[libvirt] [PATCH 0/3] Prevent removing a in-used static bridge and destroying a in-used virtual network

[Laine Stump]

On 02/03/2015 11:47 AM, Michal Privoznik wrote:
> On 02.02.2015 15:08, Lin Ma wrote:
>> * Get the live state info of a virtual network through netcf in networkGetXMLDesc.
>> * Add --system flag for net-dumpxml to show the live state info.
>> * Check the live state info in net-destroy.
>> * Add --force flag for net-destroy to forcibly destroy the virtual network.
>> * Check the transient interfaces info in iface-unbridge.
>>
>> ---
>> Lin Ma (3):
>>   bridge_driver: Return the live state info of a given virtual network
>>   virsh: prevent destroying a in-used network for net-destroy
>>   virsh: prevent removing a in-used bridge for iface-unbridge
>>
>>  include/libvirt/libvirt-network.h    |   1 +
>>  src/Makefile.am                      |   3 +
>>  src/network/bridge_driver.c          | 141 ++++++++++++++++++++++++++++++++++-
>>  src/network/bridge_driver_platform.h |   7 ++
>>  tests/Makefile.am                    |   4 +
>>  tools/virsh-interface.c              |  25 ++++++-
>>  tools/virsh-network.c                |  62 ++++++++++++++-
>>  tools/virsh.pod                      |   8 +-
>>  8 files changed, 241 insertions(+), 10 deletions(-)
>>
> So I've spent some time thinking about this. I don't really like the
> idea of producing completely different XML (it doesn't even have
> <network/> as its root element!). But I see what you're trying to
> achieve. How about putting the bridged interfaces into the network
> definition (on request signalized by a flag, of course).

I think that if we're going to add the list of connected guest devices
to a network's status, that it should just always be there - adding a
new flag for every little bit of different status sets a bad precedent
and sets us up to have an infinitely increasing number of flags.

Like I mentioned in my response to one of the patches, I think this
information is better gathered and maintained by the bridge driver as
guests connect to / disconnect from a network; this avoids the necessity
of dragging netcf into the picture and allows much better information -
the name of the domain using each interface can be included.

>
> I think we've come to a point where we need to introduce live and config
> XML (like we have for domains). We already have that to some extent, but
> not truly.

I think the situation is more that we have live and config XML, but
there may be a small problem here and there. As a matter of fact, a long
time ago virsh wasn't using the live XML to edit a network, and this was
causing problems (e.g. if you tried to edit a network twice without
restarting it, the second time you would lose all the changes from the
first edit).


>
> On the other hand, looking into virNetworkDefFormat(), there's this
> '@connections' attribute to the <network/> element which is a nonzero
> integer if the network is in use. Won't that suffice for what you're
> trying to achieve?

In my opinion, yes - that's exactly the type of thing it was intended
for :-)