Stride: Polymorphic sled detection through instruction sequence analysis

Despite considerable effort, buffer overflow attacks remain a major security threat today, especially when coupled with self-propagation mechanisms as in worms and viruses. This paper considers the problem of designing network-level mechanisms for detecting polymorphic instances of such attacks. The starting point for our work is the observation that many buffer overflow attacks require a "sled" component to transfer control of the system to the exploit code. While previous work has shown that it is possible to detect certain types of sleds, including obfuscated instances, this paper demonstrates that the proposed detection heuristics can be thwarted by more elaborate sled obfuscation techniques. To address this problem, we have designed a new sled detection heuristic, called STRIDE, that offers three main improvements over previous work: it detects several types of sleds that other techniques are blind to, has a lower rate of false positives, and is significantly more computationally efficient, and hence more suitable for use at the network-level.