Fedora xen - Feb 2007 - Centos Hangs before login prompt, and udev fails.

Hi,

  I finally got xen to work properly by creating initrd by hand, so thanks to
all the people who helped me out last time. Now I have another issue. The centos
guest is hanging just before the login prompt comes up, and I also notice that
udev is failing initially itself. I am using the images from jailtime.org.

  I created the mkinitrd by using the command below. Is there anything I should
add?

mkinitrd -f --with=xennet --builtin=aic7xxx --builtin=serverworks
--preload=xenblk --omit-raid-modules --omit-lvm-modules --fstab=myfstab 
/boot/lxxen-initrd.img 2.6.19-1.2895.fc6xen"

  Thanks.

Update on this:

 I was able to login to the domU using sshd, and found that it was the
nash-hotplugd that''s using 99% of the cpu, and I think this is
what''s interfering with the udevd and also making the mingetty not
appear on the console. So how do I turn off nash-hotplugd in the initrd?

 thanks.

Hello:

 It seems 
 
 ------------------
 - scrub another possible error message from dmraid output
  * Mon Feb 13 2006 Bill Nottingham <notting(at)redhat.com> 8.28-1
     - kill nash-hotplug before starting udev (<pjones(at)redhat.com>)
------------------

It appears you need to kill nash-hotplug before you start udev, but I think the
kill is failing and the something is screwing up nash-hotplug to consume a lot
of cpu. So how do I kill nash-hotplug before starting udev?

Thanks.

Allright, I think I got the issue. I am trying to boot a centos 4.1 on a fc6.
Now fc6 initrd, even the custom created one, has nash-hotplugd, which has to be
killed before starting udev, but this logic is not there in the initscripts of
centos-4.1. So I have to basically hand craft an initrd that will kill off this
nash-hotplug in the initrd itself, if it has to work with all the older distros.

 This is too much work, to tell the truth. I will test with fc6 guest on fc6 and
report back.

On Fri, 2 Feb 2007, K T Ligesh wrote:
>  Allright, I think I got the issue. I am trying to boot a centos 4.1 on a
fc6. Now fc6 initrd, even the custom created one, has nash-hotplugd, which has
to be killed before starting udev, but this logic is not there in the
initscripts of centos-4.1. So I have to basically hand craft an initrd that will
kill off this nash-hotplug in the initrd itself, if it has to work with all the
older distros.
I don''t think that is the reason for not getting a login prompt. The
nash-hotplug can be annoying. I''ve had to kill
it in the past too. I infact disable hotplay/udev on those older images, and
just use a MAKEDEV and create the
devices I need,

To fix your login problem do this:

mount -o loop centosfs dir
cd dir

(or login via ssh to the running centos box)

cd dev ; /dev/MAKEDEV null zero random urandom console tty pty hda xvd loop
(this uses the MAKEDEV script in the /dev/ of dom0 to create nodes in the
 /dev of the guest image)

mknod dev/xvc0 c 25 187
(cause there is some confusion about which major/minor to use)

edit inittab and change the mingetty for tty2 to xvc0, eg:

	2:2345:respawn:/sbin/mingetty xvc0

(change tty2 not tty1 because in a normal "working" situation, tty1
will be mapped
 to the xen virtual console already, you do not want to break that)

And do:

echo "xvc0" >> etc/securetty
(to allow root logins)

I also made a xenU ramdisk for my images, on dom0:

mkinitrd /boot/initrd-2.6.18-1.2849.fc6xenU.img --with=xennet --preload=xenblk

(again note to redhat: PLEASE include xennet/xenblk in the xen ramdisk, even if
 the dom0 strictly does not need it. It will allow everyone to re-use the same
  initrd when not using pygrub with disk images, but kernel=/ramdisk= statements
 with a rootfs)

Paul

On Fri, Feb 02, 2007 at 02:10:24AM +0100, Paul Wouters
wrote:
> On Fri, 2 Feb 2007, K T Ligesh wrote:
> 
> >  Allright, I think I got the issue. I am trying to boot a centos 4.1
on a fc6. Now fc6 initrd, even the custom created one, has nash-hotplugd, which
has to be killed before starting udev, but this logic is not there in the
initscripts of centos-4.1. So I have to basically hand craft an initrd that will
kill off this nash-hotplug in the initrd itself, if it has to work with all the
older distros.
> (or login via ssh to the running centos box)
> 
> cd dev ; /dev/MAKEDEV null zero random urandom console tty pty hda xvd loop
> (this uses the MAKEDEV script in the /dev/ of dom0 to create nodes in the
>  /dev of the guest image)
> 
> mknod dev/xvc0 c 25 187
> (cause there is some confusion about which major/minor to use)
> 
> edit inittab and change the mingetty for tty2 to xvc0, eg:
> 
> 	2:2345:respawn:/sbin/mingetty xvc0
> 
> (change tty2 not tty1 because in a normal "working" situation,
tty1 will be mapped
>  to the xen virtual console already, you do not want to break that)
> 
> And do:
> 
> echo "xvc0" >> etc/securetty
> (to allow root logins)
> 
> I also made a xenU ramdisk for my images, on dom0:
> 
> mkinitrd /boot/initrd-2.6.18-1.2849.fc6xenU.img --with=xennet
--preload=xenblk
> 
> (again note to redhat: PLEASE include xennet/xenblk in the xen ramdisk,
even if
>  the dom0 strictly does not need it. It will allow everyone to re-use the
same
>   initrd when not using pygrub with disk images, but kernel=/ramdisk=
statements
>  with a rootfs)
> 
 Yeah, I did that. So now xen uses xvc0 and not the ttyS0? That''s
actually good news, since that means that the tty is now freed up?

 Thanks.