On the Controller Synthesis for Finite-State Markov Decision Processes