Abstract: As more and more memory-disclosure bugs been discovered, it’s important to protect memory safety. Nowadays mainstream of this field prefers enforcing memory isolation to protect memory safety. However, none of them can achieve a fine-grained protection while easy-to-deploy solution since most solutions requires manually modification to source code, or they enforce memory isolation in a coarse-grained way. We present a novel method, which can traces and collects all possible operations that access the sensitive data in runtime, and can perform automatically code transformation during compilation. Our solution do not require manually modification, and evaluation shows that we can efficiently eliminate memory disclosure attacks, such as Heartbleed, found in OpenSSL. Meanwhile, the overhead of compilation and memory cost are lower than 1%.

Key words: memory disclosure attacks, memory data isolation