Made requested changes, and updated base revision.

I spent some time coming up with concrete use-cases to demonstrate this patch's usefulness:

Rebased and added a test.

Looks good to me! Thanks for the report.

Updated to only include the first commit after discussion with @jrm. Will make separate PRs for the other two commits.

Split the changes into two commits and added tests. Checked that the tests fail before the two commits and pass after.

Ah, I didn't know that CI ran ipfw - thanks for pointing that out. Was able to replicate the failure finally, and test that D40973 fixes it. Since that's merged already, I just updated the base revision here.

Sorry for the delay - disabled DAD and updated branch to newest main again. I'm still not able to replicate the failure, so it's not clear to me whether there is more to solve after disabling DAD. The tools you showed are helpful but I can't use them unless I can replicate the failure.

Sorry about all the pings! I diffed from an incorrect main branch and pulled in a whole bunch of "changes" that weren't actually there. This patch should fix that.

Honestly, I'm not quite sure what the problem is on your end -- I am unable to replicate this failure. The only working theory I have for this is that fc2b::f0 is somehow not being assigned to ${epair_one}b as it should be. Unfortunately it seems that if ifconfig is not able to bring an interface up with the specified IP address, it will still bring it up without that IP address, and give no indication of failure on stderr or the exit code. So it's not possible for me to check _why_ your machine is unable to assign that IP address to the epair, other than perhaps it's already in use?

Brought the branch up to date and cleaned up some of the styling. All the tests should pass again now, and the malloc-while-non-sleepable-lock should be gone. There is a "lock order reversal" warning that shows up because we have to call in_joingroup or in6_joingroup while holding a non-sleepable lock, and those functions themselves hold a sleepable lock to work. This regression does not seem to be introduced by this patch though, since I get the same warnings from carp, which I did not change.

• fixed style nits

• add pfsync ipv6 unicast support (luiz changes)
• add pfsync ipv6 multicast support
• enable usage of all multicast addresses as peersync