Re: blktap2 implementation in NetBSD

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Fri, Jul 29, 2011 at 09:44:42AM +0200, Roger Pau Monné wrote:
> > I was making more of a general observation about netbsd-iscsi-initiator and
> > such like rather than specifics of how it might work within Xen. Obviously
> > the more flexible the better as one might want to deal with qcow2 files from
> > NetBSD in general.
> 
> You where right, because block devices cannot be implemented inside
> FUSE, you can create a block device, but operations with block devices
> are handled directly by the kernel, never passed to the filesystem
> driver, so although I can create a block or character device, only
> major and minor numbers are read, and I never get the actual
> read/write operations on the block driver. The only option to create a
> virtual block device is to use a regular file inside a virtual
> filesystem and mount it using vnd, which seems overly complicated. The
> most interesting part about FUSE was to be able to create a "standard"
> implementation, that could be shared amongst other OSes supporting
> FUSE.
> 
> > It's actually part of the Google Summer of Code:
> > http://blog.netbsd.org/tnf/entry/netbsd_s_google_summer_of
> > http://www.mail-archive.com/netbsd-soc-cvs%lists.sourceforge.net@localhost/msg02811.html
> 
> What do you think about this Manuel? It's worth to implement something
> like blktap using libvdisk or FUSE+vnd?

I don't like the FUSE+vnd option. libvdisk (or using pud(4) directly)
is probably the way to go.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--