]> git.ipfire.org Git - thirdparty/xfsprogs-dev.git/blobdiff - libxfs/xfs_da_btree.c
xfs: explicitly pass buffer size to xfs_corruption_error
[thirdparty/xfsprogs-dev.git] / libxfs / xfs_da_btree.c
index a31d35380140a6778d7b82dbf2637a8d3e203a0e..467e96dc9e350522b3435fe6cf84991cb3415946 100644 (file)
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
+ * Copyright (c) 2013 Red Hat, Inc.
  * All Rights Reserved.
  *
  * This program is free software; you can redistribute it and/or
  * along with this program; if not, write the Free Software Foundation,
  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  */
-
-#include <xfs.h>
+#include "libxfs_priv.h"
+#include "xfs_fs.h"
+#include "xfs_shared.h"
+#include "xfs_format.h"
+#include "xfs_log_format.h"
+#include "xfs_trans_resv.h"
+#include "xfs_bit.h"
+#include "xfs_mount.h"
+#include "xfs_da_format.h"
+#include "xfs_da_btree.h"
+#include "xfs_dir2.h"
+#include "xfs_dir2_priv.h"
+#include "xfs_inode.h"
+#include "xfs_trans.h"
+#include "xfs_alloc.h"
+#include "xfs_bmap.h"
+#include "xfs_attr_leaf.h"
+#include "xfs_trace.h"
+#include "xfs_cksum.h"
 
 /*
  * xfs_da_btree.c
 /*
  * Routines used for growing the Btree.
  */
-STATIC int xfs_da_root_split(xfs_da_state_t *state,
+STATIC int xfs_da3_root_split(xfs_da_state_t *state,
                                            xfs_da_state_blk_t *existing_root,
                                            xfs_da_state_blk_t *new_child);
-STATIC int xfs_da_node_split(xfs_da_state_t *state,
+STATIC int xfs_da3_node_split(xfs_da_state_t *state,
                                            xfs_da_state_blk_t *existing_blk,
                                            xfs_da_state_blk_t *split_blk,
                                            xfs_da_state_blk_t *blk_to_add,
                                            int treelevel,
                                            int *result);
-STATIC void xfs_da_node_rebalance(xfs_da_state_t *state,
+STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state,
                                         xfs_da_state_blk_t *node_blk_1,
                                         xfs_da_state_blk_t *node_blk_2);
-STATIC void xfs_da_node_add(xfs_da_state_t *state,
+STATIC void xfs_da3_node_add(xfs_da_state_t *state,
                                   xfs_da_state_blk_t *old_node_blk,
                                   xfs_da_state_blk_t *new_node_blk);
 
 /*
  * Routines used for shrinking the Btree.
  */
-STATIC int xfs_da_root_join(xfs_da_state_t *state,
+STATIC int xfs_da3_root_join(xfs_da_state_t *state,
                                           xfs_da_state_blk_t *root_blk);
-STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval);
-STATIC void xfs_da_node_remove(xfs_da_state_t *state,
+STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval);
+STATIC void xfs_da3_node_remove(xfs_da_state_t *state,
                                              xfs_da_state_blk_t *drop_blk);
-STATIC void xfs_da_node_unbalance(xfs_da_state_t *state,
+STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state,
                                         xfs_da_state_blk_t *src_node_blk,
                                         xfs_da_state_blk_t *dst_node_blk);
 
 /*
  * Utility routines.
  */
-STATIC uint    xfs_da_node_lasthash(struct xfs_buf *bp, int *count);
-STATIC int     xfs_da_node_order(struct xfs_buf *node1_bp,
-                                 struct xfs_buf *node2_bp);
-STATIC int     xfs_da_blk_unlink(xfs_da_state_t *state,
+STATIC int     xfs_da3_blk_unlink(xfs_da_state_t *state,
                                  xfs_da_state_blk_t *drop_blk,
                                  xfs_da_state_blk_t *save_blk);
-STATIC void    xfs_da_state_kill_altpath(xfs_da_state_t *state);
 
-static void
-xfs_da_node_verify(
+
+kmem_zone_t *xfs_da_state_zone;        /* anchor for state struct zone */
+
+/*
+ * Allocate a dir-state structure.
+ * We don't put them on the stack since they're large.
+ */
+xfs_da_state_t *
+xfs_da_state_alloc(void)
+{
+       return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
+}
+
+/*
+ * Kill the altpath contents of a da-state structure.
+ */
+STATIC void
+xfs_da_state_kill_altpath(xfs_da_state_t *state)
+{
+       int     i;
+
+       for (i = 0; i < state->altpath.active; i++)
+               state->altpath.blk[i].bp = NULL;
+       state->altpath.active = 0;
+}
+
+/*
+ * Free a da-state structure.
+ */
+void
+xfs_da_state_free(xfs_da_state_t *state)
+{
+       xfs_da_state_kill_altpath(state);
+#ifdef DEBUG
+       memset((char *)state, 0, sizeof(*state));
+#endif /* DEBUG */
+       kmem_zone_free(xfs_da_state_zone, state);
+}
+
+static xfs_failaddr_t
+xfs_da3_node_verify(
        struct xfs_buf          *bp)
 {
        struct xfs_mount        *mp = bp->b_target->bt_mount;
-       struct xfs_da_node_hdr *hdr = bp->b_addr;
-       int                     block_ok = 0;
-
-       block_ok = hdr->info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC);
-       block_ok = block_ok &&
-                       be16_to_cpu(hdr->level) > 0 &&
-                       be16_to_cpu(hdr->count) > 0 ;
-       if (!block_ok) {
-               XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, hdr);
-               xfs_buf_ioerror(bp, EFSCORRUPTED);
+       struct xfs_da_intnode   *hdr = bp->b_addr;
+       struct xfs_da3_icnode_hdr ichdr;
+       const struct xfs_dir_ops *ops;
+
+       ops = xfs_dir_get_ops(mp, NULL);
+
+       ops->node_hdr_from_disk(&ichdr, hdr);
+
+       if (xfs_sb_version_hascrc(&mp->m_sb)) {
+               struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
+
+               if (ichdr.magic != XFS_DA3_NODE_MAGIC)
+                       return __this_address;
+
+               if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
+                       return __this_address;
+               if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
+                       return __this_address;
+               if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn)))
+                       return __this_address;
+       } else {
+               if (ichdr.magic != XFS_DA_NODE_MAGIC)
+                       return __this_address;
        }
+       if (ichdr.level == 0)
+               return __this_address;
+       if (ichdr.level > XFS_DA_NODE_MAXDEPTH)
+               return __this_address;
+       if (ichdr.count == 0)
+               return __this_address;
+
+       /*
+        * we don't know if the node is for and attribute or directory tree,
+        * so only fail if the count is outside both bounds
+        */
+       if (ichdr.count > mp->m_dir_geo->node_ents &&
+           ichdr.count > mp->m_attr_geo->node_ents)
+               return __this_address;
 
+       /* XXX: hash order check? */
+
+       return NULL;
 }
 
 static void
-xfs_da_node_write_verify(
+xfs_da3_node_write_verify(
        struct xfs_buf  *bp)
 {
-       xfs_da_node_verify(bp);
+       struct xfs_mount        *mp = bp->b_target->bt_mount;
+       struct xfs_buf_log_item *bip = bp->b_log_item;
+       struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
+       xfs_failaddr_t          fa;
+
+       fa = xfs_da3_node_verify(bp);
+       if (fa) {
+               xfs_verifier_error(bp, -EFSCORRUPTED, fa);
+               return;
+       }
+
+       if (!xfs_sb_version_hascrc(&mp->m_sb))
+               return;
+
+       if (bip)
+               hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
+
+       xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
 }
 
 /*
@@ -103,40 +203,74 @@ xfs_da_node_write_verify(
  * format of the block being read.
  */
 static void
-xfs_da_node_read_verify(
+xfs_da3_node_read_verify(
        struct xfs_buf          *bp)
 {
-       struct xfs_mount        *mp = bp->b_target->bt_mount;
        struct xfs_da_blkinfo   *info = bp->b_addr;
+       xfs_failaddr_t          fa;
 
        switch (be16_to_cpu(info->magic)) {
+               case XFS_DA3_NODE_MAGIC:
+                       if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
+                               xfs_verifier_error(bp, -EFSBADCRC,
+                                               __this_address);
+                               break;
+                       }
+                       /* fall through */
                case XFS_DA_NODE_MAGIC:
-                       xfs_da_node_verify(bp);
-                       break;
+                       fa = xfs_da3_node_verify(bp);
+                       if (fa)
+                               xfs_verifier_error(bp, -EFSCORRUPTED, fa);
+                       return;
                case XFS_ATTR_LEAF_MAGIC:
-                       bp->b_ops = &xfs_attr_leaf_buf_ops;
+               case XFS_ATTR3_LEAF_MAGIC:
+                       bp->b_ops = &xfs_attr3_leaf_buf_ops;
                        bp->b_ops->verify_read(bp);
                        return;
                case XFS_DIR2_LEAFN_MAGIC:
-                       bp->b_ops = &xfs_dir2_leafn_buf_ops;
+               case XFS_DIR3_LEAFN_MAGIC:
+                       bp->b_ops = &xfs_dir3_leafn_buf_ops;
                        bp->b_ops->verify_read(bp);
                        return;
                default:
-                       XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW,
-                                            mp, info);
-                       xfs_buf_ioerror(bp, EFSCORRUPTED);
+                       xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
                        break;
        }
 }
 
-const struct xfs_buf_ops xfs_da_node_buf_ops = {
-       .verify_read = xfs_da_node_read_verify,
-       .verify_write = xfs_da_node_write_verify,
-};
+/* Verify the structure of a da3 block. */
+static xfs_failaddr_t
+xfs_da3_node_verify_struct(
+       struct xfs_buf          *bp)
+{
+       struct xfs_da_blkinfo   *info = bp->b_addr;
+
+       switch (be16_to_cpu(info->magic)) {
+       case XFS_DA3_NODE_MAGIC:
+       case XFS_DA_NODE_MAGIC:
+               return xfs_da3_node_verify(bp);
+       case XFS_ATTR_LEAF_MAGIC:
+       case XFS_ATTR3_LEAF_MAGIC:
+               bp->b_ops = &xfs_attr3_leaf_buf_ops;
+               return bp->b_ops->verify_struct(bp);
+       case XFS_DIR2_LEAFN_MAGIC:
+       case XFS_DIR3_LEAFN_MAGIC:
+               bp->b_ops = &xfs_dir3_leafn_buf_ops;
+               return bp->b_ops->verify_struct(bp);
+       default:
+               return __this_address;
+       }
+}
 
+const struct xfs_buf_ops xfs_da3_node_buf_ops = {
+       .name = "xfs_da3_node",
+       .verify_read = xfs_da3_node_read_verify,
+       .verify_write = xfs_da3_node_write_verify,
+       .verify_struct = xfs_da3_node_verify_struct,
+};
 
 int
-xfs_da_node_read(
+xfs_da3_node_read(
        struct xfs_trans        *tp,
        struct xfs_inode        *dp,
        xfs_dablk_t             bno,
@@ -144,8 +278,37 @@ xfs_da_node_read(
        struct xfs_buf          **bpp,
        int                     which_fork)
 {
-       return xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
-                                       which_fork, &xfs_da_node_buf_ops);
+       int                     err;
+
+       err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
+                                       which_fork, &xfs_da3_node_buf_ops);
+       if (!err && tp && *bpp) {
+               struct xfs_da_blkinfo   *info = (*bpp)->b_addr;
+               int                     type;
+
+               switch (be16_to_cpu(info->magic)) {
+               case XFS_DA_NODE_MAGIC:
+               case XFS_DA3_NODE_MAGIC:
+                       type = XFS_BLFT_DA_NODE_BUF;
+                       break;
+               case XFS_ATTR_LEAF_MAGIC:
+               case XFS_ATTR3_LEAF_MAGIC:
+                       type = XFS_BLFT_ATTR_LEAF_BUF;
+                       break;
+               case XFS_DIR2_LEAFN_MAGIC:
+               case XFS_DIR3_LEAFN_MAGIC:
+                       type = XFS_BLFT_DIR_LEAFN_BUF;
+                       break;
+               default:
+                       XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW,
+                                       tp->t_mountp, info, sizeof(*info));
+                       xfs_trans_brelse(tp, *bpp);
+                       *bpp = NULL;
+                       return -EFSCORRUPTED;
+               }
+               xfs_trans_buf_set_type(tp, *bpp, type);
+       }
+       return err;
 }
 
