This will cause a HUGE performance impact! sbspace() is called often and
twiddling the extra locks is not useful. You are just looking (reading) two
variables or even one variable in the sbused case. You do not need a lock for this!

If we put the wrap
#ifdef TCPHPTS

So Hans:

Lets go ahead and not put the netflix_stats directive in front of the sysctl's.
Turns out we already had these defined if RATELIMIT is defined so lets
just add the two new ones.

Looks like I forgot to make sure we get the counters properly initialized in tcp_subr.c opps.

Take out the mlx bits so that Hans can commit them separately

Why are you removing the NOP Pad chunks? IP options *must* be a multiple of 4 which is what the pad does, taking
it out is going to make the options incorrect I would think??

Thats fine if you want to do it that way but note that this patch adds the new
function if_next_snd_tag() to if_var.h. You won't be able to add that function unless
we commit this one (without the mlx changes) first.

Not exactly the best, but I think it fixes the immediate problem. From a long term perspective we may want to
look for a better solution :)

Fix the typo's in the comment that Michael pointed out!

First of all the tricky thing to remember about inp locks is that

Opps used the wrong goto label...

Looks like BBR has the same issue, and we need to strip trailing emacs presents too :)

In D24902#550447, @rstone wrote

Wouldn't sbintime_t be easier and cheaper to work with?

I don't mind you tracking t_rtseq here, but please do not change the idle reduction in rack. I will go dig
up the right variable and change this to use that. Your are welcome to maintain t_rtseq but please don't change
the rack behavior here.

It would not be EAGAIN/EBUSY since you can't do it again and expect it to succeed. I think
EINVAL is ok here.

Ok turns out skyzaller has found yet another crash having to do with TFO for all three stacks. Lets update
the patch to fix that too.

Has discussed on the transport call, I will be updating these to have a uint64_t and a nano-second basis for the values (including t_starttime I think since this
is what you compare these against) :)

I don't think that holding just the delta is what you really want. You want to be able
to tell when the first byte came in and the first byte went out. The delta only gives
one a partial view of what happened. And we are not living in 1984 where 64k of
memory is a lot. Four bytes to get more detailed information is I think worth it.

Note that we don't want to commit these for a while since skyzaller uses these addresses but it
is something we should do.

Hmm I somehow got the wrong patch here.

Yeah the spelling error check is rather dumb :)

We should not ever do GP measurements until we are established

One little other niggle that I noticed, we should not do a goto again if the SYN bit is set ...

And even yet another issue thanks to skyzaller in bbr.. the rc_lost when doing TCP-FO was incorrect in rxt timers.

Let's also fix the accounting panic that skyzaller is finding here too. When we are in a front
state and have a SYN outstanding that should count in the ctf_outstanding size. Also if
we have sent a FIN same thing.