> +				       &bh);
> +		if (error)
> +			return error;
> +		if (gfs2_metatype_check(ip->i_sbd, bh, GFS2_METATYPE_JD)) {
> +			brelse(bh);
> +			return -EIO;
> +		}
> +	}
> +
> +	*bhp = bh;
> +	return 0;

the code is completely different for the new vs !new case, so there's no
point in merging it to a single function.

[PATCH 07/16] GFS2: Directory handling

[Steven Whitehouse]

[PATCH 07/16] GFS2: Directory handling

Directory handling code for GFS2. This has been shrunk and made
a bit faster and a lot easier to understand since it was last
posted. Suggestions for how to deal with the readdir call in a
more efficient manner are greatly welcomed.

GFS2 uses extensible hashing in the directory code which makes lookups
fast. Due to the way in which the hash tables are split during
directory expansion, the directory contents are sorted during readdir
and it would be nice to be able to eliminate this operation.


Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>


 fs/gfs2/dir.c | 1968 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 fs/gfs2/dir.h |   73 ++
 2 files changed, 2041 insertions(+)

--- /dev/null
+++ b/fs/gfs2/dir.c
@@ -0,0 +1,1968 @@
+/*
+ * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
+ * Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ */
+
+/*
+* Implements Extendible Hashing as described in:
+*   "Extendible Hashing" by Fagin, et al in
+*     __ACM Trans. on Database Systems__, Sept 1979.
+*
+*
+* Here's the layout of dirents which is essentially the same as that of ext2
+* within a single block. The field de_name_len is the number of bytes
+* actually required for the name (no null terminator). The field de_rec_len
+* is the number of bytes allocated to the dirent. The offset of the next
+* dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
+* deleted, the preceding dirent inherits its allocated space, ie
+* prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
+* by adding de_rec_len to the current dirent, this essentially causes the
+* deleted dirent to get jumped over when iterating through all the dirents.
+*
+* When deleting the first dirent in a block, there is no previous dirent so
+* the field de_ino is set to zero to designate it as deleted. When allocating
+* a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
+* first dirent has (de_ino == 0) and de_rec_len is large enough, this first
+* dirent is allocated. Otherwise it must go through all the 'used' dirents
+* searching for one in which the amount of total space minus the amount of
+* used space will provide enough space for the new dirent.
+*
+* There are two types of blocks in which dirents reside. In a stuffed dinode,
+* the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
+* the block.  In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
+* beginning of the leaf block. The dirents reside in leaves when
+*
+* dip->i_di.di_flags & GFS2_DIF_EXHASH is true
+*
+* Otherwise, the dirents are "linear", within a single stuffed dinode block.
+*
+* When the dirents are in leaves, the actual contents of the directory file are
+* used as an array of 64-bit block pointers pointing to the leaf blocks. The
+* dirents are NOT in the directory file itself. There can be more than one block
+* pointer in the array that points to the same leaf. In fact, when a directory
+* is first converted from linear to exhash, all of the pointers point to the
+* same leaf.
+*
+* When a leaf is completely full, the size of the hash table can be
+* doubled unless it is already at the maximum size which is hard coded into
+* GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
+* but never before the maximum hash table size has been reached.
+*/
+
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/completion.h>
+#include <linux/buffer_head.h>
+#include <linux/sort.h>
+#include <linux/gfs2_ondisk.h>
+#include <linux/crc32.h>
+#include <linux/vmalloc.h>
+#include <asm/semaphore.h>
+
+#include "gfs2.h"
+#include "lm_interface.h"
+#include "incore.h"
+#include "dir.h"
+#include "glock.h"
+#include "inode.h"
+#include "meta_io.h"
+#include "quota.h"
+#include "rgrp.h"
+#include "trans.h"
+#include "bmap.h"
+#include "util.h"
+
+#define IS_LEAF     1 /* Hashed (leaf) directory */
+#define IS_DINODE   2 /* Linear (stuffed dinode block) directory */
+
+#if 1
+#define gfs2_disk_hash2offset(h) (((uint64_t)(h)) >> 1)
+#define gfs2_dir_offset2hash(p) ((uint32_t)(((uint64_t)(p)) << 1))
+#else
+#define gfs2_disk_hash2offset(h) (((uint64_t)(h)))
+#define gfs2_dir_offset2hash(p) ((uint32_t)(((uint64_t)(p))))
+#endif
+
+typedef int (*leaf_call_t) (struct gfs2_inode *dip,
+			    uint32_t index, uint32_t len, uint64_t leaf_no,
+			    void *data);
+
+int gfs2_dir_get_buffer(struct gfs2_inode *ip, uint64_t block, int new,
+		         struct buffer_head **bhp)
+{
+	struct buffer_head *bh;
+	int error = 0;
+
+	if (new) {
+		bh = gfs2_meta_new(ip->i_gl, block);
+		gfs2_trans_add_bh(ip->i_gl, bh, 1);
+		gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
+		gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
+	} else {
+		error = gfs2_meta_read(ip->i_gl, block, DIO_START | DIO_WAIT,
+				       &bh);
+		if (error)
+			return error;
+		if (gfs2_metatype_check(ip->i_sbd, bh, GFS2_METATYPE_JD)) {
+			brelse(bh);
+			return -EIO;
+		}
+	}
+
+	*bhp = bh;
+	return 0;
+}
+
+
+
+static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
+				  unsigned int offset, unsigned int size)
+                               
+{
+	struct buffer_head *dibh;
+	int error;
+
+	error = gfs2_meta_inode_buffer(ip, &dibh);
+	if (error)
+		return error;
+
+	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
+	memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
+	if (ip->i_di.di_size < offset + size)
+		ip->i_di.di_size = offset + size;
+	ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
+	gfs2_dinode_out(&ip->i_di, dibh->b_data);
+
+	brelse(dibh);
+
+	return size;
+}
+
+
+
+/**
+ * gfs2_dir_write_data - Write directory information to the inode
+ * @ip: The GFS2 inode
+ * @buf: The buffer containing information to be written
+ * @offset: The file offset to start writing at
+ * @size: The amount of data to write
+ *
+ * Returns: The number of bytes correctly written or error code
+ */
+static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
+			       uint64_t offset, unsigned int size)
+{
+	struct gfs2_sbd *sdp = ip->i_sbd;
+	struct buffer_head *dibh;
+	uint64_t lblock, dblock;
+	uint32_t extlen = 0;
+	unsigned int o;
+	int copied = 0;
+	int error = 0;
+
+	if (!size)
+		return 0;
+
+	if (gfs2_is_stuffed(ip) &&
+	    offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
+		return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
+					      size);
+
+	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
+		return -EINVAL;
+
+	if (gfs2_is_stuffed(ip)) {
+		error = gfs2_unstuff_dinode(ip, NULL, NULL);
+		if (error)
+			return error;
+	}
+
+	lblock = offset;
+	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
+
+	while (copied < size) {
+		unsigned int amount;
+		struct buffer_head *bh;
+		int new;
+
+		amount = size - copied;
+		if (amount > sdp->sd_sb.sb_bsize - o)
+			amount = sdp->sd_sb.sb_bsize - o;
+
+		if (!extlen) {
+			new = 1;
+			error = gfs2_block_map(ip, lblock, &new, &dblock,
+					       &extlen);
+			if (error)
+				goto fail;
+			error = -EIO;
+			if (gfs2_assert_withdraw(sdp, dblock))
+				goto fail;
+		}
+
+		error = gfs2_dir_get_buffer(ip, dblock,
+					    (amount == sdp->sd_jbsize) ?
