Btrfs: progress on file_write

Signed-off-by: Chris Mason <chris.mason@oracle.com>

5 files changed, 394 insertions, 49 deletions

diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index 6b76fd9..43d4779 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -1249,6 +1249,63 @@ static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
 	return ret;
 }
 
+int btrfs_truncate_item(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct btrfs_path *path,
+			u32 new_size)
+{
+	int ret = 0;
+	int slot;
+	int slot_orig;
+	struct btrfs_leaf *leaf;
+	struct buffer_head *leaf_buf;
+	u32 nritems;
+	unsigned int data_end;
+	unsigned int old_data_start;
+	unsigned int old_size;
+	unsigned int size_diff;
+	int i;
+
+	slot_orig = path->slots[0];
+	leaf_buf = path->nodes[0];
+	leaf = btrfs_buffer_leaf(leaf_buf);
+
+	nritems = btrfs_header_nritems(&leaf->header);
+	data_end = leaf_data_end(root, leaf);
+
+	slot = path->slots[0];
+	old_data_start = btrfs_item_offset(leaf->items + slot);
+	old_size = btrfs_item_size(leaf->items + slot);
+	BUG_ON(old_size <= new_size);
+	size_diff = old_size - new_size;
+
+	BUG_ON(slot < 0);
+	BUG_ON(slot >= nritems);
+
+	/*
+	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
+	 */
+	/* first correct the data pointers */
+	for (i = slot; i < nritems; i++) {
+		u32 ioff = btrfs_item_offset(leaf->items + i);
+		btrfs_set_item_offset(leaf->items + i,
+				      ioff + size_diff);
+	}
+	/* shift the data */
+printk("truncate item, new_size %u old_size %u, diff %u, bufp %p, dst, %p, num %u, old_data_start %u, data_end %u\n", new_size, old_size, size_diff, leaf, btrfs_leaf_data(leaf) + data_end + size_diff, old_data_start-data_end, old_data_start, data_end);
+	btrfs_memmove(root, leaf, btrfs_leaf_data(leaf) +
+		      data_end + size_diff, btrfs_leaf_data(leaf) +
+		      data_end, old_data_start + new_size - data_end);
+	btrfs_set_item_size(leaf->items + slot, new_size);
+	btrfs_mark_buffer_dirty(leaf_buf);
+
+	ret = 0;
+	if (btrfs_leaf_free_space(root, leaf) < 0)
+		BUG();
+	check_leaf(root, path, 0);
+	return ret;
+}
+
 int btrfs_extend_item(struct btrfs_trans_handle *trans, struct btrfs_root
 		      *root, struct btrfs_path *path, u32 data_size)
 {
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 796f19d..ca3ab16 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -999,7 +999,7 @@ static inline void btrfs_mark_buffer_dirty(struct buffer_head *bh)
 	((type *)(btrfs_leaf_data(leaf) + \
 	btrfs_item_offset((leaf)->items + (slot))))
 
-/* extent-item.c */
+/* extent-tree.c */
 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *root);
 struct buffer_head *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
@@ -1013,9 +1013,16 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 		      *root, u64 blocknr, u64 num_blocks, int pin);
 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, struct
 			       btrfs_root *root);
+int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				u64 blocknr, u64 num_blocks);
 /* ctree.c */
 int btrfs_extend_item(struct btrfs_trans_handle *trans, struct btrfs_root
 		      *root, struct btrfs_path *path, u32 data_size);
+int btrfs_truncate_item(struct btrfs_trans_handle *trans,
+			struct btrfs_root *root,
+			struct btrfs_path *path,
+			u32 new_size);
 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
 		      *root, struct btrfs_key *key, struct btrfs_path *p, int
 		      ins_len, int cow);
@@ -1073,11 +1080,10 @@ int btrfs_lookup_inode(struct btrfs_trans_handle *trans, struct btrfs_root
 		       struct btrfs_key *location, int mod);
 
