Re: RFC: rtnetlink problems with Cisco enic and VFs

[David Gibson <dgibson@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 2837 bytes --]

On Wed, 23 Apr 2014 02:33:06 +0000
"Christian Benvenuti (benve)" <benve@cisco.com> wrote:

> > -----Original Message-----
> > From: netdev-owner@vger.kernel.org [mailto:netdev-
> > owner@vger.kernel.org] On Behalf Of David Miller
> > Sent: Tuesday, April 22, 2014 6:17 PM
> > To: dgibson@redhat.com
> > Cc: gregory.v.rose@intel.com; ben@decadent.org.uk;
> > netdev@vger.kernel.org; Christian Benvenuti (benve); Sujith Sankar
> > (ssujith); govindarajulu90@gmail.com; Neel Patel (neepatel);
> > nistrive@cisco.com
> > Subject: Re: RFC: rtnetlink problems with Cisco enic and VFs
> > 
> > From: David Gibson <dgibson@redhat.com>
> > Date: Wed, 23 Apr 2014 11:12:03 +1000
> > 
> > > And therein lies the problem.  I don't even know what the
> > > IFLA_VF_PORTS info is for, but presumably something uses it.  If they
> > > stop receiving it, they can be expected to break horribly.
> 
> In the case of Enic, libvirt uses IFLA_VF_PORTS in the context of the port profile
> (see virtualport section and 802.1Qbh in the libvirt network xml documentation).
> 
> As Miller said, libvirt and iproute2 are the two known users (with libvirt being the main one)
> but you never know what else may be using it.

Ah, yes, I see it in libvirt.

I don't think it's used in iproute2 though, at least not in the master
branch.

$ git remote show origin
* remote origin
  Fetch URL:
git://git.kernel.org/pub/scm/linux/kernel/git/shemminger/iproute2.git
Push  URL:
git://git.kernel.org/pub/scm/linux/kernel/git/shemminger/iproute2.git
HEAD branch: master Remote branches: iproute-3.5.1     tracked
    master            tracked
    net-next          tracked
    net-next-3.11     tracked
    net-next-for-3.13 tracked
  Local branch configured for 'git pull':
    master merges with remote master
  Local ref configured for 'git push':
    master pushes to master (up to date)
$ git rev-parse HEAD
ce3436ca05ee2a9f4bf4c5b10eb25638865772cb
$ git grep IFLA_VF_PORTS
include/linux/if_link.h:	IFLA_VF_PORTS,
include/linux/if_link.h: *		[IFLA_VF_PORTS]

Those are declaration and comment only, no actual uses.

> > We did the same thing to VFINFO list.
> 
> I guess you refer to Bugzilla 889319.
> 
> > All the users I could find already set the mask unconditionally for all device
> > dumps.
> > 
> > It's absolutely, positively, the only reasonable fix for this problem.
> 
> The fix based on IFLA_EXT_MASK seems reasonable to me
> (IFLA_EXT_MASK is in use in libvirt >= 1.0.3 and iproute2 >=3.4.0 based on a quick check).

Ok.  And conveniently for me it looks like the EXT_MASK fix is also
backported into the RHEL libvirt.

Alright, I'll whip up a patch series that makes the IFLA_VF_PORTS
information conditional on the RTEXT_FILTER_VF flag.

-- 
David Gibson <dgibson@redhat.com>

[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]