Qiang Liu is a postdoc at EPFL, working with Prof. Mathias Payer in the HexHive laboratory. He earned his Ph.D. in 2023 from Zhejiang University (ZJU) under the guidance of Prof. Yajin Zhou and a B.S. degree from the Beijing Institute of Technology (BIT) in 2018. His research in cybersecurity focuses on developing full-lifecycle security enforcement of systems, including ahead-of-release bug finding and in-production attack mitigation. His work has been recognized at all top security conferences: IEEE SSP, Usenix Security, ACM CCS, and ISOC NDSS. He received the Best Paper Awards at USENIX Security'24 and ACM RAID'24. He is also serving on the program committee for USENIX Security'25 and is a reviewer for ACM CSUR and ACM TOSEM.

CV Google Scholar

Qiang Liu's journey began with a Capture the Flag (CTF) competition, where he learned techniques like reverse engineering and exploitation. During his PhD and PostDoc, Qiang Liu pioneered innovative solutions, including auto-extracted input dependencies and a generic executor, to significantly strengthen hypervisor security. These breakthroughs led to the discovery and mitigation of nearly 100 hypervisor vulnerabilities prior to software release. TBF.

Feel free to email me about the following ongoing projects.

[statefulness]#

Statefulness is not well defined. Previous work has shown that states can be accumulated, reached, and formalized. However, interactive and configurable protocols pose challenges to the definition of state due to overlooked response packets and to the exploration of state space due to configuration-based implementation. We are going to demystify statefulness.

Regarding the definition of states: Tango (paper, ACM RAID'24, Best Paper Award) first-in-public defines states with code coverage, which scales to different software.

[GPU security]#

GPUs, new trust base, are part of a collaborative fabric, but isolation is little considered. Unluckily, the GPU hardware and software ecosystem is highly complex such as diverse architectures, proprietary drivers, and non-trivial programming models. How do we protect CPU-GPU-fusion and isolate multi-tenant/multi-task sharing?

[interpreter security]#

How do we design a more secure interpreter engine?

The good: Reflecta (paper, AsiaCCS'25) leverages reflection to infer the signatures of APIs to 3rd party libraries.

[hypervisor security]#

It is not hard at all to ensure hypervisor security. We have built a platform with auto-extracted input dependencies and a generic executor for hypervisors, so we can apply bug-finding (e.g., fuzzing) and mitigation techniques.

Regarding quality input, ViDeZZo (paper, IEEE SSP'23) first proposed intra- and inter-message dependencies, and then Truman (paper, ISOC NDSS'25) highlighted state-dependency. Truman further explained how to auto-extract intra-/inter-message, and state dependencies from the Linux kernel via static analysis.

Regarding generic executors, HyperPill (paper, USENIX Security'24, Best Paper Award) first propose a snapshot-based hypervisor dock, capable of running both open-source and closed-source and both x86 and aarch64 hypervisors.

Other tools

Open projects

  • New target: Rust-based hypervisors (e.g., firecracker)
  • New tool: Hyper-Cube for 64, HyperPill for Aarch64
  • New attacks: race condition, out-of-resources
  • Automatic exploit generation: ref 1

[firmware rehosting]#

We aim to rehost embedded Linux kernels (the Linux kernels used in embedded devices). FirmGuide (paper, ASE'21) first modeled all interrupt controllers and timers. Then, ECMO (paper, CCS'21) supported other peripherals like NIC via binary rewriting.

Other tools

  • pyqemulog is the qemu-log ported to Python. It converts the structured trace generated by QEMU with -d to JSON.
  • llbic, short for LLVM Linux Build Issues Collection, compiles old Linux kernels in LLVM bitcode. It replaces GCC to clang and adjusts cflags in the command lines to generate bitcode files, and then links them all together to a vmlinux.bc.
  • openwrt-build-docker supports automatically building the OpenWrt project given a target/subtarget of a specific OpenWrt revision from 10.03 to 19.07.1.

[android authentication (archived)]

This project evaluates existing and proposes new implicit continuous authentication approaches to serve as a second authentication factor alongside fingerprint and facial identification. I joined this project as an intern and then proposed part of it as my final project for my bachelor's degree. Relative research papers are RiskCog (TMC'20), ESPIALCOG (TMC'20), One Cycle Attack (TIFS'20), and TRAPCOG (TMC'23)).