[libvirt] [PATCH 2/2] qemu: Emit compatible XML when migrating a domain

[Jiri Denemark]

On Fri, May 04, 2012 at 17:28:39 -0600, Eric Blake wrote:
> On 05/04/2012 04:40 PM, Jiri Denemark wrote:
> > When we added the default USB controller into domain XML, we efficiently
> > broke migration to older versions of libvirt that didn't support USB
> > controllers at all (0.9.4 and earlier) even for domains that doesn't use
> 
> s/doesn't/don't/
> 
> > anything that the older libvirt can't provide. We still want to present
> > the default USB controller in any XML seen by a user/app but we can
> > safely remove it from the domain XML used during migration. If we are
> > migrating to a new enough libvirt, it will add the controller XML back,
> > while older libvirt won't be confused with it although it will still
> > tell qemu to create the controller.
> > 
> > Similar approach can be used in the future whenever we find out we
> > always enabled some kind of device without properly advertising it in
> > domain XML.
> 
> Memballoon would be such a device, if we care about migration to even
> older libvirt.

Right.

> > ---
> >  src/qemu/qemu_domain.c    |   60 ++++++++++++++++++++++++++++++++++++++++----
> >  src/qemu/qemu_domain.h    |   10 +++++--
> >  src/qemu/qemu_driver.c    |   21 +++++++++------
> >  src/qemu/qemu_migration.c |   10 ++++---
> >  src/qemu/qemu_process.c   |    8 +++---
> >  5 files changed, 83 insertions(+), 26 deletions(-)
> > 
> > diff --git a/src/qemu/qemu_domain.c b/src/qemu/qemu_domain.c
> > index b105644..3752ddf 100644
> > --- a/src/qemu/qemu_domain.c
> > +++ b/src/qemu/qemu_domain.c
> > @@ -1227,11 +1227,14 @@ int
> >  qemuDomainDefFormatBuf(struct qemud_driver *driver,
> >                         virDomainDefPtr def,
> >                         unsigned int flags,
> > +                       bool compatible,
> >                         virBuffer *buf)
> 
> Rather than adding a new 'compatible' flag and having to adjust _every_
> caller, could we instead add a new VIR_DOMAIN_XML_INTERNAL flag to be
> or'd in to flags only for the callers that care?  On the other hand,
> we'd have to filter that flag back out before calling into domain_conf,
> so I guess this approach is as good as any.

Yeah, we would need to filter it out and we would need to either move the
internal flags enum to a more public place (it's internal to domain_conf.c
now) or make another enum and make sure neither of them conflicts with the
real flags enum :-) I decided to introduce new bool parameter since the
compiler checked are places I need to change for me and it's more explicit so
that any new caller of this api needs to explicitly think about the value to
pass for compatibility.

> > +        if (usb && usb->idx == 0 && usb->model == -1) {
> 
> Just like memballoon, if we ever want to explicitly represent an XML
> with no usb controller, we will have to provide <controller type='usb'
> mode='none'>; but there, the model would not be -1, so this would
> properly be migrated.

Right and we already have a request to do just this. BTW, I don't explicitly
check the PCI address assigned to the controller because the default
automatically added USB1 controller must always have PCI address 0:0:1.2. When
a domains is configured with different address assigned to the default USB
controller, libvirt will refuse to start it.

> ACK.

Thanks. I pushed this series.

Jirka