Grub 2 hangs on boot after installing on RAID1 device (/dev/md0) but reports no problems during install

[i iordanov]

Hi,

I tried reporting this grub2 bug to the Ubuntu 12.04 team, but they
are saying that our installation is not "Supported" by Ubuntu since
their installer does not permit whole-disk arrays to be built. So, I'm
going to try reporting the bug to you. Ubuntu 12.04 uses grub version
1.99-21ubuntu3.1.

This problem is 100% reproducible with Linux md RAID metadata 0.9 and
1.0. We are installing a system with FAI (Fully Automatic
Installation), and creating a full disk RAID1 array with two devices
(/dev/sda and /dev/sdb). This results in a device /dev/md0 which we
then label with a GPT label, and partition with parted, which gives us
devices of the form /dev/md0p1, /dev/md0p2, etc.

Since we are creating the RAID array with metadata 0.9 or 1.0, the
metadata ends up at the end of the array, and so the BIOS should have
no trouble reading from each disk individually to load the
boot-loader, etc. What's more, grub2 supports GPT, and ordinarily has
no trouble booting from a GPT labeled disk.

At the end of the FAI installation, we install grub2 like this:

grub-install /dev/md0

which results in the usual "no errors reported" message that Grub2
prints out. What we expect at this point is for md to automatically
sync the grub2 changes to both /dev/sda and /dev/sdb, and for the BIOS
to be able to boot from either device.

What happens instead is that when the BIOS attempts to boot from
either device, it hangs with a cursor blinking at the top-left corner
of the screen *indefinitely*. So, it seems like although Grub 2 thinks
the installation went fine, it is not actually OK.

The interesting thing is that when we install Grub 2 to /dev/sda and
/dev/sdb individually, rather than /dev/md0, everything works as
intended. I've seen references on the web indicating that installing
to the raw devices is the right thing to do, but I could not find any
official documentation supporting the claims.

If grub is supposed to be installed to /dev/sda and /dev/sdb, then the
bug is that it doesn't complain when an attempt is made to install it
to /dev/md0.

If it should work even when installed to /dev/md0 (with metadata 0.9
and 1.0), then the bug is that it doesn't actually install properly.

In case it is relevant, here is a print out of the partition table,
even though (as mention previously), Grub 2 does install successfully
when the raw devices are specified (/dev/sda, /dev/sdb):

# parted /dev/md0 print

Model: Linux Software RAID Array (md)
Disk /dev/md0: 500GB
Sector size (logical/physical): 512B/4096B
Partition Table: gpt

Number Start End Size File system Name Flags
 1 17.4kB 8389kB 8372kB grub bios_grub
 2 8389kB 21.0GB 21.0GB ext4 root
 3 21.0GB 23.1GB 2097MB linux-swap(v1) swap
 4 23.1GB 33.6GB 10.5GB ext4 var
 5 33.6GB 54.5GB 21.0GB ext4 tmp
 6 54.5GB 75.5GB 21.0GB ext4 local
 7 75.5GB 500GB 425GB ext4 data

Here is a link to the original bug-report on launchpad which contains
the same information as what was sent by email.

https://bugs.launchpad.net/ubuntu/+source/grub2/+bug/1019389

Thank you very much!
Iordan

-- 
The conscious mind has only one thread of execution.