 /* file-item.c */
-int btrfs_alloc_file_extent(struct btrfs_trans_handle *trans,
+int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
 			       struct btrfs_root *root,
-			       u64 objectid, u64 offset,
-			       u64 num_blocks, u64 hint_block,
-			       u64 *result);
+			       u64 objectid, u64 pos, u64 offset,
+			       u64 num_blocks);
 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
 			     struct btrfs_root *root,
 			     struct btrfs_path *path, u64 objectid,
@@ -1090,6 +1096,11 @@ int btrfs_csum_file_block(struct btrfs_trans_handle *trans,
 int btrfs_csum_verify_file_block(struct btrfs_root *root,
 				 u64 objectid, u64 offset,
 				 char *data, size_t len);
+struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path,
+					  u64 objectid, u64 offset,
+					  int cow);
 /* super.c */
 extern struct subsystem btrfs_subsys;
 
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index d560831..2cee9df 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -12,8 +12,9 @@ static int finish_current_insert(struct btrfs_trans_handle *trans, struct
 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
 			       btrfs_root *extent_root);
 
-static int inc_block_ref(struct btrfs_trans_handle *trans, struct btrfs_root
-			 *root, u64 blocknr, u64 num_blocks)
+int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
+				struct btrfs_root *root,
+				u64 blocknr, u64 num_blocks)
 {
 	struct btrfs_path *path;
 	int ret;
@@ -50,8 +51,9 @@ static int inc_block_ref(struct btrfs_trans_handle *trans, struct btrfs_root
 	return 0;
 }
 
-static int lookup_block_ref(struct btrfs_trans_handle *trans, struct btrfs_root
-			    *root, u64 blocknr, u64 num_blocks, u32 *refs)
+static int lookup_extent_ref(struct btrfs_trans_handle *trans,
+			     struct btrfs_root *root, u64 blocknr,
+			     u64 num_blocks, u32 *refs)
 {
 	struct btrfs_path *path;
 	int ret;
@@ -80,7 +82,7 @@ static int lookup_block_ref(struct btrfs_trans_handle *trans, struct btrfs_root
 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *root)
 {
-	return inc_block_ref(trans, root, bh_blocknr(root->node), 1);
+	return btrfs_inc_extent_ref(trans, root, bh_blocknr(root->node), 1);
 }
 
 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
@@ -107,13 +109,13 @@ int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 				continue;
 			fi = btrfs_item_ptr(buf_leaf, i,
 					    struct btrfs_file_extent_item);
-			ret = inc_block_ref(trans, root,
+			ret = btrfs_inc_extent_ref(trans, root,
 				    btrfs_file_extent_disk_blocknr(fi),
 				    btrfs_file_extent_disk_num_blocks(fi));
 			BUG_ON(ret);
 		} else {
 			blocknr = btrfs_node_blockptr(buf_node, i);
-			ret = inc_block_ref(trans, root, blocknr, 1);
+			ret = btrfs_inc_extent_ref(trans, root, blocknr, 1);
 			BUG_ON(ret);
 		}
 	}
@@ -563,7 +565,7 @@ static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
 
