29 files changed, 4687 insertions, 2230 deletions

diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
index f27e5378..a0c8667 100644
--- a/fs/ocfs2/alloc.c
+++ b/fs/ocfs2/alloc.c
@@ -27,6 +27,7 @@
 #include <linux/types.h>
 #include <linux/slab.h>
 #include <linux/highmem.h>
+#include <linux/swap.h>
 
 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
 #include <cluster/masklog.h>
@@ -34,6 +35,7 @@
 #include "ocfs2.h"
 
 #include "alloc.h"
+#include "aops.h"
 #include "dlmglue.h"
 #include "extent_map.h"
 #include "inode.h"
@@ -47,63 +49,243 @@
 
 #include "buffer_head_io.h"
 
-static int ocfs2_extent_contig(struct inode *inode,
-			       struct ocfs2_extent_rec *ext,
-			       u64 blkno);
+static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc);
 
-static int ocfs2_create_new_meta_bhs(struct ocfs2_super *osb,
-				     handle_t *handle,
-				     struct inode *inode,
-				     int wanted,
-				     struct ocfs2_alloc_context *meta_ac,
-				     struct buffer_head *bhs[]);
+/*
+ * Structures which describe a path through a btree, and functions to
+ * manipulate them.
+ *
+ * The idea here is to be as generic as possible with the tree
+ * manipulation code.
+ */
+struct ocfs2_path_item {
+	struct buffer_head		*bh;
+	struct ocfs2_extent_list	*el;
+};
 
-static int ocfs2_add_branch(struct ocfs2_super *osb,
-			    handle_t *handle,
-			    struct inode *inode,
-			    struct buffer_head *fe_bh,
-			    struct buffer_head *eb_bh,
-			    struct buffer_head *last_eb_bh,
-			    struct ocfs2_alloc_context *meta_ac);
+#define OCFS2_MAX_PATH_DEPTH	5
 
-static int ocfs2_shift_tree_depth(struct ocfs2_super *osb,
-				  handle_t *handle,
-				  struct inode *inode,
-				  struct buffer_head *fe_bh,
-				  struct ocfs2_alloc_context *meta_ac,
-				  struct buffer_head **ret_new_eb_bh);
+struct ocfs2_path {
+	int			p_tree_depth;
+	struct ocfs2_path_item	p_node[OCFS2_MAX_PATH_DEPTH];
+};
 
-static int ocfs2_do_insert_extent(struct ocfs2_super *osb,
-				  handle_t *handle,
-				  struct inode *inode,
-				  struct buffer_head *fe_bh,
-				  u64 blkno,
-				  u32 new_clusters);
+#define path_root_bh(_path) ((_path)->p_node[0].bh)
+#define path_root_el(_path) ((_path)->p_node[0].el)
+#define path_leaf_bh(_path) ((_path)->p_node[(_path)->p_tree_depth].bh)
+#define path_leaf_el(_path) ((_path)->p_node[(_path)->p_tree_depth].el)
+#define path_num_items(_path) ((_path)->p_tree_depth + 1)
 
-static int ocfs2_find_branch_target(struct ocfs2_super *osb,
-				    struct inode *inode,
-				    struct buffer_head *fe_bh,
-				    struct buffer_head **target_bh);
+/*
+ * Reset the actual path elements so that we can re-use the structure
+ * to build another path. Generally, this involves freeing the buffer
+ * heads.
+ */
+static void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root)
+{
+	int i, start = 0, depth = 0;
+	struct ocfs2_path_item *node;
 
-static int ocfs2_find_new_last_ext_blk(struct ocfs2_super *osb,
-				       struct inode *inode,
-				       struct ocfs2_dinode *fe,
-				       unsigned int new_i_clusters,
-				       struct buffer_head *old_last_eb,
-				       struct buffer_head **new_last_eb);
+	if (keep_root)
+		start = 1;
+
+	for(i = start; i < path_num_items(path); i++) {
+		node = &path->p_node[i];
+
+		brelse(node->bh);
+		node->bh = NULL;
+		node->el = NULL;
+	}
+
+	/*
+	 * Tree depth may change during truncate, or insert. If we're
+	 * keeping the root extent list, then make sure that our path
+	 * structure reflects the proper depth.
+	 */
+	if (keep_root)
+		depth = le16_to_cpu(path_root_el(path)->l_tree_depth);
+
+	path->p_tree_depth = depth;
+}
+
+static void ocfs2_free_path(struct ocfs2_path *path)
+{
+	if (path) {
+		ocfs2_reinit_path(path, 0);
+		kfree(path);
+	}
+}
+
+/*
+ * Make the *dest path the same as src and re-initialize src path to
+ * have a root only.
+ */
+static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src)
+{
+	int i;
+
+	BUG_ON(path_root_bh(dest) != path_root_bh(src));
+
+	for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
+		brelse(dest->p_node[i].bh);
+
+		dest->p_node[i].bh = src->p_node[i].bh;
+		dest->p_node[i].el = src->p_node[i].el;
+
+		src->p_node[i].bh = NULL;
+		src->p_node[i].el = NULL;
+	}
+}
+
+/*
+ * Insert an extent block at given index.
+ *
+ * This will not take an additional reference on eb_bh.
+ */
+static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index,
+					struct buffer_head *eb_bh)
+{
+	struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data;
+
+	/*
+	 * Right now, no root bh is an extent block, so this helps
+	 * catch code errors with dinode trees. The assertion can be
+	 * safely removed if we ever need to insert extent block
+	 * structures at the root.
+	 */
+	BUG_ON(index == 0);
+
+	path->p_node[index].bh = eb_bh;
+	path->p_node[index].el = &eb->h_list;
+}
+
+static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh,
+					 struct ocfs2_extent_list *root_el)
+{
+	struct ocfs2_path *path;
+
+	BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH);
+
+	path = kzalloc(sizeof(*path), GFP_NOFS);
+	if (path) {
+		path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth);
+		get_bh(root_bh);
+		path_root_bh(path) = root_bh;
+		path_root_el(path) = root_el;
+	}
+
+	return path;
+}
+
+/*
+ * Allocate and initialize a new path based on a disk inode tree.
+ */
+static struct ocfs2_path *ocfs2_new_inode_path(struct buffer_head *di_bh)
+{
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_extent_list *el = &di->id2.i_list;
+
+	return ocfs2_new_path(di_bh, el);
+}
+
+/*
+ * Convenience function to journal all components in a path.
+ */
+static int ocfs2_journal_access_path(struct inode *inode, handle_t *handle,
+				     struct ocfs2_path *path)
+{
+	int i, ret = 0;
+
+	if (!path)
+		goto out;
+
+	for(i = 0; i < path_num_items(path); i++) {
+		ret = ocfs2_journal_access(handle, inode, path->p_node[i].bh,
+					   OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+out:
+	return ret;
+}
+
+enum ocfs2_contig_type {
+	CONTIG_NONE = 0,
+	CONTIG_LEFT,
+	CONTIG_RIGHT
+};
 
-static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc);
 
-static int ocfs2_extent_contig(struct inode *inode,
-			       struct ocfs2_extent_rec *ext,
-			       u64 blkno)
+/*
+ * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and
+ * ocfs2_extent_contig only work properly against leaf nodes!
+ */
+static int ocfs2_block_extent_contig(struct super_block *sb,
+				     struct ocfs2_extent_rec *ext,
+				     u64 blkno)
+{
+	u64 blk_end = le64_to_cpu(ext->e_blkno);
+
+	blk_end += ocfs2_clusters_to_blocks(sb,
+				    le16_to_cpu(ext->e_leaf_clusters));
+
+	return blkno == blk_end;
+}
+
+static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left,
+				  struct ocfs2_extent_rec *right)
+{
+	u32 left_range;
+
+	left_range = le32_to_cpu(left->e_cpos) +
+		le16_to_cpu(left->e_leaf_clusters);
+
+	return (left_range == le32_to_cpu(right->e_cpos));
+}
+
+static enum ocfs2_contig_type
+	ocfs2_extent_contig(struct inode *inode,
+			    struct ocfs2_extent_rec *ext,
+			    struct ocfs2_extent_rec *insert_rec)
 {
-	return blkno == (le64_to_cpu(ext->e_blkno) +
-			 ocfs2_clusters_to_blocks(inode->i_sb,
-						  le32_to_cpu(ext->e_clusters)));
+	u64 blkno = le64_to_cpu(insert_rec->e_blkno);
+
+	if (ocfs2_extents_adjacent(ext, insert_rec) &&
+	    ocfs2_block_extent_contig(inode->i_sb, ext, blkno))
+			return CONTIG_RIGHT;
+
+	blkno = le64_to_cpu(ext->e_blkno);
+	if (ocfs2_extents_adjacent(insert_rec, ext) &&
+	    ocfs2_block_extent_contig(inode->i_sb, insert_rec, blkno))
+		return CONTIG_LEFT;
+
+	return CONTIG_NONE;
 }
 
 /*
+ * NOTE: We can have pretty much any combination of contiguousness and
+ * appending.
+ *
+ * The usefulness of APPEND_TAIL is more in that it lets us know that
+ * we'll have to update the path to that leaf.
+ */
+enum ocfs2_append_type {
+	APPEND_NONE = 0,
+	APPEND_TAIL,
+};
+
+struct ocfs2_insert_type {
+	enum ocfs2_append_type	ins_appending;
+	enum ocfs2_contig_type	ins_contig;
+	int			ins_contig_index;
+	int			ins_free_records;
+	int			ins_tree_depth;
+};
+
+/*
  * How many free extents have we got before we need more meta data?
  */
 int ocfs2_num_free_extents(struct ocfs2_super *osb,
@@ -242,6 +424,28 @@ bail:
 }
 
 /*
+ * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth().
+ *
+ * Returns the sum of the rightmost extent rec logical offset and
+ * cluster count.
+ *
+ * ocfs2_add_branch() uses this to determine what logical cluster
+ * value should be populated into the leftmost new branch records.
+ *
+ * ocfs2_shift_tree_depth() uses this to determine the # clusters
+ * value for the new topmost tree record.
+ */
+static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list  *el)
+{
+	int i;
+
+	i = le16_to_cpu(el->l_next_free_rec) - 1;
+
+	return le32_to_cpu(el->l_recs[i].e_cpos) +
+		ocfs2_rec_clusters(el, &el->l_recs[i]);
+}
+
+/*
  * Add an entire tree branch to our inode. eb_bh is the extent block
  * to start at, if we don't want to start the branch at the dinode
  * structure.
@@ -250,7 +454,7 @@ bail:
  * for the new last extent block.
  *
  * the new branch will be 'empty' in the sense that every block will
- * contain a single record with e_clusters == 0.
+ * contain a single record with cluster count == 0.
  */
 static int ocfs2_add_branch(struct ocfs2_super *osb,
 			    handle_t *handle,
@@ -268,6 +472,7 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 	struct ocfs2_extent_block *eb;
 	struct ocfs2_extent_list  *eb_el;
 	struct ocfs2_extent_list  *el;
+	u32 new_cpos;
 
 	mlog_entry_void();
 
@@ -302,6 +507,9 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 		goto bail;
 	}
 
+	eb = (struct ocfs2_extent_block *)last_eb_bh->b_data;
+	new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list);
+
 	/* Note: new_eb_bhs[new_blocks - 1] is the guy which will be
 	 * linked with the rest of the tree.
 	 * conversly, new_eb_bhs[0] is the new bottommost leaf.
@@ -330,9 +538,18 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 		eb->h_next_leaf_blk = 0;
 		eb_el->l_tree_depth = cpu_to_le16(i);
 		eb_el->l_next_free_rec = cpu_to_le16(1);
-		eb_el->l_recs[0].e_cpos = fe->i_clusters;
+		/*
+		 * This actually counts as an empty extent as
+		 * c_clusters == 0
+		 */
+		eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos);
 		eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno);
-		eb_el->l_recs[0].e_clusters = cpu_to_le32(0);
+		/*
+		 * eb_el isn't always an interior node, but even leaf
+		 * nodes want a zero'd flags and reserved field so
+		 * this gets the whole 32 bits regardless of use.
+		 */
+		eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0);
 		if (!eb_el->l_tree_depth)
 			new_last_eb_blk = le64_to_cpu(eb->h_blkno);
 
@@ -376,8 +593,8 @@ static int ocfs2_add_branch(struct ocfs2_super *osb,
 	 * either be on the fe, or the extent block passed in. */
 	i = le16_to_cpu(el->l_next_free_rec);
 	el->l_recs[i].e_blkno = cpu_to_le64(next_blkno);
-	el->l_recs[i].e_cpos = fe->i_clusters;
-	el->l_recs[i].e_clusters = 0;
+	el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
+	el->l_recs[i].e_int_clusters = 0;
 	le16_add_cpu(&el->l_next_free_rec, 1);
 
 	/* fe needs a new last extent block pointer, as does the
@@ -425,6 +642,7 @@ static int ocfs2_shift_tree_depth(struct ocfs2_super *osb,
 				  struct buffer_head **ret_new_eb_bh)
 {
 	int status, i;
+	u32 new_clusters;
 	struct buffer_head *new_eb_bh = NULL;
 	struct ocfs2_dinode *fe;
 	struct ocfs2_extent_block *eb;
@@ -461,11 +679,8 @@ static int ocfs2_shift_tree_depth(struct ocfs2_super *osb,
 	/* copy the fe data into the new extent block */
 	eb_el->l_tree_depth = fe_el->l_tree_depth;
 	eb_el->l_next_free_rec = fe_el->l_next_free_rec;
-	for(i = 0; i < le16_to_cpu(fe_el->l_next_free_rec); i++) {
-		eb_el->l_recs[i].e_cpos = fe_el->l_recs[i].e_cpos;
-		eb_el->l_recs[i].e_clusters = fe_el->l_recs[i].e_clusters;
-		eb_el->l_recs[i].e_blkno = fe_el->l_recs[i].e_blkno;
-	}
+	for(i = 0; i < le16_to_cpu(fe_el->l_next_free_rec); i++)
+		eb_el->l_recs[i] = fe_el->l_recs[i];
 
 	status = ocfs2_journal_dirty(handle, new_eb_bh);
 	if (status < 0) {
@@ -480,16 +695,15 @@ static int ocfs2_shift_tree_depth(struct ocfs2_super *osb,
 		goto bail;
 	}
 
+	new_clusters = ocfs2_sum_rightmost_rec(eb_el);
+
 	/* update fe now */
 	le16_add_cpu(&fe_el->l_tree_depth, 1);
 	fe_el->l_recs[0].e_cpos = 0;
 	fe_el->l_recs[0].e_blkno = eb->h_blkno;
-	fe_el->l_recs[0].e_clusters = fe->i_clusters;
-	for(i = 1; i < le16_to_cpu(fe_el->l_next_free_rec); i++) {
-		fe_el->l_recs[i].e_cpos = 0;
-		fe_el->l_recs[i].e_clusters = 0;
-		fe_el->l_recs[i].e_blkno = 0;
-	}
+	fe_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters);
+	for(i = 1; i < le16_to_cpu(fe_el->l_next_free_rec); i++)
+		memset(&fe_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
 	fe_el->l_next_free_rec = cpu_to_le16(1);
 
 	/* If this is our 1st tree depth shift, then last_eb_blk
@@ -515,199 +729,6 @@ bail:
 }
 
 /*
- * Expects the tree to already have room in the rightmost leaf for the
- * extent.  Updates all the extent blocks (and the dinode) on the way
- * down.
- */
-static int ocfs2_do_insert_extent(struct ocfs2_super *osb,
-				  handle_t *handle,
-				  struct inode *inode,
-				  struct buffer_head *fe_bh,
-				  u64 start_blk,
-				  u32 new_clusters)
-{
-	int status, i, num_bhs = 0;
-	u64 next_blkno;
-	u16 next_free;
-	struct buffer_head **eb_bhs = NULL;
-	struct ocfs2_dinode *fe;
-	struct ocfs2_extent_block *eb;
-	struct ocfs2_extent_list  *el;
-
-	mlog_entry_void();
-
-	status = ocfs2_journal_access(handle, inode, fe_bh,
-				      OCFS2_JOURNAL_ACCESS_WRITE);
-	if (status < 0) {
-		mlog_errno(status);
-		goto bail;
-	}
-
-	fe = (struct ocfs2_dinode *) fe_bh->b_data;
-	el = &fe->id2.i_list;
-	if (el->l_tree_depth) {
-		/* This is another operation where we want to be
-		 * careful about our tree updates. An error here means
-		 * none of the previous changes we made should roll
-		 * forward. As a result, we have to record the buffers
-		 * for this part of the tree in an array and reserve a
-		 * journal write to them before making any changes. */
-		num_bhs = le16_to_cpu(fe->id2.i_list.l_tree_depth);
-		eb_bhs = kcalloc(num_bhs, sizeof(struct buffer_head *),
-				 GFP_KERNEL);
-		if (!eb_bhs) {
-			status = -ENOMEM;
-			mlog_errno(status);
-			goto bail;
-		}
-
-		i = 0;
-		while(el->l_tree_depth) {
-			next_free = le16_to_cpu(el->l_next_free_rec);
-			if (next_free == 0) {
-				ocfs2_error(inode->i_sb,
-					    "Dinode %llu has a bad extent list",
-					    (unsigned long long)OCFS2_I(inode)->ip_blkno);
-				status = -EIO;
-				goto bail;
-			}
-			next_blkno = le64_to_cpu(el->l_recs[next_free - 1].e_blkno);
-
-			BUG_ON(i >= num_bhs);
-			status = ocfs2_read_block(osb, next_blkno, &eb_bhs[i],
-						  OCFS2_BH_CACHED, inode);
-			if (status < 0) {
-				mlog_errno(status);
-				goto bail;
-			}
-			eb = (struct ocfs2_extent_block *) eb_bhs[i]->b_data;
-			if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
-				OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb,
-								 eb);
-				status = -EIO;
-				goto bail;
-			}
-
-			status = ocfs2_journal_access(handle, inode, eb_bhs[i],
-						      OCFS2_JOURNAL_ACCESS_WRITE);
-			if (status < 0) {
-				mlog_errno(status);
-				goto bail;
-			}
-
-			el = &eb->h_list;
-			i++;
-			/* When we leave this loop, eb_bhs[num_bhs - 1] will
-			 * hold the bottom-most leaf extent block. */
-		}
-		BUG_ON(el->l_tree_depth);
-
-		el = &fe->id2.i_list;
-		/* If we have tree depth, then the fe update is
-		 * trivial, and we want to switch el out for the
-		 * bottom-most leaf in order to update it with the
-		 * actual extent data below. */
-		next_free = le16_to_cpu(el->l_next_free_rec);
-		if (next_free == 0) {
-			ocfs2_error(inode->i_sb,
-				    "Dinode %llu has a bad extent list",
-				    (unsigned long long)OCFS2_I(inode)->ip_blkno);
-			status = -EIO;
-			goto bail;
-		}
-		le32_add_cpu(&el->l_recs[next_free - 1].e_clusters,
-			     new_clusters);
-		/* (num_bhs - 1) to avoid the leaf */
-		for(i = 0; i < (num_bhs - 1); i++) {
-			eb = (struct ocfs2_extent_block *) eb_bhs[i]->b_data;
-			el = &eb->h_list;
-
-			/* finally, make our actual change to the
-			 * intermediate extent blocks. */
-			next_free = le16_to_cpu(el->l_next_free_rec);
-			le32_add_cpu(&el->l_recs[next_free - 1].e_clusters,
-				     new_clusters);
-
-			status = ocfs2_journal_dirty(handle, eb_bhs[i]);
-			if (status < 0)
-				mlog_errno(status);
-		}
-		BUG_ON(i != (num_bhs - 1));
-		/* note that the leaf block wasn't touched in
-		 * the loop above */
-		eb = (struct ocfs2_extent_block *) eb_bhs[num_bhs - 1]->b_data;
-		el = &eb->h_list;
-		BUG_ON(el->l_tree_depth);
-	}
-
-	/* yay, we can finally add the actual extent now! */
-	i = le16_to_cpu(el->l_next_free_rec) - 1;
-	if (le16_to_cpu(el->l_next_free_rec) &&
-	    ocfs2_extent_contig(inode, &el->l_recs[i], start_blk)) {
-		le32_add_cpu(&el->l_recs[i].e_clusters, new_clusters);
-	} else if (le16_to_cpu(el->l_next_free_rec) &&
-		   (le32_to_cpu(el->l_recs[i].e_clusters) == 0)) {
-		/* having an empty extent at eof is legal. */
-		if (el->l_recs[i].e_cpos != fe->i_clusters) {
-			ocfs2_error(inode->i_sb,
-				    "Dinode %llu trailing extent is bad: "
-				    "cpos (%u) != number of clusters (%u)",
-				    (unsigned long long)OCFS2_I(inode)->ip_blkno,
-				    le32_to_cpu(el->l_recs[i].e_cpos),
-				    le32_to_cpu(fe->i_clusters));
-			status = -EIO;
-			goto bail;
-		}
-		el->l_recs[i].e_blkno = cpu_to_le64(start_blk);
-		el->l_recs[i].e_clusters = cpu_to_le32(new_clusters);
-	} else {
-		/* No contiguous record, or no empty record at eof, so
-		 * we add a new one. */
-
-		BUG_ON(le16_to_cpu(el->l_next_free_rec) >=
-		       le16_to_cpu(el->l_count));
-		i = le16_to_cpu(el->l_next_free_rec);
-
-		el->l_recs[i].e_blkno = cpu_to_le64(start_blk);
-		el->l_recs[i].e_clusters = cpu_to_le32(new_clusters);
-		el->l_recs[i].e_cpos = fe->i_clusters;
-		le16_add_cpu(&el->l_next_free_rec, 1);
-	}
-
-	/*
-	 * extent_map errors are not fatal, so they are ignored outside
-	 * of flushing the thing.
-	 */
-	status = ocfs2_extent_map_append(inode, &el->l_recs[i],
-					 new_clusters);
-	if (status) {
-		mlog_errno(status);
-		ocfs2_extent_map_drop(inode, le32_to_cpu(fe->i_clusters));
-	}
-
-	status = ocfs2_journal_dirty(handle, fe_bh);
-	if (status < 0)
-		mlog_errno(status);
-	if (fe->id2.i_list.l_tree_depth) {
-		status = ocfs2_journal_dirty(handle, eb_bhs[num_bhs - 1]);
-		if (status < 0)
-			mlog_errno(status);
-	}
-
-	status = 0;
-bail:
-	if (eb_bhs) {
-		for (i = 0; i < num_bhs; i++)
-			if (eb_bhs[i])
-				brelse(eb_bhs[i]);
-		kfree(eb_bhs);
-	}
-
-	mlog_exit(status);
-	return status;
-}
-
-/*
  * Should only be called when there is no space left in any of the
  * leaf nodes. What we want to do is find the lowest tree depth
  * non-leaf extent block with room for new records. There are three
@@ -807,53 +828,1548 @@ bail:
 	return status;
 }
 
-/* the caller needs to update fe->i_clusters */
-int ocfs2_insert_extent(struct ocfs2_super *osb,
-			handle_t *handle,
-			struct inode *inode,
-			struct buffer_head *fe_bh,
-			u64 start_blk,
-			u32 new_clusters,
-			struct ocfs2_alloc_context *meta_ac)
+/*
+ * This is only valid for leaf nodes, which are the only ones that can
+ * have empty extents anyway.
+ */
+static inline int ocfs2_is_empty_extent(struct ocfs2_extent_rec *rec)
 {
-	int status, i, shift;
-	struct buffer_head *last_eb_bh = NULL;
+	return !rec->e_leaf_clusters;
+}
+
+/*
+ * This function will discard the rightmost extent record.
+ */
+static void ocfs2_shift_records_right(struct ocfs2_extent_list *el)
+{
+	int next_free = le16_to_cpu(el->l_next_free_rec);
+	int count = le16_to_cpu(el->l_count);
+	unsigned int num_bytes;
+
+	BUG_ON(!next_free);
+	/* This will cause us to go off the end of our extent list. */
+	BUG_ON(next_free >= count);
+
+	num_bytes = sizeof(struct ocfs2_extent_rec) * next_free;
+
+	memmove(&el->l_recs[1], &el->l_recs[0], num_bytes);
+}
+
+static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el,
+			      struct ocfs2_extent_rec *insert_rec)
+{
+	int i, insert_index, next_free, has_empty, num_bytes;
+	u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos);
+	struct ocfs2_extent_rec *rec;
+
+	next_free = le16_to_cpu(el->l_next_free_rec);
+	has_empty = ocfs2_is_empty_extent(&el->l_recs[0]);
+
+	BUG_ON(!next_free);
+
+	/* The tree code before us didn't allow enough room in the leaf. */
+	if (el->l_next_free_rec == el->l_count && !has_empty)
+		BUG();
+
+	/*
+	 * The easiest way to approach this is to just remove the
+	 * empty extent and temporarily decrement next_free.
+	 */
+	if (has_empty) {
+		/*
+		 * If next_free was 1 (only an empty extent), this
+		 * loop won't execute, which is fine. We still want
+		 * the decrement above to happen.
+		 */
+		for(i = 0; i < (next_free - 1); i++)
+			el->l_recs[i] = el->l_recs[i+1];
+
+		next_free--;
+	}
+
+	/*
+	 * Figure out what the new record index should be.
+	 */
+	for(i = 0; i < next_free; i++) {
+		rec = &el->l_recs[i];
+
+		if (insert_cpos < le32_to_cpu(rec->e_cpos))
+			break;
+	}
+	insert_index = i;
+
+	mlog(0, "ins %u: index %d, has_empty %d, next_free %d, count %d\n",
+	     insert_cpos, insert_index, has_empty, next_free, le16_to_cpu(el->l_count));
+
+	BUG_ON(insert_index < 0);
+	BUG_ON(insert_index >= le16_to_cpu(el->l_count));
+	BUG_ON(insert_index > next_free);
+
+	/*
+	 * No need to memmove if we're just adding to the tail.
+	 */
+	if (insert_index != next_free) {
+		BUG_ON(next_free >= le16_to_cpu(el->l_count));
+
+		num_bytes = next_free - insert_index;
+		num_bytes *= sizeof(struct ocfs2_extent_rec);
+		memmove(&el->l_recs[insert_index + 1],
+			&el->l_recs[insert_index],
+			num_bytes);
+	}
+
+	/*
+	 * Either we had an empty extent, and need to re-increment or
+	 * there was no empty extent on a non full rightmost leaf node,
+	 * in which case we still need to increment.
+	 */
+	next_free++;
+	el->l_next_free_rec = cpu_to_le16(next_free);
+	/*
+	 * Make sure none of the math above just messed up our tree.
+	 */
+	BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count));
+
+	el->l_recs[insert_index] = *insert_rec;
+
+}
+
+/*
+ * Create an empty extent record .
+ *
+ * l_next_free_rec may be updated.
+ *
+ * If an empty extent already exists do nothing.
+ */
+static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el)
+{
+	int next_free = le16_to_cpu(el->l_next_free_rec);
+
+	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+
+	if (next_free == 0)
+		goto set_and_inc;
+
+	if (ocfs2_is_empty_extent(&el->l_recs[0]))
+		return;
+
+	mlog_bug_on_msg(el->l_count == el->l_next_free_rec,
+			"Asked to create an empty extent in a full list:\n"
+			"count = %u, tree depth = %u",
+			le16_to_cpu(el->l_count),
+			le16_to_cpu(el->l_tree_depth));
+
+	ocfs2_shift_records_right(el);
+
+set_and_inc:
+	le16_add_cpu(&el->l_next_free_rec, 1);
+	memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
+}
+
+/*
+ * For a rotation which involves two leaf nodes, the "root node" is
+ * the lowest level tree node which contains a path to both leafs. This
+ * resulting set of information can be used to form a complete "subtree"
+ *
+ * This function is passed two full paths from the dinode down to a
+ * pair of adjacent leaves. It's task is to figure out which path
+ * index contains the subtree root - this can be the root index itself
+ * in a worst-case rotation.
+ *
+ * The array index of the subtree root is passed back.
+ */
+static int ocfs2_find_subtree_root(struct inode *inode,
+				   struct ocfs2_path *left,
+				   struct ocfs2_path *right)
+{
+	int i = 0;
+
+	/*
+	 * Check that the caller passed in two paths from the same tree.
+	 */
+	BUG_ON(path_root_bh(left) != path_root_bh(right));
+
+	do {
+		i++;
+
+		/*
+		 * The caller didn't pass two adjacent paths.
+		 */
+		mlog_bug_on_msg(i > left->p_tree_depth,
+				"Inode %lu, left depth %u, right depth %u\n"
+				"left leaf blk %llu, right leaf blk %llu\n",
+				inode->i_ino, left->p_tree_depth,
+				right->p_tree_depth,
+				(unsigned long long)path_leaf_bh(left)->b_blocknr,
+				(unsigned long long)path_leaf_bh(right)->b_blocknr);
+	} while (left->p_node[i].bh->b_blocknr ==
+		 right->p_node[i].bh->b_blocknr);
+
+	return i - 1;
+}
+
+typedef void (path_insert_t)(void *, struct buffer_head *);
+
+/*
+ * Traverse a btree path in search of cpos, starting at root_el.
+ *
+ * This code can be called with a cpos larger than the tree, in which
+ * case it will return the rightmost path.
+ */
+static int __ocfs2_find_path(struct inode *inode,
+			     struct ocfs2_extent_list *root_el, u32 cpos,
+			     path_insert_t *func, void *data)
+{
+	int i, ret = 0;
+	u32 range;
+	u64 blkno;
 	struct buffer_head *bh = NULL;
-	struct ocfs2_dinode *fe;
 	struct ocfs2_extent_block *eb;
-	struct ocfs2_extent_list  *el;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_rec *rec;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 
-	mlog_entry_void();
+	el = root_el;
+	while (el->l_tree_depth) {
+		if (le16_to_cpu(el->l_next_free_rec) == 0) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %llu has empty extent list at "
+				    "depth %u\n",
+				    (unsigned long long)oi->ip_blkno,
+				    le16_to_cpu(el->l_tree_depth));
+			ret = -EROFS;
+			goto out;
 
-	mlog(0, "add %u clusters starting at block %llu to inode %llu\n",
-	     new_clusters, (unsigned long long)start_blk,
-	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
+		}
 
-	fe = (struct ocfs2_dinode *) fe_bh->b_data;
-	el = &fe->id2.i_list;
+		for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) {
+			rec = &el->l_recs[i];
+
+			/*
+			 * In the case that cpos is off the allocation
+			 * tree, this should just wind up returning the
+			 * rightmost record.
+			 */
+			range = le32_to_cpu(rec->e_cpos) +
+				ocfs2_rec_clusters(el, rec);
+			if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
+			    break;
+		}
 
