[ovs-dev] Possible user-space datapath regression

[Justin Pettit]

Hi, Simon.  I've confirmed the problem you found and am looking into it.  Unfortunately, I'm traveling this week for work, so I may have limited time to look at it until Friday.

Thanks for the report.

--Justin


On Nov 20, 2012, at 2:44 AM, Simon Horman <horms at verge.net.au> wrote:

> Hi,
> 
> I believe I have found a regression that affects the behaviour of the
> datapath. A quick bisection points to the following patch:
> 
> 	acf60855126bcfa79ea22d7846af5f2efe26cd30
> 	ofproto-dpif: Use a single underlying datapath across multiple bridges.
> 
> 
> For my test I run Open vSwtich inside a kvm VM. Inside the VM
> there are two ethernet interfaces, eth0 and eth1 although only
> eth1 is used in this test. The interfaces are emulated and
> the other end of eth1 is tap1 in the host.
> 
> eth1 has been added to an Open vSwtich bridge in the VM.
> The output of ovs-vsctl show is as follows:
> 
> cff268ec-4e3d-41fa-8127-efa2bb851ae7
>    Bridge "br0"
>        Port "eth1"
>            Interface "eth1"
>        Port "br0"
>            Interface "br0"
>                type: internal
>    ovs_version: "1.9.90"
> 
> When I try to from the host to the VM I obvserve that the VM
> does not respond to ARP requests.
> 
> COnversely, when I try to ping from the VM to the host it appears
> that the VM ignores ARP responses.
> 
> If I checkout acf60855126bcfa79ea22d7846af5f2efe26cd30^ then all seems well.
> If I checkout acf60855126bcfa79ea22d7846af5f2efe26cd30 I get the behaviour
> that I have described above.
> 
> I have not debuged the problem any further yet.
>