[libvirt] [PATCH v2 3/3] virsh: Add support for text based polkit authentication

[Daniel P. Berrange]

On Fri, Feb 12, 2016 at 10:04:58AM -0500, John Ferlan wrote:
> 
> 
> On 02/12/2016 08:22 AM, Daniel P. Berrange wrote:
> > On Fri, Feb 12, 2016 at 07:53:40AM -0500, John Ferlan wrote:
> >>
> >>
> >> On 02/12/2016 06:57 AM, Daniel P. Berrange wrote:
> >>> On Fri, Feb 12, 2016 at 06:49:22AM -0500, John Ferlan wrote:
> >>>> [...]
> >>>>
> >>>>>> +        err = virGetLastError();
> >>>>>> +        if (err && strstr(err->message,
> >>>>>> +                          _("no agent is available to authenticate"))) {
> >>>>>
> >>>>>> +            if (!pkagent) {
> >>>>>> +                if (!(pkagent = virPolkitAgentCreate()))
> >>>>>> +                    goto cleanup;
> >>>>>> +            }
> >>>>>> +            agentstart++;
> >>>>>> +        } else if (err && strstr(err->message, _("authentication failed:"))) {
> >>>>>
> >>>>> String matching is pretty unpleasant. I think we can match on
> >>>>> err->domain == VIR_FROM_POLKIT && err->code == VIR_ERR_AUTH_FAILED
> >>>>> for this.
> >>>>>
> >>>>
> >>>> Using VIR_ERR_AUTH_FAILED I cannot distinguish between the failure of
> >>>> available agent or access denied by policy from virPolkitCheckAuth.
> >>>> Adjusting what virPolkitCheckAuth returns means more code modification
> >>>> since the assumption is -2 has 3 possible issues of which 2 currently
> >>>> are tested by a err->message comparison.
> >>>
> >>> My point is that you don't actually need to distinguish those two
> >>> cases directly. You can do this:
> >>>
> >>>     if (err && err->code == VIR_FROM_POLKIT && err->code == VIR_ER_AUTH_FAILED) {
> >>>          if (!virDBusIsServiceRegistered(...polkit...)) {
> >>
> >> Including "virdbus.h" to get virDBusIsServiceRegistered from virsh.c
> >> sends me down the build system rabbit hole again:
> >>
> >> In file included from virsh.c:59:0:
> >> ../src/util/virdbus.h:27:25: fatal error: dbus/dbus.h: No such file or
> >> directory
> >> compilation terminated.
> >>
> >> Adding "$(DBUS_CFLAGS)" to the virsh_CLFAGS in Makefile.am still leaves
> >> me with:
> >>
> >> virsh-virsh.o: In function `virshConnect':
> >> /home/jferlan/git/libvirt.work/tools/virsh.c:183: undefined reference to
> >> `virDBusIsServiceRegistered'
> >> collect2: error: ld returned 1 exit status
> > 
> > Oh we missed it from src/libvirt_private.syms
> > 
> 
> Ahh... that's it - didn't even consider that option...
> 
> 
> However, virDBusIsServiceRegistered:
> 
> "Retruns 0 if service is registered, -1 on fatal error, or -2 if service
> is not registered"
> 
> 
> I found passing "org.freedesktop.PolicyKit1" returns 0 every time even
> whether or not virPolkitAgentCreate has been called...  Feels like
> something like the machine name code that searches by pid would be what
> would work.
> 
> As an alternative (since this is the I want to make sure the agent is
> running path), the pkttyagent also takes a --notify-fd fd parameter. I
> can work something up to use that.

Oh, I'm an idiot in suggesting this route.  So, even when logging in via
ssh, chances are that there *is* an agent running - one run by GNOME
in your desktop login. The problem is that agent isn't able to display
prompts on the ssh text console.

So I think we would have to go back to the approach of returning a
special error code to let the client detect the problem.

So when libvirtd is able to auth due to missing agent, then we should
return a new code VIR_ERR_AUTH_UNAVAILABLE, and virsh can check for
that to see if it should launch pkttyagent.

Regards,
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 :|