[libvirt] [PATCH 1/2] qemu: Support SPICE listen over unix socket

[Pavel Hrdina]

On Tue, Mar 22, 2016 at 07:52:12AM -0400, Cole Robinson wrote:
> On 03/22/2016 04:59 AM, Pavel Hrdina wrote:
> > On Tue, Mar 22, 2016 at 09:49:34AM +0100, Christophe Fergeau wrote:
> >> Hey,
> >>
> >> On Tue, Mar 22, 2016 at 09:13:44AM +0100, Pavel Hrdina wrote:
> >>> On Mon, Mar 21, 2016 at 07:30:44PM -0400, Cole Robinson wrote:
> >>>> Add support for SPICE listen over unix socket. This has been in qemu
> >>>> since v2.3. The XML is:
> >>>>
> >>>>   <spice socket='/path/to/socket'/>
> >>>>
> >>>> Which matches support for VNC listen over unix socket.
> >>>>
> >>>> https://bugzilla.redhat.com/show_bug.cgi?id=1151761
> >>
> >>>> diff --git a/tests/qemuxml2argvdata/qemuxml2argv-graphics-spice-unix.xml b/tests/qemuxml2argvdata/qemuxml2argv-graphics-spice-unix.xml
> >>>> new file mode 100644
> >>>> index 0000000..6c6be44
> >>>> --- /dev/null
> >>>> +++ b/tests/qemuxml2argvdata/qemuxml2argv-graphics-spice-unix.xml
> >>>> @@ -0,0 +1,27 @@
> >>>> +<domain type='qemu'>
> >>>> +  <name>QEMUGuest1</name>
> >>>> +  <uuid>c7a5fdbd-edaf-9455-926a-d65c16db1809</uuid>
> >>>> +  <memory unit='KiB'>219136</memory>
> >>>> +  <currentMemory unit='KiB'>219136</currentMemory>
> >>>> +  <vcpu placement='static'>1</vcpu>
> >>>> +  <os>
> >>>> +    <type arch='i686' machine='pc'>hvm</type>
> >>>> +    <boot dev='hd'/>
> >>>> +  </os>
> >>>> +  <clock offset='utc'/>
> >>>> +  <on_poweroff>destroy</on_poweroff>
> >>>> +  <on_reboot>restart</on_reboot>
> >>>> +  <on_crash>destroy</on_crash>
> >>>> +  <devices>
> >>>> +    <emulator>/usr/bin/qemu</emulator>
> >>>> +    <controller type='usb' model='none' index='0'/>
> >>>> +    <controller type='pci' index='0' model='pci-root'/>
> >>>> +    <input type='mouse' bus='ps2'/>
> >>>> +    <input type='keyboard' bus='ps2'/>
> >>>> +    <graphics type='spice' socket='/tmp/spice.socket'/>
> >>>
> >>> This is an old way to specify listen type.  It would be better to add a new
> >>> <listen type='socket' socket='/tmp/spice.socket'/>.  Actually I'm working on
> >>> this support and I have my patches almost finished but they depends on this
> >>> patch series:
> >>>
> >>> https://www.redhat.com/archives/libvir-list/2016-March/msg00631.html
> >>>
> >>> You can see my progress there:
> >>>
> >>> https://github.com/Antique/libvirt/tree/spice-unix-socket
> >>
> >> We briefly discussed about this with Cole on IRC, and what you suggest
> >> is indeed the way forward. His patch is a quick way to get
> >> feature-parity with what VNC does, and we were not aware of your series.
> >>
> >> You seem to be adding
> >> <listen type='socket' socket='/tmp/foo.sock'/>
> >>
> >> I think <listen type='unix' path='/tmp/foo.sock'/> would be more
> >> consistent with what is done for channels, /interface/source, ...
> >>
> 
> Agreed, that seems more consistent to me.
> 
> > 
> > I know, Cole's patch is an easy and quick way but we will have to deal with it
> > for the rest of libvirt's lifetime.  That's why I've ended with rather larger
> > series that cleanup a lot of staff and introduce new listen type instead.
> > 
> 
> Right, but we already have to deal with that <graphics socket=.../> format for
> VNC... not adding that to SPICE is also inconsistent. Just pointing it out,
> doesn't matter to me much either way

That's a good point actually but I would rather have this inconsistency rather
than duplicating this information.  We do the same for <listen type='address'/>
but for backward compatibility and the same rule would apply to the VNC case
just to by compatible with old applications.  This is a new design and a new
feature and I would like to avoid to having <graphics socket=.../> for SPICE.

Pavel