[GFS2] Fix warning on impossible event in eattr code
[powerpc.git] / fs / gfs2 / dir.c
index ada283a..6918a58 100644 (file)
@@ -1,6 +1,6 @@
 /*
  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
- * Copyright (C) 2004-2005 Red Hat, Inc.  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
@@ -8,60 +8,63 @@
  */
 
 /*
-* 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.
-*/
+ * 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 <asm/semaphore.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"
@@ -70,6 +73,7 @@
 #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 */
@@ -86,32 +90,36 @@ typedef int (*leaf_call_t) (struct gfs2_inode *dip,
                            uint32_t index, uint32_t len, uint64_t leaf_no,
                            void *data);
 
-static int gfs2_dir_get_buffer(struct gfs2_inode *ip, uint64_t block, int new,
-                              struct buffer_head **bhp)
+
+int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, uint64_t block,
+                           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;
-               }
-       }
 
+       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(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)
@@ -125,7 +133,7 @@ static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
                return error;
 
        gfs2_trans_add_bh(ip->i_gl, dibh, 1);
-       memcpy(dibh->b_data + offset + sizeof(struct gfs2_inode), buf, size);
+       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();
@@ -163,7 +171,8 @@ static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
 
        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);
+               return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
+                                             size);
 
        if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
                return -EINVAL;
@@ -171,7 +180,7 @@ static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
        if (gfs2_is_stuffed(ip)) {
                error = gfs2_unstuff_dinode(ip, NULL, NULL);
                if (error)
-               return error;
+                       return error;
        }
 
        lblock = offset;
@@ -188,7 +197,8 @@ static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
 
                if (!extlen) {
                        new = 1;
-                       error = gfs2_block_map(ip, lblock, &new, &dblock, &extlen);
+                       error = gfs2_extent_map(ip->i_vnode, lblock, &new,
+                                               &dblock, &extlen);
                        if (error)
                                goto fail;
                        error = -EIO;
@@ -196,7 +206,11 @@ static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
                                goto fail;
                }
 
-               error = gfs2_dir_get_buffer(ip, dblock, (amount == sdp->sd_jbsize) ? 1 : new, &bh);
+               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;
 
@@ -235,7 +249,7 @@ fail:
 }
 
 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf,
-                               unsigned int offset, unsigned int size)
+                                unsigned int offset, unsigned int size)
 {
        struct buffer_head *dibh;
        int error;
@@ -280,7 +294,8 @@ static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf,
                return 0;
 
        if (gfs2_is_stuffed(ip))
-               return gfs2_dir_read_stuffed(ip, buf, (unsigned int)offset, size);
+               return gfs2_dir_read_stuffed(ip, buf, (unsigned int)offset,
+                                            size);
 
        if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
                return -EINVAL;
@@ -299,7 +314,8 @@ static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf,
 
                if (!extlen) {
                        new = 0;
-                       error = gfs2_block_map(ip, lblock, &new, &dblock, &extlen);
+                       error = gfs2_extent_map(ip->i_vnode, lblock, &new,
+                                               &dblock, &extlen);
                        if (error)
                                goto fail;
                }
@@ -308,7 +324,10 @@ static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf,
                        gfs2_meta_ra(ip->i_gl, dblock, extlen);
 
                if (dblock) {
-                       error = gfs2_dir_get_buffer(ip, dblock, new, &bh);
+                       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++;
@@ -332,26 +351,194 @@ fail:
        return (copied) ? copied : error;
 }
 
-/**
- * int gfs2_filecmp - Compare two filenames
- * @file1: The first filename
- * @file2: The second filename
- * @len_of_file2: The length of the second file
- *
- * This routine compares two filenames and returns 1 if they are equal.
- *
- * Returns: 1 if the files are the same, otherwise 0.
+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;
+}
 
-int gfs2_filecmp(struct qstr *file1, char *file2, int len_of_file2)
+static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
+                                 const struct qstr *name,
+                                 void *opaque)
 {
-       if (file1->len != len_of_file2)
-               return 0;
-       if (memcmp(file1->name, file2, file1->len))
-               return 0;
-       return 1;
+       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
@@ -368,7 +555,7 @@ static int dirent_first(struct gfs2_inode *dip, struct buffer_head *bh,
 {
        struct gfs2_meta_header *h = (struct gfs2_meta_header *)bh->b_data;
 
-       if (be16_to_cpu(h->mh_type) == GFS2_METATYPE_LF) {
+       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 +
@@ -397,11 +584,11 @@ static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
 {
        struct gfs2_dirent *tmp, *cur;
        char *bh_end;
-       uint32_t cur_rec_len;
+       uint16_t cur_rec_len;
 
        cur = *dent;
        bh_end = bh->b_data + bh->b_size;
-       cur_rec_len = be32_to_cpu(cur->de_rec_len);
+       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) {
@@ -413,10 +600,16 @@ static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
 
        tmp = (struct gfs2_dirent *)((char *)cur + cur_rec_len);
 
-       if ((char *)tmp + be32_to_cpu(tmp->de_rec_len) > bh_end) {
+       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);
@@ -440,7 +633,7 @@ static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
                       struct gfs2_dirent *prev, struct gfs2_dirent *cur)
 {
-       uint32_t cur_rec_len, prev_rec_len;
+       uint16_t cur_rec_len, prev_rec_len;
 
        if (!cur->de_inum.no_addr) {
                gfs2_consist_inode(dip);
@@ -460,8 +653,8 @@ static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
 
        /*  Combine this dentry with the previous one.  */
 
-       prev_rec_len = be32_to_cpu(prev->de_rec_len);
-       cur_rec_len = be32_to_cpu(cur->de_rec_len);
+       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);
@@ -469,182 +662,43 @@ static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
                gfs2_consist_inode(dip);
 
        prev_rec_len += cur_rec_len;
-       prev->de_rec_len = cpu_to_be32(prev_rec_len);
+       prev->de_rec_len = cpu_to_be16(prev_rec_len);
 }
 
