[xen-tools] Trying to get centos-5 installed with xen-tools/rinse

[Henning Sprang]

Hi,

Even if I could just be happy with installing centos/rhel-5 with the
combination of xen-tools and FAI, and a self-made minimal  base image
which works just fine, I'd like to get the same done with plain
xen-tools and rinse.

So I fiddled around with it a bit these days, here the problems I ran
into (and finally, got solved):

I'm using the role and role-args stuff in a file given with --config like this:

hostname=xen-tools-test
mac=00:16:3E:3C:02:80

install=1
#role=fai-install
template=/etc/xen-tools/paravirtual.tmpl
size=5GB
dist=centos-5

install-method=rinse

role=passwd
role-args=root


1) even if I have the role "passwd" and set role-args, I seem not to
get a password set. As centos refuses to let people loging without a
password, I had to mount the filesystem into the local one, and chroot
into the system to set a password.

I didn't test it yet if it works when I give the parameters for that
on the command line - if this is the problem, I think it should be
fixed that the --config stuff works in line with command line
arguments - think of --config as a way to store a multitude of command
line args for a single host...

2) When running with the Debian Kernel, my console is flooded with
"4gb seg fixup, process login (pid 1366), cs:ip 73:b7eb5d5c"  So, we
have a problem with TLS here.

I solved it by doing this manually:

echo hwcap 0 nosegneg' > /etc/ld.so.conf.d/libc6-xen.conf && ldconfig

This should probably be done by the centos hooks in xentools, as there
is no doubt that everybody will need this :)

It _can_ lead to a message that this is superflouus when a user
installs a native centos xen kernel, because the kernel package add
another file (named with the kernel-version) that does the same, but I
think it does not harm...


3) While trying to fix the segneg problem, I remembered that I learned
my lessons about kernel-compatibility in distributions - and thought,
this might solve the problem with the segneg's.
( I forgot the "ldconfig" command after adding the file, first, whoich
was the real reason)

So I made the same I do with my xen-tools/FAI installed domains:
Pull a centos kernel image and initred into the dom0, and boot the
domU with that one.

So I mount the Filesystem again, and install the rhel-kernel inside.
But this does hang at boot, after sshd is shown to be started.

I _guess_ this is some console issue, and the rinse or xen-tools
install scripts do something with the domU to run fine with the
console access the debian kernel provides, but which fails to work
with the centos native kernel.
I even get a ssh loging prompt, and I suspect(if not tested, yet) that
running network services that do not require proper console setup will
work in this setting.

Any hint what could be done to make the install centos-kernel-compatible?

4) in debian, this is a base package, but in centos not: dhclient!
So, as in most cases net access is required (and all other things
needed for this seem to be st up automatically by xen-tools, otherwise
I guess we wouldn't have th ifcfg-script at all, it should probably
also install the dhclient package.

As the network scripts in centos do not return any error, even if you
configure the interfaces to riun with dhcp, which cannot work without
dhclient, this is a nice trap...



So far, that was it for now - with some fiddling, i got my centos with
xenm-tools and rinse working - but with some minor fixes it cpould be
easier, I guess.

Henning