Issuer field in the CRL should be byte-for-byte equivalent with that in cert

[Kathleen Wilson]

All,

I've added another Potentially Problematic Practice, as follows.

https://wiki.mozilla.org/CA:Problematic_Practices#Issuer_Encoding_in_CRL
The encoding of the Issuer field in the CRL should be byte-for-byte equivalent with the encoding of the Issuer in the certificate; that is, using the exact same string types and field contents. The specs (RFC 2459, RFC 3280, RFC 5280) permit them to mismatch, but that causes compatibility issues with various clients -- in such cases client software might not find the entry for the revoked certificate in the CRL.

As always, I will appreciate your thoughtful and constructive feedback.

We ran into this situation several times while adding entries to OneCRL for revoked intermediate certificates, because our script pulled the data from the CAs' CRLs where possible.

We have filed https://bugzilla.mozilla.org/show_bug.cgi?id=1330968 to update the OneCRL client to be encoding agnostic when doing the Issuer comparisons.

Cheers,
Kathleen

[Peter Gutmann]

Kathleen Wilson <kwi...@mozilla.com> writes:

>The encoding of the Issuer field in the CRL should be byte-for-byte
>equivalent with the encoding of the Issuer in the certificate; that is, using
>the exact same string types and field contents. The specs (RFC 2459, RFC 3280,
>RFC 5280) permit them to mismatch, but that causes compatibility issues with
>various clients -- in such cases client software might not find the entry for
>the revoked certificate in the CRL.

That's a specific case of the more general rule that there is only one
encoding mechanism and that is memcpy(), and by extension that there is only
one name/ID-comparison mechanism and that is memcmp() [0]. With a corollary
that anything that decides it's going to try and re-encode identifier fields
in a manner it considers more cromulent will quickly discover the universal
encoding mechanism.

So perhaps a better way to put it where you don't need to re-state it for
every case where it crops up is that a field or value, once encoded, should
never be re-encoded, but copied across byte-for-byte from wherever it's been
encoded to the destination location where it's used.

Peter.

[0] Anyone who wants to dispute this can go away and try and implement the
X.500 name comparison rules. Hint: These are impossible to implement
correctly, so I'm expecting a long wait...

[Jakob Bohm]

What should a CA do if they have used different encodings of the same
issuer DN in different certs that point to the same CRL location?

Wouldn't a more proper set of rules be:

1. The encoding of the Issuer field in the CRL must be the same as in
the SubjectDN in the issuing certificate.

2. The IssuerDN in all future issued certificates must the same as in
the SubjectDN in the issuing certificate.

3. If any previously issued certificates did not follow rule 2, the CA
must publish a list of the exact IssuerDN encodings used in such
certificates.

Enjoy

Jakob
--
Jakob Bohm, CIO, Partner, WiseMo A/S. https://www.wisemo.com
Transformervej 29, 2860 Søborg, Denmark. Direct +45 31 13 16 10
This public discussion message is non-binding and may contain errors.
WiseMo - Remote Service Management for PCs, Phones and Embedded

[Ryan Sleevi]

On Thu, Feb 2, 2017 at 3:59 PM, Jakob Bohm <jb-mo...@wisemo.com> wrote:

> On 02/02/2017 00:46, Kathleen Wilson wrote:
>

> What should a CA do if they have used different encodings of the same
> issuer DN in different certs that point to the same CRL location?
>
> Wouldn't a more proper set of rules be:
>
> 1. The encoding of the Issuer field in the CRL must be the same as in
> the SubjectDN in the issuing certificate.
>
> 2. The IssuerDN in all future issued certificates must the same as in
> the SubjectDN in the issuing certificate.
>
> 3. If any previously issued certificates did not follow rule 2, the CA
> must publish a list of the exact IssuerDN encodings used in such
> certificates.
>

What value does 3 provide?

[Jakob Bohm]

It provides data to allow services such as OneCRL (or even direct CRL
checking code in applications) to work around such past practices by
simply using a table of known Issuer name in certificate to actual
subject name in issuing certificate mappings, rather than trying to
implement fuzzy matching algorithms.

The data is minimal and compact, unlike CT logs. The complete dataset
from all public CAs combined is likely to be just a few kilobytes
uncompressed, much less compressed.

Potentially, this same table as a substitute for fuzzy mapping can also
be used when building/following certificate chains.

[Ryan Sleevi]

On Thu, Feb 2, 2017 at 9:37 PM Jakob Bohm <jb-mo...@wisemo.com> wrote:

> On 03/02/2017 05:22, Ryan Sleevi wrote:

> It provides data to allow services such as OneCRL (or even direct CRL
> checking code in applications) to work around such past practices by
> simply using a table of known Issuer name in certificate to actual
> subject name in issuing certificate mappings, rather than trying to
> implement fuzzy matching algorithms.

So don't you think it's useful and important to state the goal you're
trying to achieve when you propose Mozilla be prescriptive of all CAs?

Put differently, there is a considerably high bar for requiring CAs take
some action - or answer a particular question. Making sure it's clear why
can help evaluate the value of what you propose, otherwise it's very easy
to dismiss such suggestions as random non sequitors.

>
> The data is minimal and compact, unlike CT logs.

