diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index 9e138cd..643335a 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -627,7 +627,27 @@
 	/* start IO across the range first to instantiate any delalloc
 	 * extents
 	 */
-	filemap_write_and_wait_range(inode->i_mapping, start, orig_end);
+	filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
+
+	/*
+	 * So with compression we will find and lock a dirty page and clear the
+	 * first one as dirty, setup an async extent, and immediately return
+	 * with the entire range locked but with nobody actually marked with
+	 * writeback.  So we can't just filemap_write_and_wait_range() and
+	 * expect it to work since it will just kick off a thread to do the
+	 * actual work.  So we need to call filemap_fdatawrite_range _again_
+	 * since it will wait on the page lock, which won't be unlocked until
+	 * after the pages have been marked as writeback and so we're good to go
+	 * from there.  We have to do this otherwise we'll miss the ordered
+	 * extents and that results in badness.  Please Josef, do not think you
+	 * know better and pull this out at some point in the future, it is
+	 * right and you are wrong.
+	 */
+	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
+		     &BTRFS_I(inode)->runtime_flags))
+		filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
+
+	filemap_fdatawait_range(inode->i_mapping, start, orig_end);
 
 	end = orig_end;
 	found = 0;