Re: [RFC net-next 0/6] net: configurable extra TX headroom for local egress

[Tianyi Gao]

Hi David,

Thanks for the review, and for testing it on x86_64 and arm64.

On Fri, Nov 06, 2026 at 07:00:42PM +0000, David Keogh wrote:
> Tianyi, you have a second use case for this: an XDP program on a veth
> peer. Please describe it here, as it bears directly on the points
> raised above.

At Cloudflare we run some services each in its own network namespace.
A veth pair connects the namespace to the host: veth0 is in the host
and veth1 is in the namespace, with an XDP program on veth1. XDP needs
XDP_PACKET_HEADROOM (256 bytes) in front of each packet. The packets
that the host sends into veth0 have much less: 2 bytes from a UDP
socket and 64 bytes from AF_XDP. So veth1 copies each of them into a
new buffer before the program runs.

Raising needed_headroom on veth0 is not an option. The v1 cover letter
has the details.

> I suggest you post v1, with patches 1-4 from me and 5-8 from you, and
> the cover letter under your own name, since the veth XDP case and the
> new measurements are yours.

I am happy to post v1 as you suggest. I will fold in the points
you listed and post it on the latest net-next. Thanks for the TCP test
as well.

Thanks,
Tianyi