-/**
- * gfs2_dirent_alloc - Allocate a directory entry
- * @dip: The GFS2 inode
- * @bh: The buffer
- * @name_len: The length of the name
- * @dent_out: Pointer to list of dirents
- *
- * Returns: 0 on success, error code otherwise
+/*
+ * Takes a dent from which to grab space as an argument. Returns the
+ * newly created dent.
  */
-
-int gfs2_dirent_alloc(struct gfs2_inode *dip, struct buffer_head *bh,
-                     int name_len, struct gfs2_dirent **dent_out)
+static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
+                                           struct gfs2_dirent *dent,
+                                           const struct qstr *name,
+                                           struct buffer_head *bh)
 {
-       struct gfs2_dirent *dent, *new;
-       unsigned int rec_len = GFS2_DIRENT_SIZE(name_len);
-       unsigned int entries = 0, offset = 0;
-       int type;
-
-       type = dirent_first(dip, bh, &dent);
-       if (type < 0)
-               return type;
-
-       if (type == IS_LEAF) {
-               struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
-               entries = be16_to_cpu(leaf->lf_entries);
-               offset = sizeof(struct gfs2_leaf);
-       } else {
-               struct gfs2_dinode *dinode = (struct gfs2_dinode *)bh->b_data;
-               entries = be32_to_cpu(dinode->di_entries);
-               offset = sizeof(struct gfs2_dinode);
-       }
-
-       if (!entries) {
-               if (dent->de_inum.no_addr) {
-                       gfs2_consist_inode(dip);
-                       return -EIO;
-               }
-
-               gfs2_trans_add_bh(dip->i_gl, bh, 1);
-
-               dent->de_rec_len = bh->b_size - offset;
-               dent->de_rec_len = cpu_to_be32(dent->de_rec_len);
-               dent->de_name_len = name_len;
-
-               *dent_out = dent;
-               return 0;
-       }
-
-       do {
-               uint32_t cur_rec_len, cur_name_len;
-
-               cur_rec_len = be32_to_cpu(dent->de_rec_len);
-               cur_name_len = dent->de_name_len;
-
-               if ((!dent->de_inum.no_addr && cur_rec_len >= rec_len) ||
-                   (cur_rec_len >= GFS2_DIRENT_SIZE(cur_name_len) + rec_len)) {
-                       gfs2_trans_add_bh(dip->i_gl, bh, 1);
-
-                       if (dent->de_inum.no_addr) {
-                               new = (struct gfs2_dirent *)((char *)dent +
-                                                           GFS2_DIRENT_SIZE(cur_name_len));
-                               memset(new, 0, sizeof(struct gfs2_dirent));
-
-                               new->de_rec_len = cur_rec_len - GFS2_DIRENT_SIZE(cur_name_len);
-                               new->de_rec_len = cpu_to_be32(new->de_rec_len);
-                               new->de_name_len = name_len;
-
-                               dent->de_rec_len = cur_rec_len - be32_to_cpu(new->de_rec_len);
-                               dent->de_rec_len = cpu_to_be32(dent->de_rec_len);
-
-                               *dent_out = new;
-                               return 0;
-                       }
-
-                       dent->de_name_len = name_len;
-
-                       *dent_out = dent;
-                       return 0;
-               }
-       } while (dirent_next(dip, bh, &dent) == 0);
-
-       return -ENOSPC;
+       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;
 }
 
-/**
- * dirent_fits - See if we can fit a entry in this buffer
- * @dip: The GFS2 inode
- * @bh: The buffer
- * @name_len: The length of the name
- *
- * Returns: 1 if it can fit, 0 otherwise
- */
-
-static int dirent_fits(struct gfs2_inode *dip, struct buffer_head *bh,
-                      int name_len)
+static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
+                                            struct buffer_head *bh,
+                                            const struct qstr *name)
 {
        struct gfs2_dirent *dent;
-       unsigned int rec_len = GFS2_DIRENT_SIZE(name_len);
-       unsigned int entries = 0;
-       int type;
-
-       type = dirent_first(dip, bh, &dent);
-       if (type < 0)
-               return type;
-
-       if (type == IS_LEAF) {
-               struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
-               entries = be16_to_cpu(leaf->lf_entries);
-       } else {
-               struct gfs2_dinode *dinode = (struct gfs2_dinode *)bh->b_data;
-               entries = be32_to_cpu(dinode->di_entries);
-       }
-
-       if (!entries)
-               return 1;
-
-       do {
-               uint32_t cur_rec_len, cur_name_len;
-
-               cur_rec_len = be32_to_cpu(dent->de_rec_len);
-               cur_name_len = dent->de_name_len;
-
-               if ((!dent->de_inum.no_addr && cur_rec_len >= rec_len) ||
-                   (cur_rec_len >= GFS2_DIRENT_SIZE(cur_name_len) + rec_len))
-                       return 1;
-       } while (dirent_next(dip, bh, &dent) == 0);
-
-       return 0;
-}
-
-static int leaf_search(struct gfs2_inode *dip, struct buffer_head *bh,
-                      struct qstr *filename, struct gfs2_dirent **dent_out,
-                      struct gfs2_dirent **dent_prev)
-{
-       uint32_t hash;
-       struct gfs2_dirent *dent, *prev = NULL;
-       unsigned int entries = 0;
-       int type;
-
-       type = dirent_first(dip, bh, &dent);
-       if (type < 0)
-               return type;
-
-       if (type == IS_LEAF) {
-               struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
-               entries = be16_to_cpu(leaf->lf_entries);
-       } else if (type == IS_DINODE) {
-               struct gfs2_dinode *dinode = (struct gfs2_dinode *)bh->b_data;
-               entries = be32_to_cpu(dinode->di_entries);
-       }
-
-       hash = gfs2_disk_hash(filename->name, filename->len);
-
-       do {
-               if (!dent->de_inum.no_addr) {
-                       prev = dent;
-                       continue;
-               }
-
-               if (be32_to_cpu(dent->de_hash) == hash &&
-                   gfs2_filecmp(filename, (char *)(dent + 1),
-                                dent->de_name_len)) {
-                       *dent_out = dent;
-                       if (dent_prev)
-                               *dent_prev = prev;
-
-                       return 0;
-               }
-
-               prev = dent;
-       } while (dirent_next(dip, bh, &dent) == 0);
-
-       return -ENOENT;
+       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,
@@ -698,75 +752,81 @@ static int get_first_leaf(struct gfs2_inode *dip, uint32_t index,
        return error;
 }
 
-static int get_next_leaf(struct gfs2_inode *dip, struct buffer_head *bh_in,
-                        struct buffer_head **bh_out)
+static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
+                                             const struct qstr *name,
+                                             gfs2_dscan_t scan,
+                                             struct buffer_head **pbh)
 {
-       struct gfs2_leaf *leaf;
+       struct buffer_head *bh;
+       struct gfs2_dirent *dent;
+       struct gfs2_inode *ip = inode->u.generic_ip;
        int error;
 
