On Tue, Oct 20, 2015 at 09:36:29PM +0100, Peter Maydell wrote:
> On 20 October 2015 at 19:50, Daniel P. Berrange <berra...@redhat.com> wrote:
> > On Tue, Oct 20, 2015 at 06:01:37PM +0100, Peter Maydell wrote:
> >> On 20 October 2015 at 14:39, Daniel P. Berrange <berra...@redhat.com>
> >> wrote:
> >> > On Mon, Oct 19, 2015 at 04:20:09PM -0600, Eric Blake wrote:
> >> >> On 10/12/2015 05:14 AM, Daniel P. Berrange wrote:
> >> >> > The VNC code is currently using QemuOpts to configure the
> >> >> > sockets connections / listeners it needs. Convert it to
> >> >> > use SocketAddress to bring it in line with modern QAPI
> >> >> > based code elsewhere in QEMU.
> >>
> >> >>
> >> >> > + if (to) {
> >> >> > + saddr->inet->has_to = true;
> >> >> > + saddr->inet->to = to;
> >> >> > + }
> >> >> > + saddr->inet->ipv4 = saddr->inet->has_ipv4 = has_ipv4;
> >> >> > + saddr->inet->ipv6 = saddr->inet->has_ipv6 = has_ipv6;
> >> >>
> >> >> Do we want to specify has_ipvX as true even when setting inet->ipvX to
> >> >> false?
> >> >
> >> > This saddr instance is passed into 'socket_connect' which then
> >> > converts it back into a QemuOpts object using
> >> >
> >> > bool ipv4 = addr->has_ipv4 && addr->ipv4;
> >> >
> >> > So, we don't need to set has_ipvX to true, when ipvX is false.
> >>
> >> That is not the *only* thing that inet_addr_to_opts does with
> >> the has_* flags... this kind of thing can be the difference
> >> between "force ipv4" and "use ipv4 or ipv6".
> >
> > The original code which used QemuOpts directly had this logic:
> >
> > inet_listen_opts / inet_connect_opts use this logic:
> >
> > ai.ai_family = PF_UNSPEC;
> > ....
> > if (qemu_opt_get_bool(opts, "ipv4", 0))
> > ai.ai_family = PF_INET;
> > if (qemu_opt_get_bool(opts, "ipv6", 0))
> > ai.ai_family = PF_INET6;
> >
> > IOW, these methods treat ipv4 being omitted, as equivalent to ipv4=off.
> >
> > Thus the existing VNC -> qemu-sockets code has this logic:
> >
> > VNC | qemu-sockets
> > QemuOpts | ai_family
> > ----------------------------
> > ipv4 ipv6 |
> > ----------------------------
> > - - | PF_UNSPEC
> > - off | PF_UNSPEC
> > - on | PF_INET6
> > off - | PF_UNSPEC
> > off off | PF_UNSPEC
> > off on | PF_INET6
> > on - | PF_INET
> > on off | PF_INET
> > on on | PF_INET6
> >
> >
> > My patch, which also does not distinguish ipv4=off from ipv4 omitted
> > has the following logic, at various stages of conversion between
> > QemuOpts & QAPI SocketAddress:
> >
> >
> > VNC | VNC | qemu-sockets | qemu-sockets
> > QemuOpts | QAPI SocketAddress | QemuOpts | ai_family
> > -------------------------------------------------------------------------
> > ipv4 ipv6 | has_ipv4 ipv4 has_ipv6 ipv6 | ipv4 ipv6 |
> > -------------------------------------------------------------------------
> > - - | f f f f | - - | PF_UNSPEC
> > - off | f f f f | - - | PF_UNSPEC
> > - on | f f t t | off on | PF_INET6
> > off - | f f f f | - - | PF_UNSPEC
> > off off | f f f f | - - | PF_UNSPEC
> > off on | f f t t | off on | PF_INET6
> > on - | t t f f | on off | PF_INET
> > on off | t t f f | on off | PF_INET
> > on on | t t t t | on on | PF_INET6
> >
> > So, unless I've missed something, the final ai_family is set the same
> > way before & after my patch, for all combinations of VNC ipv4/ipv6
> > CLI options.
>
> I think you're right that this patch doesn't change anything,
> but doesn't it mean that vnc still has one of the bugs we fixed in
> commit b77e7c8e99f9a: if you say 'ipv6=off' it will give you a
> PF_UNSPEC rather than an PF_INET socket? (In particular on Windows
> PF_UNSPEC will get you ipv6, which is exactly what the user asked
> not to get...)
>
> Since that's a pre-existing bug I guess we can fix it in a
> subsequent patch.
Hmm, yes, you are correct in that - I can certainly look at creating a
further patch to fix that bug.
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 :|
