Bug 727908

+++ This bug was initially created as a clone of Bug #700067 +++

Created attachment 495179 [details]
patch which fixes the problem

Description of problem:

If you try to kickstart RHEL6 machine using a kickstart file with a custom hostname set through "network --hostname=<somename> ...", rhnreg_ks during registration time sends hostname obtained from dhcp instead of the one defined on network line.

The problem was fixed already for RHEL4 (BZ #134078) and RHEL5 (BZ #457953), but the fix doesn't work for RHEL6, because RHEL6 anaconda benefits of NetworkManager whose behaviour circumvent the code logic used to fix the problem for RHEL4/5.


Version-Release number of selected component (if applicable):

rhn-client-tools-1.0.0-38.el6


How reproducible:

Always


Steps to Reproduce:

1. Setup a kickstart with network line simillar to the following one:

network --bootproto=static --hostname=ananas --ip=XXX.XXX.XXX.XXX --netmask=255.255.255.0 --gateway=XXX.XXX.XXX.XXX --nameserver=XXX.XXX.XXX.XXX

2. Kickstart a machine and then find the system's profile through WebUI and check a value of 'Hostname' field.

Systems -> Systems -> [someprofile] -> Details -> Overview


Actual results:

It contains hostname obtained from dhcp during installation time.


Expected results:

It should contain the hostname set on network line.


Additional info:

The problem is caused by the fact that anaconda uses NetworkManager to setup network before installation begins, which obtains network information from dhcp and sets also hostname so that it is different than in RHEL4/5 where it was set to localhost.localdomain (everytime) and it breaks a code logic which decides how to setup the hostname.

Currently, it sends hostname defined through network line only in case a hostname stored in kernel (line #492) is localhost.localdomain (line #499, gethostname() calls uname syscall internally which returns hostname as nodename, which can be obtained also from /proc/sys/kernel/hostname):

rhn-client-tools-1.0.0/src/up2date_client/hardware.py:
488 def read_network():
489     netdict = {}
490     netdict['class'] = "NETINFO"
491 
492     netdict['hostname'] = gethostname()
493     try:
494         netdict['ipaddr'] = gethostbyname(gethostname())
495     except:
496         netdict['ipaddr'] = "127.0.0.1"
497 
498 
499     if netdict['hostname'] == 'localhost.localdomain' or \
500     netdict['ipaddr'] == "127.0.0.1":
501         hostname, ipaddr = findHostByRoute()
502 
503         if netdict['hostname'] == 'localhost.localdomain':
504             netdict['hostname'] = hostname
505         if netdict['ipaddr'] == "127.0.0.1":
506             netdict['ipaddr'] = ipaddr
507 
508     return netdict

A solution could be either to use similar changes mentioned in attached patch which sends HOSTNAME from /etc/sysconfig/network obtained from findHostByRoute() everytime it differs from the one set in kernel or fix anaconda to set hostname to 'localhost.localdomain' before processing a kickstart file.


Workaround:

The problem disappears as soon as a client/system executes rhn-profile-sync after installation reboots or someone sets hostname to localhost.localdomain in kickstart %post script or sets directly hostname using some kickstart variable and runs the rhn-profiles-sync after that:

network --hostname=$some_kickstart_variable_holding_hostname
...
%post
hostname $some_kickstart_variable_holding_hostname
rhn-profile-sync