 	WARN_ON(*level < 0);
 	WARN_ON(*level >= BTRFS_MAX_LEVEL);
-	ret = lookup_block_ref(trans, root, bh_blocknr(path->nodes[*level]),
+	ret = lookup_extent_ref(trans, root, bh_blocknr(path->nodes[*level]),
 			       1, &refs);
 	BUG_ON(ret);
 	if (refs > 1)
@@ -587,7 +589,7 @@ static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
 		}
 		blocknr = btrfs_node_blockptr(btrfs_buffer_node(cur),
 					      path->slots[*level]);
-		ret = lookup_block_ref(trans, root, blocknr, 1, &refs);
+		ret = lookup_extent_ref(trans, root, blocknr, 1, &refs);
 		BUG_ON(ret);
 		if (refs != 1) {
 			path->slots[*level]++;
diff --git a/fs/btrfs/file-item.c b/fs/btrfs/file-item.c
index 93d42d6..f49968a 100644
--- a/fs/btrfs/file-item.c
+++ b/fs/btrfs/file-item.c
@@ -6,13 +6,11 @@
 #define MAX_CSUM_ITEMS(r) ((((BTRFS_LEAF_DATA_SIZE(r) - \
 				 sizeof(struct btrfs_item)) / \
 				sizeof(struct btrfs_csum_item)) - 1))
-int btrfs_alloc_file_extent(struct btrfs_trans_handle *trans,
+int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
 			       struct btrfs_root *root,
-			       u64 objectid, u64 offset,
-			       u64 num_blocks, u64 hint_block,
-			       u64 *result)
+			       u64 objectid, u64 pos,
+			       u64 offset, u64 num_blocks)
 {
-	struct btrfs_key ins;
 	int ret = 0;
 	struct btrfs_file_extent_item *item;
 	struct btrfs_key file_key;
@@ -21,11 +19,13 @@ int btrfs_alloc_file_extent(struct btrfs_trans_handle *trans,
 	path = btrfs_alloc_path();
 	BUG_ON(!path);
 	btrfs_init_path(path);
+	/*
 	ret = btrfs_alloc_extent(trans, root, num_blocks, hint_block,
 				 (u64)-1, &ins);
+				 */
 	BUG_ON(ret);
 	file_key.objectid = objectid;
-	file_key.offset = offset;
+	file_key.offset = pos;
 	file_key.flags = 0;
 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
 
@@ -34,21 +34,22 @@ int btrfs_alloc_file_extent(struct btrfs_trans_handle *trans,
 	BUG_ON(ret);
 	item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
 			      struct btrfs_file_extent_item);
-	btrfs_set_file_extent_disk_blocknr(item, ins.objectid);
-	btrfs_set_file_extent_disk_num_blocks(item, ins.offset);
+	btrfs_set_file_extent_disk_blocknr(item, offset);
+	btrfs_set_file_extent_disk_num_blocks(item, num_blocks);
 	btrfs_set_file_extent_offset(item, 0);
-	btrfs_set_file_extent_num_blocks(item, ins.offset);
+	btrfs_set_file_extent_num_blocks(item, num_blocks);
 	btrfs_set_file_extent_generation(item, trans->transid);
 	btrfs_mark_buffer_dirty(path->nodes[0]);
-	*result = ins.objectid;
 	btrfs_release_path(root, path);
 	btrfs_free_path(path);
 	return 0;
 }
 
-static struct btrfs_csum_item *__lookup_csum_item(struct btrfs_root *root,
-						  struct btrfs_path *path,
-						  u64 objectid, u64 offset)
+struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
+					  struct btrfs_root *root,
+					  struct btrfs_path *path,
+					  u64 objectid, u64 offset,
+					  int cow)
 {
 	int ret;
 	struct btrfs_key file_key;
@@ -61,19 +62,23 @@ static struct btrfs_csum_item *__lookup_csum_item(struct btrfs_root *root,
 	file_key.offset = offset;
 	file_key.flags = 0;
 	btrfs_set_key_type(&file_key, BTRFS_CSUM_ITEM_KEY);
-	ret = btrfs_search_slot(NULL, root, &file_key, path, 0, 0);
+printk("__lookup for %Lu\n", offset);
+	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
 	if (ret < 0)
 		goto fail;
 	leaf = btrfs_buffer_leaf(path->nodes[0]);
 	if (ret > 0) {
 		ret = 1;
-		if (path->slots[0] == 0)
+		if (path->slots[0] == 0) {
+printk("fail1\n");
 			goto fail;
+		}
 		path->slots[0]--;
 		btrfs_disk_key_to_cpu(&found_key,
 				      &leaf->items[path->slots[0]].key);
 		if (btrfs_key_type(&found_key) != BTRFS_CSUM_ITEM_KEY ||
 		    found_key.objectid != objectid) {
+printk("fail2 type %u %Lu %Lu\n", btrfs_key_type(&found_key), found_key.objectid, objectid);
 			goto fail;
 		}
 		csum_offset = (offset - found_key.offset) >>
@@ -81,6 +86,7 @@ static struct btrfs_csum_item *__lookup_csum_item(struct btrfs_root *root,
 		if (csum_offset >=
 		    btrfs_item_size(leaf->items + path->slots[0]) /
 		    sizeof(struct btrfs_csum_item)) {
+printk("fail3, csum offset %lu size %u\n", csum_offset, btrfs_item_size(leaf->items + path->slots[0]) / sizeof(struct btrfs_csum_item));
 			goto fail;
 		}
 	}
@@ -89,7 +95,7 @@ static struct btrfs_csum_item *__lookup_csum_item(struct btrfs_root *root,
 	return item;
 fail:
 	if (ret > 0)
-		ret = -EIO;
+		ret = -ENOENT;
 	return ERR_PTR(ret);
 }
 
