> You need the check with snd_max to make sure
> you don't retransmit those packets. Since the intention
> of the draft was to send upto 2  'new' packets and not
> retransmissions. Your code can result in retransmissions
> after a RTO.

Ugh.  You're likely right.  I think the Rght solution to this is to
implement the fast retransmit bugfixes outlined in RFC 2582.
(Essentially use fast retransmit (and limited transmit) only once
per "loss event").  I don't think the implementation should be that
hard.  Let me poke at it and re-send bugfix+limited transmit
patches.

Err, well limuted transmit essentially as I implemented it.  If
consensus is that we should go with Sean's idea of using up t_extra
in tcp_output() then I'll likely let someone else fight that one.

But, I'll see about hacking the bugfixes, anyway.

Thanks,
allman