+					    1 : new, &bh);
+		if (error)
+			goto fail;
+
+		gfs2_trans_add_bh(ip->i_gl, bh, 1);
+		memcpy(bh->b_data + o, buf, amount);
+		brelse(bh);
+		if (error)
+			goto fail;
+
+		copied += amount;
+		lblock++;
+		dblock++;
+		extlen--;
+
+		o = sizeof(struct gfs2_meta_header);
+	}
+
+out:
+	error = gfs2_meta_inode_buffer(ip, &dibh);
+	if (error)
+		return error;
+
+	if (ip->i_di.di_size < offset + copied)
+		ip->i_di.di_size = offset + copied;
+	ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
+
+	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
+	gfs2_dinode_out(&ip->i_di, dibh->b_data);
+	brelse(dibh);
+
+	return copied;
+fail:
+	if (copied)
+		goto out;
+	return error;
+}
+
+static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf,
+				 unsigned int offset, unsigned int size)
+{
+	struct buffer_head *dibh;
+	int error;
+
+	error = gfs2_meta_inode_buffer(ip, &dibh);
+	if (!error) {
+		offset += sizeof(struct gfs2_dinode);
+		memcpy(buf, dibh->b_data + offset, size);
+		brelse(dibh);
+	}
+
+	return (error) ? error : size;
+}
+
+
+/**
+ * gfs2_dir_read_data - Read a data from a directory inode
+ * @ip: The GFS2 Inode
+ * @buf: The buffer to place result into
+ * @offset: File offset to begin jdata_readng from
+ * @size: Amount of data to transfer
+ *
+ * Returns: The amount of data actually copied or the error
+ */
+static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf,
+			      uint64_t offset, unsigned int size)
+{
+	struct gfs2_sbd *sdp = ip->i_sbd;
+	uint64_t lblock, dblock;
+	uint32_t extlen = 0;
+	unsigned int o;
+	int copied = 0;
+	int error = 0;
+
+	if (offset >= ip->i_di.di_size)
+		return 0;
+
+	if ((offset + size) > ip->i_di.di_size)
+		size = ip->i_di.di_size - offset;
+
+	if (!size)
+		return 0;
+
+	if (gfs2_is_stuffed(ip))
+		return gfs2_dir_read_stuffed(ip, buf, (unsigned int)offset,
+					     size);
+
+	if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
+		return -EINVAL;
+
+	lblock = offset;
+	o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
+
+	while (copied < size) {
+		unsigned int amount;
+		struct buffer_head *bh;
+		int new;
+
+		amount = size - copied;
+		if (amount > sdp->sd_sb.sb_bsize - o)
+			amount = sdp->sd_sb.sb_bsize - o;
+
+		if (!extlen) {
+			new = 0;
+			error = gfs2_block_map(ip, lblock, &new, &dblock,
+					       &extlen);
+			if (error)
+				goto fail;
+		}
+
+		if (extlen > 1)
+			gfs2_meta_ra(ip->i_gl, dblock, extlen);
+
+		if (dblock) {
+			error = gfs2_dir_get_buffer(ip, dblock, new, &bh);
+			if (error)
+				goto fail;
+			dblock++;
+			extlen--;
+		} else
+			bh = NULL;
+
+		memcpy(buf, bh->b_data + o, amount);
+		brelse(bh);
+		if (error)
+			goto fail;
+
+		copied += amount;
+		lblock++;
+
+		o = sizeof(struct gfs2_meta_header);
+	}
+
+	return copied;
+fail:
+	return (copied) ? copied : error;
+}
+
+typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
+			    const struct qstr *name,
+			    void *opaque);
+
+static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
+				     const struct qstr *name, int ret)
+{
+	if (dent->de_inum.no_addr != 0 &&
+	    be32_to_cpu(dent->de_hash) == name->hash &&
+	    be16_to_cpu(dent->de_name_len) == name->len &&
+	    memcmp((char *)(dent+1), name->name, name->len) == 0)
+		return ret;
+	return 0;
+}
+
+static int gfs2_dirent_find(const struct gfs2_dirent *dent,
+			    const struct qstr *name,
+			    void *opaque)
+{
+	return __gfs2_dirent_find(dent, name, 1);
+}
+
+static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
+			    const struct qstr *name,
+			    void *opaque)
+{
+	return __gfs2_dirent_find(dent, name, 2);
+}
+
+/*
+ * name->name holds ptr to start of block.
+ * name->len holds size of block.
+ */
+static int gfs2_dirent_last(const struct gfs2_dirent *dent,
+			    const struct qstr *name,
+			    void *opaque)
+{
+	const char *start = name->name;
+	const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
+	if (name->len == (end - start))
+		return 1;
+	return 0;
+}
+
+static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
+				  const struct qstr *name,
+				  void *opaque)
+{
+	unsigned required = GFS2_DIRENT_SIZE(name->len);
+	unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
+	unsigned totlen = be16_to_cpu(dent->de_rec_len);
+
+	if (!dent->de_inum.no_addr)
+		actual = GFS2_DIRENT_SIZE(0);
+	if ((totlen - actual) >= required)
+		return 1;
+	return 0;
+}
+
+struct dirent_gather {
+	const struct gfs2_dirent **pdent;
+	unsigned offset;
+};
+
+static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
+			      const struct qstr *name,
+			      void *opaque)
+{
+	struct dirent_gather *g = opaque;
+	if (dent->de_inum.no_addr) {
+		g->pdent[g->offset++] = dent;
+	}
+	return 0;
+}
+
+/*
+ * Other possible things to check:
+ * - Inode located within filesystem size (and on valid block)
+ * - Valid directory entry type
+ * Not sure how heavy-weight we want to make this... could also check
+ * hash is correct for example, but that would take a lot of extra time.
+ * For now the most important thing is to check that the various sizes
+ * are correct.
+ */
+static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
+			     unsigned int size, unsigned int len, int first)
+{
+	const char *msg = "gfs2_dirent too small";
+	if (unlikely(size < sizeof(struct gfs2_dirent)))
+		goto error;
+	msg = "gfs2_dirent misaligned";
+	if (unlikely(offset & 0x7))
+		goto error;
+	msg = "gfs2_dirent points beyond end of block";
+	if (unlikely(offset + size > len))
+		goto error;
+	msg = "zero inode number";
+	if (unlikely(!first && !dent->de_inum.no_addr))
+		goto error;
+	msg = "name length is greater than space in dirent";
+	if (dent->de_inum.no_addr &&
+	    unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
+		     size))
+		goto error;
+	return 0;
+error:
+	printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg,
+	       first ? "first in block" : "not first in block");
+	return -EIO;
+}
+
+static int gfs2_dirent_offset(const void *buf)
+{
+	const struct gfs2_meta_header *h = buf;
+	int offset;
+
+	BUG_ON(buf == NULL);
+
+	switch(be32_to_cpu(h->mh_type)) {
+	case GFS2_METATYPE_LF:
+		offset = sizeof(struct gfs2_leaf);
+		break;
+	case GFS2_METATYPE_DI:
+		offset = sizeof(struct gfs2_dinode);
