netdev - Re: IP GRO verifies csum again?

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] [day] [month] [year] [list]

Message-ID: <2154bca7-17d0-3ed3-ab58-55f33cd13ac5@gmail.com>
Date:   Thu, 18 Jul 2019 10:27:01 +0200
From:   Eric Dumazet <eric.dumazet@...il.com>
To:     Jacob Wen <jian.w.wen@...cle.com>, netdev@...r.kernel.org
Cc:     herbert@...dor.apana.org.au
Subject: Re: IP GRO verifies csum again?

On 7/18/19 9:49 AM, Jacob Wen wrote:
> Hi,
> 
> inet_gro_receive verifies IP csum but a NIC already did so and set CHECKSUM_UNNECESSARY.
> 
> 
> https://github.com/torvalds/linux/blob/v5.2/net/ipv4/af_inet.c#L1432-L1433
> 
> if (unlikely(ip_fast_csum((u8 *)iph, 5)))
> 
>         goto out_unlock;
> 
> 
> Is this a bug?
> 

This checksum validates the TCP one, which is the real cost, since we need to touch all
the packet.

We do not bother 'offloading' IPV4 checksum over 20 bytes or so, since in modern cpus,
having the cache line hot in cpu caches means the checksum is almost free.

Adding a test here would not always be a win, say for CHECKSUM_COMPLETE cases,
which we try to generalize in favor of old CHECKSUM_UNNECESSARY.