-	if (el->l_tree_depth) {
-		/* jump to end of tree */
-		status = ocfs2_read_block(osb, le64_to_cpu(fe->i_last_eb_blk),
-					  &last_eb_bh, OCFS2_BH_CACHED, inode);
-		if (status < 0) {
-			mlog_exit(status);
-			goto bail;
+		blkno = le64_to_cpu(el->l_recs[i].e_blkno);
+		if (blkno == 0) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %llu has bad blkno in extent list "
+				    "at depth %u (index %d)\n",
+				    (unsigned long long)oi->ip_blkno,
+				    le16_to_cpu(el->l_tree_depth), i);
+			ret = -EROFS;
+			goto out;
 		}
-		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
+
+		brelse(bh);
+		bh = NULL;
+		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
+				       &bh, OCFS2_BH_CACHED, inode);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) bh->b_data;
 		el = &eb->h_list;
+		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
+			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
+			ret = -EIO;
+			goto out;
+		}
+
+		if (le16_to_cpu(el->l_next_free_rec) >
+		    le16_to_cpu(el->l_count)) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %llu has bad count in extent list "
+				    "at block %llu (next free=%u, count=%u)\n",
+				    (unsigned long long)oi->ip_blkno,
+				    (unsigned long long)bh->b_blocknr,
+				    le16_to_cpu(el->l_next_free_rec),
+				    le16_to_cpu(el->l_count));
+			ret = -EROFS;
+			goto out;
+		}
+
+		if (func)
+			func(data, bh);
+	}
+
+out:
+	/*
+	 * Catch any trailing bh that the loop didn't handle.
+	 */
+	brelse(bh);
+
+	return ret;
+}
+
+/*
+ * Given an initialized path (that is, it has a valid root extent
+ * list), this function will traverse the btree in search of the path
+ * which would contain cpos.
+ *
+ * The path traveled is recorded in the path structure.
+ *
+ * Note that this will not do any comparisons on leaf node extent
+ * records, so it will work fine in the case that we just added a tree
+ * branch.
+ */
+struct find_path_data {
+	int index;
+	struct ocfs2_path *path;
+};
+static void find_path_ins(void *data, struct buffer_head *bh)
+{
+	struct find_path_data *fp = data;
+
+	get_bh(bh);
+	ocfs2_path_insert_eb(fp->path, fp->index, bh);
+	fp->index++;
+}
+static int ocfs2_find_path(struct inode *inode, struct ocfs2_path *path,
+			   u32 cpos)
+{
+	struct find_path_data data;
+
+	data.index = 1;
+	data.path = path;
+	return __ocfs2_find_path(inode, path_root_el(path), cpos,
+				 find_path_ins, &data);
+}
+
+static void find_leaf_ins(void *data, struct buffer_head *bh)
+{
+	struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data;
+	struct ocfs2_extent_list *el = &eb->h_list;
+	struct buffer_head **ret = data;
+
+	/* We want to retain only the leaf block. */
+	if (le16_to_cpu(el->l_tree_depth) == 0) {
+		get_bh(bh);
+		*ret = bh;
+	}
+}
+/*
+ * Find the leaf block in the tree which would contain cpos. No
+ * checking of the actual leaf is done.
+ *
+ * Some paths want to call this instead of allocating a path structure
+ * and calling ocfs2_find_path().
+ *
+ * This function doesn't handle non btree extent lists.
+ */
+int ocfs2_find_leaf(struct inode *inode, struct ocfs2_extent_list *root_el,
+		    u32 cpos, struct buffer_head **leaf_bh)
+{
+	int ret;
+	struct buffer_head *bh = NULL;
+
+	ret = __ocfs2_find_path(inode, root_el, cpos, find_leaf_ins, &bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	*leaf_bh = bh;
+out:
+	return ret;
+}
+
+/*
+ * Adjust the adjacent records (left_rec, right_rec) involved in a rotation.
+ *
+ * Basically, we've moved stuff around at the bottom of the tree and
+ * we need to fix up the extent records above the changes to reflect
+ * the new changes.
+ *
+ * left_rec: the record on the left.
+ * left_child_el: is the child list pointed to by left_rec
+ * right_rec: the record to the right of left_rec
+ * right_child_el: is the child list pointed to by right_rec
+ *
+ * By definition, this only works on interior nodes.
+ */
+static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec,
+				  struct ocfs2_extent_list *left_child_el,
+				  struct ocfs2_extent_rec *right_rec,
+				  struct ocfs2_extent_list *right_child_el)
+{
+	u32 left_clusters, right_end;
+
+	/*
+	 * Interior nodes never have holes. Their cpos is the cpos of
+	 * the leftmost record in their child list. Their cluster
+	 * count covers the full theoretical range of their child list
+	 * - the range between their cpos and the cpos of the record
+	 * immediately to their right.
+	 */
+	left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos);
+	left_clusters -= le32_to_cpu(left_rec->e_cpos);
+	left_rec->e_int_clusters = cpu_to_le32(left_clusters);
+
+	/*
+	 * Calculate the rightmost cluster count boundary before
+	 * moving cpos - we will need to adjust clusters after
+	 * updating e_cpos to keep the same highest cluster count.
+	 */
+	right_end = le32_to_cpu(right_rec->e_cpos);
+	right_end += le32_to_cpu(right_rec->e_int_clusters);
+
+	right_rec->e_cpos = left_rec->e_cpos;
+	le32_add_cpu(&right_rec->e_cpos, left_clusters);
+
+	right_end -= le32_to_cpu(right_rec->e_cpos);
+	right_rec->e_int_clusters = cpu_to_le32(right_end);
+}
+
+/*
+ * Adjust the adjacent root node records involved in a
+ * rotation. left_el_blkno is passed in as a key so that we can easily
+ * find it's index in the root list.
+ */
+static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el,
+				      struct ocfs2_extent_list *left_el,
+				      struct ocfs2_extent_list *right_el,
+				      u64 left_el_blkno)
+{
+	int i;
+
+	BUG_ON(le16_to_cpu(root_el->l_tree_depth) <=
+	       le16_to_cpu(left_el->l_tree_depth));
+
+	for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) {
+		if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno)
+			break;
+	}
+
+	/*
+	 * The path walking code should have never returned a root and
+	 * two paths which are not adjacent.
+	 */
+	BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1));
+
+	ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el,
+				      &root_el->l_recs[i + 1], right_el);
+}
+
+/*
+ * We've changed a leaf block (in right_path) and need to reflect that
+ * change back up the subtree.
+ *
+ * This happens in multiple places:
+ *   - When we've moved an extent record from the left path leaf to the right
+ *     path leaf to make room for an empty extent in the left path leaf.
+ *   - When our insert into the right path leaf is at the leftmost edge
+ *     and requires an update of the path immediately to it's left. This
+ *     can occur at the end of some types of rotation and appending inserts.
+ */
+static void ocfs2_complete_edge_insert(struct inode *inode, handle_t *handle,
+				       struct ocfs2_path *left_path,
+				       struct ocfs2_path *right_path,
+				       int subtree_index)
+{
+	int ret, i, idx;
+	struct ocfs2_extent_list *el, *left_el, *right_el;
+	struct ocfs2_extent_rec *left_rec, *right_rec;
+	struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
+
+	/*
+	 * Update the counts and position values within all the
+	 * interior nodes to reflect the leaf rotation we just did.
+	 *
+	 * The root node is handled below the loop.
+	 *
+	 * We begin the loop with right_el and left_el pointing to the
+	 * leaf lists and work our way up.
+	 *
+	 * NOTE: within this loop, left_el and right_el always refer
+	 * to the *child* lists.
+	 */
+	left_el = path_leaf_el(left_path);
+	right_el = path_leaf_el(right_path);
+	for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) {
+		mlog(0, "Adjust records at index %u\n", i);
+
+		/*
+		 * One nice property of knowing that all of these
+		 * nodes are below the root is that we only deal with
+		 * the leftmost right node record and the rightmost
+		 * left node record.
+		 */
+		el = left_path->p_node[i].el;
+		idx = le16_to_cpu(left_el->l_next_free_rec) - 1;
+		left_rec = &el->l_recs[idx];
+
+		el = right_path->p_node[i].el;
+		right_rec = &el->l_recs[0];
+
+		ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec,
+					      right_el);
+
+		ret = ocfs2_journal_dirty(handle, left_path->p_node[i].bh);
+		if (ret)
+			mlog_errno(ret);
+
+		ret = ocfs2_journal_dirty(handle, right_path->p_node[i].bh);
+		if (ret)
+			mlog_errno(ret);
+
+		/*
+		 * Setup our list pointers now so that the current
+		 * parents become children in the next iteration.
+		 */
+		left_el = left_path->p_node[i].el;
+		right_el = right_path->p_node[i].el;
+	}
+
+	/*
+	 * At the root node, adjust the two adjacent records which
+	 * begin our path to the leaves.
+	 */
+
+	el = left_path->p_node[subtree_index].el;
+	left_el = left_path->p_node[subtree_index + 1].el;
+	right_el = right_path->p_node[subtree_index + 1].el;
+
+	ocfs2_adjust_root_records(el, left_el, right_el,
+				  left_path->p_node[subtree_index + 1].bh->b_blocknr);
+
+	root_bh = left_path->p_node[subtree_index].bh;
+
+	ret = ocfs2_journal_dirty(handle, root_bh);
+	if (ret)
+		mlog_errno(ret);
+}
+
+static int ocfs2_rotate_subtree_right(struct inode *inode,
+				      handle_t *handle,
+				      struct ocfs2_path *left_path,
+				      struct ocfs2_path *right_path,
+				      int subtree_index)
+{
+	int ret, i;
+	struct buffer_head *right_leaf_bh;
+	struct buffer_head *left_leaf_bh = NULL;
+	struct buffer_head *root_bh;
+	struct ocfs2_extent_list *right_el, *left_el;
+	struct ocfs2_extent_rec move_rec;
+
+	left_leaf_bh = path_leaf_bh(left_path);
+	left_el = path_leaf_el(left_path);
+
+	if (left_el->l_next_free_rec != left_el->l_count) {
+		ocfs2_error(inode->i_sb,
+			    "Inode %llu has non-full interior leaf node %llu"
+			    "(next free = %u)",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			    (unsigned long long)left_leaf_bh->b_blocknr,
+			    le16_to_cpu(left_el->l_next_free_rec));
+		return -EROFS;
+	}
+
+	/*
+	 * This extent block may already have an empty record, so we
+	 * return early if so.
+	 */
+	if (ocfs2_is_empty_extent(&left_el->l_recs[0]))
+		return 0;
+
+	root_bh = left_path->p_node[subtree_index].bh;
+	BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
+
+	ret = ocfs2_journal_access(handle, inode, root_bh,
+				   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
+		ret = ocfs2_journal_access(handle, inode,
+					   right_path->p_node[i].bh,
+					   OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_journal_access(handle, inode,
+					   left_path->p_node[i].bh,
+					   OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	right_leaf_bh = path_leaf_bh(right_path);
+	right_el = path_leaf_el(right_path);
+
+	/* This is a code error, not a disk corruption. */
+	mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails "
+			"because rightmost leaf block %llu is empty\n",
+			(unsigned long long)OCFS2_I(inode)->ip_blkno,
+			(unsigned long long)right_leaf_bh->b_blocknr);
+
+	ocfs2_create_empty_extent(right_el);
+
+	ret = ocfs2_journal_dirty(handle, right_leaf_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* Do the copy now. */
+	i = le16_to_cpu(left_el->l_next_free_rec) - 1;
+	move_rec = left_el->l_recs[i];
+	right_el->l_recs[0] = move_rec;
+
+	/*
+	 * Clear out the record we just copied and shift everything
+	 * over, leaving an empty extent in the left leaf.
+	 *
+	 * We temporarily subtract from next_free_rec so that the
+	 * shift will lose the tail record (which is now defunct).
+	 */
+	le16_add_cpu(&left_el->l_next_free_rec, -1);
+	ocfs2_shift_records_right(left_el);
+	memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
+	le16_add_cpu(&left_el->l_next_free_rec, 1);
+
+	ret = ocfs2_journal_dirty(handle, left_leaf_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_complete_edge_insert(inode, handle, left_path, right_path,
+				subtree_index);
+
+out:
+	return ret;
+}
+
+/*
+ * Given a full path, determine what cpos value would return us a path
+ * containing the leaf immediately to the left of the current one.
+ *
+ * Will return zero if the path passed in is already the leftmost path.
+ */
+static int ocfs2_find_cpos_for_left_leaf(struct super_block *sb,
+					 struct ocfs2_path *path, u32 *cpos)
+{
+	int i, j, ret = 0;
+	u64 blkno;
+	struct ocfs2_extent_list *el;
+
+	BUG_ON(path->p_tree_depth == 0);
+
+	*cpos = 0;
+
+	blkno = path_leaf_bh(path)->b_blocknr;
+
+	/* Start at the tree node just above the leaf and work our way up. */
+	i = path->p_tree_depth - 1;
+	while (i >= 0) {
+		el = path->p_node[i].el;
+
+		/*
+		 * Find the extent record just before the one in our
+		 * path.
+		 */
+		for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
+			if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
+				if (j == 0) {
+					if (i == 0) {
+						/*
+						 * We've determined that the
+						 * path specified is already
+						 * the leftmost one - return a
+						 * cpos of zero.
+						 */
+						goto out;
+					}
+					/*
+					 * The leftmost record points to our
+					 * leaf - we need to travel up the
+					 * tree one level.
+					 */
+					goto next_node;
+				}
+
+				*cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos);
+				*cpos = *cpos + ocfs2_rec_clusters(el,
+							   &el->l_recs[j - 1]);
+				*cpos = *cpos - 1;
+				goto out;
+			}
+		}
+
+		/*
+		 * If we got here, we never found a valid node where
+		 * the tree indicated one should be.
+		 */
+		ocfs2_error(sb,
+			    "Invalid extent tree at extent block %llu\n",
+			    (unsigned long long)blkno);
+		ret = -EROFS;
+		goto out;
+
+next_node:
+		blkno = path->p_node[i].bh->b_blocknr;
+		i--;
+	}
+
+out:
+	return ret;
+}
+
+static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth,
+					   struct ocfs2_path *path)
+{
+	int credits = (path->p_tree_depth - subtree_depth) * 2 + 1;
+
+	if (handle->h_buffer_credits < credits)
+		return ocfs2_extend_trans(handle, credits);
+
+	return 0;
+}
+
+/*
+ * Trap the case where we're inserting into the theoretical range past
+ * the _actual_ left leaf range. Otherwise, we'll rotate a record
+ * whose cpos is less than ours into the right leaf.
+ *
+ * It's only necessary to look at the rightmost record of the left
+ * leaf because the logic that calls us should ensure that the
+ * theoretical ranges in the path components above the leaves are
+ * correct.
+ */
+static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path,
+						 u32 insert_cpos)
+{
+	struct ocfs2_extent_list *left_el;
+	struct ocfs2_extent_rec *rec;
+	int next_free;
+
+	left_el = path_leaf_el(left_path);
+	next_free = le16_to_cpu(left_el->l_next_free_rec);
+	rec = &left_el->l_recs[next_free - 1];
+
+	if (insert_cpos > le32_to_cpu(rec->e_cpos))
+		return 1;
+	return 0;
+}
+
+/*
+ * Rotate all the records in a btree right one record, starting at insert_cpos.
+ *
+ * The path to the rightmost leaf should be passed in.
+ *
+ * The array is assumed to be large enough to hold an entire path (tree depth).
+ *
+ * Upon succesful return from this function:
+ *
+ * - The 'right_path' array will contain a path to the leaf block
+ *   whose range contains e_cpos.
+ * - That leaf block will have a single empty extent in list index 0.
+ * - In the case that the rotation requires a post-insert update,
+ *   *ret_left_path will contain a valid path which can be passed to
+ *   ocfs2_insert_path().
+ */
+static int ocfs2_rotate_tree_right(struct inode *inode,
+				   handle_t *handle,
+				   u32 insert_cpos,
+				   struct ocfs2_path *right_path,
+				   struct ocfs2_path **ret_left_path)
+{
+	int ret, start;
+	u32 cpos;
+	struct ocfs2_path *left_path = NULL;
+
+	*ret_left_path = NULL;
+
+	left_path = ocfs2_new_path(path_root_bh(right_path),
+				   path_root_el(right_path));
+	if (!left_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path, &cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mlog(0, "Insert: %u, first left path cpos: %u\n", insert_cpos, cpos);
+
+	/*
+	 * What we want to do here is:
+	 *
+	 * 1) Start with the rightmost path.
+	 *
+	 * 2) Determine a path to the leaf block directly to the left
+	 *    of that leaf.
+	 *
+	 * 3) Determine the 'subtree root' - the lowest level tree node
+	 *    which contains a path to both leaves.
+	 *
+	 * 4) Rotate the subtree.
+	 *
+	 * 5) Find the next subtree by considering the left path to be
+	 *    the new right path.
+	 *
+	 * The check at the top of this while loop also accepts
+	 * insert_cpos == cpos because cpos is only a _theoretical_
+	 * value to get us the left path - insert_cpos might very well
+	 * be filling that hole.
+	 *
+	 * Stop at a cpos of '0' because we either started at the
+	 * leftmost branch (i.e., a tree with one branch and a
+	 * rotation inside of it), or we've gone as far as we can in
+	 * rotating subtrees.
+	 */
+	while (cpos && insert_cpos <= cpos) {
+		mlog(0, "Rotating a tree: ins. cpos: %u, left path cpos: %u\n",
+		     insert_cpos, cpos);
+
+		ret = ocfs2_find_path(inode, left_path, cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		mlog_bug_on_msg(path_leaf_bh(left_path) ==
+				path_leaf_bh(right_path),
+				"Inode %lu: error during insert of %u "
+				"(left path cpos %u) results in two identical "
+				"paths ending at %llu\n",
+				inode->i_ino, insert_cpos, cpos,
+				(unsigned long long)
+				path_leaf_bh(left_path)->b_blocknr);
+
+		if (ocfs2_rotate_requires_path_adjustment(left_path,
+							  insert_cpos)) {
+			mlog(0, "Path adjustment required\n");
+
+			/*
+			 * We've rotated the tree as much as we
+			 * should. The rest is up to
+			 * ocfs2_insert_path() to complete, after the
+			 * record insertion. We indicate this
+			 * situation by returning the left path.
+			 *
+			 * The reason we don't adjust the records here
+			 * before the record insert is that an error
+			 * later might break the rule where a parent
+			 * record e_cpos will reflect the actual
+			 * e_cpos of the 1st nonempty record of the
+			 * child list.
+			 */
+			*ret_left_path = left_path;
+			goto out_ret_path;
+		}
+
+		start = ocfs2_find_subtree_root(inode, left_path, right_path);
+
+		mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n",
+		     start,
+		     (unsigned long long) right_path->p_node[start].bh->b_blocknr,
+		     right_path->p_tree_depth);
+
+		ret = ocfs2_extend_rotate_transaction(handle, start,
+						      right_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_rotate_subtree_right(inode, handle, left_path,
+						 right_path, start);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * There is no need to re-read the next right path
+		 * as we know that it'll be our current left
+		 * path. Optimize by copying values instead.
+		 */
+		ocfs2_mv_path(right_path, left_path);
+
+		ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path,
+						    &cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+out:
+	ocfs2_free_path(left_path);
+
+out_ret_path:
+	return ret;
+}
+
+/*
+ * Do the final bits of extent record insertion at the target leaf
+ * list. If this leaf is part of an allocation tree, it is assumed
+ * that the tree above has been prepared.
+ */
+static void ocfs2_insert_at_leaf(struct ocfs2_extent_rec *insert_rec,
+				 struct ocfs2_extent_list *el,
+				 struct ocfs2_insert_type *insert,
+				 struct inode *inode)
+{
+	int i = insert->ins_contig_index;
+	unsigned int range;
+	struct ocfs2_extent_rec *rec;
+
+	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+
+	/*
+	 * Contiguous insert - either left or right.
+	 */
+	if (insert->ins_contig != CONTIG_NONE) {
+		rec = &el->l_recs[i];
+		if (insert->ins_contig == CONTIG_LEFT) {
+			rec->e_blkno = insert_rec->e_blkno;
+			rec->e_cpos = insert_rec->e_cpos;
+		}
+		le16_add_cpu(&rec->e_leaf_clusters,
+			     le16_to_cpu(insert_rec->e_leaf_clusters));
+		return;
+	}
+
+	/*
+	 * Handle insert into an empty leaf.
+	 */
+	if (le16_to_cpu(el->l_next_free_rec) == 0 ||
+	    ((le16_to_cpu(el->l_next_free_rec) == 1) &&
+	     ocfs2_is_empty_extent(&el->l_recs[0]))) {
+		el->l_recs[0] = *insert_rec;
+		el->l_next_free_rec = cpu_to_le16(1);
+		return;
+	}
+
+	/*
+	 * Appending insert.
+	 */
+	if (insert->ins_appending == APPEND_TAIL) {
+		i = le16_to_cpu(el->l_next_free_rec) - 1;
+		rec = &el->l_recs[i];
+		range = le32_to_cpu(rec->e_cpos)
+			+ le16_to_cpu(rec->e_leaf_clusters);
+		BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range);
+
+		mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >=
+				le16_to_cpu(el->l_count),
+				"inode %lu, depth %u, count %u, next free %u, "
+				"rec.cpos %u, rec.clusters %u, "
+				"insert.cpos %u, insert.clusters %u\n",
+				inode->i_ino,
+				le16_to_cpu(el->l_tree_depth),
+				le16_to_cpu(el->l_count),
+				le16_to_cpu(el->l_next_free_rec),
+				le32_to_cpu(el->l_recs[i].e_cpos),
+				le16_to_cpu(el->l_recs[i].e_leaf_clusters),
+				le32_to_cpu(insert_rec->e_cpos),
+				le16_to_cpu(insert_rec->e_leaf_clusters));
+		i++;
+		el->l_recs[i] = *insert_rec;
+		le16_add_cpu(&el->l_next_free_rec, 1);
+		return;
+	}
+
+	/*
+	 * Ok, we have to rotate.
+	 *
+	 * At this point, it is safe to assume that inserting into an
+	 * empty leaf and appending to a leaf have both been handled
+	 * above.
+	 *
+	 * This leaf needs to have space, either by the empty 1st
+	 * extent record, or by virtue of an l_next_rec < l_count.
+	 */
+	ocfs2_rotate_leaf(el, insert_rec);
+}
+
+static inline void ocfs2_update_dinode_clusters(struct inode *inode,
+						struct ocfs2_dinode *di,
+						u32 clusters)
+{
+	le32_add_cpu(&di->i_clusters, clusters);
+	spin_lock(&OCFS2_I(inode)->ip_lock);
+	OCFS2_I(inode)->ip_clusters = le32_to_cpu(di->i_clusters);
+	spin_unlock(&OCFS2_I(inode)->ip_lock);
+}
+
+static int ocfs2_append_rec_to_path(struct inode *inode, handle_t *handle,
+				    struct ocfs2_extent_rec *insert_rec,
+				    struct ocfs2_path *right_path,
+				    struct ocfs2_path **ret_left_path)
+{
+	int ret, i, next_free;
+	struct buffer_head *bh;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_path *left_path = NULL;
+
+	*ret_left_path = NULL;
+
+	/*
+	 * This shouldn't happen for non-trees. The extent rec cluster
+	 * count manipulation below only works for interior nodes.
+	 */
+	BUG_ON(right_path->p_tree_depth == 0);
+
+	/*
+	 * If our appending insert is at the leftmost edge of a leaf,
+	 * then we might need to update the rightmost records of the
+	 * neighboring path.
+	 */
+	el = path_leaf_el(right_path);
+	next_free = le16_to_cpu(el->l_next_free_rec);
+	if (next_free == 0 ||
+	    (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) {
+		u32 left_cpos;
+
+		ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path,
+						    &left_cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		mlog(0, "Append may need a left path update. cpos: %u, "
+		     "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos),
+		     left_cpos);
+
+		/*
+		 * No need to worry if the append is already in the
+		 * leftmost leaf.
+		 */
+		if (left_cpos) {
+			left_path = ocfs2_new_path(path_root_bh(right_path),
+						   path_root_el(right_path));
+			if (!left_path) {
+				ret = -ENOMEM;
+				mlog_errno(ret);
+				goto out;
+			}
+
+			ret = ocfs2_find_path(inode, left_path, left_cpos);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			/*
+			 * ocfs2_insert_path() will pass the left_path to the
+			 * journal for us.
+			 */
+		}
+	}
+
+	ret = ocfs2_journal_access_path(inode, handle, right_path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	el = path_root_el(right_path);
+	bh = path_root_bh(right_path);
+	i = 0;
+	while (1) {
+		struct ocfs2_extent_rec *rec;
+
+		next_free = le16_to_cpu(el->l_next_free_rec);
+		if (next_free == 0) {
+			ocfs2_error(inode->i_sb,
+				    "Dinode %llu has a bad extent list",
+				    (unsigned long long)OCFS2_I(inode)->ip_blkno);
+			ret = -EIO;
+			goto out;
+		}
+
+		rec = &el->l_recs[next_free - 1];
+
+		rec->e_int_clusters = insert_rec->e_cpos;
+		le32_add_cpu(&rec->e_int_clusters,
+			     le16_to_cpu(insert_rec->e_leaf_clusters));
+		le32_add_cpu(&rec->e_int_clusters,
+			     -le32_to_cpu(rec->e_cpos));
+
+		ret = ocfs2_journal_dirty(handle, bh);
+		if (ret)
+			mlog_errno(ret);
+
+		/* Don't touch the leaf node */
+		if (++i >= right_path->p_tree_depth)
+			break;
+
+		bh = right_path->p_node[i].bh;
+		el = right_path->p_node[i].el;
+	}
+
+	*ret_left_path = left_path;
+	ret = 0;
+out:
+	if (ret != 0)
+		ocfs2_free_path(left_path);
+
+	return ret;
+}
+
+/*
+ * This function only does inserts on an allocation b-tree. For dinode
+ * lists, ocfs2_insert_at_leaf() is called directly.
+ *
+ * right_path is the path we want to do the actual insert
+ * in. left_path should only be passed in if we need to update that
+ * portion of the tree after an edge insert.
+ */
+static int ocfs2_insert_path(struct inode *inode,
+			     handle_t *handle,
+			     struct ocfs2_path *left_path,
+			     struct ocfs2_path *right_path,
+			     struct ocfs2_extent_rec *insert_rec,
+			     struct ocfs2_insert_type *insert)
+{
+	int ret, subtree_index;
+	struct buffer_head *leaf_bh = path_leaf_bh(right_path);
+	struct ocfs2_extent_list *el;
+
+	/*
+	 * Pass both paths to the journal. The majority of inserts
+	 * will be touching all components anyway.
+	 */
+	ret = ocfs2_journal_access_path(inode, handle, right_path);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (left_path) {
+		int credits = handle->h_buffer_credits;
+
+		/*
+		 * There's a chance that left_path got passed back to
+		 * us without being accounted for in the
+		 * journal. Extend our transaction here to be sure we
+		 * can change those blocks.
+		 */
+		credits += left_path->p_tree_depth;
+
+		ret = ocfs2_extend_trans(handle, credits);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_journal_access_path(inode, handle, left_path);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	el = path_leaf_el(right_path);
+
+	ocfs2_insert_at_leaf(insert_rec, el, insert, inode);
+	ret = ocfs2_journal_dirty(handle, leaf_bh);
+	if (ret)
+		mlog_errno(ret);
+
+	if (left_path) {
+		/*
+		 * The rotate code has indicated that we need to fix
+		 * up portions of the tree after the insert.
+		 *
+		 * XXX: Should we extend the transaction here?
+		 */
+		subtree_index = ocfs2_find_subtree_root(inode, left_path,
+							right_path);
+		ocfs2_complete_edge_insert(inode, handle, left_path,
+					   right_path, subtree_index);
+	}
+
+	ret = 0;
+out:
+	return ret;
+}
+
+static int ocfs2_do_insert_extent(struct inode *inode,
+				  handle_t *handle,
+				  struct buffer_head *di_bh,
+				  struct ocfs2_extent_rec *insert_rec,
+				  struct ocfs2_insert_type *type)
+{
+	int ret, rotate = 0;
+	u32 cpos;
+	struct ocfs2_path *right_path = NULL;
+	struct ocfs2_path *left_path = NULL;
+	struct ocfs2_dinode *di;
+	struct ocfs2_extent_list *el;
+
+	di = (struct ocfs2_dinode *) di_bh->b_data;
+	el = &di->id2.i_list;
+
+	ret = ocfs2_journal_access(handle, inode, di_bh,
+				   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (le16_to_cpu(el->l_tree_depth) == 0) {
+		ocfs2_insert_at_leaf(insert_rec, el, type, inode);
+		goto out_update_clusters;
+	}
+
+	right_path = ocfs2_new_inode_path(di_bh);
+	if (!right_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Determine the path to start with. Rotations need the
+	 * rightmost path, everything else can go directly to the
+	 * target leaf.
+	 */
+	cpos = le32_to_cpu(insert_rec->e_cpos);
+	if (type->ins_appending == APPEND_NONE &&
+	    type->ins_contig == CONTIG_NONE) {
+		rotate = 1;
+		cpos = UINT_MAX;
+	}
+
+	ret = ocfs2_find_path(inode, right_path, cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Rotations and appends need special treatment - they modify
+	 * parts of the tree's above them.
+	 *
+	 * Both might pass back a path immediate to the left of the
+	 * one being inserted to. This will be cause
+	 * ocfs2_insert_path() to modify the rightmost records of
+	 * left_path to account for an edge insert.
+	 *
+	 * XXX: When modifying this code, keep in mind that an insert
+	 * can wind up skipping both of these two special cases...
+	 */
+	if (rotate) {
+		ret = ocfs2_rotate_tree_right(inode, handle,
+					      le32_to_cpu(insert_rec->e_cpos),
+					      right_path, &left_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	} else if (type->ins_appending == APPEND_TAIL
+		   && type->ins_contig != CONTIG_LEFT) {
+		ret = ocfs2_append_rec_to_path(inode, handle, insert_rec,
+					       right_path, &left_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	ret = ocfs2_insert_path(inode, handle, left_path, right_path,
+				insert_rec, type);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+out_update_clusters:
+	ocfs2_update_dinode_clusters(inode, di,
+				     le16_to_cpu(insert_rec->e_leaf_clusters));
+
+	ret = ocfs2_journal_dirty(handle, di_bh);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	ocfs2_free_path(left_path);
+	ocfs2_free_path(right_path);
+
+	return ret;
+}
+
+static void ocfs2_figure_contig_type(struct inode *inode,
+				     struct ocfs2_insert_type *insert,
+				     struct ocfs2_extent_list *el,
+				     struct ocfs2_extent_rec *insert_rec)
+{
+	int i;
+	enum ocfs2_contig_type contig_type = CONTIG_NONE;
+
+	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+
+	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
+		contig_type = ocfs2_extent_contig(inode, &el->l_recs[i],
+						  insert_rec);
+		if (contig_type != CONTIG_NONE) {
+			insert->ins_contig_index = i;
+			break;
+		}
+	}
+	insert->ins_contig = contig_type;
+}
+
+/*
+ * This should only be called against the righmost leaf extent list.
+ *
+ * ocfs2_figure_appending_type() will figure out whether we'll have to
+ * insert at the tail of the rightmost leaf.
+ *
+ * This should also work against the dinode list for tree's with 0
+ * depth. If we consider the dinode list to be the rightmost leaf node
+ * then the logic here makes sense.
+ */
+static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
+					struct ocfs2_extent_list *el,
+					struct ocfs2_extent_rec *insert_rec)
+{
+	int i;
+	u32 cpos = le32_to_cpu(insert_rec->e_cpos);
+	struct ocfs2_extent_rec *rec;
+
+	insert->ins_appending = APPEND_NONE;
+
+	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+
+	if (!el->l_next_free_rec)
+		goto set_tail_append;
+
+	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
+		/* Were all records empty? */
+		if (le16_to_cpu(el->l_next_free_rec) == 1)
+			goto set_tail_append;
 	}
 
-	/* Can we allocate without adding/shifting tree bits? */
 	i = le16_to_cpu(el->l_next_free_rec) - 1;
-	if (le16_to_cpu(el->l_next_free_rec) == 0
-	    || (le16_to_cpu(el->l_next_free_rec) < le16_to_cpu(el->l_count))
-	    || le32_to_cpu(el->l_recs[i].e_clusters) == 0
-	    || ocfs2_extent_contig(inode, &el->l_recs[i], start_blk))
-		goto out_add;
+	rec = &el->l_recs[i];
+
+	if (cpos >=
+	    (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
+		goto set_tail_append;
+
+	return;
+
+set_tail_append:
+	insert->ins_appending = APPEND_TAIL;
+}
+
+/*
+ * Helper function called at the begining of an insert.
+ *
+ * This computes a few things that are commonly used in the process of
+ * inserting into the btree:
+ *   - Whether the new extent is contiguous with an existing one.
+ *   - The current tree depth.
+ *   - Whether the insert is an appending one.
+ *   - The total # of free records in the tree.
+ *
+ * All of the information is stored on the ocfs2_insert_type
+ * structure.
+ */
+static int ocfs2_figure_insert_type(struct inode *inode,
+				    struct buffer_head *di_bh,
+				    struct buffer_head **last_eb_bh,
+				    struct ocfs2_extent_rec *insert_rec,
+				    struct ocfs2_insert_type *insert)
+{
+	int ret;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_path *path = NULL;
+	struct buffer_head *bh = NULL;
+
+	el = &di->id2.i_list;
+	insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);
+
+	if (el->l_tree_depth) {
+		/*
+		 * If we have tree depth, we read in the
+		 * rightmost extent block ahead of time as
+		 * ocfs2_figure_insert_type() and ocfs2_add_branch()
+		 * may want it later.
+		 */
+		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
+				       le64_to_cpu(di->i_last_eb_blk), &bh,
+				       OCFS2_BH_CACHED, inode);
+		if (ret) {
+			mlog_exit(ret);
+			goto out;
+		}
+		eb = (struct ocfs2_extent_block *) bh->b_data;
+		el = &eb->h_list;
+	}
+
+	/*
+	 * Unless we have a contiguous insert, we'll need to know if
+	 * there is room left in our allocation tree for another
+	 * extent record.
+	 *
+	 * XXX: This test is simplistic, we can search for empty
+	 * extent records too.
+	 */
+	insert->ins_free_records = le16_to_cpu(el->l_count) -
+		le16_to_cpu(el->l_next_free_rec);
+
+	if (!insert->ins_tree_depth) {
+		ocfs2_figure_contig_type(inode, insert, el, insert_rec);
+		ocfs2_figure_appending_type(insert, el, insert_rec);
+		return 0;
+	}
+
+	path = ocfs2_new_inode_path(di_bh);
+	if (!path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * In the case that we're inserting past what the tree
+	 * currently accounts for, ocfs2_find_path() will return for
+	 * us the rightmost tree path. This is accounted for below in
+	 * the appending code.
+	 */
+	ret = ocfs2_find_path(inode, path, le32_to_cpu(insert_rec->e_cpos));
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	el = path_leaf_el(path);
+
+	/*
+	 * Now that we have the path, there's two things we want to determine:
+	 * 1) Contiguousness (also set contig_index if this is so)
+	 *
+	 * 2) Are we doing an append? We can trivially break this up
+         *     into two types of appends: simple record append, or a
+         *     rotate inside the tail leaf.
+	 */
+	ocfs2_figure_contig_type(inode, insert, el, insert_rec);
+
+	/*
+	 * The insert code isn't quite ready to deal with all cases of
+	 * left contiguousness. Specifically, if it's an insert into
+	 * the 1st record in a leaf, it will require the adjustment of
+	 * cluster count on the last record of the path directly to it's
+	 * left. For now, just catch that case and fool the layers
+	 * above us. This works just fine for tree_depth == 0, which
+	 * is why we allow that above.
+	 */
+	if (insert->ins_contig == CONTIG_LEFT &&
+	    insert->ins_contig_index == 0)
+		insert->ins_contig = CONTIG_NONE;
+
+	/*
+	 * Ok, so we can simply compare against last_eb to figure out
+	 * whether the path doesn't exist. This will only happen in
+	 * the case that we're doing a tail append, so maybe we can
+	 * take advantage of that information somehow.
+	 */
+	if (le64_to_cpu(di->i_last_eb_blk) == path_leaf_bh(path)->b_blocknr) {
+		/*
+		 * Ok, ocfs2_find_path() returned us the rightmost
+		 * tree path. This might be an appending insert. There are
+		 * two cases:
+		 *    1) We're doing a true append at the tail:
+		 *	-This might even be off the end of the leaf
+		 *    2) We're "appending" by rotating in the tail
+		 */
+		ocfs2_figure_appending_type(insert, el, insert_rec);
+	}
+
+out:
+	ocfs2_free_path(path);
+
+	if (ret == 0)
+		*last_eb_bh = bh;
+	else
+		brelse(bh);
+	return ret;
+}
+
+/*
+ * Insert an extent into an inode btree.
+ *
+ * The caller needs to update fe->i_clusters
+ */
+int ocfs2_insert_extent(struct ocfs2_super *osb,
+			handle_t *handle,
+			struct inode *inode,
+			struct buffer_head *fe_bh,
+			u32 cpos,
+			u64 start_blk,
+			u32 new_clusters,
+			struct ocfs2_alloc_context *meta_ac)
+{
+	int status, shift;
+	struct buffer_head *last_eb_bh = NULL;
+	struct buffer_head *bh = NULL;
+	struct ocfs2_insert_type insert = {0, };
+	struct ocfs2_extent_rec rec;
+
+	mlog(0, "add %u clusters at position %u to inode %llu\n",
+	     new_clusters, cpos, (unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) &&
+			(OCFS2_I(inode)->ip_clusters != cpos),
+			"Device %s, asking for sparse allocation: inode %llu, "
+			"cpos %u, clusters %u\n",
+			osb->dev_str,
+			(unsigned long long)OCFS2_I(inode)->ip_blkno, cpos,
+			OCFS2_I(inode)->ip_clusters);
+
+	memset(&rec, 0, sizeof(rec));
+	rec.e_cpos = cpu_to_le32(cpos);
+	rec.e_blkno = cpu_to_le64(start_blk);
+	rec.e_leaf_clusters = cpu_to_le16(new_clusters);
+
+	status = ocfs2_figure_insert_type(inode, fe_bh, &last_eb_bh, &rec,
+					  &insert);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
 
-	mlog(0, "ocfs2_allocate_extent: couldn't do a simple add, traversing "
-	     "tree now.\n");
+	mlog(0, "Insert.appending: %u, Insert.Contig: %u, "
+	     "Insert.contig_index: %d, Insert.free_records: %d, "
+	     "Insert.tree_depth: %d\n",
+	     insert.ins_appending, insert.ins_contig, insert.ins_contig_index,
+	     insert.ins_free_records, insert.ins_tree_depth);
+
+	/*
+	 * Avoid growing the tree unless we're out of records and the
+	 * insert type requres one.
+	 */
+	if (insert.ins_contig != CONTIG_NONE || insert.ins_free_records)
+		goto out_add;
 
 	shift = ocfs2_find_branch_target(osb, inode, fe_bh, &bh);
 	if (shift < 0) {
@@ -866,13 +2382,9 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
 	 * and didn't find room for any more extents - we need to add
 	 * another tree level */
 	if (shift) {
-		/* if we hit a leaf, we'd better be empty :) */
-		BUG_ON(le16_to_cpu(el->l_next_free_rec) !=
-		       le16_to_cpu(el->l_count));
 		BUG_ON(bh);
-		mlog(0, "ocfs2_allocate_extent: need to shift tree depth "
-		     "(current = %u)\n",
-		     le16_to_cpu(fe->id2.i_list.l_tree_depth));
+		mlog(0, "need to shift tree depth "
+		     "(current = %d)\n", insert.ins_tree_depth);
 
 		/* ocfs2_shift_tree_depth will return us a buffer with
 		 * the new extent block (so we can pass that to
@@ -883,15 +2395,16 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
 			mlog_errno(status);
 			goto bail;
 		}
+		insert.ins_tree_depth++;
 		/* Special case: we have room now if we shifted from
 		 * tree_depth 0 */
-		if (fe->id2.i_list.l_tree_depth == cpu_to_le16(1))
+		if (insert.ins_tree_depth == 1)
 			goto out_add;
 	}
 
 	/* call ocfs2_add_branch to add the final part of the tree with
 	 * the new data. */
-	mlog(0, "ocfs2_allocate_extent: add branch. bh = %p\n", bh);
+	mlog(0, "add branch. bh = %p\n", bh);
 	status = ocfs2_add_branch(osb, handle, inode, fe_bh, bh, last_eb_bh,
 				  meta_ac);
 	if (status < 0) {
@@ -900,11 +2413,12 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
 	}
 
 out_add:
-	/* Finally, we can add clusters. */
-	status = ocfs2_do_insert_extent(osb, handle, inode, fe_bh,
-					start_blk, new_clusters);
+	/* Finally, we can add clusters. This might rotate the tree for us. */
+	status = ocfs2_do_insert_extent(inode, handle, fe_bh, &rec, &insert);
 	if (status < 0)
 		mlog_errno(status);
+	else
+		ocfs2_extent_map_insert_rec(inode, &rec);
 
 bail:
 	if (bh)
@@ -1447,168 +2961,389 @@ int ocfs2_truncate_log_init(struct ocfs2_super *osb)
  * block will be deleted, and if it will, what the new last extent
  * block will be so we can update his h_next_leaf_blk field, as well
  * as the dinodes i_last_eb_blk */
-static int ocfs2_find_new_last_ext_blk(struct ocfs2_super *osb,
-				       struct inode *inode,
-				       struct ocfs2_dinode *fe,
-				       u32 new_i_clusters,
-				       struct buffer_head *old_last_eb,
+static int ocfs2_find_new_last_ext_blk(struct inode *inode,
+				       unsigned int clusters_to_del,
+				       struct ocfs2_path *path,
 				       struct buffer_head **new_last_eb)
 {
-	int i, status = 0;
-	u64 block = 0;
+	int next_free, ret = 0;
+	u32 cpos;
+	struct ocfs2_extent_rec *rec;
 	struct ocfs2_extent_block *eb;
 	struct ocfs2_extent_list *el;
 	struct buffer_head *bh = NULL;
 
