Binary Reduction

Modern software typically includes the union of all functionality needed by any user. This means individual users end up with more functionality—and consequently more bugs, inefficiency, and attack surface—than they need. A classic example are JavaScript engines in PDF readers; the majority of users do not need them, but are still impacted by the associated increased vulnerability to malicious scripts. This problem is amplified by modern development practices, where developers pull in external libraries with more functionality than required by any one application.

Our binary reduction tools automatically rewrite binary executables and libraries to strip out unused functionality, as well as enabling users to selectively remove features. This improves efficiency and reduces attack surface. We describe two of our binary software reduction techniques below.

Conservative CFG-based Reduction

This binary rewriting software transformation takes as input a COTS binary executable and a set of entry points, and rewrites the executable to remove all code not transitively reachable from the given set of entry points. We calculate reachability by following the control flow graph of the binary and identifying symbolic references. This is particularly useful for reducing library code; libraries often include much more functionality than what is needed for any particular application.

Aggressive Test-based Reduction

This binary rewriting software transformation takes as input a COTS binary executable and a test suite, and rewrites the executable to remove all code which is not required to continue to pass the test suite. The search for a minimal subset of the code required to retain test-suite functionality is performed quickly and deterministically using delta-debugging. This is a speculative transformation which can achieve very significant reduction in program size but will likely break untested program behavior.