netdev - Re: [PATCH net-next 0/2] net: xsk: add two sysctl knobs

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAL+tcoDYiwH8nz5u=sUiYucJL+VkGx4M50q9Lc2jsPPupZ2bFg@mail.gmail.com>
Date: Tue, 17 Jun 2025 10:15:26 +0800
From: Jason Xing <kerneljasonxing@...il.com>
To: Stanislav Fomichev <stfomichev@...il.com>
Cc: davem@...emloft.net, edumazet@...gle.com, kuba@...nel.org, 
	pabeni@...hat.com, bjorn@...nel.org, magnus.karlsson@...el.com, 
	maciej.fijalkowski@...el.com, jonathan.lemon@...il.com, sdf@...ichev.me, 
	ast@...nel.org, daniel@...earbox.net, hawk@...nel.org, 
	john.fastabend@...il.com, bpf@...r.kernel.org, netdev@...r.kernel.org, 
	Jason Xing <kernelxing@...cent.com>
Subject: Re: [PATCH net-next 0/2] net: xsk: add two sysctl knobs

Hi Stanislav,

On Tue, Jun 17, 2025 at 9:11 AM Stanislav Fomichev <stfomichev@...il.com> wrote:
>
> On 06/17, Jason Xing wrote:
> > From: Jason Xing <kernelxing@...cent.com>
> >
> > Introduce a control method in the xsk path to let users have the chance
> > to tune it manually.
>
> Can you expand more on why the defaults don't work for you?

We use a user-level tcp stack with xsk to transmit packets that have
higher priorities than other normal kernel tcp flows. It turns out
that enlarging the number can minimize times of triggering sendto
sysctl, which contributes to faster transmission. it's very easy to
hit the upper bound (namely, 32) if you log the return value of
sendto. I mentioned a bit about this in the second patch, saying that
we can have a similar knob already appearing in the qdisc layer.
Furthermore, exposing important parameters can help applications
complete their AI/auto-tuning to judge which one is the best fit in
their production workload. That is also one of the promising
tendencies :)

>
> Also, can we put these settings into the socket instead of (global/ns)
> sysctl?

As to MAX_PER_SOCKET_BUDGET, it seems not easy to get its
corresponding netns? I have no strong opinion on this point for now.

Thanks,
Jason