Existing verification and validation methodologies can detect software violations very effectively but fail to provide any mechanism for correcting faults once they are detected. Detection of faults, their diagnosis and corrective actions are all essential components of any software rectification framework. In this paper, we propose a framework for correction of violations in software systems ensuring that the desired goals of the system are achieved. We describe a stochastic finite state machine used to abstract a software system along with the uncertainty in its operating environment. Safety property violations and satisfaction of functionalities are abstracted using penalties and rewards on the states, respectively. Rectification of software is then formulated as a stochastic optimal control problem over this abstraction. Algorithms polynomial in the size of the abstraction have been developed for solving this optimization problem exactly. The paper also applies the developed framework to a variety of examples from different domains.