Code-Reuse Attacks and Defenses

Exploitation of memory corruption vulnerabilities in widely used software has been a threat for almost three decades and no end seems to be in sight. In particular, code-reuse techniques such as return-oriented programming offer a robust attack technique that is extensively used to exploit memory corruption vulnerabilities in modern software programs (e.g. web browsers or document viewers). Whereas conventional control-flow attacks (runtime exploits) require the injection of malicious code, code-reuse attacks leverage code that is already present in the address space of an application to undermine the security model of data execution prevention (DEP). In addition, code-reuse attacks in conjunction with memory disclosure attack techniques circumvent the widely applied memory protection model of address space layout randomization (ASLR). To counter this ingenious attack strategy, several proposals for enforcement of control-flow integrity (CFI) and fine-grained code randomization have emerged. In this dissertation, we explore the limitations of existing defenses against code-reuse attacks. In particular, we demonstrate that various coarse-grained CFI solutions can be effectively undermined, even under weak adversarial assumptions. Moreover, we explore a new return-oriented programming attack technique that is solely based on indirect jump and call instructions to evade detection from defenses that perform integrity checks for return addresses. To tackle the limitations of existing defenses, this dissertation introduces the design and implementation of several new countermeasures. First, we present a generic and fine-grained CFI framework for mobile devices targeting ARM-based platforms. This framework preserves static code signatures by instrumenting mobile applications on-the-fly in memory. Second, we tackle the performance and security limitations of existing CFI defenses by introducing hardware-assisted CFI for embedded devices. To this end, we present a CFI-based hardware implementation for Intel Siskiyou Peak using dedicated CFI machine instructions. Lastly, we explore fine-grained code randomization techniques