On Fri, 2024-03-29 at 09:26 -0700, Alexei Starovoitov wrote:

On Thu, Mar 28, 2024 at 10:35 AM Matthieu Baerts [off-list ref] wrote:

quoted

Hi Martin,

On 27/03/2024 23:16, Martin KaFai Lau wrote:

(...)

quoted

Unrelated, is there a way to tell if a tcp_sock is a subflow?

Yes, you can use "sk_is_mptcp(sk)". Please note that this 'sk' *has* to
be a tcp_sock, this is not checked by the helper.

That's what is used with bpf_mptcp_sock_from_subflow()

https://elixir.bootlin.com/linux/latest/source/net/mptcp/bpf.c#L15

quoted

bpf prog
can use it to decide if it wants to setsockopt on a subflow or not.

I think it is important to keep the possibility to set socket options
per subflow. If the original issue discussed here is limited to
set_rcvlowat(), best to address it there.

All makes sense to me.

Paolo,
could you send an official patch?

Sure, thank you for reminding me. This was falling off my radar. I'll
send it after some testing.

Thanks!

Paolo