 /*========================================================================
@@ -156,35 +319,50 @@ xfs_da_node_read(
  * Create the initial contents of an intermediate node.
  */
 int
-xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
-                                struct xfs_buf **bpp, int whichfork)
+xfs_da3_node_create(
+       struct xfs_da_args      *args,
+       xfs_dablk_t             blkno,
+       int                     level,
+       struct xfs_buf          **bpp,
+       int                     whichfork)
 {
-       xfs_da_intnode_t *node;
-       struct xfs_buf *bp;
-       int error;
-       xfs_trans_t *tp;
+       struct xfs_da_intnode   *node;
+       struct xfs_trans        *tp = args->trans;
+       struct xfs_mount        *mp = tp->t_mountp;
+       struct xfs_da3_icnode_hdr ichdr = {0};
+       struct xfs_buf          *bp;
+       int                     error;
+       struct xfs_inode        *dp = args->dp;
 
        trace_xfs_da_node_create(args);
+       ASSERT(level <= XFS_DA_NODE_MAXDEPTH);
 
-       tp = args->trans;
-       error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork);
+       error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, whichfork);
        if (error)
-               return(error);
-       ASSERT(bp != NULL);
+               return error;
+       bp->b_ops = &xfs_da3_node_buf_ops;
+       xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
        node = bp->b_addr;
-       node->hdr.info.forw = 0;
-       node->hdr.info.back = 0;
-       node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC);
-       node->hdr.info.pad = 0;
-       node->hdr.count = 0;
-       node->hdr.level = cpu_to_be16(level);
 
+       if (xfs_sb_version_hascrc(&mp->m_sb)) {
+               struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
+
+               memset(hdr3, 0, sizeof(struct xfs_da3_node_hdr));
+               ichdr.magic = XFS_DA3_NODE_MAGIC;
+               hdr3->info.blkno = cpu_to_be64(bp->b_bn);
+               hdr3->info.owner = cpu_to_be64(args->dp->i_ino);
+               uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid);
+       } else {
+               ichdr.magic = XFS_DA_NODE_MAGIC;
+       }
+       ichdr.level = level;
+
+       dp->d_ops->node_hdr_to_disk(node, &ichdr);
        xfs_trans_log_buf(tp, bp,
-               XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
+               XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
 
-       bp->b_ops = &xfs_da_node_buf_ops;
        *bpp = bp;
-       return(0);
+       return 0;
 }
 
 /*
@@ -192,12 +370,17 @@ xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
  * intermediate nodes, rebalance, etc.
  */
 int                                                    /* error */
-xfs_da_split(xfs_da_state_t *state)
+xfs_da3_split(
+       struct xfs_da_state     *state)
 {
-       xfs_da_state_blk_t *oldblk, *newblk, *addblk;
-       xfs_da_intnode_t *node;
-       struct xfs_buf *bp;
-       int max, action, error, i;
+       struct xfs_da_state_blk *oldblk;
+       struct xfs_da_state_blk *newblk;
+       struct xfs_da_state_blk *addblk;
+       struct xfs_da_intnode   *node;
+       int                     max;
+       int                     action = 0;
+       int                     error;
+       int                     i;
 
        trace_xfs_da_split(state->args);
 
@@ -225,31 +408,33 @@ xfs_da_split(xfs_da_state_t *state)
                 */
                switch (oldblk->magic) {
                case XFS_ATTR_LEAF_MAGIC:
-                       error = xfs_attr_leaf_split(state, oldblk, newblk);
-                       if ((error != 0) && (error != ENOSPC)) {
-                               return(error);  /* GROT: attr is inconsistent */
+                       error = xfs_attr3_leaf_split(state, oldblk, newblk);
+                       if ((error != 0) && (error != -ENOSPC)) {
+                               return error;   /* GROT: attr is inconsistent */
                        }
                        if (!error) {
                                addblk = newblk;
                                break;
                        }
                        /*
-                        * Entry wouldn't fit, split the leaf again.
+                        * Entry wouldn't fit, split the leaf again. The new
+                        * extrablk will be consumed by xfs_da3_node_split if
+                        * the node is split.
                         */
                        state->extravalid = 1;
                        if (state->inleaf) {
                                state->extraafter = 0;  /* before newblk */
                                trace_xfs_attr_leaf_split_before(state->args);
-                               error = xfs_attr_leaf_split(state, oldblk,
+                               error = xfs_attr3_leaf_split(state, oldblk,
                                                            &state->extrablk);
                        } else {
                                state->extraafter = 1;  /* after newblk */
                                trace_xfs_attr_leaf_split_after(state->args);
-                               error = xfs_attr_leaf_split(state, newblk,
+                               error = xfs_attr3_leaf_split(state, newblk,
                                                            &state->extrablk);
                        }
                        if (error)
-                               return(error);  /* GROT: attr inconsistent */
+                               return error;   /* GROT: attr inconsistent */
                        addblk = newblk;
                        break;
                case XFS_DIR2_LEAFN_MAGIC:
@@ -259,11 +444,11 @@ xfs_da_split(xfs_da_state_t *state)
                        addblk = newblk;
                        break;
                case XFS_DA_NODE_MAGIC:
-                       error = xfs_da_node_split(state, oldblk, newblk, addblk,
+                       error = xfs_da3_node_split(state, oldblk, newblk, addblk,
                                                         max - i, &action);
                        addblk->bp = NULL;
                        if (error)
-                               return(error);  /* GROT: dir is inconsistent */
+                               return error;   /* GROT: dir is inconsistent */
                        /*
                         * Record the newly split block for the next time thru?
                         */
@@ -277,59 +462,61 @@ xfs_da_split(xfs_da_state_t *state)
                /*
                 * Update the btree to show the new hashval for this child.
                 */
-               xfs_da_fixhashpath(state, &state->path);
+               xfs_da3_fixhashpath(state, &state->path);
        }
        if (!addblk)
-               return(0);
+               return 0;
+
+       /*
+        * xfs_da3_node_split() should have consumed any extra blocks we added
+        * during a double leaf split in the attr fork. This is guaranteed as
+        * we can't be here if the attr fork only has a single leaf block.
+        */
+       ASSERT(state->extravalid == 0 ||
+              state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
 
        /*
         * Split the root node.
         */
        ASSERT(state->path.active == 0);
        oldblk = &state->path.blk[0];
-       error = xfs_da_root_split(state, oldblk, addblk);
+       error = xfs_da3_root_split(state, oldblk, addblk);
        if (error) {
                addblk->bp = NULL;
-               return(error);  /* GROT: dir is inconsistent */
+               return error;   /* GROT: dir is inconsistent */
        }
 
        /*
-        * Update pointers to the node which used to be block 0 and
-        * just got bumped because of the addition of a new root node.
-        * There might be three blocks involved if a double split occurred,
-        * and the original block 0 could be at any position in the list.
+        * Update pointers to the node which used to be block 0 and just got
+        * bumped because of the addition of a new root node.  Note that the
+        * original block 0 could be at any position in the list of blocks in
+        * the tree.
+        *
+        * Note: the magic numbers and sibling pointers are in the same physical
+        * place for both v2 and v3 headers (by design). Hence it doesn't matter
+        * which version of the xfs_da_intnode structure we use here as the
+        * result will be the same using either structure.
         */
-
        node = oldblk->bp->b_addr;
        if (node->hdr.info.forw) {
-               if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
-                       bp = addblk->bp;
-               } else {
-                       ASSERT(state->extravalid);
-                       bp = state->extrablk.bp;
-               }
-               node = bp->b_addr;
+               ASSERT(be32_to_cpu(node->hdr.info.forw) == addblk->blkno);
+               node = addblk->bp->b_addr;
                node->hdr.info.back = cpu_to_be32(oldblk->blkno);
-               xfs_trans_log_buf(state->args->trans, bp,
-                   XFS_DA_LOGRANGE(node, &node->hdr.info,
-                   sizeof(node->hdr.info)));
+               xfs_trans_log_buf(state->args->trans, addblk->bp,
+                                 XFS_DA_LOGRANGE(node, &node->hdr.info,
+                                 sizeof(node->hdr.info)));
        }
        node = oldblk->bp->b_addr;
        if (node->hdr.info.back) {
-               if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
-                       bp = addblk->bp;
-               } else {
-                       ASSERT(state->extravalid);
-                       bp = state->extrablk.bp;
-               }
-               node = bp->b_addr;
+               ASSERT(be32_to_cpu(node->hdr.info.back) == addblk->blkno);
+               node = addblk->bp->b_addr;
                node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
-               xfs_trans_log_buf(state->args->trans, bp,
-                   XFS_DA_LOGRANGE(node, &node->hdr.info,
-                   sizeof(node->hdr.info)));
+               xfs_trans_log_buf(state->args->trans, addblk->bp,
+                                 XFS_DA_LOGRANGE(node, &node->hdr.info,
+                                 sizeof(node->hdr.info)));
        }
        addblk->bp = NULL;
-       return(0);
+       return 0;
 }
 
 /*
@@ -338,18 +525,24 @@ xfs_da_split(xfs_da_state_t *state)
  * the EOF, extending the inode in process.
  */
 STATIC int                                             /* error */
-xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
-                                xfs_da_state_blk_t *blk2)
+xfs_da3_root_split(
+       struct xfs_da_state     *state,
+       struct xfs_da_state_blk *blk1,
+       struct xfs_da_state_blk *blk2)
 {
-       xfs_da_intnode_t *node, *oldroot;
-       xfs_da_args_t *args;
-       xfs_dablk_t blkno;
-       struct xfs_buf *bp;
-       int error, size;
-       xfs_inode_t *dp;
-       xfs_trans_t *tp;
-       xfs_mount_t *mp;
-       xfs_dir2_leaf_t *leaf;
+       struct xfs_da_intnode   *node;
+       struct xfs_da_intnode   *oldroot;
+       struct xfs_da_node_entry *btree;
+       struct xfs_da3_icnode_hdr nodehdr;
+       struct xfs_da_args      *args;
+       struct xfs_buf          *bp;
+       struct xfs_inode        *dp;
+       struct xfs_trans        *tp;
+       struct xfs_dir2_leaf    *leaf;
+       xfs_dablk_t             blkno;
+       int                     level;
+       int                     error;
+       int                     size;
 
        trace_xfs_da_root_split(state->args);
 
@@ -358,85 +551,131 @@ xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
         * to a free space somewhere.
         */
        args = state->args;
-       ASSERT(args != NULL);
        error = xfs_da_grow_inode(args, &blkno);
        if (error)
-               return(error);
+               return error;
+
        dp = args->dp;
        tp = args->trans;
-       mp = state->mp;
        error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
        if (error)
-               return(error);
-       ASSERT(bp != NULL);
+               return error;
        node = bp->b_addr;
        oldroot = blk1->bp->b_addr;
-       if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) {
-               size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] -
-                            (char *)oldroot);
+       if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
+           oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
+               struct xfs_da3_icnode_hdr icnodehdr;
+
+               dp->d_ops->node_hdr_from_disk(&icnodehdr, oldroot);
+               btree = dp->d_ops->node_tree_p(oldroot);
+               size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot);
+               level = icnodehdr.level;
+
+               /*
+                * we are about to copy oldroot to bp, so set up the type
+                * of bp while we know exactly what it will be.
+                */
+               xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
        } else {
-               ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
+               struct xfs_dir3_icleaf_hdr leafhdr;
+               struct xfs_dir2_leaf_entry *ents;
+
                leaf = (xfs_dir2_leaf_t *)oldroot;
-               size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] -
-                            (char *)leaf);
+               dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
+               ents = dp->d_ops->leaf_ents_p(leaf);
+
+               ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
+                      leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
+               size = (int)((char *)&ents[leafhdr.count] - (char *)leaf);
+               level = 0;
+
+               /*
+                * we are about to copy oldroot to bp, so set up the type
+                * of bp while we know exactly what it will be.
+                */
+               xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
        }
+
+       /*
+        * we can copy most of the information in the node from one block to
+        * another, but for CRC enabled headers we have to make sure that the
+        * block specific identifiers are kept intact. We update the buffer
+        * directly for this.
+        */
        memcpy(node, oldroot, size);
+       if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
+           oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
+               struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node;
+
+               node3->hdr.info.blkno = cpu_to_be64(bp->b_bn);
+       }
        xfs_trans_log_buf(tp, bp, 0, size - 1);
 
        bp->b_ops = blk1->bp->b_ops;
+       xfs_trans_buf_copy_type(bp, blk1->bp);
        blk1->bp = bp;
        blk1->blkno = blkno;
 
        /*
         * Set up the new root node.
         */
-       error = xfs_da_node_create(args,
-               (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0,
-               be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork);
+       error = xfs_da3_node_create(args,
+               (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0,
+               level + 1, &bp, args->whichfork);
        if (error)
-               return(error);
+               return error;
+
        node = bp->b_addr;
-       node->btree[0].hashval = cpu_to_be32(blk1->hashval);
-       node->btree[0].before = cpu_to_be32(blk1->blkno);
-       node->btree[1].hashval = cpu_to_be32(blk2->hashval);
-       node->btree[1].before = cpu_to_be32(blk2->blkno);
-       node->hdr.count = cpu_to_be16(2);
+       dp->d_ops->node_hdr_from_disk(&nodehdr, node);
+       btree = dp->d_ops->node_tree_p(node);
+       btree[0].hashval = cpu_to_be32(blk1->hashval);
+       btree[0].before = cpu_to_be32(blk1->blkno);
+       btree[1].hashval = cpu_to_be32(blk2->hashval);
+       btree[1].before = cpu_to_be32(blk2->blkno);
+       nodehdr.count = 2;
+       dp->d_ops->node_hdr_to_disk(node, &nodehdr);
 
