Bug 7603

Summary:	Installer will not use pre-existing partition on LVM on LUKS: INTERNAL ERROR: unknown device
Product:	Mageia	Reporter:	Joe Shmoe <callimera.42>
Component:	Installer	Assignee:	Pascal Terjan <pterjan>
Status:	RESOLVED FIXED	QA Contact:
Severity:	major
Priority:	release_blocker	CC:	davidwhodgins, fri
Version:	Cauldron
Target Milestone:	---
Hardware:	All
OS:	Linux
Whiteboard:	3alpha2
Source RPM:		CVE:
Status comment:

Description Joe Shmoe 2012-09-27 19:30:17 CEST

When the Mageia 3 installer tries to use a pre-existing partition on LVM on
LUKS, I get: INTERNAL ERROR: unknown device vgfoo/lvbar.

To reproduce:

1. Use a Live CD or some installed system to create an ext4 partition on LVM 
on LUKS.

2. Boot Mageia 3 installer.

3. Choose Custom Disk Partitioning.

4. Select mount point /boot for some unencrypted partition.

5. Enter LUKS key, select mount point / for the pre-existing ext4
partition and click Done (alternatively, try to format or mount it).

There is nothing relevant on tty4. tty3 repeats the error message from the 
screenshot. Going to tty2, cryptsetup status says device "is active." lvm2
lvdisplay shows lv is not available, but lvm2 vgchange -ay makes it
available. I can mount the partition on tty2, but the installer malfunctions
again as above (also if unmounting on tty2 again.)

If I click mount, instead of the INTERNAL ERROR, I just get "mounting
partition /dev/vgfoo/lvbar in directory /mnt failed" and tty3 shows it tried
to run /mount -t ext4 /dev/vgfoo/lvbar /mnt -o acl,noatime.

/root/drakx/report.bug.gz does not exist.

This is on a physical 32 bit machine. I got the same result earlier when I
tried this with Mageia-3-alpha1-dual-CD in a 64 bit VM, which I originally
erroneously reported here: https://bugs.mageia.org/show_bug.cgi?id=5505#c13

Again, as an exception, I hope it's okay that I assign this to you, Pascal,
since Marja said the bug was yours.

Joe Shmoe 2012-09-27 19:30:36 CEST

Whiteboard: (none) => 3alpha1

Comment 1 Joe Shmoe 2012-09-27 20:06:19 CEST

The installer I used now, on the physical machine, was the current current vmlinuz/all.rdz from cauldron/i586/isolinux/alt0/.

Comment 2 Dave Hodgins 2012-10-10 05:33:09 CEST

Confirmed.  Something has seriously broken lvm handling.

After finding that creating an unencrypted lvm logical volume
is also failing with "Failed to wipe start of new LV", I
started experimenting in Mageia 2.

I created a lvm vg and pv and lv using Mageia 2 x86-64 and made a
mistake with the size of the lv, so tried to delete it, and couldn't.

Manually running the lvremove command failed with the device in use.

Using telinit 1, and then using systemctl stop for the services
that didn't stop, such as colord, cups, avahi, the lvremove
still fails, with device in use.

I suspect that udev and/or udisks is grabbing the device as
soon as it's created, preventing the lvcreate from being able
to zero the volume, and preventing lvremove from deleting
it.  I had to boot into a Mageia 1 install, to delete the
partition containing the lvm physical volume.

CC: (none) => davidwhodgins

Manuel Hiebel 2012-10-10 16:33:39 CEST

Priority: Normal => release_blocker

Comment 3 Joe Shmoe 2012-10-16 11:18:09 CEST

Reproduced in 3alpha2 net installer (vmlinuz/all.rdz from 02-Oct-2012).

Whiteboard: 3alpha1 => 3alpha2

Morgan Leijström 2012-10-16 23:26:38 CEST

CC: (none) => fri

Comment 4 Joe Shmoe 2012-11-24 21:43:40 CET

Could not reproduce with current boot-nonfree.iso x86_64, i.e. post-3alpha3.

Comment 5 Manuel Hiebel 2012-11-24 21:56:11 CET

thanks, closing too then.

Status: NEW => RESOLVED
Resolution: (none) => FIXED