Re: [tcpm] tcp-auth-opt issue: replay protection

["Adam Langley" <[e-mail]>]

On Wed, Jul 30, 2008 at 4:17 PM, Joe Touch <[e-mail]> wrote:
> Can you explain the algorithm for "closest"?

Sure,

let the last 'known' value for the ESN be a uint64_t called last_esn
and the current SEQ be a uint32_t called current.

uint64_t last_top32bits = last_esn >> 32;
uint64_t possibility_1 = (last_top32bits - 1) << 32 | current;
uint64_t possibility_2 = last_top32bits << 32 | current;
uint64_t possibility_3 = (last_top32bits + 1) << 32 | current;

uint64_t delta_1 = distance(possibility_1, last_esn);
uint64_t delta_2 = distance(possibility_2, last_esn);
uint64_t delta_3 = distance(possibility_3, last_esn);

uint64_t new_esn = min(delta_1, delta_2, delta_3);

That's written to try and be clear, rather than actual code that I would write.

The idea is that, when we get a seq number, we know that the sending
side had a definite 64-bit ESN in mind when they sent it. We know the
bottom 32-bits of this, so there are 2**32 possible ESNs (the set of
64-bit numbers with those 32 bits in the lower half). We pick the one
that is closest to the last known value.

If our last known value was 0, and we get 1, we assume that the ESN is one.

If the last ESN was 2**32 - 1, and we get 1, we assume that the ESN is
2**32 + 1.

If the last ESN was 2**32 + 1, and we get 2**32 - 5 (packet
reordering), we assume that the ESN is 2**32 - 5, because it's closer
than assuming that nearly 2**32 bytes went by, unnoticed.

I hope that's clear. I'll write real code if I still being inarticulate.


AGL

-- 
Adam Langley [e-mail] http://www.imperialviolet.org
_______________________________________________
tcpm mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/tcpm