 #ifdef DEBUG
-       if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) {
-               ASSERT(blk1->blkno >= mp->m_dirleafblk &&
-                      blk1->blkno < mp->m_dirfreeblk);
-               ASSERT(blk2->blkno >= mp->m_dirleafblk &&
-                      blk2->blkno < mp->m_dirfreeblk);
+       if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
+           oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
+               ASSERT(blk1->blkno >= args->geo->leafblk &&
+                      blk1->blkno < args->geo->freeblk);
+               ASSERT(blk2->blkno >= args->geo->leafblk &&
+                      blk2->blkno < args->geo->freeblk);
        }
 #endif
 
        /* Header is already logged by xfs_da_node_create */
        xfs_trans_log_buf(tp, bp,
-               XFS_DA_LOGRANGE(node, node->btree,
-                       sizeof(xfs_da_node_entry_t) * 2));
+               XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
 
-       return(0);
+       return 0;
 }
 
 /*
  * Split the node, rebalance, then add the new entry.
  */
 STATIC int                                             /* error */
-xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
-                                xfs_da_state_blk_t *newblk,
-                                xfs_da_state_blk_t *addblk,
-                                int treelevel, int *result)
+xfs_da3_node_split(
+       struct xfs_da_state     *state,
+       struct xfs_da_state_blk *oldblk,
+       struct xfs_da_state_blk *newblk,
+       struct xfs_da_state_blk *addblk,
+       int                     treelevel,
+       int                     *result)
 {
-       xfs_da_intnode_t *node;
-       xfs_dablk_t blkno;
-       int newcount, error;
-       int useextra;
+       struct xfs_da_intnode   *node;
+       struct xfs_da3_icnode_hdr nodehdr;
+       xfs_dablk_t             blkno;
+       int                     newcount;
+       int                     error;
+       int                     useextra;
+       struct xfs_inode        *dp = state->args->dp;
 
        trace_xfs_da_node_split(state->args);
 
        node = oldblk->bp->b_addr;
-       ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
+       dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 
        /*
         * With V2 dirs the extra block is data or freespace.
@@ -446,25 +685,25 @@ xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
        /*
         * Do we have to split the node?
         */
-       if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) {
+       if (nodehdr.count + newcount > state->args->geo->node_ents) {
                /*
                 * Allocate a new node, add to the doubly linked chain of
                 * nodes, then move some of our excess entries into it.
                 */
                error = xfs_da_grow_inode(state->args, &blkno);
                if (error)
-                       return(error);  /* GROT: dir is inconsistent */
+                       return error;   /* GROT: dir is inconsistent */
 
-               error = xfs_da_node_create(state->args, blkno, treelevel,
+               error = xfs_da3_node_create(state->args, blkno, treelevel,
                                           &newblk->bp, state->args->whichfork);
                if (error)
-                       return(error);  /* GROT: dir is inconsistent */
+                       return error;   /* GROT: dir is inconsistent */
                newblk->blkno = blkno;
                newblk->magic = XFS_DA_NODE_MAGIC;
-               xfs_da_node_rebalance(state, oldblk, newblk);
-               error = xfs_da_blk_link(state, oldblk, newblk);
+               xfs_da3_node_rebalance(state, oldblk, newblk);
+               error = xfs_da3_blk_link(state, oldblk, newblk);
                if (error)
-                       return(error);
+                       return error;
                *result = 1;
        } else {
                *result = 0;
@@ -474,7 +713,7 @@ xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
         * Insert the new entry(s) into the correct block
         * (updating last hashval in the process).
         *
-        * xfs_da_node_add() inserts BEFORE the given index,
+        * xfs_da3_node_add() inserts BEFORE the given index,
         * and as a result of using node_lookup_int() we always
         * point to a valid entry (not after one), but a split
         * operation always results in a new block whose hashvals
@@ -483,27 +722,28 @@ xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
         * If we had double-split op below us, then add the extra block too.
         */
        node = oldblk->bp->b_addr;
-       if (oldblk->index <= be16_to_cpu(node->hdr.count)) {
+       dp->d_ops->node_hdr_from_disk(&nodehdr, node);
+       if (oldblk->index <= nodehdr.count) {
                oldblk->index++;
-               xfs_da_node_add(state, oldblk, addblk);
+               xfs_da3_node_add(state, oldblk, addblk);
                if (useextra) {
                        if (state->extraafter)
                                oldblk->index++;
-                       xfs_da_node_add(state, oldblk, &state->extrablk);
+                       xfs_da3_node_add(state, oldblk, &state->extrablk);
                        state->extravalid = 0;
                }
        } else {
                newblk->index++;
-               xfs_da_node_add(state, newblk, addblk);
+               xfs_da3_node_add(state, newblk, addblk);
                if (useextra) {
                        if (state->extraafter)
                                newblk->index++;
-                       xfs_da_node_add(state, newblk, &state->extrablk);
+                       xfs_da3_node_add(state, newblk, &state->extrablk);
                        state->extravalid = 0;
                }
        }
 
-       return(0);
+       return 0;
 }
 
 /*
@@ -513,33 +753,54 @@ xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
  * NOTE: if blk2 is empty, then it will get the upper half of blk1.
  */
 STATIC void
-xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
-                                    xfs_da_state_blk_t *blk2)
+xfs_da3_node_rebalance(
+       struct xfs_da_state     *state,
+       struct xfs_da_state_blk *blk1,
+       struct xfs_da_state_blk *blk2)
 {
-       xfs_da_intnode_t *node1, *node2, *tmpnode;
-       xfs_da_node_entry_t *btree_s, *btree_d;
-       int count, tmp;
-       xfs_trans_t *tp;
+       struct xfs_da_intnode   *node1;
+       struct xfs_da_intnode   *node2;
+       struct xfs_da_intnode   *tmpnode;
+       struct xfs_da_node_entry *btree1;
+       struct xfs_da_node_entry *btree2;
+       struct xfs_da_node_entry *btree_s;
+       struct xfs_da_node_entry *btree_d;
+       struct xfs_da3_icnode_hdr nodehdr1;
+       struct xfs_da3_icnode_hdr nodehdr2;
+       struct xfs_trans        *tp;
+       int                     count;
+       int                     tmp;
+       int                     swap = 0;
+       struct xfs_inode        *dp = state->args->dp;
 
        trace_xfs_da_node_rebalance(state->args);
 
        node1 = blk1->bp->b_addr;
        node2 = blk2->bp->b_addr;
+       dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
+       dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
+       btree1 = dp->d_ops->node_tree_p(node1);
+       btree2 = dp->d_ops->node_tree_p(node2);
+
        /*
         * Figure out how many entries need to move, and in which direction.
         * Swap the nodes around if that makes it simpler.
         */
-       if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
-           ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) ||
-            (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
-             be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
+       if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
+           ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
+            (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
+                       be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
                tmpnode = node1;
                node1 = node2;
                node2 = tmpnode;
+               dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
+               dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
+               btree1 = dp->d_ops->node_tree_p(node1);
+               btree2 = dp->d_ops->node_tree_p(node2);
+               swap = 1;
        }
-       ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
-       ASSERT(node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
-       count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2;
+
+       count = (nodehdr1.count - nodehdr2.count) / 2;
        if (count == 0)
                return;
        tp = state->args->trans;
@@ -550,10 +811,11 @@ xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
                /*
                 * Move elements in node2 up to make a hole.
                 */
-               if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) {
+               tmp = nodehdr2.count;
+               if (tmp > 0) {
                        tmp *= (uint)sizeof(xfs_da_node_entry_t);
-                       btree_s = &node2->btree[0];
-                       btree_d = &node2->btree[count];
+                       btree_s = &btree2[0];
+                       btree_d = &btree2[count];
                        memmove(btree_d, btree_s, tmp);
                }
 
@@ -561,12 +823,12 @@ xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
                 * Move the req'd B-tree elements from high in node1 to
                 * low in node2.
                 */
-               be16_add_cpu(&node2->hdr.count, count);
+               nodehdr2.count += count;
                tmp = count * (uint)sizeof(xfs_da_node_entry_t);
-               btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count];
-               btree_d = &node2->btree[0];
+               btree_s = &btree1[nodehdr1.count - count];
+               btree_d = &btree2[0];
                memcpy(btree_d, btree_s, tmp);
-               be16_add_cpu(&node1->hdr.count, -count);
+               nodehdr1.count -= count;
        } else {
                /*
                 * Move the req'd B-tree elements from low in node2 to
@@ -574,49 +836,59 @@ xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
                 */
                count = -count;
                tmp = count * (uint)sizeof(xfs_da_node_entry_t);
-               btree_s = &node2->btree[0];
-               btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)];
+               btree_s = &btree2[0];
+               btree_d = &btree1[nodehdr1.count];
                memcpy(btree_d, btree_s, tmp);
-               be16_add_cpu(&node1->hdr.count, count);
+               nodehdr1.count += count;
+
                xfs_trans_log_buf(tp, blk1->bp,
                        XFS_DA_LOGRANGE(node1, btree_d, tmp));
 
                /*
                 * Move elements in node2 down to fill the hole.
                 */
-               tmp  = be16_to_cpu(node2->hdr.count) - count;
+               tmp  = nodehdr2.count - count;
                tmp *= (uint)sizeof(xfs_da_node_entry_t);
-               btree_s = &node2->btree[count];
-               btree_d = &node2->btree[0];
+               btree_s = &btree2[count];
+               btree_d = &btree2[0];
                memmove(btree_d, btree_s, tmp);
-               be16_add_cpu(&node2->hdr.count, -count);
+               nodehdr2.count -= count;
        }
 
        /*
         * Log header of node 1 and all current bits of node 2.
         */
+       dp->d_ops->node_hdr_to_disk(node1, &nodehdr1);
        xfs_trans_log_buf(tp, blk1->bp,
-               XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr)));
+               XFS_DA_LOGRANGE(node1, &node1->hdr, dp->d_ops->node_hdr_size));
+
+       dp->d_ops->node_hdr_to_disk(node2, &nodehdr2);
        xfs_trans_log_buf(tp, blk2->bp,
                XFS_DA_LOGRANGE(node2, &node2->hdr,
-                       sizeof(node2->hdr) +
-                       sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count)));
+                               dp->d_ops->node_hdr_size +
+                               (sizeof(btree2[0]) * nodehdr2.count)));
 
        /*
         * Record the last hashval from each block for upward propagation.
         * (note: don't use the swapped node pointers)
         */
-       node1 = blk1->bp->b_addr;
-       node2 = blk2->bp->b_addr;
-       blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval);
-       blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval);
+       if (swap) {
+               node1 = blk1->bp->b_addr;
+               node2 = blk2->bp->b_addr;
+               dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
+               dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
+               btree1 = dp->d_ops->node_tree_p(node1);
+               btree2 = dp->d_ops->node_tree_p(node2);
+       }
+       blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
+       blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
 
        /*
         * Adjust the expected index for insertion.
         */
-       if (blk1->index >= be16_to_cpu(node1->hdr.count)) {
-               blk2->index = blk1->index - be16_to_cpu(node1->hdr.count);
-               blk1->index = be16_to_cpu(node1->hdr.count) + 1;        /* make it invalid */
+       if (blk1->index >= nodehdr1.count) {
+               blk2->index = blk1->index - nodehdr1.count;
+               blk1->index = nodehdr1.count + 1;       /* make it invalid */
        }
 }
 
@@ -624,44 +896,52 @@ xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
  * Add a new entry to an intermediate node.
  */
 STATIC void
-xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
-                              xfs_da_state_blk_t *newblk)
+xfs_da3_node_add(
+       struct xfs_da_state     *state,
+       struct xfs_da_state_blk *oldblk,
+       struct xfs_da_state_blk *newblk)
 {
-       xfs_da_intnode_t *node;
-       xfs_da_node_entry_t *btree;
-       int tmp;
+       struct xfs_da_intnode   *node;
+       struct xfs_da3_icnode_hdr nodehdr;
+       struct xfs_da_node_entry *btree;
+       int                     tmp;
+       struct xfs_inode        *dp = state->args->dp;
 
        trace_xfs_da_node_add(state->args);
 
        node = oldblk->bp->b_addr;
-       ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
-       ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count)));
+       dp->d_ops->node_hdr_from_disk(&nodehdr, node);
+       btree = dp->d_ops->node_tree_p(node);
+
+       ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
        ASSERT(newblk->blkno != 0);
        if (state->args->whichfork == XFS_DATA_FORK)
-               ASSERT(newblk->blkno >= state->mp->m_dirleafblk &&
-                      newblk->blkno < state->mp->m_dirfreeblk);
+               ASSERT(newblk->blkno >= state->args->geo->leafblk &&
+                      newblk->blkno < state->args->geo->freeblk);
 
