Retrofitting Legacy Code for Security

Jan 19, 2011

Download: MP4 Video Size: 446.4MB  
Watch on YouTube

Abstract

Research in computer security has historically advocated Design for
Security, the principle that security must be proactively integrated
into the design of a system. While examples exist in the research
literature of systems that have been designed for security, there are
few examples of such systems deployed in the real world. Economic and
practical considerations force developers to abandon security and
focus instead on functionality and performance, which are more
tangible than security. As a result, large bodies of legacy code often
have inadequate security mechanisms. Security mechanisms are added to
legacy code on-demand using ad hoc and manual techniques, and the
resulting systems are often insecure.

This talk advocates the need for techniques to retrofit
systems with security mechanisms. In particular, it focuses on the
problem of retrofitting legacy code with mechanisms for authorization
policy enforcement. It introduces a new formalism, called
fingerprints, to represent security-sensitive operations. Fingerprints
are code templates that represent accesses to security-critical
resources, and denote key steps needed to perform operations on these
resources. This talk develops both fingerprint mining and
fingerprint matching algorithms.

Fingerprint mining algorithms discover fingerprints of
security-sensitive operations by analyzing source code. This
talk presents two novel algorithms that use dynamic program
analysis and static program analysis, respectively, to mine
fingerprints. The fingerprints so mined are used by the fingerprint
matching algorithm to statically locate security-sensitive
operations. Program transformation is then employed to statically
modify source code by adding authorization policy lookups at each
location that performs a security-sensitive operation.

These techniques have been applied to three real-world systems. These
case studies demonstrate that techniques based upon program analysis
and transformation offer a principled and automated alternative to the
ad hoc and manual techniques that are currently used to retrofit
legacy software with security mechanisms. Time permitting, we will
talk about other problems in the context of retrofitting legacy code
for security. I will also indicate where ideas from model-checking
have been used in this work.