Bug#392764: Can't reproduce on PowerStack, MAC partition issue only?
On Sun, Jan 21, 2007 at 07:58:47PM +0100, Ulrich Teichert wrote:
> I tried to reproduce this bug on my Motorola PowerStack II (PPC prep),
> but did not succeed. I've used the d-i daily build from 2007-01-19,
> netinst iso and booted the netinst kernel over TFTP (yes, I know this
> isn't supported, but my bandwith doesn't allow a full netinstall).
> During manual partitioning, I created two RAID partitions:
> ut@pleddis:~$ cat /proc/partitions
> major minor #blocks name
> 8 0 8891556 sda
> 8 1 16033 sda1
> 8 2 489982 sda2
> 8 3 4184932 sda3
> 8 4 4192965 sda4
> 9 0 4184832 md0
> Which is a brain-dead setup with only one disk and a RAID1 on two partitions:
> ut@pleddis:~$ df -h
> Filesystem Size Used Avail Use% Mounted on
> /dev/md0 4.0G 390M 3.4G 11% /
> tmpfs 62M 0 62M 0% /lib/init/rw
> udev 10M 44K 10M 1% /dev
> tmpfs 62M 0 62M 0% /dev/shm
> I haven't seen any failure, which suggests that the problem only shows up
> on MAC partition types (the PowerStack uses PC-style partitions).
> Sorry, but it looks like I don't have the hardware to track this down,
yes, it is a mac partition table issue. There where actually two bugs, one
which i fixed, and was long fixed in ubuntu but the fix didn't go back to
debian until i investigated, and the other which is not in parted, but in
partman, and is still open.
It is possible to test this on any hardware, but you would need a second disk
if your firmware/bios is not able to boot from mac partition tables, like i
believe is the case for PReP boxes. This can also be tested on plain x86 boxes
with a second disk.
My believe is that since mac partitions used to not have the RAID flag, that
partman somewhere has some explicit test for x86 partitions, and sets the RAID
flag only in those cases, or something. I will try to reproduce his on pegasos
and its amiga partition tables, which are in the same case as the mac
partman actually *ERASES* the raid flag set by hand by parted, which is rather
strange, and hints strongly at some problem in partman-md (but also possibly
mirrored in partman-lvm, which has a similar flag setup).