netdev - Re: [PATCH net-next] tcp: get rid of sysctl_tcp_adv_win

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]

Date: Thu, 20 Jul 2023 17:50:04 +0200
From: Eric Dumazet <edumazet@...gle.com>
To: Paolo Abeni <pabeni@...hat.com>
Cc: "David S . Miller" <davem@...emloft.net>, Jakub Kicinski <kuba@...nel.org>, netdev@...r.kernel.org, 
	Soheil Hassas Yeganeh <soheil@...gle.com>, Neal Cardwell <ncardwell@...gle.com>, 
	Yuchung Cheng <ycheng@...gle.com>, eric.dumazet@...il.com
Subject: Re: [PATCH net-next] tcp: get rid of sysctl_tcp_adv_win_scale

On Thu, Jul 20, 2023 at 5:43 PM Paolo Abeni <pabeni@...hat.com> wrote:
>
> Hi,
>
> On Mon, 2023-07-17 at 15:29 +0000, Eric Dumazet wrote:
> > +static inline void tcp_scaling_ratio_init(struct sock *sk)
> > +{
> > +     /* Assume a conservative default of 1200 bytes of payload per 4K page.
> > +      * This may be adjusted later in tcp_measure_rcv_mss().
> > +      */
> > +     tcp_sk(sk)->scaling_ratio = (1200 << TCP_RMEM_TO_WIN_SCALE) /
> > +                                 SKB_TRUESIZE(4096);
>
> I'm giving this patch a closer look because mptcp_rcv_space_adjust
> needs to be updated on top of it. Should SKB_TRUESIZE(4096) be replaced
> with:
>
> 4096 + SKB_DATA_ALIGN(sizeof(struct skb_shared_info))
>
> to be more accurate? The page should already include the shared info,
> right?
>
> Likely not very relevant as the ratio is updated to a more accurate
> value with the first packet, mostly to try to understand the code
> correctly.

Hi Paolo.

As discussed with Soheil, I do not think the initial value for
tp->scaling_ratio is very important,
as long as it is not too small of course (otherwise we would have to
increase tcp_rmem[1] to avoid regressions for the first RTT)

Rationale for not adding sizeof(struct skb_shared_info) is because
with GRO, this extra cost tends to disappear.
(Same idea used in truesize_adjust())