Bug#639830: mdadm: alternative md-device names

Michael Tokarev mjt at tls.msk.ru
Sat May 26 10:23:44 UTC 2012


tags 639830 + moreinfo unreproducible
thanks

On 26.05.2012 13:03, Michael Tokarev wrote:
> 
> When creating initramfs, alternative MD device names (/dev/md/X) apparently are not handled correctly by mdadm initramfs hook

Ok.  After quite some digging I was able to - sort of, anyway -
reproduce the problem by using different forms of device names
in different places.  It still does not fail as described in
#639830 -- Chistian, I need some of your help here please.

What I noticed is that Christian uses /dev/md/0 in mdadm.conf,
and /dev/md0 as root device (in fstab apparently) -- too bad
he didn't include configuration details.  This is a way to
trouble, all usages should be consistent.

But even this way, mdadm assembles arrays correctly here,
and when asked to assemble /dev/md/0 it creates both md0
and md/0 nodes (the latter being a symlink, according to
udev rules).

I was able to omit array assembly from initramfs by setting
INITRDSTART= (empty) in /etc/default/mdadm.  This way, mdadm
in initramfs does not assemble any array at all, but --assemble
--scan assembles my array (but gives it name md127, even with
0.90 superblock)

I can sort of mimick the described failure by using /dev/md0
in mdadm.conf and /dev/md/0 in fstab - this way it does not
work, but it actually _never_ works this way -- when mdadm
asked to build /dev/md0, it never creates alternative name
/dev/md/0; while when asked to create /dev/md/0, both names
actually gets created (by udev rules).

So I fail to see how the issue can be reproduced, and I need
more information.  For a start, I need the output from mdadm
bugscript which were removed from the initial bugreport, which
includes contents of mdadm.conf, INITRDSTART setting and other
useful things.

Thanks,

/mjt





More information about the pkg-mdadm-devel mailing list