[libvirt] [PATCH] libvirt no longer supports qemu format=host_device

[Daniel P. Berrange]

On Mon, Nov 08, 2010 at 12:08:26PM -0600, Jamie Strandboge wrote:
> On Mon, 2010-11-08 at 10:02 +0000, Daniel P. Berrange wrote:
> > On Fri, Nov 05, 2010 at 02:04:50PM -0500, Jamie Strandboge wrote:> 
> > > This is confirmed as not working on 0.8.3 and 0.8.5. The attached patch
> > > against 0.8.5 fixes the issue and restores host_device support.
> > 
> > host_device is not a data format type, but rather a qemu driver protocol.
> > As such it is not applicable for the <driver type=XXX> field, because
> > the protocol is already specified at the <disk> element level with
> > a combo of the type+device fields. If it happened to be possible to
> > set it in the <driver>  field that is just pure chance & not something
> > that is supported. If we need to setup host_device anywhere, then this
> > needs to be dealt with at the qemu_conf.c where we generate the CLI
> > args.  
> 
> Hmm... that is not clear in the documention IMHO (and the patch was
> careful to only use this for the qemu driver). Eg from the qemu-img man
> page:
> 
> "Supported image file formats:
>     raw ...
>     host_device ...
>     qcow2 ...
>     ...
> "
> 
> Ie, the language in the qemu-img man page has 'host_device' at the same
> level as 'raw', 'qcow2', 'cow', etc as a 'file format'.
> http://libvirt.org/formatdomain.html#elementsDisks only talks about a
> 'sub-type' for the qemu driver (but it is easy to infer that it is
> looking at the qemu-img types after using libvirt for a while) and
> http://libvirt.org/storage.html does not list 'host_device' as a 'format
> type', but it does list all of the other ones from the same section of
> the qemu-img man page (except 'vdi', which is maybe an omission in the
> libvirt docs?). Finally, 'man virsh' does list 'host_device' as a 'file
> format' in its 'vol-create-as' command.

libvirt isn't intended to expose all qemu/qemu-img terminology
directly. It isn't listed in the storage.html or formatdomain.html 
pages, because 'host_device' isn't a data format, rather it is a 
block driver access method. QEMU itself has been somewhat confused
about the distinction of format vs driver access method, due to the
fact that they've both implemented with the same code inside QEMU.
You basically have two separate sets though

 formats: raw, cow, qcow, qcow2, vmdk, etc
 drivers: file, host_device, host_cdrom, host_floppy

At the XML level, there is no need to specify the QEMU driver
because this is already represented at the main <disk> element
by the type+device attributes. The fact that it happened to be
possible to indirectly use it via the <driver> format attribute
is just a lucky coincidence of the way QEMU parses the command
line arg for 'format='. It is not intentionally supported by
QEMU in this manner, nor by libvirt. Ultimately libvirt QEMU
driver never sets any driver protocol when generating QEMU
argv because thus far it has never been required, since QEMU
auto chooses the driver protocol correctly.

The virsh manpage reference is a bug unfortunately introduced in
b28a068b430d430d08b7b72fdc2ffc2deb58fa4c. 

Anyone who has tried to set host_device in the <driver> can
just remove that setting.

Regards,
Daniel
-- 
|: Red Hat, Engineering, London    -o-   http://people.redhat.com/berrange/ :|
|: http://libvirt.org -o- http://virt-manager.org -o- http://deltacloud.org :|
|: http://autobuild.org        -o-         http://search.cpan.org/~danberr/ :|
|: GnuPG: 7D3B9505  -o-   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|