[RFC SRU Trusty] tcp: make challenge acks less predictable

[Seth Forshee]

On Fri, Aug 12, 2016 at 11:32:00AM +0200, Stefan Bader wrote:
> This is a rather quick (and not the most elegant) backport of the
> upstream fix for Trusty (Precise will need the same or even more).
> The Problem is that neither WRITE_ONCE nor prandom_u32_max do exist,
> yet. The prandom_u32_max we might be able to just replace by the
> actual function body. I am not sure I would trust myself enough to
> try that for the WRITE_ONCE part.

I note that in Ben Hutching's backports for 3.16 and 3.2 he used
ACCESS_ONCE rather than WRITE_ONCE:

https://git.kernel.org/cgit/linux/kernel/git/bwh/linux-stable-queue.git/tree/queue-3.2/tcp-make-challenge-acks-less-predictable.patch
https://git.kernel.org/cgit/linux/kernel/git/bwh/linux-stable-queue.git/tree/queue-3.16/tcp-make-challenge-acks-less-predictable.patch

That looks legitimate since challenge_count is an unsigned int, i.e. it
should result in the same assignment as the WRITE_ONCE backport.

Seth