[syslinux] Linux software RAID 1.2 superblocks

[Ferenc Wagner]

"H. Peter Anvin" <hpa at zytor.com> writes:

> Actually, I have had a fairly detailed discussion with Alistair about
> LVM support, and yes, LVM support seems perfectly reasonable, but that's
> a bigger job.

Out of curiosity, was this discussion publicly archived somewhere?

> The one thing I'm still mulling over how to handle is how to deal with
> partitioned mdraids with 1.2 superblock

Do you mean entire disk partitioned mdraids or partitioned mdraids built
of partitions?  You seem to neglect the latter, and they may indeed
sound contrived, but disk size differences probably call for them.

> (0.9/1.0 mdraid superblocks are unaffected, and 1.1 superblocks are
> simply unbootable).

Even 0.9/1.0 partitioned mdraids may pose new challenges if they are
built of partitions, as their partition table resides at a nonzero
sector offset.  There's no problem with such entire disk arrays, though.
Or does the stock MBR code handle this in the general case, too?

> You can't boot a stock MBR on the mdraid volume [...] since an MBR
> generally doesn't expect a nonzero disk offset, unlike a Volume Boot
> Record (VBR), which the Syslinux boot sector is normally installed as.

I understand this as: from the disk MBR you can't chainload a stock MBR
inside a partitioned mdraid volume, because the stock MBR can't handle
relative partition locations.  Which also gives negative answer to the
previous question.

> The easiest option is to install the Syslinux boot sector in the boot
> sector of the mdraid members

You must mean in the hole before the mdraid metadata.

> and make Syslinux handle the offsetting internally.  It would,
> however, mean that active bits on the partition table inside the
> mdraid would not be honored.

I wouldn't mind that too much.  Those bits are accessible from Linux
only, and then one could just as easily rerun the installer with a
different directory.

> A more complex variant would be to write a special mdraid bootstrap code
> which would take the place of the MBR, and walk the partition table
> inside the mdraid.  That'd be a fair bit more complex, however.

And only a little bit better, but much more consistent with the 0.9/1.0
cases.  I'd think going the easy way and leaving this for later doesn't
mean considerable wasted effort now.  There isn't much competition on
this field yet, although something is cooking in Grub.  I wonder what...
-- 
Regards,
Feri.