samba - Oct 1997 - Disappearing clients in WINS list

On Mon, 20 Oct 1997, Dieter Rothacker wrote:
> Luke Kenneth Casson Leighton wrote:
>  
> > > Here is the smb.conf of the disappeared samba box:
> 
>                                 ^^^^^^^^^^^
> I do not think THIS computer should become local master !?
> (see below)
> 
> > >    wins support = no
> > >    workgroup = WORKGROUP
> > >    browse list = no
> > >    domain master = no
> > >    preferred master = no
> > 
> > here you go! set this to yes!
> > 
> > >    local master = no
> > 
> > and this!
> > 
> > >    os level = 0
> > 
> > this puts your samba server on a par with Windows for Workgroups
> > machines: _low_ priority.
> > 
> > check the Browsing.txt file etc (grep for "os level" in docs
directory).
> > 
> > put this to at least 4 (one above NT Workstation).
> 
> I am not sure if I explained my setup enough. I thought it would be best
> to let the WINS server be domain master AND local master on all its
> subnets, so that it can easily build its cross-subnet browse list.
ok, is the disappeared machine on the same broadcast-isolated subnet as 
the domain master?


 
> This WINS server
a WINS server cannot have NetBIOS names registered.

[you mean, the machine that is running samba, samba in this instance 
having been configured to run both WINS service _and_ be a domain master, 
which is a WINS client using itself for its WINS service...]
> has __MSBROWSE__<01> and workgroup<1b>, workgroup<1d>
> registered ALL the time, I check this regularly.
ok.
> This machine is the only one with direct connection to all subnets.
> Perhaps you could give me some insight why it could be advisable to
> let elections for local master happen at all ? I thought I do not need
> that with a WINS server that has direct connection to all subnets.
> 
> So the computer you are referring to above should register at this WINS
> server and should not try to collect & distribute browse lists.
> 
> The Windows clients I configured are also set to "Browse list"
> "deactivated" (or similar, I have German clients here).
ah, good.

ok.  can you do a netstat -a on one of the w95 machines for me, and let 
me know what the results are?

i need to know if they have WORKGROUP<1e> registered or not.  we've
just
made a small mod (when "local master = no", don't register
WORKGROUP<1e>.
i believe this to be faulty: i believe this to require a new parameter
"workgroup member = yes/no", which regulates whether
WORKGROUP<1e> is
registered or not). 


luke

Simon Hyde <shyde@poboxes.com> wrote:
> Almost there but not quite so simple, previously (1.9.16 and below) samba
> would connect to the password server as under the name of the client on the
How frustrating that no one responded quite so succintly about this
problem when I wrote about it on Oct 2.  Perhaps they did and I missed
it buried in a digest.  (And can't find it in the archives.)

> addresses so the other must be a ghost. This was changed in 1.9.17 to
> instead identify itself as a combination of the remote machine's
netbios
> name and the PID of the smbd process attempting to authenticate. This was
> considered the lesser of 2 evils since you can turn the errors off by
> removing station restrictions, but you can't remove the dropping of
> connections problem.
I found the patch [1] in the digest archives and I believe I understand
the problem now.  Is there something wrong with the logic of the Samba
server sending its own netbios name along with the username to 
be validated?  It would be a more acceptable solution in our 
environment to allow user 'logon' access to the Samba server than
to eliminate all access restrictions.

I've patched password.c to send the hostname returned by get_myname() 
when making a connection to the password server.  The reasons for
*not* doing this escape me, and I'm slightly annoyed that there wasn't
any quick note in the docs that access lists wouldn't work... but amen 
for source code.

Thanks to Mr. Hyde for pointing me in the right direction.  I've
included my somewhat-of-a-hack patch for others enjoyment.   

- edan


[1] http://samba.canberra.edu.au/pub/samba/digest/currentyear/0259.html


*** password.orig       Mon Oct 20 21:04:28 1997
--- password.c  Mon Oct 20 21:09:01 1997
***************
*** 1594,1606 ****
       connections if the same client name connects
       twice. Instead, synthesize a name from our pid.
       and the remote machine name.
     */
    {
!     char buf[32]; /* create name as PIDname */
!     sprintf(buf,"%d", getpid());
!     strncpy(&buf[strlen(buf)], remote_machine, 31 - strlen(buf));
!     buf[31] = '\0';
!     DEBUG(1,("negprot w/password server as %s\n",buf));
      name_mangle(buf,p,' ');
      len += name_len(p);
    }
--- 1594,1608 ----
       connections if the same client name connects
       twice. Instead, synthesize a name from our pid.
       and the remote machine name.
+ 
+      No, how about just send *our* hostname to the password
+      server?  That way we can still have access lists that work.
+      Edan Idzerda <edan@mtu.edu>
     */
    {
!     char buf[32]; 
!     get_myname(buf, NULL);
!     DEBUG(2,("negprot w/password server as %s\n",buf));
      name_mangle(buf,p,' ');
      len += name_len(p);
    }