Re: Autoconfiguring root device and fstab(5) (wedges?)

[Jean-Yves Migeon <jeanyves.migeon%free.fr@localhost>]

On 26.03.2011 15:38, Martin Husemann wrote:
> On Sat, Mar 26, 2011 at 02:44:49PM +0100, Jean-Yves Migeon wrote:
>> (please note that I am not at all familiar with disklabel/mbr formats)
> 
> Well, you have it mostly correct. A wedge is a pure runtime concept, there
> is no persistence of any kind. The only way we can reasonably use them with
> disks besides the (necessarily slightly volatile auto detection) is GPT.

By any luck, would that be possible to have a "use wedges for this
disklabel" bit saved in disklabel struct, using a member like
``d_spare''? I am willing to check if other BSDs use this struct element
for other uses, just to avoid collisions in bit choice.

> There is GPT editing code missing in sysinst, which is easy to add.
> 
> And there is the uncommited patch suggested by mlelstv that makes mounting
> of wedges via names in GPT possible. I'd like to see the latter sorted out
> first, so we can offer the most userfriendly way in sysinst.

Manuel's suggestion works, but I have to script a couple of
disklabel/raidctl/raid0.conf commands inside anita and... it should
rather be sysinst's job. So I am more willing to work on sysinst there,
and add the GPT support for sysint, then anita.

If you believe that GPT support inside sysinst is easy to add, I would
prefer to do it that way. Sadly, I spent this late afternoon trying to
make it work (using mishka's tutorial [1]), but I never managed to get
past the "Error No /boot" from fatboot.S with QEMU. Looks like the
primary stage bootloader fails looking for /boot in the EFI partition :/

[1] http://www.netbsd.org/~mishka/gptboot/howto.html

-- 
Jean-Yves Migeon
jeanyves.migeon%free.fr@localhost