Submission #29: The Case for OS-Level Website Fingerprinting Protection ======================================================================= Abstract -------- We argue the need for generic OS-level protection against Website Fingerprinting Attacks. We believe the OS kernel is the right place to implement such a protection framework because of two reasons. First, the OS is where the major WF attack surfaces, such as packet size and timing, are ultimately created. In addition to packet scheduling described earlier, the TCP implementation makes a decision about how the application-level bytestream is segmented into the packets through the generic segmentation offload interface at the lower layer. Second, the OS kernel has visibility and control of the traffic sent or received by all the transport protocol implementations running in the host, whereas individual ones do not have that. This could create opportunities of making the host traffic more robust against WF attacks by making the obfuscation decision based on other co-existing traffic. Our early results show that, based on the k-FP model, the state-of-the-art WFA algorithm, just adding as little delay as 0.25 ms to the base RTT of 0.20 ms has a significant impact on the accuracy of the attack, indicating the feasibility of OS-level traffic manipulation to prevent WFAs. Authors ------- 1. Lisa Lavrentieva (University of Edinburgh) 2. Marc Juarez (University of Edinburgh) 3. Michio Honda (University of Edinburgh)