 	*new_last_eb = NULL;
 
-	if (!OCFS2_IS_VALID_DINODE(fe)) {
-		OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
-		status = -EIO;
-		goto bail;
-	}
-
 	/* we have no tree, so of course, no last_eb. */
-	if (!fe->id2.i_list.l_tree_depth)
-		goto bail;
+	if (!path->p_tree_depth)
+		goto out;
 
 	/* trunc to zero special case - this makes tree_depth = 0
 	 * regardless of what it is.  */
-	if (!new_i_clusters)
-		goto bail;
+	if (OCFS2_I(inode)->ip_clusters == clusters_to_del)
+		goto out;
 
-	eb = (struct ocfs2_extent_block *) old_last_eb->b_data;
-	el = &(eb->h_list);
+	el = path_leaf_el(path);
 	BUG_ON(!el->l_next_free_rec);
 
-	/* Make sure that this guy will actually be empty after we
-	 * clear away the data. */
-	if (le32_to_cpu(el->l_recs[0].e_cpos) < new_i_clusters)
-		goto bail;
+	/*
+	 * Make sure that this extent list will actually be empty
+	 * after we clear away the data. We can shortcut out if
+	 * there's more than one non-empty extent in the
+	 * list. Otherwise, a check of the remaining extent is
+	 * necessary.
+	 */
+	next_free = le16_to_cpu(el->l_next_free_rec);
+	rec = NULL;
+	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
+		if (next_free > 2)
+			goto out;
 
-	/* Ok, at this point, we know that last_eb will definitely
-	 * change, so lets traverse the tree and find the second to
-	 * last extent block. */
-	el = &(fe->id2.i_list);
-	/* go down the tree, */
-	do {
-		for(i = (le16_to_cpu(el->l_next_free_rec) - 1); i >= 0; i--) {
-			if (le32_to_cpu(el->l_recs[i].e_cpos) <
-			    new_i_clusters) {
-				block = le64_to_cpu(el->l_recs[i].e_blkno);
-				break;
-			}
+		/* We may have a valid extent in index 1, check it. */
+		if (next_free == 2)
+			rec = &el->l_recs[1];
+
+		/*
+		 * Fall through - no more nonempty extents, so we want
+		 * to delete this leaf.
+		 */
+	} else {
+		if (next_free > 1)
+			goto out;
+
+		rec = &el->l_recs[0];
+	}
+
+	if (rec) {
+		/*
+		 * Check it we'll only be trimming off the end of this
+		 * cluster.
+		 */
+		if (le16_to_cpu(rec->e_leaf_clusters) > clusters_to_del)
+			goto out;
+	}
+
+	ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_leaf(inode, path_root_el(path), cpos, &bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	eb = (struct ocfs2_extent_block *) bh->b_data;
+	el = &eb->h_list;
+	if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
+		OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
+		ret = -EROFS;
+		goto out;
+	}
+
+	*new_last_eb = bh;
+	get_bh(*new_last_eb);
+	mlog(0, "returning block %llu, (cpos: %u)\n",
+	     (unsigned long long)le64_to_cpu(eb->h_blkno), cpos);
+out:
+	brelse(bh);
+
+	return ret;
+}
+
+/*
+ * Trim some clusters off the rightmost edge of a tree. Only called
+ * during truncate.
+ *
+ * The caller needs to:
+ *   - start journaling of each path component.
+ *   - compute and fully set up any new last ext block
+ */
+static int ocfs2_trim_tree(struct inode *inode, struct ocfs2_path *path,
+			   handle_t *handle, struct ocfs2_truncate_context *tc,
+			   u32 clusters_to_del, u64 *delete_start)
+{
+	int ret, i, index = path->p_tree_depth;
+	u32 new_edge = 0;
+	u64 deleted_eb = 0;
+	struct buffer_head *bh;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_rec *rec;
+
+	*delete_start = 0;
+
+	while (index >= 0) {
+		bh = path->p_node[index].bh;
+		el = path->p_node[index].el;
+
+		mlog(0, "traveling tree (index = %d, block = %llu)\n",
+		     index,  (unsigned long long)bh->b_blocknr);
+
+		BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
+
+		if (index !=
+		    (path->p_tree_depth - le16_to_cpu(el->l_tree_depth))) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %lu has invalid ext. block %llu",
+				    inode->i_ino,
+				    (unsigned long long)bh->b_blocknr);
+			ret = -EROFS;
+			goto out;
 		}
-		BUG_ON(i < 0);
 
-		if (bh) {
-			brelse(bh);
-			bh = NULL;
+find_tail_record:
+		i = le16_to_cpu(el->l_next_free_rec) - 1;
+		rec = &el->l_recs[i];
+
+		mlog(0, "Extent list before: record %d: (%u, %u, %llu), "
+		     "next = %u\n", i, le32_to_cpu(rec->e_cpos),
+		     ocfs2_rec_clusters(el, rec),
+		     (unsigned long long)le64_to_cpu(rec->e_blkno),
+		     le16_to_cpu(el->l_next_free_rec));
+
+		BUG_ON(ocfs2_rec_clusters(el, rec) < clusters_to_del);
+
+		if (le16_to_cpu(el->l_tree_depth) == 0) {
+			/*
+			 * If the leaf block contains a single empty
+			 * extent and no records, we can just remove
+			 * the block.
+			 */
+			if (i == 0 && ocfs2_is_empty_extent(rec)) {
+				memset(rec, 0,
+				       sizeof(struct ocfs2_extent_rec));
+				el->l_next_free_rec = cpu_to_le16(0);
+
+				goto delete;
+			}
+
+			/*
+			 * Remove any empty extents by shifting things
+			 * left. That should make life much easier on
+			 * the code below. This condition is rare
+			 * enough that we shouldn't see a performance
+			 * hit.
+			 */
+			if (ocfs2_is_empty_extent(&el->l_recs[0])) {
+				le16_add_cpu(&el->l_next_free_rec, -1);
+
+				for(i = 0;
+				    i < le16_to_cpu(el->l_next_free_rec); i++)
+					el->l_recs[i] = el->l_recs[i + 1];
+
+				memset(&el->l_recs[i], 0,
+				       sizeof(struct ocfs2_extent_rec));
+
+				/*
+				 * We've modified our extent list. The
+				 * simplest way to handle this change
+				 * is to being the search from the
+				 * start again.
+				 */
+				goto find_tail_record;
+			}
+
+			le16_add_cpu(&rec->e_leaf_clusters, -clusters_to_del);
+
+			/*
+			 * We'll use "new_edge" on our way back up the
+			 * tree to know what our rightmost cpos is.
+			 */
+			new_edge = le16_to_cpu(rec->e_leaf_clusters);
+			new_edge += le32_to_cpu(rec->e_cpos);
+
+			/*
+			 * The caller will use this to delete data blocks.
+			 */
+			*delete_start = le64_to_cpu(rec->e_blkno)
+				+ ocfs2_clusters_to_blocks(inode->i_sb,
+					le16_to_cpu(rec->e_leaf_clusters));
+
+			/*
+			 * If it's now empty, remove this record.
+			 */
+			if (le16_to_cpu(rec->e_leaf_clusters) == 0) {
+				memset(rec, 0,
+				       sizeof(struct ocfs2_extent_rec));
+				le16_add_cpu(&el->l_next_free_rec, -1);
+			}
+		} else {
+			if (le64_to_cpu(rec->e_blkno) == deleted_eb) {
+				memset(rec, 0,
+				       sizeof(struct ocfs2_extent_rec));
+				le16_add_cpu(&el->l_next_free_rec, -1);
+
+				goto delete;
+			}
+
+			/* Can this actually happen? */
+			if (le16_to_cpu(el->l_next_free_rec) == 0)
+				goto delete;
+
+			/*
+			 * We never actually deleted any clusters
+			 * because our leaf was empty. There's no
+			 * reason to adjust the rightmost edge then.
+			 */
+			if (new_edge == 0)
+				goto delete;
+
+			rec->e_int_clusters = cpu_to_le32(new_edge);
+			le32_add_cpu(&rec->e_int_clusters,
+				     -le32_to_cpu(rec->e_cpos));
+
+			 /*
+			  * A deleted child record should have been
+			  * caught above.
+			  */
+			 BUG_ON(le32_to_cpu(rec->e_int_clusters) == 0);
 		}
 
-		status = ocfs2_read_block(osb, block, &bh, OCFS2_BH_CACHED,
-					 inode);
-		if (status < 0) {
-			mlog_errno(status);
-			goto bail;
+delete:
+		ret = ocfs2_journal_dirty(handle, bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
 		}
-		eb = (struct ocfs2_extent_block *) bh->b_data;
-		el = &eb->h_list;
-		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
-			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
-			status = -EIO;
-			goto bail;
+
+		mlog(0, "extent list container %llu, after: record %d: "
+		     "(%u, %u, %llu), next = %u.\n",
+		     (unsigned long long)bh->b_blocknr, i,
+		     le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec),
+		     (unsigned long long)le64_to_cpu(rec->e_blkno),
+		     le16_to_cpu(el->l_next_free_rec));
+
+		/*
+		 * We must be careful to only attempt delete of an
+		 * extent block (and not the root inode block).
+		 */
+		if (index > 0 && le16_to_cpu(el->l_next_free_rec) == 0) {
+			struct ocfs2_extent_block *eb =
+				(struct ocfs2_extent_block *)bh->b_data;
+
+			/*
+			 * Save this for use when processing the
+			 * parent block.
+			 */
+			deleted_eb = le64_to_cpu(eb->h_blkno);
+
+			mlog(0, "deleting this extent block.\n");
+
+			ocfs2_remove_from_cache(inode, bh);
+
+			BUG_ON(ocfs2_rec_clusters(el, &el->l_recs[0]));
+			BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos));
+			BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno));
+
+			if (le16_to_cpu(eb->h_suballoc_slot) == 0) {
+				/*
+				 * This code only understands how to
+				 * lock the suballocator in slot 0,
+				 * which is fine because allocation is
+				 * only ever done out of that
+				 * suballocator too. A future version
+				 * might change that however, so avoid
+				 * a free if we don't know how to
+				 * handle it. This way an fs incompat
+				 * bit will not be necessary.
+				 */
+				ret = ocfs2_free_extent_block(handle,
+							      tc->tc_ext_alloc_inode,
+							      tc->tc_ext_alloc_bh,
+							      eb);
+
+				/* An error here is not fatal. */
+				if (ret < 0)
+					mlog_errno(ret);
+			}
+		} else {
+			deleted_eb = 0;
 		}
-	} while (el->l_tree_depth);
 
-	*new_last_eb = bh;
-	get_bh(*new_last_eb);
-	mlog(0, "returning block %llu\n",
-	     (unsigned long long)le64_to_cpu(eb->h_blkno));
-bail:
-	if (bh)
-		brelse(bh);
+		index--;
+	}
 
-	return status;
+	ret = 0;
+out:
+	return ret;
 }
 
 static int ocfs2_do_truncate(struct ocfs2_super *osb,
 			     unsigned int clusters_to_del,
 			     struct inode *inode,
 			     struct buffer_head *fe_bh,
-			     struct buffer_head *old_last_eb_bh,
 			     handle_t *handle,
-			     struct ocfs2_truncate_context *tc)
+			     struct ocfs2_truncate_context *tc,
+			     struct ocfs2_path *path)
 {
-	int status, i, depth;
+	int status;
 	struct ocfs2_dinode *fe;
-	struct ocfs2_extent_block *eb;
 	struct ocfs2_extent_block *last_eb = NULL;
 	struct ocfs2_extent_list *el;
-	struct buffer_head *eb_bh = NULL;
 	struct buffer_head *last_eb_bh = NULL;
-	u64 next_eb = 0;
 	u64 delete_blk = 0;
 
 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
 
-	status = ocfs2_find_new_last_ext_blk(osb,
-					     inode,
-					     fe,
-					     le32_to_cpu(fe->i_clusters) -
-					     		clusters_to_del,
-					     old_last_eb_bh,
-					     &last_eb_bh);
+	status = ocfs2_find_new_last_ext_blk(inode, clusters_to_del,
+					     path, &last_eb_bh);
 	if (status < 0) {
 		mlog_errno(status);
 		goto bail;
 	}
-	if (last_eb_bh)
-		last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
 
-	status = ocfs2_journal_access(handle, inode, fe_bh,
-				      OCFS2_JOURNAL_ACCESS_WRITE);
+	/*
+	 * Each component will be touched, so we might as well journal
+	 * here to avoid having to handle errors later.
+	 */
+	status = ocfs2_journal_access_path(inode, handle, path);
 	if (status < 0) {
 		mlog_errno(status);
 		goto bail;
 	}
+
+	if (last_eb_bh) {
+		status = ocfs2_journal_access(handle, inode, last_eb_bh,
+					      OCFS2_JOURNAL_ACCESS_WRITE);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
+	}
+
 	el = &(fe->id2.i_list);
 
+	/*
+	 * Lower levels depend on this never happening, but it's best
+	 * to check it up here before changing the tree.
+	 */
+	if (el->l_tree_depth && el->l_recs[0].e_int_clusters == 0) {
+		ocfs2_error(inode->i_sb,
+			    "Inode %lu has an empty extent record, depth %u\n",
+			    inode->i_ino, le16_to_cpu(el->l_tree_depth));
+		status = -EROFS;
+		goto bail;
+	}
+
 	spin_lock(&OCFS2_I(inode)->ip_lock);
 	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters) -
 				      clusters_to_del;
 	spin_unlock(&OCFS2_I(inode)->ip_lock);
 	le32_add_cpu(&fe->i_clusters, -clusters_to_del);
-	fe->i_mtime = cpu_to_le64(CURRENT_TIME.tv_sec);
-	fe->i_mtime_nsec = cpu_to_le32(CURRENT_TIME.tv_nsec);
-
-	i = le16_to_cpu(el->l_next_free_rec) - 1;
-
-	BUG_ON(le32_to_cpu(el->l_recs[i].e_clusters) < clusters_to_del);
-	le32_add_cpu(&el->l_recs[i].e_clusters, -clusters_to_del);
-	/* tree depth zero, we can just delete the clusters, otherwise
-	 * we need to record the offset of the next level extent block
-	 * as we may overwrite it. */
-	if (!el->l_tree_depth)
-		delete_blk = le64_to_cpu(el->l_recs[i].e_blkno)
-			+ ocfs2_clusters_to_blocks(osb->sb,
-					le32_to_cpu(el->l_recs[i].e_clusters));
-	else
-		next_eb = le64_to_cpu(el->l_recs[i].e_blkno);
 
-	if (!el->l_recs[i].e_clusters) {
-		/* if we deleted the whole extent record, then clear
-		 * out the other fields and update the extent
-		 * list. For depth > 0 trees, we've already recorded
-		 * the extent block in 'next_eb' */
-		el->l_recs[i].e_cpos = 0;
-		el->l_recs[i].e_blkno = 0;
-		BUG_ON(!el->l_next_free_rec);
-		le16_add_cpu(&el->l_next_free_rec, -1);
+	status = ocfs2_trim_tree(inode, path, handle, tc,
+				 clusters_to_del, &delete_blk);
+	if (status) {
+		mlog_errno(status);
+		goto bail;
 	}
 
-	depth = le16_to_cpu(el->l_tree_depth);
-	if (!fe->i_clusters) {
+	if (le32_to_cpu(fe->i_clusters) == 0) {
 		/* trunc to zero is a special case. */
 		el->l_tree_depth = 0;
 		fe->i_last_eb_blk = 0;
@@ -1625,12 +3360,6 @@ static int ocfs2_do_truncate(struct ocfs2_super *osb,
 		/* If there will be a new last extent block, then by
 		 * definition, there cannot be any leaves to the right of
 		 * him. */
-		status = ocfs2_journal_access(handle, inode, last_eb_bh,
-					      OCFS2_JOURNAL_ACCESS_WRITE);
-		if (status < 0) {
-			mlog_errno(status);
-			goto bail;
-		}
 		last_eb->h_next_leaf_blk = 0;
 		status = ocfs2_journal_dirty(handle, last_eb_bh);
 		if (status < 0) {
@@ -1639,123 +3368,247 @@ static int ocfs2_do_truncate(struct ocfs2_super *osb,
 		}
 	}
 
-	/* if our tree depth > 0, update all the tree blocks below us. */
-	while (depth) {
-		mlog(0, "traveling tree (depth = %d, next_eb = %llu)\n",
-		     depth,  (unsigned long long)next_eb);
-		status = ocfs2_read_block(osb, next_eb, &eb_bh,
-					  OCFS2_BH_CACHED, inode);
+	if (delete_blk) {
+		status = ocfs2_truncate_log_append(osb, handle, delete_blk,
+						   clusters_to_del);
 		if (status < 0) {
 			mlog_errno(status);
 			goto bail;
 		}
-		eb = (struct ocfs2_extent_block *)eb_bh->b_data;
-		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
-			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
-			status = -EIO;
-			goto bail;
+	}
+	status = 0;
+bail:
+
+	mlog_exit(status);
+	return status;
+}
+
+static int ocfs2_writeback_zero_func(handle_t *handle, struct buffer_head *bh)
+{
+	set_buffer_uptodate(bh);
+	mark_buffer_dirty(bh);
+	return 0;
+}
+
+static int ocfs2_ordered_zero_func(handle_t *handle, struct buffer_head *bh)
+{
+	set_buffer_uptodate(bh);
+	mark_buffer_dirty(bh);
+	return ocfs2_journal_dirty_data(handle, bh);
+}
+
+static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t isize,
+				     struct page **pages, int numpages,
+				     u64 phys, handle_t *handle)
+{
+	int i, ret, partial = 0;
+	void *kaddr;
+	struct page *page;
+	unsigned int from, to = PAGE_CACHE_SIZE;
+	struct super_block *sb = inode->i_sb;
+
+	BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
+
+	if (numpages == 0)
+		goto out;
+
+	from = isize & (PAGE_CACHE_SIZE - 1); /* 1st page offset */
+	if (PAGE_CACHE_SHIFT > OCFS2_SB(sb)->s_clustersize_bits) {
+		/*
+		 * Since 'from' has been capped to a value below page
+		 * size, this calculation won't be able to overflow
+		 * 'to'
+		 */
+		to = ocfs2_align_bytes_to_clusters(sb, from);
+
+		/*
+		 * The truncate tail in this case should never contain
+		 * more than one page at maximum. The loop below also
+		 * assumes this.
+		 */
+		BUG_ON(numpages != 1);
+	}
+
+	for(i = 0; i < numpages; i++) {
+		page = pages[i];
+
+		BUG_ON(from > PAGE_CACHE_SIZE);
+		BUG_ON(to > PAGE_CACHE_SIZE);
+
+		ret = ocfs2_map_page_blocks(page, &phys, inode, from, to, 0);
+		if (ret)
+			mlog_errno(ret);
+
+		kaddr = kmap_atomic(page, KM_USER0);
+		memset(kaddr + from, 0, to - from);
+		kunmap_atomic(kaddr, KM_USER0);
+
+		/*
+		 * Need to set the buffers we zero'd into uptodate
+		 * here if they aren't - ocfs2_map_page_blocks()
+		 * might've skipped some
+		 */
+		if (ocfs2_should_order_data(inode)) {
+			ret = walk_page_buffers(handle,
+						page_buffers(page),
+						from, to, &partial,
+						ocfs2_ordered_zero_func);
+			if (ret < 0)
+				mlog_errno(ret);
+		} else {
+			ret = walk_page_buffers(handle, page_buffers(page),
+						from, to, &partial,
+						ocfs2_writeback_zero_func);
+			if (ret < 0)
+				mlog_errno(ret);
 		}
-		el = &(eb->h_list);
 
-		status = ocfs2_journal_access(handle, inode, eb_bh,
-					      OCFS2_JOURNAL_ACCESS_WRITE);
-		if (status < 0) {
-			mlog_errno(status);
-			goto bail;
+		if (!partial)
+			SetPageUptodate(page);
+
+		flush_dcache_page(page);
+
+		/*
+		 * Every page after the 1st one should be completely zero'd.
+		 */
+		from = 0;
+	}
+out:
+	if (pages) {
+		for (i = 0; i < numpages; i++) {
+			page = pages[i];
+			unlock_page(page);
+			mark_page_accessed(page);
+			page_cache_release(page);
 		}
+	}
+}
 
-		BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
-		BUG_ON(depth != (le16_to_cpu(el->l_tree_depth) + 1));
+static int ocfs2_grab_eof_pages(struct inode *inode, loff_t isize, struct page **pages,
+				int *num, u64 *phys)
+{
+	int i, numpages = 0, ret = 0;
+	unsigned int csize = OCFS2_SB(inode->i_sb)->s_clustersize;
+	unsigned int ext_flags;
+	struct super_block *sb = inode->i_sb;
+	struct address_space *mapping = inode->i_mapping;
+	unsigned long index;
+	u64 next_cluster_bytes;
+
+	BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
+
+	/* Cluster boundary, so we don't need to grab any pages. */
+	if ((isize & (csize - 1)) == 0)
+		goto out;
 
-		i = le16_to_cpu(el->l_next_free_rec) - 1;
+	ret = ocfs2_extent_map_get_blocks(inode, isize >> sb->s_blocksize_bits,
+					  phys, NULL, &ext_flags);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
 
-		mlog(0, "extent block %llu, before: record %d: "
-		     "(%u, %u, %llu), next = %u\n",
-		     (unsigned long long)le64_to_cpu(eb->h_blkno), i,
-		     le32_to_cpu(el->l_recs[i].e_cpos),
-		     le32_to_cpu(el->l_recs[i].e_clusters),
-		     (unsigned long long)le64_to_cpu(el->l_recs[i].e_blkno),
-		     le16_to_cpu(el->l_next_free_rec));
+	/* Tail is a hole. */
+	if (*phys == 0)
+		goto out;
 
-		BUG_ON(le32_to_cpu(el->l_recs[i].e_clusters) < clusters_to_del);
-		le32_add_cpu(&el->l_recs[i].e_clusters, -clusters_to_del);
-
-		next_eb = le64_to_cpu(el->l_recs[i].e_blkno);
-		/* bottom-most block requires us to delete data.*/
-		if (!el->l_tree_depth)
-			delete_blk = le64_to_cpu(el->l_recs[i].e_blkno)
-				+ ocfs2_clusters_to_blocks(osb->sb,
-					le32_to_cpu(el->l_recs[i].e_clusters));
-		if (!el->l_recs[i].e_clusters) {
-			el->l_recs[i].e_cpos = 0;
-			el->l_recs[i].e_blkno = 0;
-			BUG_ON(!el->l_next_free_rec);
-			le16_add_cpu(&el->l_next_free_rec, -1);
-		}
-		mlog(0, "extent block %llu, after: record %d: "
-		     "(%u, %u, %llu), next = %u\n",
-		     (unsigned long long)le64_to_cpu(eb->h_blkno), i,
-		     le32_to_cpu(el->l_recs[i].e_cpos),
-		     le32_to_cpu(el->l_recs[i].e_clusters),
-		     (unsigned long long)le64_to_cpu(el->l_recs[i].e_blkno),
-		     le16_to_cpu(el->l_next_free_rec));
+	/* Tail is marked as unwritten, we can count on write to zero
+	 * in that case. */
+	if (ext_flags & OCFS2_EXT_UNWRITTEN)
+		goto out;
 
-		status = ocfs2_journal_dirty(handle, eb_bh);
-		if (status < 0) {
-			mlog_errno(status);
-			goto bail;
+	next_cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, isize);
+	index = isize >> PAGE_CACHE_SHIFT;
+	do {
+		pages[numpages] = grab_cache_page(mapping, index);
+		if (!pages[numpages]) {
+			ret = -ENOMEM;
+			mlog_errno(ret);
+			goto out;
 		}
 
-		if (!el->l_next_free_rec) {
-			mlog(0, "deleting this extent block.\n");
-
-			ocfs2_remove_from_cache(inode, eb_bh);
+		numpages++;
+		index++;
+	} while (index < (next_cluster_bytes >> PAGE_CACHE_SHIFT));
 
-			BUG_ON(el->l_recs[0].e_clusters);
-			BUG_ON(el->l_recs[0].e_cpos);
-			BUG_ON(el->l_recs[0].e_blkno);
-			if (eb->h_suballoc_slot == 0) {
-				/*
-				 * This code only understands how to
-				 * lock the suballocator in slot 0,
-				 * which is fine because allocation is
-				 * only ever done out of that
-				 * suballocator too. A future version
-				 * might change that however, so avoid
-				 * a free if we don't know how to
-				 * handle it. This way an fs incompat
-				 * bit will not be necessary.
-				 */
-				status = ocfs2_free_extent_block(handle,
-								 tc->tc_ext_alloc_inode,
-								 tc->tc_ext_alloc_bh,
-								 eb);
-				if (status < 0) {
-					mlog_errno(status);
-					goto bail;
+out:
+	if (ret != 0) {
+		if (pages) {
+			for (i = 0; i < numpages; i++) {
+				if (pages[i]) {
+					unlock_page(pages[i]);
+					page_cache_release(pages[i]);
 				}
 			}
 		}
-		brelse(eb_bh);
-		eb_bh = NULL;
-		depth--;
+		numpages = 0;
 	}
 
-	BUG_ON(!delete_blk);
-	status = ocfs2_truncate_log_append(osb, handle, delete_blk,
-					   clusters_to_del);
-	if (status < 0) {
-		mlog_errno(status);
-		goto bail;
+	*num = numpages;
+
+	return ret;
+}
+
+/*
+ * Zero the area past i_size but still within an allocated
+ * cluster. This avoids exposing nonzero data on subsequent file
+ * extends.
+ *
+ * We need to call this before i_size is updated on the inode because
+ * otherwise block_write_full_page() will skip writeout of pages past
+ * i_size. The new_i_size parameter is passed for this reason.
+ */
+int ocfs2_zero_tail_for_truncate(struct inode *inode, handle_t *handle,
+				 u64 new_i_size)
+{
+	int ret, numpages;
+	loff_t endbyte;
+	struct page **pages = NULL;
+	u64 phys;
+
+	/*
+	 * File systems which don't support sparse files zero on every
+	 * extend.
+	 */
+	if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
+		return 0;
+
+	pages = kcalloc(ocfs2_pages_per_cluster(inode->i_sb),
+			sizeof(struct page *), GFP_NOFS);
+	if (pages == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
 	}
-	status = 0;
-bail:
-	if (!status)
-		ocfs2_extent_map_trunc(inode, le32_to_cpu(fe->i_clusters));
-	else
-		ocfs2_extent_map_drop(inode, 0);
-	mlog_exit(status);
-	return status;
+
+	ret = ocfs2_grab_eof_pages(inode, new_i_size, pages, &numpages, &phys);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (numpages == 0)
+		goto out;
+
+	ocfs2_zero_cluster_pages(inode, new_i_size, pages, numpages, phys,
+				 handle);
+
+	/*
+	 * Initiate writeout of the pages we zero'd here. We don't
+	 * wait on them - the truncate_inode_pages() call later will
+	 * do that for us.
+	 */
+	endbyte = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
+	ret = do_sync_mapping_range(inode->i_mapping, new_i_size,
+				    endbyte - 1, SYNC_FILE_RANGE_WRITE);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	if (pages)
+		kfree(pages);
+
+	return ret;
 }
 
 /*
@@ -1770,82 +3623,90 @@ int ocfs2_commit_truncate(struct ocfs2_super *osb,
 			  struct ocfs2_truncate_context *tc)
 {
 	int status, i, credits, tl_sem = 0;
-	u32 clusters_to_del, target_i_clusters;
-	u64 last_eb = 0;
-	struct ocfs2_dinode *fe;
-	struct ocfs2_extent_block *eb;
+	u32 clusters_to_del, new_highest_cpos, range;
 	struct ocfs2_extent_list *el;
-	struct buffer_head *last_eb_bh;
 	handle_t *handle = NULL;
 	struct inode *tl_inode = osb->osb_tl_inode;
+	struct ocfs2_path *path = NULL;
 
 	mlog_entry_void();
 
 	down_write(&OCFS2_I(inode)->ip_alloc_sem);
 
-	target_i_clusters = ocfs2_clusters_for_bytes(osb->sb,
+	new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
 						     i_size_read(inode));
 
-	last_eb_bh = tc->tc_last_eb_bh;
-	tc->tc_last_eb_bh = NULL;
+	path = ocfs2_new_inode_path(fe_bh);
+	if (!path) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
 
-	fe = (struct ocfs2_dinode *) fe_bh->b_data;
+	ocfs2_extent_map_trunc(inode, new_highest_cpos);
 
-	if (fe->id2.i_list.l_tree_depth) {
-		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
-		el = &eb->h_list;
-	} else
-		el = &fe->id2.i_list;
-	last_eb = le64_to_cpu(fe->i_last_eb_blk);
 start:
-	mlog(0, "ocfs2_commit_truncate: fe->i_clusters = %u, "
-	     "last_eb = %llu, fe->i_last_eb_blk = %llu, "
-	     "fe->id2.i_list.l_tree_depth = %u last_eb_bh = %p\n",
-	     le32_to_cpu(fe->i_clusters), (unsigned long long)last_eb,
-	     (unsigned long long)le64_to_cpu(fe->i_last_eb_blk),
-	     le16_to_cpu(fe->id2.i_list.l_tree_depth), last_eb_bh);
-
-	if (last_eb != le64_to_cpu(fe->i_last_eb_blk)) {
-		mlog(0, "last_eb changed!\n");
-		BUG_ON(!fe->id2.i_list.l_tree_depth);
-		last_eb = le64_to_cpu(fe->i_last_eb_blk);
-		/* i_last_eb_blk may have changed, read it if
-		 * necessary. We don't have to worry about the
-		 * truncate to zero case here (where there becomes no
-		 * last_eb) because we never loop back after our work
-		 * is done. */
-		if (last_eb_bh) {
-			brelse(last_eb_bh);
-			last_eb_bh = NULL;
-		}
+	/*
+	 * Check that we still have allocation to delete.
+	 */
+	if (OCFS2_I(inode)->ip_clusters == 0) {
+		status = 0;
+		goto bail;
+	}
 
-		status = ocfs2_read_block(osb, last_eb,
-					  &last_eb_bh, OCFS2_BH_CACHED,
-					  inode);
-		if (status < 0) {
-			mlog_errno(status);
-			goto bail;
-		}
-		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
-		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
-			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
-			status = -EIO;
-			goto bail;
-		}
-		el = &(eb->h_list);
+	/*
+	 * Truncate always works against the rightmost tree branch.
+	 */
+	status = ocfs2_find_path(inode, path, UINT_MAX);
+	if (status) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	mlog(0, "inode->ip_clusters = %u, tree_depth = %u\n",
+	     OCFS2_I(inode)->ip_clusters, path->p_tree_depth);
+
+	/*
+	 * By now, el will point to the extent list on the bottom most
+	 * portion of this tree. Only the tail record is considered in
+	 * each pass.
+	 *
+	 * We handle the following cases, in order:
+	 * - empty extent: delete the remaining branch
+	 * - remove the entire record
+	 * - remove a partial record
+	 * - no record needs to be removed (truncate has completed)
+	 */
+	el = path_leaf_el(path);
+	if (le16_to_cpu(el->l_next_free_rec) == 0) {
+		ocfs2_error(inode->i_sb,
+			    "Inode %llu has empty extent block at %llu\n",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			    (unsigned long long)path_leaf_bh(path)->b_blocknr);
+		status = -EROFS;
+		goto bail;
 	}
 
-	/* by now, el will point to the extent list on the bottom most
-	 * portion of this tree. */
 	i = le16_to_cpu(el->l_next_free_rec) - 1;
-	if (le32_to_cpu(el->l_recs[i].e_cpos) >= target_i_clusters)
-		clusters_to_del = le32_to_cpu(el->l_recs[i].e_clusters);
-	else
-		clusters_to_del = (le32_to_cpu(el->l_recs[i].e_clusters) +
+	range = le32_to_cpu(el->l_recs[i].e_cpos) +
+		ocfs2_rec_clusters(el, &el->l_recs[i]);
+	if (i == 0 && ocfs2_is_empty_extent(&el->l_recs[i])) {
+		clusters_to_del = 0;
+	} else if (le32_to_cpu(el->l_recs[i].e_cpos) >= new_highest_cpos) {
+		clusters_to_del = ocfs2_rec_clusters(el, &el->l_recs[i]);
+	} else if (range > new_highest_cpos) {
+		clusters_to_del = (ocfs2_rec_clusters(el, &el->l_recs[i]) +
 				   le32_to_cpu(el->l_recs[i].e_cpos)) -
-				  target_i_clusters;
+				  new_highest_cpos;
+	} else {
+		status = 0;
+		goto bail;
+	}
 
-	mlog(0, "clusters_to_del = %u in this pass\n", clusters_to_del);
+	mlog(0, "clusters_to_del = %u in this pass, tail blk=%llu\n",
+	     clusters_to_del, (unsigned long long)path_leaf_bh(path)->b_blocknr);
+
+	BUG_ON(clusters_to_del == 0);
 
 	mutex_lock(&tl_inode->i_mutex);
 	tl_sem = 1;
@@ -1861,7 +3722,8 @@ start:
 	}
 
 	credits = ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del,
-						fe, el);
+						(struct ocfs2_dinode *)fe_bh->b_data,
+						el);
 	handle = ocfs2_start_trans(osb, credits);
 	if (IS_ERR(handle)) {
 		status = PTR_ERR(handle);
@@ -1870,13 +3732,8 @@ start:
 		goto bail;
 	}
 
-	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
-	status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
-	if (status < 0)
-		mlog_errno(status);
-
-	status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh,
-				   last_eb_bh, handle, tc);
+	status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh, handle,
+				   tc, path);
 	if (status < 0) {
 		mlog_errno(status);
 		goto bail;
@@ -1888,9 +3745,14 @@ start:
 	ocfs2_commit_trans(osb, handle);
 	handle = NULL;
 
-	BUG_ON(le32_to_cpu(fe->i_clusters) < target_i_clusters);
-	if (le32_to_cpu(fe->i_clusters) > target_i_clusters)
-		goto start;
+	ocfs2_reinit_path(path, 1);
+
+	/*
+	 * The check above will catch the case where we've truncated
+	 * away all allocation.
+	 */
+	goto start;
+
 bail:
 	up_write(&OCFS2_I(inode)->ip_alloc_sem);
 
@@ -1902,8 +3764,7 @@ bail:
 	if (handle)
 		ocfs2_commit_trans(osb, handle);
 
-	if (last_eb_bh)
-		brelse(last_eb_bh);
+	ocfs2_free_path(path);
 
 	/* This will drop the ext_alloc cluster lock for us */
 	ocfs2_free_truncate_context(tc);
@@ -1912,7 +3773,6 @@ bail:
 	return status;
 }
 
-
 /*
  * Expects the inode to already be locked. This will figure out which
  * inodes need to be locked and will put them on the returned truncate
@@ -1923,7 +3783,7 @@ int ocfs2_prepare_truncate(struct ocfs2_super *osb,
 			   struct buffer_head *fe_bh,
 			   struct ocfs2_truncate_context **tc)
 {
-	int status, metadata_delete;
+	int status, metadata_delete, i;
 	unsigned int new_i_clusters;
 	struct ocfs2_dinode *fe;
 	struct ocfs2_extent_block *eb;
@@ -1944,21 +3804,6 @@ int ocfs2_prepare_truncate(struct ocfs2_super *osb,
 	     "%llu\n", fe->i_clusters, new_i_clusters,
 	     (unsigned long long)fe->i_size);
 
