Barring deliberate sabotage, we generally assume that computer hardware will work as described. This assumption is mistaken. Poor manufacturing, errant radiation, and heat can cause malfunction. Commonly, such malfunction manifests in DRAM chips as flipped bits. Security researchers have known about the danger of such bit flips but these attacks have not been very practical. Thanks to ever-higher DRAM densities and the use of computing devices outdoors and in high-heat environments, that has changed.
This presentation will show that bit flips pose a real attack vector. First, the presentation will describe bit-squatting, an attack akin to typo-squatting, where an attacker controls domains one bit away from a commonly queried domain (e.g. mic2osoft.com vs. microsoft.com). To verify the seriousness of the issue, I bit-squatted several popular domains, and logged all HTTP and DNS traffic. The results were shocking and surprising, ranging from misdirected DNS queries to requests for Windows updates. The presentation will show an analysis of 6 months of real DNS and HTTP traffic to bit-squatted domains. The traffic will be shown in terms of affected platform, domain queried, and HTTP resources requested. Using this data, the presentation will also attempt to ascertain the cause of the bit-flip, such as corruption on the wire, in requestor RAM, or in the RAM of a third party.
The presentation will conclude with potential mitigations of bit-squatting and other bit-flip attacks, including both hardware and software solutions. By the end I hope to convince the audience that bit-squatting and other attacks enabled by bit-flip errors are practical, serious, and should be addressed by software and hardware vendors