Type of Publication: Article in Collected Edition

Address Oblivious Code Reuse: On the Effectiveness of Leakage Resilient Diversity

Rudd, Robert; Skowyra, Richard; Bigelow, David; Dedhia, Veer; Hobson, Thomas; Crane, Stephen; Liebchen, Christopher; Larsen, Per; Davi, Lucas; Franz, Michael; Sadeghi, Ahmad-Reza; Okhravi, Hamed
Title of Anthology:
Proc. of 24th Annual Network & Distributed System Security Symposium (NDSS)
Publication Date:
Link to complete version:
Download BibTeX


Memory corruption vulnerabilities not only allow modification of control data and injection of malicious payloads; they also allow adversaries to reconnoiter a diversified program, customize a payload, and ultimately bypass code randomization defenses. In response, researchers have proposed and built various leakage-resilient defenses against code reuse. Leakage-resilient defenses use memory protection techniques to prevent adversaries from directly reading code as well as pointer indirection or encryption techniques to decouple code pointers from the randomized code layout, avoiding indirect leakage. In this paper, we show that although current code pointer protections do prevent leakage per se, they are fundamentally unable to stop code reuse. Specifically, we demonstrate a new class of attacks we call address-oblivious code reuse that bypasses state-of-the-art leakage-resilience techniques by profiling and reusing protected code pointers, without leaking the code layout. We show that an attacker can accurately identify protected code pointers of interest and mount code-reuse attacks at the abstraction level of pointers without requiring any knowledge of code addresses. We analyze the prevalence of opportunities for such attacks in popular code bases and build three real-world exploits against Nginx and Apache to demonstrate their practicality. We analyze recently proposed leakage resilient defenses and show that they are vulnerable to address oblivious code reuse. Our findings indicate that because of the prevalence of code pointers in realistic programs and the fundamental need to expose them to    read    operations (even indirectly), diversity defenses face a fundamental design challenge in mitigating such attacks.