Re: [RFC] gre: conform to RFC6040 ECN progogation

[Stephen Hemminger <shemminger@vyatta.com>]

On Tue, 25 Sep 2012 00:25:36 +0200
Eric Dumazet <eric.dumazet@gmail.com> wrote:

> On Mon, 2012-09-24 at 14:44 -0700, Stephen Hemminger wrote:
> > Linux GRE was likely written before this RFC and therefore does not
> > conform to one of the rules in Section 4.2.  Default Tunnel Egress Behaviour.
> > 
> > The new code addresses:
> >  o If the inner ECN field is Not-ECT, the decapsulator MUST NOT
> >       propagate any other ECN codepoint onwards.  This is because the
> >       inner Not-ECT marking is set by transports that rely on dropped
> >       packets as an indication of congestion and would not understand or
> >       respond to any other ECN codepoint [RFC4774].  Specifically:
> > 
> >       *  If the inner ECN field is Not-ECT and the outer ECN field is
> >          CE, the decapsulator MUST drop the packet.
> > 
> >       *  If the inner ECN field is Not-ECT and the outer ECN field is
> >          Not-ECT, ECT(0), or ECT(1), the decapsulator MUST forward the
> >          outgoing packet with the ECN field cleared to Not-ECT.
> > 
> > This was caught by Chris Wright while reviewing VXLAN.
> > This code has not been tested with real ECN through tunnel.
> > 
> > Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
> 
> It seems dangerous to me without any logging ?
> 
> One could argue that the outer ECN field should not be CE if the inner
> was Not-ECT
> 
> It means :
> 1) the encapsulator set ECT(0), or ECT(1) and a congestioned hop set CE
> 2) the encapsulator set CE
> 
> 1) or 2) while inner was not-ECT (!!!)
> 
> If a router does such a thing, we should log a message to help
> diagnostics.
> 
> By the way other tunnels probably have the same issues.

Logging is a bad idea in this case since the tunnel might be from a remote
host/protocol and the log would be filled with crap.

The tunnels in general do need to have rx_dropped counter, but it looks
like that isn't being done right either.