[Dnsmasq-discuss] force --local/--server protocol

[Ed W]

On 09/01/2013 13:40, Mr Dash Four wrote:
>
>> d) dnsmasq returns the answer with the "truncated response" bit set to
>> the client, which then retries over TCP, and dnsmasq makes a TCP
>> connection to 10.1.1.1
>>
>> The net effect of d) is much the same as c), provided that client
>> behaves in the conventional way, so It may be a winner :-)
> Thanks, I'll seek to implement that, though I have another query: if 
> dnsmasq receives a request from a client, which is directed to dns1 
> and then receives a response from that server "not-implemented" what 
> happens next: does that error gets returned to the client, or is 
> dnsmasq forwarding the same query to dns2 and then forwards the 
> response to the client, provided it is OK?
>
> The reason for asking this question is this: I have one dns, which 
> only recognises type "A" records, and another one, which has 
> everything else (A, AAAA, CNAME, MX etc). I would like to use dns1 as 
> my primary dns server as it is blazingly fast and because the majority 
> of the requests from the client machines would be for type "A" 
> records, but would like queries for records other than type "A" to be 
> forwarded to the second server, which might be a bit slower, but 
> serves all types of requests.

Note I am a huge fan of dnsmasq, however, have you looked at unbound 
also?  I *think* it may have an option to force TCP upstream queries, 
it's not clear, but I would assume this means it receives on UDP and 
queries upstream on TCP (the documentation states "this may be helpful 
for tunnelling situations")

You could even incorporate both dnsmasq and unbound in order to have 
only a portion of your requests take this route? (another option is to 
write a simple DNS server in say perl and do whatever you wish, using a 
local dnsmasq/unbound/something else as your upstream proxy)

Ed W