netdev - Re: GRO disabled with IPv4 options

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]

Message-ID: <1510871691.2043.20.camel@edumazet-glaptop3.roam.corp.google.com>
Date:   Thu, 16 Nov 2017 14:34:51 -0800
From:   Eric Dumazet <eric.dumazet@...il.com>
To:     Cristian Klein <cklein@...umu.se>
Cc:     netdev@...r.kernel.org, herbert@...dor.apana.org.au,
        Ahmed Ali-Eldin <ahmeda@...umu.se>
Subject: Re: GRO disabled with IPv4 options

On Thu, 2017-11-16 at 16:12 +0100, Cristian Klein wrote:
> [CC-ing Herbert Xu, who is to 'git blame' for the code in question. :)]
> 
> Dear all,
> 
> We are working on a research prototype which, among others, adds a new 
> IPv4 option. During testing we noticed that the packets captured by 
> tcpdump shrank from 10s of KBs to the MTU, which indicates that Generic 
> Receive Offload (GRO) got disabled.
> 
> Upon further investigation, we found the following line in 
> `inet_gro_receive`:
> 
> 	if (*(u8 *)iph != 0x45)
> 		goto out_unlock;
> 
> in plain English, don't do GRO if any IPv4 options are present.
> 
> Does somebody know the rationale for this? Is it because IPv4 options 
> are rarely used, hence implementing GRO in that case does not pay off or 
> are there some caveats? Specifically would it make sense to do GRO when 
> the IPv4 options are byte-identical in consecutive packets?

I guess nobody really uses IPV4 options, and you are the first caring
enough ;)