On Wed, Jan 18, 2017 at 12:00 PM, Salz, Rich <[email protected]> wrote:

> > One can compare three logs and if they differ you know who and where the
> rogue is, right?
>
> > Hmm.... Not unless all logs are required to have every certificate (at
> least within a given scope).
> > Otherwise, the fact that a certificate is in log A but not log B doesn't
> tell you anything about log B.
>
> So far the Chrome policies requires two or more, and the draft Moz policy
> is three or more IIRC.
>
> So for a given CA, a disagreement identifies that *one* of the logs is
> wrong, right?
>

It's quite sensitive to what the precise requirement is. So consider a
requirement
that site provide an SCT from N logs out of a set of M. Without loss of
generality,
the site gets SCTs from L_1, L_2, ... L_N but not L_N+1, ... L_M. The fact
that
the site's cert appears in L_1 but not L_M is normal and doesn't tell you
that
either one of them is wrong. (This is just the basic non-adversarial case).

Now, consider the adversarial case, where an attacker as corrupted a CA and
N logs out of M. Similarly, they will get N SCTs from those corrupt logs
(which
might not be the ones that the actual legitimate site used for their SCTs,
but the
client doesn't know that [0]. And again, the lack of an SCT from the
non-corrupt
log doesn't tell you anything.

Now, in the special case where N==M, then inconsistencies are meaningful,
and this is the case with Chrome's "one of the logs must be Google" policy,
but
as far as I can tell that's not something that's true in the general case.

-Ekr

[0] Modulo some sort of log pinning.
_______________________________________________
Trans mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/trans

Reply via email to