Re: [Qemu-devel] detecting seccomp sandbox capability via QMP

["Daniel P. Berrange" <berrange@redhat.com>]

On Thu, Dec 06, 2012 at 08:00:56AM -0600, Anthony Liguori wrote:
> "Daniel P. Berrange" <berrange@redhat.com> writes:
> 
> > On Tue, Dec 04, 2012 at 03:44:54PM -0600, Anthony Liguori wrote:
> >> "Daniel P. Berrange" <berrange@redhat.com> writes:
> >> 
> >> > On Tue, Dec 04, 2012 at 01:13:46PM -0600, Anthony Liguori wrote:
> >> >> "Daniel P. Berrange" <berrange@redhat.com> writes:
> >> >> 
> >> >> >
> >> >> > In the absence of any way to detect it via QMP, libvirt should fallback
> >> >> > to hardcoding it based on the version number. This presumes that QEMU was
> >> >> > built with it enabled in configure, but we've no other option for current
> >> >> > released 1.2/1.3 versions.
> >> >> 
> >> >> echo quit | qemu -machine none -S -monitor stdio -vnc none -sandbox on
> >> >> 
> >> >> A non-zero execute means QEMU doesn't support the option.  This will
> >> >> work for any new command line option introduction and can be considered
> >> >> a "supported" way of probing for whether options are supported.
> >> >
> >> > One of the significant benefits to libvirt of the QMP based feature
> >> > detection, was that we no longer have to invoke QEMU multiple times
> >> > to query different data. I don't want to regress in this regard,
> >> > because invoking QEMU many times has a noticable performance impact
> >> > for some applications eg virt-sandbox were even 100ms delays are
> >> > relevant.  So while what you describe does work, I don't think it
> >> > is a satisfactory approach for libvirt.
> >> 
> >> Okay, so in terms of what exists today, I don't have a better option.
> >> But we could add:
> >> 
> >> { 'enum': 'ConfigEntryType',
> >>   'data': [ 'number', 'string', 'bool', 'size' ] }
> >> 
> >> { 'type': 'ConfigEntry',
> >>   'data': { 'name': 'str', 'type': 'ConfigEntryType' } }
> >> 
> >> { 'type': 'ConfigSection',
> >>   'data': { 'name': 'str', 'fields': [ 'ConfigEntry' ] } }
> >> 
> >> { 'command': 'query-config-schema',
> >>   'returns': [ 'ConfigSection' ] }
> 
> 
> >> 
> >> This technically introspects config sections but obviously could be used
> >> to detect the availability of -sandbox.
> >> 
> >> If it's useful, I can take a quick swing at implementing (or someone
> >> else certainly could).
> >
> > I'm not sure I entirely understand what information a 'ConfigSection'
> > would represent. By config here, do you mean any command line argument
> > or something else ?
> 
> We no longer should be adding command line arguments that don't use
> QemuOpts and have a equivalent -readconfig syntax.  We could even
> eliminate new options and do something like:
> 
> qemu -conf sandbox:enable=on
> 
> But that's not user friendly so we'll stick with adding higher level
> options like -sandbox.
> 
> So what I'm proposing is to introspection on what -readconfig supports
> and then from that, you can infer when new higher level command line
> arguments are added.
> 
> >  Could you give a short example of the actual JSON
> > you envisage returning for this schema. Your suggestion sounds good,
> > but I want to make sure I'm not mis-understanding things :-)
> 
> [ { 'name': 'sandbox',
>     'fields': [ { 'name': 'enable', 'type': 'bool' } ] },
>   { 'name': 'add-fd',
>     'fields': [ { 'name': 'fd', 'type': 'number' },
>                 { 'name': 'set', 'type': 'number' },
>                 { 'name': 'opaque', 'type': 'str' } ] },
>   ...
> ]

Ok, that all sounds like a good idea to me - it should address one of the
major gaps in the new QMP based capabilities detection.

Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|