User space

              +-------+
              | CARPd | <-----> Other apps
              +---+---+
                  |
                  | - netlink to control ARP LB action
                  | - netlink to talk to ct_sync
                  | - netlink to control QoS rules
                  | - PF_PACKET or raw socket to send/rcv CARP pkts
                  |
   -----------------------------------
           Kernel
        network I/O etc

Apps interface to Carpd to send CARP encapulated packets i was talking
about earlier. With that thought lest redraw the diagram:


         User space

+-------+         +-------+         +---------+
| App#X | <-----> | CARPd | <-----> | ctsyncd |
+---+---+         +---+---+         +---+-----+
                      |                 |
                      |                 |
   ------------------------------------------
           Kernel
        network I/O etc

So now ct_sync control is not owned by carpd. Rather. ctsync listens
to the netlink msgs from the kernel, builds an app msg, passes it
carpd which will send it if state is right. Note such a mesage which
can be used by apps doesnt exist today, it would need to be defined.

Apps could also register callbacks for events
such as mastership transitions.