-       leaf = (struct gfs2_leaf *)bh_in->b_data;
-
-       if (!leaf->lf_next)
-               error = -ENOENT;
-       else
-               error = get_leaf(dip, be64_to_cpu(leaf->lf_next), bh_out);
-
-       return 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);
+               }
 
-static int linked_leaf_search(struct gfs2_inode *dip, struct qstr *filename,
-                             struct gfs2_dirent **dent_out,
-                             struct gfs2_dirent **dent_prev,
-                             struct buffer_head **bh_out)
-{
-       struct buffer_head *bh = NULL, *bh_next;
-       uint32_t hsize, index;
-       uint32_t hash;
-       int error;
+               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);
 
-       hsize = 1 << dip->i_di.di_depth;
-       if (hsize * sizeof(uint64_t) != dip->i_di.di_size) {
-               gfs2_consist_inode(dip);
-               return -EIO;
+               return error ? ERR_PTR(error) : NULL;
        }
 
-       /*  Figure out the address of the leaf node.  */
-
-       hash = gfs2_disk_hash(filename->name, filename->len);
-       index = hash >> (32 - dip->i_di.di_depth);
-
-       error = get_first_leaf(dip, index, &bh_next);
+       error = gfs2_meta_inode_buffer(ip, &bh);
        if (error)
-               return error;
-
-       /*  Find the entry  */
-
-       do {
+               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 = bh_next;
-
-               error = leaf_search(dip, bh, filename, dent_out, dent_prev);
-               switch (error) {
-               case 0:
-                       *bh_out = bh;
-                       return 0;
-
-               case -ENOENT:
-                       break;
-
-               default:
-                       brelse(bh);
-                       return error;
-               }
-
-               error = get_next_leaf(dip, bh, &bh_next);
+               bh = NULL;
        }
-       while (!error);
-
-       brelse(bh);
+       *pbh = bh;
+       return dent;
+}
 
-       return error;
+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;
 }
 
 /**
@@ -776,10 +836,12 @@ static int linked_leaf_search(struct gfs2_inode *dip, struct qstr *filename,
  * Returns: 0 on success, error code otherwise
  */
 
-static int dir_make_exhash(struct gfs2_inode *dip)
+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;
@@ -791,24 +853,14 @@ static int dir_make_exhash(struct gfs2_inode *dip)
        if (error)
                return error;
 
-       /*  Allocate a new block for the first leaf node  */
-
-       bn = gfs2_alloc_meta(dip);
-
        /*  Turn over a new leaf  */
 
-       bh = gfs2_meta_new(dip->i_gl, bn);
-       gfs2_trans_add_bh(dip->i_gl, bh, 1);
-       gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
-       gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
-
-       /*  Fill in the leaf structure  */
-
-       leaf = (struct gfs2_leaf *)bh->b_data;
+       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_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
        leaf->lf_entries = cpu_to_be16(dip->i_di.di_entries);
 
        /*  Copy dirents  */
@@ -819,23 +871,28 @@ static int dir_make_exhash(struct gfs2_inode *dip)
        /*  Find last entry  */
 
        x = 0;
-       dirent_first(dip, bh, &dent);
-
-       do {
-               if (!dent->de_inum.no_addr)
-                       continue;
-               if (++x == dip->i_di.di_entries)
-                       break;
+       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);
        }
-       while (dirent_next(dip, bh, &dent) == 0);
 
        /*  Adjust the last dirent's record length
           (Remember that dent still points to the last entry.)  */
 
-       dent->de_rec_len = be32_to_cpu(dent->de_rec_len) +
+       dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
                sizeof(struct gfs2_dinode) -
-               sizeof(struct gfs2_leaf);
-       dent->de_rec_len = cpu_to_be32(dent->de_rec_len);
+               sizeof(struct gfs2_leaf));
 
        brelse(bh);
 
@@ -874,48 +931,48 @@ static int dir_make_exhash(struct gfs2_inode *dip)
  * Returns: 0 on success, error code on failure
  */
 
-static int dir_split_leaf(struct gfs2_inode *dip, uint32_t index,
-                         uint64_t leaf_no)
+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;
-       uint32_t name_len;
+       uint64_t bn, *lp, leaf_no;
+       uint32_t index;
        int x, moved = 0;
        int error;
 
-       /*  Allocate the new leaf block  */
-
-       bn = gfs2_alloc_meta(dip);
-
-       /*  Get the new leaf block  */
-
-       nbh = gfs2_meta_new(dip->i_gl, bn);
-       gfs2_trans_add_bh(dip->i_gl, nbh, 1);
-       gfs2_metatype_set(nbh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
-       gfs2_buffer_clear_tail(nbh, sizeof(struct gfs2_meta_header));
-
-       nleaf = (struct gfs2_leaf *)nbh->b_data;
-
-       nleaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
+       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)
-               goto fail;
+               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);
 
-       oleaf = (struct gfs2_leaf *)obh->b_data;
+       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;
@@ -926,24 +983,13 @@ static int dir_split_leaf(struct gfs2_inode *dip, uint32_t index,
        /* Change the pointers.
           Don't bother distinguishing stuffed from non-stuffed.
           This code is complicated enough already. */
-
-       lp = kcalloc(half_len, sizeof(uint64_t), GFP_KERNEL | __GFP_NOFAIL);
-
-       error = gfs2_dir_read_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;
-       }
-
+       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));
+                                   half_len * sizeof(uint64_t));
        if (error != half_len * sizeof(uint64_t)) {
                if (error >= 0)
                        error = -EIO;
@@ -953,11 +999,9 @@ static int dir_split_leaf(struct gfs2_inode *dip, uint32_t index,
        kfree(lp);
 
        /*  Compute the divider  */
-
        divider = (start + half_len) << (32 - dip->i_di.di_depth);
 