        /*
         * We may need to make some room before we insert the new node.
         */
        tmp = 0;
-       btree = &node->btree[ oldblk->index ];
-       if (oldblk->index < be16_to_cpu(node->hdr.count)) {
-               tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree);
-               memmove(btree + 1, btree, tmp);
+       if (oldblk->index < nodehdr.count) {
+               tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
+               memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
        }
-       btree->hashval = cpu_to_be32(newblk->hashval);
-       btree->before = cpu_to_be32(newblk->blkno);
+       btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
+       btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
        xfs_trans_log_buf(state->args->trans, oldblk->bp,
-               XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree)));
-       be16_add_cpu(&node->hdr.count, 1);
+               XFS_DA_LOGRANGE(node, &btree[oldblk->index],
+                               tmp + sizeof(*btree)));
+
+       nodehdr.count += 1;
+       dp->d_ops->node_hdr_to_disk(node, &nodehdr);
        xfs_trans_log_buf(state->args->trans, oldblk->bp,
-               XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
+               XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
 
        /*
         * Copy the last hash value from the oldblk to propagate upwards.
         */
-       oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval);
+       oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
 }
 
 /*========================================================================
@@ -673,14 +953,16 @@ xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
  * possibly deallocating that block, etc...
  */
 int
-xfs_da_join(xfs_da_state_t *state)
+xfs_da3_join(
+       struct xfs_da_state     *state)
 {
-       xfs_da_state_blk_t *drop_blk, *save_blk;
-       int action, error;
+       struct xfs_da_state_blk *drop_blk;
+       struct xfs_da_state_blk *save_blk;
+       int                     action = 0;
+       int                     error;
 
        trace_xfs_da_join(state->args);
 
-       action = 0;
        drop_blk = &state->path.blk[ state->path.active-1 ];
        save_blk = &state->altpath.blk[ state->path.active-1 ];
        ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
@@ -701,12 +983,12 @@ xfs_da_join(xfs_da_state_t *state)
                 */
                switch (drop_blk->magic) {
                case XFS_ATTR_LEAF_MAGIC:
-                       error = xfs_attr_leaf_toosmall(state, &action);
+                       error = xfs_attr3_leaf_toosmall(state, &action);
                        if (error)
-                               return(error);
+                               return error;
                        if (action == 0)
-                               return(0);
-                       xfs_attr_leaf_unbalance(state, drop_blk, save_blk);
+                               return 0;
+                       xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
                        break;
                case XFS_DIR2_LEAFN_MAGIC:
                        error = xfs_dir2_leafn_toosmall(state, &action);
@@ -721,36 +1003,36 @@ xfs_da_join(xfs_da_state_t *state)
                         * Remove the offending node, fixup hashvals,
                         * check for a toosmall neighbor.
                         */
-                       xfs_da_node_remove(state, drop_blk);
-                       xfs_da_fixhashpath(state, &state->path);
-                       error = xfs_da_node_toosmall(state, &action);
+                       xfs_da3_node_remove(state, drop_blk);
+                       xfs_da3_fixhashpath(state, &state->path);
+                       error = xfs_da3_node_toosmall(state, &action);
                        if (error)
-                               return(error);
+                               return error;
                        if (action == 0)
                                return 0;
-                       xfs_da_node_unbalance(state, drop_blk, save_blk);
+                       xfs_da3_node_unbalance(state, drop_blk, save_blk);
                        break;
                }
-               xfs_da_fixhashpath(state, &state->altpath);
-               error = xfs_da_blk_unlink(state, drop_blk, save_blk);
+               xfs_da3_fixhashpath(state, &state->altpath);
+               error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
                xfs_da_state_kill_altpath(state);
                if (error)
-                       return(error);
+                       return error;
                error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
                                                         drop_blk->bp);
                drop_blk->bp = NULL;
                if (error)
-                       return(error);
+                       return error;
        }
        /*
         * We joined all the way to the top.  If it turns out that
         * we only have one entry in the root, make the child block
         * the new root.
         */
-       xfs_da_node_remove(state, drop_blk);
-       xfs_da_fixhashpath(state, &state->path);
-       error = xfs_da_root_join(state, &state->path.blk[0]);
-       return(error);
+       xfs_da3_node_remove(state, drop_blk);
+       xfs_da3_fixhashpath(state, &state->path);
+       error = xfs_da3_root_join(state, &state->path.blk[0]);
+       return error;
 }
 
 #ifdef DEBUG
@@ -761,9 +1043,13 @@ xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
 
        if (level == 1) {
                ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
-                      magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
-       } else
-               ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
+                      magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
+                      magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
+                      magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
+       } else {
+               ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
+                      magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
+       }
        ASSERT(!blkinfo->forw);
        ASSERT(!blkinfo->back);
 }
@@ -776,55 +1062,66 @@ xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
  * the old root to block 0 as the new root node.
  */
 STATIC int
-xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
+xfs_da3_root_join(
+       struct xfs_da_state     *state,
+       struct xfs_da_state_blk *root_blk)
 {
-       xfs_da_intnode_t *oldroot;
-       xfs_da_args_t *args;
-       xfs_dablk_t child;
-       struct xfs_buf *bp;
-       int error;
+       struct xfs_da_intnode   *oldroot;
+       struct xfs_da_args      *args;
+       xfs_dablk_t             child;
+       struct xfs_buf          *bp;
+       struct xfs_da3_icnode_hdr oldroothdr;
+       struct xfs_da_node_entry *btree;
+       int                     error;
+       struct xfs_inode        *dp = state->args->dp;
 
        trace_xfs_da_root_join(state->args);
 
-       args = state->args;
-       ASSERT(args != NULL);
        ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
+
+       args = state->args;
        oldroot = root_blk->bp->b_addr;
-       ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
-       ASSERT(!oldroot->hdr.info.forw);
-       ASSERT(!oldroot->hdr.info.back);
+       dp->d_ops->node_hdr_from_disk(&oldroothdr, oldroot);
+       ASSERT(oldroothdr.forw == 0);
+       ASSERT(oldroothdr.back == 0);
 
        /*
         * If the root has more than one child, then don't do anything.
         */
-       if (be16_to_cpu(oldroot->hdr.count) > 1)
-               return(0);
+       if (oldroothdr.count > 1)
+               return 0;
 
        /*
         * Read in the (only) child block, then copy those bytes into
         * the root block's buffer and free the original child block.
         */
-       child = be32_to_cpu(oldroot->btree[0].before);
+       btree = dp->d_ops->node_tree_p(oldroot);
+       child = be32_to_cpu(btree[0].before);
        ASSERT(child != 0);
-       error = xfs_da_node_read(args->trans, args->dp, child, -1, &bp,
+       error = xfs_da3_node_read(args->trans, dp, child, -1, &bp,
                                             args->whichfork);
        if (error)
-               return(error);
-       ASSERT(bp != NULL);
-       xfs_da_blkinfo_onlychild_validate(bp->b_addr,
-                                       be16_to_cpu(oldroot->hdr.level));
+               return error;
+       xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
 
        /*
         * This could be copying a leaf back into the root block in the case of
         * there only being a single leaf block left in the tree. Hence we have
         * to update the b_ops pointer as well to match the buffer type change
-        * that could occur.
+        * that could occur. For dir3 blocks we also need to update the block
+        * number in the buffer header.
         */
-       memcpy(root_blk->bp->b_addr, bp->b_addr, state->blocksize);
+       memcpy(root_blk->bp->b_addr, bp->b_addr, args->geo->blksize);
        root_blk->bp->b_ops = bp->b_ops;
-       xfs_trans_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
+       xfs_trans_buf_copy_type(root_blk->bp, bp);
+       if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) {
+               struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr;
+               da3->blkno = cpu_to_be64(root_blk->bp->b_bn);
+       }
+       xfs_trans_log_buf(args->trans, root_blk->bp, 0,
+                         args->geo->blksize - 1);
        error = xfs_da_shrink_inode(args, child, bp);
-       return(error);
+       return error;
 }
 
 /*
@@ -837,14 +1134,22 @@ xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
  * If nothing can be done, return 0.
  */
 STATIC int
-xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
+xfs_da3_node_toosmall(
+       struct xfs_da_state     *state,
+       int                     *action)
 {
-       xfs_da_intnode_t *node;
-       xfs_da_state_blk_t *blk;
-       xfs_da_blkinfo_t *info;
-       int count, forward, error, retval, i;
-       xfs_dablk_t blkno;
-       struct xfs_buf *bp;
+       struct xfs_da_intnode   *node;
+       struct xfs_da_state_blk *blk;
+       struct xfs_da_blkinfo   *info;
+       xfs_dablk_t             blkno;
+       struct xfs_buf          *bp;
+       struct xfs_da3_icnode_hdr nodehdr;
+       int                     count;
+       int                     forward;
+       int                     error;
+       int                     retval;
+       int                     i;
+       struct xfs_inode        *dp = state->args->dp;
 
        trace_xfs_da_node_toosmall(state->args);
 
@@ -855,12 +1160,11 @@ xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
         */
        blk = &state->path.blk[ state->path.active-1 ];
        info = blk->bp->b_addr;
-       ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
        node = (xfs_da_intnode_t *)info;
-       count = be16_to_cpu(node->hdr.count);
-       if (count > (state->node_ents >> 1)) {
+       dp->d_ops->node_hdr_from_disk(&nodehdr, node);
+       if (nodehdr.count > (state->args->geo->node_ents >> 1)) {
                *action = 0;    /* blk over 50%, don't try to join */
-               return(0);      /* blk over 50%, don't try to join */
+               return 0;       /* blk over 50%, don't try to join */
        }
 
        /*
@@ -869,23 +1173,23 @@ xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
         * coalesce it with a sibling block.  We choose (arbitrarily)
         * to merge with the forward block unless it is NULL.
         */
-       if (count == 0) {
+       if (nodehdr.count == 0) {
                /*
                 * Make altpath point to the block we want to keep and
                 * path point to the block we want to drop (this one).
                 */
                forward = (info->forw != 0);
                memcpy(&state->altpath, &state->path, sizeof(state->path));
-               error = xfs_da_path_shift(state, &state->altpath, forward,
+               error = xfs_da3_path_shift(state, &state->altpath, forward,
                                                 0, &retval);
                if (error)
-                       return(error);
+                       return error;
                if (retval) {
                        *action = 0;
                } else {
                        *action = 2;
                }
-               return(0);
+               return 0;
        }
 
        /*
@@ -895,35 +1199,35 @@ xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
         * We prefer coalescing with the lower numbered sibling so as
         * to shrink a directory over time.
         */
+       count  = state->args->geo->node_ents;
+       count -= state->args->geo->node_ents >> 2;
+       count -= nodehdr.count;
+
        /* start with smaller blk num */
-       forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
+       forward = nodehdr.forw < nodehdr.back;
        for (i = 0; i < 2; forward = !forward, i++) {
+               struct xfs_da3_icnode_hdr thdr;
                if (forward)
-                       blkno = be32_to_cpu(info->forw);
+                       blkno = nodehdr.forw;
                else
-                       blkno = be32_to_cpu(info->back);
+                       blkno = nodehdr.back;
                if (blkno == 0)
                        continue;
-               error = xfs_da_node_read(state->args->trans, state->args->dp,
+               error = xfs_da3_node_read(state->args->trans, dp,
                                        blkno, -1, &bp, state->args->whichfork);
                if (error)
-                       return(error);
-               ASSERT(bp != NULL);
+                       return error;
 
-               node = (xfs_da_intnode_t *)info;
-               count  = state->node_ents;
-               count -= state->node_ents >> 2;
-               count -= be16_to_cpu(node->hdr.count);
                node = bp->b_addr;
-               ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
-               count -= be16_to_cpu(node->hdr.count);
+               dp->d_ops->node_hdr_from_disk(&thdr, node);
                xfs_trans_brelse(state->args->trans, bp);
-               if (count >= 0)
+
+               if (count - thdr.count >= 0)
                        break;  /* fits with at least 25% to spare */
        }
        if (i >= 2) {
                *action = 0;
-               return(0);
+               return 0;
        }
 
        /*
@@ -932,28 +1236,43 @@ xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
         */
        memcpy(&state->altpath, &state->path, sizeof(state->path));
        if (blkno < blk->blkno) {
-               error = xfs_da_path_shift(state, &state->altpath, forward,
+               error = xfs_da3_path_shift(state, &state->altpath, forward,
                                                 0, &retval);
-               if (error) {
-                       return(error);
-               }
-               if (retval) {
-                       *action = 0;
-                       return(0);
-               }
        } else {
-               error = xfs_da_path_shift(state, &state->path, forward,
+               error = xfs_da3_path_shift(state, &state->path, forward,
                                                 0, &retval);
-               if (error) {
-                       return(error);
-               }
-               if (retval) {
-                       *action = 0;
-                       return(0);
-               }
+       }
+       if (error)
+               return error;
+       if (retval) {
+               *action = 0;
+               return 0;
        }
        *action = 1;
-       return(0);
+       return 0;
+}
+
+/*
+ * Pick up the last hashvalue from an intermediate node.
+ */
+STATIC uint
+xfs_da3_node_lasthash(
+       struct xfs_inode        *dp,
+       struct xfs_buf          *bp,
+       int                     *count)
+{
+       struct xfs_da_intnode    *node;
+       struct xfs_da_node_entry *btree;
+       struct xfs_da3_icnode_hdr nodehdr;
+
+       node = bp->b_addr;
+       dp->d_ops->node_hdr_from_disk(&nodehdr, node);
+       if (count)
+               *count = nodehdr.count;
+       if (!nodehdr.count)
+               return 0;
+       btree = dp->d_ops->node_tree_p(node);
+       return be32_to_cpu(btree[nodehdr.count - 1].hashval);
 }
 
