WINS multihomed registration debugging

[Kevin (HxPro) Wheatley]

HI,

I've got a slight problem with my samba setup here, I'm trying to debug
exactly what is happening when some of our machines register with the
WINS server (samba PDC).

I belive the problem is related to registering an interface address that
can only do point-to-point connections (HIPPI) or those which the WINS
server doesn't have a route to. I've tried limiting the interfaces
registered by the client using:

        name resolve order = host lmhost wins
        interfaces = ef0 fa0 lo0
        bind interfaces only = yes

I've also attempted to stop and start the WINS nmbd processes, but the
WINS server only has the following for the client:

        Name = SHIVA<00>        Source = REGISTER_NAME
b_flags = 44    death_time = Sat Jan  5 14:45:50 2002
        refresh_time = Wed Jan  2 15:36:34 2002

                number of IPS = 1       10.x.y.z

Which is the address of the hippi card...

Other machines that do not have the HIPPI card but are multihomed appear
in the database correctly.

        Name = ALBERT<00>       Source = REGISTER_NAME
b_flags = 44    death_time = Sat Jan  5 18:08:27 2002
        refresh_time = Wed Jan  2 18:28:27 2002

                number of IPS = 2       10.a.b.c    192.168.d.e

I've looked at packet traces and the samba nmbd running on shiva only
sends multi-homed requests for the two expected addresses and not the
hippi one, but the WINS server still has it in the db, I then noticed
that wins.dat only has the single IP address registered in it so I shut
down the nmbd; moved the file and started it up again, and now it's
fine...

Details

IRIX 6.5.14f based samba 2.2.2 release from the binary on the samba
site.

Questions

1) Is there a better way to force the removal of the WINS entries
without this stop/move/start ? 

2) Any ideas why the HIPPI interface address prevented the other
interfaces from being in the WINS database?

3) Is there scope for a BIND 'sortlist' style address sorting of the
returned addresses from the WINS server? i.e. more than simply matching
based upon the source address of the query (which appears to happen for
directly attatched clients at least calling nmblookup with different
interface addresses on the command line at least - not tested on a real
Windows machine yet...).

shiva 136# /usr/samba/bin/nmblookup -U atm-bonky -R -T 'shiva'
querying shiva on 192.168.X.X
Got a positive name query response from 192.168.X.X ( 192.168.X.X
10.123.X.X 10.66.X.X )
atm-shiva.london.cinesite.com, 192.168.X.X shiva<00>
shiva.london.cinesite.com, 10.123.X.X shiva<00>
gige-shiva.london.cinesite.com, 10.66.X.X shiva<00>

shiva 137# /usr/samba/bin/nmblookup -U bonky -R -T 'shiva'
querying shiva on 10.123.X.X
Got a positive name query response from 10.123.X.X ( 10.123.X.X
10.66.X.X 192.168.X.X )
shiva.london.cinesite.com, 10.123.X.X shiva<00>
gige-shiva.london.cinesite.com, 10.66.X.X shiva<00>
atm-shiva.london.cinesite.com, 192.168.X.X shiva<00>


If anybody has anything they would like me to do to test/try this then
I've probably got a day or so before we are back in full production. We
have several machines with the problem of not totally registering

Thanks

Kevin

-- 
| Kevin Wheatley             | These are the opinions of nobody   |
| Technical Services Manager | and are not shared by my employers |
| Cinesite Digital Studios   |                                    |