On Wed, Feb 18, 2026 at 2:03 PM PRANAB DAS via lists.fd.io <pkdas.boston=
[email protected]> wrote:

> Hi,
>
> I want host traffic to be always seen by the VPP IP stack for
> ACL/NAT/IPsec.
> But in L2 (tap) mode, linux-cp-x-ip4 bypasses VPP IP stack/FIB.
>

Hi,

linux-cp-xc-ip4 should not do anything to bypass ACL, NAT, or IPsec if they
are configured as output features on the VPP interface. If there are
feature nodes enabled on interface-output, they should be traversed. It
should just bypass the FIB lookup.

Is there a way (config) linux-cp-x-ip4 to forward the host packet to VPP
> ip4-lookup?
>

Not currently. You would have to modify the code in your build to do that.

For instance there could be a SPD/SA on the VPP egress ethernet interface
> which is an internal interface for pipelining/service-chaining.
>

As noted, output features should work correctly. If some output feature is
not working, I could try to make a guess at the cause if you provide a
packet trace and more information about your topology and how you have VPP
configured.


> Should I  create the lcp interface in l3 (tun) mode? I am afraid that
> in L3 mode, some of the Linux control plane functionality will not be
> available.
>

The mode that you create the pair with depends on what type of VPP
interface you are pairing with. If it's an interface type that passes
ethernet frames (e.g. hardware/DPDK, GRE in TEB mode, vhost-user, memif in
L2 mode), you should create the pair in tap mode. If it's a tunnel L3
interface (e.g. IP-IP, GRE in L3 mode), you should create the pair in tun
mode.

It's unlikely to work if you try to create a pair in tun mode with a VPP
ethernet interface.

-Matt
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#26824): https://lists.fd.io/g/vpp-dev/message/26824
Mute This Topic: https://lists.fd.io/mt/117881329/21656
Group Owner: [email protected]
Unsubscribe: https://lists.fd.io/g/vpp-dev/leave/14379924/21656/631435203/xyzzy 
[[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

Reply via email to