Re: SCSI DVD-RAM partitions

[Christoph Hellwig <hch@infradead.org>]

On Tue, Jul 12, 2005 at 10:23:17AM +0200, Thomas Heinz wrote:
> Hi Christoph
> 
> You wrote:
> >While adding support for partitions on sr is trivial it has a huge
> >drawback: it's chaning the dev_t space by using up device numbers
> >for partitions, so /dev/sr0 ff will have different device numbers
> >with that change applied.  I have an old patch that's supposed to
> >enable support for partitioned scsi removable devices at
> >http://rechner.lst.de/~hch/hacks/sr-parts.diff, I'm not sure it
> >actually ever worked (but you should get the basic idea from it)
> 
> Ok, thanks for your valuable input. In fact, I thought about making
> the device available both as /dev/srX and /dev/sdX at the same time
> in order to support partitions. In my case it would even suffice to
> make it available as /dev/sdX instead of /dev/srX.

That doesn't make sense because sd is a very different driver from sd.
Besides that aliasing different dev_ts to the same underlying blockdevice
can't work, it's cause all sorts of aliasing problems.

> Since I have no expert knowledge about this topic, I would be
> interested in the general attitude towards "partitions on SCSI
> DVD-RAM media / SCSI removable devices":
> - Are partitions intentionally not supported? If so, why?

Historical coincidence.

> - Does it usually work but not with my specific DVD-RAM model?
>   If so, why?
> - Do you think that this should be fixed?

There's no nice way to fix it unfortunately.

> Please note that personally, I can live with the "losetup hack"
> since it is easy enough to write a program which encapsulates
> partition mounting. However, there might be people which would
> prefer plugging in a (possibly pre-)partitioned medium and
> having the partitions work out of the box in the expected way.

It would probably be better to use device-mapper than the loop device.
I think there's already userland partition parsing code for dm, and
having a simple command line tool to do that, and maybe even automatically
run through udev and creating /dev/sr<num>p<partition> devices would
be very nice to have as an almost invisible workaround.