Virtual values for taint and information flow analysis
Prakasam Kannan, Thomas H. Austin, Mark Stamp, Tim Disney, Cormac Flanagan · San José State University ScholarWorks (San Jose State University) · 2016
Security controls such as taint analysis and information flow analysis can be powerful tools to protect against many common attacks. However, incorporating these controls into a language such as JavaScript is challenging. Native implementations require the support of all JavaScript VMs. Code rewriting requires developers to reason about the entire abstract syntax of JavaScript. In this paper, we demonstrate how virtual values may be used to more easily integrate these security controls. Virtual values provide hooks to alter the behavior of primitive operations, allowing programmers to create the desired security controls in a more declarative fashion, facilitating more rapid prototyping. We demonstrate how virtual values may be encoded in JavaScript using a combination of JavaScript object proxies and the Sweet.js macro library, and use that implementation to build taint and information flow controls into JavaScript. Finally, we show some benchmark results to demonstrate the overhead of this approach.