[Dnsmasq-discuss] CNAME trouble with no AAAA
simon at thekelleys.org.uk
Fri Oct 25 21:48:41 BST 2019
On 20/10/2019 17:55, Dominick C. Pastore wrote:
> I apologize for continuing the discussion on this. The patch (applied on top of 2.80-1 provided by Debian Buster) completely solved the issues I was having, but I did notice a couple other things.
> First, locally configured CNAMEs and records other than A or AAAA do not seem to play well together. For example, MX and TXT requests still get forwarded upstream, even after the patch. I played around with this a bit and discovered:
> 1. Unlike "host-record", "txt-record" and "mx-host" on the target are not enough to keep Dnsmasq from ignoring a locally defined CNAME. (I did not try others, like "srv-host".)
This is true, and difficult to fix for very obscure reasons. It should
be more explicitly documented, or better, fixed.
> 2. In fact, Dnsmasq never follows a CNAME for MX or TXT requests, even when the CNAME does point to a host Dnsmasq knows locally. (I assume this is the reason for #1.)
Actually it's not, it just that the CNAME code was never generalised to
handle stuff not in the cache. I've spent an enjoyable afternoon down
the rabbit-hole testing and rewriting, and this should be fixed now. The
prohibition on mixing local and upstream continues, but you can now
define a TXT/MX/SRV locally and a local CNAME pointing to it, and as
long as you define an A or AAAA record of the same name, all will be fine.
> Second, it seems that when Dnsmasq caches a NXDOMAIN response from upstream, it starts giving a NODATA response for other request types on the same name. Strangely, log-queries indicates the requests are forwarded, but right after a SIGHUP to clear the cache, sending one of the NODATA queries results in NXDOMAIN.
I can't reproduce this. Could you provide a simple example?
> Neither of these are actually causing problems in my case, but I suspect this isn't intended behavior either, so it seemed worth mentioning.
> On Sat, Oct 19, 2019, at 12:19 PM, Dominick C. Pastore wrote:
>> On Sat, Oct 19, 2019, at 6:16 AM, Simon Kelley wrote:
>>> The restriction still applies. indeed the patch relies on it.
>>> The origin of this is that, for architectural reasons, dnsmasq can only
>>> supply a reply which originates completely from locally known data, or
>>> completely from a reply from upstream. Since a local CNAME to a target
>>> in the public DNS necessarily has both, it's not possible.
>>> What the patch does, is allow a reply consisting only of the CNAME, of
>>> the target isn't locally defined for the type in question. This would be
>>> in error if the target was defined for the type in the public DNS, hence
>>> the condition disallowing that. The second version of the patch only
>>> does this for a locally defined CNAME, otherwise, you get the situation
>>> where a CNAME to a A record is cached from upstream, and then a query
>>> for an AAAA record on the CNAME name returns just the CNAME, rather than
>>> sending it upstream, because the AAAA record for the CNAME target it not
>>> currently cached.
>>>> I ask because in the former case, that could mean Dnsmasq would send a
>>>> NODATA reply if the target only exists in public DNS, correct? I'm not
>>>> familiar enough with the intricacies of DNS to know if that would
>>>> cause a problem for clients.
>>> Such a reply could, in theory, cause a client to cache the Nodata status
>>> of the CNAME target, whereas, if it queried the target directly, it
>>> would get public data. A cabeat about that should, possibly be added to
>>> the current disclaimer :(
>> Thanks for the detailed explanation. That makes sense and sounds very
>> reasonable. Thank you for all the help!
> Dnsmasq-discuss mailing list
> Dnsmasq-discuss at lists.thekelleys.org.uk
More information about the Dnsmasq-discuss