Verification State-Space Reduction through Restricted Parsing Environments

Abstract

We discuss the potential for significant reduction in size and complexity of verification tasks for input-handling software when such software is constructed according to Lang Sec principles, i.e., Is designed as a recognizer for a particular language of valid inputs and is compiled for a suitably limited computational model no stronger than needed for the recognition task. We introduce Crema, a programming language and restricted execution environment of sub-Turing power, and conduct a case study to estimate and compare the respective sizes of verification tasks for the qmail SMTP parsing code fragments when executed natively vs in Crema -- using LLVM and KLEE. We also study the application of the same principles to the verification of reference monitors.