Re: Multicast and UDP checksums

[Frank Kardel <kardel%netbsd.org@localhost>]

On 08/21/12 08:58, Christos Zoulas wrote:
> In article<50315ED7.6030203%netbsd.org@localhost>,
> Frank Kardel<kardel%netbsd.org@localhost>  wrote:
> > Hi,
> >
> > while debugging ntp multicast issues I found a strange behaviour on 6.99.5.
> >
> > ntpd in machine A (amd64, wm interface) sends multicast packets to
> > 224.0.1.1.
> >
> > tcpdump on machine B (amd64, wm interface) sees both packets types
> > fine (no complaints about the checksum).
> >
> > ntpd on machine B never sees the multicast packets, because they get
> > dropped due to 'bad checksum' in the udp4_realinput() path.
> >
> > ntpd on machine B sees the plain broadcast packets, because the checksum
> > is fine.
> >
> > Any ideas why the ntp multicast packets get dropped, but tcpdump
> > finds nothing wrong about them ?
> Do you have hardware checksumming on, on either side?
>
> christos
On and off.
    started with on on both sides - bad
    then off on receiver - bad
    then off on sender and receiver - bad
so, no improvement.

broadcasts make it through with no problems (sane checksums).

tcpdump does not complain when looking at the packets (multicast and 
broadcast) and I believe tcpdump makes these checks also.

I have yet to look very close at packet data and checksums (earliest 
next WE) .
I seem to remember, that multicast reception was working some time ago.

Frank