On 3/22/2019 6:59 AM, Ammu wrote:
Hello,

As suggested, I tried the same with *openvswitch-2.10.1-3.el7.x86_64.rpm*.
Kernel details: *Linux 3.10.0-514.el7.x86_64*
*
*
I still see the DF bit set in the outer IP layer of the tunnelled packets. (Be it either in GRE/VxLAN configuration)

Herewith, I am attaching the OVS configuration that I have made.

And, I have attached a sample packet captured for both vxlan and gre tunnelling.

I'll have a look at it.  I note that you're using a Red Hat distribution kernel and openvswitch package.  I'll see if
there is any difference in that source package from what we expect as well.

Thanks,

- Greg


-
Keerthana



On Thu, Mar 21, 2019 at 7:40 PM Ammu <[email protected] <mailto:[email protected]>> wrote:

    Hi Greg,

    As I am doing unwell today, I didn't try it with the latest version.

    I will do it tomorrow and update you on that.

    Thank you!

    -
    Keerthana

    On Thu, Mar 21, 2019 at 4:31 AM Gregory Rose <[email protected]
    <mailto:[email protected]>> wrote:


        On 3/19/2019 6:25 PM, Ben Pfaff wrote:
        > On Tue, Mar 19, 2019 at 05:19:42PM +0530, Ammu wrote:
        >>
        
https://stackoverflow.com/questions/55223517/allow-df-not-to-be-set-on-gre-vxlan-tunnels
        > That question reads:
        >
        >      I have created a bridge with a tunnel interface(be it
        either
        >      vxlan/gre) and an internal interface. When incoming
        packets get
        >      encapsulated with either vxlan/gre tunnel header, I
        don't want the
        >      DF bit set on the outer IP layer of tunnelled packet.
        >
        >      Despite setting df_default tunnel option to False while
        creating the
        >      tunnel interface, I get the DF bit set on the outer IP
        layer of the
        >      tunnelled packet.
        >
        > Glancing through the userspace and datapath code in the
        tree, it looks
        > to me like the DF setting propagates through the whole
        stack, so I'm
        > surprised there's a bug.  Perhaps Greg can take a look at
        some point, if
        > he has time.

        I'll review the code but as you've pointed out, the code bases
        used are
        very old.  I'd suggest 2.10 at least since
        that's when we did a total review and upgrade of gre tunneling
        code to
        support erspan as well.  A lot has
        changed since 2.9.0

        Thanks,

        - Greg

        >
        > We probably need to know what version of OVS you're using,
        what version
        > of the kernel you're running, and whether you're using the
        Linux kernel
        > built-in version of the OVS kernel module or the one that is
        shipped
        > with OVS itself.


_______________________________________________
discuss mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-discuss

Reply via email to