-	if (le32_to_cpu(fe->i_clusters) <= new_i_clusters) {
-		ocfs2_error(inode->i_sb, "Dinode %llu has cluster count "
-			    "%u and size %llu whereas struct inode has "
-			    "cluster count %u and size %llu which caused an "
-			    "invalid truncate to %u clusters.",
-			    (unsigned long long)le64_to_cpu(fe->i_blkno),
-			    le32_to_cpu(fe->i_clusters),
-			    (unsigned long long)le64_to_cpu(fe->i_size),
-			    OCFS2_I(inode)->ip_clusters, i_size_read(inode),
-			    new_i_clusters);
-		mlog_meta_lvb(ML_ERROR, &OCFS2_I(inode)->ip_meta_lockres);
-		status = -EIO;
-		goto bail;
-	}
-
 	*tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL);
 	if (!(*tc)) {
 		status = -ENOMEM;
@@ -1986,7 +3831,15 @@ int ocfs2_prepare_truncate(struct ocfs2_super *osb,
 			goto bail;
 		}
 		el = &(eb->h_list);
-		if (le32_to_cpu(el->l_recs[0].e_cpos) >= new_i_clusters)
+
+		i = 0;
+		if (ocfs2_is_empty_extent(&el->l_recs[0]))
+			i = 1;
+		/*
+		 * XXX: Should we check that next_free_rec contains
+		 * the extent?
+		 */
+		if (le32_to_cpu(el->l_recs[i].e_cpos) >= new_i_clusters)
 			metadata_delete = 1;
 	}
 
diff --git a/fs/ocfs2/alloc.h b/fs/ocfs2/alloc.h
index 0b82e80..fbcb593 100644
--- a/fs/ocfs2/alloc.h
+++ b/fs/ocfs2/alloc.h
@@ -31,7 +31,8 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
 			handle_t *handle,
 			struct inode *inode,
 			struct buffer_head *fe_bh,
-			u64 blkno,
+			u32 cpos,
+			u64 start_blk,
 			u32 new_clusters,
 			struct ocfs2_alloc_context *meta_ac);
 int ocfs2_num_free_extents(struct ocfs2_super *osb,
@@ -70,6 +71,8 @@ struct ocfs2_truncate_context {
 	struct buffer_head *tc_last_eb_bh;
 };
 
+int ocfs2_zero_tail_for_truncate(struct inode *inode, handle_t *handle,
+				 u64 new_i_size);
 int ocfs2_prepare_truncate(struct ocfs2_super *osb,
 			   struct inode *inode,
 			   struct buffer_head *fe_bh,
@@ -79,4 +82,26 @@ int ocfs2_commit_truncate(struct ocfs2_super *osb,
 			  struct buffer_head *fe_bh,
 			  struct ocfs2_truncate_context *tc);
 
+int ocfs2_find_leaf(struct inode *inode, struct ocfs2_extent_list *root_el,
+		    u32 cpos, struct buffer_head **leaf_bh);
+
+/*
+ * Helper function to look at the # of clusters in an extent record.
+ */
+static inline unsigned int ocfs2_rec_clusters(struct ocfs2_extent_list *el,
+					      struct ocfs2_extent_rec *rec)
+{
+	/*
+	 * Cluster count in extent records is slightly different
+	 * between interior nodes and leaf nodes. This is to support
+	 * unwritten extents which need a flags field in leaf node
+	 * records, thus shrinking the available space for a clusters
+	 * field.
+	 */
+	if (el->l_tree_depth)
+		return le32_to_cpu(rec->e_int_clusters);
+	else
+		return le16_to_cpu(rec->e_leaf_clusters);
+}
+
 #endif /* OCFS2_ALLOC_H */
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index 875c114..56963e6 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -24,6 +24,8 @@
 #include <linux/highmem.h>
 #include <linux/pagemap.h>
 #include <asm/byteorder.h>
+#include <linux/swap.h>
+#include <linux/pipe_fs_i.h>
 
 #define MLOG_MASK_PREFIX ML_FILE_IO
 #include <cluster/masklog.h>
@@ -37,6 +39,7 @@
 #include "file.h"
 #include "inode.h"
 #include "journal.h"
+#include "suballoc.h"
 #include "super.h"
 #include "symlink.h"
 
@@ -134,7 +137,9 @@ static int ocfs2_get_block(struct inode *inode, sector_t iblock,
 			   struct buffer_head *bh_result, int create)
 {
 	int err = 0;
+	unsigned int ext_flags;
 	u64 p_blkno, past_eof;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 
 	mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
 		   (unsigned long long)iblock, bh_result, create);
@@ -149,17 +154,8 @@ static int ocfs2_get_block(struct inode *inode, sector_t iblock,
 		goto bail;
 	}
 
-	/* this can happen if another node truncs after our extend! */
-	spin_lock(&OCFS2_I(inode)->ip_lock);
-	if (iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
-					       OCFS2_I(inode)->ip_clusters))
-		err = -EIO;
-	spin_unlock(&OCFS2_I(inode)->ip_lock);
-	if (err)
-		goto bail;
-
-	err = ocfs2_extent_map_get_blocks(inode, iblock, 1, &p_blkno,
-					  NULL);
+	err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, NULL,
+					  &ext_flags);
 	if (err) {
 		mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
 		     "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
@@ -167,22 +163,39 @@ static int ocfs2_get_block(struct inode *inode, sector_t iblock,
 		goto bail;
 	}
 
-	map_bh(bh_result, inode->i_sb, p_blkno);
-
-	if (bh_result->b_blocknr == 0) {
-		err = -EIO;
-		mlog(ML_ERROR, "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
-		     (unsigned long long)iblock,
-		     (unsigned long long)p_blkno,
-		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
-	}
+	/*
+	 * ocfs2 never allocates in this function - the only time we
+	 * need to use BH_New is when we're extending i_size on a file
+	 * system which doesn't support holes, in which case BH_New
+	 * allows block_prepare_write() to zero.
+	 */
+	mlog_bug_on_msg(create && p_blkno == 0 && ocfs2_sparse_alloc(osb),
+			"ino %lu, iblock %llu\n", inode->i_ino,
+			(unsigned long long)iblock);
+
+	/* Treat the unwritten extent as a hole for zeroing purposes. */
+	if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
+		map_bh(bh_result, inode->i_sb, p_blkno);
+
+	if (!ocfs2_sparse_alloc(osb)) {
+		if (p_blkno == 0) {
+			err = -EIO;
+			mlog(ML_ERROR,
+			     "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
+			     (unsigned long long)iblock,
+			     (unsigned long long)p_blkno,
+			     (unsigned long long)OCFS2_I(inode)->ip_blkno);
+			mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
+			dump_stack();
+		}
 
-	past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
-	mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino,
-	     (unsigned long long)past_eof);
+		past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
+		mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino,
+		     (unsigned long long)past_eof);
 
-	if (create && (iblock >= past_eof))
-		set_buffer_new(bh_result);
+		if (create && (iblock >= past_eof))
+			set_buffer_new(bh_result);
+	}
 
 bail:
 	if (err < 0)
@@ -276,8 +289,11 @@ static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
 	return ret;
 }
 
-/* This can also be called from ocfs2_write_zero_page() which has done
- * it's own cluster locking. */
+/*
+ * This is called from ocfs2_write_zero_page() which has handled it's
+ * own cluster locking and has ensured allocation exists for those
+ * blocks to be written.
+ */
 int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
 			       unsigned from, unsigned to)
 {
@@ -292,44 +308,17 @@ int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
 	return ret;
 }
 
-/*
- * ocfs2_prepare_write() can be an outer-most ocfs2 call when it is called
- * from loopback.  It must be able to perform its own locking around
- * ocfs2_get_block().
- */
-static int ocfs2_prepare_write(struct file *file, struct page *page,
-			       unsigned from, unsigned to)
-{
-	struct inode *inode = page->mapping->host;
-	int ret;
-
-	mlog_entry("(0x%p, 0x%p, %u, %u)\n", file, page, from, to);
-
-	ret = ocfs2_meta_lock_with_page(inode, NULL, 0, page);
-	if (ret != 0) {
-		mlog_errno(ret);
-		goto out;
-	}
-
-	ret = ocfs2_prepare_write_nolock(inode, page, from, to);
-
-	ocfs2_meta_unlock(inode, 0);
-out:
-	mlog_exit(ret);
-	return ret;
-}
-
 /* Taken from ext3. We don't necessarily need the full blown
  * functionality yet, but IMHO it's better to cut and paste the whole
  * thing so we can avoid introducing our own bugs (and easily pick up
  * their fixes when they happen) --Mark */
-static int walk_page_buffers(	handle_t *handle,
-				struct buffer_head *head,
-				unsigned from,
-				unsigned to,
-				int *partial,
-				int (*fn)(	handle_t *handle,
-						struct buffer_head *bh))
+int walk_page_buffers(	handle_t *handle,
+			struct buffer_head *head,
+			unsigned from,
+			unsigned to,
+			int *partial,
+			int (*fn)(	handle_t *handle,
+					struct buffer_head *bh))
 {
 	struct buffer_head *bh;
 	unsigned block_start, block_end;
@@ -388,95 +377,6 @@ out:
 	return handle;
 }
 
-static int ocfs2_commit_write(struct file *file, struct page *page,
-			      unsigned from, unsigned to)
-{
-	int ret;
-	struct buffer_head *di_bh = NULL;
-	struct inode *inode = page->mapping->host;
-	handle_t *handle = NULL;
-	struct ocfs2_dinode *di;
-
-	mlog_entry("(0x%p, 0x%p, %u, %u)\n", file, page, from, to);
-
-	/* NOTE: ocfs2_file_aio_write has ensured that it's safe for
-	 * us to continue here without rechecking the I/O against
-	 * changed inode values.
-	 *
-	 * 1) We're currently holding the inode alloc lock, so no
-	 *    nodes can change it underneath us.
-	 *
-	 * 2) We've had to take the metadata lock at least once
-	 *    already to check for extending writes, suid removal, etc.
-	 *    The meta data update code then ensures that we don't get a
-	 *    stale inode allocation image (i_size, i_clusters, etc).
-	 */
-
-	ret = ocfs2_meta_lock_with_page(inode, &di_bh, 1, page);
-	if (ret != 0) {
-		mlog_errno(ret);
-		goto out;
-	}
-
-	ret = ocfs2_data_lock_with_page(inode, 1, page);
-	if (ret != 0) {
-		mlog_errno(ret);
-		goto out_unlock_meta;
-	}
-
-	handle = ocfs2_start_walk_page_trans(inode, page, from, to);
-	if (IS_ERR(handle)) {
-		ret = PTR_ERR(handle);
-		goto out_unlock_data;
-	}
-
-	/* Mark our buffer early. We'd rather catch this error up here
-	 * as opposed to after a successful commit_write which would
-	 * require us to set back inode->i_size. */
-	ret = ocfs2_journal_access(handle, inode, di_bh,
-				   OCFS2_JOURNAL_ACCESS_WRITE);
-	if (ret < 0) {
-		mlog_errno(ret);
-		goto out_commit;
-	}
-
-	/* might update i_size */
-	ret = generic_commit_write(file, page, from, to);
-	if (ret < 0) {
-		mlog_errno(ret);
-		goto out_commit;
-	}
-
-	di = (struct ocfs2_dinode *)di_bh->b_data;
-
-	/* ocfs2_mark_inode_dirty() is too heavy to use here. */
-	inode->i_mtime = inode->i_ctime = CURRENT_TIME;
-	di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
-	di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
-
-	inode->i_blocks = ocfs2_align_bytes_to_sectors((u64)(i_size_read(inode)));
-	di->i_size = cpu_to_le64((u64)i_size_read(inode));
-
-	ret = ocfs2_journal_dirty(handle, di_bh);
-	if (ret < 0) {
-		mlog_errno(ret);
-		goto out_commit;
-	}
-
-out_commit:
-	ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
-out_unlock_data:
-	ocfs2_data_unlock(inode, 1);
-out_unlock_meta:
-	ocfs2_meta_unlock(inode, 1);
-out:
-	if (di_bh)
-		brelse(di_bh);
-
-	mlog_exit(ret);
-	return ret;
-}
-
 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
 {
 	sector_t status;
@@ -499,8 +399,7 @@ static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
 		down_read(&OCFS2_I(inode)->ip_alloc_sem);
 	}
 
-	err = ocfs2_extent_map_get_blocks(inode, block, 1, &p_blkno,
-					  NULL);
+	err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, NULL);
 
 	if (!INODE_JOURNAL(inode)) {
 		up_read(&OCFS2_I(inode)->ip_alloc_sem);
@@ -540,8 +439,8 @@ static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
 				     struct buffer_head *bh_result, int create)
 {
 	int ret;
-	u64 p_blkno, inode_blocks;
-	int contig_blocks;
+	u64 p_blkno, inode_blocks, contig_blocks;
+	unsigned int ext_flags;
 	unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
 	unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
 
@@ -549,33 +448,20 @@ static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
 	 * nicely aligned and of the right size, so there's no need
 	 * for us to check any of that. */
 
-	spin_lock(&OCFS2_I(inode)->ip_lock);
-	inode_blocks = ocfs2_clusters_to_blocks(inode->i_sb,
-						OCFS2_I(inode)->ip_clusters);
-
-	/*
-	 * For a read which begins past the end of file, we return a hole.
-	 */
-	if (!create && (iblock >= inode_blocks)) {
-		spin_unlock(&OCFS2_I(inode)->ip_lock);
-		ret = 0;
-		goto bail;
-	}
+	inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
 
 	/*
 	 * Any write past EOF is not allowed because we'd be extending.
 	 */
 	if (create && (iblock + max_blocks) > inode_blocks) {
-		spin_unlock(&OCFS2_I(inode)->ip_lock);
 		ret = -EIO;
 		goto bail;
 	}
-	spin_unlock(&OCFS2_I(inode)->ip_lock);
 
 	/* This figures out the size of the next contiguous block, and
 	 * our logical offset */
-	ret = ocfs2_extent_map_get_blocks(inode, iblock, 1, &p_blkno,
-					  &contig_blocks);
+	ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
+					  &contig_blocks, &ext_flags);
 	if (ret) {
 		mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
 		     (unsigned long long)iblock);
@@ -583,7 +469,37 @@ static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
 		goto bail;
 	}
 
-	map_bh(bh_result, inode->i_sb, p_blkno);
+	if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)) && !p_blkno) {
+		ocfs2_error(inode->i_sb,
+			    "Inode %llu has a hole at block %llu\n",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			    (unsigned long long)iblock);
+		ret = -EROFS;
+		goto bail;
+	}
+
+	/*
+	 * get_more_blocks() expects us to describe a hole by clearing
+	 * the mapped bit on bh_result().
+	 *
+	 * Consider an unwritten extent as a hole.
+	 */
+	if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
+		map_bh(bh_result, inode->i_sb, p_blkno);
+	else {
+		/*
+		 * ocfs2_prepare_inode_for_write() should have caught
+		 * the case where we'd be filling a hole and triggered
+		 * a buffered write instead.
+		 */
+		if (create) {
+			ret = -EIO;
+			mlog_errno(ret);
+			goto bail;
+		}
+
+		clear_buffer_mapped(bh_result);
+	}
 
 	/* make sure we don't map more than max_blocks blocks here as
 	   that's all the kernel will handle at this point. */
@@ -606,12 +522,17 @@ static void ocfs2_dio_end_io(struct kiocb *iocb,
 			     void *private)
 {
 	struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
+	int level;
 
 	/* this io's submitter should not have unlocked this before we could */
 	BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
+
 	ocfs2_iocb_clear_rw_locked(iocb);
-	up_read(&inode->i_alloc_sem);
-	ocfs2_rw_unlock(inode, 0);
+
+	level = ocfs2_iocb_rw_locked_level(iocb);
+	if (!level)
+		up_read(&inode->i_alloc_sem);
+	ocfs2_rw_unlock(inode, level);
 }
 
 /*
@@ -647,23 +568,27 @@ static ssize_t ocfs2_direct_IO(int rw,
 
 	mlog_entry_void();
 
-	/*
-	 * We get PR data locks even for O_DIRECT.  This allows
-	 * concurrent O_DIRECT I/O but doesn't let O_DIRECT with
-	 * extending and buffered zeroing writes race.  If they did
-	 * race then the buffered zeroing could be written back after
-	 * the O_DIRECT I/O.  It's one thing to tell people not to mix
-	 * buffered and O_DIRECT writes, but expecting them to
-	 * understand that file extension is also an implicit buffered
-	 * write is too much.  By getting the PR we force writeback of
-	 * the buffered zeroing before proceeding.
-	 */
-	ret = ocfs2_data_lock(inode, 0);
-	if (ret < 0) {
-		mlog_errno(ret);
-		goto out;
+	if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
+		/*
+		 * We get PR data locks even for O_DIRECT.  This
+		 * allows concurrent O_DIRECT I/O but doesn't let
+		 * O_DIRECT with extending and buffered zeroing writes
+		 * race.  If they did race then the buffered zeroing
+		 * could be written back after the O_DIRECT I/O.  It's
+		 * one thing to tell people not to mix buffered and
+		 * O_DIRECT writes, but expecting them to understand
+		 * that file extension is also an implicit buffered
+		 * write is too much.  By getting the PR we force
+		 * writeback of the buffered zeroing before
+		 * proceeding.
+		 */
+		ret = ocfs2_data_lock(inode, 0);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+		ocfs2_data_unlock(inode, 0);
 	}
-	ocfs2_data_unlock(inode, 0);
 
 	ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
 					    inode->i_sb->s_bdev, iov, offset,
@@ -675,11 +600,715 @@ out:
 	return ret;
 }
 
+static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
+					    u32 cpos,
+					    unsigned int *start,
+					    unsigned int *end)
+{
+	unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
+
+	if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
+		unsigned int cpp;
+
+		cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
+
+		cluster_start = cpos % cpp;
+		cluster_start = cluster_start << osb->s_clustersize_bits;
+
+		cluster_end = cluster_start + osb->s_clustersize;
+	}
+
+	BUG_ON(cluster_start > PAGE_SIZE);
+	BUG_ON(cluster_end > PAGE_SIZE);
+
+	if (start)
+		*start = cluster_start;
+	if (end)
+		*end = cluster_end;
+}
+
+/*
+ * 'from' and 'to' are the region in the page to avoid zeroing.
+ *
+ * If pagesize > clustersize, this function will avoid zeroing outside
+ * of the cluster boundary.
+ *
+ * from == to == 0 is code for "zero the entire cluster region"
+ */
+static void ocfs2_clear_page_regions(struct page *page,
+				     struct ocfs2_super *osb, u32 cpos,
+				     unsigned from, unsigned to)
+{
+	void *kaddr;
+	unsigned int cluster_start, cluster_end;
+
+	ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
+
+	kaddr = kmap_atomic(page, KM_USER0);
+
+	if (from || to) {
+		if (from > cluster_start)
+			memset(kaddr + cluster_start, 0, from - cluster_start);
+		if (to < cluster_end)
+			memset(kaddr + to, 0, cluster_end - to);
+	} else {
+		memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
+	}
+
+	kunmap_atomic(kaddr, KM_USER0);
+}
+
+/*
+ * Some of this taken from block_prepare_write(). We already have our
+ * mapping by now though, and the entire write will be allocating or
+ * it won't, so not much need to use BH_New.
+ *
+ * This will also skip zeroing, which is handled externally.
+ */
+int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
+			  struct inode *inode, unsigned int from,
+			  unsigned int to, int new)
+{
+	int ret = 0;
+	struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
+	unsigned int block_end, block_start;
+	unsigned int bsize = 1 << inode->i_blkbits;
+
+	if (!page_has_buffers(page))
+		create_empty_buffers(page, bsize, 0);
+
+	head = page_buffers(page);
+	for (bh = head, block_start = 0; bh != head || !block_start;
+	     bh = bh->b_this_page, block_start += bsize) {
+		block_end = block_start + bsize;
+
+		/*
+		 * Ignore blocks outside of our i/o range -
+		 * they may belong to unallocated clusters.
+		 */
+		if (block_start >= to || block_end <= from) {
+			if (PageUptodate(page))
+				set_buffer_uptodate(bh);
+			continue;
+		}
+
+		/*
+		 * For an allocating write with cluster size >= page
+		 * size, we always write the entire page.
+		 */
+
+		if (buffer_new(bh))
+			clear_buffer_new(bh);
+
+		if (!buffer_mapped(bh)) {
+			map_bh(bh, inode->i_sb, *p_blkno);
+			unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
+		}
+
+		if (PageUptodate(page)) {
+			if (!buffer_uptodate(bh))
+				set_buffer_uptodate(bh);
+		} else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
+		     (block_start < from || block_end > to)) {
+			ll_rw_block(READ, 1, &bh);
+			*wait_bh++=bh;
+		}
+
+		*p_blkno = *p_blkno + 1;
+	}
+
+	/*
+	 * If we issued read requests - let them complete.
+	 */
+	while(wait_bh > wait) {
+		wait_on_buffer(*--wait_bh);
+		if (!buffer_uptodate(*wait_bh))
+			ret = -EIO;
+	}
+
+	if (ret == 0 || !new)
+		return ret;
+
+	/*
+	 * If we get -EIO above, zero out any newly allocated blocks
+	 * to avoid exposing stale data.
+	 */
+	bh = head;
+	block_start = 0;
+	do {
+		void *kaddr;
+
+		block_end = block_start + bsize;
+		if (block_end <= from)
+			goto next_bh;
+		if (block_start >= to)
+			break;
+
+		kaddr = kmap_atomic(page, KM_USER0);
+		memset(kaddr+block_start, 0, bh->b_size);
+		flush_dcache_page(page);
+		kunmap_atomic(kaddr, KM_USER0);
+		set_buffer_uptodate(bh);
+		mark_buffer_dirty(bh);
+
+next_bh:
+		block_start = block_end;
+		bh = bh->b_this_page;
+	} while (bh != head);
+
+	return ret;
+}
+
+/*
+ * This will copy user data from the buffer page in the splice
+ * context.
+ *
+ * For now, we ignore SPLICE_F_MOVE as that would require some extra
+ * communication out all the way to ocfs2_write().
+ */
+int ocfs2_map_and_write_splice_data(struct inode *inode,
+				  struct ocfs2_write_ctxt *wc, u64 *p_blkno,
+				  unsigned int *ret_from, unsigned int *ret_to)
+{
+	int ret;
+	unsigned int to, from, cluster_start, cluster_end;
+	char *src, *dst;
+	struct ocfs2_splice_write_priv *sp = wc->w_private;
+	struct pipe_buffer *buf = sp->s_buf;
+	unsigned long bytes, src_from;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	ocfs2_figure_cluster_boundaries(osb, wc->w_cpos, &cluster_start,
+					&cluster_end);
+
+	from = sp->s_offset;
+	src_from = sp->s_buf_offset;
+	bytes = wc->w_count;
+
+	if (wc->w_large_pages) {
+		/*
+		 * For cluster size < page size, we have to
+		 * calculate pos within the cluster and obey
+		 * the rightmost boundary.
+		 */
+		bytes = min(bytes, (unsigned long)(osb->s_clustersize
+				   - (wc->w_pos & (osb->s_clustersize - 1))));
+	}
+	to = from + bytes;
+
+	if (wc->w_this_page_new)
+		ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
+					    cluster_start, cluster_end, 1);
+	else
+		ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
+					    from, to, 0);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	BUG_ON(from > PAGE_CACHE_SIZE);
+	BUG_ON(to > PAGE_CACHE_SIZE);
+	BUG_ON(from > osb->s_clustersize);
+	BUG_ON(to > osb->s_clustersize);
+
+	src = buf->ops->map(sp->s_pipe, buf, 1);
+	dst = kmap_atomic(wc->w_this_page, KM_USER1);
+	memcpy(dst + from, src + src_from, bytes);
+	kunmap_atomic(wc->w_this_page, KM_USER1);
+	buf->ops->unmap(sp->s_pipe, buf, src);
+
+	wc->w_finished_copy = 1;
+
+	*ret_from = from;
+	*ret_to = to;
+out:
+
+	return bytes ? (unsigned int)bytes : ret;
+}
+
+/*
+ * This will copy user data from the iovec in the buffered write
+ * context.
+ */
+int ocfs2_map_and_write_user_data(struct inode *inode,
+				  struct ocfs2_write_ctxt *wc, u64 *p_blkno,
+				  unsigned int *ret_from, unsigned int *ret_to)
+{
+	int ret;
+	unsigned int to, from, cluster_start, cluster_end;
+	unsigned long bytes, src_from;
+	char *dst;
+	struct ocfs2_buffered_write_priv *bp = wc->w_private;
+	const struct iovec *cur_iov = bp->b_cur_iov;
+	char __user *buf;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	ocfs2_figure_cluster_boundaries(osb, wc->w_cpos, &cluster_start,
+					&cluster_end);
+
+	buf = cur_iov->iov_base + bp->b_cur_off;
+	src_from = (unsigned long)buf & ~PAGE_CACHE_MASK;
+
+	from = wc->w_pos & (PAGE_CACHE_SIZE - 1);
+
+	/*
+	 * This is a lot of comparisons, but it reads quite
+	 * easily, which is important here.
+	 */
+	/* Stay within the src page */
+	bytes = PAGE_SIZE - src_from;
+	/* Stay within the vector */
+	bytes = min(bytes,
+		    (unsigned long)(cur_iov->iov_len - bp->b_cur_off));
+	/* Stay within count */
+	bytes = min(bytes, (unsigned long)wc->w_count);
+	/*
+	 * For clustersize > page size, just stay within
+	 * target page, otherwise we have to calculate pos
+	 * within the cluster and obey the rightmost
+	 * boundary.
+	 */
+	if (wc->w_large_pages) {
+		/*
+		 * For cluster size < page size, we have to
+		 * calculate pos within the cluster and obey
+		 * the rightmost boundary.
+		 */
+		bytes = min(bytes, (unsigned long)(osb->s_clustersize
+				   - (wc->w_pos & (osb->s_clustersize - 1))));
+	} else {
+		/*
+		 * cluster size > page size is the most common
+		 * case - we just stay within the target page
+		 * boundary.
+		 */
+		bytes = min(bytes, PAGE_CACHE_SIZE - from);
+	}
+
+	to = from + bytes;
+
+	if (wc->w_this_page_new)
+		ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
+					    cluster_start, cluster_end, 1);
+	else
+		ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
+					    from, to, 0);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	BUG_ON(from > PAGE_CACHE_SIZE);
+	BUG_ON(to > PAGE_CACHE_SIZE);
+	BUG_ON(from > osb->s_clustersize);
+	BUG_ON(to > osb->s_clustersize);
+
+	dst = kmap(wc->w_this_page);
+	memcpy(dst + from, bp->b_src_buf + src_from, bytes);
+	kunmap(wc->w_this_page);
+
+	/*
+	 * XXX: This is slow, but simple. The caller of
+	 * ocfs2_buffered_write_cluster() is responsible for
+	 * passing through the iovecs, so it's difficult to
+	 * predict what our next step is in here after our
+	 * initial write. A future version should be pushing
+	 * that iovec manipulation further down.
+	 *
+	 * By setting this, we indicate that a copy from user
+	 * data was done, and subsequent calls for this
+	 * cluster will skip copying more data.
+	 */
+	wc->w_finished_copy = 1;
+
+	*ret_from = from;
+	*ret_to = to;
+out:
+
+	return bytes ? (unsigned int)bytes : ret;
+}
+
+/*
+ * Map, fill and write a page to disk.
+ *
+ * The work of copying data is done via callback.  Newly allocated
+ * pages which don't take user data will be zero'd (set 'new' to
+ * indicate an allocating write)
+ *
+ * Returns a negative error code or the number of bytes copied into
+ * the page.
+ */
+int ocfs2_write_data_page(struct inode *inode, handle_t *handle,
+			  u64 *p_blkno, struct page *page,
+			  struct ocfs2_write_ctxt *wc, int new)
+{
+	int ret, copied = 0;
+	unsigned int from = 0, to = 0;
+	unsigned int cluster_start, cluster_end;
+	unsigned int zero_from = 0, zero_to = 0;
+
+	ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), wc->w_cpos,
+					&cluster_start, &cluster_end);
+
+	if ((wc->w_pos >> PAGE_CACHE_SHIFT) == page->index
+	    && !wc->w_finished_copy) {
+
+		wc->w_this_page = page;
+		wc->w_this_page_new = new;
+		ret = wc->w_write_data_page(inode, wc, p_blkno, &from, &to);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		copied = ret;
+
+		zero_from = from;
+		zero_to = to;
+		if (new) {
+			from = cluster_start;
+			to = cluster_end;
+		}
+	} else {
+		/*
+		 * If we haven't allocated the new page yet, we
+		 * shouldn't be writing it out without copying user
+		 * data. This is likely a math error from the caller.
+		 */
+		BUG_ON(!new);
+
+		from = cluster_start;
+		to = cluster_end;
+
+		ret = ocfs2_map_page_blocks(page, p_blkno, inode,
+					    cluster_start, cluster_end, 1);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	/*
+	 * Parts of newly allocated pages need to be zero'd.
+	 *
+	 * Above, we have also rewritten 'to' and 'from' - as far as
+	 * the rest of the function is concerned, the entire cluster
+	 * range inside of a page needs to be written.
+	 *
+	 * We can skip this if the page is up to date - it's already
+	 * been zero'd from being read in as a hole.
+	 */
+	if (new && !PageUptodate(page))
+		ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
+					 wc->w_cpos, zero_from, zero_to);
+
+	flush_dcache_page(page);
+
+	if (ocfs2_should_order_data(inode)) {
+		ret = walk_page_buffers(handle,
+					page_buffers(page),
+					from, to, NULL,
+					ocfs2_journal_dirty_data);
+		if (ret < 0)
+			mlog_errno(ret);
+	}
+
+	/*
+	 * We don't use generic_commit_write() because we need to
+	 * handle our own i_size update.
+	 */
+	ret = block_commit_write(page, from, to);
+	if (ret)
+		mlog_errno(ret);
+out:
+
+	return copied ? copied : ret;
+}
+
+/*
+ * Do the actual write of some data into an inode. Optionally allocate
+ * in order to fulfill the write.
+ *
+ * cpos is the logical cluster offset within the file to write at
+ *
+ * 'phys' is the physical mapping of that offset. a 'phys' value of
+ * zero indicates that allocation is required. In this case, data_ac
+ * and meta_ac should be valid (meta_ac can be null if metadata
+ * allocation isn't required).
+ */
+static ssize_t ocfs2_write(struct file *file, u32 phys, handle_t *handle,
+			   struct buffer_head *di_bh,
+			   struct ocfs2_alloc_context *data_ac,
+			   struct ocfs2_alloc_context *meta_ac,
+			   struct ocfs2_write_ctxt *wc)
+{
+	int ret, i, numpages = 1, new;
+	unsigned int copied = 0;
+	u32 tmp_pos;
+	u64 v_blkno, p_blkno;
+	struct address_space *mapping = file->f_mapping;
+	struct inode *inode = mapping->host;
+	unsigned long index, start;
+	struct page **cpages;
+
+	new = phys == 0 ? 1 : 0;
+
+	/*
+	 * Figure out how many pages we'll be manipulating here. For
+	 * non allocating write, we just change the one
+	 * page. Otherwise, we'll need a whole clusters worth.
+	 */
+	if (new)
+		numpages = ocfs2_pages_per_cluster(inode->i_sb);
+
+	cpages = kzalloc(sizeof(*cpages) * numpages, GFP_NOFS);
+	if (!cpages) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		return ret;
+	}
+
+	/*
+	 * Fill our page array first. That way we've grabbed enough so
+	 * that we can zero and flush if we error after adding the
+	 * extent.
+	 */
+	if (new) {
+		start = ocfs2_align_clusters_to_page_index(inode->i_sb,
+							   wc->w_cpos);
+		v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, wc->w_cpos);
+	} else {
+		start = wc->w_pos >> PAGE_CACHE_SHIFT;
+		v_blkno = wc->w_pos >> inode->i_sb->s_blocksize_bits;
+	}
+
+	for(i = 0; i < numpages; i++) {
+		index = start + i;
+
+		cpages[i] = grab_cache_page(mapping, index);
+		if (!cpages[i]) {
+			ret = -ENOMEM;
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (new) {
+		/*
+		 * This is safe to call with the page locks - it won't take
+		 * any additional semaphores or cluster locks.
+		 */
+		tmp_pos = wc->w_cpos;
+		ret = ocfs2_do_extend_allocation(OCFS2_SB(inode->i_sb), inode,
+						 &tmp_pos, 1, di_bh, handle,
+						 data_ac, meta_ac, NULL);
+		/*
+		 * This shouldn't happen because we must have already
+		 * calculated the correct meta data allocation required. The
+		 * internal tree allocation code should know how to increase
+		 * transaction credits itself.
+		 *
+		 * If need be, we could handle -EAGAIN for a
+		 * RESTART_TRANS here.
+		 */
+		mlog_bug_on_msg(ret == -EAGAIN,
+				"Inode %llu: EAGAIN return during allocation.\n",
+				(unsigned long long)OCFS2_I(inode)->ip_blkno);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
+					  NULL);
+	if (ret < 0) {
+
+		/*
+		 * XXX: Should we go readonly here?
+		 */
+
+		mlog_errno(ret);
+		goto out;
+	}
+
+	BUG_ON(p_blkno == 0);
+
+	for(i = 0; i < numpages; i++) {
+		ret = ocfs2_write_data_page(inode, handle, &p_blkno, cpages[i],
+					    wc, new);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		copied += ret;
+	}
+
+out:
+	for(i = 0; i < numpages; i++) {
+		unlock_page(cpages[i]);
+		mark_page_accessed(cpages[i]);
+		page_cache_release(cpages[i]);
+	}
+	kfree(cpages);
+
+	return copied ? copied : ret;
+}
+
+static void ocfs2_write_ctxt_init(struct ocfs2_write_ctxt *wc,
+				  struct ocfs2_super *osb, loff_t pos,
+				  size_t count, ocfs2_page_writer *cb,
+				  void *cb_priv)
+{
+	wc->w_count = count;
+	wc->w_pos = pos;
+	wc->w_cpos = wc->w_pos >> osb->s_clustersize_bits;
+	wc->w_finished_copy = 0;
+
+	if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
+		wc->w_large_pages = 1;
+	else
+		wc->w_large_pages = 0;
+
+	wc->w_write_data_page = cb;
+	wc->w_private = cb_priv;
+}
+
+/*
+ * Write a cluster to an inode. The cluster may not be allocated yet,
+ * in which case it will be. This only exists for buffered writes -
+ * O_DIRECT takes a more "traditional" path through the kernel.
+ *
+ * The caller is responsible for incrementing pos, written counts, etc
+ *
+ * For file systems that don't support sparse files, pre-allocation
+ * and page zeroing up until cpos should be done prior to this
+ * function call.
+ *
+ * Callers should be holding i_sem, and the rw cluster lock.
+ *
+ * Returns the number of user bytes written, or less than zero for
+ * error.
+ */
+ssize_t ocfs2_buffered_write_cluster(struct file *file, loff_t pos,
+				     size_t count, ocfs2_page_writer *actor,
+				     void *priv)
+{
+	int ret, credits = OCFS2_INODE_UPDATE_CREDITS;
+	ssize_t written = 0;
+	u32 phys;
+	struct inode *inode = file->f_mapping->host;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dinode *di;
+	struct ocfs2_alloc_context *data_ac = NULL;
+	struct ocfs2_alloc_context *meta_ac = NULL;
+	handle_t *handle;
+	struct ocfs2_write_ctxt wc;
+
+	ocfs2_write_ctxt_init(&wc, osb, pos, count, actor, priv);
+
+	ret = ocfs2_meta_lock(inode, &di_bh, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	/*
+	 * Take alloc sem here to prevent concurrent lookups. That way
+	 * the mapping, zeroing and tree manipulation within
+	 * ocfs2_write() will be safe against ->readpage(). This
+	 * should also serve to lock out allocation from a shared
+	 * writeable region.
+	 */
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+	ret = ocfs2_get_clusters(inode, wc.w_cpos, &phys, NULL, NULL);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_meta;
+	}
+
+	/* phys == 0 means that allocation is required. */
+	if (phys == 0) {
+		ret = ocfs2_lock_allocators(inode, di, 1, &data_ac, &meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_meta;
+		}
+
+		credits = ocfs2_calc_extend_credits(inode->i_sb, di, 1);
+	}
+
+	ret = ocfs2_data_lock(inode, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_meta;
+	}
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out_data;
+	}
+
+	written = ocfs2_write(file, phys, handle, di_bh, data_ac,
+			      meta_ac, &wc);
+	if (written < 0) {
+		ret = written;
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = ocfs2_journal_access(handle, inode, di_bh,
+				   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	pos += written;
+	if (pos > inode->i_size) {
+		i_size_write(inode, pos);
+		mark_inode_dirty(inode);
+	}
+	inode->i_blocks = ocfs2_inode_sector_count(inode);
+	di->i_size = cpu_to_le64((u64)i_size_read(inode));
+	inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+	di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
+	di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
+
+	ret = ocfs2_journal_dirty(handle, di_bh);
+	if (ret)
+		mlog_errno(ret);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out_data:
+	ocfs2_data_unlock(inode, 1);
+
+out_meta:
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+	ocfs2_meta_unlock(inode, 1);
+
+out:
+	brelse(di_bh);
+	if (data_ac)
+		ocfs2_free_alloc_context(data_ac);
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+
+	return written ? written : ret;
+}
+
 const struct address_space_operations ocfs2_aops = {
 	.readpage	= ocfs2_readpage,
 	.writepage	= ocfs2_writepage,
-	.prepare_write	= ocfs2_prepare_write,
-	.commit_write	= ocfs2_commit_write,
 	.bmap		= ocfs2_bmap,
 	.sync_page	= block_sync_page,
 	.direct_IO	= ocfs2_direct_IO,
diff --git a/fs/ocfs2/aops.h b/fs/ocfs2/aops.h
index f446a15..45821d4 100644
--- a/fs/ocfs2/aops.h
+++ b/fs/ocfs2/aops.h
@@ -30,12 +30,83 @@ handle_t *ocfs2_start_walk_page_trans(struct inode *inode,
 							 unsigned from,
 							 unsigned to);
 
