The BPF-programmable network device
lwn.net
lwn.net
However, I fail to see the benefits vs. an XDP probe attached to a VM's TAP interface. It already bypasses the host's stack, and can already redirect to other network devices. Those can also modify/resize the packet to add e.g. encapsulation.
Perhaps those get passed a full SKB instead of a "small" packet? As with virtio offloads, guests can pass a full 64k "packet" for a full TCP buffer, instead of already segmented packets, offloading those to the physical card.
It'll be interesting to see who adopts this first & who adopts it strongest. There's already been a ton of performance wins by using eBPF but making the host & guest driver be a big eBPF shim is pretty next level!
Wasn’t netkit an userspace networking thing already, some ~10 years ago?
But apparently it renamed and got generally restructured?
I remember it as a very nice tool to learn about linux networking
Apparently the name has changed to Khatara: