Pre-Conference Talk by LIN Yan | SafeStack+: Enhanced Dual Stack to Combat Data-Flow Hijacking

SafeStack+: Enhanced Dual Stack to Combat Data-Flow Hijacking

SafeStack, initially proposed as a key component of Code Pointer Integrity (CPI), separates the program stack into two distinct regions to provide a safe region for sensitive code pointers. SafeStack can prevent buffer overflow attacks that overwrite sensitive code pointers, e.g., return addresses, to hijack control flow of the program, and has been incorporated into the Clang project of LLVM as a C-based language frontend. In this paper, we propose and implement SafeStack+, an enhanced dual stack LLVM plug-in that further protects programs from data-flow hijacking. SafeStack+ locates data flow sensitive variables on the unsafe stack that could potentially affect evaluation of branching conditions, and adds canaries of random sizes and values to them to detect malicious overwriting. We implement SafeStack+ as a plugin on LLVM 3.8 and perform extensive experiments to justify a lazy checking mechanism that adds on average 3.0% of runtime and 5.3% of memory overhead on top of SafeStack on SPEC CPU2006 benchmark programs. Our security analysis confirms that SafeStack+ is effective in detecting data-flow hijacking attacks.

This is a pre-conference talk for 22nd Australasian Conference on Information Security and Privacy (ACISP 2017).