 /*
@@ -961,13 +1280,17 @@ xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
  * when we stop making changes, return.
  */
 void
-xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
+xfs_da3_fixhashpath(
+       struct xfs_da_state     *state,
+       struct xfs_da_state_path *path)
 {
-       xfs_da_state_blk_t *blk;
-       xfs_da_intnode_t *node;
-       xfs_da_node_entry_t *btree;
-       xfs_dahash_t lasthash=0;
-       int level, count;
+       struct xfs_da_state_blk *blk;
+       struct xfs_da_intnode   *node;
+       struct xfs_da_node_entry *btree;
+       xfs_dahash_t            lasthash=0;
+       int                     level;
+       int                     count;
+       struct xfs_inode        *dp = state->args->dp;
 
        trace_xfs_da_fixhashpath(state->args);
 
@@ -980,28 +1303,31 @@ xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
                        return;
                break;
        case XFS_DIR2_LEAFN_MAGIC:
-               lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count);
+               lasthash = xfs_dir2_leaf_lasthash(dp, blk->bp, &count);
                if (count == 0)
                        return;
                break;
        case XFS_DA_NODE_MAGIC:
-               lasthash = xfs_da_node_lasthash(blk->bp, &count);
+               lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
                if (count == 0)
                        return;
                break;
        }
        for (blk--, level--; level >= 0; blk--, level--) {
+               struct xfs_da3_icnode_hdr nodehdr;
+
                node = blk->bp->b_addr;
-               ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
-               btree = &node->btree[ blk->index ];
-               if (be32_to_cpu(btree->hashval) == lasthash)
+               dp->d_ops->node_hdr_from_disk(&nodehdr, node);
+               btree = dp->d_ops->node_tree_p(node);
+               if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
                        break;
                blk->hashval = lasthash;
-               btree->hashval = cpu_to_be32(lasthash);
+               btree[blk->index].hashval = cpu_to_be32(lasthash);
                xfs_trans_log_buf(state->args->trans, blk->bp,
-                                 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
+                                 XFS_DA_LOGRANGE(node, &btree[blk->index],
+                                                 sizeof(*btree)));
 
-               lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
+               lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
        }
 }
 
@@ -1009,104 +1335,122 @@ xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
  * Remove an entry from an intermediate node.
  */
 STATIC void
-xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk)
+xfs_da3_node_remove(
+       struct xfs_da_state     *state,
+       struct xfs_da_state_blk *drop_blk)
 {
-       xfs_da_intnode_t *node;
-       xfs_da_node_entry_t *btree;
-       int tmp;
+       struct xfs_da_intnode   *node;
+       struct xfs_da3_icnode_hdr nodehdr;
+       struct xfs_da_node_entry *btree;
+       int                     index;
+       int                     tmp;
+       struct xfs_inode        *dp = state->args->dp;
 
        trace_xfs_da_node_remove(state->args);
 
        node = drop_blk->bp->b_addr;
-       ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count));
+       dp->d_ops->node_hdr_from_disk(&nodehdr, node);
+       ASSERT(drop_blk->index < nodehdr.count);
        ASSERT(drop_blk->index >= 0);
 
        /*
         * Copy over the offending entry, or just zero it out.
         */
-       btree = &node->btree[drop_blk->index];
-       if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) {
-               tmp  = be16_to_cpu(node->hdr.count) - drop_blk->index - 1;
+       index = drop_blk->index;
+       btree = dp->d_ops->node_tree_p(node);
+       if (index < nodehdr.count - 1) {
+               tmp  = nodehdr.count - index - 1;
                tmp *= (uint)sizeof(xfs_da_node_entry_t);
-               memmove(btree, btree + 1, tmp);
+               memmove(&btree[index], &btree[index + 1], tmp);
                xfs_trans_log_buf(state->args->trans, drop_blk->bp,
-                   XFS_DA_LOGRANGE(node, btree, tmp));
-               btree = &node->btree[be16_to_cpu(node->hdr.count)-1];
+                   XFS_DA_LOGRANGE(node, &btree[index], tmp));
+               index = nodehdr.count - 1;
        }
-       memset((char *)btree, 0, sizeof(xfs_da_node_entry_t));
+       memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
        xfs_trans_log_buf(state->args->trans, drop_blk->bp,
-           XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
-       be16_add_cpu(&node->hdr.count, -1);
+           XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
+       nodehdr.count -= 1;
+       dp->d_ops->node_hdr_to_disk(node, &nodehdr);
        xfs_trans_log_buf(state->args->trans, drop_blk->bp,
-           XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
+           XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
 
        /*
         * Copy the last hash value from the block to propagate upwards.
         */
-       btree--;
-       drop_blk->hashval = be32_to_cpu(btree->hashval);
+       drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
 }
 
 /*
- * Unbalance the btree elements between two intermediate nodes,
+ * Unbalance the elements between two intermediate nodes,
  * move all Btree elements from one node into another.
  */
 STATIC void
-xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
-                                    xfs_da_state_blk_t *save_blk)
+xfs_da3_node_unbalance(
+       struct xfs_da_state     *state,
+       struct xfs_da_state_blk *drop_blk,
+       struct xfs_da_state_blk *save_blk)
 {
-       xfs_da_intnode_t *drop_node, *save_node;
-       xfs_da_node_entry_t *btree;
-       int tmp;
-       xfs_trans_t *tp;
+       struct xfs_da_intnode   *drop_node;
+       struct xfs_da_intnode   *save_node;
+       struct xfs_da_node_entry *drop_btree;
+       struct xfs_da_node_entry *save_btree;
+       struct xfs_da3_icnode_hdr drop_hdr;
+       struct xfs_da3_icnode_hdr save_hdr;
+       struct xfs_trans        *tp;
+       int                     sindex;
+       int                     tmp;
+       struct xfs_inode        *dp = state->args->dp;
 
        trace_xfs_da_node_unbalance(state->args);
 
        drop_node = drop_blk->bp->b_addr;
        save_node = save_blk->bp->b_addr;
-       ASSERT(drop_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
-       ASSERT(save_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
+       dp->d_ops->node_hdr_from_disk(&drop_hdr, drop_node);
+       dp->d_ops->node_hdr_from_disk(&save_hdr, save_node);
+       drop_btree = dp->d_ops->node_tree_p(drop_node);
+       save_btree = dp->d_ops->node_tree_p(save_node);
        tp = state->args->trans;
 
        /*
         * If the dying block has lower hashvals, then move all the
         * elements in the remaining block up to make a hole.
         */
-       if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) ||
-           (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) <
-            be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval)))
-       {
-               btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)];
-               tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
-               memmove(btree, &save_node->btree[0], tmp);
-               btree = &save_node->btree[0];
+       if ((be32_to_cpu(drop_btree[0].hashval) <
+                       be32_to_cpu(save_btree[0].hashval)) ||
+           (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
+                       be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
+               /* XXX: check this - is memmove dst correct? */
+               tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
+               memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
+
+               sindex = 0;
                xfs_trans_log_buf(tp, save_blk->bp,
-                       XFS_DA_LOGRANGE(save_node, btree,
-                               (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) *
-                               sizeof(xfs_da_node_entry_t)));
+                       XFS_DA_LOGRANGE(save_node, &save_btree[0],
+                               (save_hdr.count + drop_hdr.count) *
+                                               sizeof(xfs_da_node_entry_t)));
        } else {
-               btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)];
+               sindex = save_hdr.count;
                xfs_trans_log_buf(tp, save_blk->bp,
-                       XFS_DA_LOGRANGE(save_node, btree,
-                               be16_to_cpu(drop_node->hdr.count) *
-                               sizeof(xfs_da_node_entry_t)));
+                       XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
+                               drop_hdr.count * sizeof(xfs_da_node_entry_t)));
        }
 
        /*
         * Move all the B-tree elements from drop_blk to save_blk.
         */
-       tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
-       memcpy(btree, &drop_node->btree[0], tmp);
-       be16_add_cpu(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count));
+       tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
+       memcpy(&save_btree[sindex], &drop_btree[0], tmp);
+       save_hdr.count += drop_hdr.count;
 
+       dp->d_ops->node_hdr_to_disk(save_node, &save_hdr);
        xfs_trans_log_buf(tp, save_blk->bp,
                XFS_DA_LOGRANGE(save_node, &save_node->hdr,
-                       sizeof(save_node->hdr)));
+                               dp->d_ops->node_hdr_size));
 
        /*
         * Save the last hashval in the remaining block for upward propagation.
         */
-       save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval);
+       save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
 }
 
 /*========================================================================
@@ -1125,16 +1469,26 @@ xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
  * pruned depth-first tree search.
  */
 int                                                    /* error */
-xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
+xfs_da3_node_lookup_int(
+       struct xfs_da_state     *state,
+       int                     *result)
 {
-       xfs_da_state_blk_t *blk;
-       xfs_da_blkinfo_t *curr;
-       xfs_da_intnode_t *node;
-       xfs_da_node_entry_t *btree;
-       xfs_dablk_t blkno;
-       int probe, span, max, error, retval;
-       xfs_dahash_t hashval, btreehashval;
-       xfs_da_args_t *args;
+       struct xfs_da_state_blk *blk;
+       struct xfs_da_blkinfo   *curr;
+       struct xfs_da_intnode   *node;
+       struct xfs_da_node_entry *btree;
+       struct xfs_da3_icnode_hdr nodehdr;
+       struct xfs_da_args      *args;
+       xfs_dablk_t             blkno;
+       xfs_dahash_t            hashval;
+       xfs_dahash_t            btreehashval;
+       int                     probe;
+       int                     span;
+       int                     max;
+       int                     error;
+       int                     retval;
+       unsigned int            expected_level = 0;
+       struct xfs_inode        *dp = state->args->dp;
 
        args = state->args;
 
@@ -1142,7 +1496,7 @@ xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
         * Descend thru the B-tree searching each level for the right
         * node to use, until the right hashval is found.
         */
-       blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0;
+       blkno = args->geo->leafblk;
        for (blk = &state->path.blk[0], state->path.active = 1;
                         state->path.active <= XFS_DA_NODE_MAXDEPTH;
                         blk++, state->path.active++) {
@@ -1150,78 +1504,107 @@ xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
                 * Read the next node down in the tree.
                 */
                blk->blkno = blkno;
-               error = xfs_da_node_read(args->trans, args->dp, blkno,
+               error = xfs_da3_node_read(args->trans, args->dp, blkno,
                                        -1, &blk->bp, args->whichfork);
                if (error) {
                        blk->blkno = 0;
                        state->path.active--;
-                       return(error);
+                       return error;
                }
                curr = blk->bp->b_addr;
                blk->magic = be16_to_cpu(curr->magic);
-               ASSERT(blk->magic == XFS_DA_NODE_MAGIC ||
-                      blk->magic == XFS_DIR2_LEAFN_MAGIC ||
-                      blk->magic == XFS_ATTR_LEAF_MAGIC);
+
+               if (blk->magic == XFS_ATTR_LEAF_MAGIC ||
+                   blk->magic == XFS_ATTR3_LEAF_MAGIC) {
+                       blk->magic = XFS_ATTR_LEAF_MAGIC;
+                       blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
+                       break;
+               }
+
+               if (blk->magic == XFS_DIR2_LEAFN_MAGIC ||
+                   blk->magic == XFS_DIR3_LEAFN_MAGIC) {
+                       blk->magic = XFS_DIR2_LEAFN_MAGIC;
+                       blk->hashval = xfs_dir2_leaf_lasthash(args->dp,
+                                                             blk->bp, NULL);
+                       break;
+               }
+
+               blk->magic = XFS_DA_NODE_MAGIC;
+
 
                /*
                 * Search an intermediate node for a match.
                 */
-               if (blk->magic == XFS_DA_NODE_MAGIC) {
-                       node = blk->bp->b_addr;
-                       max = be16_to_cpu(node->hdr.count);
-                       blk->hashval = be32_to_cpu(node->btree[max-1].hashval);
+               node = blk->bp->b_addr;
+               dp->d_ops->node_hdr_from_disk(&nodehdr, node);
+               btree = dp->d_ops->node_tree_p(node);
+
+               /* Tree taller than we can handle; bail out! */
+               if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH)
+                       return -EFSCORRUPTED;
+
+               /* Check the level from the root. */
+               if (blkno == args->geo->leafblk)
+                       expected_level = nodehdr.level - 1;
+               else if (expected_level != nodehdr.level)
+                       return -EFSCORRUPTED;
+               else
+                       expected_level--;
 
