[feature request] `portmap` support `masquerade-all` option
When k8s cluster use ipvlan L3/L2, macvlan or other underlay network plugin. The traffic come back from pod to host maybe not go through conntrack in host, and can not un-snat to hostip which client requested.
Masquerad all traffic can make sure the pod reply come back to host and go through conntrack in host.
I'm having trouble understanding what is going on.
What addresses live where?
@BSWANG can you clarify the diagram? What are 1.1.1.1 and 2.2.2.2?
Is the square a node in a cluster? And is 192.168.0.1 and 0.2 NICs on the node?
Is 10.0.0.1 the container?
In your use-case, what is the source-ip and dest-ip of the incoming packet from client -> pod?
When the pod replies, what is the source-ip and dest-ip of hte outgoing packet?
Are you able to give the output of 'ip r' on the host node?
@dcbw @MikeZappa87
Thanks for reply. I have updated the description, for some underlay plugins scenarios.