CrowdFlow: Efficient Information Flow Security

摘要

The widespread use of JavaScript (JS) as the dominant web programming language opens the door to attacks such as Cross Site Scripting that steal sensitive information from users. Information flow tracking successfully addresses current browser security shortcomings, but current implementations incur a significant runtime overhead cost that prevents adoption. We present a novel approach to information flow security that distributes the tracking workload across all page visitors by probabilistically switching between two JavaScript execution modes. Our framework reports attempts to steal information from a user's browser to a third party that maintains a blacklist of malicious URLs. Participating users can then benefit from receiving warnings about blacklisted URLs, similar to anti-phishing filters. Our measurements indicate that our approach is both efficient and effective. First, our technique is efficient because it reduces performance impact by an order of magnitude. Second, our system is effective, i.e., it detects 99.45% of all information flow violations on the Alexa Top 500 pages using a conservative 5 % sampling rate. Most sites need fewer samples in practice; and will therefore incur even less overhead.