+		break;
+	default:
+		goto wrong_type;
+	}
+	return offset;
+wrong_type:
+	printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n",
+	       be32_to_cpu(h->mh_type));
+	return -1;
+}
+
+static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode,
+					    void *buf,
+					    unsigned int len, gfs2_dscan_t scan,
+					    const struct qstr *name,
+					    void *opaque)
+{
+	struct gfs2_dirent *dent, *prev;
+	unsigned offset;
+	unsigned size;
+	int ret = 0;
+
+	ret = gfs2_dirent_offset(buf);
+	if (ret < 0)
+		goto consist_inode;
+
+	offset = ret;
+	prev = NULL;
+	dent = (struct gfs2_dirent *)(buf + offset);
+	size = be16_to_cpu(dent->de_rec_len);
+	if (gfs2_check_dirent(dent, offset, size, len, 1))
+		goto consist_inode;
+	do {
+		ret = scan(dent, name, opaque);
+		if (ret)
+			break;
+		offset += size;
+		if (offset == len)
+			break;
+		prev = dent;
+		dent = (struct gfs2_dirent *)(buf + offset);
+		size = be16_to_cpu(dent->de_rec_len);
+		if (gfs2_check_dirent(dent, offset, size, len, 0))
+			goto consist_inode;
+	} while(1);
+
+	switch(ret) {
+	case 0:
+		return NULL;
+	case 1:
+		return dent;
+	case 2:
+		return prev ? prev : dent;
+	default:
+		BUG_ON(ret > 0);
+		return ERR_PTR(ret);
+	}
+
+consist_inode:
+	gfs2_consist_inode(inode->u.generic_ip);
+	return ERR_PTR(-EIO);
+}
+
+
+/**
+ * dirent_first - Return the first dirent
+ * @dip: the directory
+ * @bh: The buffer
+ * @dent: Pointer to list of dirents
+ *
+ * return first dirent whether bh points to leaf or stuffed dinode
+ *
+ * Returns: IS_LEAF, IS_DINODE, or -errno
+ */
+
+static int dirent_first(struct gfs2_inode *dip, struct buffer_head *bh,
+			struct gfs2_dirent **dent)
+{
+	struct gfs2_meta_header *h = (struct gfs2_meta_header *)bh->b_data;
+
+	if (be32_to_cpu(h->mh_type) == GFS2_METATYPE_LF) {
+		if (gfs2_meta_check(dip->i_sbd, bh))
+			return -EIO;
+		*dent = (struct gfs2_dirent *)(bh->b_data +
+					       sizeof(struct gfs2_leaf));
+		return IS_LEAF;
+	} else {
+		if (gfs2_metatype_check(dip->i_sbd, bh, GFS2_METATYPE_DI))
+			return -EIO;
+		*dent = (struct gfs2_dirent *)(bh->b_data +
+					       sizeof(struct gfs2_dinode));
+		return IS_DINODE;
+	}
+}
+
+/**
+ * dirent_next - Next dirent
+ * @dip: the directory
+ * @bh: The buffer
+ * @dent: Pointer to list of dirents
+ *
+ * Returns: 0 on success, error code otherwise
+ */
+
+static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
+		       struct gfs2_dirent **dent)
+{
+	struct gfs2_dirent *tmp, *cur;
+	char *bh_end;
+	uint16_t cur_rec_len;
+
+	cur = *dent;
+	bh_end = bh->b_data + bh->b_size;
+	cur_rec_len = be16_to_cpu(cur->de_rec_len);
+
+	if ((char *)cur + cur_rec_len >= bh_end) {
+		if ((char *)cur + cur_rec_len > bh_end) {
+			gfs2_consist_inode(dip);
+			return -EIO;
+		}
+		return -ENOENT;
+	}
+
+	tmp = (struct gfs2_dirent *)((char *)cur + cur_rec_len);
+
+	if ((char *)tmp + be16_to_cpu(tmp->de_rec_len) > bh_end) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+	if (cur_rec_len == 0) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+        /* Only the first dent could ever have de_inum.no_addr == 0 */
+	if (!tmp->de_inum.no_addr) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+	*dent = tmp;
+
+	return 0;
+}
+
+/**
+ * dirent_del - Delete a dirent
+ * @dip: The GFS2 inode
+ * @bh: The buffer
+ * @prev: The previous dirent
+ * @cur: The current dirent
+ *
+ */
+
+static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
+		       struct gfs2_dirent *prev, struct gfs2_dirent *cur)
+{
+	uint16_t cur_rec_len, prev_rec_len;
+
+	if (!cur->de_inum.no_addr) {
+		gfs2_consist_inode(dip);
+		return;
+	}
+
+	gfs2_trans_add_bh(dip->i_gl, bh, 1);
+
+	/* If there is no prev entry, this is the first entry in the block.
+	   The de_rec_len is already as big as it needs to be.  Just zero
+	   out the inode number and return.  */
+
+	if (!prev) {
+		cur->de_inum.no_addr = 0;	/* No endianess worries */
+		return;
+	}
+
+	/*  Combine this dentry with the previous one.  */
+
+	prev_rec_len = be16_to_cpu(prev->de_rec_len);
+	cur_rec_len = be16_to_cpu(cur->de_rec_len);
+
+	if ((char *)prev + prev_rec_len != (char *)cur)
+		gfs2_consist_inode(dip);
+	if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
+		gfs2_consist_inode(dip);
+
+	prev_rec_len += cur_rec_len;
+	prev->de_rec_len = cpu_to_be16(prev_rec_len);
+}
+
+/*
+ * Takes a dent from which to grab space as an argument. Returns the
+ * newly created dent.
+ */
+struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
+				     struct gfs2_dirent *dent,
+				     const struct qstr *name,
+				     struct buffer_head *bh)
+{
+	struct gfs2_inode *ip = inode->u.generic_ip;
+	struct gfs2_dirent *ndent;
+	unsigned offset = 0, totlen;
+
+	if (dent->de_inum.no_addr)
+		offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
+	totlen = be16_to_cpu(dent->de_rec_len);
+	BUG_ON(offset + name->len > totlen);
+	gfs2_trans_add_bh(ip->i_gl, bh, 1);
+	ndent = (struct gfs2_dirent *)((char *)dent + offset);
+	dent->de_rec_len = cpu_to_be16(offset);
+	gfs2_qstr2dirent(name, totlen - offset, ndent);
+	return ndent;
+}
+
+static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
+					     struct buffer_head *bh,
+					     const struct qstr *name)
+{
+	struct gfs2_dirent *dent;
+	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, 
+				gfs2_dirent_find_space, name, NULL);
+	if (!dent || IS_ERR(dent))
+		return dent;
+	return gfs2_init_dirent(inode, dent, name, bh);
+}
+
+static int get_leaf(struct gfs2_inode *dip, uint64_t leaf_no,
+		    struct buffer_head **bhp)
+{
+	int error;
+
+	error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_START | DIO_WAIT, bhp);
+	if (!error && gfs2_metatype_check(dip->i_sbd, *bhp, GFS2_METATYPE_LF))
+		error = -EIO;
+
+	return error;
+}
+
+/**
+ * get_leaf_nr - Get a leaf number associated with the index
+ * @dip: The GFS2 inode
+ * @index:
+ * @leaf_out:
+ *
+ * Returns: 0 on success, error code otherwise
+ */
+
+static int get_leaf_nr(struct gfs2_inode *dip, uint32_t index,
+		       uint64_t *leaf_out)
+{
+	uint64_t leaf_no;
+	int error;
+
+	error = gfs2_dir_read_data(dip, (char *)&leaf_no,
+				    index * sizeof(uint64_t),
+				    sizeof(uint64_t));
+	if (error != sizeof(uint64_t))
+		return (error < 0) ? error : -EIO;
