Re: em0 hijacking traffic to port 623

On 7/13/07, Ian FREISLICH <ianf_at_clue.co.za> wrote:
> John Baldwin wrote:
> > On Monday 21 May 2007 07:17:07 pm Jack Vogel wrote:
> > > On 5/21/07, Sten Spans <sten_at_blinkenlights.nl> wrote:
> > > > On Mon, 21 May 2007, Ian FREISLICH wrote:
> > > >
> > > > > Hi
> > > > >
> > > > > We've noticed an issue on our firewalls where the first em device
> > > > > in the system hijacks inbound port 623 tcp and udp.  The OS never
> > > > > sees this traffic.  Interestingly, em1 and em2 do not appear to be
> > > > > afflicted by this problem.  Some reading I've done points to a
> > > > > similar conclusion:
> > > > >
> > > > > http://blogs.sun.com/shepler/entry/port_623_or_the_mount
> > > > >
> > > > > I've looked at the bios, but I can't find any settings that remotely
> > > > > hint IPMI or RMCP+ or serial-over-lan.
> > > > >
> > > > > Does anyone know how I can stop the card or system from stealing
> > > > > port 623 in hardware or must I just stop using em0 (and/or Intel NICS)?
> > > >
> > > > Does "ifconfig em0 promisc" help ?
> > > > That fixed firmware related vanishing ipv6 packets on fxp and em.
> > >
> > > Is this happening even with the latest CURRENT driver,  there is code in
> > > it now that is supposed to stop the firmware from doing that, at least
> > > that was the theory :)
> >
> > We still see this at work.  We use this workaround in /etc/sysctl.conf:
> >
> > net.inet.ip.portrange.lowlast=665
> >
> > It seems that the em0 interface always snoops 623 looking for RCMP
> > packets for IPMI (or ASF).
>
> ===================================================================
> RCS file: /home/ncvs/src/sys/dev/em/e1000_82540.c,v
> retrieving revision 1.3
> diff -u -d -r1.3 e1000_82540.c
> --- e1000_82540.c       16 May 2007 00:14:23 -0000      1.3
> +++ e1000_82540.c       25 May 2007 13:40:19 -0000
> _at__at_ -316,6 +316,7 _at__at_
>         /* Disable HW ARPs on ASF enabled adapters */
>         manc = E1000_READ_REG(hw, E1000_MANC);
>         manc &= ~E1000_MANC_ARP_EN;
> +       manc &= ~(E1000_MANC_RMCP_EN | E1000_MANC_0298_EN);
>         E1000_WRITE_REG(hw, E1000_MANC, manc);
>
>         E1000_WRITE_REG(hw, E1000_IMC, 0xffffffff);
>
> Fixes it for me.  But, of course, I'm not interested in ASF on my
> gateway hosts.
>
> I didn't look further into the initalization of the chip.  The
> datasheet I downloaded suggests that you can filter these packets
> on IP address as well and I suspect the option is turned on with
> some default in the EEPROM register that equates to 0.0.0.0/0.

Hardcoding this change into shared code is not the right place
to do it, however I'll take a look at that and figure out a more
appropriate approach.

Jack