View article

A widely used technique for securing computer systems is to execute programs inside protection domains that enforce established security policies. These containers, often referred to as sandboxes, come in a variety of forms. Although current sandboxing techniques have individual strengths, they also have limitations that reduce the scope of their applicability. In this paper, we give a detailed analysis of the options available to designers of sandboxing mechanisms. As we discuss the tradeoffs of various design choices, we present a sandboxing facility that combines the strengths of a wide variety of design alternatives. Our design provides a set of simple yet powerful primitives that serve as a flexible, general-purpose framework for confining untrusted programs. As we present our work, we compare and contrast it with the work of others and give preliminary results.