Last tested on 2.6.38-rc2+ 70d1f365568e0cdbc9f4ab92428e1830fdb09ab0

tcp_tw_recycle seems to break ipv6 badly.

The script tcp6br.rb basically spawns two threads, one server one client. It continuously connects from client to server, and does some checking outside of those threads to try to spot strange behavior.

After several minutes, it seems to fubar the tcp6 layer.

I observer intermittent net unreachable errors, and after a few minutes, usually <10min, the test that connects to the closed port starts timing out. At that stage, the tcp6 layer seems to be completely hosed.

I have yet to see any net unreachable or connect timeout errors from tcp4 under the same conditions.

compare: ruby tcp6br.rb ::1 5555 ruby tcp6br.rb 127.0.0.1 5555