qemu-discuss
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-discuss] qemu guests using 802.1q vlans bridged on host


From: Vlad Yasevich
Subject: Re: [Qemu-discuss] qemu guests using 802.1q vlans bridged on host
Date: Wed, 11 Sep 2013 17:28:54 -0400
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:17.0) Gecko/20130805 Thunderbird/17.0.8

On 09/11/2013 03:13 PM, Stephan von Krawczynski wrote:
On Wed, 11 Sep 2013 12:50:11 -0400
Vlad Yasevich <address@hidden> wrote:

Rally? I mean isn't using tagged 802.1q vlans something pretty normal? I
cannot believe that linux is incapable of doing what every 10 bucks desktop
switch and its bridge can...


Yes, it is normal.  I just tried the same on 3.10.10 kernel and it works
fine.  My config was:

#brctl show br0
bridge name     bridge id               STP enabled     interfaces
br0             8000.5254001f7aef       no              eth0
                                                          vnet0


vnet0 is just a tap interface on top of which VM is running.

Inside VM, vlan100 is configured with an address.  Another host
configured vlan100 as well and I can send traffic between the two
just fine.

-vlad

And what type of network card do you use for eth0 ?

Test was done over e1000e driver. Here are the appropriate vlan offloads that were set:

sudo ethtool -k eth0 | grep vlan
rx-vlan-offload: on
tx-vlan-offload: on
rx-vlan-filter: off [fixed]
vlan-challenged: off [fixed]
tx-vlan-stag-hw-insert: off [fixed]
rx-vlan-stag-hw-parse: off [fixed]
rx-vlan-stag-filter: off [fixed]


-vlad

--
Regards,
Stephan





--
Regards
Stephan


On Thu, 22 Aug 2013 10:58:08 -0700
Tony Su <address@hidden> wrote:

I haven't investigated what you describe so can't offer much help...

But my reaction is that if it's not possible to configure some kind of
"master vlan tag" I'd consider "packaging" all the VLANs through a VPN
just long enough to pass through any major obstacles (technical or
onerous work). Of course such an approach would likely come with
significant overhead but it's a matter of trade-offs.

Or, I suppose that you could attempt to script the creation of your
bridges and just deal with them all.

Tony

On Thu, Aug 22, 2013 at 10:49 AM, Stephan von Krawczynski
<address@hidden> wrote:
Sorry, you misunderstood my writing. I am talking of several hundred vlans
with - of course - different ids and quite some guests (around 50).
There is no way to simplify this setup besides the trivial way of a bridge
that carries all vlan-tagged interfaces. The trivial thing about it is all
these different vlans come in through one trunk. So if vlan-tagged bridging
worked I would have only one bridge interface with 50 guests connected ...

--
Regards,
Stephan



On Thu, 22 Aug 2013 10:29:59 -0700
Tony Su <address@hidden> wrote:

If you're configuring the all your "hundreds" of guests to connect to
the same VLAN, then you should able to simply configure all guests to
connect to the same working bridge device without further
configuration.

You're surely not trying to configure hundreds of individual vlans,
separate ones for each guest?

Tony

On Thu, Aug 22, 2013 at 10:04 AM, Stephan von Krawczynski
<address@hidden> wrote:
Hello Tony,

thank you for answering, my comments are inline. Just as an additional hint to
what I've tested so far. Since I found vlan bridging not working I configured
the vlan on the host and put that interface to a bridge and over to a virtio
device (non-vlan-tagged) in the guest. As you might expect this works
perfectly. Unfortunately it is not useable for me, because if you want several
hundred vlans to several guests you will end up configuring hundreds of
bridges and interfaces.


On Thu, 22 Aug 2013 09:32:42 -0700
Tony Su <address@hidden> wrote:

Have you
- Tested without VLAN tags?

Yes, works perfectly.

- Verified IP Forwarding is enabled, I usually see this implemented in
/etc/sysctl.conf and not written directly to the /proc files

Yes, forwarding is active.

- Disabled all the transparent bridge filters, typicallly at
/proc/sys/net/bridge/* again, although you can write directly to these
files I'd recommend you simply add the commands to your sysctl.conf

Yes, I played with these a bit but found out that there is no effect on my
problem.

- Verified any personal FW is configured properly.

There is none.

Tony

On Thu, Aug 22, 2013 at 7:39 AM, Stephan von Krawczynski
<address@hidden> wrote:
Hello all,

I'd like to do something very simple - at least that's what I thought
I want a guest to have access to a network just as if he was connected to the
real card, but set up as bridge on the host and virtio network driver. The
guest should be able to configure and use some or maybe even many 802.1q vlans
on this network and the traffic should go out tagged.

So I setup the hosts bridge and connected an intel network card and a qemu
virtio card. Now the problem: No vlan-tagged traffic from the physical
interface reaches the guest at all, and no vlan-tagged traffic from the guest
reaches the physical net over the bridge. One major reason for this is the
vlan offloading by the host interface card (intel). Another seems to be that
arp requests are somehow not going through the bridge for the vlans.

I hope that someone here has used 802.1q vlans inside guests before and can
share some tips how to make this work. Because out-of-the-box it does not. All
system are linux of course and with latest kernels (3.10.9 currently).
qemu is 1.5.2.
Thanks for any hints.

--
Regards,
Stephan







reply via email to

[Prev in Thread] Current Thread [Next in Thread]