Re: [tcpm] question about if a recent Linux patch is compliant with RFC7323 on window scaling

["Cui, Cheng" <[e-mail]>]

On 9/4/16, 6:55 AM, "Yoshifumi Nishida" <[e-mail]> wrote:
> I think the Linux patch at least follows the following principle in RFC793.
> 
>   The mechanisms provided allow a TCP to advertise a large window and to
>   subsequently advertise a much smaller window without having accepted
>   that much data.  This, so called ³shrinking the window,² is strongly
>   discouraged.  The robustness principle dictates that TCPs will not
>   shrink the window themselves, but will be prepared for such behavior
>   on the part of other TCPs.
> 
> I personally think a real question would be how FreeBSD prepares for
> shrinking window.
> Linux chose one way for it, but I guess all implementations don¹t need
> to follow the same way.
> 
> --
> Yoshi

Hi Yoshi,

Thanks a lot for the reply.

I agree that's a choice for Linux. But I think FreeBSD has to follow the same
way; otherwise, Linux has to do something at its sender side to send a correct
th_seq number.

I think this forum may not be a best place to talk about FreeBSD
implementation. But talking about FreeBSD, I did not find anything in FreeBSD
to round-up the shrinking window due to window scale precision loss, neither
did I find anything about how FreeBSD can tolerate the incoming left-shifted
th_seq ("tp->snd_una+tp->snd_wnd") in an ACK packet sent by Linux in this
particular case, as Linux had sent all the data in a previous larger window. I
guess the FreeBSD will just drop the Linux ACK packet as a DUP (see "todrop =
tp->rcv_nxt - th->th_seq;").

I also sent a question to the FreeBSD mailing lists. But so far no reply yet:
https://lists.freebsd.org/pipermail/freebsd-transport/2016-August/000141.html
https://lists.freebsd.org/pipermail/freebsd-net/2016-September/046048.html

Therefore, I think my best plan is to send separate patches to Linux and
FreeBSD. For Linux, the patch will be correct the th_seq number at sender side
if window scale precision loss is detected. For FreebSD, I think the patch has
to round-up the receive-advertise-window at receiver side.

-- Cheng