        /*  Copy the entries  */
-
        dirent_first(dip, obh, &dent);
 
        do {
@@ -967,48 +1011,37 @@ static int dir_split_leaf(struct gfs2_inode *dip, uint32_t index,
 
                if (dent->de_inum.no_addr &&
                    be32_to_cpu(dent->de_hash) < divider) {
-                       name_len = dent->de_name_len;
-
-                       gfs2_dirent_alloc(dip, nbh, name_len, &new);
+                       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_hash = dent->de_hash; /* No endian worries */
                        new->de_type = dent->de_type; /* No endian worries */
-                       memcpy((char *)(new + 1), (char *)(dent + 1),
-                              name_len);
-
-                       nleaf->lf_entries = be16_to_cpu(nleaf->lf_entries)+1;
-                       nleaf->lf_entries = cpu_to_be16(nleaf->lf_entries);
+                       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 = be16_to_cpu(oleaf->lf_entries)-1;
-                       oleaf->lf_entries = cpu_to_be16(oleaf->lf_entries);
+                       oleaf->lf_entries = cpu_to_be16(be16_to_cpu(oleaf->lf_entries)-1);
 
                        if (!prev)
                                prev = dent;
 
                        moved = 1;
-               } else
+               } else {
                        prev = dent;
-
+               }
                dent = next;
-       }
-       while (dent);
-
-       /* If none of the entries got moved into the new leaf,
-          artificially fill in the first entry. */
-
-       if (!moved) {
-               gfs2_dirent_alloc(dip, nbh, 0, &new);
-               new->de_inum.no_addr = 0;
-       }
+       } while (dent);
 
-       oleaf->lf_depth = be16_to_cpu(oleaf->lf_depth) + 1;
-       oleaf->lf_depth = cpu_to_be16(oleaf->lf_depth);
-       nleaf->lf_depth = oleaf->lf_depth;
+       oleaf->lf_depth = nleaf->lf_depth;
 
        error = gfs2_meta_inode_buffer(dip, &dibh);
        if (!gfs2_assert_withdraw(dip->i_sbd, !error)) {
@@ -1022,13 +1055,11 @@ static int dir_split_leaf(struct gfs2_inode *dip, uint32_t index,
 
        return error;
 
- fail_lpfree:
+fail_lpfree:
        kfree(lp);
 
- fail_brelse:
+fail_brelse:
        brelse(obh);
-
- fail:
        brelse(nbh);
        return error;
 }
@@ -1125,20 +1156,18 @@ static int compare_dents(const void *a, const void *b)
        int ret = 0;
 
        dent_a = *(struct gfs2_dirent **)a;
-       hash_a = dent_a->de_hash;
-       hash_a = be32_to_cpu(hash_a);
+       hash_a = be32_to_cpu(dent_a->de_hash);
 
        dent_b = *(struct gfs2_dirent **)b;
-       hash_b = dent_b->de_hash;
-       hash_b = be32_to_cpu(hash_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 = dent_a->de_name_len;
-               unsigned int len_b = dent_b->de_name_len;
+               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;
@@ -1173,10 +1202,10 @@ static int compare_dents(const void *a, const void *b)
 
 static int do_filldir_main(struct gfs2_inode *dip, uint64_t *offset,
                           void *opaque, gfs2_filldir_t filldir,
-                          struct gfs2_dirent **darr, uint32_t entries,
+                          const struct gfs2_dirent **darr, uint32_t entries,
                           int *copied)
 {
-       struct gfs2_dirent *dent, *dent_next;
+       const struct gfs2_dirent *dent, *dent_next;
        struct gfs2_inum inum;
        uint64_t off, off_next;
        unsigned int x, y;
@@ -1217,9 +1246,9 @@ static int do_filldir_main(struct gfs2_inode *dip, uint64_t *offset,
                gfs2_inum_in(&inum, (char *)&dent->de_inum);
 
                error = filldir(opaque, (char *)(dent + 1),
-                               dent->de_name_len,
+                               be16_to_cpu(dent->de_name_len),
                                off, &inum,
-                               dent->de_type);
+                               be16_to_cpu(dent->de_type));
                if (error)
                        return 1;
 
@@ -1230,685 +1259,167 @@ static int do_filldir_main(struct gfs2_inode *dip, uint64_t *offset,
           do_filldir fxn, we get the next entry instead of the last one in the
           current leaf */
 
