A Binary-Centric Approach to Vulnerability Analysis and Defense


Abstract

This thesis takes a binary-centric approach for defending against vulnerabilities. A binary-centric approach is attractive for two reasons. First, most users only have access to programs in binary form (e.g., executables), thus security defenses based upon binary analysis are likely widely applicable. Second, a binary itself is often the most faithful encoding of security-relevant details since it is what is actually executed on hardware. This faithfulness allows us to provide strong guarantees about what sort of overall security can be offered. In this thesis, we first show that the binary-centric approach allows us to automatically generate exploits based upon vendor-provided patches for vulnerabilities. An immediate consequence of this line of research is current vendor patching practices are insecure because they allow attackers to create new exploits before all vulnerable hosts can receive a patch. Next, we show how to protect against exploits by automatically generating vulnerability-based signatures, which are signatures based upon a vulnerability instead of exploits. Our binary-centric approach to signature generation provides the first comprehensive framework for vulnerability-based signature generation and evaluation. For example, we are the first to show how to generate signatures that are guaranteed to never have false positives in a variety of settings. A key component of this research is development of a binary infrastructure to demonstrate the utility of our techniques and the application to our problem domain. The binary analysis techniques and architecture are of independent interest, and are currently being used by many other research projects at various universities and labs. In this proposal, we detail our initial research in binary analysis and vulnerability defense, and then propose future research directions.

Full Version: PDF

Proposal Date: Dec 11, 3PM EST, Wean 4623.

Committee


Back to David's Home Page