cross-subnet browsing problems

[Luke Leighton]

>On Thu, 24 Jul 1997 mac at nibsc.ac.uk wrote:
>
>> I think the problem is very simple.  The cross-subnet browsing of Samba
>> only support one workgroup.  (The one the samba server is in).

yes, however if there are other samba servers or NT servers (or,
basically, any domain master browsers of _any_ kind) around, the lists
forwarded to all hosts, including these other DMBs, contain other
workgroups.

this encourages other domain master browsers to propagate workgroups
across all subnets, which in turn will get propagated to all browse
clients, which will enable users to _find_ the domain master browsers
for those other workgroups.

NOTE: if there _isn't_ a DMB for a workgroup propagated across all
subnets, you will find that you get a 30 second timeout.  this occurs in
ANY tcp/ip multi-subnet environment, and would and does occur in an
NT-only environment.  in other words, it's something that you, as a
network administrator, need to sort out, regardless of whether you're
using NT or samba.

NOTE: samba-1.9.16 does _not_ propagate workgroups that it finds to
other DMBs.  it only informs people about _its_ workgroup.  it is quite
uncooperative in this respect.


>> 
>> A 'quick and dirty' solution is therefore to put all your machines (clients
>> and servers) in one workgroup, and all machines will appear.

this is necessary, rather than a quick and dirty solution.  if you want
workgroups to be propagated across several subnets, you will one unique
DMB (regardless of whether they are samba or NT server or other) for
each workgroup.
 
>Both Samba servers and the machines I'm interested in are in the same
>workgroup.  Right now, it looks like:
>(diagram shamelessly ripped from BROWSING.txt)

oh, _good_ :-)  documentation that's _useful_!

>           All machines in the diagram are in the workgroup SLING
>
>                                                    (DMB/LMB - samba)
>             (NT)       (NT)        (NT)       (NT)       (WINS)
>            sling1     sling2      sling3     sling4    slingblade
>              |          |           |          |           |
>          -------------------------------------------------------
>                          |          subnet 1 (158.187.108.0)
>                        +---+
>                        |R1 | Router 1
>                        +---+
>                          |
>                        +---+
>                        |R2 | Router 2
>                        +---+
>                          |
>                          |  subnet 2 (158.187.46.0)
>                --------------------------
>                |           |
>            davisson      blalor
>             (W95)     (LMB - samba)
>
>
>On subnet 1, the workgroup SLING appears with the machines sling1-4 and
>slingblade.  On subnet 2, the workgroup SLING appears with the machines
>davisson and blalor.  I had originally said that sling1-4 appeared in
>browse.dat on both machines.  As I look this morning, these 4 machines do
>not appear in either machine's browse lists (browse.dat).  Every 15
>minutes, slingblade contacts blalor, presumably to swap browse
>information.  

blalor sends slingblade a "LocalMasterAnnounce" packet, and slingblade,
in response, does a NetServerEnum2 call to it, to grab slingblade's
browse list.    blalor, at some small time later, should contact
slingblade, to grab the compiled browse list, which should include the
entries it's just forwarded to slingblade.   

so the action is initiated by blalor, but should result in this double
NetServerEnum2 transfer by slingblade and then blalor.  if you're _not_
seeing a two-way transfer, initiated by blalor's "LocalMasterAnnounce"
packet, then this is the problem.  

ok.  try switching blalor _off_ as an LMB, and see what happens.  do
this by setting "local master = no", or by simply killing it.

do some network neighbourhood happy-clicking on SLING, see if you get a
full list.

leave it for 15 minutes, see if davisson contacts slingblade to get a
full domain list (keep the packet traces going!).

leave it for about a further 30 minutes, see if every 15 mins davisson
still contacts slingblade on a 15 minute basis.

actually, it should be every 12 minutes...

>Since davisson is currently off, I should at least see
>blalor in the Network Neighborhood of sling4, but I'm not.
>
>Any suggestions?

sodding browsing.  it's _chewing_ developer time.

are all clients, _including_ the samba server blalor, set to "wins
server = ip_address_of_slingblade"?  slingblade has "wins support =
yes", and does not have "wins server = its_own_ip_address"?

do an nbtstat -a on _all_ the machines.  check the status of each and
every machine:

SLING           <00> - SMB client (i think)
SLING           <20> - SMB server (i think)
SLING           <1e> - workgroup member
>^1^2__MSBROWSE^2<01> - backup local master browser
>SLING           <1d> - local master browser
>SLING           <1c> - domain logon server (backup domain controller)
SLING           <1b> - domain master browser (primary domain controller)


what i'm expecting is:

- blalor and slingblade to both have SLING<1d> - no-one else.

- all hosts to have <00>, <20> and <1e>.

- slingblade to have SLING<1c> and SLING<1b> - no-one else.

- a random distribution of the msbrowse backup name.  possibly.  doesn't
matter who owns this one, really, at the moment.  i hope.


then, do a network neighbourhood search on davisson and sling1.  then,
do an nbtstat -c on each of these machines.  this will show the netbios
name cache on davisson and sling1.

i expect:

- sling1 to have the ip address of slingblade in its cache

- davisson to have the ip address of both blalor and slingblade in its
cache.



what fun diagnosing browsing can be.  isn't that right, jeremy?

luke