ARMore: Pushing Love Back Into Binaries

Static rewriting enables late-state code changes (e.g., to add mitigations, to remove unnecessary code, or to instrument for code coverage) at low overhead in security-critical environments. Most research on static rewriting has so far focused on the x86 architecture. However, the prevalence and proliferation of ARM-based devices along with a large amount of personal data (e.g., health and sensor data) that they process calls for efficient introspection and analysis capabilities on the ARM platform. Addressing the unique challenges on aarch64, we introduce ARMore, the first efficient, robust, and heuristic-free static binary rewriter for arbitrary aarch64 binaries that produces reassembleable assembly. The key improvements introduced by ARMore make the recovery of indirect control flow an option rather than a necessity. Instead of crashing, the cost of an uncovered target only causes the small overhead of an additional branch. ARMore can rewrite binaries from different languages and compilers (even arbitrary hand-written assembly), both on PIC and non-PIC code, with or without symbols, including exception handling for C++ and Go binaries, and also including binaries with mixed data and text. ARMore is sound as it does not rely on any assumptions about the input binary. ARMore is also efficient: it does not employ any expensive dynamic translation techniques, incurring negligible overhead (<1% in our evaluated benchmarks). Our AFL++ coverage instrumentation pass enables fuzzing of closed-source aarch64 binaries at three times the speed compared to the state-of-the-art (AFL-QEMU), and we found 58 unique crashes in closed-source software. ARMore is the only static rewriter whose rewritten binaries correctly pass all SQLite3 and core-utils test cases and autopkgtest of 97.5% Debian packages.