eBPFとXDPに基づくカーネル内ネットワーク機能

本研究では,eBPF,XDP,uBPFなどを活用し,高速性と柔軟性を兼ね備えたネットワーク機能の実現を目指しています.

従来,高度なネットワーク機能は主にユーザ空間で実装されており,パケットをカーネル空間とユーザ空間の間で受け渡す際にオーバーヘッドが発生します.一方,eBPFやXDPを用いることで,カーネル内で直接パケットを処理できるため,ユーザ空間へのパケット転送に伴うオーバーヘッドを削減できます.しかし,eBPFには実行環境やプログラムに関する制約があるため,複雑な機械学習アルゴリズムなどをカーネル内に実装することは容易ではありません.

このような背景の下で,軽量なニューラルネットワークや決定木をeBPF/XDP上に実装し,機械学習に基づくカーネル内侵入検知システム(Intrusion Detection System: IDS)の実用性を評価しました.さらに,カーネルバイパス技術の一つであるAF_XDPを活用し,カーネル空間とユーザ空間のパケット処理を組み合わせることで,高速性と柔軟性を両立する手法を検討しました123.

また,ユーザ空間でeBPFプログラムを実行できるuBPFに着目し,機械学習アルゴリズムを活用したネットワーク侵入検知・防御システム(Intrusion Detection and Prevention System: IDPS)の実現に取り組んでいます.特に,分散型サービス拒否(Distributed Denial of Service: DDoS)攻撃を対象として,eBPFとuBPFに基づく侵入検知・防御システムの実用性を評価し4,さらに,uBPFを用いた機械学習ベースのネットワーク侵入検知・防御システムについて検討しています5.

加えて,Service Function Chaining(SFC)におけるネットワーク機能の実行経路の信頼性を確保するため,eBPFを用いたOrdered Proof of Transit(OPT)による経路検証手法についても研究しています6.

In-Kernel Network Functions with eBPF and XDP

This research aims to realize high-performance and flexible network functions using eBPF, XDP, and uBPF.

Traditionally, advanced network functions have primarily been implemented in user space, incurring overhead when transferring packets between kernel space and user space. In contrast, eBPF and XDP enable direct packet processing within the kernel, reducing the overhead associated with packet transfers to user space. However, implementing advanced network functions, such as complex machine learning algorithms, in the kernel remains challenging due to restrictions imposed by the eBPF execution environment and programming model123.

To address these challenges, we implemented lightweight neural networks and decision trees using eBPF/XDP and evaluated the practicality of machine learning-based in-kernel Intrusion Detection Systems (IDSs). Furthermore, we investigated an approach that combines in-kernel and user-space packet processing using AF_XDP, a kernel-bypass technology, to achieve both high performance and flexibility.

We have also investigated network Intrusion Detection and Prevention Systems (IDPSs) based on machine learning algorithms using uBPF, which enables the execution of eBPF programs in user space. In particular, we evaluated the practicality of eBPF/uBPF-based intrusion detection and prevention systems against Distributed Denial of Service (DDoS) attacks4 and further investigated machine learning-based network intrusion detection and prevention systems using uBPF5.

In addition, to ensure the trustworthiness of network function traversal in Service Function Chaining (SFC), we have investigated an eBPF-based path verification method using Ordered Proof of Transit (OPT)6.


  1. T. Hara and M. Sasabe, “On Practicality of Kernel Packet Processing Empowered by Lightweight Neural Network and Decision Tree,” in Proc. of International Conference on Network of the Future (NoF), Oct. 2023, pp. 89–97. doi: 10.1109/NoF58724.2023.10302811. ↩︎ ↩︎

  2. T. Hara and M. Sasabe, “Practicality of In-Kernel/User-Space Packet Processing Empowered by Lightweight Neural Network and Decision Tree,” Computer Networks, vol. 240, p. 110188, Feb. 2024, doi: 10.1016/j.comnet.2024.110188. ↩︎ ↩︎

  3. H. Taguchi, T. Hara, and S. Kasahara, “Unsupervised Real-Time In-Kernel Intrusion Detection System Using Autoencoders and eBPF,” IEICE Transactions on Communications, pp. 1–12, 2025, doi: 10.23919/transcom.2025EBP3028. ↩︎ ↩︎

  4. T. Hara and S. Kasahara “Practicality Analysis of eBPF/uBPF-based Network Intrusion Detection and Prevention Systems for Distributed Denial of Service Attacks,” in Proc. of the First Workshop on AI Native Distributed Intelligence for 6G Networks (6G AI-RAN 2026) organized in conjunction with IEEE International Conference on Computer Communications (INFOCOM 2026), pp.1-6, May 2026. doi:10.1109/INFOCOM59046.2026.11571593 ↩︎ ↩︎

  5. T. Hara and S. Kasahara, “uBPF-based Network Intrusion Detection and Prevention Systems Empowered by Machine Learning Algorithms,” Computer Networks, vol. 292, pp. 112868:1-112868:13, Jan. 2027. doi: 10.1016/j.comnet.2026.112868 ↩︎ ↩︎

  6. T. Hara and M. Sasabe, “eBPF-Based Ordered Proof of Transit for Trustworthy Service Function Chaining,” IEEE Transactions on Network and Service Management, pp. 1–1, 2025, doi: 10.1109/TNSM.2025.3550333. ↩︎ ↩︎