netdev
[Top] [All Lists]

Re: pppoe and receive checksum offload

To: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
Subject: Re: pppoe and receive checksum offload
From: Alexey Kuznetsov <kuznet@xxxxxxxxxxxxx>
Date: Mon, 28 Feb 2005 17:04:01 +0300
Cc: Stephen Hemminger <shemminger@xxxxxxxx>, "David S. Miller" <davem@xxxxxxxxxxxxx>, mostrows@xxxxxxxxxxxxx, Alexey Kuznetsov <kuznet@xxxxxxxxxxxxx>, netdev@xxxxxxxxxxx
Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=s1024; d=ms2.inr.ac.ru; b=VgYN1eISPKq08gHHZpJRjm7hXFTCuuj1VqgsheI1BiN+5ddldfY6o5JtXPjht/YatDWI1rLelAmVw8UbAl27iQD/gnhaFvMqVnrrbuTNO15RqIo3621/WrH3vRdGhWTvDhMNiSGdC/4zIRpFGkoECEOATLxpIApN037yZ3OQqyc=;
In-reply-to: <20050228113938.GA3393@xxxxxxxxxxxxxxxxxxx>
References: <20050224155906.73890361@xxxxxxxxxxxxxxxxx> <20050228113106.GA3268@xxxxxxxxxxxxxxxxxxx> <20050228113938.GA3393@xxxxxxxxxxxxxxxxxxx>
Sender: netdev-bounce@xxxxxxxxxxx
User-agent: Mutt/1.5.6i
Hello!

> Actually ip_gre is probably right.  It seems that CHECKSUM_UNNECESSARY
> should not be set for PPPOE/GRE packets at all.  So it would be a bug in
> the sk* driver.
> 
> Alexey/Dave, is this interpretation of ip_summed correct?

What's about CHECKSUM_UNNECESSARY, yes, it is set only for TCP/UDP packets,
when device verifies the checksum itself. It is not the case for
PPOE frames. CHECKSUM_HW is more flxible, ipip/gre tunnels use this adjusting
skb->csum by checksum of stripped headers.

ppp_input() could do the same thing. It does not, hence suggested patch
is corrrect minimal solution.

Alexey

<Prev in Thread] Current Thread [Next in Thread>