A SAMBA gotcha I think should be included in docs and/or technical change

Chris and Val Johnson cjohnson at mint.net
Wed Dec 2 03:38:05 GMT 1998


I installed linux and samba on a new server using a test IP address,
configuring it to be domain logon server and wins server.

I reconfigured eth0 with the production IP address, fixed /etc/hosts,
hosts.allow in smb.conf, etc.

Samba worked fine for unix smbclient and smbfs, but win95 fails domain
logon.  Cancelling past domain logon I find that "nbtstat -a servername"
also fails but "nbtstat -A server_ip_addr" works. And ping and telnet
work. Seems that nmbd is not right but it's running and debugging doc
tips work through step 4 but lead nowhere useful in step 5.  Wanted a
packet trace but no tool handy.

Spent hours examining everything that might be wrong compared to known
good samba implementation on an old server even trying a back-level
samba to no avail. Search FAQs and docs.  Search dejanews.  Nothing.

Finally I set debug level = 3 and start/stop samba.  Examining log.nmb I
find wins_initialise of servername has the test IP address.  I
double-check all network configuration files wondering where that is
coming from.  Eventually I discover the /var/lock/samba/wins.dat that
has the old address.  Apparently samba does not pick up the new address
of the local server name on its own.  I delete wins.dat and repeat the
start/stop to get a new log.nmb.  All addresses are right and after
turning off debug level and starting samba everything works.

SOOO... Whenever your samba is a wins server and you change its IP
address you must delete /var/lock/samba/wins.dat too.

Perhaps nmbd should recheck local name in wins cache during
initialization.  Perhaps this should be added to the doc.  In any case I
hope I can save someone else a few hours figuring this one out.

Chris



More information about the samba-technical mailing list