Do this instead of filemap_fdatawrite() - this way we sync only the
range between i_size and the cluster boundary.
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
                                 u64 new_i_size)
 {
        int ret, numpages;
+       loff_t endbyte;
        struct page **pages = NULL;
        u64 phys;
 
         * wait on them - the truncate_inode_pages() call later will
         * do that for us.
         */
-       ret = filemap_fdatawrite(inode->i_mapping);
+       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);