Quantifying Persistent Browser Cache Poisoning

Web browsers rely on caching for improving perfor-mance and for reducing bandwidth use. Cache poisoning poses alarming security concerns in light of HTTP’s lack of an integrity guarantee in conjunction with the proper-ties of its caching behavior. In our previous study we demonstrated the simplicity of replacing objects in the browser cache with malicious code to enable a persistent attack. This paper expands on this topic with a quantitative analysis of the impact of this threat. Based on full-packet traces from two distinct environments – a large research lab in California and a network in rural northern India – we conduct an empiri-cal study showing that (i) an attacker can with high prob-ability achieve a long-lived attack vector when poison-ing a web object picked at random and, (ii) that the high degree of object sharing, especially of executable code, enables an attacker to achieve high-coverage attack vec-tors by only poisoning a small set of intensively shared objects. We believe that the increasing popularity of Web 2.0 mash-ups will increase the degree of sharing, making the discussed attack extremely wide in scope. In par-ticular, we note the conceptual security shortcomings and risks of JavaScript content distribution networks (CDNs), techniques used by cellular carriers for com-pressing content on-the-fly, advertising networks, and popular services to track users ’ surfing behavior.