Wednesday 30 September 16:30 - 17:00, Red room
Costin Ionescu (Broadcom)
The security industry is well aware of the importance of delivering secure solutions as part of their software offering given that their applications run with elevated privileges. Sometimes, even with the best intentions and carefully thought out precautions, blunders still happen.
This highlights the importance of a systemic approach that minimizes the impact of human (or soon AI-overlord) mistakes.
In this paper we'll briefly go over some of the existing approaches for hardening software - from virtualization, containerization and sandboxing to smaller-scale in-process focused solutions using a combination of hardware-assisted features with OS support and/or compiler features such as secure enclaves (Intel SGX), memory protection extensions (Intel MPX), ASLR in all modern OSes, GNU RELRO in GNU/Linux, Microsoft's Code Flow Guard, Control-flow Enforcement Technology (CET) Shadow Stack.
We'll also review similar securing efforts used in the browser world (such as Google's NaCl/PNaCl, asm.js, Mozilla's WASM) and discuss the evolution of safety features in modern languages (C++, Rust).
We can then discuss the approach that the security engines team in Symantec (now a division of Broadcom) took a few years back to secure new engine releases. We'll describe how clang/LLVM instrumentation is used to produce hardened binaries which dramatically reduce the risk of remote code execution, denial of service and other attacks, by severely mitigating the impact of bugs such as buffer overflows, unbounded recursion (stack exhaustion), infinite loops. The hardened binaries are then used as templates for micro in-process VMs (where we designate a portion of the process address space to be used as host for a separate execution unit).
We will demonstrate using some proof-of-concept code how this approach handles a few of the typical bugs mentioned above.
Costin Ionescu Costin is a security researcher and security software developer with Symantec - now a division of Broadcom. |
Minhee Lee (Financial Security Institute)
Daegyu Kang (Financial Security Institute)