For example, this appears as a nonsequitor, because no one has proposed
sending CT logs to clients (the contextual way to read this suggestion,
given the statement that follows). Another way to read it (ignoring what
follows) suggests perhaps confusion between a CT log (which clients don't
download) and an SCT (which is minimal and compact for what it is). Or it
could be read that you're suggesting someone simply comb the CT logs,
extract the CRL URLs, and compare those against the issuer fields present.

While it seems highly unlikely you're proposing the last option, given your
expressed disdain and occasional confusion about CT, I mention it because
it's example of where the data can be independently obtained, rather than
using the threat of force (which is, implicitly, what is behind every CA
communication - the threat of removal), and arguably in a way that more
accurately captures practical impact.

The complete dataset
> from all public CAs combined is likely to be just a few kilobytes
> uncompressed, much less compressed.


>
> Potentially, this same table as a substitute for fuzzy mapping can also
> be used when building/following certificate chains.

Right, so just to follow:

You're proposing that, rather than list it as a problematic practice for an
edge case from a few CAs, Mozilla use a CA communication - possibly with a
change to the Baseline Requirements - so that they can write additional
code that no other client implements and ship several kilobytes (at least)
of data and code.

It does seem like Kathleen's current proposal is far more reasonable, and
so no, your proposal does not seem like a proper set of rules.

[Jakob Bohm]

The stated goal was to be *less* onerous than Kathleens proposal, which
was the the CRL issuer DN should be encoded exactly as it is in *every*
certificate referring to that issuer.

For the majority of CAs where this was always the case, neither
requirement would do involve any action at all. For the cases where a
CA has in the past issued certificates and CRLs with different IssuerDN
encoding, but has done so consistently, my proposal would allow the CA
to end its problematic practice without having to roll a new
intermediary CA with a different CRL service URL.

For the third case (which may or may not be as common as the second
one) where a CA has in the past issued certificates and CRLs with
different IssuerDN encoding, but has already mended its ways so more
recently issued certificates have the same IssuerDN encoding as in the
CAcert and CRL, the original proposal would have required *the same
CRL* to contain two conflicting IssuerDN encodings while my proposal
would allow such a CA to simply make a public statement that this is
the situation (with some specific details) and then continue.

>
> Put differently, there is a considerably high bar for requiring CAs take
> some action - or answer a particular question. Making sure it's clear why
> can help evaluate the value of what you propose, otherwise it's very easy
> to dismiss such suggestions as random non sequitors.
>
>>
>> The data is minimal and compact, unlike CT logs.
>
>
> For example, this appears as a nonsequitor, because no one has proposed
> sending CT logs to clients (the contextual way to read this suggestion,
> given the statement that follows). Another way to read it (ignoring what
> follows) suggests perhaps confusion between a CT log (which clients don't
> download) and an SCT (which is minimal and compact for what it is). Or it
> could be read that you're suggesting someone simply comb the CT logs,
> extract the CRL URLs, and compare those against the issuer fields present.

I was indeed suggesting the impracticality of both these causes of
action.

>
> While it seems highly unlikely you're proposing the last option, given your
> expressed disdain and occasional confusion about CT, I mention it because
> it's example of where the data can be independently obtained, rather than
> using the threat of force (which is, implicitly, what is behind every CA
> communication - the threat of removal), and arguably in a way that more
> accurately captures practical impact.
>
> The complete dataset
>> from all public CAs combined is likely to be just a few kilobytes
>> uncompressed, much less compressed.
>
>
>>
>> Potentially, this same table as a substitute for fuzzy mapping can also
>> be used when building/following certificate chains.
>
>
> Right, so just to follow:
>
> You're proposing that, rather than list it as a problematic practice for an
> edge case from a few CAs, Mozilla use a CA communication - possibly with a
> change to the Baseline Requirements - so that they can write additional
> code that no other client implements and ship several kilobytes (at least)
> of data and code.

No, I am suggesting that while *still* listing it as a problematic
practice for an edge case from a few few CAs, Mozilla offers those few
CAs an easier way out, while at the same time obtaining for both itself
and any other implementors (such as Google's BoringSSL and Microsoft's
CNG) a table of the only values that code for that edge case will need
to handle.

I was also suggesting, that if, after gathering data, the resulting
table is very small, using the table in code might be easier than
coding an algorithm that matches certificates to issuers and CRLs for
all the needed non-identical cases. This however would be an
implementation choice, as any other algorithm giving correct results
would solve the problem.

>
> It does seem like Kathleen's current proposal is far more reasonable, and
> so no, your proposal does not seem like a proper set of rules.
>

[Kathleen Wilson]

On Friday, February 3, 2017 at 7:26:14 AM UTC-8, Jakob Bohm wrote:
>
> No, I am suggesting that while *still* listing it as a problematic
> practice for an edge case from a few few CAs, Mozilla offers those few
> CAs an easier way out, while at the same time obtaining for both itself
> and any other implementors (such as Google's BoringSSL and Microsoft's
> CNG) a table of the only values that code for that edge case will need
> to handle.
>
> I was also suggesting, that if, after gathering data, the resulting
> table is very small, using the table in code might be easier than
> coding an algorithm that matches certificates to issuers and CRLs for
> all the needed non-identical cases. This however would be an
> implementation choice, as any other algorithm giving correct results
> would solve the problem.
>

There is a bug to "Make OneCRL name comparisons encoding agnostic"...
https://bugzilla.mozilla.org/show_bug.cgi?id=1330968

But we'll still call it out as a problematic practice.

Cheers,
Kathleen