Improving operating system security

Abstract

Existing operating system security models implicity measure security with an invalid measurement function, by assuming that denial of access necessarily increases security. Security should be measured against the level of security desired by the owners of the information. A system that is too restrictive can result in breaches of security, as well as a system that is not restrictive enough. This paper shows how this concept should be incorporated into existing security models by applying conventional security access limits to executing files. This new system adds the flexibility to control access according to users and applications. The changes are shown to yield better security than existing operating system security applications. The model has four major improvements over existing models. First, the model considers the implications of user preferences and their reaction to inflexible systems. Second, the paper identifies the lack of granularity as a significant difficulty with existing theories. Third, it explicitly describes a measurement function that allows discussion of the level of security provided by different models. Fourth, the paper shows how the model can stop a Trojan Horse, even in the presence of “covert channels”.