DNS (named) quirks/bugs/features on Solaris 2.6

[John Biggs]

Hello all.
I encountered an unusual symptom when we moved the primary master DNS
server
to Solaris 2.6. We have/had an *ancient* AIX box running internal DNS.
(I
apologize for not having exact version numbers for everything at the
moment,
but during the first rush of troubleshooting that wasn't relevant) I
know it
was bind 4.x on the AIX box because of the file layouts.
The symptom: when the new (solaris 2.6) box is on line, queries from
clients
(Win95) for unqualified hostnames, e.g. "ping server" fail. If they
try
"ping server.domain.com" or "ping 123.0.0.23" they are successful.
What's
odd about this is that if the Solaris box is swapped with the AIX box,
"ping
server" then graciously returns "reply from server.domain.com..."
My first response to the manager and my co workers was: "This is a
client
configuration issue, either they're missing the domain suffix
"domain.com"
in their IP configuration, or from the "suffix search order" options,
etc.
The clients obtain this config information from DHCP, and it *seems*
to be
there. This begs the question, however, of why it worked with the AIX
box!
Changing the client config wasn't an option as there are *hundreds* of
clients connecting from all across the country.
 
I know the configuation, zone files, etc. is OK, because:
    1. if you manually enter the "domain.com" suffix in the
appropriate location on the client, it works.
    2. nslookup queries *do* work properly, as do resolver requests on
the sun box!! ex.
            nslookup...server server.domain.com...set type=any...fubar
            response: fubar.domain.com        internet address =
123.0.0.10
...etc.
    3. The 2 things that "DNS and BIND" suggested to try, hostname
(change
from "hostname server" to "hostname server.domain.com") and verifying
contents of resolv.conf, made no difference.

To distill the above, it seems that our build/compile/configuration of
named
on Solaris 2.6 is relying on the client's resolver to append the
domain
suffix, rather than "trying" the name with the suffix and passing it
to the
client.

I've done a fair amount of DNS stuff (not that you can tell from this
post!)
but not much debugging, as it hasn't been required (yet)--and I
haven't seen
this before.

If anyone has seen this, or can offer any help,
thanks,

Lee