-       (*offset)++;
-
-       return 0;
-}
-
-/**
- * do_filldir_single - Read directory entries out of a single block
- * @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
- * @bh: the block
- * @entries: the number of entries in the block
- * @copied: pointer to int that's non-zero if a entry has been copied out
- *
- * Returns: errno, >0 on exception from filldir
- */
-
-static int do_filldir_single(struct gfs2_inode *dip, uint64_t *offset,
-                            void *opaque, gfs2_filldir_t filldir,
-                            struct buffer_head *bh, uint32_t entries,
-                            int *copied)
-{
-       struct gfs2_dirent **darr;
-       struct gfs2_dirent *de;
-       unsigned int e = 0;
-       int error;
-
-       if (!entries)
-               return 0;
-
-       darr = kcalloc(entries, sizeof(struct gfs2_dirent *), GFP_KERNEL);
-       if (!darr)
-               return -ENOMEM;
-
-       dirent_first(dip, bh, &de);
-       do {
-               if (!de->de_inum.no_addr)
-                       continue;
-               if (e >= entries) {
-                       gfs2_consist_inode(dip);
-                       error = -EIO;
-                       goto out;
-               }
-               darr[e++] = de;
-       }
-       while (dirent_next(dip, bh, &de) == 0);
-
-       if (e != entries) {
-               gfs2_consist_inode(dip);
-               error = -EIO;
-               goto out;
-       }
-
-       error = do_filldir_main(dip, offset, opaque, filldir, darr,
-                               entries, copied);
-
- out:
-       kfree(darr);
-
-       return error;
-}
-
-/**
- * do_filldir_multi - Read directory entries out of a linked leaf list
- * @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
- * @bh: the first leaf in the list
- * @copied: pointer to int that's non-zero if a entry has been copied out
- *
- * Returns: errno, >0 on exception from filldir
- */
-
-static int do_filldir_multi(struct gfs2_inode *dip, uint64_t *offset,
-                           void *opaque, gfs2_filldir_t filldir,
-                           struct buffer_head *bh, int *copied)
-{
-       struct buffer_head **larr = NULL;
-       struct gfs2_dirent **darr;
-       struct gfs2_leaf *leaf;
-       struct buffer_head *tmp_bh;
-       struct gfs2_dirent *de;
-       unsigned int entries, e = 0;
-       unsigned int leaves = 0, l = 0;
-       unsigned int x;
-       uint64_t ln;
-       int error = 0;
-
-       /*  Count leaves and entries  */
-
-       leaf = (struct gfs2_leaf *)bh->b_data;
-       entries = be16_to_cpu(leaf->lf_entries);
-       ln = leaf->lf_next;
-
-       while (ln) {
-               ln = be64_to_cpu(ln);
-
-               error = get_leaf(dip, ln, &tmp_bh);
-               if (error)
-                       return error;
-
-               leaf = (struct gfs2_leaf *)tmp_bh->b_data;
-               if (leaf->lf_entries) {
-                       entries += be16_to_cpu(leaf->lf_entries);
-                       leaves++;
-               }
-               ln = leaf->lf_next;
-
-               brelse(tmp_bh);
-       }
-
-       if (!entries)
-               return 0;
-
-       if (leaves) {
-               larr = kcalloc(leaves, sizeof(struct buffer_head *),GFP_KERNEL);
-               if (!larr)
-                       return -ENOMEM;
-       }
-
-       darr = kcalloc(entries, sizeof(struct gfs2_dirent *), GFP_KERNEL);
-       if (!darr) {
-               kfree(larr);
-               return -ENOMEM;
-       }
-
-       leaf = (struct gfs2_leaf *)bh->b_data;
-       if (leaf->lf_entries) {
-               dirent_first(dip, bh, &de);
-               do {
-                       if (!de->de_inum.no_addr)
-                               continue;
-                       if (e >= entries) {
-                               gfs2_consist_inode(dip);
-                               error = -EIO;
-                               goto out;
-                       }
-                       darr[e++] = de;
-               }
-               while (dirent_next(dip, bh, &de) == 0);
-       }
-       ln = leaf->lf_next;
-
-       while (ln) {
-               ln = be64_to_cpu(ln);
-
-               error = get_leaf(dip, ln, &tmp_bh);
-               if (error)
-                       goto out;
-
-               leaf = (struct gfs2_leaf *)tmp_bh->b_data;
-               if (leaf->lf_entries) {
-                       dirent_first(dip, tmp_bh, &de);
-                       do {
-                               if (!de->de_inum.no_addr)
-                                       continue;
-                               if (e >= entries) {
-                                       gfs2_consist_inode(dip);
-                                       error = -EIO;
-                                       goto out;
-                               }
-                               darr[e++] = de;
-                       }
-                       while (dirent_next(dip, tmp_bh, &de) == 0);
-
-                       larr[l++] = tmp_bh;
-
-                       ln = leaf->lf_next;
-               } else {
-                       ln = leaf->lf_next;
-                       brelse(tmp_bh);
-               }
-       }
-
-       if (gfs2_assert_withdraw(dip->i_sbd, l == leaves)) {
-               error = -EIO;
-               goto out;
-       }
-       if (e != entries) {
-               gfs2_consist_inode(dip);
-               error = -EIO;
-               goto out;
-       }
-
-       error = do_filldir_main(dip, offset, opaque, filldir, darr,
-                               entries, copied);
-
- out:
-       kfree(darr);
-       for (x = 0; x < l; x++)
-               brelse(larr[x]);
-       kfree(larr);
-
-       return error;
-}
-
-/**
- * dir_e_search - Search exhash (leaf) dir for inode matching name
- * @dip: The GFS2 inode
- * @filename: Filename string
- * @inode: If non-NULL, function fills with formal inode # and block address
- * @type: If non-NULL, function fills with DT_... dinode type
- *
- * Returns:
- */
-
-static int dir_e_search(struct gfs2_inode *dip, struct qstr *filename,
-                       struct gfs2_inum *inum, unsigned int *type)
-{
-       struct buffer_head *bh;
-       struct gfs2_dirent *dent;
-       int error;
-
-       error = linked_leaf_search(dip, filename, &dent, NULL, &bh);
-       if (error)
-               return error;
-
-       if (inum)
-               gfs2_inum_in(inum, (char *)&dent->de_inum);
-       if (type)
-               *type = dent->de_type;
-
-       brelse(bh);
-
-       return 0;
-}
-
-static int dir_e_add(struct gfs2_inode *dip, struct qstr *filename,
-                    struct gfs2_inum *inum, unsigned int type)
-{
-       struct buffer_head *bh, *nbh, *dibh;
-       struct gfs2_leaf *leaf, *nleaf;
-       struct gfs2_dirent *dent;
-       uint32_t hsize, index;
-       uint32_t hash;
-       uint64_t leaf_no, bn;
-       int error;
-
- restart:
-       hsize = 1 << dip->i_di.di_depth;
-       if (hsize * sizeof(uint64_t) != dip->i_di.di_size) {
-               gfs2_consist_inode(dip);
-               return -EIO;
-       }
-
-       /*  Figure out the address of the leaf node.  */
-
-       hash = gfs2_disk_hash(filename->name, filename->len);
-       index = hash >> (32 - dip->i_di.di_depth);
-
-       error = get_leaf_nr(dip, index, &leaf_no);
-       if (error)
-               return error;
-
-       /*  Add entry to the leaf  */
-
-       for (;;) {
-               error = get_leaf(dip, leaf_no, &bh);
-               if (error)
-                       return error;
-
-               leaf = (struct gfs2_leaf *)bh->b_data;
-
-               if (gfs2_dirent_alloc(dip, bh, filename->len, &dent)) {
-
-                       if (be16_to_cpu(leaf->lf_depth) < dip->i_di.di_depth) {
-                               /* Can we split the leaf? */
-
-                               brelse(bh);
-
-                               error = dir_split_leaf(dip, index, leaf_no);
-                               if (error)
-                                       return error;
-
-                               goto restart;
-
-                       } else if (dip->i_di.di_depth < GFS2_DIR_MAX_DEPTH) {
-                               /* Can we double the hash table? */
-
-                               brelse(bh);
-
-                               error = dir_double_exhash(dip);
-                               if (error)
-                                       return error;
-
-                               goto restart;
-
-                       } else if (leaf->lf_next) {
-                               /* Can we try the next leaf in the list? */
-                               leaf_no = be64_to_cpu(leaf->lf_next);
-                               brelse(bh);
