RE: Gen-ART review of draft-ietf-pkix-rfc2560bis-15

Stefan,

Is this important enough to do that?


IMHO, yes - the "running code" aspects of existing responder 
behavior/limitations
are definitely important enough for an RFC like this that revises a protocol 
spec,
and the alternatives to "revoked" feel like an important complement to those
aspects (discussion what to do instead when responder behavior/limitations are
encountered).

I appreciate the level of work that may be involved in capturing this, as
I've had my share of contentious discussions in WGs that I chair - FWIW,
I'm currently chairing my 4th and 5th WGs.  OTOH, when a WG has put that much
time/effort into reaching a (compromise) decision, it really is valuable
to record why the decision was reached to avoid recovering that ground
in the future and (specific to this situation) to give implementers some
more context/information on how the protocol is likely to work in practice.

Thanks,
--David

-----Original Message-----
From: Stefan Santesson [mailto:stefan(_at_)aaa-sec(_dot_)com]
Sent: Wednesday, March 27, 2013 11:38 AM
To: Black, David; sts(_at_)aaa-sec(_dot_)com; mmyers(_at_)fastq(_dot_)com; 
ambarish(_at_)gmail(_dot_)com;
slava(_dot_)galperin(_at_)gmail(_dot_)com; 
cadams(_at_)eecs(_dot_)uottawa(_dot_)ca; gen-art(_at_)ietf(_dot_)org
Cc: pkix(_at_)ietf(_dot_)org; Sean Turner; ietf(_at_)ietf(_dot_)org
Subject: Re: Gen-ART review of draft-ietf-pkix-rfc2560bis-15

I could.

My worry is just that this is such a contentious subject and it took us x
hundreds of emails to reach this state, that if I add more explanations,
people will start disagreeing with it and that we end up in a long debate
on how to correctly express this.

Is this important enough to do that?

/Stefan

On 3/27/13 3:30 PM, "Black, David" <david(_dot_)black(_at_)emc(_dot_)com> 
wrote:

Hi Stefan,

Does this answer your question?


Yes, please add some of that explanation to the next version of the draft
;-).
Coverage of existing responder behavior/limitations (important "running
code"
concerns, IMHO) and alternatives to using "revoked" ("have a number of
tools
to prevent the client from accepting a bad certificate") seem particularly
relevant.

Thanks,
--David

-----Original Message-----
From: Stefan Santesson [mailto:stefan(_at_)aaa-sec(_dot_)com]
Sent: Wednesday, March 27, 2013 7:44 AM
To: Black, David; sts(_at_)aaa-sec(_dot_)com; mmyers(_at_)fastq(_dot_)com; 
ambarish(_at_)gmail(_dot_)com;
slava(_dot_)galperin(_at_)gmail(_dot_)com; 
cadams(_at_)eecs(_dot_)uottawa(_dot_)ca; gen-art(_at_)ietf(_dot_)org
Cc: pkix(_at_)ietf(_dot_)org; Sean Turner; ietf(_at_)ietf(_dot_)org
Subject: Re: Gen-ART review of draft-ietf-pkix-rfc2560bis-15

Hi David,

Yes I missed to respond to that aspect.

This is a bit complicated, but we have a large legacy to take into
account
where some responders implements just RFC 2560, while some deliver
pre-generated responses according to RFC 5019 (Light-weight OCSP). LW
responders are not capable of producing a signed response at the time of
responding and in case such responder finds a request for a certificate
where no pre-produced response exists, it will reply with an unsigned
error response "unauthorized", which also is a legitimate way to
respond.
So the actual OCSP responder may actually know that the certificate was
never issued, but since it delivers pre-produced responder through a
CDN,
it can not provide a revoked response in real time.

So the major aim with the current writing is to declare that the revoked
response is a MAY because there are other valid alternatives.

We also want to avoid putting down a SHOULD respond revoked if a
certificate is known to be not-issued, because that would require us to
define what "known to be non-issued" actually means. And that could be
quite tricky as OCSP responders by no means are required to have this
knowledge.

The OCSP responder simply have a number of tools to prevent the client
from accepting a bad certificate.
This update of OCSP simply allows responders to use the "revoked"
response
as a preventive measure, without mandating it.

