More questions about WINS, interfaces, etc

[Philip A. Prindeville]

> On Thu, 18 Sep 1997, Philip A. Prindeville wrote:
> 
> > I have yet another question about WINS and DNS...  Why can't the
> > DNS be used as the database of reference for name-to-address
> > mappings when using NB over TCP/IP?
> 
> we're working on that, for samba-2.

I would be interested in collaborating on that.

> >  Assuming you could easily
> > make incremental changes to the name server and propagate them
> > out to all of the secondaries, then why bother having nmbd (or
> > whatever) keep in-memory tables at all?
> 
> cacheing, to reduce network traffic.  also, nmbd is a single process.  
> we'd need to add a non-blocking, state-based way to resolve dns names.  
> (gethostbyname() is blocking...)

At least under Solaris, there is a thread-safe gethostbyname_r()
function that could be used.  I don't know how portable this is,
though.

> > On the subject of that, who would be responsible for adding the
> > name to the DNS:  Samba or the DHCP server?  What about
> > deletions?
> 
> the DNS administrator.  NOT samba.  and NOT the DHCP server.  samba has 
> nothing directly to do with DHCP daemons or named daemons.

I'm assuming that a machine, say that has a network identity of
"FIDO", gets an address lease from a DHCP server, and then tries to
defend the name via NetBIOS.  Having succeeded, the name should
probably be re-integrated into the DNS with the leased address.
Of course, this happens too often and too dynamically to involve
the intervention of a hostmaster.

> > I've modified source/interface.c to handle interface names instead
> > of addresses/netmasks...  I thought this might be nicer since if
> > you renumber a network or change a host's address, the fewer
> > places you have to change its IP address the better.  So with these
> > patches (I'm working with 1.9.16p11 -- sorry)
> 
> i don't care: the samba team deals with patches from different versions 
> all the time.  we prefer diff -u, if you can generate it.  patch < 
> diff_file usually deals with things automatically, and creates .rej files 
> for rejected bits, which have to be added by hand - no problem there.

Hmmmm...  My diff doesn't understand -u.  Where can I get one that
does?

> > you can say something
> > like:
> > 
> > 	interfaces = le0 le1
> > 
> > or
> > 
> > 	interfaces = all
> 
> oh, _good_!  an excellent addition.

I'm just not sure about one thing, however:  in the original code
they seem to ignore the broadcast address and force the address to
be the all-ones host address.  Which is what the broadcast address
should be anyway, unless you have misconfigured or incredibly old
machines.

> Luke Kenneth Casson Leighton (lkcl at switchboard.net)
> "Deal with difficult problems while they are still easy"

-Philip