-                               continue;
-
-                       } else {
-                               /* Create a new leaf and add it to the list. */
-
-                               bn = gfs2_alloc_meta(dip);
-
-                               nbh = gfs2_meta_new(dip->i_gl, bn);
-                               gfs2_trans_add_bh(dip->i_gl, nbh, 1);
-                               gfs2_metatype_set(nbh,
-                                                GFS2_METATYPE_LF,
-                                                GFS2_FORMAT_LF);
-                               gfs2_buffer_clear_tail(nbh,
-                                       sizeof(struct gfs2_meta_header));
-
-                               gfs2_trans_add_bh(dip->i_gl, bh, 1);
-                               leaf->lf_next = cpu_to_be64(bn);
-
-                               nleaf = (struct gfs2_leaf *)nbh->b_data;
-                               nleaf->lf_depth = leaf->lf_depth;
-                               nleaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
-
-                               gfs2_dirent_alloc(dip, nbh, filename->len,
-                                                 &dent);
-
-                               dip->i_di.di_blocks++;
-
-                               brelse(bh);
-
-                               bh = nbh;
-                               leaf = nleaf;
-                       }
-               }
-
-               /* If the gfs2_dirent_alloc() succeeded, it pinned the "bh" */
-
-               gfs2_inum_out(inum, (char *)&dent->de_inum);
-               dent->de_hash = cpu_to_be32(hash);
-               dent->de_type = type;
-               memcpy((char *)(dent + 1), filename->name, filename->len);
-
-               leaf->lf_entries = be16_to_cpu(leaf->lf_entries) + 1;
-               leaf->lf_entries = cpu_to_be16(leaf->lf_entries);
-
-               brelse(bh);
-
-               error = gfs2_meta_inode_buffer(dip, &dibh);
-               if (error)
-                       return error;
-
-               dip->i_di.di_entries++;
-               dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
-
-               gfs2_trans_add_bh(dip->i_gl, dibh, 1);
-               gfs2_dinode_out(&dip->i_di, dibh->b_data);
-               brelse(dibh);
-
-               return 0;
-       }
-
-       return -ENOENT;
-}
-
-static int dir_e_del(struct gfs2_inode *dip, struct qstr *filename)
-{
-       struct buffer_head *bh, *dibh;
-       struct gfs2_dirent *dent, *prev;
-       struct gfs2_leaf *leaf;
-       unsigned int entries;
-       int error;
-
-       error = linked_leaf_search(dip, filename, &dent, &prev, &bh);
-       if (error == -ENOENT) {
-               gfs2_consist_inode(dip);
-               return -EIO;
-       }
-       if (error)
-               return error;
-
-       dirent_del(dip, bh, prev, dent); /* Pins bh */
-
-       leaf = (struct gfs2_leaf *)bh->b_data;
-       entries = be16_to_cpu(leaf->lf_entries);
-       if (!entries)
-               gfs2_consist_inode(dip);
-       entries--;
-       leaf->lf_entries = cpu_to_be16(entries);
-
-       brelse(bh);
-
-       error = gfs2_meta_inode_buffer(dip, &dibh);
-       if (error)
-               return error;
-
-       if (!dip->i_di.di_entries)
-               gfs2_consist_inode(dip);
-       dip->i_di.di_entries--;
-       dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
-
-       gfs2_trans_add_bh(dip->i_gl, dibh, 1);
-       gfs2_dinode_out(&dip->i_di, dibh->b_data);
-       brelse(dibh);
-
-       return 0;
-}
-
-/**
- * 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 gfs2_inode *dip, uint64_t *offset, void *opaque,
-                     gfs2_filldir_t filldir)
-{
-       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 hash, index;
-       uint64_t *lp;
-       int copied = 0;
-       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;
-       }
-
-       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 = get_leaf(dip, be64_to_cpu(lp[lp_offset]), &bh);
-               if (error)
-                       goto out;
-
-               gfs2_leaf_in(&leaf, bh->b_data);
-
-               if (leaf.lf_next)
-                       error = do_filldir_multi(dip, offset, opaque, filldir,
-                                                bh, &copied);
-               else
-                       error = do_filldir_single(dip, offset, opaque, filldir,
-                                                 bh, leaf.lf_entries, &copied);
-
-               brelse(bh);
-
-               if (error) {
-                       if (error > 0)
-                               error = 0;
-                       goto out;
-               }
-
-               len = 1 << (dip->i_di.di_depth - leaf.lf_depth);
-               index = (index & ~(len - 1)) + len;
-       }
-
- out:
-       kfree(lp);
-
-       return error;
-}
-
-static int dir_e_mvino(struct gfs2_inode *dip, struct qstr *filename,
-                      struct gfs2_inum *inum, unsigned int new_type)
-{
-       struct buffer_head *bh, *dibh;
-       struct gfs2_dirent *dent;
-       int error;
-
-       error = linked_leaf_search(dip, filename, &dent, NULL, &bh);
-       if (error == -ENOENT) {
-               gfs2_consist_inode(dip);
-               return -EIO;
-       }
-       if (error)
-               return error;
-
-       gfs2_trans_add_bh(dip->i_gl, bh, 1);
-
-       gfs2_inum_out(inum, (char *)&dent->de_inum);
-       dent->de_type = new_type;
-
-       brelse(bh);
-
-       error = gfs2_meta_inode_buffer(dip, &dibh);
-       if (error)
-               return error;
-
-       dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
-
-       gfs2_trans_add_bh(dip->i_gl, dibh, 1);
-       gfs2_dinode_out(&dip->i_di, dibh->b_data);
-       brelse(dibh);
-
-       return 0;
-}
-
-/**
- * dir_l_search - Search linear (stuffed dinode) dir for inode matching name
- * @dip: The GFS2 inode
- * @filename: Filename string
- * @inode: If non-NULL, function fills with formal inode # and block address
- * @type: If non-NULL, function fills with DT_... dinode type
- *
- * Returns:
- */
-
-static int dir_l_search(struct gfs2_inode *dip, struct qstr *filename,
-                       struct gfs2_inum *inum, unsigned int *type)
-{
-       struct buffer_head *dibh;
-       struct gfs2_dirent *dent;
-       int error;
-
-       if (!gfs2_is_stuffed(dip)) {
-               gfs2_consist_inode(dip);
-               return -EIO;
-       }
-
-       error = gfs2_meta_inode_buffer(dip, &dibh);
-       if (error)
-               return error;
-
-       error = leaf_search(dip, dibh, filename, &dent, NULL);
-       if (!error) {
-               if (inum)
-                       gfs2_inum_in(inum, (char *)&dent->de_inum);
-               if (type)
-                       *type = dent->de_type;
-       }
-
-       brelse(dibh);
-
-       return error;
-}
-
-static int dir_l_add(struct gfs2_inode *dip, struct qstr *filename,
-                    struct gfs2_inum *inum, unsigned int type)
-{
-       struct buffer_head *dibh;
-       struct gfs2_dirent *dent;
-       int error;
-
-       if (!gfs2_is_stuffed(dip)) {
-               gfs2_consist_inode(dip);
-               return -EIO;
-       }
-
-       error = gfs2_meta_inode_buffer(dip, &dibh);
-       if (error)
-               return error;
-
-       if (gfs2_dirent_alloc(dip, dibh, filename->len, &dent)) {
-               brelse(dibh);
-
-               error = dir_make_exhash(dip);
-               if (!error)
-                       error = dir_e_add(dip, filename, inum, type);
-
-               return error;
-       }
-
-       /*  gfs2_dirent_alloc() pins  */
-
-       gfs2_inum_out(inum, (char *)&dent->de_inum);
-       dent->de_hash = gfs2_disk_hash(filename->name, filename->len);
-       dent->de_hash = cpu_to_be32(dent->de_hash);
-       dent->de_type = type;
-       memcpy((char *)(dent + 1), filename->name, filename->len);
-
-       dip->i_di.di_entries++;
-       dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
-
-       gfs2_dinode_out(&dip->i_di, dibh->b_data);
-       brelse(dibh);
-
-       return 0;
-}
-
-static int dir_l_del(struct gfs2_inode *dip, struct qstr *filename)
-{
-       struct buffer_head *dibh;
-       struct gfs2_dirent *dent, *prev;
-       int error;
-
-       if (!gfs2_is_stuffed(dip)) {
-               gfs2_consist_inode(dip);
-               return -EIO;
-       }
-
-       error = gfs2_meta_inode_buffer(dip, &dibh);
-       if (error)
-               return error;
-
-       error = leaf_search(dip, dibh, filename, &dent, &prev);
-       if (error == -ENOENT) {
-               gfs2_consist_inode(dip);
-               error = -EIO;
-               goto out;
-       }
-       if (error)
-               goto out;
+       (*offset)++;
 