@@ -105,7 +111,7 @@ int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
 	int cow = mod != 0;
 	struct btrfs_csum_item *csum_item;
 
-	csum_item = __lookup_csum_item(root, path, objectid, offset);
+	csum_item = btrfs_lookup_csum(trans, root, path, objectid, offset, 0);
 	if (IS_ERR(csum_item))
 		return PTR_ERR(csum_item);
 	file_key.objectid = objectid;
@@ -113,7 +119,9 @@ int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
 	file_key.flags = 0;
 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
 	btrfs_release_path(root, path);
+printk("lookup file extent searches for %Lu\n", file_key.offset);
 	ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
+printk("ret is %d\n", ret);
 	return ret;
 }
 
@@ -134,17 +142,23 @@ int btrfs_csum_file_block(struct btrfs_trans_handle *trans,
 	path = btrfs_alloc_path();
 	BUG_ON(!path);
 	btrfs_init_path(path);
+
+	item = btrfs_lookup_csum(trans, root, path, objectid, offset, 0);
+	if (!IS_ERR(item))
+		goto found;
+	btrfs_release_path(root, path);
 	file_key.objectid = objectid;
 	file_key.offset = offset;
 	file_key.flags = 0;
 	btrfs_set_key_type(&file_key, BTRFS_CSUM_ITEM_KEY);
+printk("searching for csum %Lu %Lu\n", objectid, offset);
 	ret = btrfs_search_slot(trans, root, &file_key, path,
 				sizeof(struct btrfs_csum_item), 1);
+printk("ret %d\n", ret);
 	if (ret < 0)
 		goto fail;
 	if (ret == 0) {
-		csum_offset = 0;
-		goto csum;
+		BUG();
 	}
 	if (path->slots[0] == 0) {
 		btrfs_release_path(root, path);
@@ -153,12 +167,15 @@ int btrfs_csum_file_block(struct btrfs_trans_handle *trans,
 	path->slots[0]--;
 	leaf = btrfs_buffer_leaf(path->nodes[0]);
 	btrfs_disk_key_to_cpu(&found_key, &leaf->items[path->slots[0]].key);
+printk("found key %Lu %Lu %u\n", found_key.objectid, found_key.offset, found_key.flags);
 	csum_offset = (offset - found_key.offset) >>
 			root->fs_info->sb->s_blocksize_bits;
+printk("csum_offset %Lu\n", csum_offset);
 	if (btrfs_key_type(&found_key) != BTRFS_CSUM_ITEM_KEY ||
 	    found_key.objectid != objectid ||
 	    csum_offset >= MAX_CSUM_ITEMS(root)) {
 		btrfs_release_path(root, path);
+printk("insert1\n");
 		goto insert;
 	}
 	if (csum_offset >= btrfs_item_size(leaf->items + path->slots[0]) /
@@ -166,11 +183,13 @@ int btrfs_csum_file_block(struct btrfs_trans_handle *trans,
 		ret = btrfs_extend_item(trans, root, path,
 					sizeof(struct btrfs_csum_item));
 		BUG_ON(ret);
+printk("item extended\n");
 		goto csum;
 	}
 
