FAI and portmapper problems

[Alexander Bugl]

Hi Michael, hi Holger,

thanks for your replies. To answer your questions:

> > Every time a package postinst tries to add a user or a group, there
> > comes the error message
> > "rpcinfo: can't contact portmapper: RPC: Remote system error -
> > Connection timed out"
> > It looks like the portmapper on the installation client is not running
> > as it should -- when using addgroup, there is a process
> > "/usr/bin/rpcinfo -p", and this one seems to hang.
> - Could you find out, where rpcinfo -p is invoked? What happens, if you run it
>   yourself?

I investigated a little bit more: My observation was that the
installations hangs every time a user or group is added. This is done
through the scripts /usr/sbin/adduser and /usr/sbin/addgroup from Debian.
Both are written in Perl and include
/usr/share/perl5/Debian/AdduserCommon.pm, in which you can find the
lines ($nisconfig is true, as the package nis has been installed):

if(defined($nisconfig) && -f "/var/yp/Makefile" &&
   -x "/usr/bin/rpcinfo" && grep(/ypserv/, qx{/usr/bin/rpcinfo -p}))

The last call should be the one I can see in the process list:
11434 tty1     S+     0:00 /bin/sh -e /var/lib/dpkg/info/ssh.postinst configure
11470 tty1     S+     0:00 adduser
11474 tty1     S+     0:00 /usr/bin/rpcinfo -p

If I run rpcinfo -p on the install client, there is no answer, too, it
hangs.

> - Which kernel are you using for your setup?

The standard FAI 2.6 install kernel:
:~# uname -a
Linux 2.6.8-fai #1 SMP Fri Apr 8 16:44:39 CEST 2005 i686 GNU/Linux

> - Can you find anything unusual in the output of dmesg?

No. And the same machine installed fine before, no problems.

> - Could you test an installation using the example-configuration (simple)
>   provided with FAI?

I will (have to) try something like this if I do find no other solution,
as the installation needs with the hanging portmapper more than 2 hours
(instead of under 15 min before). But my setup is based on the simple
example, I just added some things.

> PS.: Judging from your output of strace, portmap starts up fine - it just forks ...

But there seem to be some problems with the portmapper: If I kill every
instance and try to restart it, it won't run:
:~# ps ax |grep portmap
11531 pts/0    R+     0:00 grep portmap
:~# /etc/init.d/portmap start
Starting portmap daemon: portmap.
:~# ps ax |grep portmap
13861 pts/0    S+     0:00 grep portmap

Holger suggested:
> did you check if portmap runs on the requiered interfaces ? IIRC there
> was a portmap (package) version which per default binded (past of bind ?) only 
> to the loopback interface. But IIRC this change has been postponed for
> post-sarge...

As the portmapper is not running correctly, I can't check (see above).
But it looks like it should, if I interpret the man-page correctly:
:~# cat /etc/default/portmap
# By default listen on all interfaces
#OPTIONS="-i 127.0.0.1"
And the option -i is not specified.
And if I strace the start of the portmapper, there are lines containing
sin_addr=inet_addr("0.0.0.0"), which looks like the portmapper binds
itself to all addresses:
bind(5, {sa_family=AF_INET, sin_port=htons(111), sin_addr=inet_addr("0.0.0.0")}, 16) = 0

So, any further suggestions? Anything else I can test/provide?

-- 
Alexander Bugl,  Central IT Services, ZMAW
Max  Planck  Institute   for   Meteorology
Bundesstrasse 53, D-20146 Hamburg, Germany
tel +49-40-41173-351, fax -298, room PE042