Reproduce “When eBPF Meets Machine Learning: On-the-fly OS Kernel Compartmentalization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zicheng Wang, Tiejin Chen, Qinrun Dai, Yueqi Chen, Hua Wei, Qingkai Zeng
Researchers from Nanjing University, University of Colorado Boulder, and Arizona State University developed O2C, a system enabling on-the-fly OS kernel compartmentalization through eBPF and embedded machine learning, to immediately apply security measures without system recompilation or reboot. The system maintains system-wide performance with overheads between -1.26% and 3.81% on LMbench, while its eBPF-embedded decision tree model achieves over 99% accuracy for distinguishing compartment-owned objects.
Key insights
- eBPF can be leveraged to achieve on-the-fly kernel compartmentalization by dynamically instrumenting security policies, thus bypassing traditional offline requirements.
- Machine learning models, specifically customized decision trees, can be embedded within the eBPF kernel space to intelligently infer object types and ensure data integrity during dynamic transitions.
- Effective on-the-fly compartmentalization necessitates a multi-phase workflow to address challenges like managing intertwined assets and auditing untracked objects during the active enforcement transition.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.