Automatic synthesis of filters to discard buffer overflow attacks: A step towards realizing selfhealing systems

Buffer overflows have become the most common target for network-based attacks. They are also the primary propagation mechanism used by worms. Although many techniques (such as StackGuard) have been developed to protect servers from being compromised by buffer over-flow attacks, these techniques cause the server to crash. In the face of automated, repetitive attacks such as those due to worms, these protection mechanisms lead to re-peated restarts of the victim application, rendering its service unavailable. In contrast, we present a promis-ing new approach that learns the characteristics of inputs associated with attacks, and filters them out in the future. It can be implemented without changing the server code, or even having access to its source. Since attack-bearing inputs are dropped before they corrupt the victim pro-cess, there is no need to restart the victim; as a result, recovery from attacks can be very fast. We tested our ap-proach on 8 buffer overflow attacks reported in the past few years on securityfocus.com and were available with working exploit code, and found that it generated accurate filters for 7 out of these 8 attacks.