On Fri, Sep 16, 2016 at 05:19:55PM -0700, Jacob Hoffman-Andrews wrote:
> On 09/14/2016 01:14 AM, Ron wrote:
> > doing some sort of legal theatre lite where the client blindly sends
> 
> For better or for worse, there is some "legal theatre lite" required to
> get a certificate in the Web PKI. The Baseline Requirements require it.
> The question is: Do we automate it or do we not automate it?

And if the answer to that is "we do automate it", then the important
question is: Does this even remotely satisfy the actual requirement?

AIUI the CABF requirement is for an explicit agreement that is legally
binding to the end user in the applicable jurisdiction.

The problem you ran into with the existing system is that some (many?
all?) extant clients simply hardcoded:

 "agreement":"https://letsencrypt.org/documents/LE-SA-v1.0.1-July-27-2015.pdf";

into their source, without even looking at what the directory reported.
And that others required user intervention before performing the actions
to accept a changed ToS URL.

And the solution you are proposing to 'fix' that is for them to instead
hardcode "agreement":"yes", so that it still 'works' no matter what
random text is indicated in the directory at any point in time - and so
that doing it once implicitly agrees to any future ToS in advance.

Which is roughly equivalent to adding a line to my client software that
sends "btw":"You now agree to pay Ron One Beelion dollars tomorrow".

When the client blindly includes that in a protocol message without the
end user ever having seen it, you won't have to be a bookkeeper or lawyer
to correctly guess the chance of any court anywhere considering either
of those to form a binding contract.


Will it satisfy the CABF?  Who knows, they turn a blind eye to all sorts
of things that are inconvenient - but in terms of protocol design, it's
not any sort of mechanism that provides people options for meeting the
requirement, or for indicating what you are agreeing to.  It just says
you're generically agreeable to anything.  And doesn't give you a way
to post-facto query what it was that some employee of yours actually
agreed to on your company's behalf, possibly without your knowledge or
consent or permission or authorisation.


I would like to see a good way to fix the problem(s) you ran into in
practice when LE changed its ToS.  What I don't want to see is that
we try to fix what was a relatively simple client implementation bug
by using the protocol to mandate people implement something which is
fatally flawed and unfit for purpose from the outset - because it will
be harder and uglier to fix that later than if we just punted on this
and did nothing at all.


> > a hard coded "sure, whatever you say, whenever you say it" flag.
> 
> I am *not* proposing a "sure, whatever you say, whenever you say it"
> flag. I am proposing that ACME only needs to know how to agree on terms
> of service when creating an account. If the server later wants their
> users to agree to new terms, that can be readily be implemented with
> error codes and out-of-band URLs.
> 
> You are proposing that ACME specify not only initial agreement, but also
> how to negotiate subsequent agreements. There is no need for this,
> specifically because those subsequent agreements cannot be automated.
> ACME is the Automated Certificate Management Environment. Why would we
> include extra plumbing that can only be used manually?

I don't see how it could possibly be worse to have _one_ well specified
mechanism which is suitable for _both_ initial agreement and later
re-agreement if a CA decides that is necessary, that can be implemented
_once_ in the same client and server software?

One or both of us seems to be missing something in what the other is
saying or thinking here ...  since we seem to be in violent agreement
about the goal, but not about what might actually achieve it.

If the protocol was to remain as it presently is, where the server
declared the applicable ToS in the directory, and the client responds
with the explicit URL of what they have agreed to (and that becomes
state in the registration object) - then whether or not a change in
the ToS can be fully automated is *entirely* the decision of of the
client side.

Which IMO is as it should be.  The end user should be the one to
decide whether a 'blind' change to the ToS is acceptable to them
or not, not the service provider.

And all the client software needs to do to provide that choice is to
give the end user a configuration option that indicates whether or
not to automatically register acceptance if the ToS ever does change,
using the same protocol mechanism they used the first time.


Having already implemented this myself, I know that it works and
that it's possible :)

What you are suggesting is that I'd have to replace that with some
entirely unknown, and unspecified, and possibly different process
for every server that ever needed this.  So unless I'm missing some
key part of what you are saying, _that_ is the option which would
in fact be entirely impossible to automate.