+
+	*leaf_out = be64_to_cpu(leaf_no);
+
+	return 0;
+}
+
+static int get_first_leaf(struct gfs2_inode *dip, uint32_t index,
+			  struct buffer_head **bh_out)
+{
+	uint64_t leaf_no;
+	int error;
+
+	error = get_leaf_nr(dip, index, &leaf_no);
+	if (!error)
+		error = get_leaf(dip, leaf_no, bh_out);
+
+	return error;
+}
+
+static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
+					      const struct qstr *name,
+					      gfs2_dscan_t scan,
+					      struct buffer_head **pbh)
+{
+	struct buffer_head *bh;
+	struct gfs2_dirent *dent;
+	struct gfs2_inode *ip = inode->u.generic_ip;
+	int error;
+
+	if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
+		struct gfs2_leaf *leaf;
+		unsigned hsize = 1 << ip->i_di.di_depth;
+		unsigned index;
+		u64 ln;
+		if (hsize * sizeof(u64) != ip->i_di.di_size) {
+			gfs2_consist_inode(ip);
+			return ERR_PTR(-EIO);
+		}
+
+		index = name->hash >> (32 - ip->i_di.di_depth);
+		error = get_first_leaf(ip, index, &bh);
+		if (error)
+			return ERR_PTR(error);
+		do {
+			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
+						scan, name, NULL);
+			if (dent)
+				goto got_dent;
+			leaf = (struct gfs2_leaf *)bh->b_data;
+			ln = be64_to_cpu(leaf->lf_next);
+			brelse(bh);
+			if (!ln)
+				break;
+			error = get_leaf(ip, ln, &bh);
+		} while(!error);
+
+		return error ? ERR_PTR(error) : NULL;
+	}
+
+	error = gfs2_meta_inode_buffer(ip, &bh);
+	if (error)
+		return ERR_PTR(error);
+	dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
+got_dent:
+	if (unlikely(dent == NULL || IS_ERR(dent))) {
+		brelse(bh);
+		bh = NULL;
+	}
+	*pbh = bh;
+	return dent;
+}
+
+static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
+{
+	struct gfs2_inode *ip = inode->u.generic_ip;
+	u64 bn = gfs2_alloc_meta(ip);
+	struct buffer_head *bh = gfs2_meta_new(ip->i_gl, bn);
+	struct gfs2_leaf *leaf;
+	struct gfs2_dirent *dent;
+	struct qstr name = { .name = "", .len = 0, .hash = 0 };
+	if (!bh)
+		return NULL;
+	gfs2_trans_add_bh(ip->i_gl, bh, 1);
+	gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
+	leaf = (struct gfs2_leaf *)bh->b_data;
+	leaf->lf_depth = cpu_to_be16(depth);
+	leaf->lf_entries = cpu_to_be16(0);
+	leaf->lf_dirent_format = cpu_to_be16(GFS2_FORMAT_DE);
+	leaf->lf_next = cpu_to_be64(0);
+	memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved));
+	dent = (struct gfs2_dirent *)(leaf+1);
+	gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
+	*pbh = bh;
+	return leaf;
+}
+
+/**
+ * dir_make_exhash - Convert a stuffed directory into an ExHash directory
+ * @dip: The GFS2 inode
+ *
+ * Returns: 0 on success, error code otherwise
+ */
+
+static int dir_make_exhash(struct inode *inode)
+{
+	struct gfs2_inode *dip = inode->u.generic_ip;
+	struct gfs2_sbd *sdp = dip->i_sbd;
+	struct gfs2_dirent *dent;
+	struct qstr args;
+	struct buffer_head *bh, *dibh;
+	struct gfs2_leaf *leaf;
+	int y;
+	uint32_t x;
+	uint64_t *lp, bn;
+	int error;
+
+	error = gfs2_meta_inode_buffer(dip, &dibh);
+	if (error)
+		return error;
+
+	/*  Turn over a new leaf  */
+
+	leaf = new_leaf(inode, &bh, 0);
+	if (!leaf)
+		return -ENOSPC;
+	bn = bh->b_blocknr;
+
+	gfs2_assert(sdp, dip->i_di.di_entries < (1 << 16));
+	leaf->lf_entries = cpu_to_be16(dip->i_di.di_entries);
+
+	/*  Copy dirents  */
+
+	gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
+			     sizeof(struct gfs2_dinode));
+
+	/*  Find last entry  */
+
+	x = 0;
+	args.len = bh->b_size - sizeof(struct gfs2_dinode) +
+		   sizeof(struct gfs2_leaf);
+	args.name = bh->b_data;
+	dent = gfs2_dirent_scan(dip->i_vnode, bh->b_data, bh->b_size,
+				gfs2_dirent_last, &args, NULL);
+	if (!dent) {
+		brelse(bh);
+		brelse(dibh);
+		return -EIO;
+	}
+	if (IS_ERR(dent)) {
+		brelse(bh);
+		brelse(dibh);
+		return PTR_ERR(dent);
+	}
+
+	/*  Adjust the last dirent's record length
+	   (Remember that dent still points to the last entry.)  */
+
+	dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
+		sizeof(struct gfs2_dinode) -
+		sizeof(struct gfs2_leaf));
+
+	brelse(bh);
+
+	/*  We're done with the new leaf block, now setup the new
+	    hash table.  */
+
+	gfs2_trans_add_bh(dip->i_gl, dibh, 1);
+	gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
+
+	lp = (uint64_t *)(dibh->b_data + sizeof(struct gfs2_dinode));
+
+	for (x = sdp->sd_hash_ptrs; x--; lp++)
+		*lp = cpu_to_be64(bn);
+
+	dip->i_di.di_size = sdp->sd_sb.sb_bsize / 2;
+	dip->i_di.di_blocks++;
+	dip->i_di.di_flags |= GFS2_DIF_EXHASH;
+	dip->i_di.di_payload_format = 0;
+
+	for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
+	dip->i_di.di_depth = y;
+
+	gfs2_dinode_out(&dip->i_di, dibh->b_data);
+
+	brelse(dibh);
+
+	return 0;
+}
+
+/**
+ * dir_split_leaf - Split a leaf block into two
+ * @dip: The GFS2 inode
+ * @index:
+ * @leaf_no:
+ *
+ * Returns: 0 on success, error code on failure
+ */
+
+static int dir_split_leaf(struct inode *inode, const struct qstr *name)
+{
+	struct gfs2_inode *dip = inode->u.generic_ip;
+	struct buffer_head *nbh, *obh, *dibh;
+	struct gfs2_leaf *nleaf, *oleaf;
+	struct gfs2_dirent *dent, *prev = NULL, *next = NULL, *new;
+	uint32_t start, len, half_len, divider;
+	uint64_t bn, *lp, leaf_no;
+	uint32_t index;
+	int x, moved = 0;
+	int error;
+
+	index = name->hash >> (32 - dip->i_di.di_depth);
+	error = get_leaf_nr(dip, index, &leaf_no);
+	if (error)
+		return error;
+
+	/*  Get the old leaf block  */
+	error = get_leaf(dip, leaf_no, &obh);
+	if (error)
+		return error;
+
+	oleaf = (struct gfs2_leaf *)obh->b_data;
+	if (dip->i_di.di_depth == be16_to_cpu(oleaf->lf_depth)) {
+		brelse(obh);
+		return 1; /* can't split */
+	}
+
+	gfs2_trans_add_bh(dip->i_gl, obh, 1);
+
+	nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
+	if (!nleaf) {
+		brelse(obh);
+		return -ENOSPC;
+	}
+	bn = nbh->b_blocknr;
+
+	/*  Compute the start and len of leaf pointers in the hash table.  */
+	len = 1 << (dip->i_di.di_depth - be16_to_cpu(oleaf->lf_depth));
+	half_len = len >> 1;
+	if (!half_len) {
+		printk(KERN_WARNING "di_depth %u lf_depth %u index %u\n", dip->i_di.di_depth, be16_to_cpu(oleaf->lf_depth), index);
+		gfs2_consist_inode(dip);
+		error = -EIO;
+		goto fail_brelse;
+	}
+
+	start = (index & ~(len - 1));
+
+	/* Change the pointers.
+	   Don't bother distinguishing stuffed from non-stuffed.
