On Fri, 31 Mar 2023 at 22:23, Thomas Zermeno <[email protected]> wrote:
>
> In response to
> https://groups.google.com/a/ccadb.org/g/public/c/mTUkK-gkHPE/m/WBaf7QUnAwAJ,
> starting from the Matthias’ last statement we would like to point to our
> actions as explained in [1], [2] and [3], which demonstrate SSL.com’s ability
> to monitor, analyze and take informed decisions, addressing particular cases
> in accordance with the BRs.
>
> More specifically, our immediate actions demonstrate:
>
> i) active monitoring of industry channels such as m.d.s.p., which
> allowed capturing a case which was not reported via the required channels, as
> published in our CP/CPS section 4.9.3 per provisions of BRs section 4.9.3;
Problem reports are not the only way a CA can become aware of problems
that exist in certificates. The CA is required to act in the time
specified in BR section 4.9.1.1, regardless of the way they became
aware of the poblem. The only thing that CPR brings onto the table is
a clear and objective way to determine when the clock starts without
depending on the CA for the rather vague moment of "becoming aware" or
"receives evidence": the moment that the message containing the CPR
was received by the CA, regardless of other processing steps.
Note, again, that neither of us seems to consider the original mail a
Certificate Problem Report as per the BR. That's also why I don't
think BRs4.9.5's start time stipulation is relevant - this time only
4.9.1.1 specifies the time windows.
> ii) ability to respond swiftly with expedited deployment and testing of
> technical control updates (pre-issuance linting), effectively blocking any
> similar issuances;
>
> iii) provident care to retrospectively test against the entire corpus of
> certificates, confirming no other occurrences exist;
>
> iv) due diligence by promptly analyzing the technical and compliance
> aspects of the issue at hand, considering the relevant literature and the
> applicable requirements, thus leading to informed decisions instead of acting
> hastily;
This all is commendable, yet these tasks are part of handling a
noncompliance issue under 6.1.1.3, not 4.9.1.1. The sections are
related, but are distinct in function: One is to prevent issuance, and
one is to revoke once issued. Being known not to comply with 6.1.1.3
doesn't mean that the certificates issued won't have to be revoked
with 4.9.1.1-specified timelines.
> v) [...]
> In our opinion, completing the above-mentioned actions (analysis, decision,
> notification, revocation, retrospection and mitigation) within 26 hours, is
> not a sign of “SSL.com's lack of urgency regarding detecting, handling, and
> responding” as stated in Matthias opening statement.
To me, this "26 hours for the above-mentioned actions (including
revocation)" feels like it refers a time period including both the
first mail from SSL.com on the m.d.s.p. forum's topic [0] and the
revocation timestamp of the certificate [1], which already spans at
least 25 hours and 50 minutes.
The first mail already claims to have completed testing the database
for other vulnerable certificates, which would mean that the database
was scanned and a new linter version was installed within 10 minutes,
which to me seems extremely fast for a database containing hundreds of
thousands of certificates (which at the time of writing is 205648
unexpired certificates for the issuer CA of the problematic
certificate alone), and a deployment process that needs sign-off of at
least 2 people, which is why I doubt that your claim of 26 hours is
accurate.
Again, from an outsider's perspective the timeline looks something like this:
2022-10-29 20:45 UTC - the problem was sent in a mail to m.d.s.p
No specific moment known, but implied to be before 2022-11-02 15:00
UTC: SSL.com notices the mail in m.d.s.p, becomes aware of the problem
and starts acting on the issue.
No specific moment known, but implied to be before 2022-11-02 15:00
UTC: SSL.com expedites their zlint update, and scans its database for
problematic certificates, completing before the next item.
2022-11-02 15:00 UTC - SSL.com publicly posts on m.d.s.p that they
have updated zlint, scanned their database for the issue, proposed a
CA/B Forum ballot, and have started planning to replace the
certificate.
2022-11-03 16:50 UTC - the timestamp in CRL / OCSP for the revoked certificate.
In whatever manner, if the certificate itself wasn't considered to be
vulnerable before the database scan, the certificate should have
popped up in the search of the database for vulnerable public keys and
at that point the timeline required by 4.9.1.1(4) starts. I don't see
how the "needs more context" changes the requirements - the public key
has a known vulnerability, you knew ("was aware") of the vulnerability
of the key at-or-before 2022-11-02 15:00 UTC due to _at least_ the
database scan, and failed to revoke within the stipulated 24 hours.
A CA misunderstanding the BR does not remove their need to comply with
the BR, that'd be absurd.
> It is worth explaining the rationale of our actions, and more specifically
> why we didn’t make an earlier revocation decision. Context played a decisive
> role here. If we had received a Certificate Problem Report (CPR) per section
> 4.9.5 of the BRs, we would be forced to follow the exact deadlines specified
> in the CPR procedure (confirmation, revocation decision and execution within
> 24 hours of receipt of the CPR).
Although I'm willing to listen to your reasoning for the actions
taken, it isn't going to change the point I'm making. The BR don't
require a CPR for the 24-hour requirement to be applicable.
If a CA is otherwise notified or becomes aware of problematic
certificates, regardless of the method, the clock starts from the
moment they become aware. The clause for the CPR endpoint in 4.9.5
only makes it clear that a CPR being received by the CA at their
specified CPR endpoint is equivalent to the CA becoming aware, which
starts the timer. Other methods of becoming aware of issues still
co-exist with the CPR endpoint, and although they are less clear about
the awareness timelines, they still have the 24-hour time limit.
