Given a set of forbidden states in a plant model, an optimal Petri net supervisor is developed. The optimality is twofold. First, it provides a minimal supervisory control structure in the sense of the number of monitors that are used to prevent the occurrences of the forbidden states in a plant model. A monitor is designed by associating a P-semiflow with other places in the plant. Second, the supervisor is maximally permissive, i.e., no admissible state is excluded. If a maximally permissive supervisor does not exist, a most permissive supervisor can be similarly designed. A partial reachability graph of a plant is generated by considering the given forbidden states, from which first-met bad markings are then identified. Integer linear programming problems are employed to offer monitor solutions to ensure that the first-met bad markings are not reachable. Generalized mutual exclusion constraints are considered in this work.