On Tue, Mar 8, 2022 at 8:02 AM Numan Siddique <[email protected]> wrote:
>
> On Tue, Mar 8, 2022 at 9:51 AM Dumitru Ceara <[email protected]> wrote:
> >
> > On 3/7/22 18:46, [email protected] wrote:
> > > From: Numan Siddique <[email protected]>
> > >
> > > Presently for ACLs and LBs we do the following in the logical switch
> > > ingress pipeline
> > >
> > >    1.  Send the packet to conntrack.
> > >    2.  Apply ACLs (from-lport)
> > >    3a. If the packet is a new connection and it is destined to the LB
> > >        VIP, then select a backend (and commit to conntrack with DNAT).
> > >    3b. If the packet is a new connection and it doesn't match 3a, then
> > >        commit to conntrack.
> > >
> > > With the above approach, we cannot address the scenario of applying
> > > ACLs after the load balancing.  There can be ACLs which could match
> > > on the load balancer backend ips.
> > >
> > > This patch addresses this usecase by
> > >
> > >    1. Send the packet to conntrack.
> > >    2. Apply ACLs (from-lport, not configured with apply-after-lb=true)
> > >    3. If the packet is a new connection and it is destined to the LB
> > >       VIP, then select a backend (and commit to conntrack with DNAT).
> > >    4. Apply ACLs (from-lport, configured with apply-after-lb=true)
> > >    5. If the packet is a new connection and it didn't match (2), then
> > >       commit to conntrack.
> > >
> > > In order to support this usecase, this patch supports an option
> > > "apply-after-lb=true" in the ACL table.  This option is valid
> > > only for "from-lport" ACLs.
> > >
> > > Suggested-by: Dumitru Ceara <[email protected]>
> > > Signed-off-by: Numan Siddique <[email protected]>
> > > ---
> >
> > Aside from Han's comments, the rest looks good to me, thanks!
> >
> > Acked-by: Dumitru Ceara <[email protected]>
>
> Thanks Han and Dumitru.
>
> I applied this patch to the main branch with the below changes.

Thanks Numan. Since this is an important feature, would it be ok to
backport to branch-21.12? cc @Mark Michelson <[email protected]>

Han

>
> --
>
> diff --git a/ovn-nb.xml b/ovn-nb.xml
> index cea105b545..4d7a23c527 100644
> --- a/ovn-nb.xml
> +++ b/ovn-nb.xml
> @@ -2040,19 +2040,20 @@
>          </p>
>
>          <p>
> -          The main usecase of this option is to support ACLs matching on
> +          The main use case of this option is to support ACLs matching on
>            the destination IP address of the packet for the backend IPs
>            of load balancers.
>          </p>
>
>          <p>
> -          <code>OVN</code> will apply the <code>from-lport</code>ACLs in
two
> +          <code>OVN</code> will apply the <code>from-lport</code> ACLs
in two
>            stages.  ACLs without this option <code>apply-after-lb</code>
>            set, will be applied before the load balancer stage and ACLs
>            with this option set will be applied after the load balancer
> -          stage.  Hence CMS should be extra careful when using this
option
> -          and should carefully evaluate the priorities of all the ACLs
and
> -          the default deny/allow ACLs if any.
> +          stage.  The priorities are indepedent between these stages and
> +          may not be obvious to the CMS.  Hence CMS should be extra
careful
> +          when using this option and should carefully evaluate the
priorities
> +          of all the ACLs and the default deny/allow ACLs if any.
>          </p>
>        </column>
>      </group>
>
>
>
> Numan
>
> >
> > _______________________________________________
> > dev mailing list
> > [email protected]
> > https://mail.openvswitch.org/mailman/listinfo/ovs-dev
> >
> _______________________________________________
> dev mailing list
> [email protected]
> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to