cannot create md devices with the installer

[Danilo Piazzalunga]

I was testing an installation of a current image with Qemu.

Just after the creation of RAID partitions with partman, I was trying to
configure md devices, but mdcfg stopped with an error (it said that there are no
unused partitions of type "Linux RAID autodetect"). I am sure that the
partitions don't contain any file system.

At first, I thought the problem was due to the absence of the /dev/md* device
files, but creating them by hand didn't change anything. The only way to create
md devices at installation time was by opening a shell and resorting to mdadm.

The same test was performed with a current netinst image from Debian proper, and
it worked flawlessly.

[Colin Watson (cjwatson)]

When you say "current image", do you mean Warty or Hoary?

[Danilo Piazzalunga (danilopiazza)]

Hoary. Sorry, should have said "current Hoary daily image". (As for the Debian
image, it was a post-rc2 daily image too.)

The same test case worked perfectly with Warty.

[Jowilly (jowilly)]

Same problem here.

System:

- 2 SATA HDs (sda,sdb) with 4 raid partition = 2 md (md0,md1), were setup and
working with Gentoo.

When trying to add the raid partitions, it also says no unused partitions of
type "Linux RAID autodetect".

I have also tried to make 2 new raid partitions (that would be partition 5 and
6) with the hoary installer. When trying to add them to a raid its the same problem.

This was tested with the Hoary image dated 28.01.2005.

[Colin Watson (cjwatson)]

Duplicated; the issue in question is a bug in partconf, but there's a further
bug in mdcfg which also needs to be fixed.

[Colin Watson (cjwatson)]

I believe I've fixed this now. When Array CD 5 is released (currently planned
for Wednesday), please test with that, and reopen if not fixed:

partconf (1.05ubuntu2) hoary; urgency=low

  * get_all_partitions()'s symlink handling in /dev/discs was incorrect in
    places, and broke when using udev. Replace it with use of
    canonicalize_file_name().

 -- Colin Watson <email address hidden> Mon, 14 Feb 2005 17:27:31 +0000

mdcfg (1.09ubuntu1) hoary; urgency=low

  * Use --auto=yes parameter to mdadm when creating arrays, to avoid failing
    with udev.

 -- Colin Watson <email address hidden> Mon, 14 Feb 2005 20:06:35 +0000

[Danilo Piazzalunga (danilopiazza)]

I tested it with the 20050215.3 daily image: it appears to succeed, but silently
fails(*). Manually creating the needed /dev/md[0-9] devices makes it work. I'm
not reopening, since the 1.09ubuntu2 upload may have fixed this already. I will
test with Array-5 and let you know.

(*) Maybe I should report this too, marking it as NEEDINFO; but I fear that
would be abusing of bugzilla.

[Colin Watson (cjwatson)]

(In reply to comment #6)
> I tested it with the 20050215.3 daily image: it appears to succeed, but silently
> fails(*). Manually creating the needed /dev/md[0-9] devices makes it work. I'm
> not reopening, since the 1.09ubuntu2 upload may have fixed this already. I will
> test with Array-5 and let you know.

This is certainly why I said "Array CD 5" rather than tomorrow, to give myself
time to test :-) Thanks for testing, though. I spotted some similar issues
myself, and made the following further uploads to (I think) fix it:

mdcfg (1.09ubuntu2) hoary; urgency=low

  * Create /dev/md at mdcfg startup. (Perhaps this should be done in mdadm
    instead?)

 -- Colin Watson <email address hidden> Tue, 15 Feb 2005 17:55:17 +0000

rootskel (1.11ubuntu7) hoary; urgency=low

  * Restore /.dev bind-mount, since I've discovered something that needs it
    (mdrun).

 -- Colin Watson <email address hidden> Tue, 15 Feb 2005 19:25:36 +0000

mdadm (1.8.1-1ubuntu3) hoary; urgency=low

  * debian/mdrun:
    - Remove Ubuntu-specific 'shift' in the /.dev case that broke at least
      with busybox sh.
    - $BASE was being unnecessarily set twice if $AUTOCREATE, and the second
      time this overwrote the previous check for a /dev/md directory; remove
      the duplication.

 -- Colin Watson <email address hidden> Tue, 15 Feb 2005 19:31:59 +0000

[Danilo Piazzalunga (danilopiazza)]

Sorry for the premature testing. However, I did it again ;-) and now I can
confirm that, with the changes you mentioned, it works perfectly. Thank you
again for your interest!

[Colin Watson (cjwatson)]

(In reply to comment #8)
> Sorry for the premature testing.

Don't let me discourage you; better more testing than less.

> However, I did it again ;-) and now I can
> confirm that, with the changes you mentioned, it works perfectly. Thank you
> again for your interest!

Fantastic; this fix'll look good in the release notes. Thanks for testing!