Re: 6.0.1 upgrade dhcp problem

[Robert Elz <kre%munnari.OZ.AU@localhost>]

    Date:        Fri, 1 Feb 2013 17:21:59 +1000
    From:        Ray Phillips <r.phillips%uq.edu.au@localhost>
    Message-ID:  <p05200f0ecd3111466bc3@[192.168.37.112]>

  | These commands were executed after using 6.0.1's  e: Utilities -> c: 
  | Configure network  menu item:
  | 
  | # ifconfig fxp0
  | fxp0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
  |          capabilities=1400<TCP4CSUM_Rx,UDP4CSUM_Rx>
  |          enabled=0
  |          address: 00:10:dc:7b:92:a9
  |          media: Ethernet autoselect (100baseTX full-duplex)
  |          status: active
  |          inet6 fe80::210:dcff:fe7b:92a9%fxp0 prefixlen 64 scopeid 0x3

That's just an unconfigured (but enabled) interface, it has no address
at all.

  | Configuring the NIC manually was successful:

This means that the kernel & hardware is all fine, it is a problem with
dhcpcd and its scripts.

  | # ifconfig fxp0 inet 192.168.36.135/23 up
  | # route add default 192.168.36.30
  | route: writing to routing socket: File exists
  | add net default: gateway 192.168.36.30: File exists

What's more, those errors indicate that the default route was correctly
added previously, and is still there.   So it is just the interface
configuration that's failing.

It is also interesting, because to configure the default route we would
need a route to its destination.  That route is an interface route that
would get installed when the interface is configured (and removed when the
interface configuration vanished).

This indicates that the interface configuration from dhcpcd probably worked
when it was initially done - then something has removed it (it is kind of
a netbsd bug that routes that refer to an interface that has been unconfigured
don't just go away - it is also a kind of a feature - here it helps with
diagnostics, in that we have information because of that behaviour.)

martin%duskware.de@localhost said:
  | FWIW, I just downloaded an official NetBSD-6.0.1-i386.iso and booted it in a
  | virtual box. The DHCP network configuration just works for me, and the
  | scripts are present. 

Yes, Ray said previously that the came CD he is using works correctly on
a different system (what's more, one with more or less the same interface
hardware).

It can't be that the CD is no good (it works elsewhere) or that the CD
drive is no good (it works for manual config), so there must be some other
difference.

I wonder if perhaps the system with the problem is connected to an
ethernet switch that has spanning tree enabled - the switch might start
the spanning tree algorithm when the port is enabled (but allow the
DHCP packets to pass through) and then reset the port after the algorithm
timeout - when reset the driver could be removing all the configuration
info.   That could be a difference between dhcpcd and dhclient - the former
might be making more interface resets happen.

This is all wild speculation - none of it even sounds plausible to me, but
I am running out of rational explanations to offer.

As a next step, I think I'd try running dhcpcd manually (after sysinst
has failed to correctly configure the network - just as you did the manual
configuration this time), but have an ifconfig loop running in the
background.

So get through sysinst (doesn't really matter if you have it attempt to
configure the network or not), get a sh prompt

        while sleep 1; do ifconfig fxp0; done &
        dhcpcd fxp0

and see if you can observe anything from the ifconfig outputs, as dhcpcd
is running.   I'm kind of expecting that you will see the IP address appear,
and then later go away again.

kre