+int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
+			  struct inode *inode, unsigned int from,
+			  unsigned int to, int new);
+
+int walk_page_buffers(	handle_t *handle,
+			struct buffer_head *head,
+			unsigned from,
+			unsigned to,
+			int *partial,
+			int (*fn)(	handle_t *handle,
+					struct buffer_head *bh));
+
+struct ocfs2_write_ctxt;
+typedef int (ocfs2_page_writer)(struct inode *, struct ocfs2_write_ctxt *,
+				u64 *, unsigned int *, unsigned int *);
+
+ssize_t ocfs2_buffered_write_cluster(struct file *file, loff_t pos,
+				     size_t count, ocfs2_page_writer *actor,
+				     void *priv);
+
+struct ocfs2_write_ctxt {
+	size_t				w_count;
+	loff_t				w_pos;
+	u32				w_cpos;
+	unsigned int			w_finished_copy;
+
+	/* This is true if page_size > cluster_size */
+	unsigned int			w_large_pages;
+
+	/* Filler callback and private data */
+	ocfs2_page_writer		*w_write_data_page;
+	void				*w_private;
+
+	/* Only valid for the filler callback */
+	struct page			*w_this_page;
+	unsigned int			w_this_page_new;
+};
+
+struct ocfs2_buffered_write_priv {
+	char				*b_src_buf;
+	const struct iovec		*b_cur_iov; /* Current iovec */
+	size_t				b_cur_off; /* Offset in the
+						    * current iovec */
+};
+int ocfs2_map_and_write_user_data(struct inode *inode,
+				  struct ocfs2_write_ctxt *wc,
+				  u64 *p_blkno,
+				  unsigned int *ret_from,
+				  unsigned int *ret_to);
+
+struct ocfs2_splice_write_priv {
+	struct splice_desc		*s_sd;
+	struct pipe_buffer		*s_buf;
+	struct pipe_inode_info		*s_pipe;
+	/* Neither offset value is ever larger than one page */
+	unsigned int			s_offset;
+	unsigned int			s_buf_offset;
+};
+int ocfs2_map_and_write_splice_data(struct inode *inode,
+				    struct ocfs2_write_ctxt *wc,
+				    u64 *p_blkno,
+				    unsigned int *ret_from,
+				    unsigned int *ret_to);
+
 /* all ocfs2_dio_end_io()'s fault */
 #define ocfs2_iocb_is_rw_locked(iocb) \
 	test_bit(0, (unsigned long *)&iocb->private)
-#define ocfs2_iocb_set_rw_locked(iocb) \
-	set_bit(0, (unsigned long *)&iocb->private)
+static inline void ocfs2_iocb_set_rw_locked(struct kiocb *iocb, int level)
+{
+	set_bit(0, (unsigned long *)&iocb->private);
+	if (level)
+		set_bit(1, (unsigned long *)&iocb->private);
+	else
+		clear_bit(1, (unsigned long *)&iocb->private);
+}
 #define ocfs2_iocb_clear_rw_locked(iocb) \
 	clear_bit(0, (unsigned long *)&iocb->private)
-
+#define ocfs2_iocb_rw_locked_level(iocb) \
+	test_bit(1, (unsigned long *)&iocb->private)
 #endif /* OCFS2_FILE_H */
diff --git a/fs/ocfs2/cluster/quorum.c b/fs/ocfs2/cluster/quorum.c
index 4705d65..bbacf7d 100644
--- a/fs/ocfs2/cluster/quorum.c
+++ b/fs/ocfs2/cluster/quorum.c
@@ -46,6 +46,7 @@
 #include <linux/kernel.h>
 #include <linux/slab.h>
 #include <linux/workqueue.h>
+#include <linux/reboot.h>
 
 #include "heartbeat.h"
 #include "nodemanager.h"
@@ -72,7 +73,9 @@ static void o2quo_fence_self(void)
 	/* panic spins with interrupts enabled.  with preempt
 	 * threads can still schedule, etc, etc */
 	o2hb_stop_all_regions();
-	panic("ocfs2 is very sorry to be fencing this system by panicing\n");
+
+	printk("ocfs2 is very sorry to be fencing this system by restarting\n");
+	emergency_restart();
 }
 
 /* Indicate that a timeout occured on a hearbeat region write. The
diff --git a/fs/ocfs2/cluster/tcp_internal.h b/fs/ocfs2/cluster/tcp_internal.h
index 4dae5df..9606111 100644
--- a/fs/ocfs2/cluster/tcp_internal.h
+++ b/fs/ocfs2/cluster/tcp_internal.h
@@ -38,6 +38,9 @@
  * locking semantics of the file system using the protocol.  It should 
  * be somewhere else, I'm sure, but right now it isn't.
  *
+ * New in version 8:
+ * 	- Replace delete inode votes with a cluster lock
+ *
  * New in version 7:
  * 	- DLM join domain includes the live nodemap
  *
@@ -57,7 +60,7 @@
  * 	- full 64 bit i_size in the metadata lock lvbs
  * 	- introduction of "rw" lock and pushing meta/data locking down
  */
-#define O2NET_PROTOCOL_VERSION 7ULL
+#define O2NET_PROTOCOL_VERSION 8ULL
 struct o2net_handshake {
 	__be64	protocol_version;
 	__be64	connector_id;
diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
index 66821e1..67e6866 100644
--- a/fs/ocfs2/dir.c
+++ b/fs/ocfs2/dir.c
@@ -358,15 +358,17 @@ int ocfs2_do_extend_dir(struct super_block *sb,
 {
 	int status;
 	int extend;
-	u64 p_blkno;
+	u64 p_blkno, v_blkno;
 
 	spin_lock(&OCFS2_I(dir)->ip_lock);
 	extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
 	spin_unlock(&OCFS2_I(dir)->ip_lock);
 
 	if (extend) {
-		status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, 1,
-						    parent_fe_bh, handle,
+		u32 offset = OCFS2_I(dir)->ip_clusters;
+
+		status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, &offset,
+						    1, parent_fe_bh, handle,
 						    data_ac, meta_ac, NULL);
 		BUG_ON(status == -EAGAIN);
 		if (status < 0) {
@@ -375,9 +377,8 @@ int ocfs2_do_extend_dir(struct super_block *sb,
 		}
 	}
 
-	status = ocfs2_extent_map_get_blocks(dir, (dir->i_blocks >>
-						   (sb->s_blocksize_bits - 9)),
-					     1, &p_blkno, NULL);
+	v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
+	status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
 	if (status < 0) {
 		mlog_errno(status);
 		goto bail;
@@ -486,7 +487,7 @@ static int ocfs2_extend_dir(struct ocfs2_super *osb,
 
 	dir_i_size += dir->i_sb->s_blocksize;
 	i_size_write(dir, dir_i_size);
-	dir->i_blocks = ocfs2_align_bytes_to_sectors(dir_i_size);
+	dir->i_blocks = ocfs2_inode_sector_count(dir);
 	status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
 	if (status < 0) {
 		mlog_errno(status);
diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
index c558442..d836b98 100644
--- a/fs/ocfs2/dlm/dlmdomain.c
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -430,11 +430,10 @@ redo_bucket:
 
 			dlm_lockres_put(res);
 
-			cond_resched_lock(&dlm->spinlock);
-
 			if (dropped)
 				goto redo_bucket;
 		}
+		cond_resched_lock(&dlm->spinlock);
 		num += n;
 		mlog(0, "%s: touched %d lockreses in bucket %d "
 		     "(tot=%d)\n", dlm->name, n, i, num);
@@ -1035,7 +1034,7 @@ static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
 {
 	int status = 0, tmpstat, node;
 	struct domain_join_ctxt *ctxt;
-	enum dlm_query_join_response response;
+	enum dlm_query_join_response response = JOIN_DISALLOW;
 
 	mlog_entry("%p", dlm);
 
diff --git a/fs/ocfs2/dlm/dlmrecovery.c b/fs/ocfs2/dlm/dlmrecovery.c
index 6d4a83d..c1807a4 100644
--- a/fs/ocfs2/dlm/dlmrecovery.c
+++ b/fs/ocfs2/dlm/dlmrecovery.c
@@ -611,6 +611,7 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node)
 			}
 		} while (status != 0);
 
+		spin_lock(&dlm_reco_state_lock);
 		switch (ndata->state) {
 			case DLM_RECO_NODE_DATA_INIT:
 			case DLM_RECO_NODE_DATA_FINALIZE_SENT:
@@ -641,6 +642,7 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node)
 				     ndata->node_num, dead_node);
 				break;
 		}
+		spin_unlock(&dlm_reco_state_lock);
 	}
 
 	mlog(0, "done requesting all lock info\n");
diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c
index e335541..27e43b0 100644
--- a/fs/ocfs2/dlmglue.c
+++ b/fs/ocfs2/dlmglue.c
@@ -225,11 +225,17 @@ static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
 	.flags		= 0,
 };
 
+static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
+	.get_osb	= ocfs2_get_inode_osb,
+	.flags		= 0,
+};
+
 static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
 {
 	return lockres->l_type == OCFS2_LOCK_TYPE_META ||
 		lockres->l_type == OCFS2_LOCK_TYPE_DATA ||
-		lockres->l_type == OCFS2_LOCK_TYPE_RW;
+		lockres->l_type == OCFS2_LOCK_TYPE_RW ||
+		lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
 }
 
 static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
@@ -373,6 +379,9 @@ void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
 		case OCFS2_LOCK_TYPE_DATA:
 			ops = &ocfs2_inode_data_lops;
 			break;
+		case OCFS2_LOCK_TYPE_OPEN:
+			ops = &ocfs2_inode_open_lops;
+			break;
 		default:
 			mlog_bug_on_msg(1, "type: %d\n", type);
 			ops = NULL; /* thanks, gcc */
@@ -1129,6 +1138,12 @@ int ocfs2_create_new_inode_locks(struct inode *inode)
 		goto bail;
 	}
 
+	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
+	if (ret) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
 bail:
 	mlog_exit(ret);
 	return ret;
@@ -1182,6 +1197,99 @@ void ocfs2_rw_unlock(struct inode *inode, int write)
 	mlog_exit_void();
 }
 
+/*
+ * ocfs2_open_lock always get PR mode lock.
+ */
+int ocfs2_open_lock(struct inode *inode)
+{
+	int status = 0;
+	struct ocfs2_lock_res *lockres;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	BUG_ON(!inode);
+
+	mlog_entry_void();
+
+	mlog(0, "inode %llu take PRMODE open lock\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	if (ocfs2_mount_local(osb))
+		goto out;
+
+	lockres = &OCFS2_I(inode)->ip_open_lockres;
+
+	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
+				    LKM_PRMODE, 0, 0);
+	if (status < 0)
+		mlog_errno(status);
+
+out:
+	mlog_exit(status);
+	return status;
+}
+
+int ocfs2_try_open_lock(struct inode *inode, int write)
+{
+	int status = 0, level;
+	struct ocfs2_lock_res *lockres;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	BUG_ON(!inode);
+
+	mlog_entry_void();
+
+	mlog(0, "inode %llu try to take %s open lock\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+	     write ? "EXMODE" : "PRMODE");
+
+	if (ocfs2_mount_local(osb))
+		goto out;
+
+	lockres = &OCFS2_I(inode)->ip_open_lockres;
+
+	level = write ? LKM_EXMODE : LKM_PRMODE;
+
+	/*
+	 * The file system may already holding a PRMODE/EXMODE open lock.
+	 * Since we pass LKM_NOQUEUE, the request won't block waiting on
+	 * other nodes and the -EAGAIN will indicate to the caller that
+	 * this inode is still in use.
+	 */
+	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
+				    level, LKM_NOQUEUE, 0);
+
+out:
+	mlog_exit(status);
+	return status;
+}
+
+/*
+ * ocfs2_open_unlock unlock PR and EX mode open locks.
+ */
+void ocfs2_open_unlock(struct inode *inode)
+{
+	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	mlog_entry_void();
+
+	mlog(0, "inode %llu drop open lock\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	if (ocfs2_mount_local(osb))
+		goto out;
+
+	if(lockres->l_ro_holders)
+		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
+				     LKM_PRMODE);
+	if(lockres->l_ex_holders)
+		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
+				     LKM_EXMODE);
+
+out:
+	mlog_exit_void();
+}
+
 int ocfs2_data_lock_full(struct inode *inode,
 			 int write,
 			 int arg_flags)
@@ -1387,8 +1495,7 @@ static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
 	if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
 		inode->i_blocks = 0;
 	else
-		inode->i_blocks =
-			ocfs2_align_bytes_to_sectors(i_size_read(inode));
+		inode->i_blocks = ocfs2_inode_sector_count(inode);
 
 	inode->i_uid     = be32_to_cpu(lvb->lvb_iuid);
 	inode->i_gid     = be32_to_cpu(lvb->lvb_igid);
@@ -1479,12 +1586,15 @@ static int ocfs2_meta_lock_update(struct inode *inode,
 {
 	int status = 0;
 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
-	struct ocfs2_lock_res *lockres = NULL;
+	struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres;
 	struct ocfs2_dinode *fe;
 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 
 	mlog_entry_void();
 
+	if (ocfs2_mount_local(osb))
+		goto bail;
+
 	spin_lock(&oi->ip_lock);
 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
 		mlog(0, "Orphaned inode %llu was deleted while we "
@@ -1496,22 +1606,16 @@ static int ocfs2_meta_lock_update(struct inode *inode,
 	}
 	spin_unlock(&oi->ip_lock);
 
-	if (!ocfs2_mount_local(osb)) {
-		lockres = &oi->ip_meta_lockres;
-
-		if (!ocfs2_should_refresh_lock_res(lockres))
-			goto bail;
-	}
+	if (!ocfs2_should_refresh_lock_res(lockres))
+		goto bail;
 
 	/* This will discard any caching information we might have had
 	 * for the inode metadata. */
 	ocfs2_metadata_cache_purge(inode);
 
-	/* will do nothing for inode types that don't use the extent
-	 * map (directories, bitmap files, etc) */
 	ocfs2_extent_map_trunc(inode, 0);
 
-	if (lockres && ocfs2_meta_lvb_is_trustable(inode, lockres)) {
+	if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
 		mlog(0, "Trusting LVB on inode %llu\n",
 		     (unsigned long long)oi->ip_blkno);
 		ocfs2_refresh_inode_from_lvb(inode);
@@ -1558,8 +1662,7 @@ static int ocfs2_meta_lock_update(struct inode *inode,
 
 	status = 0;
 bail_refresh:
-	if (lockres)
-		ocfs2_complete_lock_res_refresh(lockres, status);
+	ocfs2_complete_lock_res_refresh(lockres, status);
 bail:
 	mlog_exit(status);
 	return status;
@@ -1630,7 +1733,6 @@ int ocfs2_meta_lock_full(struct inode *inode,
 		wait_event(osb->recovery_event,
 			   ocfs2_node_map_is_empty(osb, &osb->recovery_map));
 
-	acquired = 0;
 	lockres = &OCFS2_I(inode)->ip_meta_lockres;
 	level = ex ? LKM_EXMODE : LKM_PRMODE;
 	dlm_flags = 0;
@@ -2458,13 +2560,20 @@ int ocfs2_drop_inode_locks(struct inode *inode)
 	 * ocfs2_clear_inode has done it for us. */
 
 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
-			      &OCFS2_I(inode)->ip_data_lockres);
+			      &OCFS2_I(inode)->ip_open_lockres);
 	if (err < 0)
 		mlog_errno(err);
 
 	status = err;
 
 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
+			      &OCFS2_I(inode)->ip_data_lockres);
+	if (err < 0)
+		mlog_errno(err);
+	if (err < 0 && !status)
+		status = err;
+
+	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
 			      &OCFS2_I(inode)->ip_meta_lockres);
 	if (err < 0)
 		mlog_errno(err);
diff --git a/fs/ocfs2/dlmglue.h b/fs/ocfs2/dlmglue.h
index c343fca..59cb566 100644
--- a/fs/ocfs2/dlmglue.h
+++ b/fs/ocfs2/dlmglue.h
@@ -80,6 +80,9 @@ void ocfs2_data_unlock(struct inode *inode,
 		       int write);
 int ocfs2_rw_lock(struct inode *inode, int write);
 void ocfs2_rw_unlock(struct inode *inode, int write);
+int ocfs2_open_lock(struct inode *inode);
+int ocfs2_try_open_lock(struct inode *inode, int write);
+void ocfs2_open_unlock(struct inode *inode);
 int ocfs2_meta_lock_atime(struct inode *inode,
 			  struct vfsmount *vfsmnt,
 			  int *level);
diff --git a/fs/ocfs2/extent_map.c b/fs/ocfs2/extent_map.c
index 80ac69f..ba2b2ab 100644
--- a/fs/ocfs2/extent_map.c
+++ b/fs/ocfs2/extent_map.c
@@ -3,8 +3,7 @@
  *
  * extent_map.c
  *
- * In-memory extent map for OCFS2.  Man, this code was prettier in
- * the library.
+ * Block/Cluster mapping functions
  *
  * Copyright (C) 2004 Oracle.  All rights reserved.
  *
@@ -26,1016 +25,528 @@
 #include <linux/fs.h>
 #include <linux/init.h>
 #include <linux/types.h>
-#include <linux/slab.h>
-#include <linux/rbtree.h>
 
 #define MLOG_MASK_PREFIX ML_EXTENT_MAP
 #include <cluster/masklog.h>
 
 #include "ocfs2.h"
 
+#include "alloc.h"
 #include "extent_map.h"
 #include "inode.h"
 #include "super.h"
 
 #include "buffer_head_io.h"
 
-
 /*
- * SUCK SUCK SUCK
- * Our headers are so bad that struct ocfs2_extent_map is in ocfs.h
- */
-
-struct ocfs2_extent_map_entry {
-	struct rb_node e_node;
-	int e_tree_depth;
-	struct ocfs2_extent_rec e_rec;
-};
-
-struct ocfs2_em_insert_context {
-	int need_left;
-	int need_right;
-	struct ocfs2_extent_map_entry *new_ent;
-	struct ocfs2_extent_map_entry *old_ent;
-	struct ocfs2_extent_map_entry *left_ent;
-	struct ocfs2_extent_map_entry *right_ent;
-};
-
-static struct kmem_cache *ocfs2_em_ent_cachep = NULL;
-
-
-static struct ocfs2_extent_map_entry *
-ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
-			u32 cpos, u32 clusters,
-			struct rb_node ***ret_p,
-			struct rb_node **ret_parent);
-static int ocfs2_extent_map_insert(struct inode *inode,
-				   struct ocfs2_extent_rec *rec,
-				   int tree_depth);
-static int ocfs2_extent_map_insert_entry(struct ocfs2_extent_map *em,
-					 struct ocfs2_extent_map_entry *ent);
-static int ocfs2_extent_map_find_leaf(struct inode *inode,
-				      u32 cpos, u32 clusters,
-				      struct ocfs2_extent_list *el);
-static int ocfs2_extent_map_lookup_read(struct inode *inode,
-					u32 cpos, u32 clusters,
-					struct ocfs2_extent_map_entry **ret_ent);
-static int ocfs2_extent_map_try_insert(struct inode *inode,
-				       struct ocfs2_extent_rec *rec,
-				       int tree_depth,
-				       struct ocfs2_em_insert_context *ctxt);
-
-/* returns 1 only if the rec contains all the given clusters -- that is that
- * rec's cpos is <= the cluster cpos and that the rec endpoint (cpos +
- * clusters) is >= the argument's endpoint */
-static int ocfs2_extent_rec_contains_clusters(struct ocfs2_extent_rec *rec,
-					      u32 cpos, u32 clusters)
-{
-	if (le32_to_cpu(rec->e_cpos) > cpos)
-		return 0;
-	if (cpos + clusters > le32_to_cpu(rec->e_cpos) + 
-			      le32_to_cpu(rec->e_clusters))
-		return 0;
-	return 1;
-}
-
-
-/*
- * Find an entry in the tree that intersects the region passed in.
- * Note that this will find straddled intervals, it is up to the
- * callers to enforce any boundary conditions.
- *
- * Callers must hold ip_lock.  This lookup is not guaranteed to return
- * a tree_depth 0 match, and as such can race inserts if the lock
- * were not held.
+ * The extent caching implementation is intentionally trivial.
  *
- * The rb_node garbage lets insertion share the search.  Trivial
- * callers pass NULL.
+ * We only cache a small number of extents stored directly on the
+ * inode, so linear order operations are acceptable. If we ever want
+ * to increase the size of the extent map, then these algorithms must
+ * get smarter.
  */
-static struct ocfs2_extent_map_entry *
-ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
-			u32 cpos, u32 clusters,
-			struct rb_node ***ret_p,
-			struct rb_node **ret_parent)
+
+void ocfs2_extent_map_init(struct inode *inode)
 {
-	struct rb_node **p = &em->em_extents.rb_node;
-	struct rb_node *parent = NULL;
-	struct ocfs2_extent_map_entry *ent = NULL;
-
-	while (*p)
-	{
-		parent = *p;
-		ent = rb_entry(parent, struct ocfs2_extent_map_entry,
-			       e_node);
-		if ((cpos + clusters) <= le32_to_cpu(ent->e_rec.e_cpos)) {
-			p = &(*p)->rb_left;
-			ent = NULL;
-		} else if (cpos >= (le32_to_cpu(ent->e_rec.e_cpos) +
-				    le32_to_cpu(ent->e_rec.e_clusters))) {
-			p = &(*p)->rb_right;
-			ent = NULL;
-		} else
-			break;
-	}
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 
-	if (ret_p != NULL)
-		*ret_p = p;
-	if (ret_parent != NULL)
-		*ret_parent = parent;
-	return ent;
+	oi->ip_extent_map.em_num_items = 0;
+	INIT_LIST_HEAD(&oi->ip_extent_map.em_list);
 }
 
-/*
- * Find the leaf containing the interval we want.  While we're on our
- * way down the tree, fill in every record we see at any depth, because
- * we might want it later.
- *
- * Note that this code is run without ip_lock.  That's because it
- * sleeps while reading.  If someone is also filling the extent list at
- * the same time we are, we might have to restart.
- */
-static int ocfs2_extent_map_find_leaf(struct inode *inode,
-				      u32 cpos, u32 clusters,
-				      struct ocfs2_extent_list *el)
+static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
+				      unsigned int cpos,
+				      struct ocfs2_extent_map_item **ret_emi)
 {
-	int i, ret;
-	struct buffer_head *eb_bh = NULL;
-	u64 blkno;
-	u32 rec_end;
-	struct ocfs2_extent_block *eb;
-	struct ocfs2_extent_rec *rec;
-
-	/*
-	 * The bh data containing the el cannot change here, because
-	 * we hold alloc_sem.  So we can do this without other
-	 * locks.
-	 */
-	while (el->l_tree_depth)
-	{
-		blkno = 0;
-		for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
-			rec = &el->l_recs[i];
-			rec_end = (le32_to_cpu(rec->e_cpos) +
-				   le32_to_cpu(rec->e_clusters));
-
-			ret = -EBADR;
-			if (rec_end > OCFS2_I(inode)->ip_clusters) {
-				mlog_errno(ret);
-				ocfs2_error(inode->i_sb,
-					    "Extent %d at e_blkno %llu of inode %llu goes past ip_clusters of %u\n",
-					    i,
-					    (unsigned long long)le64_to_cpu(rec->e_blkno),
-					    (unsigned long long)OCFS2_I(inode)->ip_blkno,
-					    OCFS2_I(inode)->ip_clusters);
-				goto out_free;
-			}
-
-			if (rec_end <= cpos) {
-				ret = ocfs2_extent_map_insert(inode, rec,
-						le16_to_cpu(el->l_tree_depth));
-				if (ret && (ret != -EEXIST)) {
-					mlog_errno(ret);
-					goto out_free;
-				}
-				continue;
-			}
-			if ((cpos + clusters) <= le32_to_cpu(rec->e_cpos)) {
-				ret = ocfs2_extent_map_insert(inode, rec,
-						le16_to_cpu(el->l_tree_depth));
-				if (ret && (ret != -EEXIST)) {
-					mlog_errno(ret);
-					goto out_free;
-				}
-				continue;
-			}
+	unsigned int range;
+	struct ocfs2_extent_map_item *emi;
 
-			/*
-			 * We've found a record that matches our
-			 * interval.  We don't insert it because we're
-			 * about to traverse it.
-			 */
-
-			/* Check to see if we're stradling */
-			ret = -ESRCH;
-			if (!ocfs2_extent_rec_contains_clusters(rec,
-							        cpos,
-								clusters)) {
-				mlog_errno(ret);
-				goto out_free;
-			}
+	*ret_emi = NULL;
 
-			/*
-			 * If we've already found a record, the el has
-			 * two records covering the same interval.
-			 * EEEK!
-			 */
-			ret = -EBADR;
-			if (blkno) {
-				mlog_errno(ret);
-				ocfs2_error(inode->i_sb,
-					    "Multiple extents for (cpos = %u, clusters = %u) on inode %llu; e_blkno %llu and rec %d at e_blkno %llu\n",
-					    cpos, clusters,
-					    (unsigned long long)OCFS2_I(inode)->ip_blkno,
-					    (unsigned long long)blkno, i,
-					    (unsigned long long)le64_to_cpu(rec->e_blkno));
-				goto out_free;
-			}
+	list_for_each_entry(emi, &em->em_list, ei_list) {
+		range = emi->ei_cpos + emi->ei_clusters;
 
-			blkno = le64_to_cpu(rec->e_blkno);
-		}
+		if (cpos >= emi->ei_cpos && cpos < range) {
+			list_move(&emi->ei_list, &em->em_list);
 
-		/*
-		 * We don't support holes, and we're still up
-		 * in the branches, so we'd better have found someone
-		 */
-		ret = -EBADR;
-		if (!blkno) {
-			ocfs2_error(inode->i_sb,
-				    "No record found for (cpos = %u, clusters = %u) on inode %llu\n",
-				    cpos, clusters,
-				    (unsigned long long)OCFS2_I(inode)->ip_blkno);
-			mlog_errno(ret);
-			goto out_free;
-		}
-
-		if (eb_bh) {
-			brelse(eb_bh);
-			eb_bh = NULL;
-		}
-		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
-				       blkno, &eb_bh, OCFS2_BH_CACHED,
-				       inode);
-		if (ret) {
-			mlog_errno(ret);
-			goto out_free;
-		}
-		eb = (struct ocfs2_extent_block *)eb_bh->b_data;
-		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
-			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
-			ret = -EIO;
-			goto out_free;
+			*ret_emi = emi;
+			break;
 		}
-		el = &eb->h_list;
 	}
+}
 
-	BUG_ON(el->l_tree_depth);
-
-	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
-		rec = &el->l_recs[i];
-
-		if ((le32_to_cpu(rec->e_cpos) + le32_to_cpu(rec->e_clusters)) >
-		    OCFS2_I(inode)->ip_clusters) {
-			ret = -EBADR;
-			mlog_errno(ret);
-			ocfs2_error(inode->i_sb,
-				    "Extent %d at e_blkno %llu of inode %llu goes past ip_clusters of %u\n",
-				    i,
-				    (unsigned long long)le64_to_cpu(rec->e_blkno),
-				    (unsigned long long)OCFS2_I(inode)->ip_blkno,
-				    OCFS2_I(inode)->ip_clusters);
-			return ret;
-		}
-
-		ret = ocfs2_extent_map_insert(inode, rec,
-					      le16_to_cpu(el->l_tree_depth));
-		if (ret && (ret != -EEXIST)) {
-			mlog_errno(ret);
-			goto out_free;
-		}
+static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
+				   unsigned int *phys, unsigned int *len,
+				   unsigned int *flags)
+{
+	unsigned int coff;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_extent_map_item *emi;
+
+	spin_lock(&oi->ip_lock);
+
+	__ocfs2_extent_map_lookup(&oi->ip_extent_map, cpos, &emi);
+	if (emi) {
+		coff = cpos - emi->ei_cpos;
+		*phys = emi->ei_phys + coff;
+		if (len)
+			*len = emi->ei_clusters - coff;
+		if (flags)
+			*flags = emi->ei_flags;
 	}
 
-	ret = 0;
+	spin_unlock(&oi->ip_lock);
 
-out_free:
-	if (eb_bh)
-		brelse(eb_bh);
+	if (emi == NULL)
+		return -ENOENT;
 
-	return ret;
+	return 0;
 }
 
 /*
- * This lookup actually will read from disk.  It has one invariant:
- * It will never re-traverse blocks.  This means that all inserts should
- * be new regions or more granular regions (both allowed by insert).
+ * Forget about all clusters equal to or greater than cpos.
  */
-static int ocfs2_extent_map_lookup_read(struct inode *inode,
-					u32 cpos,
-					u32 clusters,
-					struct ocfs2_extent_map_entry **ret_ent)
+void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
 {
-	int ret;
-	u64 blkno;
-	struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
-	struct ocfs2_extent_map_entry *ent;
-	struct buffer_head *bh = NULL;
-	struct ocfs2_extent_block *eb;
-	struct ocfs2_dinode *di;
-	struct ocfs2_extent_list *el;
-
-	spin_lock(&OCFS2_I(inode)->ip_lock);
-	ent = ocfs2_extent_map_lookup(em, cpos, clusters, NULL, NULL);
-	if (ent) {
-		if (!ent->e_tree_depth) {
-			spin_unlock(&OCFS2_I(inode)->ip_lock);
-			*ret_ent = ent;
-			return 0;
-		}
-		blkno = le64_to_cpu(ent->e_rec.e_blkno);
-		spin_unlock(&OCFS2_I(inode)->ip_lock);
-
-		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno, &bh,
-				       OCFS2_BH_CACHED, inode);
-		if (ret) {
-			mlog_errno(ret);
-			if (bh)
-				brelse(bh);
-			return ret;
+	struct list_head *p, *n;
+	struct ocfs2_extent_map_item *emi;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_extent_map *em = &oi->ip_extent_map;
+	LIST_HEAD(tmp_list);
+	unsigned int range;
+
+	spin_lock(&oi->ip_lock);
+	list_for_each_safe(p, n, &em->em_list) {
+		emi = list_entry(p, struct ocfs2_extent_map_item, ei_list);
+
+		if (emi->ei_cpos >= cpos) {
+			/* Full truncate of this record. */
+			list_move(&emi->ei_list, &tmp_list);
+			BUG_ON(em->em_num_items == 0);
+			em->em_num_items--;
+			continue;
 		}
-		eb = (struct ocfs2_extent_block *)bh->b_data;
-		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
-			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
-			brelse(bh);
-			return -EIO;
-		}
-		el = &eb->h_list;
-	} else {
-		spin_unlock(&OCFS2_I(inode)->ip_lock);
 
-		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
-				       OCFS2_I(inode)->ip_blkno, &bh,
-				       OCFS2_BH_CACHED, inode);
-		if (ret) {
-			mlog_errno(ret);
-			if (bh)
-				brelse(bh);
-			return ret;
+		range = emi->ei_cpos + emi->ei_clusters;
+		if (range > cpos) {
+			/* Partial truncate */
+			emi->ei_clusters = cpos - emi->ei_cpos;
 		}
-		di = (struct ocfs2_dinode *)bh->b_data;
-		if (!OCFS2_IS_VALID_DINODE(di)) {
-			brelse(bh);
-			OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, di);
-			return -EIO;
-		}
-		el = &di->id2.i_list;
-	}
-
-	ret = ocfs2_extent_map_find_leaf(inode, cpos, clusters, el);
-	brelse(bh);
-	if (ret) {
-		mlog_errno(ret);
-		return ret;
 	}
+	spin_unlock(&oi->ip_lock);
 
-	ent = ocfs2_extent_map_lookup(em, cpos, clusters, NULL, NULL);
-	if (!ent) {
-		ret = -ESRCH;
-		mlog_errno(ret);
-		return ret;
+	list_for_each_safe(p, n, &tmp_list) {
+		emi = list_entry(p, struct ocfs2_extent_map_item, ei_list);
+		list_del(&emi->ei_list);
+		kfree(emi);
 	}
-
-	/* FIXME: Make sure this isn't a corruption */
-	BUG_ON(ent->e_tree_depth);
-
-	*ret_ent = ent;
-
-	return 0;
 }
 
 /*
- * Callers must hold ip_lock.  This can insert pieces of the tree,
- * thus racing lookup if the lock weren't held.
+ * Is any part of emi2 contained within emi1
  */
-static int ocfs2_extent_map_insert_entry(struct ocfs2_extent_map *em,
-					 struct ocfs2_extent_map_entry *ent)
+static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item *emi1,
+				 struct ocfs2_extent_map_item *emi2)
 {
-	struct rb_node **p, *parent;
-	struct ocfs2_extent_map_entry *old_ent;
+	unsigned int range1, range2;
 
-	old_ent = ocfs2_extent_map_lookup(em, le32_to_cpu(ent->e_rec.e_cpos),
-					  le32_to_cpu(ent->e_rec.e_clusters),
-					  &p, &parent);
-	if (old_ent)
-		return -EEXIST;
+	/*
+	 * Check if logical start of emi2 is inside emi1
+	 */
+	range1 = emi1->ei_cpos + emi1->ei_clusters;
+	if (emi2->ei_cpos >= emi1->ei_cpos && emi2->ei_cpos < range1)
+		return 1;
 
-	rb_link_node(&ent->e_node, parent, p);
-	rb_insert_color(&ent->e_node, &em->em_extents);
+	/*
+	 * Check if logical end of emi2 is inside emi1
+	 */
+	range2 = emi2->ei_cpos + emi2->ei_clusters;
+	if (range2 > emi1->ei_cpos && range2 <= range1)
+		return 1;
 
 	return 0;
 }
 
+static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item *dest,
+				  struct ocfs2_extent_map_item *src)
+{
+	dest->ei_cpos = src->ei_cpos;
+	dest->ei_phys = src->ei_phys;
+	dest->ei_clusters = src->ei_clusters;
+	dest->ei_flags = src->ei_flags;
+}
 
