grub doesn't properly migrate from incorrectly-detected-as-evms root filesystems

[Gerv]

Binary package hint: mdadm

Upgrading to the beta has near-bricked my IBM X40 laptop :-( Note that my machine has a single, 40GB disk, and I've never asked for or used RAID of any sort.

A straight boot puts up the splash screen; the progress bar gets 2px in and hangs. Ctrl-Alt-F1 shows:

Loading, please wait...
libgcc_s.so.1 must be installed for pthread_cancel to work

Apr 02 22:53:59 (none) _0_ Engine: signal_handler: ***
Apr 02 22:53:59 (none) _0_ Engine: signal_handler: *** Signal: Aborted
Apr 02 22:53:59 (none) _0_ Engine: signal_handler: ***
Aborted
mdadm: No devices listed in conf file were found.

From reading various bugs, I learnt about using the "break=mount" boot option. That gets me to a (initramfs) prompt.

(initramfs) cat /etc/mdadm/mdadm.conf
DEVICE partitions

(initramfs) ./scripts/local-top/mdadm
mdadm: No arrays found in config file or automatically

(initramfs) ./scripts/local-top/mdadm from-udev
beginning: Loading MD modules
success: loaded module linear.
success: loaded module multipath.
success: loaded module raid0.
success: loaded module raid1.
success: loaded module raid456.
success: loaded module raid5.
success: loaded module raid6.
success: loaded module raid10.
done.
beginning: Assembling all MD arrays
mdadm: No arrays found in config file or automatically
./scripts/local-top/mdadm: ./scripts/local-top/mdadm: 120: log_failure_msg: not found

(initramfs) lvm vgscan
No volume groups found

That's just a random selection of commands I've found from the several bugs which seem to be related. None of the "magic solutions" other people propose seem to work; they all seem to have non-working ancillary RAID arrays, whereas it's my single, only, root drive which won't mount.

I'm hamstrung a bit because busybox doesn't seem to come with an editor.

According to bug 75681, something has very recently fixed in this area - so if I could boot once, I could perhaps do an apt-get to find the upgraded packages. But I can't boot even once!

My laptop is hosed. Please help :-(

Gerv

[Gerv (gerv)]

Subscribing a couple of people who were involved in the related bugs. I hope that's not impolite...

Gerv

[Scott James Remnant (Canonical) (canonical-scott)]

Please provide the output of the following:

  dpkg-query -W dmsetup libdevmapper1.02 udev volumeid
libvolume-id0 lvm2 lvm-common mdadm

[Gerv (gerv)]

(initramfs) dpkg-query
/bin/sh: dpkg-query: not found

Like I said, I can't boot to a normal prompt. I can only get to the (initramfs) one using the "break=mount" boot option. :-(

The following is all I can gather, and hopefully may be of use:

(initramfs) lvm version
LVM version: 2.02.05 (2006-05-12)
Library version: 1.02.08 (2006-07-17)
Driver version: 4.11.0

(initramfs) mdadm --version
mdadm - v2.5.6 - 9 November 2006

(initramfs) ls /lib
...
libdevmapper.so.1.02
libvolume_id.so.0
...

I couldn't find any tell-tale information for the other packages you mention.

Gerv

[Gerv (gerv)]

I've downloaded, burned and booted with a nightly Feisty LiveCD, so I can get the machine started. But I don't know how to access my hard disk (I can't see it mounted) and I don't know how I can run the dpkg-query command without getting the results from the LiveCD rather than the installed software. If this is a good step forward, help in this area would be appreciated.

Gerv

[Gerv (gerv)]

I figured out how to mount my drive, and how to get dpkg-query to use a different database. Here are the results:

ubuntu@ubuntu:/tmp/hd$ dpkg-query --admindir=/tmp/hd/var/lib/dpkg -W dmsetup libdevmapper1.02 udev volumeid libvolume-id0 lvm2 lvm-common mdadm
dmsetup 2:1.02.08-1ubuntu5
libdevmapper1.02 2:1.02.08-1ubuntu5
udev 103-0ubuntu14
volumeid 103-0ubuntu14
libvolume-id0 103-0ubuntu14
lvm2 2.02.06-2ubuntu8
lvm-common 1.5.20ubuntu11
mdadm 2.5.6-7ubuntu4
ubuntu@ubuntu:/tmp/hd$

Hope that helps.

Gerv

[Scott James Remnant (Canonical) (canonical-scott)]

Updating to the latest package versions at the end of the day should cure your problems, if not, please set the bug back to confirmed

[Gerv (gerv)]

I booted into the live CD, manually mounted /dev/sda1, chrooted into it and did:

sudo apt-get update
sudo apt-get upgrade

135 packages later, I rebooted. No improvement, same error. :-((

dpkg-query, executed in the same way as last time, now says:

ubuntu@ubuntu:/tmp/hd$ dpkg-query --admindir=/tmp/hd/var/lib/dpkg -W dmsetup libdevmapper1.02 udev volumeid libvolume-id0 lvm2 lvm-common mdadm
dmsetup 2:1.02.08-1ubuntu10
libdevmapper1.02 2:1.02.08-1ubuntu10
udev 108-0ubuntu3
volumeid 108-0ubuntu3
libvolume-id0 108-0ubuntu3
lvm2 2.02.06-2ubuntu9
lvm-common 1.5.20ubuntu12
mdadm 2.5.6-7ubuntu4
ubuntu@ubuntu:/tmp/hd$

So all packages except mdadm have changed versions.

What do I do now?

Gerv

[Gerv (gerv)]

Resetting status, as requested.

Gerv

[Colin Watson (cjwatson)]

After discussion with Gerv on IRC, this turned out to be because his /boot/grub/menu.lst said root=/dev/evms/hda1 rather than /dev/sda1 or a UUID= version; this in turn was because he had originally installed dapper, where evms was installed by default. grub's migration code needs to handle this case.

[Colin Watson (cjwatson)]

(Gerv also says he's available to help with further debugging)

[Colin Watson (cjwatson)]

Here's a very tentative untested patch.

[Colin Watson (cjwatson)]

grub (0.97-20ubuntu6) feisty; urgency=low

  * Migrate root=/dev/evms/[hs]d[a-z][0-9]* to UUIDs, since systems upgraded
    from dapper to edgy may find themselves with such a root= option by
    mistake (LP: #102148).

 -- Colin Watson <email address hidden> Tue, 10 Apr 2007 13:37:42 +0100

[Colin Watson (cjwatson)]

Note that this has to be installed before you reboot to the new kernel or it won't work; you'll have to fix root= by hand if you've already rebooted. This should fix upgrades from edgy->feisty for afflicted systems, though, which is the main thing.