CITATION.cff

@inproceedings{10.1145/3672197.3673434,
author = {Zheng, Yusheng and Yang, Yiwei and Chen, Maolin and Quinn, Andrew},
title = {Kgent: Kernel Extensions Large Language Model Agent},
year = {2024},
isbn = {9798400707124},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3672197.3673434},
doi = {10.1145/3672197.3673434},
abstract = {The extended Berkeley Packet Filters (eBPF) ecosystem allows for the extension of Linux and Windows kernels, but writing eBPF programs is challenging due to the required knowledge of OS internals and programming limitations enforced by the eBPF verifier. These limitations ensure that only expert kernel developers can extend their kernels, making it difficult for junior sys admins, patch makers, and DevOps personnel to maintain extensions. This paper presents Kgent, an alternative framework that alleviates the difficulty of writing an eBPF program by allowing Kernel Extensions to be written in Natural language. Kgent uses recent advances in large language models (LLMs) to synthesize an eBPF program given a user's English language prompt. To ensure that LLM's output is semantically equivalent to the user's prompt, Kgent employs a combination of LLM-empowered program comprehension, symbolic execution, and a series of feedback loops. Kgent's key novelty is the combination of these techniques. In particular, the system uses symbolic execution in a novel structure that allows it to combine the results of program synthesis and program comprehension and build on the recent success that LLMs have shown for each of these tasks individually.To evaluate Kgent, we develop a new corpus of natural language prompts for eBPF programs. We show that Kgent produces correct eBPF programs on 80\%---which is an improvement of a factor of 2.67 compared to GPT-4 program synthesis baseline. Moreover, we find that Kgent very rarely synthesizes "false positive" eBPF programs--- i.e., eBPF programs that Kgent verifies as correct but manual inspection reveals to be semantically incorrect for the input prompt. The code for Kgent is publicly accessible at https://github.com/eunomia-bpf/KEN.},
booktitle = {Proceedings of the ACM SIGCOMM 2024 Workshop on EBPF and Kernel Extensions},
pages = {30–36},
numpages = {7},
keywords = {Large Language Model, Symbolic Execution, eBPF},
location = {Sydney, NSW, Australia},
series = {eBPF '24}
}