Subject: Re: wedges vs. not-quite-wedges, was > 1T filesystems, disklabels,
To: Klaus Heinz <k.heinz.dez.zwei@onlinehome.de>
From: Bill Studenmund <wrstuden@netbsd.org>
List: tech-kern
Date: 12/22/2002 15:30:19
On Sat, 21 Dec 2002, Klaus Heinz wrote:

> David Laight wrote:
>
> > If you can save enough stuff to generate a 'reasonably likely'
> > disk fingerprint, you should be able to detect a disk being moved
> > to a different controller - and reassign it the same logical disk
> > number.
>
> In the discussions cited at the start of this thread there was a
> suggestion by Ty Sarna to introduce readable names for partitions (akin
> to the AmigaOS concept, please bear with me :-). Couldn't this solve the
> problem of moving disks? And being able to refer to partitions also with
> their assigned name would be real convenient (IMO).

I have two concerns with this. Note that as I understand it, the Intel
partitioning scheme (need to dig its name out) does add identifiers to
partition names, so we'll have to confront this either way.

1) While it's nice where we can use it, we can't use it everywhere, and
this proposal is to fix things up in a manner that works for all ports.
Not all partition types support this, so I am very concerned about
depending on it.

2) Do we have a requirement in the partitioning layout that the names be
unique? Further, do we have a requirement that an OS maintains uniqueness
of names even in the space of partitions it doesn't understand?

i.e. For these names to really work, we need ALL OSs that mess with the
partition map to respect them. Also, we need to watch for the case say
where windows lets a user rename a Windows disk to match the existing name
of a NetBSD partition, merely because since Windows doesn't understand it,
it ignored it. ??

> I faintly remember Linux also knows names for partitions but I think
> they use some random strings which are not suitable for humans to
> remember.

That would be a pain. But at least that you could have an alias config
file.

Take care,

Bill