netdev - Re: Question related to GSO6 checksum magic

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:   Tue, 11 Feb 2020 13:01:44 -0800
From:   Alexander Duyck <alexander.h.duyck@...ux.intel.com>
To:     Heiner Kallweit <hkallweit1@...il.com>,
        Eric Dumazet <edumazet@...gle.com>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: Re: Question related to GSO6 checksum magic

On Tue, 2020-02-11 at 20:48 +0100, Heiner Kallweit wrote:
> Few network drivers like Intel e1000e or r8169 have the following in the
> GSO6 tx path:
> 
> ipv6_hdr(skb)->payload_len = 0;
> tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
> 				       &ipv6_hdr(skb)->daddr,
> 				       0, IPPROTO_TCP, 0);
> (partially also w/o the payload_len assignment)
> 
> This sounds like we should factor it out to a helper.
> The code however leaves few questions to me, but I'm not familiar enough
> with the net core low-level details to answer them:
> 
> - This code is used in a number of drivers, so is it something that
>   should be moved to the core? If yes, where would it belong to?
> 
> - Is clearing payload_len needed? IOW, can it be a problem if drivers
>   miss this?
> 
> Thanks, Heiner

The hardware is expecting the TCP header to contain the partial checksum
minus the length. It does this because it reuses the value when it
computes the checksum for the header of outgoing TCP frames and it will
add the payload length as it is segmenting the frames.

An alternative approach would be to pull the original checksum value out
and simply do the checksum math to subtract the length from it. If I am
not mistaken there are some drivers that take that approach for some of
the headers.

- Alex