+	   This code is complicated enough already. */
+	lp = kmalloc(half_len * sizeof(uint64_t), GFP_NOFS | __GFP_NOFAIL);
+	/*  Change the pointers  */
+	for (x = 0; x < half_len; x++)
+		lp[x] = cpu_to_be64(bn);
+
+	error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(uint64_t),
+				    half_len * sizeof(uint64_t));
+	if (error != half_len * sizeof(uint64_t)) {
+		if (error >= 0)
+			error = -EIO;
+		goto fail_lpfree;
+	}
+
+	kfree(lp);
+
+	/*  Compute the divider  */
+	divider = (start + half_len) << (32 - dip->i_di.di_depth);
+
+	/*  Copy the entries  */
+	dirent_first(dip, obh, &dent);
+
+	do {
+		next = dent;
+		if (dirent_next(dip, obh, &next))
+			next = NULL;
+
+		if (dent->de_inum.no_addr &&
+		    be32_to_cpu(dent->de_hash) < divider) {
+			struct qstr str;
+			str.name = (char*)(dent+1);
+			str.len = be16_to_cpu(dent->de_name_len);
+			str.hash = be32_to_cpu(dent->de_hash);
+			new = gfs2_dirent_alloc(inode, nbh, &str);
+			if (IS_ERR(new)) {
+				error = PTR_ERR(new);
+				break;
+			}
+
+			new->de_inum = dent->de_inum; /* No endian worries */
+			new->de_type = dent->de_type; /* No endian worries */
+			nleaf->lf_entries = cpu_to_be16(be16_to_cpu(nleaf->lf_entries)+1);
+
+			dirent_del(dip, obh, prev, dent);
+
+			if (!oleaf->lf_entries)
+				gfs2_consist_inode(dip);
+			oleaf->lf_entries = cpu_to_be16(be16_to_cpu(oleaf->lf_entries)-1);
+
+			if (!prev)
+				prev = dent;
+
+			moved = 1;
+		} else {
+			prev = dent;
+		}
+		dent = next;
+	} while (dent);
+
+	oleaf->lf_depth = nleaf->lf_depth;
+
+	error = gfs2_meta_inode_buffer(dip, &dibh);
+	if (!gfs2_assert_withdraw(dip->i_sbd, !error)) {
+		dip->i_di.di_blocks++;
+		gfs2_dinode_out(&dip->i_di, dibh->b_data);
+		brelse(dibh);
+	}
+
+	brelse(obh);
+	brelse(nbh);
+
+	return error;
+
+fail_lpfree:
+	kfree(lp);
+
+fail_brelse:
+	brelse(obh);
+	brelse(nbh);
+	return error;
+}
+
+/**
+ * dir_double_exhash - Double size of ExHash table
+ * @dip: The GFS2 dinode
+ *
+ * Returns: 0 on success, error code on failure
+ */
+
+static int dir_double_exhash(struct gfs2_inode *dip)
+{
+	struct gfs2_sbd *sdp = dip->i_sbd;
+	struct buffer_head *dibh;
+	uint32_t hsize;
+	uint64_t *buf;
+	uint64_t *from, *to;
+	uint64_t block;
+	int x;
+	int error = 0;
+
+	hsize = 1 << dip->i_di.di_depth;
+	if (hsize * sizeof(uint64_t) != dip->i_di.di_size) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+	/*  Allocate both the "from" and "to" buffers in one big chunk  */
+
+	buf = kcalloc(3, sdp->sd_hash_bsize, GFP_KERNEL | __GFP_NOFAIL);
+
+	for (block = dip->i_di.di_size >> sdp->sd_hash_bsize_shift; block--;) {
+		error = gfs2_dir_read_data(dip, (char *)buf,
+					    block * sdp->sd_hash_bsize,
+					    sdp->sd_hash_bsize);
+		if (error != sdp->sd_hash_bsize) {
+			if (error >= 0)
+				error = -EIO;
+			goto fail;
+		}
+
+		from = buf;
+		to = (uint64_t *)((char *)buf + sdp->sd_hash_bsize);
+
+		for (x = sdp->sd_hash_ptrs; x--; from++) {
+			*to++ = *from;	/*  No endianess worries  */
+			*to++ = *from;
+		}
+
+		error = gfs2_dir_write_data(dip,
+					     (char *)buf + sdp->sd_hash_bsize,
+					     block * sdp->sd_sb.sb_bsize,
+					     sdp->sd_sb.sb_bsize);
+		if (error != sdp->sd_sb.sb_bsize) {
+			if (error >= 0)
+				error = -EIO;
+			goto fail;
+		}
+	}
+
+	kfree(buf);
+
+	error = gfs2_meta_inode_buffer(dip, &dibh);
+	if (!gfs2_assert_withdraw(sdp, !error)) {
+		dip->i_di.di_depth++;
+		gfs2_dinode_out(&dip->i_di, dibh->b_data);
+		brelse(dibh);
+	}
+
+	return error;
+
+ fail:
+	kfree(buf);
+
+	return error;
+}
+
+/**
+ * compare_dents - compare directory entries by hash value
+ * @a: first dent
+ * @b: second dent
+ *
+ * When comparing the hash entries of @a to @b:
+ *   gt: returns 1
+ *   lt: returns -1
+ *   eq: returns 0
+ */
+
+static int compare_dents(const void *a, const void *b)
+{
+	struct gfs2_dirent *dent_a, *dent_b;
+	uint32_t hash_a, hash_b;
+	int ret = 0;
+
+	dent_a = *(struct gfs2_dirent **)a;
+	hash_a = be32_to_cpu(dent_a->de_hash);
+
+	dent_b = *(struct gfs2_dirent **)b;
+	hash_b = be32_to_cpu(dent_b->de_hash);
+
+	if (hash_a > hash_b)
+		ret = 1;
+	else if (hash_a < hash_b)
+		ret = -1;
+	else {
+		unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
+		unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
+
+		if (len_a > len_b)
+			ret = 1;
+		else if (len_a < len_b)
+			ret = -1;
+		else
+			ret = memcmp((char *)(dent_a + 1),
+				     (char *)(dent_b + 1),
+				     len_a);
+	}
+
+	return ret;
+}
+
+/**
+ * do_filldir_main - read out directory entries
+ * @dip: The GFS2 inode
+ * @offset: The offset in the file to read from
+ * @opaque: opaque data to pass to filldir
+ * @filldir: The function to pass entries to
+ * @darr: an array of struct gfs2_dirent pointers to read
+ * @entries: the number of entries in darr
+ * @copied: pointer to int that's non-zero if a entry has been copied out
+ *
+ * Jump through some hoops to make sure that if there are hash collsions,
+ * they are read out at the beginning of a buffer.  We want to minimize
+ * the possibility that they will fall into different readdir buffers or
+ * that someone will want to seek to that location.
