[libvirt] [PATCH 31/32] qemu: Use the 'device_id' property of SCSI disks to avoid regressing

[Daniel P. Berrangé]

On Thu, Jan 31, 2019 at 01:56:57PM +0100, Peter Krempa wrote:
> On Thu, Jan 31, 2019 at 12:31:58 +0000, Daniel Berrange wrote:
> > On Thu, Jan 31, 2019 at 01:26:24PM +0100, Peter Krempa wrote:
> > > On Tue, Jan 29, 2019 at 16:06:35 +0000, Daniel Berrange wrote:
> > > > On Mon, Jan 28, 2019 at 05:19:00PM +0100, Peter Krempa wrote:
> > > > > QEMU accidentally exposed the id of -drive (or same value as disk
> > > > > serial, if provided) in one of the identifiers visible from the guest.
> > > > > 
> > > > > To avoid regression in case when -blockdev will be used we need to
> > > > > always specify it ourselves.
> > > > > 
> > > > > Signed-off-by: Peter Krempa <pkrempa at redhat.com>
> > > > > ---
> > > > >  src/qemu/qemu_command.c                       | 22 +++++++++++++++++++
> > > > >  .../controller-virtio-scsi.x86_64-latest.args | 20 ++++++++---------
> > > > >  .../disk-cache.x86_64-latest.args             |  4 ++--
> > > > >  .../disk-scsi-device-auto.x86_64-latest.args  |  3 ++-
> > > > >  .../disk-scsi.x86_64-latest.args              | 16 ++++++++------
> > > > >  .../disk-shared.x86_64-latest.args            |  5 +++--
> > > > >  ...threads-virtio-scsi-pci.x86_64-latest.args |  4 ++--
> > > > >  7 files changed, 50 insertions(+), 24 deletions(-)
> > > 
> > > [...]
> > > 
> > > > I rather wish there was a way we could avoid exposing the alias ID to every
> > > > guest forever more.
> > > > 
> > > > QEMU could achieve that with machine type versioning, so that it only exposes
> > > > the drive ID to guests with legacy machine types for sake of backport. We need
> > > > to explicitly set this though, as with -blockdev QEMU can't do the right thing
> > > > for legacy machine types as it lacks tie drive ID entirely. Once we set it,
> > > > we enable it for non-legacy machine types too :-(  Annoyingly I don't see a
> > > > way out of this mess such that libvirt only enables the back compat for
> > > > existing guests.
> > > 
> > > Do you mean that for any new machine type we should not put the -drive
> > > ID as the 'device_id' property? AFAIK qemu should now handle that
> > > correctly by not creating any ID.
> > > 
> > > In that case we could tie -blockdev to the new machine type only and
> > > thus will not need to add any code to format the backward compatibility
> > > stuff.
> > 
> > The tricky thing is how to tie the -blockdev to the new machine type.
> > Libvirt generally aims to treat the version part of the machine type
> > as an opaque string, because it can be arbitrarily changed by distros,
> > so there's no reliable way to determine "newest" version for a machine
> > type. At best we know that one of them is the default and so the latest,
> > but we don't know anything about ordering of others.
> 
> Well, if we don't want to look at the machine type I don't see much other
> options than to just always add the property forever.

The only thing I can imagine working is if QEMU extends the 'query-machines'
QMP data to add a flag "use-blockdev" to indicate whether it is ok to use
blockdev or not. This feels like too much of a special case hack to
warrant exposing in QMP though.

So I think we'll just have to live with the gross property forever

Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|