 /*
- * Simple rule: on any return code other than -EAGAIN, anything left
- * in the insert_context will be freed.
- *
- * Simple rule #2: A return code of -EEXIST from this function or
- * its calls to ocfs2_extent_map_insert_entry() signifies that another
- * thread beat us to the insert.  It is not an actual error, but it
- * tells the caller we have no more work to do.
+ * Try to merge emi with ins. Returns 1 if merge succeeds, zero
+ * otherwise.
  */
-static int ocfs2_extent_map_try_insert(struct inode *inode,
-				       struct ocfs2_extent_rec *rec,
-				       int tree_depth,
-				       struct ocfs2_em_insert_context *ctxt)
+static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item *emi,
+					 struct ocfs2_extent_map_item *ins)
 {
-	int ret;
-	struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
-	struct ocfs2_extent_map_entry *old_ent;
-
-	ctxt->need_left = 0;
-	ctxt->need_right = 0;
-	ctxt->old_ent = NULL;
-
-	spin_lock(&OCFS2_I(inode)->ip_lock);
-	ret = ocfs2_extent_map_insert_entry(em, ctxt->new_ent);
-	if (!ret) {
-		ctxt->new_ent = NULL;
-		goto out_unlock;
-	}
-
-	/* Since insert_entry failed, the map MUST have old_ent */
-	old_ent = ocfs2_extent_map_lookup(em, le32_to_cpu(rec->e_cpos),
-					  le32_to_cpu(rec->e_clusters),
-					  NULL, NULL);
-
-	BUG_ON(!old_ent);
-
-	if (old_ent->e_tree_depth < tree_depth) {
-		/* Another thread beat us to the lower tree_depth */
-		ret = -EEXIST;
-		goto out_unlock;
-	}
-
-	if (old_ent->e_tree_depth == tree_depth) {
-		/*
-		 * Another thread beat us to this tree_depth.
-		 * Let's make sure we agree with that thread (the
-		 * extent_rec should be identical).
-		 */
-		if (!memcmp(rec, &old_ent->e_rec,
-			    sizeof(struct ocfs2_extent_rec)))
-			ret = 0;
-		else
-			/* FIXME: Should this be ESRCH/EBADR??? */
-			ret = -EEXIST;
-
-		goto out_unlock;
-	}
-
 	/*
-	 * We do it in this order specifically so that no actual tree
-	 * changes occur until we have all the pieces we need.  We
-	 * don't want malloc failures to leave an inconsistent tree.
-	 * Whenever we drop the lock, another process could be
-	 * inserting.  Also note that, if another process just beat us
-	 * to an insert, we might not need the same pieces we needed
-	 * the first go round.  In the end, the pieces we need will
-	 * be used, and the pieces we don't will be freed.
+	 * Handle contiguousness
 	 */
-	ctxt->need_left = !!(le32_to_cpu(rec->e_cpos) >
-			     le32_to_cpu(old_ent->e_rec.e_cpos));
-	ctxt->need_right = !!((le32_to_cpu(old_ent->e_rec.e_cpos) +
-			       le32_to_cpu(old_ent->e_rec.e_clusters)) >
-			      (le32_to_cpu(rec->e_cpos) + le32_to_cpu(rec->e_clusters)));
-	ret = -EAGAIN;
-	if (ctxt->need_left) {
-		if (!ctxt->left_ent)
-			goto out_unlock;
-		*(ctxt->left_ent) = *old_ent;
-		ctxt->left_ent->e_rec.e_clusters =
-			cpu_to_le32(le32_to_cpu(rec->e_cpos) -
-				    le32_to_cpu(ctxt->left_ent->e_rec.e_cpos));
-	}
-	if (ctxt->need_right) {
-		if (!ctxt->right_ent)
-			goto out_unlock;
-		*(ctxt->right_ent) = *old_ent;
-		ctxt->right_ent->e_rec.e_cpos =
-			cpu_to_le32(le32_to_cpu(rec->e_cpos) +
-				    le32_to_cpu(rec->e_clusters));
-		ctxt->right_ent->e_rec.e_clusters =
-			cpu_to_le32((le32_to_cpu(old_ent->e_rec.e_cpos) +
-				     le32_to_cpu(old_ent->e_rec.e_clusters)) -
-				    le32_to_cpu(ctxt->right_ent->e_rec.e_cpos));
-	}
-
-	rb_erase(&old_ent->e_node, &em->em_extents);
-	/* Now that he's erased, set him up for deletion */
-	ctxt->old_ent = old_ent;
-
-	if (ctxt->need_left) {
-		ret = ocfs2_extent_map_insert_entry(em,
-						    ctxt->left_ent);
-		if (ret)
-			goto out_unlock;
-		ctxt->left_ent = NULL;
+	if (ins->ei_phys == (emi->ei_phys + emi->ei_clusters) &&
+	    ins->ei_cpos == (emi->ei_cpos + emi->ei_clusters) &&
+	    ins->ei_flags == emi->ei_flags) {
+		emi->ei_clusters += ins->ei_clusters;
+		return 1;
+	} else if ((ins->ei_phys + ins->ei_clusters) == emi->ei_phys &&
+		   (ins->ei_cpos + ins->ei_clusters) == emi->ei_phys &&
+		   ins->ei_flags == emi->ei_flags) {
+		emi->ei_phys = ins->ei_phys;
+		emi->ei_cpos = ins->ei_cpos;
+		emi->ei_clusters += ins->ei_clusters;
+		return 1;
 	}
 
-	if (ctxt->need_right) {
-		ret = ocfs2_extent_map_insert_entry(em,
-						    ctxt->right_ent);
-		if (ret)
-			goto out_unlock;
-		ctxt->right_ent = NULL;
+	/*
+	 * Overlapping extents - this shouldn't happen unless we've
+	 * split an extent to change it's flags. That is exceedingly
+	 * rare, so there's no sense in trying to optimize it yet.
+	 */
+	if (ocfs2_ei_is_contained(emi, ins) ||
+	    ocfs2_ei_is_contained(ins, emi)) {
+		ocfs2_copy_emi_fields(emi, ins);
+		return 1;
 	}
 
-	ret = ocfs2_extent_map_insert_entry(em, ctxt->new_ent);
-
-	if (!ret)
-		ctxt->new_ent = NULL;
-
-out_unlock:
-	spin_unlock(&OCFS2_I(inode)->ip_lock);
-
-	return ret;
+	/* No merge was possible. */
+	return 0;
 }
 
-
-static int ocfs2_extent_map_insert(struct inode *inode,
-				   struct ocfs2_extent_rec *rec,
-				   int tree_depth)
+/*
+ * In order to reduce complexity on the caller, this insert function
+ * is intentionally liberal in what it will accept.
+ *
+ * The only rule is that the truncate call *must* be used whenever
+ * records have been deleted. This avoids inserting overlapping
+ * records with different physical mappings.
+ */
+void ocfs2_extent_map_insert_rec(struct inode *inode,
+				 struct ocfs2_extent_rec *rec)
 {
-	int ret;
-	struct ocfs2_em_insert_context ctxt = {0, };
-
-	if ((le32_to_cpu(rec->e_cpos) + le32_to_cpu(rec->e_clusters)) >
-	    OCFS2_I(inode)->ip_map.em_clusters) {
-		ret = -EBADR;
-		mlog_errno(ret);
-		return ret;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_extent_map *em = &oi->ip_extent_map;
+	struct ocfs2_extent_map_item *emi, *new_emi = NULL;
+	struct ocfs2_extent_map_item ins;
+
+	ins.ei_cpos = le32_to_cpu(rec->e_cpos);
+	ins.ei_phys = ocfs2_blocks_to_clusters(inode->i_sb,
+					       le64_to_cpu(rec->e_blkno));
+	ins.ei_clusters = le16_to_cpu(rec->e_leaf_clusters);
+	ins.ei_flags = rec->e_flags;
+
+search:
+	spin_lock(&oi->ip_lock);
+
+	list_for_each_entry(emi, &em->em_list, ei_list) {
+		if (ocfs2_try_to_merge_extent_map(emi, &ins)) {
+			list_move(&emi->ei_list, &em->em_list);
+			spin_unlock(&oi->ip_lock);
+			goto out;
+		}
 	}
 
-	/* Zero e_clusters means a truncated tail record.  It better be EOF */
-	if (!rec->e_clusters) {
-		if ((le32_to_cpu(rec->e_cpos) + le32_to_cpu(rec->e_clusters)) !=
-		    OCFS2_I(inode)->ip_map.em_clusters) {
-			ret = -EBADR;
-			mlog_errno(ret);
-			ocfs2_error(inode->i_sb,
-				    "Zero e_clusters on non-tail extent record at e_blkno %llu on inode %llu\n",
-				    (unsigned long long)le64_to_cpu(rec->e_blkno),
-				    (unsigned long long)OCFS2_I(inode)->ip_blkno);
-			return ret;
-		}
+	/*
+	 * No item could be merged.
+	 *
+	 * Either allocate and add a new item, or overwrite the last recently
+	 * inserted.
+	 */
 
-		/* Ignore the truncated tail */
-		return 0;
-	}
+	if (em->em_num_items < OCFS2_MAX_EXTENT_MAP_ITEMS) {
+		if (new_emi == NULL) {
+			spin_unlock(&oi->ip_lock);
 
-	ret = -ENOMEM;
-	ctxt.new_ent = kmem_cache_alloc(ocfs2_em_ent_cachep,
-					GFP_NOFS);
-	if (!ctxt.new_ent) {
-		mlog_errno(ret);
-		return ret;
-	}
+			new_emi = kmalloc(sizeof(*new_emi), GFP_NOFS);
+			if (new_emi == NULL)
+				goto out;
 
-	ctxt.new_ent->e_rec = *rec;
-	ctxt.new_ent->e_tree_depth = tree_depth;
-
-	do {
-		ret = -ENOMEM;
-		if (ctxt.need_left && !ctxt.left_ent) {
-			ctxt.left_ent =
-				kmem_cache_alloc(ocfs2_em_ent_cachep,
-						 GFP_NOFS);
-			if (!ctxt.left_ent)
-				break;
-		}
-		if (ctxt.need_right && !ctxt.right_ent) {
-			ctxt.right_ent =
-				kmem_cache_alloc(ocfs2_em_ent_cachep,
-						 GFP_NOFS);
-			if (!ctxt.right_ent)
-				break;
+			goto search;
 		}
 
-		ret = ocfs2_extent_map_try_insert(inode, rec,
-						  tree_depth, &ctxt);
-	} while (ret == -EAGAIN);
-
-	if ((ret < 0) && (ret != -EEXIST))
-		mlog_errno(ret);
+		ocfs2_copy_emi_fields(new_emi, &ins);
+		list_add(&new_emi->ei_list, &em->em_list);
+		em->em_num_items++;
+		new_emi = NULL;
+	} else {
+		BUG_ON(list_empty(&em->em_list) || em->em_num_items == 0);
+		emi = list_entry(em->em_list.prev,
+				 struct ocfs2_extent_map_item, ei_list);
+		list_move(&emi->ei_list, &em->em_list);
+		ocfs2_copy_emi_fields(emi, &ins);
+	}
 
-	if (ctxt.left_ent)
-		kmem_cache_free(ocfs2_em_ent_cachep, ctxt.left_ent);
-	if (ctxt.right_ent)
-		kmem_cache_free(ocfs2_em_ent_cachep, ctxt.right_ent);
-	if (ctxt.old_ent)
-		kmem_cache_free(ocfs2_em_ent_cachep, ctxt.old_ent);
-	if (ctxt.new_ent)
-		kmem_cache_free(ocfs2_em_ent_cachep, ctxt.new_ent);
+	spin_unlock(&oi->ip_lock);
 
-	return ret;
+out:
+	if (new_emi)
+		kfree(new_emi);
 }
 
 /*
- * Append this record to the tail of the extent map.  It must be
- * tree_depth 0.  The record might be an extension of an existing
- * record, and as such that needs to be handled.  eg:
- *
- * Existing record in the extent map:
- *
- *	cpos = 10, len = 10
- *	|---------|
- *
- * New Record:
- *
- *	cpos = 10, len = 20
- *	|------------------|
- *
- * The passed record is the new on-disk record.  The new_clusters value
- * is how many clusters were added to the file.  If the append is a
- * contiguous append, the new_clusters has been added to
- * rec->e_clusters.  If the append is an entirely new extent, then
- * rec->e_clusters is == new_clusters.
+ * Return the 1st index within el which contains an extent start
+ * larger than v_cluster.
  */
-int ocfs2_extent_map_append(struct inode *inode,
-			    struct ocfs2_extent_rec *rec,
-			    u32 new_clusters)
+static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
+				       u32 v_cluster)
 {
-	int ret;
-	struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
-	struct ocfs2_extent_map_entry *ent;
-	struct ocfs2_extent_rec *old;
-
-	BUG_ON(!new_clusters);
-	BUG_ON(le32_to_cpu(rec->e_clusters) < new_clusters);
+	int i;
+	struct ocfs2_extent_rec *rec;
 
-	if (em->em_clusters < OCFS2_I(inode)->ip_clusters) {
-		/*
-		 * Size changed underneath us on disk.  Drop any
-		 * straddling records and update our idea of
-		 * i_clusters
-		 */
-		ocfs2_extent_map_drop(inode, em->em_clusters - 1);
-		em->em_clusters = OCFS2_I(inode)->ip_clusters;
-	}
+	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
+		rec = &el->l_recs[i];
 
-	mlog_bug_on_msg((le32_to_cpu(rec->e_cpos) +
-			 le32_to_cpu(rec->e_clusters)) !=
-			(em->em_clusters + new_clusters),
-			"Inode %llu:\n"
-			"rec->e_cpos = %u + rec->e_clusters = %u = %u\n"
-			"em->em_clusters = %u + new_clusters = %u = %u\n",
-			(unsigned long long)OCFS2_I(inode)->ip_blkno,
-			le32_to_cpu(rec->e_cpos), le32_to_cpu(rec->e_clusters),
-			le32_to_cpu(rec->e_cpos) + le32_to_cpu(rec->e_clusters),
-			em->em_clusters, new_clusters,
-			em->em_clusters + new_clusters);
-
-	em->em_clusters += new_clusters;
-
-	ret = -ENOENT;
-	if (le32_to_cpu(rec->e_clusters) > new_clusters) {
-		/* This is a contiguous append */
-		ent = ocfs2_extent_map_lookup(em, le32_to_cpu(rec->e_cpos), 1,
-					      NULL, NULL);
-		if (ent) {
-			old = &ent->e_rec;
-			BUG_ON((le32_to_cpu(rec->e_cpos) +
-				le32_to_cpu(rec->e_clusters)) !=
-				 (le32_to_cpu(old->e_cpos) +
-				  le32_to_cpu(old->e_clusters) +
-				  new_clusters));
-			if (ent->e_tree_depth == 0) {
-				BUG_ON(le32_to_cpu(old->e_cpos) !=
-				       le32_to_cpu(rec->e_cpos));
-				BUG_ON(le64_to_cpu(old->e_blkno) !=
-				       le64_to_cpu(rec->e_blkno));
-				ret = 0;
-			}
-			/*
-			 * Let non-leafs fall through as -ENOENT to
-			 * force insertion of the new leaf.
-			 */
-			le32_add_cpu(&old->e_clusters, new_clusters);
-		}
+		if (v_cluster < le32_to_cpu(rec->e_cpos))
+			break;
 	}
 
-	if (ret == -ENOENT)
-		ret = ocfs2_extent_map_insert(inode, rec, 0);
-	if (ret < 0)
-		mlog_errno(ret);
-	return ret;
+	return i;
 }
 
-#if 0
-/* Code here is included but defined out as it completes the extent
- * map api and may be used in the future. */
-
 /*
- * Look up the record containing this cluster offset.  This record is
- * part of the extent map.  Do not free it.  Any changes you make to
- * it will reflect in the extent map.  So, if your last extent
- * is (cpos = 10, clusters = 10) and you truncate the file by 5
- * clusters, you can do:
+ * Figure out the size of a hole which starts at v_cluster within the given
+ * extent list.
  *
- * ret = ocfs2_extent_map_get_rec(em, orig_size - 5, &rec);
- * rec->e_clusters -= 5;
+ * If there is no more allocation past v_cluster, we return the maximum
+ * cluster size minus v_cluster.
  *
- * The lookup does not read from disk.  If the map isn't filled in for
- * an entry, you won't find it.
- *
- * Also note that the returned record is valid until alloc_sem is
- * dropped.  After that, truncate and extend can happen.  Caveat Emptor.
+ * If we have in-inode extents, then el points to the dinode list and
+ * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
+ * containing el.
  */
-int ocfs2_extent_map_get_rec(struct inode *inode, u32 cpos,
-			     struct ocfs2_extent_rec **rec,
-			     int *tree_depth)
+static int ocfs2_figure_hole_clusters(struct inode *inode,
+				      struct ocfs2_extent_list *el,
+				      struct buffer_head *eb_bh,
+				      u32 v_cluster,
+				      u32 *num_clusters)
 {
-	int ret = -ENOENT;
-	struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
-	struct ocfs2_extent_map_entry *ent;
+	int ret, i;
+	struct buffer_head *next_eb_bh = NULL;
+	struct ocfs2_extent_block *eb, *next_eb;
 
-	*rec = NULL;
+	i = ocfs2_search_for_hole_index(el, v_cluster);
 
-	if (cpos >= OCFS2_I(inode)->ip_clusters)
-		return -EINVAL;
+	if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
+		eb = (struct ocfs2_extent_block *)eb_bh->b_data;
 
-	if (cpos >= em->em_clusters) {
 		/*
-		 * Size changed underneath us on disk.  Drop any
-		 * straddling records and update our idea of
-		 * i_clusters
+		 * Check the next leaf for any extents.
 		 */
-		ocfs2_extent_map_drop(inode, em->em_clusters - 1);
-		em->em_clusters = OCFS2_I(inode)->ip_clusters ;
-	}
-
-	ent = ocfs2_extent_map_lookup(&OCFS2_I(inode)->ip_map, cpos, 1,
-				      NULL, NULL);
 
-	if (ent) {
-		*rec = &ent->e_rec;
-		if (tree_depth)
-			*tree_depth = ent->e_tree_depth;
-		ret = 0;
-	}
+		if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
+			goto no_more_extents;
 
-	return ret;
-}
+		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
+				       le64_to_cpu(eb->h_next_leaf_blk),
+				       &next_eb_bh, OCFS2_BH_CACHED, inode);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+		next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
 
-int ocfs2_extent_map_get_clusters(struct inode *inode,
-				  u32 v_cpos, int count,
-				  u32 *p_cpos, int *ret_count)
-{
-	int ret;
-	u32 coff, ccount;
-	struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
-	struct ocfs2_extent_map_entry *ent = NULL;
+		if (!OCFS2_IS_VALID_EXTENT_BLOCK(next_eb)) {
+			ret = -EROFS;
+			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, next_eb);
+			goto out;
+		}
 
-	*p_cpos = ccount = 0;
+		el = &next_eb->h_list;
 
-	if ((v_cpos + count) > OCFS2_I(inode)->ip_clusters)
-		return -EINVAL;
+		i = ocfs2_search_for_hole_index(el, v_cluster);
+	}
 
-	if ((v_cpos + count) > em->em_clusters) {
+no_more_extents:
+	if (i == le16_to_cpu(el->l_next_free_rec)) {
 		/*
-		 * Size changed underneath us on disk.  Drop any
-		 * straddling records and update our idea of
-		 * i_clusters
+		 * We're at the end of our existing allocation. Just
+		 * return the maximum number of clusters we could
+		 * possibly allocate.
 		 */
-		ocfs2_extent_map_drop(inode, em->em_clusters - 1);
-		em->em_clusters = OCFS2_I(inode)->ip_clusters;
+		*num_clusters = UINT_MAX - v_cluster;
+	} else {
+		*num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
 	}
 
+	ret = 0;
+out:
+	brelse(next_eb_bh);
+	return ret;
+}
 
-	ret = ocfs2_extent_map_lookup_read(inode, v_cpos, count, &ent);
-	if (ret)
-		return ret;
+/*
+ * Return the index of the extent record which contains cluster #v_cluster.
+ * -1 is returned if it was not found.
+ *
+ * Should work fine on interior and exterior nodes.
+ */
+static int ocfs2_search_extent_list(struct ocfs2_extent_list *el,
+				    u32 v_cluster)
+{
+	int ret = -1;
+	int i;
+	struct ocfs2_extent_rec *rec;
+	u32 rec_end, rec_start, clusters;
 
-	if (ent) {
-		/* We should never find ourselves straddling an interval */
-		if (!ocfs2_extent_rec_contains_clusters(&ent->e_rec,
-							v_cpos,
-							count))
-			return -ESRCH;
+	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
+		rec = &el->l_recs[i];
 
-		coff = v_cpos - le32_to_cpu(ent->e_rec.e_cpos);
-		*p_cpos = ocfs2_blocks_to_clusters(inode->i_sb,
-				le64_to_cpu(ent->e_rec.e_blkno)) +
-			  coff;
+		rec_start = le32_to_cpu(rec->e_cpos);
+		clusters = ocfs2_rec_clusters(el, rec);
 
-		if (ret_count)
-			*ret_count = le32_to_cpu(ent->e_rec.e_clusters) - coff;
+		rec_end = rec_start + clusters;
 
-		return 0;
+		if (v_cluster >= rec_start && v_cluster < rec_end) {
+			ret = i;
+			break;
+		}
 	}
 
-
-	return -ENOENT;
+	return ret;
 }
 
-#endif  /*  0  */
-
-int ocfs2_extent_map_get_blocks(struct inode *inode,
-				u64 v_blkno, int count,
-				u64 *p_blkno, int *ret_count)
+int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
+		       u32 *p_cluster, u32 *num_clusters,
+		       unsigned int *extent_flags)
 {
-	int ret;
-	u64 boff;
-	u32 cpos, clusters;
-	int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
-	struct ocfs2_extent_map_entry *ent = NULL;
-	struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
+	int ret, i;
+	unsigned int flags = 0;
+	struct buffer_head *di_bh = NULL;
+	struct buffer_head *eb_bh = NULL;
+	struct ocfs2_dinode *di;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list *el;
 	struct ocfs2_extent_rec *rec;
+	u32 coff;
 
-	*p_blkno = 0;
-
-	cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
-	clusters = ocfs2_blocks_to_clusters(inode->i_sb,
-					    (u64)count + bpc - 1);
-	if ((cpos + clusters) > OCFS2_I(inode)->ip_clusters) {
-		ret = -EINVAL;
-		mlog_errno(ret);
-		return ret;
-	}
-
-	if ((cpos + clusters) > em->em_clusters) {
-		/*
-		 * Size changed underneath us on disk.  Drop any
-		 * straddling records and update our idea of
-		 * i_clusters
-		 */
-		ocfs2_extent_map_drop(inode, em->em_clusters - 1);
-		em->em_clusters = OCFS2_I(inode)->ip_clusters;
-	}
+	ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
+				      num_clusters, extent_flags);
+	if (ret == 0)
+		goto out;
 
-	ret = ocfs2_extent_map_lookup_read(inode, cpos, clusters, &ent);
+	ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), OCFS2_I(inode)->ip_blkno,
+			       &di_bh, OCFS2_BH_CACHED, inode);
 	if (ret) {
 		mlog_errno(ret);
-		return ret;
+		goto out;
 	}
 
-	if (ent)
-	{
-		rec = &ent->e_rec;
+	di = (struct ocfs2_dinode *) di_bh->b_data;
+	el = &di->id2.i_list;
 
-		/* We should never find ourselves straddling an interval */
-		if (!ocfs2_extent_rec_contains_clusters(rec, cpos, clusters)) {
-			ret = -ESRCH;
+	if (el->l_tree_depth) {
+		ret = ocfs2_find_leaf(inode, el, v_cluster, &eb_bh);
+		if (ret) {
 			mlog_errno(ret);
-			return ret;
+			goto out;
 		}
 
-		boff = ocfs2_clusters_to_blocks(inode->i_sb, cpos -
-						le32_to_cpu(rec->e_cpos));
-		boff += (v_blkno & (u64)(bpc - 1));
-		*p_blkno = le64_to_cpu(rec->e_blkno) + boff;
+		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+		el = &eb->h_list;
 
-		if (ret_count) {
-			*ret_count = ocfs2_clusters_to_blocks(inode->i_sb,
-					le32_to_cpu(rec->e_clusters)) - boff;
+		if (el->l_tree_depth) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %lu has non zero tree depth in "
+				    "leaf block %llu\n", inode->i_ino,
+				    (unsigned long long)eb_bh->b_blocknr);
+			ret = -EROFS;
+			goto out;
 		}
-
-		return 0;
 	}
 
-	return -ENOENT;
-}
-
-int ocfs2_extent_map_init(struct inode *inode)
-{
-	struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
-
-	em->em_extents = RB_ROOT;
-	em->em_clusters = 0;
-
-	return 0;
-}
-
-/* Needs the lock */
-static void __ocfs2_extent_map_drop(struct inode *inode,
-				    u32 new_clusters,
-				    struct rb_node **free_head,
-				    struct ocfs2_extent_map_entry **tail_ent)
-{
-	struct rb_node *node, *next;
-	struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
-	struct ocfs2_extent_map_entry *ent;
+	i = ocfs2_search_extent_list(el, v_cluster);
+	if (i == -1) {
+		/*
+		 * A hole was found. Return some canned values that
+		 * callers can key on. If asked for, num_clusters will
+		 * be populated with the size of the hole.
+		 */
+		*p_cluster = 0;
+		if (num_clusters) {
+			ret = ocfs2_figure_hole_clusters(inode, el, eb_bh,
+							 v_cluster,
+							 num_clusters);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+	} else {
+		rec = &el->l_recs[i];
 
-	*free_head = NULL;
+		BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
 
-	ent = NULL;
-	node = rb_last(&em->em_extents);
-	while (node)
-	{
-		next = rb_prev(node);
+		if (!rec->e_blkno) {
+			ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
+				    "record (%u, %u, 0)", inode->i_ino,
+				    le32_to_cpu(rec->e_cpos),
+				    ocfs2_rec_clusters(el, rec));
+			ret = -EROFS;
+			goto out;
+		}
 
-		ent = rb_entry(node, struct ocfs2_extent_map_entry,
-			       e_node);
-		if (le32_to_cpu(ent->e_rec.e_cpos) < new_clusters)
-			break;
+		coff = v_cluster - le32_to_cpu(rec->e_cpos);
 
-		rb_erase(&ent->e_node, &em->em_extents);
+		*p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
+						    le64_to_cpu(rec->e_blkno));
+		*p_cluster = *p_cluster + coff;
 
-		node->rb_right = *free_head;
-		*free_head = node;
+		if (num_clusters)
+			*num_clusters = ocfs2_rec_clusters(el, rec) - coff;
 
-		ent = NULL;
-		node = next;
-	}
+		flags = rec->e_flags;
 
-	/* Do we have an entry straddling new_clusters? */
-	if (tail_ent) {
-		if (ent &&
-		    ((le32_to_cpu(ent->e_rec.e_cpos) +
-		      le32_to_cpu(ent->e_rec.e_clusters)) > new_clusters))
-			*tail_ent = ent;
-		else
-			*tail_ent = NULL;
+		ocfs2_extent_map_insert_rec(inode, rec);
 	}
-}
-
-static void __ocfs2_extent_map_drop_cleanup(struct rb_node *free_head)
-{
-	struct rb_node *node;
-	struct ocfs2_extent_map_entry *ent;
 
-	while (free_head) {
-		node = free_head;
-		free_head = node->rb_right;
+	if (extent_flags)
+		*extent_flags = flags;
 
-		ent = rb_entry(node, struct ocfs2_extent_map_entry,
-			       e_node);
-		kmem_cache_free(ocfs2_em_ent_cachep, ent);
-	}
+out:
+	brelse(di_bh);
+	brelse(eb_bh);
+	return ret;
 }
 
 /*
- * Remove all entries past new_clusters, inclusive of an entry that
- * contains new_clusters.  This is effectively a cache forget.
- *
- * If you want to also clip the last extent by some number of clusters,
- * you need to call ocfs2_extent_map_trunc().
- * This code does not check or modify ip_clusters.
+ * This expects alloc_sem to be held. The allocation cannot change at
+ * all while the map is in the process of being updated.
  */
-int ocfs2_extent_map_drop(struct inode *inode, u32 new_clusters)
+int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
+				u64 *ret_count, unsigned int *extent_flags)
 {
-	struct rb_node *free_head = NULL;
-	struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
-	struct ocfs2_extent_map_entry *ent;
-
-	spin_lock(&OCFS2_I(inode)->ip_lock);
+	int ret;
+	int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
+	u32 cpos, num_clusters, p_cluster;
+	u64 boff = 0;
 
-	__ocfs2_extent_map_drop(inode, new_clusters, &free_head, &ent);
+	cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
 
-	if (ent) {
-		rb_erase(&ent->e_node, &em->em_extents);
-		ent->e_node.rb_right = free_head;
-		free_head = &ent->e_node;
+	ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
+				 extent_flags);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
 	}
 
-	spin_unlock(&OCFS2_I(inode)->ip_lock);
-
-	if (free_head)
-		__ocfs2_extent_map_drop_cleanup(free_head);
-
-	return 0;
-}
-
-/*
- * Remove all entries past new_clusters and also clip any extent
- * straddling new_clusters, if there is one.  This does not check
- * or modify ip_clusters
- */
-int ocfs2_extent_map_trunc(struct inode *inode, u32 new_clusters)
-{
-	struct rb_node *free_head = NULL;
-	struct ocfs2_extent_map_entry *ent = NULL;
-
-	spin_lock(&OCFS2_I(inode)->ip_lock);
-
-	__ocfs2_extent_map_drop(inode, new_clusters, &free_head, &ent);
-
-	if (ent)
-		ent->e_rec.e_clusters = cpu_to_le32(new_clusters -
-					       le32_to_cpu(ent->e_rec.e_cpos));
-
-	OCFS2_I(inode)->ip_map.em_clusters = new_clusters;
-
-	spin_unlock(&OCFS2_I(inode)->ip_lock);
-
-	if (free_head)
-		__ocfs2_extent_map_drop_cleanup(free_head);
-
-	return 0;
-}
+	/*
+	 * p_cluster == 0 indicates a hole.
+	 */
+	if (p_cluster) {
+		boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
+		boff += (v_blkno & (u64)(bpc - 1));
+	}
 
-int __init init_ocfs2_extent_maps(void)
-{
-	ocfs2_em_ent_cachep =
-		kmem_cache_create("ocfs2_em_ent",
-				  sizeof(struct ocfs2_extent_map_entry),
-				  0, SLAB_HWCACHE_ALIGN, NULL, NULL);
-	if (!ocfs2_em_ent_cachep)
-		return -ENOMEM;
+	*p_blkno = boff;
 
-	return 0;
-}
+	if (ret_count) {
+		*ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
+		*ret_count -= v_blkno & (u64)(bpc - 1);
+	}
 
-void exit_ocfs2_extent_maps(void)
-{
-	kmem_cache_destroy(ocfs2_em_ent_cachep);
+out:
+	return ret;
 }
diff --git a/fs/ocfs2/extent_map.h b/fs/ocfs2/extent_map.h
index fa3745e..de91e3e 100644
--- a/fs/ocfs2/extent_map.h
+++ b/fs/ocfs2/extent_map.h
@@ -25,22 +25,29 @@
 #ifndef _EXTENT_MAP_H
 #define _EXTENT_MAP_H
 
-int init_ocfs2_extent_maps(void);
-void exit_ocfs2_extent_maps(void);
+struct ocfs2_extent_map_item {
+	unsigned int			ei_cpos;
+	unsigned int			ei_phys;
+	unsigned int			ei_clusters;
+	unsigned int			ei_flags;
 
-/*
- * EVERY CALL here except _init, _trunc, and _drop expects alloc_sem
- * to be held.  The allocation cannot change at all while the map is
- * in the process of being updated.
- */
-int ocfs2_extent_map_init(struct inode *inode);
-int ocfs2_extent_map_append(struct inode *inode,
-			    struct ocfs2_extent_rec *rec,
-			    u32 new_clusters);
-int ocfs2_extent_map_get_blocks(struct inode *inode,
-				u64 v_blkno, int count,
-				u64 *p_blkno, int *ret_count);
-int ocfs2_extent_map_drop(struct inode *inode, u32 new_clusters);
-int ocfs2_extent_map_trunc(struct inode *inode, u32 new_clusters);
+	struct list_head		ei_list;
+};
+
+#define OCFS2_MAX_EXTENT_MAP_ITEMS			3
+struct ocfs2_extent_map {
+	unsigned int			em_num_items;
+	struct list_head		em_list;
+};
+
+void ocfs2_extent_map_init(struct inode *inode);
+void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cluster);
+void ocfs2_extent_map_insert_rec(struct inode *inode,
+				 struct ocfs2_extent_rec *rec);
+
+int ocfs2_get_clusters(struct inode *inode, u32 v_cluster, u32 *p_cluster,
+		       u32 *num_clusters, unsigned int *extent_flags);
+int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
+				u64 *ret_count, unsigned int *extent_flags);
 
 #endif  /* _EXTENT_MAP_H */
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index f2cd3bf..520a2a6 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -33,6 +33,7 @@
 #include <linux/sched.h>
 #include <linux/pipe_fs_i.h>
 #include <linux/mount.h>
+#include <linux/writeback.h>
 
 #define MLOG_MASK_PREFIX ML_INODE
 #include <cluster/masklog.h>
@@ -215,7 +216,7 @@ int ocfs2_set_inode_size(handle_t *handle,
 
 	mlog_entry_void();
 	i_size_write(inode, new_i_size);
-	inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size);
+	inode->i_blocks = ocfs2_inode_sector_count(inode);
 	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 
 	status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
@@ -261,6 +262,7 @@ static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 {
 	int status;
 	handle_t *handle;
+	struct ocfs2_dinode *di;
 
 	mlog_entry_void();
 
@@ -274,12 +276,39 @@ static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 		goto out;
 	}
 
-	status = ocfs2_set_inode_size(handle, inode, fe_bh, new_i_size);
+	status = ocfs2_journal_access(handle, inode, fe_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	/*
+	 * Do this before setting i_size.
+	 */
+	status = ocfs2_zero_tail_for_truncate(inode, handle, new_i_size);
+	if (status) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	i_size_write(inode, new_i_size);
+	inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size);
+	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
+
+	di = (struct ocfs2_dinode *) fe_bh->b_data;
+	di->i_size = cpu_to_le64(new_i_size);
+	di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
+	di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+
+	status = ocfs2_journal_dirty(handle, fe_bh);
 	if (status < 0)
 		mlog_errno(status);
 
+out_commit:
 	ocfs2_commit_trans(osb, handle);
 out:
+
 	mlog_exit(status);
 	return status;
 }
@@ -342,19 +371,6 @@ static int ocfs2_truncate_file(struct inode *inode,
 		mlog_errno(status);
 		goto bail;
 	}
-	ocfs2_data_unlock(inode, 1);
-
-	if (le32_to_cpu(fe->i_clusters) ==
-	    ocfs2_clusters_for_bytes(osb->sb, new_i_size)) {
-		mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n",
-		     fe->i_clusters);
-		/* No allocation change is required, so lets fast path
-		 * this truncate. */
-		status = ocfs2_simple_size_update(inode, di_bh, new_i_size);
-		if (status < 0)
-			mlog_errno(status);
-		goto bail;
-	}
 
 	/* alright, we're going to need to do a full blown alloc size
 	 * change. Orphan the inode so that recovery can complete the
@@ -363,22 +379,25 @@ static int ocfs2_truncate_file(struct inode *inode,
 	status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 	if (status < 0) {
 		mlog_errno(status);
-		goto bail;
+		goto bail_unlock_data;
 	}
 
 	status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
 	if (status < 0) {
 		mlog_errno(status);
-		goto bail;
+		goto bail_unlock_data;
 	}
 
 	status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
 	if (status < 0) {
 		mlog_errno(status);
-		goto bail;
+		goto bail_unlock_data;
 	}
 
 	/* TODO: orphan dir cleanup here. */
+bail_unlock_data:
+	ocfs2_data_unlock(inode, 1);
+
 bail:
 
 	mlog_exit(status);
@@ -397,6 +416,7 @@ bail:
  */
 int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 			       struct inode *inode,
+			       u32 *logical_offset,
 			       u32 clusters_to_add,
 			       struct buffer_head *fe_bh,
 			       handle_t *handle,
@@ -460,18 +480,14 @@ int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 	block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
 	mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
 	     num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
-	status = ocfs2_insert_extent(osb, handle, inode, fe_bh, block,
-				     num_bits, meta_ac);
+	status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
+				     *logical_offset, block, num_bits,
+				     meta_ac);
 	if (status < 0) {
 		mlog_errno(status);
 		goto leave;
 	}
 