 insert:
 	csum_offset = 0;
+printk("inserting item %Lu %Lu %u\n", file_key.objectid, file_key.offset, file_key.flags);
 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 				      sizeof(struct btrfs_csum_item));
 	if (ret != 0 && ret != -EEXIST)
@@ -180,12 +199,14 @@ csum:
 			      struct btrfs_csum_item);
 	ret = 0;
 	item += csum_offset;
+found:
 	ret = btrfs_csum_data(root, data, len, item->csum);
 	btrfs_set_csum_extent_offset(item, extent_offset);
 	btrfs_mark_buffer_dirty(path->nodes[0]);
 fail:
 	btrfs_release_path(root, path);
 	btrfs_free_path(path);
+printk("return ret %d\n", ret);
 	return ret;
 }
 
@@ -208,7 +229,7 @@ int btrfs_csum_verify_file_block(struct btrfs_root *root,
 	btrfs_set_key_type(&file_key, BTRFS_CSUM_ITEM_KEY);
 	mutex_lock(&root->fs_info->fs_mutex);
 
-	item = __lookup_csum_item(root, path, objectid, offset);
+	item = btrfs_lookup_csum(NULL, root, path, objectid, offset, 0);
 	if (IS_ERR(item)) {
 		ret = PTR_ERR(item);
 		goto fail;
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index ec68999..6a56416 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -1089,7 +1089,6 @@ static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
 
 	extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
 	extent_start = extent_start >> inode->i_blkbits;
-	extent_start += btrfs_file_extent_offset(item);
 	extent_end = extent_start + btrfs_file_extent_num_blocks(item);
 	if (iblock >= extent_start && iblock < extent_end) {
 		err = 0;
@@ -1103,6 +1102,7 @@ allocate:
 		err = 0;
 		goto out;
 	}
+#if 0
 	ret = btrfs_alloc_file_extent(trans, root, objectid,
 				      iblock << inode->i_blkbits,
 				      1, extent_end, &blocknr);
@@ -1115,9 +1115,11 @@ allocate:
 	map_bh(result, inode->i_sb, blocknr);
 
 	btrfs_map_bh_to_logical(root, result, blocknr);
+#endif
 out:
 	btrfs_release_path(root, path);
 	btrfs_free_path(path);
+printk("mapping iblock %lu to %lu\n", iblock, result->b_blocknr);
 	if (trans)
 		btrfs_end_transaction(trans, root);
 	return err;
@@ -1273,8 +1275,244 @@ failed:
 	return err;
 }
 
