Automatic software instrumentation for the detection of non-control-data attacks

Abstract

To detect intrusions resulting of an attack that corrupted data items used by a program to perform its computation, we propose an approach that automatically instruments programs to control a data-based behavior model during their execution. We build our model by discovering the sets of data the system calls depend on and which constraints these sets must verify at runtime. We have implemented our approach using a static analysis framework called Frama-C and we present the results of experimentations on a vulnerable version of OpenSSH. © 2009 Springer Berlin Heidelberg.