+ *
+ * Returns: errno, >0 on exception from filldir
+ */
+
+static int do_filldir_main(struct gfs2_inode *dip, uint64_t *offset,
+			   void *opaque, gfs2_filldir_t filldir,
+			   const struct gfs2_dirent **darr, uint32_t entries,
+			   int *copied)
+{
+	const struct gfs2_dirent *dent, *dent_next;
+	struct gfs2_inum inum;
+	uint64_t off, off_next;
+	unsigned int x, y;
+	int run = 0;
+	int error = 0;
+
+	sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
+
+	dent_next = darr[0];
+	off_next = be32_to_cpu(dent_next->de_hash);
+	off_next = gfs2_disk_hash2offset(off_next);
+
+	for (x = 0, y = 1; x < entries; x++, y++) {
+		dent = dent_next;
+		off = off_next;
+
+		if (y < entries) {
+			dent_next = darr[y];
+			off_next = be32_to_cpu(dent_next->de_hash);
+			off_next = gfs2_disk_hash2offset(off_next);
+
+			if (off < *offset)
+				continue;
+			*offset = off;
+
+			if (off_next == off) {
+				if (*copied && !run)
+					return 1;
+				run = 1;
+			} else
+				run = 0;
+		} else {
+			if (off < *offset)
+				continue;
+			*offset = off;
+		}
+
+		gfs2_inum_in(&inum, (char *)&dent->de_inum);
+
+		error = filldir(opaque, (char *)(dent + 1),
+				be16_to_cpu(dent->de_name_len),
+				off, &inum,
+				be16_to_cpu(dent->de_type));
+		if (error)
+			return 1;
+
+		*copied = 1;
+	}
+
+	/* Increment the *offset by one, so the next time we come into the
+	   do_filldir fxn, we get the next entry instead of the last one in the
+	   current leaf */
+
+	(*offset)++;
+
+	return 0;
+}
+
+static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque,
+			      gfs2_filldir_t filldir, int *copied,
+			      unsigned *depth, u64 leaf_no)
+{
+	struct gfs2_inode *ip = inode->u.generic_ip;
+	struct buffer_head *bh;
+	struct gfs2_leaf *lf;
+	unsigned entries = 0;
+	unsigned leaves = 0;
+	const struct gfs2_dirent **darr, *dent;
+	struct dirent_gather g;
+	struct buffer_head **larr;
+	int leaf = 0;
+	int error, i;
+	u64 lfn = leaf_no;
+
+	do {
+		error = get_leaf(ip, lfn, &bh);
+		if (error)
+			goto out;
+		lf = (struct gfs2_leaf *)bh->b_data;
+		if (leaves == 0)
+			*depth = be16_to_cpu(lf->lf_depth);
+		entries += be16_to_cpu(lf->lf_entries);
+		leaves++;
+		lfn = be64_to_cpu(lf->lf_next);
+		brelse(bh);
+	} while(lfn);
+
+	if (!entries)
+		return 0;
+
+	error = -ENOMEM;
+	larr = vmalloc((leaves + entries) * sizeof(void*));
+	if (!larr)
+		goto out;
+	darr = (const struct gfs2_dirent **)(larr + leaves);
+	g.pdent = darr;
+	g.offset = 0;
+	lfn = leaf_no;
+
+	do {
+		error = get_leaf(ip, lfn, &bh);
+		if (error)
+			goto out_kfree;
+		lf = (struct gfs2_leaf *)bh->b_data;
+		lfn = be64_to_cpu(lf->lf_next);
+		if (lf->lf_entries) {
+			dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
+						gfs2_dirent_gather, NULL, &g);
+			error = PTR_ERR(dent);
+			if (IS_ERR(dent)) {
+				goto out_kfree;
+			}
+			error = 0;
+			larr[leaf++] = bh;
+		} else {
+			brelse(bh);
+		}
+	} while(lfn);
+
+	error = do_filldir_main(ip, offset, opaque, filldir, darr,
+				entries, copied);
+out_kfree:
+	for(i = 0; i < leaf; i++)
+		brelse(larr[i]);
+	vfree(larr);
+out:
+	return error;
+}
+
+/**
+ * dir_e_read - Reads the entries from a directory into a filldir buffer
+ * @dip: dinode pointer
+ * @offset: the hash of the last entry read shifted to the right once
+ * @opaque: buffer for the filldir function to fill
+ * @filldir: points to the filldir function to use
+ *
+ * Returns: errno
+ */
+
+static int dir_e_read(struct inode *inode, uint64_t *offset, void *opaque,
+		      gfs2_filldir_t filldir)
+{
+	struct gfs2_inode *dip = inode->u.generic_ip;
+	struct gfs2_sbd *sdp = dip->i_sbd;
+	uint32_t hsize, len = 0;
+	uint32_t ht_offset, lp_offset, ht_offset_cur = -1;
+	uint32_t hash, index;
+	uint64_t *lp;
+	int copied = 0;
+	int error = 0;
+	unsigned depth;
+
+	hsize = 1 << dip->i_di.di_depth;
+	if (hsize * sizeof(uint64_t) != dip->i_di.di_size) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+	hash = gfs2_dir_offset2hash(*offset);
+	index = hash >> (32 - dip->i_di.di_depth);
+
+	lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
+	if (!lp)
+		return -ENOMEM;
+
+	while (index < hsize) {
+		lp_offset = index & (sdp->sd_hash_ptrs - 1);
+		ht_offset = index - lp_offset;
+
+		if (ht_offset_cur != ht_offset) {
+			error = gfs2_dir_read_data(dip, (char *)lp,
+						ht_offset * sizeof(uint64_t),
+						sdp->sd_hash_bsize);
+			if (error != sdp->sd_hash_bsize) {
+				if (error >= 0)
+					error = -EIO;
+				goto out;
+			}
+			ht_offset_cur = ht_offset;
+		}
+
+		error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
+					   &copied, &depth,
+					   be64_to_cpu(lp[lp_offset]));
+		if (error)
+			break;
+
+		len = 1 << (dip->i_di.di_depth - depth);
+		index = (index & ~(len - 1)) + len;
+	}
+
+out:
+	kfree(lp);
+	if (error > 0)
+		error = 0;
+	return error;
+}
+
+int gfs2_dir_read(struct inode *inode, uint64_t *offset, void *opaque,
+		  gfs2_filldir_t filldir)
+{
+	struct gfs2_inode *dip = inode->u.generic_ip;
+	struct dirent_gather g;
+	const struct gfs2_dirent **darr, *dent;
+	struct buffer_head *dibh;
+	int copied = 0;
+	int error;
+
+	if (!dip->i_di.di_entries)
+		return 0;
+
+	if (dip->i_di.di_flags & GFS2_DIF_EXHASH)
+		return dir_e_read(inode, offset, opaque, filldir);
+
+	if (!gfs2_is_stuffed(dip)) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+	error = gfs2_meta_inode_buffer(dip, &dibh);
+	if (error)
+		return error;
+
+	error = -ENOMEM;
+	darr = kmalloc(dip->i_di.di_entries * sizeof(struct gfs2_dirent *),
+		       GFP_KERNEL);
+	if (darr) {
+		g.pdent = darr;
+		g.offset = 0;
+		dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
+					gfs2_dirent_gather, NULL, &g);
+		if (IS_ERR(dent)) {
+			error = PTR_ERR(dent);
+			goto out;
+		}
+		error = do_filldir_main(dip, offset, opaque, filldir, darr,
+					dip->i_di.di_entries, &copied);
+out:
+		kfree(darr);
+	}
+
+	if (error > 0)
+		error = 0;
+
+	brelse(dibh);
+
+	return error;
+}
+
+/**
+ * gfs2_dir_search - Search a directory
+ * @dip: The GFS2 inode
+ * @filename:
+ * @inode:
+ *
+ * This routine searches a directory for a file or another directory.
+ * Assumes a glock is held on dip.
