Configuration question..

Keith.Neufeld at wichita.edu Keith.Neufeld at wichita.edu
Tue Sep 12 18:04:24 UTC 2006


> with this config I get a WARNING now when I restart my dhcpd..

Yes, code has recently been added to the server to produce a warning about 
this very topic.

David Hankins:  Was the warning possibly added for hosts inside subnet 
declarations, but inadvertently overlooked for hosts within pool 
declarations?


> I have a host dilvish in my 192.168.1.0 subnet::pool,
> I move that to the 10.0.0.0 subnet::pool, and remove the routers 
option..
> 
> restart dhcpd, have the win2k box relase and then renew it's IP, it gets 

> the same 192.168.1.250 IP but it gets all the things associated w/ the 
> 10. (I think this is what you were saying would happen.. )

Yes.  The "gets all the things associated with the 10" is a part that has 
only recently become widely understood.  See below for my attempt at 
explaining it further.

> When I comment out the entry (now existing only in the 10 subnet) and 
> then restart dhcpd and then have the client renew it gets an ip in the 
> 192.168.1.22 range.  (I am not sure if you said this would happen.. )

I didn't mention that, but that's exactly right.  It's no longer "known," 
so it gets an address from your unknown pool.

> This is understandable.. (sort of) as this host is coming to the dhcpd 
> server on the 192.168.1.x subnet, and like you said "..known (listed 
> anywhere in the configuration).. ".

Yes, that's it.  "known" means there's a host declaration for it 
_anywhere_.  "unknown" means there isn't.


> What I am looking for is that when hosts come in through the dhcp-relay 
> on the router that they get routed to the right subnet and follow the 
> correct known/unknown settings that I have provided for them.

The DHCP server automatically handles matching the relay's gateway address 
to the subnets you've defined in order to give an appropriate address to 
the client.  You don't have to do anything special, other than fully and 
accurately describe your subnet and shared-network layout to the server 
(including the subnet(s) the server is attached to, *even if it's not 
serving any clients there*).  The known and unknown configurations, you 
also seem to have a good handle on.


Okay, here's an attempt at explaining where to put different configuration 
settings.

*       If you want settings to apply to all clients everywhere, at least 
as default values that can be overridden elsewhere in the configuration, 
place those settings in the global scope, outside of all other blocks.

*       If you want settings to apply to all clients on a VLAN (again, at 
least as a default for that VLAN), place those settings in the 
shared-network declaration for that VLAN, outside of the subnet 
declarations.

*       If you want settings to apply to all clients in a particular IP 
subnet (which may coexist with other logical subnets on the same VLAN), 
place those settings in that subnet declaration (which may be inside a 
shared-network declaration along with other subnet declarations, or may 
stand alone if the logical subnet is alone in that VLAN).

*       If you want settings to apply to _some_ clients *when they operate 
in a particular IP subnet,* create a means to identify those clients 
(known/unknown, class/rule, class/subclass) and pools that allow/deny 
clients identified in that fashion, then place those settings in the pool 
declarations that allow those clients.

*       If you want settings to apply to some clients *no matter where 
they operate,* either place those settings inside the host declarations 
for those clients (if there are only one or two such clients), or create a 
group declaration enclosing those clients' host declarations and place the 
settings inside the group declaration.

*       Always place host declarations at the global scope, outside of all 
other blocks, except possibly for group declarations enclosing the host 
declarations. (+)


(+)  The recent thread on the mailing list about host declaration 
placement has illuminated the behavior that a host declared inside a 
subnet or pool declaration inherits attributes from the subnet/pool even 
when it travels elsewhere.  Thus carefully-crafted configurations may take 
advantage of this, in contrast with my suggestion always to place 
host/group declarations at the global scope.  However, this inheritance 
mechanism is IMHO more likely to cause confusion for the casual reader, 
and the consensus seems to be that it's simpler and safer to declare 
hosts/groups globally.

-- 
Keith Neufeld
Lead Network Engineer
Wichita State University


More information about the dhcp-users mailing list