oo7: Low-overhead Defense against Spectre Attacks via Binary Analysis.

Abstract

The Spectre vulnerability in modern processors has been reported earlier this year (2018). The key insight in this vulnerability is that speculative execution in processors can be misused to access secrets speculatively. Subsequently even though the speculatively executed states are squashed, the secret may linger in micro-architectural data structures such as cache, and hence can be potentially accessed by an attacker via side channels. In this report, we propose oo7, a binary analysis framework to check and fix code snippets against potential vulnerability to Spectre attacks. Our solution employs control flow extraction, taint analysis and address analysis to detect tainted conditional branches and their ability to impact memory accesses. Fixing is achieved by selectively inserting a small number of fences, instead of inserting fences after every conditional branch. Due to the accuracy of our analysis, oo7 suggests inserting less fences, and is shown experimentally to impose acceptably low performance overheads; less than 2% performance overhead is observed in our experiments on GNU Core utilities. Moreover, the accuracy of the analysis allows oo7 to effectively detect fourteen (14) out of the fifteen (15) Spectre vulnerable code patterns proposed by Paul Kocher, a feat that could not be achieved by the Spectre mitigation in C/C++ compiler proposed by Microsoft. While oo7 is both low-overhead and effective, for large scale deployment of our solution we need to investigate and optimize the time taken by our compile-time analysis. Finally, we show that similar binary analysis solutions are possible for detecting and fixing Meltdown.