-                       /*
-                        * Binary search.  (note: small blocks will skip loop)
-                        */
-                       probe = span = max / 2;
-                       hashval = args->hashval;
-                       for (btree = &node->btree[probe]; span > 4;
-                                  btree = &node->btree[probe]) {
-                               span /= 2;
-                               btreehashval = be32_to_cpu(btree->hashval);
-                               if (btreehashval < hashval)
-                                       probe += span;
-                               else if (btreehashval > hashval)
-                                       probe -= span;
-                               else
-                                       break;
-                       }
-                       ASSERT((probe >= 0) && (probe < max));
-                       ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval));
+               max = nodehdr.count;
+               blk->hashval = be32_to_cpu(btree[max - 1].hashval);
 
-                       /*
-                        * Since we may have duplicate hashval's, find the first
-                        * matching hashval in the node.
-                        */
-                       while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) {
-                               btree--;
-                               probe--;
-                       }
-                       while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) {
-                               btree++;
-                               probe++;
-                       }
+               /*
+                * Binary search.  (note: small blocks will skip loop)
+                */
+               probe = span = max / 2;
+               hashval = args->hashval;
+               while (span > 4) {
+                       span /= 2;
+                       btreehashval = be32_to_cpu(btree[probe].hashval);
+                       if (btreehashval < hashval)
+                               probe += span;
+                       else if (btreehashval > hashval)
+                               probe -= span;
+                       else
+                               break;
+               }
+               ASSERT((probe >= 0) && (probe < max));
+               ASSERT((span <= 4) ||
+                       (be32_to_cpu(btree[probe].hashval) == hashval));
 
-                       /*
-                        * Pick the right block to descend on.
-                        */
-                       if (probe == max) {
-                               blk->index = max-1;
-                               blkno = be32_to_cpu(node->btree[max-1].before);
-                       } else {
-                               blk->index = probe;
-                               blkno = be32_to_cpu(btree->before);
-                       }
-               } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
-                       blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
-                       break;
-               } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
-                       blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL);
-                       break;
+               /*
+                * Since we may have duplicate hashval's, find the first
+                * matching hashval in the node.
+                */
+               while (probe > 0 &&
+                      be32_to_cpu(btree[probe].hashval) >= hashval) {
+                       probe--;
+               }
+               while (probe < max &&
+                      be32_to_cpu(btree[probe].hashval) < hashval) {
+                       probe++;
+               }
+
+               /*
+                * Pick the right block to descend on.
+                */
+               if (probe == max) {
+                       blk->index = max - 1;
+                       blkno = be32_to_cpu(btree[max - 1].before);
+               } else {
+                       blk->index = probe;
+                       blkno = be32_to_cpu(btree[probe].before);
                }
+
+               /* We can't point back to the root. */
+               if (blkno == args->geo->leafblk)
+                       return -EFSCORRUPTED;
        }
 
+       if (expected_level != 0)
+               return -EFSCORRUPTED;
+
        /*
         * A leaf block that ends in the hashval that we are interested in
         * (final hashval == search hashval) means that the next block may
@@ -1233,47 +1616,85 @@ xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
                        retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
                                                        &blk->index, state);
                } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
-                       retval = xfs_attr_leaf_lookup_int(blk->bp, args);
+                       retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
                        blk->index = args->index;
                        args->blkno = blk->blkno;
                } else {
                        ASSERT(0);
-                       return XFS_ERROR(EFSCORRUPTED);
+                       return -EFSCORRUPTED;
                }
-               if (((retval == ENOENT) || (retval == ENOATTR)) &&
+               if (((retval == -ENOENT) || (retval == -ENOATTR)) &&
                    (blk->hashval == args->hashval)) {
-                       error = xfs_da_path_shift(state, &state->path, 1, 1,
+                       error = xfs_da3_path_shift(state, &state->path, 1, 1,
                                                         &retval);
                        if (error)
-                               return(error);
+                               return error;
                        if (retval == 0) {
                                continue;
                        } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
                                /* path_shift() gives ENOENT */
-                               retval = XFS_ERROR(ENOATTR);
+                               retval = -ENOATTR;
                        }
                }
                break;
        }
        *result = retval;
-       return(0);
+       return 0;
 }
 
 /*========================================================================
  * Utility routines.
  *========================================================================*/
 
+/*
+ * Compare two intermediate nodes for "order".
+ */
+STATIC int
+xfs_da3_node_order(
+       struct xfs_inode *dp,
+       struct xfs_buf  *node1_bp,
+       struct xfs_buf  *node2_bp)
+{
+       struct xfs_da_intnode   *node1;
+       struct xfs_da_intnode   *node2;
+       struct xfs_da_node_entry *btree1;
+       struct xfs_da_node_entry *btree2;
+       struct xfs_da3_icnode_hdr node1hdr;
+       struct xfs_da3_icnode_hdr node2hdr;
+
+       node1 = node1_bp->b_addr;
+       node2 = node2_bp->b_addr;
+       dp->d_ops->node_hdr_from_disk(&node1hdr, node1);
+       dp->d_ops->node_hdr_from_disk(&node2hdr, node2);
+       btree1 = dp->d_ops->node_tree_p(node1);
+       btree2 = dp->d_ops->node_tree_p(node2);
+
+       if (node1hdr.count > 0 && node2hdr.count > 0 &&
+           ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
+            (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
+             be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
+               return 1;
+       }
+       return 0;
+}
+
 /*
  * Link a new block into a doubly linked list of blocks (of whatever type).
  */
 int                                                    /* error */
-xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
-                              xfs_da_state_blk_t *new_blk)
+xfs_da3_blk_link(
+       struct xfs_da_state     *state,
+       struct xfs_da_state_blk *old_blk,
+       struct xfs_da_state_blk *new_blk)
 {
-       xfs_da_blkinfo_t *old_info, *new_info, *tmp_info;
-       xfs_da_args_t *args;
-       int before=0, error;
-       struct xfs_buf *bp;
+       struct xfs_da_blkinfo   *old_info;
+       struct xfs_da_blkinfo   *new_info;
+       struct xfs_da_blkinfo   *tmp_info;
+       struct xfs_da_args      *args;
+       struct xfs_buf          *bp;
+       int                     before = 0;
+       int                     error;
+       struct xfs_inode        *dp = state->args->dp;
 
        /*
         * Set up environment.
@@ -1285,19 +1706,16 @@ xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
        ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
               old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
               old_blk->magic == XFS_ATTR_LEAF_MAGIC);
-       ASSERT(old_blk->magic == be16_to_cpu(old_info->magic));
-       ASSERT(new_blk->magic == be16_to_cpu(new_info->magic));
-       ASSERT(old_blk->magic == new_blk->magic);
 
        switch (old_blk->magic) {
        case XFS_ATTR_LEAF_MAGIC:
                before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
                break;
        case XFS_DIR2_LEAFN_MAGIC:
-               before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp);
+               before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
                break;
        case XFS_DA_NODE_MAGIC:
-               before = xfs_da_node_order(old_blk->bp, new_blk->bp);
+               before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
                break;
        }
 
@@ -1312,14 +1730,14 @@ xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
                new_info->forw = cpu_to_be32(old_blk->blkno);
                new_info->back = old_info->back;
                if (old_info->back) {
-                       error = xfs_da_node_read(args->trans, args->dp,
+                       error = xfs_da3_node_read(args->trans, dp,
                                                be32_to_cpu(old_info->back),
                                                -1, &bp, args->whichfork);
                        if (error)
-                               return(error);
+                               return error;
                        ASSERT(bp != NULL);
                        tmp_info = bp->b_addr;
-                       ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic));
+                       ASSERT(tmp_info->magic == old_info->magic);
                        ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
                        tmp_info->forw = cpu_to_be32(new_blk->blkno);
                        xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
@@ -1333,11 +1751,11 @@ xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
                new_info->forw = old_info->forw;
                new_info->back = cpu_to_be32(old_blk->blkno);
                if (old_info->forw) {
-                       error = xfs_da_node_read(args->trans, args->dp,
+                       error = xfs_da3_node_read(args->trans, dp,
                                                be32_to_cpu(old_info->forw),
                                                -1, &bp, args->whichfork);
                        if (error)
-                               return(error);
+                               return error;
                        ASSERT(bp != NULL);
                        tmp_info = bp->b_addr;
                        ASSERT(tmp_info->magic == old_info->magic);
@@ -1350,63 +1768,24 @@ xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
 
        xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
        xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
-       return(0);
-}
-
-/*
- * Compare two intermediate nodes for "order".
- */
-STATIC int
-xfs_da_node_order(
-       struct xfs_buf  *node1_bp,
-       struct xfs_buf  *node2_bp)
-{
-       xfs_da_intnode_t *node1, *node2;
-
-       node1 = node1_bp->b_addr;
-       node2 = node2_bp->b_addr;
-       ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) &&
-              node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
-       if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
-           ((be32_to_cpu(node2->btree[0].hashval) <
-             be32_to_cpu(node1->btree[0].hashval)) ||
-            (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
-             be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
-               return(1);
-       }
-       return(0);
-}
-
-/*
- * Pick up the last hashvalue from an intermediate node.
- */
-STATIC uint
-xfs_da_node_lasthash(
-       struct xfs_buf  *bp,
-       int             *count)
-{
-       xfs_da_intnode_t *node;
-
-       node = bp->b_addr;
-       ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
-       if (count)
-               *count = be16_to_cpu(node->hdr.count);
-       if (!node->hdr.count)
-               return(0);
-       return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
+       return 0;
 }
 
 /*
  * Unlink a block from a doubly linked list of blocks.
  */
 STATIC int                                             /* error */
-xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
-                                xfs_da_state_blk_t *save_blk)
+xfs_da3_blk_unlink(
+       struct xfs_da_state     *state,
+       struct xfs_da_state_blk *drop_blk,
+       struct xfs_da_state_blk *save_blk)
 {
-       xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info;
-       xfs_da_args_t *args;
-       struct xfs_buf *bp;
-       int error;
+       struct xfs_da_blkinfo   *drop_info;
+       struct xfs_da_blkinfo   *save_info;
+       struct xfs_da_blkinfo   *tmp_info;
+       struct xfs_da_args      *args;
+       struct xfs_buf          *bp;
+       int                     error;
 
        /*
         * Set up environment.
@@ -1418,8 +1797,6 @@ xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
        ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
               save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
               save_blk->magic == XFS_ATTR_LEAF_MAGIC);
-       ASSERT(save_blk->magic == be16_to_cpu(save_info->magic));
-       ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic));
        ASSERT(save_blk->magic == drop_blk->magic);
        ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
               (be32_to_cpu(save_info->back) == drop_blk->blkno));
@@ -1433,11 +1810,11 @@ xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
                trace_xfs_da_unlink_back(args);
                save_info->back = drop_info->back;
                if (drop_info->back) {
-                       error = xfs_da_node_read(args->trans, args->dp,
+                       error = xfs_da3_node_read(args->trans, args->dp,
                                                be32_to_cpu(drop_info->back),
                                                -1, &bp, args->whichfork);
                        if (error)
-                               return(error);
+                               return error;
                        ASSERT(bp != NULL);
                        tmp_info = bp->b_addr;
                        ASSERT(tmp_info->magic == save_info->magic);
@@ -1450,11 +1827,11 @@ xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
                trace_xfs_da_unlink_forward(args);
                save_info->forw = drop_info->forw;
                if (drop_info->forw) {
-                       error = xfs_da_node_read(args->trans, args->dp,
+                       error = xfs_da3_node_read(args->trans, args->dp,
                                                be32_to_cpu(drop_info->forw),
                                                -1, &bp, args->whichfork);
                        if (error)
-                               return(error);
+                               return error;
                        ASSERT(bp != NULL);
                        tmp_info = bp->b_addr;
                        ASSERT(tmp_info->magic == save_info->magic);
@@ -1466,7 +1843,7 @@ xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
        }
 
        xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
-       return(0);
+       return 0;
 }
 
 /*
@@ -1478,15 +1855,24 @@ xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
  * the new bottom and the root.
  */
 int                                                    /* error */
-xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
-                                int forward, int release, int *result)
+xfs_da3_path_shift(
+       struct xfs_da_state     *state,
+       struct xfs_da_state_path *path,
+       int                     forward,
+       int                     release,
+       int                     *result)
 {
-       xfs_da_state_blk_t *blk;
-       xfs_da_blkinfo_t *info;
-       xfs_da_intnode_t *node;
-       xfs_da_args_t *args;
-       xfs_dablk_t blkno=0;
-       int level, error;
+       struct xfs_da_state_blk *blk;
+       struct xfs_da_blkinfo   *info;
+       struct xfs_da_intnode   *node;
+       struct xfs_da_args      *args;
+       struct xfs_da_node_entry *btree;
+       struct xfs_da3_icnode_hdr nodehdr;
+       struct xfs_buf          *bp;
+       xfs_dablk_t             blkno = 0;
+       int                     level;
+       int                     error;
+       struct xfs_inode        *dp = state->args->dp;
 
        trace_xfs_da_path_shift(state->args);
 
