install debian on fake raid/bios raid
pjones at redhat.com
Sat Jul 16 05:47:49 UTC 2005
On Sat, 2005-07-16 at 01:38 +0200, Gerte Hoogewerf wrote:
> Some 'small' things. The most important problem is: handling changes to
> the partition table. This is just different from any other case and
> simply needs attention. Normally when a partition table on a disk
> changes, a program like fdisk tells the kernel to re-read the partition
> table (ugh!). Re-mapping dmraid partitions can be done by re-running
> dmraid, the kernel can't help us here.
Well, maybe it can... it's not wired up to do it now, but it seems like
it should be relatively easy to make the ioctl code for dm devices
trigger a hotplug event when it gets BLKRRPART on a dm device...
> The second problem is: bootloaders. For lilo a specially crafted config
> file is needed. A patched lilo with device-mapper support (like the
> patch that Mandriva uses by default) breaks any chance to install
> dmraid. I've written a patch for lilo to make installation on
> device-mapper Raids easier, but my patch on its turn breaks LVM and EVMS.
> For grub it's important to load a correct device map. This can easily be
> done on the the grub commandline OR by having some shell script output a
> working device map to a file (and call grub to use it).
For raid1 you'd really want to do more than that, though -- you want
grub-install to put the boot block on each volume, and to use the
kernel/initrd local to that volume. See the patches for raid in the RH
grub package. (said patches haven't been updated for dm devices yet,
but it shouldn't be much work.)
More information about the Ataraid-list