Re: devfs

Richard Gooch (rgooch@atnf.CSIRO.AU)
Fri, 16 Jan 1998 09:53:07 +1100

Michael Neuffer writes:
> On Wed, 14 Jan 1998, Leonard N. Zubkoff wrote:
> > I guess I could live with that. It would actually bring things more
> > into line with Solaris 2 (except that we have a bus identifier as
> > well).
> > However, since you raise this again, Leonard, what about the issue of
> > slices within a partition? It appears that Solaris two places multiple
> > slices inside a standard MS-DOS-style partition. What would you
> > suggest to handle that? How about:
> > /dev/sd/c0b0t0d0p1s6 :-)
> >
> > Ugh. I sure hope we don't need both "p" and "s". My only Solaris experience
> > is on the SPARC, and there's no "p" in use there. Do we really need both?
>
> If you look at Solaris/Intel and *BSD/Intel and want to be able to
> keep a consistent naming sheme we need it. All of them have slices
> _within_ partitions.

What does *BSD/Intel have as it's naming scheme for partitions and for
slices?

Pre-emptive proposal: it we have to deal with partitions *and* slices,
one possibility is to munge the partition ID and slice ID
together. The first N "subdevices" (s) would be real partitions, the
remaining subdevices would be slices within partitions.
However, this I consider messy. You would have to know how many real
partitions you have and which ones have how many slices.
So, it would be more logical to have a naming scheme for
controller,bus,target,device(lun),partition thus:
/dev/sd/c0b0t0d0p1 a whole partition

Then, if any slices are found in partitions, you would also have:
/dev/sd/c0b0t0d0p1s3 a single Solaris slice within a partition

and of course:
/dev/sd/c0b0t0d0 a whole disc

Leonard: what do you think of this? I don't think we can ignore the
fact that partitions and slices are different. Munging them together
only partially hides it.
Also, if we were to go back to "p" for partition, how about "u" for
LUN?

Regards,

Richard....