High Coverage Detection of Input-Related Security Faults

Improperly bounded program inputs present a major class of program defects. In secure applications, these bugs can be exploited by malicious users, allowing them to overwrite buffers and execute harmful code. In this paper, we present a high coverage dynamic technique for detecting software faults caused by improperly bounded program inputs. Our approach is novel in that it retains the advantages of dynamic bug detection, scope and precision; while at the same time, relaxing the requirement that the user specify the input that exposes the bug. To implement our approach, inputs are shadowed by additional state that characterize the allowed bounds of input-derived variables. Program operations and decision points may alter the shadowed state associated with input variables. Potentially hazardous program sites, such as an array references and string functions , are checked against the entire range of values that the user might specify. The approach found several bugs including two high-risk security bugs in a recent version of OpenSSH.