DHCP server 'leaking' across server machine

[Andy Rabagliati]

Folks,

  I have a difficult problem that is a problem to troubleshoot because I
  am not on location - so apologies in advance for some lack of detail.

  The setup is a school lab environment here in South Africa.

  It is a fairly standard thin client LTSP setup, with a fat server
  managing about 20 thin clients.

  eth0 (172.16.60.0/24) on the server points towards the lab, and runs
  ISC dhcpd on Ubuntu dapper, with a tftp server and an X server to back it up.

  eth1 (192.168.8.0/24) is a second interface on the server, and is only
  connected to a wireless router, that I am in the process of using to
  provision Email and other services.

  The wireless router runs a DHCP server, dnsmasq. I do not have
  administrative access to this router, though I can coordinate with
  someone else and get it changed.

  * With no default route set, the lab boots and runs local services
    just fine.

  * As soon as I add a default route via eth1 to 192.168.8.1, the lab
    will not boot. Logs indicate that the DHCP server sees the
    DHCPDISCOVER, DHCPOFFER, and even the DHCPREQUEST, DHCPACK, but the
    error reported on the client is that no offer was received.

    Details are a little sketchy, as I am talking by phone to the lab
    manager, who is not a unix admin, but undoubtably the lab does not
    work.

  * If I add a static route to the server, enough for me to access the
    box remotely, the lab also works fine.

  To summarise, this setup works :-

root at tuxlab:~# route -n
Kernel IP routing table
Destination     Gateway         Genmask         Flags Metric Ref    Use Iface
41.206.193.0    192.168.8.1     255.255.255.0   UG    0      0        0 eth1
172.16.0.0      192.168.8.1     255.255.255.0   UG    0      0        0 eth1
172.16.60.0     0.0.0.0         255.255.255.0   U     0      0        0 eth0   <<-- Lab
192.168.8.0     0.0.0.0         255.255.255.0   U     0      0        0 eth1


  And this setup does not :-

root at tuxlab:~# route -n
Kernel IP routing table
Destination     Gateway         Genmask         Flags Metric Ref    Use Iface
41.206.193.0    192.168.8.1     255.255.255.0   UG    0      0        0 eth1
172.16.0.0      192.168.8.1     255.255.255.0   UG    0      0        0 eth1
172.16.60.0     0.0.0.0         255.255.255.0   U     0      0        0 eth0   <<-- Lab
192.168.8.0     0.0.0.0         255.255.255.0   U     0      0        0 eth1
0.0.0.0         192.168.8.1     0.0.0.0         UG    0      0        0 eth1   <<-- Default route


  I suspect the two DHCP servers are interfering with each other, and the clients
  are getting duff information from the other DHCP server.

  Is this possible ?

  The invocation line is as follows :-

  start-stop-daemon --start --quiet --pidfile $DHCPDPID \
	-exec /usr/sbin/dhcpd3 -- -q $INTERFACES -pf $DHCPDPID -cf $CONFIG_FILE

  And I set INTERFACES to eth0, and I even put an empty subnet declaration for
  192.168.8.0/24, to no avail.

  /proc/sys/net/ipv4/ip_forward says 0, if that is any help ..

  I could ask the ISP to disable DHCP on (all 60) wireless routers, but it
  is nice to have things like routing an nameservers delivered from the
  router rather than being statically set.

  I get a little lost on networking when dealing with UDP broadcasts, and I think
  this is my problem area.

  Is there anything I can do to have my cake and eat it too ?

Cheers,    Andy!