+ *
+ * Returns: errno
+ */
+
+int gfs2_dir_search(struct inode *dir, const struct qstr *name,
+		    struct gfs2_inum *inum, unsigned int *type)
+{
+	struct buffer_head *bh;
+	struct gfs2_dirent *dent;
+
+	dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
+	if (dent) {
+		if (IS_ERR(dent))
+			return PTR_ERR(dent);
+		if (inum)
+			gfs2_inum_in(inum, (char *)&dent->de_inum);
+		if (type)
+			*type = be16_to_cpu(dent->de_type);
+		brelse(bh);
+		return 0;
+	}
+	return -ENOENT;
+}
+
+static int dir_new_leaf(struct inode *inode, const struct qstr *name)
+{
+	struct buffer_head *bh, *obh;
+	struct gfs2_inode *ip = inode->u.generic_ip;
+	struct gfs2_leaf *leaf, *oleaf;
+	int error;
+	u32 index;
+	u64 bn;
+
+	index = name->hash >> (32 - ip->i_di.di_depth);
+	error = get_first_leaf(ip, index, &obh);
+	if (error)
+		return error;
+	do {
+		oleaf = (struct gfs2_leaf *)obh->b_data;
+		bn = be64_to_cpu(oleaf->lf_next);
+		if (!bn)
+			break;
+		brelse(obh);
+		error = get_leaf(ip, bn, &obh);
+		if (error)
+			return error;
+	} while(1);
+
+	gfs2_trans_add_bh(ip->i_gl, obh, 1);
+
+	leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
+	if (!leaf) {
+		brelse(obh);
+		return -ENOSPC;
+	}
+	oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
+	brelse(bh);
+	brelse(obh);
+
+	error = gfs2_meta_inode_buffer(ip, &bh);
+	if (error)
+		return error;
+	gfs2_trans_add_bh(ip->i_gl, bh, 1);
+	ip->i_di.di_blocks++;
+	gfs2_dinode_out(&ip->i_di, bh->b_data);
+	brelse(bh);
+	return 0;
+}
+
+/**
+ * gfs2_dir_add - Add new filename into directory
+ * @dip: The GFS2 inode
+ * @filename: The new name
+ * @inode: The inode number of the entry
+ * @type: The type of the entry
+ *
+ * Returns: 0 on success, error code on failure
+ */
+
+int gfs2_dir_add(struct inode *inode, const struct qstr *name,
+		 const struct gfs2_inum *inum, unsigned type)
+{
+	struct gfs2_inode *ip = inode->u.generic_ip;
+	struct buffer_head *bh;
+	struct gfs2_dirent *dent;
+	struct gfs2_leaf *leaf;
+	int error;
+
+	while(1) {
+		dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space,
+					  &bh);
+		if (dent) {
+			if (IS_ERR(dent))
+				return PTR_ERR(dent);
+			dent = gfs2_init_dirent(inode, dent, name, bh);
+			gfs2_inum_out(inum, (char *)&dent->de_inum);
+			dent->de_type = cpu_to_be16(type);
+			if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
+				leaf = (struct gfs2_leaf *)bh->b_data;
+				leaf->lf_entries = cpu_to_be16(be16_to_cpu(leaf->lf_entries) + 1);
+			}
+			brelse(bh);
+			error = gfs2_meta_inode_buffer(ip, &bh);
+			if (error)
+				break;
+			gfs2_trans_add_bh(ip->i_gl, bh, 1);
+			ip->i_di.di_entries++;
+			ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
+			gfs2_dinode_out(&ip->i_di, bh->b_data);
+			brelse(bh);
+			error = 0;
+			break;
+		}
+		if (!(ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
+			error = dir_make_exhash(inode);
+			if (error)
+				break;
+			continue;
+		}
+		error = dir_split_leaf(inode, name);
+		if (error == 0)
+			continue;
+		if (error < 0)
+			break;
+		if (ip->i_di.di_depth < GFS2_DIR_MAX_DEPTH) {
+			error = dir_double_exhash(ip);
+			if (error)
+				break;
+			error = dir_split_leaf(inode, name);
+			if (error < 0)
+				break;
+			if (error == 0)
+				continue;
+		}
+		error = dir_new_leaf(inode, name);
+		if (!error)
+			continue;
+		error = -ENOSPC;
+		break;
+	}
+	return error;
+}
+
+
+/**
+ * gfs2_dir_del - Delete a directory entry
+ * @dip: The GFS2 inode
+ * @filename: The filename
+ *
+ * Returns: 0 on success, error code on failure
+ */
+
+int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name)
+{
+	struct gfs2_dirent *dent, *prev = NULL;
+	struct buffer_head *bh;
+	int error;
+
+	/* Returns _either_ the entry (if its first in block) or the
+	   previous entry otherwise */
+	dent = gfs2_dirent_search(dip->i_vnode, name, gfs2_dirent_prev, &bh);
+	if (!dent) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+	if (IS_ERR(dent)) {
+		gfs2_consist_inode(dip);
+		return PTR_ERR(dent);
+	}
+	/* If not first in block, adjust pointers accordingly */
+	if (gfs2_dirent_find(dent, name, NULL) == 0) {
+		prev = dent;
+		dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
+	}
+
+	dirent_del(dip, bh, prev, dent);
+	if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
+		struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
+		u16 entries = be16_to_cpu(leaf->lf_entries);
+		if (!entries)
+			gfs2_consist_inode(dip);
+		leaf->lf_entries = cpu_to_be16(--entries);
+	}
+	brelse(bh);
+
+	error = gfs2_meta_inode_buffer(dip, &bh);
+	if (error)
+		return error;
+
+	if (!dip->i_di.di_entries)
+		gfs2_consist_inode(dip);
+	gfs2_trans_add_bh(dip->i_gl, bh, 1);
+	dip->i_di.di_entries--;
+	dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
+	gfs2_dinode_out(&dip->i_di, bh->b_data);
+	brelse(bh);
+
+	return error;
+}
+
+/**
+ * gfs2_dir_mvino - Change inode number of directory entry
+ * @dip: The GFS2 inode
+ * @filename:
+ * @new_inode:
+ *
+ * This routine changes the inode number of a directory entry.  It's used
+ * by rename to change ".." when a directory is moved.
+ * Assumes a glock is held on dvp.
+ *
+ * Returns: errno
+ */
+
+int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
+		   struct gfs2_inum *inum, unsigned int new_type)
+{
+	struct buffer_head *bh;
+	struct gfs2_dirent *dent;
+	int error;
+
+	dent = gfs2_dirent_search(dip->i_vnode, filename, gfs2_dirent_find, &bh);
+	if (!dent) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+	if (IS_ERR(dent))
+		return PTR_ERR(dent);
+
+	gfs2_trans_add_bh(dip->i_gl, bh, 1);
+	gfs2_inum_out(inum, (char *)&dent->de_inum);
+	dent->de_type = cpu_to_be16(new_type);
+
+	if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
+		brelse(bh);
+		error = gfs2_meta_inode_buffer(dip, &bh);
+		if (error)
+			return error;
+		gfs2_trans_add_bh(dip->i_gl, bh, 1);
+	}
+
+	dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
+	gfs2_dinode_out(&dip->i_di, bh->b_data);
+	brelse(bh);
+	return 0;
+}
+
+/**
+ * foreach_leaf - call a function for each leaf in a directory
+ * @dip: the directory
+ * @lc: the function to call for each each
+ * @data: private data to pass to it
+ *
+ * Returns: errno
+ */
+
+static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data)
+{
+	struct gfs2_sbd *sdp = dip->i_sbd;
+	struct buffer_head *bh;
+	struct gfs2_leaf *leaf;
+	uint32_t hsize, len;
+	uint32_t ht_offset, lp_offset, ht_offset_cur = -1;
+	uint32_t index = 0;
+	uint64_t *lp;
+	uint64_t leaf_no;
+	int error = 0;
+
+	hsize = 1 << dip->i_di.di_depth;
+	if (hsize * sizeof(uint64_t) != dip->i_di.di_size) {
+		gfs2_consist_inode(dip);
+		return -EIO;
+	}
+
+	lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
+	if (!lp)
+		return -ENOMEM;
+
+	while (index < hsize) {
+		lp_offset = index & (sdp->sd_hash_ptrs - 1);
+		ht_offset = index - lp_offset;
+
+		if (ht_offset_cur != ht_offset) {
+			error = gfs2_dir_read_data(dip, (char *)lp,
+						ht_offset * sizeof(uint64_t),
+						sdp->sd_hash_bsize);
+			if (error != sdp->sd_hash_bsize) {
+				if (error >= 0)
+					error = -EIO;
+				goto out;
+			}
+			ht_offset_cur = ht_offset;
+		}
+
+		leaf_no = be64_to_cpu(lp[lp_offset]);
+		if (leaf_no) {
+			error = get_leaf(dip, leaf_no, &bh);
