Debian Bug report logs - #738863
openvswitch-switch: TCP does not work properly with virtio-net when 3.12 kernel and tx/rx offloading is on

Toggle useless messages

Message #5 received at submit@bugs.debian.org (full text, mbox, reply):

From: Gary Hawkins <gary.hawkins@garyhawkins.me.uk>

To: Debian Bug Tracking System <submit@bugs.debian.org>

Subject: openvswitch-switch: TCP does not work properly with virtio-net when 3.12 kernel and tx/rx offloading is on

Date: Thu, 13 Feb 2014 16:05:53 +0000

Source: openvswitch-switch
Version: 1.9.3+git20131029-1.1
Severity: important

Dear Maintainer,

   * What led up to the situation?

I ran apt-get upgrade/aptitude safe-upgrade on the host to upgrade my testing box to latest testing packages.

   * What exactly did you do (or not do) that was effective (or
     ineffective)?

After installing the latest 3.12 kernel (linux-image-3.12-1-amd64, version 3.12.9-1) from testing causes TCP connections to stop working on qemu/kvm connected to openvswitch by way of a virtio-net tap 
device.  Packet traces show lots of TCP retransmission errors.  Very occasionally a TCP connection works, but is dropped soon after.  Most often the connection stops working entirely.

If I run the command:
ethtool -K <virtio-net tap device> tso off gro off gso off tx off rx off

everything starts working again.

Going back to 3.11 kernel also makes things work.

   * What was the outcome of this action?

Using 3.12 kernel from testing causes TCP connections to stop working on qemu/kvm guests and generates large numbers of TCP retransmission errors

   * What outcome did you expect instead?

Using 3.12 kernel from testing should not break TCP connections on qemu/kvm guests



-- System Information:
Debian Release: jessie/sid
  APT prefers testing
  APT policy: (990, 'testing'), (500, 'unstable'), (500, 'stable')
Architecture: amd64 (x86_64)

Kernel: Linux 3.12-1-amd64 (SMP w/1 CPU core)
Locale: LANG=en_GB.UTF-8, LC_CTYPE=en_GB.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/dash

Message #8 received at 738863@bugs.debian.org (full text, mbox, reply):

From: Andrei POPESCU <andreimpopescu@gmail.com>

To: Gary Hawkins <gary.hawkins@garyhawkins.me.uk>, 738863@bugs.debian.org

Cc: openvswitch-switch@packages.debian.org

Subject: Re: Bug#738863: openvswitch-switch: TCP does not work properly with virtio-net when 3.12 kernel and tx/rx offloading is on

Date: Thu, 13 Feb 2014 19:56:30 +0200

[Message part 1 (text/plain, inline)]

Control: reassign -1 src:openvswitch 1.9.3+git20131029-1.1

On Jo, 13 feb 14, 16:05:53, Gary Hawkins wrote:
> Source: openvswitch-switch
> Version: 1.9.3+git20131029-1.1
> Severity: important
> 
> Dear Maintainer,
> 
>    * What led up to the situation?
> 
> I ran apt-get upgrade/aptitude safe-upgrade on the host to upgrade my testing box to latest testing packages.
> 
>    * What exactly did you do (or not do) that was effective (or
>      ineffective)?
> 
> After installing the latest 3.12 kernel (linux-image-3.12-1-amd64, version 3.12.9-1) from testing causes TCP connections to stop working on qemu/kvm connected to openvswitch by way of a virtio-net tap 
> device.  Packet traces show lots of TCP retransmission errors.  Very occasionally a TCP connection works, but is dropped soon after.  Most often the connection stops working entirely.
> 
> If I run the command:
> ethtool -K <virtio-net tap device> tso off gro off gso off tx off rx off
> 
> everything starts working again.
> 
> Going back to 3.11 kernel also makes things work.
> 
>    * What was the outcome of this action?
> 
> Using 3.12 kernel from testing causes TCP connections to stop working on qemu/kvm guests and generates large numbers of TCP retransmission errors
> 
>    * What outcome did you expect instead?
> 
> Using 3.12 kernel from testing should not break TCP connections on qemu/kvm guests
> 
> 
> 
> -- System Information:
> Debian Release: jessie/sid
>   APT prefers testing
>   APT policy: (990, 'testing'), (500, 'unstable'), (500, 'stable')
> Architecture: amd64 (x86_64)
> 
> Kernel: Linux 3.12-1-amd64 (SMP w/1 CPU core)
> Locale: LANG=en_GB.UTF-8, LC_CTYPE=en_GB.UTF-8 (charmap=UTF-8)
> Shell: /bin/sh linked to /bin/dash

-- 
http://wiki.debian.org/FAQsFromDebianUser
Offtopic discussions among Debian users and developers:
http://lists.alioth.debian.org/mailman/listinfo/d-community-offtopic
http://nuvreauspam.ro/gpg-transition.txt

[signature.asc (application/pgp-signature, inline)]

Message #19 received at 738863@bugs.debian.org (full text, mbox, reply):

From: "Bernhard M. Wiedemann" <bwiedemann@suse.de>

To: 738863@bugs.debian.org

Subject: Re: Bug#738863

Date: Tue, 01 Jul 2014 15:24:59 +0200

We hit the same bug with a Linux-3.12 kernel and openvswitch in SLES-12
and I built a reproducer without openvswitch with just two network
interfaces, one VLAN and one DNAT+SNAT rule.
Also some more analysis of what goes wrong

http://www.spinics.net/lists/netdev/msg287640.html

=> this is actually a kernel bug that can be triggered by openvswitch

Message #24 received at 738863-done@bugs.debian.org (full text, mbox, reply):

From: Thomas Goirand <zigo@debian.org>

To: 738863-done@bugs.debian.org

Subject: Closing this bug

Date: Fri, 2 Feb 2018 17:44:44 +0100

Since this is a kernel bug, and that we're not even shipping this
version 3.12 in Debian (Wheezy has 3.2, Jessie has 3.16), this bug can
be closed.

Cheers,

Thomas Goirand (zigo)

Send a report that this bug log contains spam.