diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 30a5094fffa71b86e65d3465f52a1250738c6696..db996f0edf0b5b3feeb6c3ca47f16afb50f8cac4 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -731,6 +731,7 @@ u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
 	u64 found_owner;
 	u64 root_objectid = root->root_key.objectid;
 	u32 total_count = 0;
+	u32 extent_refs;
 	u32 cur_count;
 	u32 nritems;
 	int ret;
@@ -767,6 +768,7 @@ u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
 	}
 
 	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
+	extent_refs = btrfs_extent_refs(l, item);
 	while (1) {
 		l = path->nodes[0];
 		nritems = btrfs_header_nritems(l);
@@ -800,10 +802,28 @@ u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
 				total_count = 2;
 				goto out;
 			}
+			/*
+			 * nasty.  we don't count a reference held by
+			 * the running transaction.  This allows nodatacow
+			 * to avoid cow most of the time
+			 */
+			if (found_owner >= BTRFS_FIRST_FREE_OBJECTID &&
+			    btrfs_ref_generation(l, ref_item) ==
+			    root->fs_info->generation) {
+				extent_refs--;
+			}
 		}
 		total_count = 1;
 		path->slots[0]++;
 	}
+	/*
+	 * if there is more than one reference against a data extent,
+	 * we have to assume the other ref is another snapshot
+	 */
+	if (level == -1 && extent_refs > 1) {
+		total_count = 2;
+		goto out;
+	}
 	if (cur_count == 0) {
 		total_count = 0;
 		goto out;
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 1bf37d15b17411c77f2428dd2afa40692720f5d3..a492fd238c887a145af5a9634d33c7417e275861 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -207,9 +207,8 @@ static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 	found_type = btrfs_key_type(&found_key);
 	if (found_key.objectid != inode->i_ino ||
-	    found_type != BTRFS_EXTENT_DATA_KEY) {
+	    found_type != BTRFS_EXTENT_DATA_KEY)
 		goto not_found;
-	}
 
 	found_type = btrfs_file_extent_type(leaf, item);
 	extent_start = found_key.offset;
@@ -245,7 +244,6 @@ static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 		if (!block_group || block_group->ro)
 			goto not_found;
 
-
 		start = extent_end;
 	} else {
 		goto not_found;
@@ -260,8 +258,8 @@ static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 	goto again;
 
 not_found:
-	cow_file_range(inode, start, cow_end);
-	start = cow_end + 1;
+	cow_file_range(inode, start, end);
+	start = end + 1;
 	goto loop;
 }