This is also related to activities in the CA Browser Forum where they
put
down requirements on responders complying with CAB rules to not respond
"good" to certificates that were never issued.
With this update in OCSP, they can now mandate in their policies both
the
fact that their responders MUST know if a certificate was never issued
and
MUST respond "revoked".

So we allow other communities to raise the bar even if the base standard
defines the response as optional.

In theory we could possibly say that responding revoked is optional, but
if you choose between revoked and unknown then you SHOULD favour revoked
over unknown. But such nested requirements just feels bad and impossible
to test compliance against. I'd much rather just leave it optional. I
think the Note gives a clear recommendation on this and the rationale
without spelling it out as a requirement.

Does this answer your question?

On 3/27/13 12:51 AM, "Black, David" <david(_dot_)black(_at_)emc(_dot_)com> 
wrote:

Hi Stefan,

This looks good - thank you for the prompt response.

It looks like my speculation on item [1] was wrong, so could you

respond

to the question below, please?:

[1] Section 2.2:

      NOTE: The "revoked" state for known non-issued certificate 
serial
              numbers is allowed in order to reduce the risk of 
relying
              parties using CRLs as a fall back mechanism, which 
would be
              considerably higher if an "unknown" response was 
returned.

Given this explanation, I'm surprised that the use of "revoked"

instead of

"unknown" for a known non-issued certificate is a "MAY" requirement

and

not a "SHOULD" requirement.  Why is that the case?


--------------

Beyond that, the proposed actions (or proposed non-actions) on items
[2]-[5]
are fine with me, Sean's taken care of the author permissions item from
idnits, and I assume someone has or will check the ASN.1 .

Thanks,
--David

-----Original Message-----
From: Stefan Santesson [mailto:stefan(_at_)aaa-sec(_dot_)com]
Sent: Monday, March 25, 2013 10:21 PM
To: Black, David; sts(_at_)aaa-sec(_dot_)com; 
mmyers(_at_)fastq(_dot_)com;

ambarish(_at_)gmail(_dot_)com;

slava(_dot_)galperin(_at_)gmail(_dot_)com; 
cadams(_at_)eecs(_dot_)uottawa(_dot_)ca; gen-art(_at_)ietf(_dot_)org
Cc: pkix(_at_)ietf(_dot_)org; Sean Turner; ietf(_at_)ietf(_dot_)org
Subject: Re: Gen-ART review of draft-ietf-pkix-rfc2560bis-15

Hi David,

Thanks for the review.
My reply in line.

On 3/26/13 1:25 AM, "Black, David" 
<david(_dot_)black(_at_)emc(_dot_)com> wrote:

Authors,

I am the assigned Gen-ART reviewer for this draft. For background on
Gen-ART, please
see the FAQ at

<http://wiki.tools.ietf.org/area/gen/trac/wiki/GenArtfaq>.


Please resolve these comments along with any other Last Call

comments

you

may receive.

Document: draft-ietf-pkix-rfc2560bis-15
Reviewer: David L. Black
Review Date: March 25, 2013
IETF LC End Date: March 27, 2013

Summary:
This draft is on the right track but has open issues, described in

the

review.

This draft updates the OCSP protocol for obtaining certificate

status

with some minor extensions.

Because this is a "bis" draft, I reviewed the diffs against RFC

2560.


I did not check the ASN.1.  I also did not see a writeup for this

draft

in the data tracker, and so will rely on the document shepherd to
ensure that the ASN.1 has been checked when the writeup is prepared.

I found five open issues, all of which are minor, plus one idnits

item

that is probably ok, but should be double-checked.

Minor issues:

[1] Section 2.2:

      NOTE: The "revoked" state for known non-issued certificate 
serial
              numbers is allowed in order to reduce the risk of 
relying
              parties using CRLs as a fall back mechanism, which 
would be
              considerably higher if an "unknown" response was 
returned.

Given this explanation, I'm surprised that the use of "revoked"

instead of

"unknown" for a known non-issued certificate is a "MAY" requirement

and

not a "SHOULD" requirement.  Why is that the case?

It appears that the reason is that the use of "revoked" in this

situation

may be dangerous when serial numbers can be predicted for

certificates

that
will be issued in the future.  If that's what's going on, this

concern

is

already explained in the security considerations section, but it

should

also be mentioned here for completeness.


