Re: [PATCH] grub-1.98: support for partitionable array v1.2

[Colin Watson]

On Tue, Jul 20, 2010 at 05:56:46PM +0200, Gaëtan Schmidt wrote:
> I modify grub2 to understand mdadm 1.2 partition array without the
> need to map either the metadevice or the partitions in the device.map
> file. Just the physical devices are required.  As the meta devices are
> added, they will be recognized automatically. Only raid 1 is supported
> for the moment but it can be extended to other layouts.

You've sent this *just* after I committed patches to trunk to understand
mdadm 1.x metadata.  Could you please look at trunk instead of 1.98?

> mdadm 1.1 is not fit to be bootable, its superblock overlaps the boot
> sector 0.

Only if you RAID entire disk devices; it works fine if you RAID
partitions.

> I don't use mdX and md/X syntaxes. md_dX is more relevant because of
> the mdadm default 1.x kernel device scheme /dev/md_dX.

We already discussed this and agreed to use md/*.  The current mdadm(8)
says:

  From kernel version, 2.6.28 the "non-partitioned array" can actually
  be partitioned.  So the "md_dNN" names are no longer needed, and
  partitions such as "/dev/mdNNpXX" are possible.

I prefer the explicit namespacing of md/*, and it makes it easier to
support named arrays in a clear way.

> grub-setup.c and grub-probe.c have been heavily altered to understand
> mdadm 1.2 partitionable array. That means, grub-setup.c does not
> support embedded core image in the MBR first sectors gap (<64k)
> anymore.
> 
> To avoid any confusion, grub-setup should be renamed grub-mdraid-setup
> or something like that.

No renames were necessary with the branch recently merged into trunk.
If you need further work to support partitionable arrays, then let's
discuss that; I don't think it's desirable to create a forked version of
grub-setup or grub-probe, though.

> For the try and fix version of the patches, please forget it. It is
> really messy. I am using git for the development and i was not able to
> convert the bzr repository to git, so i don't have a valid tree.
> Anyway, if you are interested in it, please let me know.

Just use bzr instead. :-)

Regards,

-- 
Colin Watson                                       address@hidden