JavaPairRDD<String, PortableDataStream> jrdd = javaSparkContext.binaryFiles("<path_to_file>");
Map<String, PortableDataStream> mp = jrdd.collectAsMap();
OutputStream os = new FileOutputStream(f);
mp.values().forEach(pd -> {
    try {
        os.write(pd.toArray());
    } catch (IOException e) {
        throw new RuntimeException(e);
    }
});
os.flush();

And then supplying file to jaxb unmarshaller. Not sure if there is a better way.