- Add rtfree_remote(), which send rtentry to its owner CPU to be freed.
Print backtrace, if net.route.remote_free_panic is 0. This function
only serves as _temporary_ workaround.
- In ip_output(), if passed rtentry cache's owner CPU is not the current,
then keep it untouched and use the stack variable to relocate the route.
Add comment about it. This at least could happen in the following
callgraph on a TCP socket, on which listen(2) has been called:
ip_input() -> tcp_input() -> tcp_respond() -> ip_output()
Related to Dragonfly-bug: <http://bugs.dragonflybsd.org/issue1134>