-       dirent_del(dip, dibh, prev, dent);
+       return 0;
+}
 
-       /*  dirent_del() pins  */
+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;
 
-       if (!dip->i_di.di_entries)
-               gfs2_consist_inode(dip);
-       dip->i_di.di_entries--;
+       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);
 
-       dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
+       if (!entries)
+               return 0;
 
-       gfs2_dinode_out(&dip->i_di, dibh->b_data);
+       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;
 
- out:
-       brelse(dibh);
+       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;
 }
 
-static int dir_l_read(struct gfs2_inode *dip, uint64_t *offset, void *opaque,
+/**
+ * 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 buffer_head *dibh;
+       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;
+       int error = 0;
+       unsigned depth;
 
-       if (!gfs2_is_stuffed(dip)) {
+       hsize = 1 << dip->i_di.di_depth;
+       if (hsize * sizeof(uint64_t) != dip->i_di.di_size) {
                gfs2_consist_inode(dip);
                return -EIO;
        }
 
-       if (!dip->i_di.di_entries)
-               return 0;
+       hash = gfs2_dir_offset2hash(*offset);
+       index = hash >> (32 - dip->i_di.di_depth);
 
-       error = gfs2_meta_inode_buffer(dip, &dibh);
-       if (error)
-               return error;
+       lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
+       if (!lp)
+               return -ENOMEM;
 
-       error = do_filldir_single(dip, offset,
-                                 opaque, filldir,
-                                 dibh, dip->i_di.di_entries,
-                                 &copied);
-       if (error > 0)
-               error = 0;
+       while (index < hsize) {
+               lp_offset = index & (sdp->sd_hash_ptrs - 1);
+               ht_offset = index - lp_offset;
 
-       brelse(dibh);
+               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;
 }
 
-static int dir_l_mvino(struct gfs2_inode *dip, struct qstr *filename,
-                      struct gfs2_inum *inum, unsigned int new_type)
+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;
-       struct gfs2_dirent *dent;
+       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;
@@ -1918,25 +1429,27 @@ static int dir_l_mvino(struct gfs2_inode *dip, struct qstr *filename,
        if (error)
                return error;
 
-       error = leaf_search(dip, dibh, filename, &dent, NULL);
-       if (error == -ENOENT) {
-               gfs2_consist_inode(dip);
-               error = -EIO;
-               goto out;
+       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)
-               goto out;
-
-       gfs2_trans_add_bh(dip->i_gl, dibh, 1);
-
-       gfs2_inum_out(inum, (char *)&dent->de_inum);
-       dent->de_type = new_type;
 
-       dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
-
-       gfs2_dinode_out(&dip->i_di, dibh->b_data);
+       if (error > 0)
+               error = 0;
 
- out:
        brelse(dibh);
 
        return error;
@@ -1954,17 +1467,69 @@ static int dir_l_mvino(struct gfs2_inode *dip, struct qstr *filename,
  * Returns: errno
  */
 
-int gfs2_dir_search(struct gfs2_inode *dip, struct qstr *filename,
+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;
 
-       if (dip->i_di.di_flags & GFS2_DIF_EXHASH)
-               error = dir_e_search(dip, filename, inum, type);
-       else
-               error = dir_l_search(dip, filename, inum, type);
+       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);
 
-       return error;
+       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;
 }
 
 /**
@@ -1977,19 +1542,71 @@ int gfs2_dir_search(struct gfs2_inode *dip, struct qstr *filename,
  * Returns: 0 on success, error code on failure
  */
 
-int gfs2_dir_add(struct gfs2_inode *dip, struct qstr *filename,
-                struct gfs2_inum *inum, unsigned int type)
+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;
 
-       if (dip->i_di.di_flags & GFS2_DIF_EXHASH)
-               error = dir_e_add(dip, filename, inum, type);
-       else
-               error = dir_l_add(dip, filename, inum, type);
-
+       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
@@ -1998,27 +1615,50 @@ int gfs2_dir_add(struct gfs2_inode *dip, struct qstr *filename,
  * Returns: 0 on success, error code on failure
  */
 
-int gfs2_dir_del(struct gfs2_inode *dip, struct qstr *filename)
+int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name)
 {
+       struct gfs2_dirent *dent, *prev = NULL;
+       struct buffer_head *bh;
        int error;
 
-       if (dip->i_di.di_flags & GFS2_DIF_EXHASH)
-               error = dir_e_del(dip, filename);
-       else
-               error = dir_l_del(dip, filename);
+       /* 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));
+       }
 