> > We could just provide a directory entry (which could even be optional)
> > that indicates where the terms can be found
> 
> This exists, and is what my proposal is based on.

Sorry, maybe I wasn't clear there - I meant *only* provide the directory
entry, as a convenience to client software, so the user doesn't have to
hunt for it at some out of band location.  And make acceptance implicit
on proceeding with registration.

Which would have about as much legal weight as including a hidden
"I accept" constant in the protocol - would delete useless code like
you were advocating - and has a lot more precedent in existing practice
advocated by the lawyers of network service providers.

If that gets ruled invalid in a court, then a lot more than the CABF
contracts will fall down in a screaming heap.


> > that any other use of the service indicates acceptance of those terms
> > (just like pretty much every other network service in existence does
> > without needing to kludge a contentless 'accept' bit into the protocol).
> 
> I understand this proposal is an unappealing compromise. I feel the same
> way. I'd prefer to go all the way and remove explicit agreement to
> terms. Unfortunately, the BRs don't seem compatible with this.

And I understand (and sympathise with) the barrel you are over with this.
I realise you are looking for the simplest and most foolproof thing that
will satisfy the CABF BRs - but I'm having a hard time seeing how this
would even roughly achieve either of the technical or legal aims.

We'd still have useless cruft that people need to implement, in possibly
buggy ways, and it would have even less legal standing than what we
currently have for the purpose of meeting the BRs.

It's the sort of compromise that has us trying to share a cocktail
umbrella to keep us out of the rain.  It ties up one hand without
actually keeping anyone any drier, while I think we do have other
viable options here.


> In another thread, Rich has asked if we are close to consensus. How
> strong are your objections? Are you willing to join the CA/Browser Forum
> as an Interested Party and propose the necessary change to the BRs?

I certainly don't want to be a curmudgeon standing in the way of
reaching a consensus here, and the only thing I really object to
with passion is "do something half-assed" - I could be just as happy
with either of "provide a real mechanism" or "excise this from the
ACME protocol entirely".  You said you'd talk to your legal folk
about whether they thought removing it was an option, and I don't
yet know for sure what the result of that was, which is why I
hadn't been pushing for one of those options over the other yet
and wanted to wait until we had more feedback again from you.

If proceeding with registration on its own constitutes explicit
acceptance regardless of the content of the protocol message which
does that, then I'd probably tip toward removing it.  If that's
somehow not enough, then I think we do need a real mechanism in
the protocol that has some chance of being flexible enough to work
across not only a range of legal jurisdictions and desired use
cases - but might also survive a change in interpretation of the
CABF too.

Even if you and I were to lobby the CABF for a change to their BRs,
or were to agree that we think an extra magic constant in the
protocol really would add any legal weight to satisfying them as
they are today - the opposite could also happen, that enough of
the CABF concludes that LE is an existential threat to them and
so tightens their rules in the other direction.  Or a court actually
rules that "no way is this sort of thing even remotely binding".


Which is why I think that for making decisions about the protocol,
we need to think less about "what is the absolute minimum we can
get away with for the current BRs" - and more about "do we actually
need a mechanism for this at all or not?".

If we do need it, it should be a real mechanism for maintaining
this as shared state between client and server.  If we don't,
we should get rid of it so that we do have a clean slate to add
a proper mechanism later if it ever does become needed and what
is needed becomes clearer.

The worst option I see is that we clutter the protocol now with
a lip-service policy decision that becomes a legacy wart stuck
on the side of whatever is actually needed to really solve this
problem.


I have a really hard time seeing how the magic constant string
that you proposed adding would have any more legal weight than
just saying in the RFC that "a protocol message with a curly
brace in it indicates acceptance of the terms of service".

I understand what you are _trying_ to do with that, but I think
a lot of people would have to pretend pretty hard for it to have
any substantial difference from the "do nothing" option beyond
being an additional (re)implementation burden.

If we can get past that somehow, we should be able to converge
on agreement about all this pretty quickly.

  Best,
  Ron


_______________________________________________
Acme mailing list
Acme@ietf.org
https://www.ietf.org/mailman/listinfo/acme

Reply via email to