-	le32_add_cpu(&fe->i_clusters, num_bits);
-	spin_lock(&OCFS2_I(inode)->ip_lock);
-	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
-	spin_unlock(&OCFS2_I(inode)->ip_lock);
-
 	status = ocfs2_journal_dirty(handle, fe_bh);
 	if (status < 0) {
 		mlog_errno(status);
@@ -479,6 +495,7 @@ int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 	}
 
 	clusters_to_add -= num_bits;
+	*logical_offset += num_bits;
 
 	if (clusters_to_add) {
 		mlog(0, "need to alloc once more, clusters = %u, wanted = "
@@ -494,14 +511,87 @@ leave:
 	return status;
 }
 
+/*
+ * For a given allocation, determine which allocators will need to be
+ * accessed, and lock them, reserving the appropriate number of bits.
+ *
+ * Called from ocfs2_extend_allocation() for file systems which don't
+ * support holes, and from ocfs2_write() for file systems which
+ * understand sparse inodes.
+ */
+int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
+			  u32 clusters_to_add,
+			  struct ocfs2_alloc_context **data_ac,
+			  struct ocfs2_alloc_context **meta_ac)
+{
+	int ret, num_free_extents;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	*meta_ac = NULL;
+	*data_ac = NULL;
+
+	mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
+	     "clusters_to_add = %u\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
+	     le32_to_cpu(di->i_clusters), clusters_to_add);
+
+	num_free_extents = ocfs2_num_free_extents(osb, inode, di);
+	if (num_free_extents < 0) {
+		ret = num_free_extents;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Sparse allocation file systems need to be more conservative
+	 * with reserving room for expansion - the actual allocation
+	 * happens while we've got a journal handle open so re-taking
+	 * a cluster lock (because we ran out of room for another
+	 * extent) will violate ordering rules.
+	 *
+	 * Most of the time we'll only be seeing this 1 cluster at a time
+	 * anyway.
+	 */
+	if (!num_free_extents ||
+	    (ocfs2_sparse_alloc(osb) && num_free_extents < clusters_to_add)) {
+		ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
+		if (ret < 0) {
+			if (ret != -ENOSPC)
+				mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
+	if (ret < 0) {
+		if (ret != -ENOSPC)
+			mlog_errno(ret);
+		goto out;
+	}
+
+out:
+	if (ret) {
+		if (*meta_ac) {
+			ocfs2_free_alloc_context(*meta_ac);
+			*meta_ac = NULL;
+		}
+
+		/*
+		 * We cannot have an error and a non null *data_ac.
+		 */
+	}
+
+	return ret;
+}
+
 static int ocfs2_extend_allocation(struct inode *inode,
 				   u32 clusters_to_add)
 {
 	int status = 0;
 	int restart_func = 0;
 	int drop_alloc_sem = 0;
-	int credits, num_free_extents;
-	u32 prev_clusters;
+	int credits;
+	u32 prev_clusters, logical_start;
 	struct buffer_head *bh = NULL;
 	struct ocfs2_dinode *fe = NULL;
 	handle_t *handle = NULL;
@@ -512,6 +602,12 @@ static int ocfs2_extend_allocation(struct inode *inode,
 
 	mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
 
+	/*
+	 * This function only exists for file systems which don't
+	 * support holes.
+	 */
+	BUG_ON(ocfs2_sparse_alloc(osb));
+
 	status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
 				  OCFS2_BH_CACHED, inode);
 	if (status < 0) {
@@ -526,39 +622,11 @@ static int ocfs2_extend_allocation(struct inode *inode,
 		goto leave;
 	}
 
+	logical_start = OCFS2_I(inode)->ip_clusters;
+
 restart_all:
 	BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 
-	mlog(0, "extend inode %llu, i_size = %lld, fe->i_clusters = %u, "
-	     "clusters_to_add = %u\n",
-	     (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
-	     fe->i_clusters, clusters_to_add);
-
-	num_free_extents = ocfs2_num_free_extents(osb,
-						  inode,
-						  fe);
-	if (num_free_extents < 0) {
-		status = num_free_extents;
-		mlog_errno(status);
-		goto leave;
-	}
-
-	if (!num_free_extents) {
-		status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac);
-		if (status < 0) {
-			if (status != -ENOSPC)
-				mlog_errno(status);
-			goto leave;
-		}
-	}
-
-	status = ocfs2_reserve_clusters(osb, clusters_to_add, &data_ac);
-	if (status < 0) {
-		if (status != -ENOSPC)
-			mlog_errno(status);
-		goto leave;
-	}
-
 	/* blocks peope in read/write from reading our allocation
 	 * until we're done changing it. We depend on i_mutex to block
 	 * other extend/truncate calls while we're here. Ordering wrt
@@ -566,6 +634,13 @@ restart_all:
 	down_write(&OCFS2_I(inode)->ip_alloc_sem);
 	drop_alloc_sem = 1;
 
+	status = ocfs2_lock_allocators(inode, fe, clusters_to_add, &data_ac,
+				       &meta_ac);
+	if (status) {
+		mlog_errno(status);
+		goto leave;
+	}
+
 	credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
 	handle = ocfs2_start_trans(osb, credits);
 	if (IS_ERR(handle)) {
@@ -590,6 +665,7 @@ restarted_transaction:
 
 	status = ocfs2_do_extend_allocation(osb,
 					    inode,
+					    &logical_start,
 					    clusters_to_add,
 					    bh,
 					    handle,
@@ -778,7 +854,7 @@ static int ocfs2_extend_file(struct inode *inode,
 			     size_t tail_to_skip)
 {
 	int ret = 0;
-	u32 clusters_to_add;
+	u32 clusters_to_add = 0;
 
 	BUG_ON(!tail_to_skip && !di_bh);
 
@@ -790,6 +866,11 @@ static int ocfs2_extend_file(struct inode *inode,
   		goto out;
 	BUG_ON(new_i_size < i_size_read(inode));
 
+	if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
+		BUG_ON(tail_to_skip != 0);
+		goto out_update_size;
+	}
+
 	clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) - 
 		OCFS2_I(inode)->ip_clusters;
 
@@ -825,6 +906,7 @@ static int ocfs2_extend_file(struct inode *inode,
 		goto out_unlock;
 	}
 
+out_update_size:
 	if (!tail_to_skip) {
 		/* We're being called from ocfs2_setattr() which wants
 		 * us to update i_size */
@@ -834,7 +916,8 @@ static int ocfs2_extend_file(struct inode *inode,
 	}
 
 out_unlock:
-	ocfs2_data_unlock(inode, 1);
+	if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
+		ocfs2_data_unlock(inode, 1);
 
 out:
 	return ret;
@@ -972,7 +1055,8 @@ int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd)
 
 	ret = ocfs2_meta_lock(inode, NULL, 0);
 	if (ret) {
-		mlog_errno(ret);
+		if (ret != -ENOENT)
+			mlog_errno(ret);
 		goto out;
 	}
 
@@ -1035,10 +1119,49 @@ out:
 	return ret;
 }
 
+/*
+ * Will look for holes and unwritten extents in the range starting at
+ * pos for count bytes (inclusive).
+ */
+static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
+				       size_t count)
+{
+	int ret = 0;
+	unsigned int extent_flags;
+	u32 cpos, clusters, extent_len, phys_cpos;
+	struct super_block *sb = inode->i_sb;
+
+	cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
+	clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
+
+	while (clusters) {
+		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
+					 &extent_flags);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
+			ret = 1;
+			break;
+		}
+
+		if (extent_len > clusters)
+			extent_len = clusters;
+
+		clusters -= extent_len;
+		cpos += extent_len;
+	}
+out:
+	return ret;
+}
+
 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 					 loff_t *ppos,
 					 size_t count,
-					 int appending)
+					 int appending,
+					 int *direct_io)
 {
 	int ret = 0, meta_level = appending;
 	struct inode *inode = dentry->d_inode;
@@ -1089,6 +1212,49 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 		} else {
 			saved_pos = *ppos;
 		}
+
+		if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
+			loff_t end = saved_pos + count;
+
+			/*
+			 * Skip the O_DIRECT checks if we don't need
+			 * them.
+			 */
+			if (!direct_io || !(*direct_io))
+				break;
+
+			/*
+			 * Allowing concurrent direct writes means
+			 * i_size changes wouldn't be synchronized, so
+			 * one node could wind up truncating another
+			 * nodes writes.
+			 */
+			if (end > i_size_read(inode)) {
+				*direct_io = 0;
+				break;
+			}
+
+			/*
+			 * We don't fill holes during direct io, so
+			 * check for them here. If any are found, the
+			 * caller will have to retake some cluster
+			 * locks and initiate the io as buffered.
+			 */
+			ret = ocfs2_check_range_for_holes(inode, saved_pos,
+							  count);
+			if (ret == 1) {
+				*direct_io = 0;
+				ret = 0;
+			} else if (ret < 0)
+				mlog_errno(ret);
+			break;
+		}
+
+		/*
+		 * The rest of this loop is concerned with legacy file
+		 * systems which don't support sparse files.
+		 */
+
 		newsize = count + saved_pos;
 
 		mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
@@ -1141,55 +1307,264 @@ out:
 	return ret;
 }
 
+static inline void
+ocfs2_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes)
+{
+	const struct iovec *iov = *iovp;
+	size_t base = *basep;
+
+	do {
+		int copy = min(bytes, iov->iov_len - base);
+
+		bytes -= copy;
+		base += copy;
+		if (iov->iov_len == base) {
+			iov++;
+			base = 0;
+		}
+	} while (bytes);
+	*iovp = iov;
+	*basep = base;
+}
+
+static struct page * ocfs2_get_write_source(struct ocfs2_buffered_write_priv *bp,
+					    const struct iovec *cur_iov,
+					    size_t iov_offset)
+{
+	int ret;
+	char *buf;
+	struct page *src_page = NULL;
+
+	buf = cur_iov->iov_base + iov_offset;
+
+	if (!segment_eq(get_fs(), KERNEL_DS)) {
+		/*
+		 * Pull in the user page. We want to do this outside
+		 * of the meta data locks in order to preserve locking
+		 * order in case of page fault.
+		 */
+		ret = get_user_pages(current, current->mm,
+				     (unsigned long)buf & PAGE_CACHE_MASK, 1,
+				     0, 0, &src_page, NULL);
+		if (ret == 1)
+			bp->b_src_buf = kmap(src_page);
+		else
+			src_page = ERR_PTR(-EFAULT);
+	} else {
+		bp->b_src_buf = buf;
+	}
+
+	return src_page;
+}
+
+static void ocfs2_put_write_source(struct ocfs2_buffered_write_priv *bp,
+				   struct page *page)
+{
+	if (page) {
+		kunmap(page);
+		page_cache_release(page);
+	}
+}
+
+static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos,
+					 const struct iovec *iov,
+					 unsigned long nr_segs,
+					 size_t count,
+					 ssize_t o_direct_written)
+{
+	int ret = 0;
+	ssize_t copied, total = 0;
+	size_t iov_offset = 0;
+	const struct iovec *cur_iov = iov;
+	struct ocfs2_buffered_write_priv bp;
+	struct page *page;
+
+	/*
+	 * handle partial DIO write.  Adjust cur_iov if needed.
+	 */
+	ocfs2_set_next_iovec(&cur_iov, &iov_offset, o_direct_written);
+
+	do {
+		bp.b_cur_off = iov_offset;
+		bp.b_cur_iov = cur_iov;
+
+		page = ocfs2_get_write_source(&bp, cur_iov, iov_offset);
+		if (IS_ERR(page)) {
+			ret = PTR_ERR(page);
+			goto out;
+		}
+
+		copied = ocfs2_buffered_write_cluster(file, *ppos, count,
+						      ocfs2_map_and_write_user_data,
+						      &bp);
+
+		ocfs2_put_write_source(&bp, page);
+
+		if (copied < 0) {
+			mlog_errno(copied);
+			ret = copied;
+			goto out;
+		}
+
+		total += copied;
+		*ppos = *ppos + copied;
+		count -= copied;
+
+		ocfs2_set_next_iovec(&cur_iov, &iov_offset, copied);
+	} while(count);
+
+out:
+	return total ? total : ret;
+}
+
+static int ocfs2_check_iovec(const struct iovec *iov, size_t *counted,
+			     unsigned long *nr_segs)
+{
+	size_t ocount;		/* original count */
+	unsigned long seg;
+
+	ocount = 0;
+	for (seg = 0; seg < *nr_segs; seg++) {
+		const struct iovec *iv = &iov[seg];
+
+		/*
+		 * If any segment has a negative length, or the cumulative
+		 * length ever wraps negative then return -EINVAL.
+		 */
+		ocount += iv->iov_len;
+		if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
+			return -EINVAL;
+		if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
+			continue;
+		if (seg == 0)
+			return -EFAULT;
+		*nr_segs = seg;
+		ocount -= iv->iov_len;	/* This segment is no good */
+		break;
+	}
+
+	*counted = ocount;
+	return 0;
+}
+
 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
 				    const struct iovec *iov,
 				    unsigned long nr_segs,
 				    loff_t pos)
 {
-	int ret, rw_level, have_alloc_sem = 0;
-	struct file *filp = iocb->ki_filp;
-	struct inode *inode = filp->f_path.dentry->d_inode;
-	int appending = filp->f_flags & O_APPEND ? 1 : 0;
-
-	mlog_entry("(0x%p, %u, '%.*s')\n", filp,
+	int ret, direct_io, appending, rw_level, have_alloc_sem  = 0;
+	int can_do_direct, sync = 0;
+	ssize_t written = 0;
+	size_t ocount;		/* original count */
+	size_t count;		/* after file limit checks */
+	loff_t *ppos = &iocb->ki_pos;
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file->f_path.dentry->d_inode;
+
+	mlog_entry("(0x%p, %u, '%.*s')\n", file,
 		   (unsigned int)nr_segs,
-		   filp->f_path.dentry->d_name.len,
-		   filp->f_path.dentry->d_name.name);
+		   file->f_path.dentry->d_name.len,
+		   file->f_path.dentry->d_name.name);
 
-	/* happy write of zero bytes */
 	if (iocb->ki_left == 0)
 		return 0;
 
+	ret = ocfs2_check_iovec(iov, &ocount, &nr_segs);
+	if (ret)
+		return ret;
+
+	count = ocount;
+
+	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
+
+	appending = file->f_flags & O_APPEND ? 1 : 0;
+	direct_io = file->f_flags & O_DIRECT ? 1 : 0;
+
 	mutex_lock(&inode->i_mutex);
+
+relock:
 	/* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
-	if (filp->f_flags & O_DIRECT) {
-		have_alloc_sem = 1;
+	if (direct_io) {
 		down_read(&inode->i_alloc_sem);
+		have_alloc_sem = 1;
 	}
 
 	/* concurrent O_DIRECT writes are allowed */
-	rw_level = (filp->f_flags & O_DIRECT) ? 0 : 1;
+	rw_level = !direct_io;
 	ret = ocfs2_rw_lock(inode, rw_level);
 	if (ret < 0) {
-		rw_level = -1;
 		mlog_errno(ret);
-		goto out;
+		goto out_sems;
 	}
 
-	ret = ocfs2_prepare_inode_for_write(filp->f_path.dentry, &iocb->ki_pos,
-					    iocb->ki_left, appending);
+	can_do_direct = direct_io;
+	ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
+					    iocb->ki_left, appending,
+					    &can_do_direct);
 	if (ret < 0) {
 		mlog_errno(ret);
 		goto out;
 	}
 
-	/* communicate with ocfs2_dio_end_io */
-	ocfs2_iocb_set_rw_locked(iocb);
+	/*
+	 * We can't complete the direct I/O as requested, fall back to
+	 * buffered I/O.
+	 */
+	if (direct_io && !can_do_direct) {
+		ocfs2_rw_unlock(inode, rw_level);
+		up_read(&inode->i_alloc_sem);
+
+		have_alloc_sem = 0;
+		rw_level = -1;
 
-	ret = generic_file_aio_write_nolock(iocb, iov, nr_segs, iocb->ki_pos);
+		direct_io = 0;
+		sync = 1;
+		goto relock;
+	}
+
+	if (!sync && ((file->f_flags & O_SYNC) || IS_SYNC(inode)))
+		sync = 1;
+
+	/*
+	 * XXX: Is it ok to execute these checks a second time?
+	 */
+	ret = generic_write_checks(file, ppos, &count, S_ISBLK(inode->i_mode));
+	if (ret)
+		goto out;
+
+	/*
+	 * Set pos so that sync_page_range_nolock() below understands
+	 * where to start from. We might've moved it around via the
+	 * calls above. The range we want to actually sync starts from
+	 * *ppos here.
+	 *
+	 */
+	pos = *ppos;
+
+	/* communicate with ocfs2_dio_end_io */
+	ocfs2_iocb_set_rw_locked(iocb, rw_level);
+
+	if (direct_io) {
+		written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
+						    ppos, count, ocount);
+		if (written < 0) {
+			ret = written;
+			goto out_dio;
+		}
+	} else {
+		written = ocfs2_file_buffered_write(file, ppos, iov, nr_segs,
+						    count, written);
+		if (written < 0) {
+			ret = written;
+			if (ret != -EFAULT || ret != -ENOSPC)
+				mlog_errno(ret);
+			goto out;
+		}
+	}
 
+out_dio:
 	/* buffered aio wouldn't have proper lock coverage today */
-	BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
+	BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
 
 	/* 
 	 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
@@ -1207,13 +1582,102 @@ static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
 	}
 
 out:
+	if (rw_level != -1)
+		ocfs2_rw_unlock(inode, rw_level);
+
+out_sems:
 	if (have_alloc_sem)
 		up_read(&inode->i_alloc_sem);
-	if (rw_level != -1) 
-		ocfs2_rw_unlock(inode, rw_level);
+
+	if (written > 0 && sync) {
+		ssize_t err;
+
+		err = sync_page_range_nolock(inode, file->f_mapping, pos, count);
+		if (err < 0)
+			written = err;
+	}
+
 	mutex_unlock(&inode->i_mutex);
 
 	mlog_exit(ret);
+	return written ? written : ret;
+}
+
+static int ocfs2_splice_write_actor(struct pipe_inode_info *pipe,
+				    struct pipe_buffer *buf,
+				    struct splice_desc *sd)
+{
+	int ret, count, total = 0;
+	ssize_t copied = 0;
+	struct ocfs2_splice_write_priv sp;
+
+	ret = buf->ops->pin(pipe, buf);
+	if (ret)
+		goto out;
+
+	sp.s_sd = sd;
+	sp.s_buf = buf;
+	sp.s_pipe = pipe;
+	sp.s_offset = sd->pos & ~PAGE_CACHE_MASK;
+	sp.s_buf_offset = buf->offset;
+
+	count = sd->len;
+	if (count + sp.s_offset > PAGE_CACHE_SIZE)
+		count = PAGE_CACHE_SIZE - sp.s_offset;
+
+	do {
+		/*
+		 * splice wants us to copy up to one page at a
+		 * time. For pagesize > cluster size, this means we
+		 * might enter ocfs2_buffered_write_cluster() more
+		 * than once, so keep track of our progress here.
+		 */
+		copied = ocfs2_buffered_write_cluster(sd->file,
+						      (loff_t)sd->pos + total,
+						      count,
+						      ocfs2_map_and_write_splice_data,
+						      &sp);
+		if (copied < 0) {
+			mlog_errno(copied);
+			ret = copied;
+			goto out;
+		}
+
+		count -= copied;
+		sp.s_offset += copied;
+		sp.s_buf_offset += copied;
+		total += copied;
+	} while (count);
+
+	ret = 0;
+out:
+
+	return total ? total : ret;
+}
+
+static ssize_t __ocfs2_file_splice_write(struct pipe_inode_info *pipe,
+					 struct file *out,
+					 loff_t *ppos,
+					 size_t len,
+					 unsigned int flags)
+{
+	int ret, err;
+	struct address_space *mapping = out->f_mapping;
+	struct inode *inode = mapping->host;
+
+	ret = __splice_from_pipe(pipe, out, ppos, len, flags,
+				 ocfs2_splice_write_actor);
+	if (ret > 0) {
+		*ppos += ret;
+
+		if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
+			err = generic_osync_inode(inode, mapping,
+						  OSYNC_METADATA|OSYNC_DATA);
+			if (err)
+				ret = err;
+		}
+	}
+
 	return ret;
 }
 
@@ -1239,14 +1703,15 @@ static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
 		goto out;
 	}
 
-	ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0);
+	ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
+					    NULL);
 	if (ret < 0) {
 		mlog_errno(ret);
 		goto out_unlock;
 	}
 
 	/* ok, we're done with i_size and alloc work */
-	ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags);
+	ret = __ocfs2_file_splice_write(pipe, out, ppos, len, flags);
 
 out_unlock:
 	ocfs2_rw_unlock(inode, 1);
@@ -1323,7 +1788,7 @@ static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
 		}
 		rw_level = 0;
 		/* communicate with ocfs2_dio_end_io */
-		ocfs2_iocb_set_rw_locked(iocb);
+		ocfs2_iocb_set_rw_locked(iocb, rw_level);
 	}
 
 	/*
diff --git a/fs/ocfs2/file.h b/fs/ocfs2/file.h
index cc973f0..2c4460f 100644
--- a/fs/ocfs2/file.h
+++ b/fs/ocfs2/file.h
@@ -39,12 +39,17 @@ enum ocfs2_alloc_restarted {
 };
 int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
 			       struct inode *inode,
+			       u32 *cluster_start,
 			       u32 clusters_to_add,
 			       struct buffer_head *fe_bh,
 			       handle_t *handle,
 			       struct ocfs2_alloc_context *data_ac,
 			       struct ocfs2_alloc_context *meta_ac,
 			       enum ocfs2_alloc_restarted *reason);
+int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
+			  u32 clusters_to_add,
+			  struct ocfs2_alloc_context **data_ac,
+			  struct ocfs2_alloc_context **meta_ac);
 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr);
 int ocfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
 		  struct kstat *stat);
diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c
index 28ab56f..21a6050 100644
--- a/fs/ocfs2/inode.c
+++ b/fs/ocfs2/inode.c
@@ -89,24 +89,6 @@ void ocfs2_set_inode_flags(struct inode *inode)
 		inode->i_flags |= S_DIRSYNC;
 }
 
-struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb,
-				     u64 blkno,
-				     int delete_vote)
-{
-	struct ocfs2_find_inode_args args;
-
-	/* ocfs2_ilookup_for_vote should *only* be called from the
-	 * vote thread */
-	BUG_ON(current != osb->vote_task);
-
-	args.fi_blkno = blkno;
-	args.fi_flags = OCFS2_FI_FLAG_NOWAIT;
-	if (delete_vote)
-		args.fi_flags |= OCFS2_FI_FLAG_DELETE;
-	args.fi_ino = ino_from_blkno(osb->sb, blkno);
-	return ilookup5(osb->sb, args.fi_ino, ocfs2_find_actor, &args);
-}
-
 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
 {
 	struct inode *inode = NULL;
@@ -182,28 +164,6 @@ static int ocfs2_find_actor(struct inode *inode, void *opaque)
 	if (oi->ip_blkno != args->fi_blkno)
 		goto bail;
 
-	/* OCFS2_FI_FLAG_NOWAIT is *only* set from
-	 * ocfs2_ilookup_for_vote which won't create an inode for one
-	 * that isn't found. The vote thread which doesn't want to get
-	 * an inode which is in the process of going away - otherwise
-	 * the call to __wait_on_freeing_inode in find_inode_fast will
-	 * cause it to deadlock on an inode which may be waiting on a
-	 * vote (or lock release) in delete_inode */
-	if ((args->fi_flags & OCFS2_FI_FLAG_NOWAIT) &&
-	    (inode->i_state & (I_FREEING|I_CLEAR))) {
-		/* As stated above, we're not going to return an
-		 * inode.  In the case of a delete vote, the voting
-		 * code is going to signal the other node to go
-		 * ahead. Mark that state here, so this freeing inode
-		 * has the state when it gets to delete_inode. */
-		if (args->fi_flags & OCFS2_FI_FLAG_DELETE) {
-			spin_lock(&oi->ip_lock);
-			ocfs2_mark_inode_remotely_deleted(inode);
-			spin_unlock(&oi->ip_lock);
-		}
-		goto bail;
-	}
-
 	ret = 1;
 bail:
 	mlog_exit(ret);
@@ -261,6 +221,9 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 		goto bail;
 	}
 
+	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
+	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
+
 	inode->i_version = 1;
 	inode->i_generation = le32_to_cpu(fe->i_generation);
 	inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
@@ -272,8 +235,7 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 	if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
 		inode->i_blocks = 0;
 	else
-		inode->i_blocks =
-			ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size));
+		inode->i_blocks = ocfs2_inode_sector_count(inode);
 	inode->i_mapping->a_ops = &ocfs2_aops;
 	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
 	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
@@ -288,10 +250,6 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
 		     (unsigned long long)fe->i_blkno);
 
-	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
-	OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT;
-	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
-
 	inode->i_nlink = le16_to_cpu(fe->i_links_count);
 
 	if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
@@ -347,6 +305,9 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 
 		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
 					  OCFS2_LOCK_TYPE_META, 0, inode);
+
+		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
+					  OCFS2_LOCK_TYPE_OPEN, 0, inode);
 	}
 
 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
@@ -421,7 +382,7 @@ static int ocfs2_read_locked_inode(struct inode *inode,
 	 * cluster lock before trusting anything anyway.
 	 */
 	can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
-		&& !(args->fi_flags & OCFS2_FI_FLAG_NOLOCK)
+		&& !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
 		&& !ocfs2_mount_local(osb);
 
 	/*
@@ -438,7 +399,17 @@ static int ocfs2_read_locked_inode(struct inode *inode,
 				  OCFS2_LOCK_TYPE_META,
 				  generation, inode);
 
+	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
+				  OCFS2_LOCK_TYPE_OPEN,
+				  0, inode);
+
 	if (can_lock) {
+		status = ocfs2_open_lock(inode);
+		if (status) {
+			make_bad_inode(inode);
+			mlog_errno(status);
+			return status;
+		}
 		status = ocfs2_meta_lock(inode, NULL, 0);
 		if (status) {
 			make_bad_inode(inode);
@@ -447,6 +418,14 @@ static int ocfs2_read_locked_inode(struct inode *inode,
 		}
 	}
 
+	if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
+		status = ocfs2_try_open_lock(inode, 0);
+		if (status) {
+			make_bad_inode(inode);	
+			return status;
+		}
+	}
+
 	status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
 				  can_lock ? inode : NULL);
 	if (status < 0) {
@@ -507,50 +486,56 @@ static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
 				     struct buffer_head *fe_bh)
 {
 	int status = 0;
-	handle_t *handle = NULL;
 	struct ocfs2_truncate_context *tc = NULL;
 	struct ocfs2_dinode *fe;
+	handle_t *handle = NULL;
 
 	mlog_entry_void();
 
 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
 
-	/* zero allocation, zero truncate :) */
-	if (!fe->i_clusters)
-		goto bail;
+	if (fe->i_clusters) {
+		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+		if (IS_ERR(handle)) {
+			status = PTR_ERR(handle);
+			mlog_errno(status);
+			goto out;
+		}
 
-	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
-	if (IS_ERR(handle)) {
-		status = PTR_ERR(handle);
-		handle = NULL;
-		mlog_errno(status);
-		goto bail;
-	}
+		status = ocfs2_journal_access(handle, inode, fe_bh,
+					      OCFS2_JOURNAL_ACCESS_WRITE);
+		if (status < 0) {
+			mlog_errno(status);
+			goto out;
+		}
 
-	status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL);
-	if (status < 0) {
-		mlog_errno(status);
-		goto bail;
-	}
+		i_size_write(inode, 0);
 
-	ocfs2_commit_trans(osb, handle);
-	handle = NULL;
+		status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
+		if (status < 0) {
+			mlog_errno(status);
+			goto out;
+		}
 
-	status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
-	if (status < 0) {
-		mlog_errno(status);
-		goto bail;
-	}
+		ocfs2_commit_trans(osb, handle);
+		handle = NULL;
 
-	status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
-	if (status < 0) {
-		mlog_errno(status);
-		goto bail;
+		status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
+		if (status < 0) {
+			mlog_errno(status);
+			goto out;
+		}
+
+		status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
+		if (status < 0) {
+			mlog_errno(status);
+			goto out;
+		}
 	}
-bail:
+
+out:
 	if (handle)
 		ocfs2_commit_trans(osb, handle);
-
 	mlog_exit(status);
 	return status;
 }
@@ -678,10 +663,10 @@ static int ocfs2_wipe_inode(struct inode *inode,
 	struct inode *orphan_dir_inode = NULL;
 	struct buffer_head *orphan_dir_bh = NULL;
 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_dinode *di;
 
-	/* We've already voted on this so it should be readonly - no
-	 * spinlock needed. */
-	orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot;
+	di = (struct ocfs2_dinode *) di_bh->b_data;
+	orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
 
 	status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
 	if (status)
@@ -839,11 +824,20 @@ static int ocfs2_query_inode_wipe(struct inode *inode,
 		goto bail;
 	}
 
-	status = ocfs2_request_delete_vote(inode);
-	/* -EBUSY means that other nodes are still using the
-	 * inode. We're done here though, so avoid doing anything on
-	 * disk and let them worry about deleting it. */
-	if (status == -EBUSY) {
+	/*
+	 * This is how ocfs2 determines whether an inode is still live
+	 * within the cluster. Every node takes a shared read lock on
+	 * the inode open lock in ocfs2_read_locked_inode(). When we
+	 * get to ->delete_inode(), each node tries to convert it's
+	 * lock to an exclusive. Trylocks are serialized by the inode
+	 * meta data lock. If the upconvert suceeds, we know the inode
+	 * is no longer live and can be deleted.
+	 *
+	 * Though we call this with the meta data lock held, the
+	 * trylock keeps us from ABBA deadlock.
+	 */
+	status = ocfs2_try_open_lock(inode, 1);
+	if (status == -EAGAIN) {
 		status = 0;
 		mlog(0, "Skipping delete of %llu because it is in use on"
 		     "other nodes\n", (unsigned long long)oi->ip_blkno);
@@ -854,21 +848,10 @@ static int ocfs2_query_inode_wipe(struct inode *inode,
 		goto bail;
 	}
 
-	spin_lock(&oi->ip_lock);
-	if (oi->ip_orphaned_slot == OCFS2_INVALID_SLOT) {
-		/* Nobody knew which slot this inode was orphaned
-		 * into. This may happen during node death and
-		 * recovery knows how to clean it up so we can safely
-		 * ignore this inode for now on. */
-		mlog(0, "Nobody knew where inode %llu was orphaned!\n",
-		     (unsigned long long)oi->ip_blkno);
-	} else {
-		*wipe = 1;
-
-		mlog(0, "Inode %llu is ok to wipe from orphan dir %d\n",
-		     (unsigned long long)oi->ip_blkno, oi->ip_orphaned_slot);
-	}
-	spin_unlock(&oi->ip_lock);
+	*wipe = 1;
+	mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
+	     (unsigned long long)oi->ip_blkno,
+	     le16_to_cpu(di->i_orphaned_slot));
 
 bail:
 	return status;
@@ -1001,11 +984,16 @@ void ocfs2_clear_inode(struct inode *inode)
 	mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
 			"Inode=%lu\n", inode->i_ino);
 
+	/* For remove delete_inode vote, we hold open lock before,
+	 * now it is time to unlock PR and EX open locks. */
+	ocfs2_open_unlock(inode);
+
 	/* Do these before all the other work so that we don't bounce
 	 * the vote thread while waiting to destroy the locks. */
 	ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
 	ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
 	ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
+	ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
 
 	/* We very well may get a clear_inode before all an inodes
 	 * metadata has hit disk. Of course, we can't drop any cluster
@@ -1020,8 +1008,7 @@ void ocfs2_clear_inode(struct inode *inode)
 			"Clear inode of %llu, inode has io markers\n",
 			(unsigned long long)oi->ip_blkno);
 
-	ocfs2_extent_map_drop(inode, 0);
-	ocfs2_extent_map_init(inode);
+	ocfs2_extent_map_trunc(inode, 0);
 
 	status = ocfs2_drop_inode_locks(inode);
 	if (status < 0)
@@ -1030,6 +1017,7 @@ void ocfs2_clear_inode(struct inode *inode)
 	ocfs2_lock_res_free(&oi->ip_rw_lockres);
 	ocfs2_lock_res_free(&oi->ip_meta_lockres);
 	ocfs2_lock_res_free(&oi->ip_data_lockres);
+	ocfs2_lock_res_free(&oi->ip_open_lockres);
 
 	ocfs2_metadata_cache_purge(inode);
 
@@ -1086,9 +1074,6 @@ void ocfs2_drop_inode(struct inode *inode)
 	mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
 	     (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
 
-	/* Testing ip_orphaned_slot here wouldn't work because we may
-	 * not have gotten a delete_inode vote from any other nodes
-	 * yet. */
 	if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
 		generic_delete_inode(inode);
 	else
@@ -1121,8 +1106,8 @@ struct buffer_head *ocfs2_bread(struct inode *inode,
 		return NULL;
 	}
 
-	tmperr = ocfs2_extent_map_get_blocks(inode, block, 1,
-					     &p_blkno, NULL);
+	tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
+					     NULL);
 	if (tmperr < 0) {
 		mlog_errno(tmperr);
 		goto fail;
@@ -1259,7 +1244,7 @@ void ocfs2_refresh_inode(struct inode *inode,
 	if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
 		inode->i_blocks = 0;
 	else
-		inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode));
+		inode->i_blocks = ocfs2_inode_sector_count(inode);
 	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
 	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
 	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
diff --git a/fs/ocfs2/inode.h b/fs/ocfs2/inode.h
index 1a7dd29..03ae075 100644
--- a/fs/ocfs2/inode.h
+++ b/fs/ocfs2/inode.h
@@ -26,6 +26,8 @@
 #ifndef OCFS2_INODE_H
 #define OCFS2_INODE_H
 
+#include "extent_map.h"
+
 /* OCFS2 Inode Private Data */
 struct ocfs2_inode_info
 {
@@ -34,6 +36,7 @@ struct ocfs2_inode_info
 	struct ocfs2_lock_res		ip_rw_lockres;
 	struct ocfs2_lock_res		ip_meta_lockres;
 	struct ocfs2_lock_res		ip_data_lockres;
+	struct ocfs2_lock_res		ip_open_lockres;
 
 	/* protects allocation changes on this inode. */
 	struct rw_semaphore		ip_alloc_sem;
@@ -42,9 +45,7 @@ struct ocfs2_inode_info
 	spinlock_t			ip_lock;
 	u32				ip_open_count;
 	u32				ip_clusters;
-	struct ocfs2_extent_map		ip_map;
 	struct list_head		ip_io_markers;
-	int				ip_orphaned_slot;
 
 	struct mutex			ip_io_mutex;
 
@@ -64,6 +65,8 @@ struct ocfs2_inode_info
 
 	struct ocfs2_caching_info	ip_metadata_cache;
 
+	struct ocfs2_extent_map		ip_extent_map;
+
 	struct inode			vfs_inode;
 };
 
@@ -117,14 +120,9 @@ void ocfs2_delete_inode(struct inode *inode);
 void ocfs2_drop_inode(struct inode *inode);
 
 /* Flags for ocfs2_iget() */
-#define OCFS2_FI_FLAG_NOWAIT	0x1
-#define OCFS2_FI_FLAG_DELETE	0x2
-#define OCFS2_FI_FLAG_SYSFILE	0x4
-#define OCFS2_FI_FLAG_NOLOCK	0x8
+#define OCFS2_FI_FLAG_SYSFILE		0x4
+#define OCFS2_FI_FLAG_ORPHAN_RECOVERY	0x8
 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 feoff, int flags);
-struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb,
-				     u64 blkno,
-				     int delete_vote);
 int ocfs2_inode_init_private(struct inode *inode);
 int ocfs2_inode_revalidate(struct dentry *dentry);
 int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
@@ -144,4 +142,11 @@ int ocfs2_aio_write(struct file *file, struct kiocb *req, struct iocb *iocb);
 
 void ocfs2_set_inode_flags(struct inode *inode);
 
