Re: Mdadm, udev and fakeraid?

[NeilBrown <neilb@suse.de>]

On Fri, 22 Apr 2011 19:37:40 -0700 Dan Williams <dan.j.williams@gmail.com>
wrote:

> On Sun, Apr 17, 2011 at 5:38 PM, NeilBrown <neilb@suse.de> wrote:

> > As has been mentioned elsewhere, mdadm only recognised IMSM arrays on
> > machines with IMSM hardware.  I'm not entirely happy about this and may well
> > change it.
> 
> I have trouble answering the "least surprise" question in this area.
> 
> Is it more surprising to go into your BIOS, explicitly turn off raid
> support and still see raid devices showing up?

Is the "RAID support has been explicitly turned off" state visible from a
running kernel? or is it indistinguishable from "platform does not have RAID
support"?

> 
> Or is it more surprising to take a raid array from a raid enabled
> system to raid disabled system and wonder why things won't assemble?
> 
> For safety I think it is better if mdadm not perform operations that
> might be incompatible with the platform option-rom.  But if you need
> to recover to a usb attached drive, or some other
> platform-incompatible configuration, you can use the environment
> variable in a pinch.

There are 3 interesting cases:  create, assemble, examine.
(grow might be interesting too, but for now it would be confusing).

I am perfectly happy for 'create' to be arbitrarily hard if platform support
is not available.  One is unlikely to want to create an array in that case
anyway.
I think 'examine' should always show whatever it can, which is the case for
3.2.1.  Possibly it should also give a warning about  any difficulty that
might be experienced in assembling the array.

Assemble in the interesting case.  The law of least surprise requires it to
either work or give a good error message.  Your suggestion that it possibly
should not work in some cases seems defensible, so at least a very clear
error message would be good.
As for how to over-ride the default caution - I would prefer --force to
achieve it rather than requiring an environment variable.  I would possibly
accept --force-platform (or similar) but I think --force should be sufficient.

What think you?

Thanks,
NeilBrown
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html