-static int prepare_pages(struct btrfs_trans_handle *trans,
-			 struct btrfs_root *root,
+static int drop_csums(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  struct inode *inode,
+			  u64 start, u64 end)
+{
+	struct btrfs_path *path;
+	struct btrfs_leaf *leaf;
+	struct btrfs_key key;
+	int slot;
+	struct btrfs_csum_item *item;
+	char *old_block = NULL;
+	u64 cur = start;
+	u64 found_end;
+	u64 num_csums;
+	u64 item_size;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	while(cur < end) {
+		item = btrfs_lookup_csum(trans, root, path,
+					 inode->i_ino, cur, 1);
+		if (IS_ERR(item)) {
+			cur += root->blocksize;
+			continue;
+		}
+		leaf = btrfs_buffer_leaf(path->nodes[0]);
+		slot = path->slots[0];
+		btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
+		item_size = btrfs_item_size(leaf->items + slot);
+		num_csums = item_size / sizeof(struct btrfs_csum_item);
+		found_end = key.offset + (num_csums << inode->i_blkbits);
+		cur = found_end;
+
+		if (found_end > end) {
+			char *src;
+			old_block = kmalloc(root->blocksize, GFP_NOFS);
+			src = btrfs_item_ptr(leaf, slot, char);
+			memcpy(old_block, src, item_size);
+		}
+		if (key.offset < start) {
+			u64 new_size = (start - key.offset) >>
+					inode->i_blkbits;
+			new_size *= sizeof(struct btrfs_csum_item);
+			ret = btrfs_truncate_item(trans, root, path, new_size);
+			BUG_ON(ret);
+		} else {
+			btrfs_del_item(trans, root, path);
+		}
+		btrfs_release_path(root, path);
+		if (found_end > end) {
+			char *dst;
+			int i;
+			int new_size;
+
+			num_csums = (found_end - end) >> inode->i_blkbits;
+			new_size = num_csums * sizeof(struct btrfs_csum_item);
+			key.offset = end;
+			ret = btrfs_insert_empty_item(trans, root, path,
+						      &key, new_size);
+			BUG_ON(ret);
+			dst = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
+					     path->slots[0], char);
+			memcpy(dst, old_block + item_size - new_size,
+			       new_size);
+			item = (struct btrfs_csum_item *)dst;
+			for (i = 0; i < num_csums; i++) {
+				btrfs_set_csum_extent_offset(item, end);
+				item++;
+			}
+			mark_buffer_dirty(path->nodes[0]);
+			kfree(old_block);
+			break;
+		}
+	}
+	btrfs_free_path(path);
+	return 0;
+}
+
+static int drop_extents(struct btrfs_trans_handle *trans,
+			  struct btrfs_root *root,
+			  struct inode *inode,
+			  u64 start, u64 end)
+{
+	int ret;
+	struct btrfs_key key;
+	struct btrfs_leaf *leaf;
+	int slot;
+	struct btrfs_file_extent_item *extent;
+	u64 extent_end;
+	int keep;
+	struct btrfs_file_extent_item old;
+	struct btrfs_path *path;
+	u64 search_start = start;
+	int bookend;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+search_again:
+printk("drop extent inode %lu start %Lu end %Lu\n", inode->i_ino, start, end);
+	ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
+				       search_start, -1);
+	if (ret != 0) {
+printk("lookup failed\n");
+		goto out;
+	}
+	while(1) {
+		keep = 0;
+		bookend = 0;
+		leaf = btrfs_buffer_leaf(path->nodes[0]);
+		slot = path->slots[0];
+		btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
+
+printk("found key %Lu %Lu %u\n", key.objectid, key.offset, key.flags);
+
+		extent = btrfs_item_ptr(leaf, slot,
+					struct btrfs_file_extent_item);
+		extent_end = key.offset +
+			(btrfs_file_extent_num_blocks(extent) <<
+			 inode->i_blkbits);
+printk("extent end is %Lu\n", extent_end);
+		if (key.offset >= end || key.objectid != inode->i_ino) {
+			ret = 0;
+			goto out;
+		}
+		if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
+			goto next_leaf;
+
+		if (end < extent_end && end >= key.offset) {
+			memcpy(&old, extent, sizeof(old));
+			ret = btrfs_inc_extent_ref(trans, root,
+				   btrfs_file_extent_disk_blocknr(&old),
+				   btrfs_file_extent_disk_num_blocks(&old));
+			BUG_ON(ret);
+			bookend = 1;
+		}
+
+		if (start > key.offset) {
+			u64 new_num;
+			/* truncate existing extent */
+			keep = 1;
+			WARN_ON(start & (root->blocksize - 1));
+			new_num = (start - key.offset) >> inode->i_blkbits;
+printk("truncating existing extent, was %Lu ", btrfs_file_extent_num_blocks(extent));
+			btrfs_set_file_extent_num_blocks(extent, new_num);
+printk("now %Lu\n", btrfs_file_extent_num_blocks(extent));
+
+			mark_buffer_dirty(path->nodes[0]);
+		}
