Host declarations with fixed-address treated as unknown for dynamic pools?

[Ingen Schenau, Jeroen van (ICTS)]

Hi all,

We're in the process of replacing our old DHCP servers (dhcpd 3.0.7)
with a newer version (4.2.1). While testing with our config, there seems
to be a difference in behaviour that I don't understand...

We have various subnets defined, where we assign a certain amount of
space to a failover pool and leave the rest for host declarations with
static assignment. eg (all addresses fictional):

subnet 192.168.1.0 netmask 255.255.255.0 {
        option routers 192.168.1.1;
        pool {
                deny dynamic bootp clients;
                deny unknown-clients;
                failover peer "dhcp-failover";
                range 192.168.1.128 192.168.1.254;
        }
}

subnet 192.168.2.0 netmask 255.255.255.0 {
        option routers 192.168.2.1;
        pool {
                deny dynamic bootp clients;
                deny unknown-clients;
                failover peer "dhcp-failover";
                range 192.168.2.128 192.168.2.254;
        }
}

host host1 {hardware ethernet 00:a0:73:de:ad:01; fixed-address
192.168.1.2;}
host host2 {hardware ethernet 00:a0:73:02:02:02; fixed-address
192.168.2.5;}
host dyn1 {hardware ethernet 00:a0:73:fe:dc:ba;};


With our current version, all three hosts will get an address, no matter
which subnet they're in. "host1" will get its fixed-address in the first
network if it's active there and a dynamic assignment if it's in the
second subnet; for "host2" it's exactly the other way around and "dyn1"
will always get a dynamic assignment.


Now when I run the same config with DHCPd 4.2.1, "host1" in the example
doesn't get an address if it's in the 192.168.2.0/24 subnet and "host2"
doesn't get an address if it's in the 192.168.1.0/24 subnet. In those
cases, both DHCP servers log something like "DHCPDISCOVER from
00:a0:73:02:02:02 via 192.168.1.1: unknown client".

This was quite unexpected and I'm wondering: can anyone tell me whether
this is a bug or an intentional change in behaviour? Has the definition
of a "known" client been changed in a way that it can be relative to the
subnet?


Regards,

Jeroen van Ingen
ICT Service Centre
University of Twente, P.O.Box 217, 7500 AE Enschede, The Netherlands