+			if (error)
+				goto out;
+			leaf = (struct gfs2_leaf *)bh->b_data;
+			brelse(bh);
+
+			len = 1 << (dip->i_di.di_depth - be16_to_cpu(leaf->lf_depth));
+
+			error = lc(dip, index, len, leaf_no, data);
+			if (error)
+				goto out;
+
+			index = (index & ~(len - 1)) + len;
+		} else
+			index++;
+	}
+
+	if (index != hsize) {
+		gfs2_consist_inode(dip);
+		error = -EIO;
+	}
+
+ out:
+	kfree(lp);
+
+	return error;
+}
+
+/**
+ * leaf_dealloc - Deallocate a directory leaf
+ * @dip: the directory
+ * @index: the hash table offset in the directory
+ * @len: the number of pointers to this leaf
+ * @leaf_no: the leaf number
+ * @data: not used
+ *
+ * Returns: errno
+ */
+
+static int leaf_dealloc(struct gfs2_inode *dip, uint32_t index, uint32_t len,
+			uint64_t leaf_no, void *data)
+{
+	struct gfs2_sbd *sdp = dip->i_sbd;
+	struct gfs2_leaf *tmp_leaf;
+	struct gfs2_rgrp_list rlist;
+	struct buffer_head *bh, *dibh;
+	uint64_t blk, nblk;
+	unsigned int rg_blocks = 0, l_blocks = 0;
+	char *ht;
+	unsigned int x, size = len * sizeof(uint64_t);
+	int error;
+
+	memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
+
+	ht = kzalloc(size, GFP_KERNEL);
+	if (!ht)
+		return -ENOMEM;
+
+	gfs2_alloc_get(dip);
+
+	error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
+	if (error)
+		goto out;
+
+	error = gfs2_rindex_hold(sdp, &dip->i_alloc.al_ri_gh);
+	if (error)
+		goto out_qs;
+
+	/*  Count the number of leaves  */
+
+	for (blk = leaf_no; blk; blk = nblk) {
+		error = get_leaf(dip, blk, &bh);
+		if (error)
+			goto out_rlist;
+		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
+		nblk = be64_to_cpu(tmp_leaf->lf_next);
+		brelse(bh);
+
+		gfs2_rlist_add(sdp, &rlist, blk);
+		l_blocks++;
+	}
+
+	gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, 0);
+
+	for (x = 0; x < rlist.rl_rgrps; x++) {
+		struct gfs2_rgrpd *rgd;
+		rgd = rlist.rl_ghs[x].gh_gl->gl_object;
+		rg_blocks += rgd->rd_ri.ri_length;
+	}
+
+	error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
+	if (error)
+		goto out_rlist;
+
+	error = gfs2_trans_begin(sdp,
+			rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
+			RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
+	if (error)
+		goto out_rg_gunlock;
+
+	for (blk = leaf_no; blk; blk = nblk) {
+		error = get_leaf(dip, blk, &bh);
+		if (error)
+			goto out_end_trans;
+		tmp_leaf = (struct gfs2_leaf *)bh->b_data;
+		nblk = be64_to_cpu(tmp_leaf->lf_next);
+		brelse(bh);
+
+		gfs2_free_meta(dip, blk, 1);
+
+		if (!dip->i_di.di_blocks)
+			gfs2_consist_inode(dip);
+		dip->i_di.di_blocks--;
+	}
+
+	error = gfs2_dir_write_data(dip, ht, index * sizeof(uint64_t), size);
+	if (error != size) {
+		if (error >= 0)
+			error = -EIO;
+		goto out_end_trans;
+	}
+
+	error = gfs2_meta_inode_buffer(dip, &dibh);
+	if (error)
+		goto out_end_trans;
+
+	gfs2_trans_add_bh(dip->i_gl, dibh, 1);
+	gfs2_dinode_out(&dip->i_di, dibh->b_data);
+	brelse(dibh);
+
+ out_end_trans:
+	gfs2_trans_end(sdp);
+
+ out_rg_gunlock:
+	gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
+
+ out_rlist:
+	gfs2_rlist_free(&rlist);
+	gfs2_glock_dq_uninit(&dip->i_alloc.al_ri_gh);
+
+ out_qs:
+	gfs2_quota_unhold(dip);
+
+ out:
+	gfs2_alloc_put(dip);
+	kfree(ht);
+
+	return error;
+}
+
+/**
+ * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
+ * @dip: the directory
+ *
+ * Dealloc all on-disk directory leaves to FREEMETA state
+ * Change on-disk inode type to "regular file"
+ *
+ * Returns: errno
+ */
+
+int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
+{
+	struct gfs2_sbd *sdp = dip->i_sbd;
+	struct buffer_head *bh;
+	int error;
+
+	/* Dealloc on-disk leaves to FREEMETA state */
+	error = foreach_leaf(dip, leaf_dealloc, NULL);
+	if (error)
+		return error;
+
+	/* Make this a regular file in case we crash.
+	   (We don't want to free these blocks a second time.)  */
+
+	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
+	if (error)
+		return error;
+
+	error = gfs2_meta_inode_buffer(dip, &bh);
+	if (!error) {
+		gfs2_trans_add_bh(dip->i_gl, bh, 1);
+		((struct gfs2_dinode *)bh->b_data)->di_mode =
+						cpu_to_be32(S_IFREG);
+		brelse(bh);
+	}
+
+	gfs2_trans_end(sdp);
+
+	return error;
+}
+
+/**
+ * gfs2_diradd_alloc_required - find if adding entry will require an allocation
+ * @ip: the file being written to
+ * @filname: the filename that's going to be added
+ *
+ * Returns: 1 if alloc required, 0 if not, -ve on error
+ */
+
+int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name)
+{
+	struct gfs2_dirent *dent;
+	struct buffer_head *bh;
+
+	dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
+	if (!dent) {
+		return 1;
+	}
+	if (IS_ERR(dent))
+		return PTR_ERR(dent);
+	brelse(bh);
+	return 0;
+}
+
--- /dev/null
+++ b/fs/gfs2/dir.h
@@ -0,0 +1,73 @@
+/*
+ * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
+ * Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ */
+
+#ifndef __DIR_DOT_H__
+#define __DIR_DOT_H__
+
+/**
+ * gfs2_filldir_t - Report a directory entry to the caller of gfs2_dir_read()
+ * @opaque: opaque data used by the function
+ * @name: the name of the directory entry
+ * @length: the length of the name
+ * @offset: the entry's offset in the directory
+ * @inum: the inode number the entry points to
+ * @type: the type of inode the entry points to
+ *
+ * Returns: 0 on success, 1 if buffer full
+ */
+
+typedef int (*gfs2_filldir_t) (void *opaque,
+			      const char *name, unsigned int length,
+			      uint64_t offset,
+			      struct gfs2_inum *inum, unsigned int type);
+
+int gfs2_dir_search(struct inode *dir, const struct qstr *filename,
+		    struct gfs2_inum *inum, unsigned int *type);
+int gfs2_dir_add(struct inode *inode, const struct qstr *filename,
+		 const struct gfs2_inum *inum, unsigned int type);
+int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *filename);
+int gfs2_dir_read(struct inode *inode, uint64_t * offset, void *opaque,
+		  gfs2_filldir_t filldir);
+int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
+		   struct gfs2_inum *new_inum, unsigned int new_type);
+
+int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip);
+
+int gfs2_diradd_alloc_required(struct inode *dir,
+			       const struct qstr *filename);
+int gfs2_dir_get_buffer(struct gfs2_inode *ip, uint64_t block, int new,
+                        struct buffer_head **bhp);
+
+static inline uint32_t gfs2_disk_hash(const char *data, int len)
+{
+        return crc32_le(0xFFFFFFFF, data, len) ^ 0xFFFFFFFF;
+}
+
+
+static inline void gfs2_str2qstr(struct qstr *name, const char *fname)
+{
+	name->name = fname;
+	name->len = strlen(fname);
+	name->hash = gfs2_disk_hash(name->name, name->len);
+}
+
+/* N.B. This probably ought to take inum & type as args as well */
+static inline void gfs2_qstr2dirent(const struct qstr *name, u16 reclen, struct gfs2_dirent *dent)
+{
+	dent->de_inum.no_addr = cpu_to_be64(0);
+	dent->de_inum.no_formal_ino = cpu_to_be64(0);
+	dent->de_hash = cpu_to_be32(name->hash);
+	dent->de_rec_len = cpu_to_be16(reclen);
+	dent->de_name_len = cpu_to_be16(name->len);
+	dent->de_type = cpu_to_be16(0);
+	memset(dent->__pad, 0, sizeof(dent->__pad));
+	memcpy((char*)(dent+1), name->name, name->len);
+}
+
+#endif /* __DIR_DOT_H__ */