[ovs-discuss] adding MPTCP code breaks OVS kmod datapath creation

Jesse Gross jesse at nicira.com
Fri Jan 20 22:39:04 UTC 2012


On Fri, Jan 20, 2012 at 2:30 PM, Ben Pfaff <blp at nicira.com> wrote:
> On Fri, Jan 20, 2012 at 02:21:09PM -0800, Jesse Gross wrote:
>> On Fri, Jan 20, 2012 at 1:50 AM, Brandon Heller <brandonh at stanford.edu> wrote:
>> > I'm trying to run a single system that has OVS as well as the draft Linux
>> > implementation of Multi-Path TCP. ??Briefly, MPTCP uses multiple subflows to
>> > take advantage of multiple interfaces (such as WiFi + WiFi + WiMax) or
>> > multiple paths (within a Fat Tree in an ECMP-based data center). ??The
>> > project is fairly young, and more details are available
>> > at??http://mptcp.info.ucl.ac.be/. ??I did not write any of this code.
>> >
>> > On an Ubuntu 3.0.0-15-generic kernel on an Ubuntu 11.10 system w/OVS
>> > {1.2.2/1.3.0}, I have no issues w/OVS. ??When using 3.0.0-15-mptcp, which has
>> > about 11K lines of MPTCP-specific changes, plus a few small config changes,
>> > I cannot instantiate a datapath:
>>
>> It's hard to say.  They made a lot of changes to their local copy of
>> the Linux tree and likely they did something incompatible, so it's not
>> really possible to support a version of Linux with arbitrary
>> modifications to it.
>
> That was my first reaction too, but the traceback implies that the
> following WARN_ON in datapath.c:dp_ifinfo_notify() triggered and I
> really don't understand how that would happen:
>
>                        /* -EMSGSIZE implies BUG in br_nlmsg_size() */
>                        WARN_ON(err == -EMSGSIZE);
>                        goto err;

I don't know either.  Perhaps they changed a constant in one place and
didn't update it elsewhere.



More information about the discuss mailing list