Cyberdyne: Automatic bug-finding at scale

Peter Goodman

Creating a scalable, distributed bug-finding system that is more than just the sum of its parts is challenging. Finding bugs that occur deep within a program’s execution requires the application of multiple bug-finding approaches (e.g. fuzzing, symbolic execution, static analysis). This talk will describe the practical aspects of how to design and implement a bug-finding system that combines multiple bug-finding approaches, using Cyberdyne as a running example.

Cyberdyne is a distributed, automatic bug-finding system, originally developed to compete in the DARPA Cyber Grand Challenge (CGC). Cyberdyne finds and fixes bugs in program binaries, without human intervention. Cyberdyne combines off-the-shelf and custom bug-finding tools into a unified, scalable system.

The first half of this talk describes Cyberdyne’s exoskeleton: the service-oriented architecture (SOA) that coordinates Cyberdyne’s bug-finding tools, triages and patches bugs, and validates that patches maintain program functionality. The second half of this talk describes Cyberdyne’s “machine intelligence”: the individual bug-finding tools, and the mechanism by which they cooperate to find deep program bugs.