Recent studies in secure computer svstems have shown that lattice structured information flow policies have properties which lead to simple and efficient enforcement mechanisms. This paper outlines a method for transforming nonlattice structured policies into lattices while Dreserving the validity of all : flows.
Dynamic multiprogramming memory management strategies are classified and compared using extant test data. Conclusions about program behavior are then drawn.
The implementation of programming systems that cannot leak confidential information Is examined. Unless severe restrictions are placed on their form, programs of such systems cannot even be permitted to output apparently nonconfldential information unless they have been proved error-free.
The protection mechanisms of computer systems control the access to objects, especially information objects. The range of responsibilities of these mechanisms includes at one extreme completely isolating executing programs from each other, and at the other extreme permitting complete cooperation and shared access among executing programs. Within this range one can identify at least seven levels at which protection mechanisms can be conceived as being required, each level being more difficult than its predecessor to implement: 1. No sharing at all (complete isolation). 2. Sharing copies of programs or data files. 3. Sharing originals of programs or data files. 4. Sharing programming systems or subsystems. 5. Permitting the cooperation of mutually suspicious subsystems---e.g., as with debugging or proprietary subsystems. 6. Providing "memoryless" subsystems---i.e., systems which, having performed their tasks, are guaranteed to have kept no secret record of the task performed (an income-tax computing service, for example, must be allowed to keep billing information on its use by customers but not to store information secretly on customers' incomes). 7. Providing "certified" subsystems---i.e., those whose correctness has been completely validated and is guaranteed a priori .
