[Dnsmasq-discuss] dnsmasq giving new addresses despite of leases file

[Simon Kelley]

Leonardo Rodrigues wrote:
> Em 08/03/2010 12:20, Simon Kelley escreveu:
>>>      shouldnt dnsmasq provide the IP present on its lease file, given the
>>> fact that lease was still valid ???
>>>
>>>      
>> Good question: I can give you a partial answer, the lease gets deleted
>> as part of the response to the first DHCPREQUEST. Here's the code in
>> question.
>>
>>                  {
>>                    message = _("wrong address");
>>                    /* avoid loops when client brain-dead */
>>                    lease_prune(lease, now);
>>                    lease = NULL;
>>                  }
>>
>> What I can't answer is what the "brain-dead" behaviour is/was that's
>> getting countered here. The "lease prune" call was added in version
>> 2.41, but there's no explanation in the changelog :-(  and I can't
>> remember why it was added.
>>
>> I suspect that some client somewhere kept trying with the wrong address
>> instead of falling back to a DHCPDISCOVER.
>>
>>    
> 
> 
>      Hi Simon,
> 
>      I've checked logs and found no other 'wrong address' cases than 
> this one. In fact all machines on this network are desktops and this one 
> which is giving the problem is the only notebook on the network.
> 
>      anyway, do you think enabling the '--log-dhcp' option can provide 
> some useful information ? Is there any way of getting debug logging 
> enabled AND going to background ? I've found the debug options, but 
> seems all of them keep process in the foreground, which unfortunelly is 
> not possible to me, as this is a small device running OpenWRT, i have no 
> monitor/keyboard attached to this machine.
> 
>      anyway, do you think debug modes could provide some useful 
> information ?
> 
I understand what is happening, what I'm less sure of is why that code
to delete the lease was added: without it, the behaviour would be
exactly what you want (and seems quite correct).

The code in question was added about 2 and half years ago, and I've now
found the relevant emails in my archive. It seems that the "braindead
client" was a particular buggy release of dhcpcd, which is long-ago
fixed. It should therefore be safe to remove the lease deletion code again.

Can you build new dnsmasq packages from source for your WRT box? If so
I'll get a test release to you with the changes implemented.

Cheers,

Simon.