Hiren's SoC student has completed a project with packetdrill for a TCP regression suite. It has a local mode so it might be possible to automate with Kyua. There is a set of simple tests in the project. The tests have a result of pass or fail. The student intends to continue to add tests and to extend this suite. George will pass this on to the folks at Isilon and Steve K. has passed it along to the folks at Juniper. Hiren will propose a session at the MeetBSD Dev Summit. Hiren has tried to reach the folks at Google on their tests but has not gotten ahold of them as yet.
The route caching code is now in both HEAD and 11. There was a an issue with this with respect to IPv6 which has now been addressed. It is not used for MTU discovery.
The plan is to remove the flowtable code during the next few months.
Ryan has put in his fixes for IN_BROADCAST. There has been some discussion of this with bms@ but the results of that are inconclusive. There is a way to incrementally move the drivers from the rwlock to an rmlock by having both locks and having the old drivers use rwlock until they use the rmlock. Drew suggests that we can optimize all of this by having two lists and swapping them, perhaps once per second. The overall goal is to actually make this lockless, but it may require a bit more thought. Ryan will continue to look at this.
bz@ points out that his optimizations to UDP and v6 has been in phabricator for a year and that he will update and push that.