Control-Flow Hijacking: Are We Making Progress?

Memory corruption errors in C/C++ programs remain the most common source of security vulnerabilities in today's systems. Over the last 10+ years the security community developed several defenses [4]. Data Execution Prevention (DEP) protects against code injection -- eradicating this attack vector. Yet, control-flow hijacking and code reuse remain challenging despite wide deployment of Address Space Layout Randomization (ASLR) and stack canaries. These defenses are probabilistic and rely on information hiding.

The deployed defenses complicate attacks, yet control-flow hijack attacks (redirecting execution to a location that would not be reached in a benign execution) are still prevalent. Attacks reuse existing gadgets (short sequences of code), often leveraging information disclosures to learn the location of the desired gadgets. Strong defense mechanisms have not yet been widely deployed due to (i) the time it takes to roll out a security mechanism, (ii) incompatibility with specific features, and (iii) performance overhead. In the meantime, only a set of low-overhead but incomplete mitigations has been deployed in practice.

Control-Flow Integrity (CFI) [1,2] and Code-Pointer Integrity (CPI) [3] are two promising upcoming defense mechanisms, protecting against control-flow hijacking. CFI guarantees that the runtime control flow follows the statically determined control-flow graph. An attacker may reuse any of the valid transitions at any control-flow transfer. We compare a broad range of CFI mechanisms using a unified nomenclature based on (i) a qualitative discussion of the conceptual security guarantees, (ii) a quantitative security evaluation, and (iii)~an empirical evaluation of their performance in the same test environment. For each mechanism, we evaluate (i) protected types of control-flow transfers, (ii) the precision of the protection for forward and backward edges. For open-source compiler-based implementations, we additionally evaluate (iii) the generated equivalence classes and target sets, and (iv) the runtime performance. CPI on the other hand is a dynamic property that enforces selective memory safety through bounds checks for code pointers by separating code pointers from regular data.