Samba troubleshooting tips and techniques

[Andrew Tridgell]

> One of the chapters I am responsible for in the book I am helping with is
> on troubleshooting.

much like Diagnosis.txt

> 1. Is Samba running?
> 
>     ps ax | grep mbd # look for nmbd (possibly 2) and 1+n smbd

won't work on sysV systems. You are better off running netstat to see
that you are listening on the 3 ports (UDP/137, UDP/138 and
TCP/139). That will also work with inetd installations.

For example "netstat -a | grep netbios" 

> 2. Is the problem with Samba, the client, or the network?
> 
>     standard network debugging stuff here, ping from client to server
>     if all works, is name server working OK, is WINS configured correctly,
>     and so on.

"standard network debugging" is what users have trouble with!

You may need to lay out specific steps that build on the previous steps.

>     Ie, a ping of the NetBIOS name of the server from a Windows client will
>     determine if NetBIOS Name Services are working or not. (OK, as long as 
>     you have DNS switched off and the name is not in the hosts or the lmhosts
>     file.)

better using nbtstat or nmblookup. ping invokes too many weird things
as far as name mapping goes.

> 3. Stopping debugging for all clients so you can focus on one client
> 
>     include = /wherever/%m.smb.conf
> 
>    and in /wherever/%m.smb.conf add
> 
>     debug level = 5 ; or more
> 
>    then look at the log files for that machine
> 
> 4. Check the log files looking for specific message types
> 
> 5. Use smbclient to probe your Samba server to make sure it looks right
> 
> 6. Use testparm to check that your smb.conf file is OK and does not have
>    problems

these four are all in the wrong order. 
 

What you need to get across is that there are 4 different basic things
that can cause a Samba box to not be accessible from a windows box:

1) browsing broken means it doesn't appear in the GUI
2) name resolution broken so the client can't get a IP
3) smb (TCP/139) connection failing
4) authentication failing

the rest is just sugar. 

What you need is a series of steps that _independently_ tests these 4
things. That was what the original design of Diagnosis.txt did. I
haven't looked at it much lately and I know John and other have
changed it so I don't know if that original design has been preserved.

The independence of the tests is critical. You need to be able to test
each of the above 4 without relying on the other 3 working. Or at the
very least don't rely on a facility that an earlier test hasn't proven
to be working. That is easy with appropriate use of "net view",
nmblookup, smbclient, ping and telnet.

Once you've tested those 4 and know which of them is working and which
isn't you can recommend solutions. Anything else is just witchcraft.

If all 4 don't work then suggest that they install Samba before trying
again. (I have had people come to me complaining that Samba doesn't
work when they haven't installed it yet!)