@@ -1501,23 +1887,24 @@ xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
        ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
        level = (path->active-1) - 1;   /* skip bottom layer in path */
        for (blk = &path->blk[level]; level >= 0; blk--, level--) {
-               ASSERT(blk->bp != NULL);
                node = blk->bp->b_addr;
-               ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
-               if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) {
+               dp->d_ops->node_hdr_from_disk(&nodehdr, node);
+               btree = dp->d_ops->node_tree_p(node);
+
+               if (forward && (blk->index < nodehdr.count - 1)) {
                        blk->index++;
-                       blkno = be32_to_cpu(node->btree[blk->index].before);
+                       blkno = be32_to_cpu(btree[blk->index].before);
                        break;
                } else if (!forward && (blk->index > 0)) {
                        blk->index--;
-                       blkno = be32_to_cpu(node->btree[blk->index].before);
+                       blkno = be32_to_cpu(btree[blk->index].before);
                        break;
                }
        }
        if (level < 0) {
-               *result = XFS_ERROR(ENOENT);    /* we're out of our tree */
+               *result = -ENOENT;      /* we're out of our tree */
                ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
-               return(0);
+               return 0;
        }
 
        /*
@@ -1526,55 +1913,73 @@ xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
         */
        for (blk++, level++; level < path->active; blk++, level++) {
                /*
-                * Release the old block.
-                * (if it's dirty, trans won't actually let go)
+                * Read the next child block into a local buffer.
                 */
-               if (release)
-                       xfs_trans_brelse(args->trans, blk->bp);
+               error = xfs_da3_node_read(args->trans, dp, blkno, -1, &bp,
+                                         args->whichfork);
+               if (error)
+                       return error;
 
                /*
-                * Read the next child block.
+                * Release the old block (if it's dirty, the trans doesn't
+                * actually let go) and swap the local buffer into the path
+                * structure. This ensures failure of the above read doesn't set
+                * a NULL buffer in an active slot in the path.
                 */
+               if (release)
+                       xfs_trans_brelse(args->trans, blk->bp);
                blk->blkno = blkno;
-               error = xfs_da_node_read(args->trans, args->dp, blkno, -1,
-                                       &blk->bp, args->whichfork);
-               if (error)
-                       return(error);
-               ASSERT(blk->bp != NULL);
+               blk->bp = bp;
+
                info = blk->bp->b_addr;
                ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
+                      info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
                       info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
-                      info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
-               blk->magic = be16_to_cpu(info->magic);
-               if (blk->magic == XFS_DA_NODE_MAGIC) {
+                      info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
+                      info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
+                      info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
+
+
+               /*
+                * Note: we flatten the magic number to a single type so we
+                * don't have to compare against crc/non-crc types elsewhere.
+                */
+               switch (be16_to_cpu(info->magic)) {
+               case XFS_DA_NODE_MAGIC:
+               case XFS_DA3_NODE_MAGIC:
+                       blk->magic = XFS_DA_NODE_MAGIC;
                        node = (xfs_da_intnode_t *)info;
-                       blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
+                       dp->d_ops->node_hdr_from_disk(&nodehdr, node);
+                       btree = dp->d_ops->node_tree_p(node);
+                       blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
                        if (forward)
                                blk->index = 0;
                        else
-                               blk->index = be16_to_cpu(node->hdr.count)-1;
-                       blkno = be32_to_cpu(node->btree[blk->index].before);
-               } else {
+                               blk->index = nodehdr.count - 1;
+                       blkno = be32_to_cpu(btree[blk->index].before);
+                       break;
+               case XFS_ATTR_LEAF_MAGIC:
+               case XFS_ATTR3_LEAF_MAGIC:
+                       blk->magic = XFS_ATTR_LEAF_MAGIC;
                        ASSERT(level == path->active-1);
                        blk->index = 0;
-                       switch(blk->magic) {
-                       case XFS_ATTR_LEAF_MAGIC:
-                               blk->hashval = xfs_attr_leaf_lasthash(blk->bp,
-                                                                     NULL);
-                               break;
-                       case XFS_DIR2_LEAFN_MAGIC:
-                               blk->hashval = xfs_dir2_leafn_lasthash(blk->bp,
-                                                                      NULL);
-                               break;
-                       default:
-                               ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC ||
-                                      blk->magic == XFS_DIR2_LEAFN_MAGIC);
-                               break;
-                       }
+                       blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
+                       break;
+               case XFS_DIR2_LEAFN_MAGIC:
+               case XFS_DIR3_LEAFN_MAGIC:
+                       blk->magic = XFS_DIR2_LEAFN_MAGIC;
+                       ASSERT(level == path->active-1);
+                       blk->index = 0;
+                       blk->hashval = xfs_dir2_leaf_lasthash(args->dp,
+                                                             blk->bp, NULL);
+                       break;
+               default:
+                       ASSERT(0);
+                       break;
                }
        }
        *result = 0;
-       return(0);
+       return 0;
 }
 
 
@@ -1588,7 +1993,7 @@ xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
  * This is implemented with some source-level loop unrolling.
  */
 xfs_dahash_t
-xfs_da_hashname(const __uint8_t *name, int namelen)
+xfs_da_hashname(const uint8_t *name, int namelen)
 {
        xfs_dahash_t hash;
 
@@ -1646,7 +2051,7 @@ xfs_da_grow_inode_int(
        struct xfs_trans        *tp = args->trans;
        struct xfs_inode        *dp = args->dp;
        int                     w = args->whichfork;
-       xfs_drfsbno_t           nblks = dp->i_d.di_nblocks;
+       xfs_rfsblock_t          nblks = dp->i_d.di_nblocks;
        struct xfs_bmbt_irec    map, *mapp;
        int                     nmap, error, got, i, mapi;
 
@@ -1665,7 +2070,7 @@ xfs_da_grow_inode_int(
        error = xfs_bmapi_write(tp, dp, *bno, count,
                        xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
                        args->firstblock, args->total, &map, &nmap,
-                       args->flist);
+                       args->dfops);
        if (error)
                return error;
 
@@ -1688,7 +2093,7 @@ xfs_da_grow_inode_int(
                        error = xfs_bmapi_write(tp, dp, b, c,
                                        xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
                                        args->firstblock, args->total,
-                                       &mapp[mapi], &nmap, args->flist);
+                                       &mapp[mapi], &nmap, args->dfops);
                        if (error)
                                goto out_free_map;
                        if (nmap < 1)
@@ -1710,7 +2115,7 @@ xfs_da_grow_inode_int(
        if (got != count || mapp[0].br_startoff != *bno ||
            mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
            *bno + count) {
-               error = XFS_ERROR(ENOSPC);
+               error = -ENOSPC;
                goto out_free_map;
        }
 
@@ -1733,20 +2138,12 @@ xfs_da_grow_inode(
        xfs_dablk_t             *new_blkno)
 {
        xfs_fileoff_t           bno;
-       int                     count;
        int                     error;
 
        trace_xfs_da_grow_inode(args);
 
-       if (args->whichfork == XFS_DATA_FORK) {
-               bno = args->dp->i_mount->m_dirleafblk;
-               count = args->dp->i_mount->m_dirblkfsbs;
-       } else {
-               bno = 0;
-               count = 1;
-       }
-
-       error = xfs_da_grow_inode_int(args, &bno, count);
+       bno = args->geo->leafblk;
+       error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount);
        if (!error)
                *new_blkno = (xfs_dablk_t)bno;
        return error;
@@ -1761,73 +2158,96 @@ xfs_da_grow_inode(
  * a bmap btree split to do that.
  */
 STATIC int
-xfs_da_swap_lastblock(
-       xfs_da_args_t   *args,
-       xfs_dablk_t     *dead_blknop,
-       struct xfs_buf  **dead_bufp)
+xfs_da3_swap_lastblock(
+       struct xfs_da_args      *args,
+       xfs_dablk_t             *dead_blknop,
+       struct xfs_buf          **dead_bufp)
 {
-       xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno;
-       struct xfs_buf *dead_buf, *last_buf, *sib_buf, *par_buf;
-       xfs_fileoff_t lastoff;
-       xfs_inode_t *ip;
-       xfs_trans_t *tp;
-       xfs_mount_t *mp;
-       int error, w, entno, level, dead_level;
-       xfs_da_blkinfo_t *dead_info, *sib_info;
-       xfs_da_intnode_t *par_node, *dead_node;
-       xfs_dir2_leaf_t *dead_leaf2;
-       xfs_dahash_t dead_hash;
+       struct xfs_da_blkinfo   *dead_info;
+       struct xfs_da_blkinfo   *sib_info;
+       struct xfs_da_intnode   *par_node;
+       struct xfs_da_intnode   *dead_node;
+       struct xfs_dir2_leaf    *dead_leaf2;
+       struct xfs_da_node_entry *btree;
+       struct xfs_da3_icnode_hdr par_hdr;
+       struct xfs_inode        *dp;
+       struct xfs_trans        *tp;
+       struct xfs_mount        *mp;
+       struct xfs_buf          *dead_buf;
+       struct xfs_buf          *last_buf;
+       struct xfs_buf          *sib_buf;
+       struct xfs_buf          *par_buf;
+       xfs_dahash_t            dead_hash;
+       xfs_fileoff_t           lastoff;
+       xfs_dablk_t             dead_blkno;
+       xfs_dablk_t             last_blkno;
+       xfs_dablk_t             sib_blkno;
+       xfs_dablk_t             par_blkno;
+       int                     error;
+       int                     w;
+       int                     entno;
+       int                     level;
+       int                     dead_level;
 
        trace_xfs_da_swap_lastblock(args);
 
