On Thu, 14 Jul 2016 02:52:41 -0700 (PDT) Nick Lamb <[email protected]> wrote:
> On Thursday, 14 July 2016 05:18:20 UTC+1, Andrew Ayer wrote: > > Revocation does not address the risk that this mis-issuance has > > caused to the ecosystem, since collided certificates (the ones we > > cannot see, and need to be worried about) have different serial > > numbers and therefore do not appear revoked. > > Unless I'm missing something if Symantec produced these certificates > in a sensible way, any potential attacker outside Symantec (including > TSYS) will not have known the serial numbers of the certificates in > advance of their being produced, let alone been able to choose them. > Without knowing the serial numbers in advance there is no way to > perform the MD-style chosen prefix collision attack on SHA-1 that I > can see. Do you suppose that attackers might have a pre-image attack > instead? I'm concerned about collision attacks. A truly unpredictable serial number does prevent chosen prefix attacks, but Mozilla and the other root store operators have decided that the correct way to deal with collision attacks is to deprecate the hash algorithm instead of relying on serial number entropy. This is the prudent approach. Cryptographically-secure random number generation is easy to get wrong, and mistakes are hard to spot. Symantec's CPS only guarantees 20 bits of entropy, which provides a poor security margin, and Symantec has a history of using predictable serial numbers even during a period when Microsoft's root program required 64 bits of entropy. We should therefore consider the possibility, even if slight, that an attacker was able to predict at least one of these serial numbers. Regards, Andrew _______________________________________________ dev-security-policy mailing list [email protected] https://lists.mozilla.org/listinfo/dev-security-policy