+		if (!keep) {
+			u64 disk_blocknr;
+			u64 disk_num_blocks;
+printk("del old\n");
+			disk_blocknr = btrfs_file_extent_disk_blocknr(extent);
+			disk_num_blocks =
+				btrfs_file_extent_disk_num_blocks(extent);
+			search_start = key.offset +
+				(btrfs_file_extent_num_blocks(extent) <<
+				inode->i_blkbits);
+			ret = btrfs_del_item(trans, root, path);
+			BUG_ON(ret);
+			btrfs_release_path(root, path);
+
+			ret = btrfs_free_extent(trans, root, disk_blocknr,
+						disk_num_blocks, 0);
+
+			BUG_ON(ret);
+			if (!bookend && search_start >= end) {
+				ret = 0;
+				goto out;
+			}
+			if (!bookend)
+				goto search_again;
+		}
+		if (bookend) {
+			/* create bookend */
+			struct btrfs_key ins;
+printk("bookend! extent end %Lu\n", extent_end);
+			ins.objectid = inode->i_ino;
+			ins.offset = end;
+			ins.flags = 0;
+			btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
+
+			btrfs_release_path(root, path);
+			ret = drop_csums(trans, root, inode, start, end);
+			BUG_ON(ret);
+			ret = btrfs_insert_empty_item(trans, root, path, &ins,
+						      sizeof(*extent));
+			BUG_ON(ret);
+			extent = btrfs_item_ptr(
+				    btrfs_buffer_leaf(path->nodes[0]),
+				    path->slots[0],
+				    struct btrfs_file_extent_item);
+			btrfs_set_file_extent_disk_blocknr(extent,
+				    btrfs_file_extent_disk_blocknr(&old));
+			btrfs_set_file_extent_disk_num_blocks(extent,
+				    btrfs_file_extent_disk_num_blocks(&old));
+
+			btrfs_set_file_extent_offset(extent,
+				    btrfs_file_extent_offset(&old) +
+				    ((end - key.offset) >> inode->i_blkbits));
+			WARN_ON(btrfs_file_extent_num_blocks(&old) <
+				(end - key.offset) >> inode->i_blkbits);
+			btrfs_set_file_extent_num_blocks(extent,
+				    btrfs_file_extent_num_blocks(&old) -
+				    ((end - key.offset) >> inode->i_blkbits));
+
+			btrfs_set_file_extent_generation(extent,
+				    btrfs_file_extent_generation(&old));
+printk("new bookend at offset %Lu, file_extent_offset %Lu, file_extent_num_blocks %Lu\n", end, btrfs_file_extent_offset(extent), btrfs_file_extent_num_blocks(extent));
+			btrfs_mark_buffer_dirty(path->nodes[0]);
+			ret = 0;
+			goto out_nocsum;
+		}
+next_leaf:
+		if (slot >= btrfs_header_nritems(&leaf->header) - 1) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret) {
+				ret = 0;
+				goto out;
+			}
+		} else {
+			path->slots[0]++;
+		}
+	}
+
+out:
+	ret = drop_csums(trans, root, inode, start, end);
+	BUG_ON(ret);
+
+out_nocsum:
+	btrfs_free_path(path);
+	return ret;
+}
+
+static int prepare_pages(struct btrfs_root *root,
 			 struct file *file,
 			 struct page **pages,
 			 size_t num_pages,
@@ -1289,7 +1527,6 @@ static int prepare_pages(struct btrfs_trans_handle *trans,
 	struct inode *inode = file->f_path.dentry->d_inode;
 	int offset;
 	int err = 0;
-	int ret;
 	int this_write;
 	struct buffer_head *bh;
 	struct buffer_head *head;
@@ -1305,18 +1542,21 @@ static int prepare_pages(struct btrfs_trans_handle *trans,
 		}
 		offset = pos & (PAGE_CACHE_SIZE -1);
 		this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
-		if (!PageUptodate(pages[i]) &&
-		   (pages[i]->index == first_index ||
-		    pages[i]->index == last_index) && pos < isize) {
+#if 0
+		if ((pages[i]->index == first_index ||
+		    pages[i]->index == last_index) && pos < isize &&
+		    !PageUptodate(pages[i])) {
 			ret = mpage_readpage(pages[i], btrfs_get_block);
 			BUG_ON(ret);
 			lock_page(pages[i]);
 		}
+#endif
 		create_empty_buffers(pages[i], root->fs_info->sb->s_blocksize,
 				     (1 << BH_Uptodate));
 		head = page_buffers(pages[i]);
 		bh = head;
 		do {
+printk("mapping page %lu to block %Lu\n", pages[i]->index, alloc_extent_start);
 			err = btrfs_map_bh_to_logical(root, bh,
 						      alloc_extent_start);
 			BUG_ON(err);
@@ -1351,7 +1591,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 	int ret = 0;
 	struct inode *inode = file->f_path.dentry->d_inode;
 	struct btrfs_root *root = BTRFS_I(inode)->root;
-	struct page *pages[1];
+	struct page *pages[8];
 	unsigned long first_index;
 	unsigned long last_index;
 	u64 start_pos;
@@ -1359,6 +1599,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 	u64 alloc_extent_start;
 	u64 orig_extent_start;
 	struct btrfs_trans_handle *trans;
+	struct btrfs_key ins;
 
