[ovs-discuss] Adding a new vlan will caused the host lost connection

[Alex Wang]

Hey Kris,

There was a bug in the "add_vsp()" function in "ofproto/ofproto-dpif.c".
And the fix is in this patch
http://git.openvswitch.org/cgi-bin/gitweb.cgi?p=openvswitch;a=commit;h=deea120099d23fac3f687ec302351e38a21ee353
.

I think this may be the reason for your problem.

Kind Regards,
Alex Wang


On Tue, Jun 4, 2013 at 8:07 AM, Kris zhang <zhang.kris at gmail.com> wrote:

> Thanks Jesse, I solved this issue through upgrade the NIC driver igb from
> 2.1 to 2.4, now it is ok even if use Linux bond as the external port of ovs
> bridge. And i will try to use OVS bonding once i have a chance. Also i hope
> FAQ can list which NIC driver has vlan problem, and should be upgrade to
> which version.
>
> Thanks,
> Kris
>
>
> On Thu, May 30, 2013 at 7:22 AM, Jesse Gross <jesse at nicira.com> wrote:
>
>> Linux bonds don't pass vlan information through to the driver. If you
>> use OVS bonding then you shouldn't have this problem.
>>
>> On Wed, May 29, 2013 at 11:08 AM, Kris zhang <zhang.kris at gmail.com>
>> wrote:
>> > I found the reason, the problem is not caused by vlan splinters, it
>> caused
>> > by bond0. I don't know why, but if remove the bond0, and ovs-br0
>> directly
>> > connect to eth0, the problem disappears.
>> >
>> >
>> >
>> >
>> > On Mon, May 27, 2013 at 10:10 AM, Kris zhang <zhang.kris at gmail.com>
>> wrote:
>> >>
>> >> Hi Jesse,
>> >>
>> >> Because if i don't use the vlan splinters, the VMs' network traffic
>> will
>> >> be nearly 0 M/s (but they can ping each other). I read FAQ, it says the
>> >> reason maybe the NIC driver or Linux kernel version problems, and it
>> also
>> >> says the vlan splinters may solve this issue, but they don't said the
>> ping
>> >> will miss some packets when add a new vlan. is it a bug for vlan
>> splinters?
>> >> Please see the images in the attachment.
>> >>
>> >> Thanks,
>> >> Kris
>> >>
>> >>
>> >> On Thu, May 23, 2013 at 11:25 PM, Jesse Gross <jesse at nicira.com>
>> wrote:
>> >>>
>> >>> It seems that the original issue was that you were using VLANs when
>> >>> you shouldn't have. In that case, why are you trying to use VLAN
>> >>> splinters?
>> >>>
>> >>> On Thu, May 23, 2013 at 2:44 AM, Kris zhang <zhang.kris at gmail.com>
>> wrote:
>> >>> > Thanks Jesse, but i still cannot fix my issue. Because if i use vlan
>> >>> > splinters, the above issue will be happened, if i don't use vlan
>> >>> > splinters,
>> >>> > the packets between the VMs are very slow (ping is ok), i use
>> ovs-dpctl
>> >>> > dump-flows br0, get the following result:
>> >>> >
>> >>> >
>> >>> >
>> in_port(1),eth(src=c6:b0:ea:37:29:47,dst=c6:b0:3a:c0:0d:55),eth_type(0x8100),vlan(vid=3000,pcp=0),encap(eth_type(0x0800),ipv4(src=20.1.120.12,dst=20.1.120.13,proto=1,tos=1,ttl=64,frag=no),icmp(type=8,code=0)),
>> >>> > packets:101, bytes:10302, used:4.341s, actions=pop_vlan,3
>> >>> >
>> >>> >
>> in_port(3),eth(src=c6:b0:3a:c0:0d:55,dst=c6:b0:ea:37:29:47),eth_type(0x0800),ipv4(src=20.1.120.13,dst=20.1.120.12,proto=1,tos=0,ttl=64,frag=no),icmp(type=0,code=0)),
>> >>> > packets:101, bytes:9898, used:4.341s,
>> >>> > actions=push_vlan(vid=3000,pcp=0),1
>> >>> >
>> >>> > The port 1 is eth1, and the port 3 is the tap of VM.
>> >>> > Does the iptables affect the ovs? If not, i have to upgrade the
>> linux
>> >>> > kernel, or upgrade NIC driver.
>> >>> >
>> >>> > Thanks,
>> >>> > Kris
>> >>> >
>> >>> >
>> >>> >
>> >>> >
>> >>> > On Tue, May 21, 2013 at 3:32 AM, Jesse Gross <jesse at nicira.com>
>> wrote:
>> >>> >>
>> >>> >> There's an extensive section in the FAQ about vlans that I would
>> >>> >> recommend reading.
>> >>> >>
>> >>> >> On Mon, May 20, 2013 at 8:51 AM, Kris zhang <zhang.kris at gmail.com>
>> >>> >> wrote:
>> >>> >> > No, so you mean it maybe not caused by vlan splinters?
>> >>> >> >
>> >>> >> >
>> >>> >> > On Mon, May 20, 2013 at 11:18 PM, Jesse Gross <jesse at nicira.com>
>> >>> >> > wrote:
>> >>> >> >>
>> >>> >> >> On Mon, May 20, 2013 at 4:37 AM, Kris zhang <
>> zhang.kris at gmail.com>
>> >>> >> >> wrote:
>> >>> >> >> > Hi guys,
>> >>> >> >> >
>> >>> >> >> > I use ovs-1.7.1, and i run ovs on a single NIC host (CentOS):
>> >>> >> >> >
>> >>> >> >> > # ovs-vsctl add-br br0
>> >>> >> >> > # ovs-vsctl add-port br0 eth0
>> >>> >> >> >
>> >>> >> >> > Then i setup the eth0 interface's other_config:
>> >>> >> >> > enable-vlan-splinters="true".
>> >>> >> >> >
>> >>> >> >> > Last I ping this host by another PC, and at same time i do
>> this:
>> >>> >> >> >
>> >>> >> >> > # ovs-vsctl add-port taptest br0 tag=100
>> >>> >> >> >
>> >>> >> >> > I found there are 4 "Request timed out." happened. That means
>> the
>> >>> >> >> > host
>> >>> >> >> > lost
>> >>> >> >> > connect about 4 seconds.
>> >>> >> >> >
>> >>> >> >> > If i remove the vlan splinters on interface eth0. it won't
>> >>> >> >> > happen.
>> >>> >> >> > So does anybody know the reason?
>> >>> >> >>
>> >>> >> >> Is the other machine actually on that VLAN?
>> >>> >> >
>> >>> >> >
>> >>> >
>> >>> >
>> >>
>> >>
>> >
>>
>
>
> _______________________________________________
> discuss mailing list
> discuss at openvswitch.org
> http://openvswitch.org/mailman/listinfo/discuss
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20130604/c97f6432/attachment.html>