Interprocedural analysis for privileged code placement and tainted variable detection

Abstract

In Java 2 and Microsoft .NET Common Language Runtime (CLR), trusted code has often been programmed to perform access-restricted operations not explicitly requested by its untrusted clients. Since an untrusted client will be on the call stack when access control is enforced, an access-restricted operation will not succeed unless the client is authorized. To avoid this, a portion of the trusted code can be made "privileged." When access control is enforced, privileged code causes the stack traversal to stop at the trusted code frame, and the untrusted code stack frames will not be checked for authorization. For large programs, manually understanding which portions of code should be made privileged is a difficult task. Developers must understand which authorizations will implicitly be extended to client code and make sure that the values of the variables used by the privileged code are not "tainted" by client code. This paper presents an interprocedural analysis for Java bytecode to automatically identify which portions of trusted code should be made privileged, ensure that there are no tainted variables in privileged code, and detect "unnecessary" and "redundant" privileged code. We implemented the algorithm and present the results of our analyses on a set of large programs. While the analysis techniques are in the context of Java code, the basic concepts are also applicable to non-Java systems with a similar authorization model. © Springer-Verlag Berlin Heidelberg 2005.