[SSL Observatory] SSL CA compromise in the wild

Erwann ABALEA erwann at abalea.com
Thu Mar 24 09:05:14 PDT 2011


I missed the "reply to all" button :(

2011/3/24 Erwann ABALEA <erwann at abalea.com>:
> Bonjour,
>
> 2011/3/24 Peter Gutmann <pgut001 at cs.auckland.ac.nz>:
>> - The blacklist-based controls used in PKI (CRLs and OCSP) don't work, and the
>>  vendors agree (Jacob Appelbaum has pointed this out too).  AFAIK every
>>  single one of them pushed out updates that hardcode the certs to be rejected
>>  into their browsers.  Looked at the other way round, not one single vendor
>>  trusts the mechanisms that PKI is supposed to use to deal with these
>>  certificates.  So if you want to go through the motions for compliance
>>  purposes, issue a CRL or OCSP.  If you really care about the status of a
>>  cert, do something else.
>
> I made some tests a few weeks ago, on several browsers and OS
> combinations. It appeared that the NSS library (used by Firefox on
> every platform, and by Chrome on Linux) doesn't check anything (CRL or
> OCSP) for non-EV certificates. From memory, MSCAPI, Opera, and the
> MacOSX crypto toolkit do a better job, checking either the OCSP
> responder or the CRL, depending on the level in the hierarchy and the
> software used.
> More checks could be done.
>
> In that specific case, since the emitting CA is a root one, it can't
> be revoked. And if it is suppressed from the trust store, since it was
> cross-signed by another root (AddTrust External Root CA) and the good
> URI is placed in the AIA extension, a chain can still be built.
>
>>  (If you can issue your own certs then it's even worse, just fit them with a
>>  CRLDP extension pointing to an OCSP responder that you control and those
>>  certs can never be revoked.  It's another case of PKI relying on mechanisms
>>  that involve asking the drunk whether he's drunk).
>
> I'd create a long-lived OCSP responder certificate with the
> OCSPNoCheck extension. This kind of certificate can't be revoked *at
> all*, and has the same power as a CRL-signing key (which can be
> revoked).

-- 
Erwann.



More information about the Observatory mailing list