-       return error;
-}
+       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);
 
-int gfs2_dir_read(struct gfs2_inode *dip, uint64_t *offset, void *opaque,
-                 gfs2_filldir_t filldir)
-{
-       int error;
+       error = gfs2_meta_inode_buffer(dip, &bh);
+       if (error)
+               return error;
 
-       if (dip->i_di.di_flags & GFS2_DIF_EXHASH)
-               error = dir_e_read(dip, offset, opaque, filldir);
-       else
-               error = dir_l_read(dip, offset, opaque, filldir);
+       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;
 }
@@ -2036,17 +1676,37 @@ int gfs2_dir_read(struct gfs2_inode *dip, uint64_t *offset, void *opaque,
  * Returns: errno
  */
 
-int gfs2_dir_mvino(struct gfs2_inode *dip, struct qstr *filename,
+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;
 
-       if (dip->i_di.di_flags & GFS2_DIF_EXHASH)
-               error = dir_e_mvino(dip, filename, inum, new_type);
-       else
-               error = dir_l_mvino(dip, filename, inum, new_type);
+       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);
 
-       return error;
+       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;
 }
 
 /**
@@ -2062,7 +1722,7 @@ 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;
+       struct gfs2_leaf *leaf;
        uint32_t hsize, len;
        uint32_t ht_offset, lp_offset, ht_offset_cur = -1;
        uint32_t index = 0;
@@ -2101,10 +1761,10 @@ static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data)
                        error = get_leaf(dip, leaf_no, &bh);
                        if (error)
                                goto out;
-                       gfs2_leaf_in(&leaf, bh->b_data);
+                       leaf = (struct gfs2_leaf *)bh->b_data;
                        brelse(bh);
 
-                       len = 1 << (dip->i_di.di_depth - leaf.lf_depth);
+                       len = 1 << (dip->i_di.di_depth - be16_to_cpu(leaf->lf_depth));
 
                        error = lc(dip, index, len, leaf_no, data);
                        if (error)
@@ -2141,10 +1801,10 @@ 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_leaf *tmp_leaf;
        struct gfs2_rgrp_list rlist;
        struct buffer_head *bh, *dibh;
-       uint64_t blk;
+       uint64_t blk, nblk;
        unsigned int rg_blocks = 0, l_blocks = 0;
        char *ht;
        unsigned int x, size = len * sizeof(uint64_t);
@@ -2168,11 +1828,12 @@ static int leaf_dealloc(struct gfs2_inode *dip, uint32_t index, uint32_t len,
 
        /*  Count the number of leaves  */
 
-       for (blk = leaf_no; blk; blk = tmp_leaf.lf_next) {
+       for (blk = leaf_no; blk; blk = nblk) {
                error = get_leaf(dip, blk, &bh);
                if (error)
                        goto out_rlist;
-               gfs2_leaf_in(&tmp_leaf, (bh)->b_data);
+               tmp_leaf = (struct gfs2_leaf *)bh->b_data;
+               nblk = be64_to_cpu(tmp_leaf->lf_next);
                brelse(bh);
 
                gfs2_rlist_add(sdp, &rlist, blk);
@@ -2183,7 +1844,7 @@ static int leaf_dealloc(struct gfs2_inode *dip, uint32_t index, uint32_t len,
 
        for (x = 0; x < rlist.rl_rgrps; x++) {
                struct gfs2_rgrpd *rgd;
-               rgd = get_gl2rgd(rlist.rl_ghs[x].gh_gl);
+               rgd = rlist.rl_ghs[x].gh_gl->gl_object;
                rg_blocks += rgd->rd_ri.ri_length;
        }
 
@@ -2192,16 +1853,17 @@ static int leaf_dealloc(struct gfs2_inode *dip, uint32_t index, uint32_t len,
                goto out_rlist;
 
        error = gfs2_trans_begin(sdp,
-                       rg_blocks + (DIV_RU(size, sdp->sd_jbsize) + 1) +
+                       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 = tmp_leaf.lf_next) {
+       for (blk = leaf_no; blk; blk = nblk) {
                error = get_leaf(dip, blk, &bh);
                if (error)
                        goto out_end_trans;
-               gfs2_leaf_in(&tmp_leaf, bh->b_data);
+               tmp_leaf = (struct gfs2_leaf *)bh->b_data;
+               nblk = be64_to_cpu(tmp_leaf->lf_next);
                brelse(bh);
 
                gfs2_free_meta(dip, blk, 1);
@@ -2277,7 +1939,8 @@ int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
        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);
+               ((struct gfs2_dinode *)bh->b_data)->di_mode =
+                                               cpu_to_be32(S_IFREG);
                brelse(bh);
        }
 
@@ -2290,63 +1953,22 @@ int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
  * 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
- * @alloc_required: set to 1 if an alloc is required, 0 otherwise
  *
- * Returns: errno
+ * Returns: 1 if alloc required, 0 if not, -ve on error
  */
 
-int gfs2_diradd_alloc_required(struct gfs2_inode *dip, struct qstr *filename,
-                              int *alloc_required)
+int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name)
 {
-       struct buffer_head *bh = NULL, *bh_next;
-       uint32_t hsize, hash, index;
-       int error = 0;
-
-       *alloc_required = 0;
-
-       if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
-               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_disk_hash(filename->name, filename->len);
-               index = hash >> (32 - dip->i_di.di_depth);
-
-               error = get_first_leaf(dip, index, &bh_next);
-               if (error)
-                       return error;
-
-               do {
-                       brelse(bh);
-
-                       bh = bh_next;
-
-                       if (dirent_fits(dip, bh, filename->len))
-                               break;
-
-                       error = get_next_leaf(dip, bh, &bh_next);
-                       if (error == -ENOENT) {
-                               *alloc_required = 1;
-                               error = 0;
-                               break;
-                       }
-               }
-               while (!error);
-
-               brelse(bh);
-       } else {
-               error = gfs2_meta_inode_buffer(dip, &bh);
-               if (error)
-                       return error;
-
-               if (!dirent_fits(dip, bh, filename->len))
-                       *alloc_required = 1;
+       struct gfs2_dirent *dent;
+       struct buffer_head *bh;
 
-               brelse(bh);
+       dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
+       if (!dent) {
+               return 1;
        }
-
-       return error;
+       if (IS_ERR(dent))
+               return PTR_ERR(dent);
+       brelse(bh);
+       return 0;
 }