[comp.unix.aix] puzzling tcpip problem. ODM messed up?

eliot@engr.washington.edu (Eliot Lim) (06/12/91)

I've been called in to fix an RS6000 which has been fubar'd by people
who didn't know what they were doing.  It concerns tcpip.  First, some
details:

- running 3005/3.1.5 code


Symtoms:

- machine can go out on the network anywhere and do all tcpip services.

- nothing outside can come in.  only ping from outside gets a response.

- nfs works fine.

- yes, inetd is running, and inetd.conf has all the stuff uncommented.

- the machine cannot even ping/telnet/rlogin/finger itself.  what makes
  this even more wierd is that i can rlogin to another machine out on the
  net and this remote machine can ping it.

- I've noticed that telnetd/rlogind do not get fired up when a remote
  machine is attempting connection.  it seems that inetd does not
  recognise the connection request.

- yes, the ip addresses match in the smit-tcpip menu, /etc/hosts and
  ifconfig.  I have tried reconfiguring tcpip from scratch, erasing
  /etc/hosts & resolv.conf, even replacing the inetd executable... no
  luck.

- ifconfig shows all the right flags, broadcast, netmask etc.  there's
  a sun 4 on the same wire with the same set up that works a-ok.

- i've tried flushing the routing table, running routed etc, no luck.

- it seems that this machine has an identity crisis.  Since all the ascii
  files look alright, the finger drifts towards the mysterious ODM.  I
  wonder if the information in there is messed up.  If so, how would I
  clear it up?  I have tried the smit-tcpip-minimum configuration bit,
  but no success.  Anyone out there have any clues?  I hate to reinstall,
  since 3005 is an update and not an install.  ie. i will have to put on
  3003 then apply 3005.

Thanks in advance,

Eliot Lim
Univ. of Washington
Seattle

johnson@horta.rice.edu (Bruce Johnson) (06/12/91)

In article <1991Jun11.203549.4994@milton.u.washington.edu> eliot@engr.washington.edu (Eliot Lim) writes:
>
>I've been called in to fix an RS6000 which has been fubar'd by people
>who didn't know what they were doing.  It concerns tcpip.  First, some
>details:
>
>- running 3005/3.1.5 code
>
>- nothing outside can come in.  only ping from outside gets a response.

No guarantees that this will work, but I mentioned to yates@a.chem.upenn.edu
that several people had been complaining about no incoming calls after
going to 3005.  His response:

#You mentioned a problem telnetting in after the upgrade. Well, this morning
#we observed the same thing. However, the cure is simple. Go into smit,
#System Environments and Processes, and Change Number of licensed users
#to more than 1-2 . We set ours to 2-32.

It would be a cruel twist if a simple default flag like this was the
culprit.  Let us know if it is or not.

Bruce Johnson
johnson@horta.rice.edu

eliot@engr.washington.edu (Eliot Lim) (06/12/91)

In article <1991Jun12.001444.11707@rice.edu>, johnson@horta.rice.edu (Bruce Johnson) writes:

> >- nothing outside can come in.  only ping from outside gets a response.
> 
> #You mentioned a problem telnetting in after the upgrade. Well, this morning
> #we observed the same thing. However, the cure is simple. Go into smit,
> #System Environments and Processes, and Change Number of licensed users
> #to more than 1-2 . We set ours to 2-32.
> 
> It would be a cruel twist if a simple default flag like this was the
> culprit.  Let us know if it is or not.


Unfortunately this is not the problem.  If the licensed users bomb
hits, what you get is a "connection closed" message.  In my problem
there is no response.  The command hangs until it timeouts.  Commands
like finger should not be affected by # of licensed users but in
this case it is.  Also, this machine was upgraded from 3003, where the
1-2 licensed users bomb was planted.  That was fixed a long time ago.
Also, how would one explain that the machine cannot ping itself, while
machines on the net can ping it?

Anyway, thanks for responding, I will post if I ever find out what
the problem is.


Eliot