No, this is not the main reason. The main reason is the one stated

as a

Note: in this section:

NOTE: The "revoked" state for known non-issued certificate serial
numbers
is allowed in order to reduce the risk of relying parties using CRLs

as

a
fall back mechanism, which would be considerably higher if an

"unknown"

response was returned.


[2] Section 4.2.2.2:

      The key that signs a certificate's status information need not 
be

the

      same key that signed the certificate. It is necessary however to
      ensure that the entity signing this information is authorized to

do

      so.  Therefore, a certificate's issuer MAY either sign the OCSP
      responses itself or it MAY explicitly designate this authority 
to
      another entity.

The two instances of "MAY" in the above text were both "MUST" in RFC

2560.


The RFC 2560 text construction of "MUST" or "MUST" is a bit odd, but

the two

"MAY"s in this draft are even worse, as they allow "MAY do something

else

entirely", despite being enclosed in an either-or construct.  I

strongly

suspect that the latter was not intended, so the following would be
clearer:

      The key that signs a certificate's status information need not 
be

the

      same key that signed the certificate. It is necessary however to
      ensure that the entity signing this information is authorized to

do

      so.  Therefore, a certificate's issuer MUST do one of the

following:

              - sign the OCSP responses itself, or
              - explicitly designate this authority to another entity.



I Agree. I will adopt your text.


[3] Section 4.3:

Is the "SHOULD" requirement still appropriate for the DSA with SHA-1

combo

(vs. a "MAY" requirement)?  This requirement was a "MUST" in RFC

2560,

but

I wonder about actual usage of DSA in practice.


The change in algorithm requirements was provided by RFC 6277, and
further
refined in this draft in accordance with requests from Sean Turner.


[4] Section 5, last paragraph:

      Responding a "revoked" state to certificate that has never been
      issued may enable someone to obtain a revocation response for a
      certificate that is not yet issued, but soon will be issued, if

the

      CA issues certificates using sequential certificate serial 
number
      assignment.

The above text after starting with the "if" is too narrow - it

should

say:


      if the certificate serial number of the certificate that
      will be issued can be predicted or guessed by the requester.
      Such prediction is easy for a CA that issues certificates
      using sequential certificate serial number assignment.

There's also a nit in original text - its first line should be:

      Responding with a "revoked" state for a certificate that has 
never

been

Good suggestions. I will update accordingly.


[5] Section 5.1.1:

      In archival applications it is quite possible that an OCSP

responder

      might be asked to report the validity of a certificate on a date

in

      the distant past. Such a certificate might employ a signing 
method
      that is no longer considered acceptably secure. In such
      circumstances the responder MUST NOT generate a signature using 
a
      signing mechanism that is not considered acceptably secure.

This could use an additional warning that certificate archival

should

not rely solely on signatures in archived certificates for ensuring

the

validity and integrity of the archived certificates because the

signature

algorithm(s) may transition to no longer being considered acceptably
secure at some point after the certificates are archived.


This note if I remember correctly is imported from RFC 6277, which is
incorporated into this document. The reason behind the text is only

to

avoid usages of insecure algorithms.
Historical validation is a real can of worms that I really would

like to

keep a tight lid on. I really want to avoid doing recommendations in
this
space as it may trigger a whole flood of things that could be equally
important to say about this subject.


Nits:

idnits 2.12.15 said:

 -- The document seems to lack a disclaimer for pre-RFC5378 work,

but

may

    have content which was first submitted before 10 November 2008.

If

you
    have contacted all the original authors and they are all

willing

to

grant
    the BCP78 rights to the IETF Trust, then this is fine, and you

can

ignore
    this comment.  If not, you may need to add the pre-RFC5378
disclaimer.
    (See the Legal Provisions document at
    http://trustee.ietf.org/license-info for more information.)

This looks like it's ok because all the authors of RFC 2560 are also
authors of
this draft, but it should be double-checked.



I defer this one to Sean. I think he has this one under control.


Thanks again for the review.

/Stefan


Thanks,
--David
----------------------------------------------------
David L. Black, Distinguished Engineer
EMC Corporation, 176 South St., Hopkinton, MA  01748
+1 (508) 293-7953             FAX: +1 (508) 293-7786
david(_dot_)black(_at_)emc(_dot_)com        Mobile: +1 (978) 394-7754
----------------------------------------------------