On 02/21/2014 03:58 PM, Laine Stump wrote:
Although the immediate reason for all these patches is $subject, it
really is something that should have been done a long time ago (I just
hadn't convinced myself it was the right thing to do). These patches
will allow a management application to easily learn exactly what
physical hardware is being used by a domain's interface, which wasn't
previously possible (e.g., it will be simple to learn which SRIOV VF
is being used by a domain interface that is configured as <interface
type='network'> where the network is a pool of VFs).
Laine Stump (7):
conf: clarify what is returned for actual bandwidth and vlan
conf: handle null pointer in virNetDevVlanFormat
conf: make virDomainNetDefFormat a public function
conf: re-situate <bandwidth> element in <interface>
conf: new function virDomainActualNetDefContentsFormat
conf: output actual netdev status in <interface> XML
network: include plugged interface XML in "plugged" network hook
I just made the few commit log and code changes pointed out, and pushed
these.
There are two other bugs that I've noticed while testing out the network
hooks, and that I'm hoping to fix before the release:
* If a domain start fails, all interfaces get their "unplugged"
hook called,even if they were never "plugged". This implies that
the networkReleaseActualDevice has been called possibly when it
shouldn't (or that networkReleaseActualDevice() is doing more
than itshould with interfaces that were never allocated).
* It's not allowed to add "floor" in a bandwidth within a portgroup,
i.e. the validation treats it as if it were a network <bandwidth>
element rather than an interface <bandwidth>.