Practical Keystroke Timing Attacks in Sandboxed JavaScript

摘要

Keystrokes trigger interrupts which can be detected through software side channels to reconstruct keystroke timings. Keystroke timing attacks use these side channels to infer typed words, passphrases, or create user fingerprints. While keystroke timing attacks are considered harmful, they typically require native code execution to exploit the side channels and, thus, may not be practical in many scenarios. In this paper, we present the first generic keystroke timing attack in sandboxed JavaScript, targeting arbitrary other tabs, processes and programs. This violates same-origin policy, HTTPS security model, and process isolation. Our attack is based on the interrupt-timing side channel which has previously only been exploited using native code. In contrast to previous attacks, we do not require the victim to run a malicious binary or interact with the malicious website. Instead, our attack runs in a background tab, possibly in a minimized browser window, displaying a malicious online advertisement. We show that we can observe the exact inter-keystroke timings for a user's PIN or password, infer URLs entered by the user, and distinguish different users time-sharing a computer. Our attack works on personal computers, laptops and smartphones, with different operating systems and browsers. As a solution against all known JavaScript timing attacks, we propose a fine-grained permission model.