[Samba] nbp hell

[Andrew Bartlett]

On Mon, Dec 29, 2003 at 08:36:30PM -0500, Romeyn Prescott wrote:
> Greetings,
> 
> I spent the better part of a day tracking down a rather annoying 
> problem.  I would like someone more knowledgeable than myself to 
> confirm this diagnosis.
> 
> I have a RH 8.0 Linux box with three NICS.  I had been developing it 
> testing with an XP box.  Everything seemed groovy.
> 
> Today I went to put it into "production" and Win98 clients were 
> having all sorts of problems.
> 
> eth2 is the system's "default" interface.  eth1 is used for netatalk, 
> and eth2 for samba.  I have used the interfaces and nind interfaces 
> only options in smb.conf.

You have 3 interfaces onto the same LAN?  You fill find that you are
only really using one of them, as the system will accept ARP requests
on all for all IPs.

> After more hairpulling, I did a packet capture on a hub with the Win98 client.
> 
> It seems that the client does an nbp query.  My server responds, but 
> the response comes from eth2, not eth0!  

Sounds standard for UDP.

> Despite the packet's payload 
> having the correct information (that the server the client seeks is 
> at the IP address bound to eth0), Win98 decides that the server is at 
> the IP address associated with the packet informing it of such.
> 
> I am told that this is a violation of the smb protocol.
> 
> Can anyone suggest a fix?  Should I just relegate samba to the 
> server's "default" interface and not worry about this anymore?

Ditch the multiple network cards, if they are to the same LAN.  You
would do much better to upgrade to Gigabit if you really need it, or
make them 'one card' with channel bonding and an fancy switch.

Andrew Bartlett