bind 8.1.2 vs bind 8.2.2 on SunOS 4.1.X

[Ian G Batten]

[[ Platinum Beta people, I think the Sun DNS people were talking about
shipping Bind 8.latest with Solaris 8, and 8.2.2 is the latest.  I
suspect that Solaris machines are disproportionately likely to be on a
net with a 4.1.X machine... ]]

I've taken my eye off the bind ball over the past twelve months, but I
need dynamic update so I've started in on moving to 8.2.2 from (largely)
8.1.2.  My Auspex doesn't like it: the ancient resolver that ships with
SunOS 4.1.4 and is used by ypserv appears to choke.

My Auspex is called ``kevin.ftel.co.uk''.  I'm using a resolv.conf that
lists nameserver 127.0.0.1, domain ftel.co.uk and has the `hosts' yellow
pages map empty to redirect through to the DNS.

Using 8.1.2, ``ypmatch foo hosts'' and ``ypmatch foo.ftel.co.uk hosts''
both work.  Using 8.2.2, if I have the domain line in resolv.conf, I can
only look up unqualified names, with it out I can only look up fqdns.

Bumping the resolver tracing up a bit, for the case of ``lookup
foo.ftel.co.uk, domain set to ftel.co.uk'' I see a query for
``foo.ftel.co.uk.ftel.co.uk'' come in, for which a no is sent back, but
I then don't see the expected foo.ftel.co.uk.co.uk and so on.  Under
8.1.2, I do.  I'm lost at this point, but I presume there's a difference
in the nature of the null response which causes the resolver to give up.
For a non-existant foo, I think the pattern is clearer.

8.1.2:

Version = named 8.1.2 Wed Nov 18 15:24:54 GMT 1998
        igb at kevin.ftel.co.uk:/export/staff-home/igb/src/networking/bind-8.1.2/src/SunOS-4.1.4-aushp/bin/named
conffile = /etc/named.conf
datagram from [127.0.0.1].1086, fd 36, len 48
req: nlookup(djkfakjs.ftel.co.uk.ftel.co.uk) id 134 type=1 class=1
req: found 'djkfakjs.ftel.co.uk.ftel.co.uk' as 'ftel.co.uk' (cname=0)
ns_req: answer -> [127.0.0.1].1086 fd=36 id=134 size=110
datagram from [127.0.0.1].1086, fd 36, len 43
req: nlookup(djkfakjs.ftel.co.uk.co.uk) id 135 type=1 class=1
req: found 'djkfakjs.ftel.co.uk.co.uk' as 'co.uk.co.uk' (cname=0)
ns_req: answer -> [127.0.0.1].1086 fd=36 id=135 size=111
datagram from [127.0.0.1].1086, fd 36, len 37
req: nlookup(djkfakjs.ftel.co.uk) id 136 type=1 class=1
req: found 'djkfakjs.ftel.co.uk' as 'ftel.co.uk' (cname=0)
ns_req: answer -> [127.0.0.1].1086 fd=36 id=136 size=99
Debug off

Now, I run 8.2.2 without touching any other files:

Debug level 1
Version = named 8.2.2-REL Wed Oct 27 16:56:40 BST 1999
        igb at kevin.ftel.co.uk:/export/staff-home/igb/src/networking/bind-8.2.2/src/auspex/bin/named
conffile = /etc/named.conf
 [[ much select debugging deleted ]]
datagram from [127.0.0.1].1118, fd 36, len 54
req: nlookup(dskjfhasdkjdhk.ftel.co.uk.ftel.co.uk) id 142 type=1 class=1
req: found 'dskjfhasdkjdhk.ftel.co.uk.ftel.co.uk' as 'ftel.co.uk' (cname=0)
ns_req: answer -> [127.0.0.1].1118 fd=36 id=142 size=54 rc=2
datagram from [193.123.211.5].1087, fd 22, len 37

And that's it: game over from the resolver's point of view.

Any thoughts from anyone?  As things stand, this blocks a move to 8.2.2
until I get SunOS 4.1.4 out of my network (which is happening in a few
weeks) but I also need to test against Solari with old resolvers (I have
production 2.4 machines).

ian