+static inline blkcnt_t ocfs2_inode_sector_count(struct inode *inode)
+{
+	int c_to_s_bits = OCFS2_SB(inode->i_sb)->s_clustersize_bits - 9;
+
+	return (blkcnt_t)(OCFS2_I(inode)->ip_clusters << c_to_s_bits);
+}
+
 #endif /* OCFS2_INODE_H */
diff --git a/fs/ocfs2/journal.c b/fs/ocfs2/journal.c
index 825cb0a..5a8a90d 100644
--- a/fs/ocfs2/journal.c
+++ b/fs/ocfs2/journal.c
@@ -649,29 +649,20 @@ bail:
 static int ocfs2_force_read_journal(struct inode *inode)
 {
 	int status = 0;
-	int i, p_blocks;
-	u64 v_blkno, p_blkno;
-#define CONCURRENT_JOURNAL_FILL 32
+	int i;
+	u64 v_blkno, p_blkno, p_blocks, num_blocks;
+#define CONCURRENT_JOURNAL_FILL 32ULL
 	struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
 
 	mlog_entry_void();
 
-	BUG_ON(inode->i_blocks !=
-		     ocfs2_align_bytes_to_sectors(i_size_read(inode)));
-
 	memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
 
-	mlog(0, "Force reading %llu blocks\n",
-		(unsigned long long)(inode->i_blocks >>
-			(inode->i_sb->s_blocksize_bits - 9)));
-
+	num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size);
 	v_blkno = 0;
-	while (v_blkno <
-	       (inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9))) {
-
+	while (v_blkno < num_blocks) {
 		status = ocfs2_extent_map_get_blocks(inode, v_blkno,
-						     1, &p_blkno,
-						     &p_blocks);
+						     &p_blkno, &p_blocks, NULL);
 		if (status < 0) {
 			mlog_errno(status);
 			goto bail;
@@ -1306,7 +1297,7 @@ static int ocfs2_queue_orphans(struct ocfs2_super *osb,
 				continue;
 
 			iter = ocfs2_iget(osb, le64_to_cpu(de->inode),
-					  OCFS2_FI_FLAG_NOLOCK);
+					  OCFS2_FI_FLAG_ORPHAN_RECOVERY);
 			if (IS_ERR(iter))
 				continue;
 
@@ -1418,7 +1409,6 @@ static int ocfs2_recover_orphans(struct ocfs2_super *osb,
 		/* Set the proper information to get us going into
 		 * ocfs2_delete_inode. */
 		oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
-		oi->ip_orphaned_slot = slot;
 		spin_unlock(&oi->ip_lock);
 
 		iput(inode);
diff --git a/fs/ocfs2/journal.h b/fs/ocfs2/journal.h
index d026b4f..3db5de4 100644
--- a/fs/ocfs2/journal.h
+++ b/fs/ocfs2/journal.h
@@ -390,7 +390,7 @@ static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
 	/* We may be deleting metadata blocks, so metadata alloc dinode +
 	   one desc. block for each possible delete. */
 	if (tree_depth && next_free == 1 &&
-	    le32_to_cpu(last_el->l_recs[i].e_clusters) == clusters_to_del)
+	    ocfs2_rec_clusters(last_el, &last_el->l_recs[i]) == clusters_to_del)
 		credits += 1 + tree_depth;
 
 	/* update to the truncate log. */
diff --git a/fs/ocfs2/mmap.c b/fs/ocfs2/mmap.c
index 51b0204..af01158 100644
--- a/fs/ocfs2/mmap.c
+++ b/fs/ocfs2/mmap.c
@@ -85,8 +85,11 @@ int ocfs2_mmap(struct file *file, struct vm_area_struct *vma)
 	int ret = 0, lock_level = 0;
 	struct ocfs2_super *osb = OCFS2_SB(file->f_dentry->d_inode->i_sb);
 
-	/* We don't want to support shared writable mappings yet. */
-	if (!ocfs2_mount_local(osb) &&
+	/*
+	 * Only support shared writeable mmap for local mounts which
+	 * don't know about holes.
+	 */
+	if ((!ocfs2_mount_local(osb) || ocfs2_sparse_alloc(osb)) &&
 	    ((vma->vm_flags & VM_SHARED) || (vma->vm_flags & VM_MAYSHARE)) &&
 	    ((vma->vm_flags & VM_WRITE) || (vma->vm_flags & VM_MAYWRITE))) {
 		mlog(0, "disallow shared writable mmaps %lx\n", vma->vm_flags);
diff --git a/fs/ocfs2/namei.c b/fs/ocfs2/namei.c
index 28dd757..2bcf353 100644
--- a/fs/ocfs2/namei.c
+++ b/fs/ocfs2/namei.c
@@ -175,8 +175,6 @@ static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 
 	inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
 	if (IS_ERR(inode)) {
-		mlog(ML_ERROR, "Unable to create inode %llu\n",
-		     (unsigned long long)blkno);
 		ret = ERR_PTR(-EACCES);
 		goto bail_unlock;
 	}
@@ -189,7 +187,6 @@ static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 	 * unlink. */
 	spin_lock(&oi->ip_lock);
 	oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
-	oi->ip_orphaned_slot = OCFS2_INVALID_SLOT;
 	spin_unlock(&oi->ip_lock);
 
 bail_add:
@@ -288,7 +285,7 @@ static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 
 	i_size_write(inode, inode->i_sb->s_blocksize);
 	inode->i_nlink = 2;
-	inode->i_blocks = ocfs2_align_bytes_to_sectors(inode->i_sb->s_blocksize);
+	inode->i_blocks = ocfs2_inode_sector_count(inode);
 	status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 	if (status < 0) {
 		mlog_errno(status);
@@ -1486,8 +1483,7 @@ static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
 	struct buffer_head **bhs = NULL;
 	const char *c;
 	struct super_block *sb = osb->sb;
-	u64 p_blkno;
-	int p_blocks;
+	u64 p_blkno, p_blocks;
 	int virtual, blocks, status, i, bytes_left;
 
 	bytes_left = i_size_read(inode) + 1;
@@ -1514,8 +1510,8 @@ static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
 		goto bail;
 	}
 
-	status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
-					     &p_blocks);
+	status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
+					     NULL);
 	if (status < 0) {
 		mlog_errno(status);
 		goto bail;
@@ -1674,8 +1670,11 @@ static int ocfs2_symlink(struct inode *dir,
 	inode->i_rdev = 0;
 	newsize = l - 1;
 	if (l > ocfs2_fast_symlink_chars(sb)) {
+		u32 offset = 0;
+
 		inode->i_op = &ocfs2_symlink_inode_operations;
-		status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
+		status = ocfs2_do_extend_allocation(osb, inode, &offset, 1,
+						    new_fe_bh,
 						    handle, data_ac, NULL,
 						    NULL);
 		if (status < 0) {
@@ -1689,7 +1688,7 @@ static int ocfs2_symlink(struct inode *dir,
 			goto bail;
 		}
 		i_size_write(inode, newsize);
-		inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
+		inode->i_blocks = ocfs2_inode_sector_count(inode);
 	} else {
 		inode->i_op = &ocfs2_fast_symlink_inode_operations;
 		memcpy((char *) fe->id2.i_symlink, symname, l);
@@ -2222,9 +2221,7 @@ static int ocfs2_orphan_add(struct ocfs2_super *osb,
 	/* Record which orphan dir our inode now resides
 	 * in. delete_inode will use this to determine which orphan
 	 * dir to lock. */
-	spin_lock(&OCFS2_I(inode)->ip_lock);
-	OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
-	spin_unlock(&OCFS2_I(inode)->ip_lock);
+	fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
 
 	mlog(0, "Inode %llu orphaned in slot %d\n",
 	     (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
diff --git a/fs/ocfs2/ocfs2.h b/fs/ocfs2/ocfs2.h
index db8e77c..82cc92d 100644
--- a/fs/ocfs2/ocfs2.h
+++ b/fs/ocfs2/ocfs2.h
@@ -46,11 +46,6 @@
 #include "endian.h"
 #include "ocfs2_lockid.h"
 
-struct ocfs2_extent_map {
-	u32		em_clusters;
-	struct rb_root	em_extents;
-};
-
 /* Most user visible OCFS2 inodes will have very few pieces of
  * metadata, but larger files (including bitmaps, etc) must be taken
  * into account when designing an access scheme. We allow a small
@@ -303,6 +298,13 @@ static inline int ocfs2_should_order_data(struct inode *inode)
 	return 1;
 }
 
+static inline int ocfs2_sparse_alloc(struct ocfs2_super *osb)
+{
+	if (osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_SPARSE_ALLOC)
+		return 1;
+	return 0;
+}
+
 /* set / clear functions because cluster events can make these happen
  * in parallel so we want the transitions to be atomic. this also
  * means that any future flags osb_flags must be protected by spinlock
@@ -461,6 +463,49 @@ static inline unsigned long ocfs2_align_bytes_to_sectors(u64 bytes)
 	return (unsigned long)((bytes + 511) >> 9);
 }
 
+static inline unsigned int ocfs2_page_index_to_clusters(struct super_block *sb,
+							unsigned long pg_index)
+{
+	u32 clusters = pg_index;
+	unsigned int cbits = OCFS2_SB(sb)->s_clustersize_bits;
+
+	if (unlikely(PAGE_CACHE_SHIFT > cbits))
+		clusters = pg_index << (PAGE_CACHE_SHIFT - cbits);
+	else if (PAGE_CACHE_SHIFT < cbits)
+		clusters = pg_index >> (cbits - PAGE_CACHE_SHIFT);
+
+	return clusters;
+}
+
+/*
+ * Find the 1st page index which covers the given clusters.
+ */
+static inline unsigned long ocfs2_align_clusters_to_page_index(struct super_block *sb,
+							u32 clusters)
+{
+	unsigned int cbits = OCFS2_SB(sb)->s_clustersize_bits;
+	unsigned long index = clusters;
+
+	if (PAGE_CACHE_SHIFT > cbits) {
+		index = clusters >> (PAGE_CACHE_SHIFT - cbits);
+	} else if (PAGE_CACHE_SHIFT < cbits) {
+		index = clusters << (cbits - PAGE_CACHE_SHIFT);
+	}
+
+	return index;
+}
+
+static inline unsigned int ocfs2_pages_per_cluster(struct super_block *sb)
+{
+	unsigned int cbits = OCFS2_SB(sb)->s_clustersize_bits;
+	unsigned int pages_per_cluster = 1;
+
+	if (PAGE_CACHE_SHIFT < cbits)
+		pages_per_cluster = 1 << (cbits - PAGE_CACHE_SHIFT);
+
+	return pages_per_cluster;
+}
+
 #define ocfs2_set_bit ext2_set_bit
 #define ocfs2_clear_bit ext2_clear_bit
 #define ocfs2_test_bit ext2_test_bit
diff --git a/fs/ocfs2/ocfs2_fs.h b/fs/ocfs2/ocfs2_fs.h
index e61e218..7130647 100644
--- a/fs/ocfs2/ocfs2_fs.h
+++ b/fs/ocfs2/ocfs2_fs.h
@@ -86,7 +86,8 @@
 	OCFS2_SB(sb)->s_feature_incompat &= ~(mask)
 
 #define OCFS2_FEATURE_COMPAT_SUPP	OCFS2_FEATURE_COMPAT_BACKUP_SB
-#define OCFS2_FEATURE_INCOMPAT_SUPP	OCFS2_FEATURE_INCOMPAT_LOCAL_MOUNT
+#define OCFS2_FEATURE_INCOMPAT_SUPP	(OCFS2_FEATURE_INCOMPAT_LOCAL_MOUNT \
+					 | OCFS2_FEATURE_INCOMPAT_SPARSE_ALLOC)
 #define OCFS2_FEATURE_RO_COMPAT_SUPP	0
 
 /*
@@ -155,6 +156,12 @@
 #define OCFS2_FL_MODIFIABLE	(0x000100FF)	/* User modifiable flags */
 
 /*
+ * Extent record flags (e_node.leaf.flags)
+ */
+#define OCFS2_EXT_UNWRITTEN	(0x01)	/* Extent is allocated but
+					 * unwritten */
+
+/*
  * ioctl commands
  */
 #define OCFS2_IOC_GETFLAGS	_IOR('f', 1, long)
@@ -282,10 +289,21 @@ static unsigned char ocfs2_type_by_mode[S_IFMT >> S_SHIFT] = {
 /*
  * On disk extent record for OCFS2
  * It describes a range of clusters on disk.
+ *
+ * Length fields are divided into interior and leaf node versions.
+ * This leaves room for a flags field (OCFS2_EXT_*) in the leaf nodes.
  */
 struct ocfs2_extent_rec {
 /*00*/	__le32 e_cpos;		/* Offset into the file, in clusters */
-	__le32 e_clusters;	/* Clusters covered by this extent */
+	union {
+		__le32 e_int_clusters; /* Clusters covered by all children */
+		struct {
+			__le16 e_leaf_clusters; /* Clusters covered by this
+						   extent */
+			__u8 e_reserved1;
+			__u8 e_flags; /* Extent flags */
+		};
+	};
 	__le64 e_blkno;		/* Physical disk offset, in blocks */
 /*10*/
 };
@@ -311,7 +329,10 @@ struct ocfs2_extent_list {
 /*00*/	__le16 l_tree_depth;		/* Extent tree depth from this
 					   point.  0 means data extents
 					   hang directly off this
-					   header (a leaf) */
+					   header (a leaf)
+					   NOTE: The high 8 bits cannot be
+					   used - tree_depth is never that big.
+					*/
 	__le16 l_count;			/* Number of extent records */
 	__le16 l_next_free_rec;		/* Next unused extent slot */
 	__le16 l_reserved1;
@@ -446,7 +467,9 @@ struct ocfs2_dinode {
 	__le32 i_ctime_nsec;
 	__le32 i_mtime_nsec;
 	__le32 i_attr;
-	__le32 i_reserved1;
+	__le16 i_orphaned_slot;		/* Only valid when OCFS2_ORPHANED_FL
+					   was set in i_flags */
+	__le16 i_reserved1;
 /*70*/	__le64 i_reserved2[8];
 /*B8*/	union {
 		__le64 i_pad1;		/* Generic way to refer to this
diff --git a/fs/ocfs2/ocfs2_lockid.h b/fs/ocfs2/ocfs2_lockid.h
index 4d5d565..4ca02b1 100644
--- a/fs/ocfs2/ocfs2_lockid.h
+++ b/fs/ocfs2/ocfs2_lockid.h
@@ -44,6 +44,7 @@ enum ocfs2_lock_type {
 	OCFS2_LOCK_TYPE_RENAME,
 	OCFS2_LOCK_TYPE_RW,
 	OCFS2_LOCK_TYPE_DENTRY,
+	OCFS2_LOCK_TYPE_OPEN,
 	OCFS2_NUM_LOCK_TYPES
 };
 
@@ -69,6 +70,9 @@ static inline char ocfs2_lock_type_char(enum ocfs2_lock_type type)
 		case OCFS2_LOCK_TYPE_DENTRY:
 			c = 'N';
 			break;
+		case OCFS2_LOCK_TYPE_OPEN:
+			c = 'O';
+			break;
 		default:
 			c = '\0';
 	}
@@ -85,6 +89,7 @@ static char *ocfs2_lock_type_strings[] = {
 	 * important job it does, anyway. */
 	[OCFS2_LOCK_TYPE_RW] = "Write/Read",
 	[OCFS2_LOCK_TYPE_DENTRY] = "Dentry",
+	[OCFS2_LOCK_TYPE_OPEN] = "Open",
 };
 
 static inline const char *ocfs2_lock_type_string(enum ocfs2_lock_type type)
diff --git a/fs/ocfs2/slot_map.c b/fs/ocfs2/slot_map.c
index 2d3ac32..d921a28 100644
--- a/fs/ocfs2/slot_map.c
+++ b/fs/ocfs2/slot_map.c
@@ -197,7 +197,7 @@ int ocfs2_init_slot_info(struct ocfs2_super *osb)
 		goto bail;
 	}
 
-	status = ocfs2_extent_map_get_blocks(inode, 0ULL, 1, &blkno, NULL);
+	status = ocfs2_extent_map_get_blocks(inode, 0ULL, &blkno, NULL, NULL);
 	if (status < 0) {
 		mlog_errno(status);
 		goto bail;
diff --git a/fs/ocfs2/suballoc.c b/fs/ocfs2/suballoc.c
index 6dbb117..0da655a 100644
--- a/fs/ocfs2/suballoc.c
+++ b/fs/ocfs2/suballoc.c
@@ -381,8 +381,7 @@ static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
 					     le32_to_cpu(fe->i_clusters)));
 	spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
 	i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
-	alloc_inode->i_blocks =
-		ocfs2_align_bytes_to_sectors(i_size_read(alloc_inode));
+	alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
 
 	status = 0;
 bail:
diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
index 6534f92..5c9e824 100644
--- a/fs/ocfs2/super.c
+++ b/fs/ocfs2/super.c
@@ -806,9 +806,6 @@ static int __init ocfs2_init(void)
 
 	ocfs2_print_version();
 
-	if (init_ocfs2_extent_maps())
-		return -ENOMEM;
-
 	status = init_ocfs2_uptodate_cache();
 	if (status < 0) {
 		mlog_errno(status);
@@ -837,7 +834,6 @@ leave:
 	if (status < 0) {
 		ocfs2_free_mem_caches();
 		exit_ocfs2_uptodate_cache();
-		exit_ocfs2_extent_maps();
 	}
 
 	mlog_exit(status);
@@ -863,8 +859,6 @@ static void __exit ocfs2_exit(void)
 
 	unregister_filesystem(&ocfs2_fs_type);
 
-	exit_ocfs2_extent_maps();
-
 	exit_ocfs2_uptodate_cache();
 
 	mlog_exit_void();
@@ -963,6 +957,7 @@ static void ocfs2_inode_init_once(void *data,
 		ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
 		ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
 		ocfs2_lock_res_init_once(&oi->ip_data_lockres);
+		ocfs2_lock_res_init_once(&oi->ip_open_lockres);
 
 		ocfs2_metadata_cache_init(&oi->vfs_inode);
 
diff --git a/fs/ocfs2/vote.c b/fs/ocfs2/vote.c
index f30e63b..4f82a2f 100644
--- a/fs/ocfs2/vote.c
+++ b/fs/ocfs2/vote.c
@@ -63,17 +63,10 @@ struct ocfs2_msg_hdr
 	__be32 h_node_num;    /* node sending this particular message. */
 };
 
-/* OCFS2_MAX_FILENAME_LEN is 255 characters, but we want to align this
- * for the network. */
-#define OCFS2_VOTE_FILENAME_LEN 256
 struct ocfs2_vote_msg
 {
 	struct ocfs2_msg_hdr v_hdr;
-	union {
-		__be32 v_generic1;
-		__be32 v_orphaned_slot;	/* Used during delete votes */
-		__be32 v_nlink;		/* Used during unlink votes */
-	} md1;				/* Message type dependant 1 */
+	__be32 v_reserved1;
 };
 
 /* Responses are given these values to maintain backwards
@@ -86,7 +79,6 @@ struct ocfs2_response_msg
 {
 	struct ocfs2_msg_hdr r_hdr;
 	__be32 r_response;
-	__be32 r_orphaned_slot;
 };
 
 struct ocfs2_vote_work {
@@ -96,7 +88,6 @@ struct ocfs2_vote_work {
 
 enum ocfs2_vote_request {
 	OCFS2_VOTE_REQ_INVALID = 0,
-	OCFS2_VOTE_REQ_DELETE,
 	OCFS2_VOTE_REQ_MOUNT,
 	OCFS2_VOTE_REQ_UMOUNT,
 	OCFS2_VOTE_REQ_LAST
@@ -151,135 +142,23 @@ static void ocfs2_process_umount_request(struct ocfs2_super *osb,
 	ocfs2_node_map_set_bit(osb, &osb->umount_map, node_num);
 }
 
-void ocfs2_mark_inode_remotely_deleted(struct inode *inode)
-{
-	struct ocfs2_inode_info *oi = OCFS2_I(inode);
-
-	assert_spin_locked(&oi->ip_lock);
-	/* We set the SKIP_DELETE flag on the inode so we don't try to
-	 * delete it in delete_inode ourselves, thus avoiding
-	 * unecessary lock pinging. If the other node failed to wipe
-	 * the inode as a result of a crash, then recovery will pick
-	 * up the slack. */
-	oi->ip_flags |= OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE;
-}
-
-static int ocfs2_process_delete_request(struct inode *inode,
-					int *orphaned_slot)
-{
-	int response = OCFS2_RESPONSE_BUSY;
-
-	mlog(0, "DELETE vote on inode %lu, read lnk_cnt = %u, slot = %d\n",
-	     inode->i_ino, inode->i_nlink, *orphaned_slot);
-
-	spin_lock(&OCFS2_I(inode)->ip_lock);
-
-	/* Whatever our vote response is, we want to make sure that
-	 * the orphaned slot is recorded properly on this node *and*
-	 * on the requesting node. Technically, if the requesting node
-	 * did not know which slot the inode is orphaned in but we
-	 * respond with BUSY he doesn't actually need the orphaned
-	 * slot, but it doesn't hurt to do it here anyway. */
-	if ((*orphaned_slot) != OCFS2_INVALID_SLOT) {
-		mlog_bug_on_msg(OCFS2_I(inode)->ip_orphaned_slot !=
-				OCFS2_INVALID_SLOT &&
-				OCFS2_I(inode)->ip_orphaned_slot !=
-				(*orphaned_slot),
-				"Inode %llu: This node thinks it's "
-				"orphaned in slot %d, messaged it's in %d\n",
-				(unsigned long long)OCFS2_I(inode)->ip_blkno,
-				OCFS2_I(inode)->ip_orphaned_slot,
-				*orphaned_slot);
-
-		mlog(0, "Setting orphaned slot for inode %llu to %d\n",
-		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
-		     *orphaned_slot);
-
-		OCFS2_I(inode)->ip_orphaned_slot = *orphaned_slot;
-	} else {
-		mlog(0, "Sending back orphaned slot %d for inode %llu\n",
-		     OCFS2_I(inode)->ip_orphaned_slot,
-		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
-
-		*orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot;
-	}
-
-	/* vote no if the file is still open. */
-	if (OCFS2_I(inode)->ip_open_count) {
-		mlog(0, "open count = %u\n",
-		     OCFS2_I(inode)->ip_open_count);
-		spin_unlock(&OCFS2_I(inode)->ip_lock);
-		goto done;
-	}
-	spin_unlock(&OCFS2_I(inode)->ip_lock);
-
-	/* directories are a bit ugly... What if someone is sitting in
-	 * it? We want to make sure the inode is removed completely as
-	 * a result of the iput in process_vote. */
-	if (S_ISDIR(inode->i_mode) && (atomic_read(&inode->i_count) != 1)) {
-		mlog(0, "i_count = %u\n", atomic_read(&inode->i_count));
-		goto done;
-	}
-
-	if (filemap_fdatawrite(inode->i_mapping)) {
-		mlog(ML_ERROR, "Could not sync inode %llu for delete!\n",
-		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
-		goto done;
-	}
-	sync_mapping_buffers(inode->i_mapping);
-	truncate_inode_pages(inode->i_mapping, 0);
-	ocfs2_extent_map_trunc(inode, 0);
-
-	spin_lock(&OCFS2_I(inode)->ip_lock);
-	/* double check open count - someone might have raced this
-	 * thread into ocfs2_file_open while we were writing out
-	 * data. If we're to allow a wipe of this inode now, we *must*
-	 * hold the spinlock until we've marked it. */
-	if (OCFS2_I(inode)->ip_open_count) {
-		mlog(0, "Raced to wipe! open count = %u\n",
-		     OCFS2_I(inode)->ip_open_count);
-		spin_unlock(&OCFS2_I(inode)->ip_lock);
-		goto done;
-	}
-
-	/* Mark the inode as being wiped from disk. */
-	ocfs2_mark_inode_remotely_deleted(inode);
-	spin_unlock(&OCFS2_I(inode)->ip_lock);
-
-	/* Not sure this is necessary anymore. */
-	d_prune_aliases(inode);
-
-	/* If we get here, then we're voting 'yes', so commit the
-	 * delete on our side. */
-	response = OCFS2_RESPONSE_OK;
-done:
-	return response;
-}
-
 static void ocfs2_process_vote(struct ocfs2_super *osb,
 			       struct ocfs2_vote_msg *msg)
 {
 	int net_status, vote_response;
-	int orphaned_slot = 0;
-	unsigned int node_num, generation;
+	unsigned int node_num;
 	u64 blkno;
 	enum ocfs2_vote_request request;
-	struct inode *inode = NULL;
 	struct ocfs2_msg_hdr *hdr = &msg->v_hdr;
 	struct ocfs2_response_msg response;
 
 	/* decode the network mumbo jumbo into local variables. */
 	request = be32_to_cpu(hdr->h_request);
 	blkno = be64_to_cpu(hdr->h_blkno);
-	generation = be32_to_cpu(hdr->h_generation);
 	node_num = be32_to_cpu(hdr->h_node_num);
-	if (request == OCFS2_VOTE_REQ_DELETE)
-		orphaned_slot = be32_to_cpu(msg->md1.v_orphaned_slot);
 
-	mlog(0, "processing vote: request = %u, blkno = %llu, "
-	     "generation = %u, node_num = %u, priv1 = %u\n", request,
-	     (unsigned long long)blkno, generation, node_num,
-	     be32_to_cpu(msg->md1.v_generic1));
+	mlog(0, "processing vote: request = %u, blkno = %llu, node_num = %u\n",
+	     request, (unsigned long long)blkno, node_num);
 
 	if (!ocfs2_is_valid_vote_request(request)) {
 		mlog(ML_ERROR, "Invalid vote request %d from node %u\n",
@@ -302,52 +181,6 @@ static void ocfs2_process_vote(struct ocfs2_super *osb,
 		break;
 	}
 
-	/* We cannot process the remaining message types before we're
-	 * fully mounted. It's perfectly safe however to send a 'yes'
-	 * response as we can't possibly have any of the state they're
-	 * asking us to modify yet. */
-	if (atomic_read(&osb->vol_state) == VOLUME_INIT)
-		goto respond;
-
-	/* If we get here, then the request is against an inode. */
-	inode = ocfs2_ilookup_for_vote(osb, blkno,
-				       request == OCFS2_VOTE_REQ_DELETE);
-
-	/* Not finding the inode is perfectly valid - it means we're
-	 * not interested in what the other node is about to do to it
-	 * so in those cases we automatically respond with an
-	 * affirmative. Cluster locking ensures that we won't race
-	 * interest in the inode with this vote request. */
-	if (!inode)
-		goto respond;
-
-	/* Check generation values. It's possible for us to get a
-	 * request against a stale inode. If so then we proceed as if
-	 * we had not found an inode in the first place. */
-	if (inode->i_generation != generation) {
-		mlog(0, "generation passed %u != inode generation = %u, "
-		     "ip_flags = %x, ip_blkno = %llu, msg %llu, i_count = %u, "
-		     "message type = %u\n", generation, inode->i_generation,
-		     OCFS2_I(inode)->ip_flags,
-		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
-		     (unsigned long long)blkno, atomic_read(&inode->i_count),
-		     request);
-		iput(inode);
-		inode = NULL;
-		goto respond;
-	}
-
-	switch (request) {
-	case OCFS2_VOTE_REQ_DELETE:
-		vote_response = ocfs2_process_delete_request(inode,
-							     &orphaned_slot);
-		break;
-	default:
-		mlog(ML_ERROR, "node %u, invalid request: %u\n",
-		     node_num, request);
-		vote_response = OCFS2_RESPONSE_BAD_MSG;
-	}
-
 respond:
 	/* Response struture is small so we just put it on the stack
 	 * and stuff it inline. */
@@ -357,7 +190,6 @@ respond:
 	response.r_hdr.h_generation = hdr->h_generation;
 	response.r_hdr.h_node_num = cpu_to_be32(osb->node_num);
 	response.r_response = cpu_to_be32(vote_response);
-	response.r_orphaned_slot = cpu_to_be32(orphaned_slot);
 
 	net_status = o2net_send_message(OCFS2_MESSAGE_TYPE_RESPONSE,
 					osb->net_key,
@@ -373,9 +205,6 @@ respond:
 	    && net_status != -ENOTCONN)
 		mlog(ML_ERROR, "message to node %u fails with error %d!\n",
 		     node_num, net_status);
-
-	if (inode)
-		iput(inode);
 }
 
 static void ocfs2_vote_thread_do_work(struct ocfs2_super *osb)
@@ -634,8 +463,7 @@ bail:
 static struct ocfs2_vote_msg * ocfs2_new_vote_request(struct ocfs2_super *osb,
 						      u64 blkno,
 						      unsigned int generation,
-						      enum ocfs2_vote_request type,
-						      u32 priv)
+						      enum ocfs2_vote_request type)
 {
 	struct ocfs2_vote_msg *request;
 	struct ocfs2_msg_hdr *hdr;
@@ -651,8 +479,6 @@ static struct ocfs2_vote_msg * ocfs2_new_vote_request(struct ocfs2_super *osb,
 		hdr->h_request = cpu_to_be32(type);
 		hdr->h_blkno = cpu_to_be64(blkno);
 		hdr->h_generation = cpu_to_be32(generation);
-
-		request->md1.v_generic1 = cpu_to_be32(priv);
 	}
 
 	return request;
@@ -664,7 +490,7 @@ static int ocfs2_do_request_vote(struct ocfs2_super *osb,
 				 struct ocfs2_vote_msg *request,
 				 struct ocfs2_net_response_cb *callback)
 {
-	int status, response;
+	int status, response = -EBUSY;
 	unsigned int response_id;
 	struct ocfs2_msg_hdr *hdr;
 
@@ -686,109 +512,12 @@ bail:
 	return status;
 }
 
-static int ocfs2_request_vote(struct inode *inode,
-			      struct ocfs2_vote_msg *request,
-			      struct ocfs2_net_response_cb *callback)
-{
-	int status;
-	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
-
-	if (ocfs2_inode_is_new(inode))
-		return 0;
-
-	status = -EAGAIN;
-	while (status == -EAGAIN) {
-		if (!(osb->s_mount_opt & OCFS2_MOUNT_NOINTR) &&
-		    signal_pending(current))
-			return -ERESTARTSYS;
-
-		status = ocfs2_super_lock(osb, 0);
-		if (status < 0) {
-			mlog_errno(status);
-			break;
-		}
-
-		status = 0;
-		if (!ocfs2_node_map_is_only(osb, &osb->mounted_map,
-					   osb->node_num))
-			status = ocfs2_do_request_vote(osb, request, callback);
-
-		ocfs2_super_unlock(osb, 0);
-	}
-	return status;
-}
-
-static void ocfs2_delete_response_cb(void *priv,
-				     struct ocfs2_response_msg *resp)
-{
-	int orphaned_slot, node;
-	struct inode *inode = priv;
-
-	orphaned_slot = be32_to_cpu(resp->r_orphaned_slot);
-	node = be32_to_cpu(resp->r_hdr.h_node_num);
-	mlog(0, "node %d tells us that inode %llu is orphaned in slot %d\n",
-	     node, (unsigned long long)OCFS2_I(inode)->ip_blkno,
-	     orphaned_slot);
-
-	/* The other node may not actually know which slot the inode
-	 * is orphaned in. */
-	if (orphaned_slot == OCFS2_INVALID_SLOT)
-		return;
-
-	/* Ok, the responding node knows which slot this inode is
-	 * orphaned in. We verify that the information is correct and
-	 * then record this in the inode. ocfs2_delete_inode will use
-	 * this information to determine which lock to take. */
-	spin_lock(&OCFS2_I(inode)->ip_lock);
-	mlog_bug_on_msg(OCFS2_I(inode)->ip_orphaned_slot != orphaned_slot &&
-			OCFS2_I(inode)->ip_orphaned_slot
-			!= OCFS2_INVALID_SLOT, "Inode %llu: Node %d says it's "
-			"orphaned in slot %d, we think it's in %d\n",
-			(unsigned long long)OCFS2_I(inode)->ip_blkno,
-			be32_to_cpu(resp->r_hdr.h_node_num),
-			orphaned_slot, OCFS2_I(inode)->ip_orphaned_slot);
-
-	OCFS2_I(inode)->ip_orphaned_slot = orphaned_slot;
-	spin_unlock(&OCFS2_I(inode)->ip_lock);
-}
-
-int ocfs2_request_delete_vote(struct inode *inode)
-{
-	int orphaned_slot, status;
-	struct ocfs2_net_response_cb delete_cb;
-	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
-	struct ocfs2_vote_msg *request;
-
-	spin_lock(&OCFS2_I(inode)->ip_lock);
-	orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot;
-	spin_unlock(&OCFS2_I(inode)->ip_lock);
-
-	delete_cb.rc_cb = ocfs2_delete_response_cb;
-	delete_cb.rc_priv = inode;
-
-	mlog(0, "Inode %llu, we start thinking orphaned slot is %d\n",
-	     (unsigned long long)OCFS2_I(inode)->ip_blkno, orphaned_slot);
-
-	status = -ENOMEM;
-	request = ocfs2_new_vote_request(osb, OCFS2_I(inode)->ip_blkno,
-					 inode->i_generation,
-					 OCFS2_VOTE_REQ_DELETE, orphaned_slot);
-	if (request) {
-		status = ocfs2_request_vote(inode, request, &delete_cb);
-
-		kfree(request);
-	}
-
-	return status;
-}
-
 int ocfs2_request_mount_vote(struct ocfs2_super *osb)
 {
 	int status;
 	struct ocfs2_vote_msg *request = NULL;
 
-	request = ocfs2_new_vote_request(osb, 0ULL, 0,
-					 OCFS2_VOTE_REQ_MOUNT, 0);
+	request = ocfs2_new_vote_request(osb, 0ULL, 0, OCFS2_VOTE_REQ_MOUNT);
 	if (!request) {
 		status = -ENOMEM;
 		goto bail;
@@ -821,8 +550,7 @@ int ocfs2_request_umount_vote(struct ocfs2_super *osb)
 	int status;
 	struct ocfs2_vote_msg *request = NULL;
 
-	request = ocfs2_new_vote_request(osb, 0ULL, 0,
-					 OCFS2_VOTE_REQ_UMOUNT, 0);
+	request = ocfs2_new_vote_request(osb, 0ULL, 0, OCFS2_VOTE_REQ_UMOUNT);
 	if (!request) {
 		status = -ENOMEM;
 		goto bail;
@@ -969,7 +697,6 @@ static int ocfs2_handle_vote_message(struct o2net_msg *msg,
 	     be32_to_cpu(work->w_msg.v_hdr.h_generation));
 	mlog(0, "h_node_num = %u\n",
 	     be32_to_cpu(work->w_msg.v_hdr.h_node_num));
-	mlog(0, "v_generic1 = %u\n", be32_to_cpu(work->w_msg.md1.v_generic1));
 
 	spin_lock(&osb->vote_task_lock);
 	list_add_tail(&work->w_list, &osb->vote_list);
diff --git a/fs/ocfs2/vote.h b/fs/ocfs2/vote.h
index 53ebc1c6..9ea46f6 100644
--- a/fs/ocfs2/vote.h
+++ b/fs/ocfs2/vote.h
@@ -38,14 +38,11 @@ static inline void ocfs2_kick_vote_thread(struct ocfs2_super *osb)
 	wake_up(&osb->vote_event);
 }
 
-int ocfs2_request_delete_vote(struct inode *inode);
 int ocfs2_request_mount_vote(struct ocfs2_super *osb);
 int ocfs2_request_umount_vote(struct ocfs2_super *osb);
 int ocfs2_register_net_handlers(struct ocfs2_super *osb);
 void ocfs2_unregister_net_handlers(struct ocfs2_super *osb);
 
-void ocfs2_mark_inode_remotely_deleted(struct inode *inode);
-
 void ocfs2_remove_node_from_vote_queues(struct ocfs2_super *osb,
 					int node_num);
 #endif