[Dnsmasq-discuss] No new lease for Option 82 requests until old one times out

[Kincl, Jason C.]

Hi,

Sorry for bringing up such an old thread but I was circling back onto this problem and I wonder if we could revisit it.

> On 17/09/14 09:49, Joachim Nilsson wrote:
>> Hi Simon,
>> 
>> I've found a little problem with how Option 82 circuit-id/remote-id
>> works. Everything is fine in the below setup until I replace the client
>> with a replacement unit that has a different MAC.
>> 
>>    [client]---LAN1---[dhcrelay]---LAN2---[dnsmasq]
>> 
>> dhcp-range=subnet0,tag:!static,192.168.2.100,192.168.2.199,255.255.255.0,864000
>> 
>> 
>>    dhcp-circuitid=set:cid0,"Eth6"
>>    tag-if=set:static,set:tag0,tag:cid0
>> dhcp-range=tag:tag0,192.168.2.99,192.168.2.99,255.255.255.0,864000
>> 
>> When 'client' is replaced the request from the new client reaches
>> dnsmasq, which responds with "no address available".  I figured this is
>> because the "pool" for the static IP only has one entry, bound to the
>> old client's MAC.  Indeed, it is not until the lease for the old client
>> times out that the new client receives an offer. I guess this behavior
>> is by design ...
>> 
>> I was thinking that adding 'tag:tag0' to the --dhcp-host setting would
>> have been perfect for Option 82, since they're basically static leases
>> anyway -- the same port on a given switch should always receive the same
>> IP ... so adding tag support, in addition to the already existing mac
>> and client-id, seemed at least to me useful.
>> 
>> What do you think?
> 
> Allowing dhcp-host matching by tag is a good idea, but it's not the
> complete solution to this problem. You've spotted the catch, which is
> that the address you want is already bound to a different MAC/client-id.
> 
> If you could select a dhcp-host line by tag, ie
> 
> dhcp-circuitid=set:cid0,"Eth6"
> dhcp-host=tag:cid0,192.168.2.99
> 
> then when the MAC address changed, you just get a different error:
> 
> "not using configured address 192.168.2.99 because it is leased to <old
> MAC address>"
> 
> 
> To make this work, you'd need some extra semantics, either explicit or
> implicit, to enable the old binding to be abandoned. Abandoning a
> binding is dangerous, since when it granted the lease, the server was
> promising the client exclusive use of the IP address. There is precedent
> for this: the form of dhcp-host which has more than one MAC address
> allows exactly then abandonment of a lease to one MAC address in favour
> of the other one.
> 
> It would be possible to define the dhcp-host=tag:cid0,192.168.2.99 form
> as having the same properties, or require some special keyword to enable
> this behaviour.

What about dhcp-host=*,tag:cid0,192.168.2.99? This syntax already exists for disregarding the client id and relying only on the MAC address only and it is similar to what we are saying, disregard the MAC and match on tag only. This should give us enough to be explicit about abandoning the old binding. 

What do you think?

> 
> I think this covers Neil's points as well, part from the PXE one. There
> is already some logic that tries to do the right thing when the same MAC
> address sometimes presents a client-id, and sometimes doesn't. It
> doesn't solve the case of two different client-ids at differnet points
> in the netboot, AFAIK.
> 
> 
> Cheers,
> 
> Simon.
> 
> 
> 
>> 
>> Regards
>> /Joachim
>> 
>> 
>> _______________________________________________
>> Dnsmasq-discuss mailing list
>> Dnsmasq-discuss at lists.thekelleys.org.uk
>> http://lists.thekelleys.org.uk/mailman/listinfo/dnsmasq-discuss
>> 

Thanks,

Jason Kincl