I'm kind of late to the game, but it looks good to me (as do most of the inline suggestions)

TIL rc.subr has reverse_list, so the old patch didn't need to to it by hand!

There's a right way to do it, which is to use the natural ordering of the jail command (including the depend parameter). But that would require re-writing the rc script to run the jail command once for all the jails, instead of trying to get the ordering on its own. Unfortunately, backward compatibility (e.g. ez-jail wanting these pid files) makes the right way difficult.

Looks good to me :-)

You may want to do the same with jid. You can give the jail.conf block a numeric tag, which will set the jid instead of the name, but will will still set j->name to that tag string. I would expect the same sort of problem from then setting jid=x inside the block.

I'd prefer it do a little less: first, the old shell-based parameters for /etc/rc.d/jail are for backward compatibility, and anyone who's going to be using new parameters should also be using the "new" config file. Also, the mount.linprocfs/linsysfs parameters don't need to exist. They exist mostly to cover things that were already in the old rc.d system, and can as easily be covered by the more general "mount" parameter.

I'm not particularly invested in keeping the command line the same as rc.conf; if rc.conf reverses and the command line doesn't, that's fine by me.

In D2130#16, @robak wrote:

I've made some more testing to figure out why you say it works, when I clearly see it doesnt, and I think I've found a problem: when jail -r jailname is used, then indeed, the jail command cleans up the mounts it created. However, when jail -r jid is used, the mounts are staying after killing the jail and cluttering the host by future mounts in the same place.

I'd suggest adding a functionality to jail -r jid which seems to be not reading the config when jailname is not used, to check the mounts in killed jail and remove them, in some similar way that the patch is doing it, to clean up after the jail. In addition to that, creating a jail -c could verify if there are any leftover mounts in the jail's target mountpoint to do something about them (cleanup? warning? error?).

In D2130#11, @jilles wrote:

If you care just about devfs, IP_MOUNT_DEVFS appears to be capable of an umount command already.

We need to reverse a list that comes from rc.conf, since there's only the one list for both directions. I think it would be confusing to reverse it in one situation and not in the other.

It works just as expected:

It does the right thing in its place. I would change the "_ _j" loop variable should just be _j though - or if you really want the double underscore add it to the local declaration.

Sounds good - thanks for looking into those. I had just kind of looked around at things, mostly because I wasn't all that familiar with osreldate (osrelease is a little easier to digest).

I have a few little picky notes :-) ...

One small change for the part in config.c that strips the extra parameters from KP_IP4_ADDR: if you put those two lines of code after line 591 (instead of after line 587) then you can reuse cs (as the current 592 does) instead of having to define the new ss. Or perhaps better after 603, to keep things in order.