> We would have a significantly different approach (lower level of analysis
> depth, lower priority in updating preventive controls, etc.). Since this case
> was reported in a public mailing list, we considered it better to do a more
> in-depth analysis as due diligence and use this time to prioritize other
> equally important actions (preventive controls, retrospection).
Although I applaud the effort that is being put into the handling of
the non-compliance with BR 6.1.1.3, I feel like you're failing to see
the forest for the trees. The issue is not that you handled your whole
certificate corpus in a great manner, but that you did not meet the
BR-stipulated timeline for the one certificate that showed the
problem.
I see three distinct ways to interpret the publicly available timeline:
1. Until (at the earliest) 2022-11-02 16:50 UTC (which is after
SSL.com's m.d.s.p mail on the topic) SSL.com did not know that for
that certificate there was "a demonstrated or proven method that can
easily compute the Subscriber’s Private Key based on the Public Key in
the Certificate", with in this case "a proven method" being Fermat's
Factorization Method.
2. SSL.com did know that the certificate was vulnerable to Fermat's
Factorization Method before 2022-11-02 16:50 UTC
3. as above, but SSL.com didn't think 4.9.1.1 (4) to apply to Close
Primes / Fermat's Factorization Method
If the first, then I worry about the CA's ability to act on clear
reports from side-channels and linters about problematic certificates
- the report refers to the certificate by a public id and reference,
and which attack it is vulnerable to, and you reply with what looks
like an acknowledgement of the attack vector and an acknowledgement
that the database was scanned for the vulnerability. At that point,
you should have known that the certificate must be revoked within 24
hours under the 4.9.1.1(4) requirements.
If the second, then the required timeline of 24 hours (as per BR
4.9.1.1(4)) was not met, and in that case the CA does seem to be
trying to escape the obligations that come with being part of the
Mozilla Root Store under MRSP 2.4 by explaining their actions.
If the last, then that is also quite problematic: what other (new)
attack vectors to RSA could the CA be aware of without considering the
public keys vulnerable to those attacks as in need to be revoked? A CA
not being able to put 1 and 1 together in the timeframe it said it
would seems quite concerning.
I agree that explaining actions is important to understand them, but
if mistakes/incidents aren't considered a problem, then there is no
value in explaining the actions that lead to that incident, as the
problematic behaviour would persist.
> Overall, we would like to state that this is not a case of a CA trying to
> escape its obligations.
See above "I see three distinct [...]" through "If the last [...]".
> With [3] we made clear that, in our opinion, and in accordance with the
> expectations of the industry, a CA may not arbitrarily extend the period of
> investigation and use it as an excuse to delay revocation.
That is my understanding as well; yet to my knowledge this is not
directly related to the issue at hand. The mail containing information
that implied that SSL.com had knowledge of the vulnerable certificate
was released more than 24 hours before the certificate was ultimately
revoked.
> We have also been completely transparent [...]
It would be greatly appreciated if you could share your timeline of
the events on that issue, and specifically the moments where each of
your above i) to v) was handled regarding that issue.
> [We] tried to detail the rationale of our actions.
Yes, I understand and (mostly) agree with the rationale of the actions
you took, except the one where you stated that attacks not in the BR
don't trigger the 4.9.1.1(4) requirement - I think they do.
> As stated in [2], if the root store owners considered this a violation or
> found any value in filing an incident report, we would do so. However, no
> such request was made (for the more than two months after our December
> response, which further explained our rationale); on the contrary, [4] is a
> clear indication that our actions were considered reasonable and compliant.
I can't speak for Ben, but that's not what I read in that message.
What I read is that Ben considered that the extensive discussion on
the issue dying out warranted the bug to be closed; not specifically
as Invalid, and had probably missed that there was no incident report
published yet.
> As a publicly-trusted CA, we have handled numerous cases of CPRs reporting
> vulnerable or other forms of compromised keys and we have followed the
> requirements to the letter. Our annual audits confirm our compliance with
> these requirements.
The latest standard audit report of SSL.com that I could find ([1],
for the period of July 1, 2021 to June 30, 2022) does, whilst not
modifying the opinion, refer to other non-compliance issues, including
more late revocations; implying that your "following to the letter"
also includes "failing to comply". In that case, I'd rather have you
not "follow the requirements to the letter", but instead be compliant
on all aspects of the requirements.
So, in short: I still see no reason to believe SSL.com was _not_ aware
of the certificate's issue w.r.t. vulnerable public key by the time
they sent their mail, yet by all metrics they failed to comply with
the BR. They then try to confuse us with context of them solving their
non-compliance with BR 6.1.1.3 (which is another compliance issue)
without considering their non-compliance with BRs4.9.1.1 as an issue.
I'm not confident that SSL.com's understanding of the BR is to the
level that we can expect of a CA, let alone one that already has root
certificates in various root stores.
Kind regards,
Matthias van de Meent
[0]
https://groups.google.com/a/mozilla.org/g/dev-security-policy/c/TlhjCJm610I/m/LmlR3tJCBAAJ
[1]
https://www.cpacanada.ca/generichandlers/CPACHandler.ashx?attachmentid=4abb6f0c-a72d-40ca-90a1-1c40b140f75a
--
You received this message because you are subscribed to the Google Groups
"CCADB Public" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion on the web visit
https://groups.google.com/a/ccadb.org/d/msgid/public/CAAT_OQvN2f3BOcERijRPuXDAsWNrrZBXOc2vzuqssrK%3DQAssEg%40mail.gmail.com.
