I have two OpenBSD 4.2 machines, box1 and box2. They have public IPs 1.1.1.1 and 1.1.1.2 and a direct Ethernet connection with 10.88.0.1 and 10.88.0.2. They share a carp interface configured for 1.1.1.5 (box1 is the master; preempt is enabled), and pfsync is configured on the direct Ethernet.
box1 is running a TCP server that prints out its hostname once a second on 10.88.0.1:5555, and a PF rule for "rdr on $ext_if to 1.1.1.5 port 5555 -> 10.88.0.1". Starting with both machines up and running, I run "nc 1.1.1.5 5555" and get output once a second. I run "ifconfig -g carp carpdemote 5" on box1, and box2 takes over the carp interface. I keep receiving output, and then kill nc. box1's PF state changes to TIME_WAIT, but box2's stays in ESTABLISHED. I expect the state on box2 to change to TIME_WAIT as this is what happens if I do not run "ifconfig -g carp carpdemote 5" in the above process, or if I also run "ifconfig -g carp -carpdemote 5" before killing the TCP connection. (There are other details to my configuration like a second carp interface for load balancing, another server running on 10.88.0.2:5555, and an internal network, but I don't think those are relevant to this problem.)
