Re: [PATCH] Fix grub-probe partition naming on FreeBSD

[Grégoire Sutre]

Hi Colin,

> The following patch is aimed at fixing this Debian bug:
>
>    http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=585068
>
> I've tested it on Debian GNU/kFreeBSD and it seems to be producing sensible
> output now.

In another thread [1], it was observed that offsets are not absolute in
FreeBSD disklabels, whereas they are absolute in NetBSD disklabels.
I believe that find_partition_start is supposed to return absolute
offsets.  Does DIOCGDINFO convert the on-disk label into absolute
offsets on FreeBSD?

> The one glitch is that if you ask it to probe /dev/ad0s1a, it returns
> (hd0,msdos1) rather than (hd0,msdos1,bsd1): this is because both /dev/ad0s1
> and /dev/ad0s1a have the same start sector, and it just uses the first one
> it finds.  When I set prefix to (hd0,msdos1)/boot/grub, GRUB can read from
> that perfectly well, so can I ignore this glitch on the basis that it
> doesn't cause a practical problem?

We get a similar behavior on NetBSD.  As I mentioned in [2], this may
have an impact when the kernel is (a) loaded with the multiboot
protocol, and (b) relies on the MBI boot_device field to find its
root -- which it shouldn't anyway, so it's not a big deal.  I am not
aware of other impacts.

I know that hybrid msdos+gpt are not recommended, but, for the record,
I guess that another glitch could happen if /dev/ad0s1X (msdos) and
/dev/ad0p1Y (gpt) are actually the same partition: grub-probe would
return the same answer for both, thus ignoring the user's desire to use
a specific partitioning scheme.  I believe that such a partitioning is
supported by FreeBSD, but I may be wrong (please tell me if so), I did
not test this myself.

Grégoire

[1] http://lists.gnu.org/archive/html/grub-devel/2010-05/msg00065.html
[2] http://lists.gnu.org/archive/html/grub-devel/2010-06/msg00075.html