        dead_buf = *dead_bufp;
        dead_blkno = *dead_blknop;
        tp = args->trans;
-       ip = args->dp;
+       dp = args->dp;
        w = args->whichfork;
        ASSERT(w == XFS_DATA_FORK);
-       mp = ip->i_mount;
-       lastoff = mp->m_dirfreeblk;
-       error = xfs_bmap_last_before(tp, ip, &lastoff, w);
+       mp = dp->i_mount;
+       lastoff = args->geo->freeblk;
+       error = xfs_bmap_last_before(tp, dp, &lastoff, w);
        if (error)
                return error;
        if (unlikely(lastoff == 0)) {
                XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
                                 mp);
-               return XFS_ERROR(EFSCORRUPTED);
+               return -EFSCORRUPTED;
        }
        /*
         * Read the last block in the btree space.
         */
-       last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
-       error = xfs_da_node_read(tp, ip, last_blkno, -1, &last_buf, w);
+       last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount;
+       error = xfs_da3_node_read(tp, dp, last_blkno, -1, &last_buf, w);
        if (error)
                return error;
        /*
         * Copy the last block into the dead buffer and log it.
         */
-       memcpy(dead_buf->b_addr, last_buf->b_addr, mp->m_dirblksize);
-       xfs_trans_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
+       memcpy(dead_buf->b_addr, last_buf->b_addr, args->geo->blksize);
+       xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1);
        dead_info = dead_buf->b_addr;
        /*
         * Get values from the moved block.
         */
-       if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) {
+       if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
+           dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
+               struct xfs_dir3_icleaf_hdr leafhdr;
+               struct xfs_dir2_leaf_entry *ents;
+
                dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
+               dp->d_ops->leaf_hdr_from_disk(&leafhdr, dead_leaf2);
+               ents = dp->d_ops->leaf_ents_p(dead_leaf2);
                dead_level = 0;
-               dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval);
+               dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
        } else {
-               ASSERT(dead_info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
+               struct xfs_da3_icnode_hdr deadhdr;
+
                dead_node = (xfs_da_intnode_t *)dead_info;
-               dead_level = be16_to_cpu(dead_node->hdr.level);
-               dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval);
+               dp->d_ops->node_hdr_from_disk(&deadhdr, dead_node);
+               btree = dp->d_ops->node_tree_p(dead_node);
+               dead_level = deadhdr.level;
+               dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
        }
        sib_buf = par_buf = NULL;
        /*
         * If the moved block has a left sibling, fix up the pointers.
         */
        if ((sib_blkno = be32_to_cpu(dead_info->back))) {
-               error = xfs_da_node_read(tp, ip, sib_blkno, -1, &sib_buf, w);
+               error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
                if (error)
                        goto done;
                sib_info = sib_buf->b_addr;
@@ -1836,7 +2256,7 @@ xfs_da_swap_lastblock(
                    sib_info->magic != dead_info->magic)) {
                        XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
                                         XFS_ERRLEVEL_LOW, mp);
-                       error = XFS_ERROR(EFSCORRUPTED);
+                       error = -EFSCORRUPTED;
                        goto done;
                }
                sib_info->forw = cpu_to_be32(dead_blkno);
@@ -1849,7 +2269,7 @@ xfs_da_swap_lastblock(
         * If the moved block has a right sibling, fix up the pointers.
         */
        if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
-               error = xfs_da_node_read(tp, ip, sib_blkno, -1, &sib_buf, w);
+               error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
                if (error)
                        goto done;
                sib_info = sib_buf->b_addr;
@@ -1858,7 +2278,7 @@ xfs_da_swap_lastblock(
                       sib_info->magic != dead_info->magic)) {
                        XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
                                         XFS_ERRLEVEL_LOW, mp);
-                       error = XFS_ERROR(EFSCORRUPTED);
+                       error = -EFSCORRUPTED;
                        goto done;
                }
                sib_info->back = cpu_to_be32(dead_blkno);
@@ -1867,37 +2287,37 @@ xfs_da_swap_lastblock(
                                        sizeof(sib_info->back)));
                sib_buf = NULL;
        }
-       par_blkno = mp->m_dirleafblk;
+       par_blkno = args->geo->leafblk;
        level = -1;
        /*
         * Walk down the tree looking for the parent of the moved block.
         */
        for (;;) {
-               error = xfs_da_node_read(tp, ip, par_blkno, -1, &par_buf, w);
+               error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
                if (error)
                        goto done;
                par_node = par_buf->b_addr;
-               if (unlikely(par_node->hdr.info.magic !=
-                   cpu_to_be16(XFS_DA_NODE_MAGIC) ||
-                   (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) {
+               dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
+               if (level >= 0 && level != par_hdr.level + 1) {
                        XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
                                         XFS_ERRLEVEL_LOW, mp);
-                       error = XFS_ERROR(EFSCORRUPTED);
+                       error = -EFSCORRUPTED;
                        goto done;
                }
-               level = be16_to_cpu(par_node->hdr.level);
+               level = par_hdr.level;
+               btree = dp->d_ops->node_tree_p(par_node);
                for (entno = 0;
-                    entno < be16_to_cpu(par_node->hdr.count) &&
-                    be32_to_cpu(par_node->btree[entno].hashval) < dead_hash;
+                    entno < par_hdr.count &&
+                    be32_to_cpu(btree[entno].hashval) < dead_hash;
                     entno++)
                        continue;
-               if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) {
+               if (entno == par_hdr.count) {
                        XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
                                         XFS_ERRLEVEL_LOW, mp);
-                       error = XFS_ERROR(EFSCORRUPTED);
+                       error = -EFSCORRUPTED;
                        goto done;
                }
-               par_blkno = be32_to_cpu(par_node->btree[entno].before);
+               par_blkno = be32_to_cpu(btree[entno].before);
                if (level == dead_level + 1)
                        break;
                xfs_trans_brelse(tp, par_buf);
@@ -1909,42 +2329,42 @@ xfs_da_swap_lastblock(
         */
        for (;;) {
                for (;
-                    entno < be16_to_cpu(par_node->hdr.count) &&
-                    be32_to_cpu(par_node->btree[entno].before) != last_blkno;
+                    entno < par_hdr.count &&
+                    be32_to_cpu(btree[entno].before) != last_blkno;
                     entno++)
                        continue;
-               if (entno < be16_to_cpu(par_node->hdr.count))
+               if (entno < par_hdr.count)
                        break;
-               par_blkno = be32_to_cpu(par_node->hdr.info.forw);
+               par_blkno = par_hdr.forw;
                xfs_trans_brelse(tp, par_buf);
                par_buf = NULL;
                if (unlikely(par_blkno == 0)) {
                        XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
                                         XFS_ERRLEVEL_LOW, mp);
-                       error = XFS_ERROR(EFSCORRUPTED);
+                       error = -EFSCORRUPTED;
                        goto done;
                }
-               error = xfs_da_node_read(tp, ip, par_blkno, -1, &par_buf, w);
+               error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
                if (error)
                        goto done;
                par_node = par_buf->b_addr;
-               if (unlikely(
-                   be16_to_cpu(par_node->hdr.level) != level ||
-                   par_node->hdr.info.magic != cpu_to_be16(XFS_DA_NODE_MAGIC))) {
+               dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
+               if (par_hdr.level != level) {
                        XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
                                         XFS_ERRLEVEL_LOW, mp);
-                       error = XFS_ERROR(EFSCORRUPTED);
+                       error = -EFSCORRUPTED;
                        goto done;
                }
+               btree = dp->d_ops->node_tree_p(par_node);
                entno = 0;
        }
        /*
         * Update the parent entry pointing to the moved block.
         */
-       par_node->btree[entno].before = cpu_to_be32(dead_blkno);
+       btree[entno].before = cpu_to_be32(dead_blkno);
        xfs_trans_log_buf(tp, par_buf,
-               XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before,
-                               sizeof(par_node->btree[entno].before)));
+               XFS_DA_LOGRANGE(par_node, &btree[entno].before,
+                               sizeof(btree[entno].before)));
        *dead_blknop = last_blkno;
        *dead_bufp = last_buf;
        return 0;
@@ -1969,31 +2389,27 @@ xfs_da_shrink_inode(
        xfs_inode_t *dp;
        int done, error, w, count;
        xfs_trans_t *tp;
-       xfs_mount_t *mp;
 
        trace_xfs_da_shrink_inode(args);
 
        dp = args->dp;
        w = args->whichfork;
        tp = args->trans;
-       mp = dp->i_mount;
-       if (w == XFS_DATA_FORK)
-               count = mp->m_dirblkfsbs;
-       else
-               count = 1;
+       count = args->geo->fsbcount;
        for (;;) {
                /*
                 * Remove extents.  If we get ENOSPC for a dir we have to move
                 * the last block to the place we want to kill.
                 */
-               if ((error = xfs_bunmapi(tp, dp, dead_blkno, count,
-                               xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
-                               0, args->firstblock, args->flist,
-                               &done)) == ENOSPC) {
+               error = xfs_bunmapi(tp, dp, dead_blkno, count,
+                                   xfs_bmapi_aflag(w), 0, args->firstblock,
+                                   args->dfops, &done);
+               if (error == -ENOSPC) {
                        if (w != XFS_DATA_FORK)
                                break;
-                       if ((error = xfs_da_swap_lastblock(args, &dead_blkno,
-                                       &dead_buf)))
+                       error = xfs_da3_swap_lastblock(args, &dead_blkno,
+                                                     &dead_buf);
+                       if (error)
                                break;
                } else {
                        break;
@@ -2042,9 +2458,9 @@ static int
 xfs_buf_map_from_irec(
        struct xfs_mount        *mp,
        struct xfs_buf_map      **mapp,
-       unsigned int            *nmaps,
+       int                     *nmaps,
        struct xfs_bmbt_irec    *irecs,
-       unsigned int            nirecs)
+       int                     nirecs)
 {
        struct xfs_buf_map      *map;
        int                     i;
@@ -2053,9 +2469,10 @@ xfs_buf_map_from_irec(
        ASSERT(nirecs >= 1);
 
        if (nirecs > 1) {
-               map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map), KM_SLEEP);
+               map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map),
+                                 KM_SLEEP | KM_NOFS);
                if (!map)
-                       return ENOMEM;
+                       return -ENOMEM;
                *mapp = map;
        }
 
@@ -2080,7 +2497,6 @@ xfs_buf_map_from_irec(
  */
 static int
 xfs_dabuf_map(
-       struct xfs_trans        *trans,
        struct xfs_inode        *dp,
        xfs_dablk_t             bno,
        xfs_daddr_t             mappedbno,
@@ -2098,7 +2514,10 @@ xfs_dabuf_map(
        ASSERT(map && *map);
        ASSERT(*nmaps == 1);
 
-       nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1;
+       if (whichfork == XFS_DATA_FORK)
+               nfsb = mp->m_dir_geo->fsbcount;
+       else
+               nfsb = mp->m_attr_geo->fsbcount;
 
        /*
         * Caller doesn't have a mapping.  -2 means don't complain
@@ -2109,7 +2528,8 @@ xfs_dabuf_map(
                 * Optimize the one-block case.
                 */
                if (nfsb != 1)
-                       irecs = kmem_zalloc(sizeof(irec) * nfsb, KM_SLEEP);
+                       irecs = kmem_zalloc(sizeof(irec) * nfsb,
+                                           KM_SLEEP | KM_NOFS);
 
                nirecs = nfsb;
                error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs,
@@ -2125,8 +2545,8 @@ xfs_dabuf_map(
        }
 
        if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) {
-               error = mappedbno == -2 ? -1 : XFS_ERROR(EFSCORRUPTED);
-               if (unlikely(error == EFSCORRUPTED)) {
+               error = mappedbno == -2 ? -1 : -EFSCORRUPTED;
+               if (unlikely(error == -EFSCORRUPTED)) {
                        if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
                                int i;
                                xfs_alert(mp, "%s: bno %lld dir: inode %lld",
@@ -2175,7 +2595,7 @@ xfs_da_get_buf(
        *bpp = NULL;
        mapp = &map;
        nmap = 1;
-       error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork,
+       error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
                                &mapp, &nmap);
        if (error) {
                /* mapping a hole is not an error, but we don't continue */
@@ -2186,9 +2606,10 @@ xfs_da_get_buf(
 
        bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp,
                                    mapp, nmap, 0);
-       error = bp ? bp->b_error : XFS_ERROR(EIO);
+       error = bp ? bp->b_error : -EIO;
        if (error) {
-               xfs_trans_brelse(trans, bp);
+               if (bp)
+                       xfs_trans_brelse(trans, bp);
                goto out_free;
        }
 
@@ -2223,7 +2644,7 @@ xfs_da_read_buf(
        *bpp = NULL;
        mapp = &map;
        nmap = 1;
-       error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork,
+       error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
                                &mapp, &nmap);
        if (error) {
                /* mapping a hole is not an error, but we don't continue */
@@ -2242,38 +2663,6 @@ xfs_da_read_buf(
                xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
        else
                xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
-
-       /*
-        * This verification code will be moved to a CRC verification callback
-        * function so just leave it here unchanged until then.
-        */
-       {
-               xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
-               xfs_dir2_free_t         *free = bp->b_addr;
-               xfs_da_blkinfo_t        *info = bp->b_addr;
-               uint                    magic, magic1;
-               struct xfs_mount        *mp = dp->i_mount;
-
-               magic = be16_to_cpu(info->magic);
-               magic1 = be32_to_cpu(hdr->magic);
-               if (unlikely(
-                   XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
-                                  (magic != XFS_ATTR_LEAF_MAGIC) &&
-                                  (magic != XFS_DIR2_LEAF1_MAGIC) &&
-                                  (magic != XFS_DIR2_LEAFN_MAGIC) &&
-                                  (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
-                                  (magic1 != XFS_DIR2_DATA_MAGIC) &&
-                                  (free->hdr.magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC)),
-                               mp, XFS_ERRTAG_DA_READ_BUF,
-                               XFS_RANDOM_DA_READ_BUF))) {
-                       trace_xfs_da_btree_corrupt(bp, _RET_IP_);
-                       XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
-                                            XFS_ERRLEVEL_LOW, mp, info);
-                       error = XFS_ERROR(EFSCORRUPTED);
-                       xfs_trans_brelse(trans, bp);
-                       goto out_free;
-               }
-       }
        *bpp = bp;
 out_free:
        if (mapp != &map)
@@ -2285,9 +2674,8 @@ out_free:
 /*
  * Readahead the dir/attr block.
  */
-xfs_daddr_t
+int
 xfs_da_reada_buf(
-       struct xfs_trans        *trans,
        struct xfs_inode        *dp,
        xfs_dablk_t             bno,
        xfs_daddr_t             mappedbno,
@@ -2301,7 +2689,7 @@ xfs_da_reada_buf(
 
        mapp = &map;
        nmap = 1;
-       error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork,
+       error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
                                &mapp, &nmap);
        if (error) {
                /* mapping a hole is not an error, but we don't continue */
@@ -2317,45 +2705,5 @@ out_free:
        if (mapp != &map)
                kmem_free(mapp);
 
-       if (error)
-               return -1;
-       return mappedbno;
-}
-
-kmem_zone_t *xfs_da_state_zone;        /* anchor for state struct zone */
-
-/*
- * Allocate a dir-state structure.
- * We don't put them on the stack since they're large.
- */
-xfs_da_state_t *
-xfs_da_state_alloc(void)
-{
-       return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
-}
-
-/*
- * Kill the altpath contents of a da-state structure.
- */
-STATIC void
-xfs_da_state_kill_altpath(xfs_da_state_t *state)
-{
-       int     i;
-
-       for (i = 0; i < state->altpath.active; i++)
-               state->altpath.blk[i].bp = NULL;
-       state->altpath.active = 0;
-}
-
-/*
- * Free a da-state structure.
- */
-void
-xfs_da_state_free(xfs_da_state_t *state)
-{
-       xfs_da_state_kill_altpath(state);
-#ifdef DEBUG
-       memset((char *)state, 0, sizeof(*state));
-#endif /* DEBUG */
-       kmem_zone_free(xfs_da_state_zone, state);
+       return error;
 }