 	if (file->f_flags & O_DIRECT)
 		return -EINVAL;
@@ -1390,16 +1631,24 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 	trans = btrfs_start_transaction(root, 1);
 	if (!trans) {
 		err = -ENOMEM;
+		mutex_unlock(&root->fs_info->fs_mutex);
 		goto out_unlock;
 	}
-	ret = btrfs_alloc_file_extent(trans, root, inode->i_ino,
-				      start_pos, num_blocks, 1,
-				      &alloc_extent_start);
-	BUG_ON(ret);
-
+	if (start_pos < inode->i_size) {
+		ret = drop_extents(trans, root, inode,
+				   start_pos,
+				   (pos + count + root->blocksize -1) &
+				   ~(root->blocksize - 1));
+	}
 	orig_extent_start = start_pos;
-	ret = btrfs_end_transaction(trans, root);
+	ret = btrfs_alloc_extent(trans, root, num_blocks, 1,
+				 (u64)-1, &ins);
+	BUG_ON(ret);
+	ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
+				       start_pos, ins.objectid, ins.offset);
 	BUG_ON(ret);
+	alloc_extent_start = ins.objectid;
+	ret = btrfs_end_transaction(trans, root);
 	mutex_unlock(&root->fs_info->fs_mutex);
 
 	while(count > 0) {
@@ -1407,16 +1656,21 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 		size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
 		size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 					PAGE_CACHE_SHIFT;
-		ret = prepare_pages(NULL, root, file, pages, num_pages,
+printk("num_pages is %lu\n", num_pages);
+
+		memset(pages, 0, sizeof(pages));
+		ret = prepare_pages(root, file, pages, num_pages,
 				    pos, first_index, last_index,
 				    write_bytes, alloc_extent_start);
 		BUG_ON(ret);
+
 		/* FIXME blocks != pagesize */
 		alloc_extent_start += num_pages;
 		ret = btrfs_copy_from_user(pos, num_pages,
 					   write_bytes, pages, buf);
 		BUG_ON(ret);
 
+printk("2num_pages is %lu\n", num_pages);
 		ret = dirty_and_release_pages(NULL, root, file, pages,
 					      num_pages, orig_extent_start,
 					      pos, write_bytes);