Systems & Kernel Engineering · Linux 6.12 LTS

Hey, I'm Hyunbin (bomberrie).

I'm a systems engineer in the United Kingdom working on Linux networking, packet processing, and storage. Most of my time is spent where high-level abstractions break down—tracking down subtle kernel bugs, writing eBPF tools to see what the network stack is actually doing, and making sure packets move at line rate.

Outside of code, I spend way too much time with my NixOS home lab, testing routing hardware on my own network, and listening to Rammstein. I like fixing problems at the root rather than papering over them in userspace.

Live PresenceOnline
Hyunbin
Hyunbin@bomberrie
Tracing netfilter flowtables in Linux 6.12
LAST PLAYED ON SPOTIFY
Recovery — Rival Consoles
Ambient Electronic · Erased Tapes
Engineering Philosophy

Root Causes over Band-Aids

I prefer isolating and fixing regressions directly in the kernel tree over patching symptoms in userspace. If a packet drops or a disk stalls, I want to see the exact root cause in the driver or subsystem.

Linux 6.12 LTSnet-nextNixOS FlakeseBPF / XDPio_uring
Local TimeUK / GMT
23:57:12
Norfolk, United Kingdom (GMT)
Daily OS:NixOS (Flakes)
Kernel Tree:6.12 LTS / net-next
Editor & Shell:Neovim · Zsh · tmux
Systems & Kernel Work

Selected Systems & Kernel Engineering

Selected kernel patches, packet telemetry tools, and storage experiments.

4 Core Telemetry Indicators
FLOWTABLE BYPASSNF_FLOW
10 Gbps+

Zero conntrack L2 neighbor misroutes

XDP PROBE LATENCYNDO_BPF
850 ns

Direct driver-layer zero-copy BPF hook

IO_URING THROUGHPUTRING_SQ
480k IOPS

Lock-free LSM memtable flush rate

FAILOVER CONVERGENCEWG_MESH
< 240 ms

WireGuard dynamic shortest-path mesh

OpenWrt 24.10 Kernel Audit & Upstream Patch

Resolving Flowtable PBR Egress MAC Corruption

Packet Simulation: Switch between unpatched regression and verified fix
Interactive Pipeline (5 Stages)← Swipe to explore →
Stage 01
NIC Ingress
eth0: 10GbE
→
Stage 02
Flowtable Hook
Ingress Priority
→
Stage 03
PBR Routing
FIB Table 200
→
Stage 04 (Core)
dst_neigh_lookup()
Valid Gateway MAC
→
Stage 05
Line-Rate Egress
eth1 (WAN)
PACKET TRACE:dst_neigh_lookup resolved L2 neighbor -> Gateway MAC valid (52:54:00:12:34:56) -> Wire-speed Forwarding
Patch Verified: Explicit dst_neigh_lookup Resolves Gateway MAC

Our upstream kernel patch invokes dst_neigh_lookup(&rt->dst, daddr) directly in the flow offload egress path. This ensures that the destination neighbor MAC address matches the active PBR routing table entry, completely eliminating L2 frame corruption while preserving 10 Gbps+ line-rate flowtable hardware offload speeds.

Verified reproducible on Linux 6.6 and 6.12 LTS kernelsView Upstream Patch Repository
Open-Source Systems Infrastructure

Open-Source Codebases & Engineering

Production kernel patches, packet telemetry tools, and lock-free storage engines backing our benchmark metrics.

3 Core RepositoriesAll Repositories
C / Kernel C
PRODUCTION_PATCH
openwrt-flowoffload-pbr
C · Linux Kernel · Netfilter

Upstream kernel patch resolving L2 destination MAC corruption on multi-WAN Policy-Based Routing with hardware flowtable offload enabled.

Line-Rate Flowtable
10 Gbps+
Subsystem
netfilter / nft_flow_offload
eBPF / Aya Rust
ACTIVE_CORE
ebpf-xdp-telemetry
C · eBPF / XDP · Rust

Driver-layer zero-copy packet inspection and ringbuffer telemetry probe operating at 850ns latency before sk_buff allocation.

Driver Probe Latency
850 ns
Subsystem
XDP_DRV / BPF RingBuffer
Rust / Systems
EXPERIMENTAL
wireguard-dynamic-mesh & io-uring-lsm
Rust · io_uring · NixOS

Lock-free NVMe storage engine experiments and sub-240ms dynamic shortest-path WireGuard failover daemon.

< 240ms Convergence
480k IOPS
Subsystem
io_uring SQ / WG Netlink
BARE-METAL INFRASTRUCTURE

Hardware & Home Lab

The physical routers, storage nodes, and test benches where I reproduce bugs and benchmark kernel patches.

All Lab Nodes Active
Explore NixOS Flakes on GitHub
Physical Lab Interconnect & Topology
WAN INGRESSOPTICAL
10GbE SFP+ Fiber
Direct ISP AS Transit
10 Gbps Line-Rate
GATEWAYACTIVE
edge-gw-01.lan
NixOS · Flowtable Bypass
Bird2 BGP · Linux 6.12
STORAGENVMe ZFS
storage-node-01.lan
4x PCIe 4.0 NVMe · io_uring
> 480k IOPS Flush Rate
BENCHISOLATED
netns Test Bench
veth pairs · Intel X520
eBPF/XDP Driver Hooks
PACKET FLOW:WAN Ingress→edge-gw-01→storage-node-01 & netns bench
EDGE ROUTING APPLIANCEONLINE

edge-gw-01.lan

Multi-WAN routing gateway & verification host

Kernel:Linux 6.12 LTS
Architecture:x86_64 / aarch64
Orchestration:NixOS 24.11 Flakes
Routing Transit:Bird2 BGP Peered
Low-latency direct transit upstream10GbE SFP+
STORAGE & COMPUTE NODEZFS RAIDZ

storage-node-01.lan

io_uring NVMe storage testbed & LSM bench

Drive Topology:4x Samsung 990 Pro PCIe 4.0
File System:ZFS native encryption
Async Subsystem:io_uring polled SQ
Throughput Target:> 480k IOPS
NVMe-backed LSM flush engineio_uring SQ
SYNTHETIC TEST BENCHSYNTHETIC

Synthetic Test Harness

Linux network namespace reproducer suite

NIC Hardware:Intel X520 10GbE (ixgbe)
Offload Engine:Flowtable Hardware Bypass
Isolation:netns veth pairs
BPF Execution:XDP Driver Mode
Reproducible bare-metal closures100% Hermetic