From: "Adam Langley" <redacted>
Date: Wed, 18 Jun 2008 16:24:33 -0700

quoted

The retransmit timeout timer has to fire eventually, and when
that happens we should clear all of our SACK state.

Ok, that's good at least. There were other variables in my test that could have
screwed that up.

Attached is a patch which pulls the options logic together into a single
function of SYN/SYNACK and one for established. At the moment, it chooses
SACK over timestamps for MD5 connections, but it's very easy to change. (And
that's the point of this patch - the logic isn't spread out)

Not signed off, because I want to do more testing, and to sleep on it first,
but pretty close to final.

Thanks for doing this work, I'll have a closer look at this
patch later this evening.