[cabfpub] Draft Ballot 185 - Limiting the Lifetime of Certificates: User input

Thu Feb 9 20:44:06 UTC 2017

To Jody's point:

Consider the following sites (all times Pacific, because I'm on a Mac right
now and I'm too lazy to adjust back to GMT from the UI)

https://amd.com (Symantec; expires 2017-07-21, issued 2014-07-21;
https://crt.sh/?id=4638163 )
https://americancrew.com (Symantec; expires 2018-07-10, issued 2014-06-09;
https://crt.sh/?id=4324896 )
https://www.ashevillechamber.org/ (GoDaddy; expires 2017-03-06, issued
2012-03-06; https://crt.sh/?id=982926 )
https://bitcoin.hu (Comodo; expires 2019-08-14, issued 2014-08-14;
https://crt.sh/?id=4816214 )
https://gm.ca (DigiCert/Verizon; expires 2017-09-18, issued 2014-09-18;
https://crt.sh/?id=5021956 )

Despite the Forum's efforts, all of these certificates are SHA-1 and valid
through 2017-01-01. That means these sites break when you disable SHA-1,
such as Chrome has. I'm sure, if members want to keep debating this as if
it's not an issue, I can find certificates from most of the CA members
opposed that demonstrate these sorts of issues.

You might be wondering where I came up with this list. It wasn't
Certificate Transparency. It was courtesy of Mozilla, by way of
understanding what's currently preventing them from disabling SHA-1 on
Firefox 51 at 100% - specifically,
https://bugzilla.mozilla.org/show_bug.cgi?id=1330043#c7

Until Mozilla disables SHA-1, their users are at risk. They're at risk from
malicious parties having taking advantages of CA's past reticence to adopt
serial entropy - and I'm happy to name the specific CAs I'm thinking about,
if members would like - and they're at risk of CAs misissuing SHA-1 - of
which, again, I'm happy to name some specific members who have already
proven this is a valid concern.

But this also demonstrates the balance that browsers face when charged with
protecting their users - do we block access to these sites (as Chrome is
doing, and as Microsoft will around February 15) or not? If we do block
these sites, we run the risk of causing the average user to see too many of
these messages a day, thus succumbing to warning fatigue, and causing them
to ignore these warnings when their information is truly at risk. This
makes everyone less secure - either through warning fatigue or through lack
of automatic protection.

We can talk about how revocation is the answer - but this is clearly
irrelevant, because every one of these certificates is unrevoked. We can
argue about how many outgoing links to those specific addresses exist and
whether they're "used" - for example, https://www.amd.com has corrected the
error, but https://amd.com has not - but that's irrelevant, because
browsers intentionally don't collect data to that degree, precisely because
it's invasive to users' privacy. So we - the Forum - have no way of
determining whether or not our users care about these sites, whether or not
the server operators care about these sites, whether or not the CAs care
about these sites - but we know these sites will and do break.

For those who are unfortunately arguing we should "phase these requirements
in" - we did that, with 60 month certs to 39 month certs. Notice how those
60 month certs are still causing pain in the examples I gave above? That
phase in approach doesn't meaningfully work - at least not for validity
periods.

What would the world look like if we had required 398 day (nee 13 month)
certificates from the get-go? Not a single one of those certificates would
be valid beyond 2017-02-01. Browsers - such as Chrome, Firefox, Edge,
Safari, Opera, Qihoo360 - could all make an informed decision that they can
enforce "no SHA-1" on 2017-02-01, and know *it will not break a single
site*.

What an amazing world that would be.
Because it would mean our users were protected from SHA-1 issues - whether
past or future.
That's the world we want to move to.
That's the world some member CAs are opposing.

That's why this is about security.

On Thu, Feb 9, 2017 at 11:59 AM, Christian Heutger via Public <
public at cabforum.org> wrote:

> I’m following the discussion already for a while. I understand the
> reasons, but I also see the customers situation out there. I also saw the
> votes recently and as a compromise I believe a two year timeframe + up to
> 90 days on renewal (so a total of 27 months or 820 days to be clear) would
> be a reasonable timeframe for most customers. However, I still expect some
> trouble, but maybe some exceptions would be granted. For the need to
> eliminate any X fast, OCSP still should be the solution, to get an
> evolution process running, two years should be enough.
>
> Am 09.02.17, 18:47 schrieb "Jody Cloutier" <jodycl at microsoft.com>:
>
>     I'm the first to admit that I haven't been following this thread as
> closely as I would like, but, from Microsoft's perspective, we want shorter
> certificates and not longer. We would certainly endorse a ballot that would
> mandate shorter certificate life for the very reason stated below: if we
> want to eliminate X we would know exactly when the last cert will expire.
> We've gone so far as to consider mandating this as a program requirement.
> Anyway, that's our .02.
>
>     -----Original Message-----
>     From: Public [mailto:public-bounces at cabforum.org] On Behalf Of
> Christian Heutger via Public
>     Sent: Thursday, February 9, 2017 9:31 AM
>     To: CA/Browser Forum Public Discussion List <public at cabforum.org>
>     Cc: Christian Heutger <ch at psw.net>
>     Subject: Re: [cabfpub] Draft Ballot 185 - Limiting the Lifetime of
> Certificates: User input
>
>     > I can see why there's some confusion here :-) Ryan is not arguing
> that we should switch to 13 months so that we will always in future move
> from "let's eliminate Algorithm X" to "Algorithm X is gone" in 13 months.
> One would always consider lots of data points in setting such a timetable.
> His point is that 3.25 > year certs make it very hard to move faster than
> that in _any_ deprecation scenario, whether simple or complex.
>
>     I don’t believe, moving faster is required for normal situations. If
> there are issues arising needing faster reaction, revocation and reissue is
> still a possible way. For normal situations, enterprises need to be able to
> react and they can’t move faster. Why are most enterprises skipping one
> Windows version and roll out the next one? As they are not able to move
> faster in controlled enterprise security environments.
>
>     > I don't agree that replacing your certificates once a year requires
> automation. It's made easier by automation, but it doesn't require it.
>
>     As I understood the discussion, 1 year is the first step on a road to
> months or weeks.
>
>     > I'm sure there are plenty of CAs, big and small, who would assert
> their automation solutions are secure. :-)
>
>     But as you know, there is nothing, which is 100% secure and if we talk
> about certificates in their sense of encryption and(!) identity assurance,
> such job shouldn’t be based on automatism.
>
>
>
>     PSW Secure Mail Gateway Info - http://www.psw.net/
>
>     - The message was not encrypted and not digitally signed
>     PSW Secure Mail Gateway Info - http://www.psw.net/
>
>     - The message was not encrypted and not digitally signed
>
>
> _______________________________________________
> Public mailing list
> Public at cabforum.org
> https://cabforum.org/mailman/listinfo/public
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cabforum.org/pipermail/public/attachments/20170209/5e12cb83/attachment-0003.html>