Re: Resolver not always resolving hostnames

On 2006.10.27 02:56:00 +0900, Hiroki Sato wrote:
> "Jiawei Ye" <leafy7382_at_gmail.com> wrote
>   in <c21e92e20610240824n7e94e4c1jf454697cab4120a9_at_mail.gmail.com>:
> 
> le> On 10/24/06, Hajimu UMEMOTO <ume_at_freebsd.org> wrote:
> le> > leafy7382> This is much better. It would be great if this is commited.
> le> >
> le> > I've committed it into 7-CURRENT, and I'll MFC it into RELENG_6 after
> le> > 3 days.
> le> >
> le> > The fix should help in certain case.  However, it seems doesn't help
> le> > for nork-san's case.  He could reproduce it, and sent me his ktrace
> le> > output.  As far as I can see his output, his local name server (BIND
> le> > 9.3.2-P1) returned the response with no answer record.  As a result,
> le> > his csup ended up with following error:
> le> >
> le> >         Name lookup failure for "cvsup.jp.freebsd.org": hostname nor
> le> >         servname provided, or not known
> le> >
> le> > I suspect that the BIND9's named have some problem.
> le> >
> le> >        "Name lookup failure for "cvsup.jp.freebsd.org": hostname nor servname \
> le> >         provided, or not known
> le> >        "
> le> You are correct, I just ran into that a few times in the last hour.
> le> But it is a lot less frequent than before. I also run a local BIND9
> le> for our company. I think some commits happening after between Sept 30
> le> and Oct 15th caused it, because I did not see such symptom before late
> le> Sept and I rebuilt my system around mid Oct.
> 
>  According to log files on my public cvsup server which is suffering
>  from this problem, this happened on 22 September at least.  I have
>  not looked into the source files closely yet, but is it possible that
>  the 9.3.2-P1 patch has a bug that can return 0/0/0 with NOERROR
>  answer packet?  I confirmed that behaviors of 4.x's resolver and
>  6.x's against such a malformed packet are the same, so I think it is
>  not likely a bug in the resolver.

After I upgraded my development server to include the fix I started a
tcpdump of all my DNS traffic in case the resolving issue happened
again, and it did :-/.  So, the above fix does not fix the issue and
as hrs_at_ notes it actually seems to be a a bug in the nameserver and
not resolver.

When my latest cvsup failed this is the DNS packets captured:

15:06:45.036710 IP eddie.54684 > ns.nitro.dk.domain:  51917+ A? cvsup3.dk.freebsd.org. (39)
15:06:45.224002 IP ns.nitro.dk.domain > eddie.54684:  51917 0/0/0 (39)
15:06:45.224796 IP eddie.50804 > ns.nitro.dk.domain:  51918+ A? cvsup3.dk.freebsd.org.nitro.dk. (48)
15:06:45.226255 IP ns.nitro.dk.domain > eddie.50804:  51918 NXDomain* 0/1/0 (98)

I still have the pcap file around if it's needed.

The nameserver in question is running BIND 9.3.2 from 6.1-SECURITY
base.

-- 
Simon L. Nielsen