CCFI: Cryptographically Enforced Control Flow Integrity

Control flow integrity (CFI) restricts jumps and branches within a program to prevent attackers from executing arbitrary code in vulnerable programs. However, traditional CFI still offers attackers too much freedom to chose between valid jump targets, as seen in recent attacks. We present a new approach to CFI based on cryptographic message authentication codes (MACs). Our approach, called cryptographic CFI (CCFI), uses MACs to protect control flow elements such as return addresses, function pointers, and vtable pointers. Through dynamic checks, CCFI enables much finer-grained classification of sensitive pointers than previous approaches, thwarting all known attacks and resisting even attackers with arbitrary access to program memory. We implemented CCFI in Clang/LLVM, taking advantage of recently available cryptographic CPU instructions (AES-NI). We evaluate our system on several large software packages (including nginx, Apache and memcache) as well as all their dependencies. The cost of protection ranges from a 3--18% decrease in server request rate. We also expect this overhead to shrink as Intel improves the performance AES-NI.