Re: [libvirt] [PATCH 2/3] qemu: remove unnecessary setting of tap device online state
25 Авг 2016 г. 21:38 пользователь "Laine Stump" <laine(a)laine.org>
написал:
On 08/25/2016 02:19 PM, Vasiliy Tolstov wrote:
>
> 25 Авг 2016 г. 19:18 пользователь "Laine Stump" <laine(a)laine.org>
написал:
> >
> > On 08/25/2016 05:42 AM, Vasiliy Tolstov wrote:
> >>
> >> 25 Авг 2016 г. 12:34 пользователь Vasiliy Tolstov
<v.tolstov(a)selfip.ru>
написал:
> >> >
> >> > 25 Авг 2016 г. 8:58 пользователь "Laine Stump"
<laine(a)laine.org>
написал:
> >> > >
> >> > > The linkstate setting of an <interface> is only meant to
change
the
> >> > > online status reported to the guest system by
the emulated network
> >> > > device driver in qemu,
> >> >
> >> I need to set host side status of interface. Without this live
migration with dinamic routing software (ospf with quagga or bird) bring
packet drops. Because on dest interface in up state and kernel try to
forward packets to it, but guest CPU is not running.
> >
> >
> > That shouldn't be a problem, since the IPInfo isn't added to the tap
device until immediately before the guest CPU is started on the destination
(that's the purpose of qemuInterfaceStartDevice()).
> >
> >
> >
> >> Also host side status needed for easy blackhole traffic to guest ip.
> >
> >
> > Is this something you need to do while the guest is already running?
If
not, then I think we don't need anything extra.
> >
> >
> >
> >> May be create inside the source link state attribute for host side
link status? So it consistent with ip and route elements?
> >
> >
> > If necessary, that might be the right solution, although I still think
it's better to not set the tap device offline, in case it's connected to a
bridge - we wouldn't want to trigger an STP forward delay. Maybe just
delete (and later re-add) the IPInfo would be less disruptive? (Or it might
be *more* disruptive, we'd have to try both).
> >
> >
> Thanks for info.
> Why not add ability to specify device state from host side? If attribute
is
empty think that device is up. This is reasonable default. I'm use link
status when vm running, for example if we have ddos - I down tap and via
ospf route deleted and traffic blackholed.
Yeah, I can see the utility of that. And as long as the default is up,
then nobody
is surprised by the results.
So what we're talking about is a new subelement of <source> for any
tap-based interface type (at least):
<interface type='ethernet|bridge|network'>
<source>
<link state='down'/>
</source>
Since it also needs to be supported for qemuDomainChangeNet(), I'm
doubtful
this can be done prior to the freeze tonight or early tomorrow,
since DV is in China) though. So will it be okay to have the patches I've
made in this release (which should handle proper operation for everything
except the "need to modify the state while the guest is running" case)? I
don't think any of that will need to be *un*done to support this new
attribute, and it would be nice to have something in the next release that
works at least in the default situation...
Thanks, I'm happy with this.
Attachments:
- attachment.html (text/html — 4.1 KB)