On 06/03/17 17:44, Ryan Sleevi wrote:
> I'm assuming as with previous discussions, you'd like to keep the
> discussion on the list.

Yes, please :-)

> Overall: I would suggest every "should" be replaced with either a "must" or
> a "shall" RFC2119 style, to avoid any "best practice" vs "required mandate"
> confusion.

That is a good idea; however, doing so might run the risk of
accidentally making normative changes, which I want to avoid with this
update. After all, the direct replacement linguistically for "should" is
SHOULD, which may not be what we meant most of the time, but may
sometimes be.

What do people think - is this a safe enough change to make?

> 1.1 Scope
>   Item 2:
>     Bullet 1: This would allow the anyEKU to be considered 'out of scope'.
> Is that intentional? (notwithstanding Section 5.3.1)
>     Bullet 2: This potentially leads to confusion as to what it means to
> 'not allow' such types, given that nameConstraints only apply to the type
> for which they're present. That is, the absence of an iPAddress
> nameConstraint means there's no restriction, while the presence has to be
> constructed in a way to exclude all IP addresses in the excludedSubtrees.
> Similarly, as captured during the SRVName discussions in the CA/Browser
> Forum, there's uncertainty as to how to capture such an exclusion with an
> SRVName nameConstraint.
> 
>   I don't know how best to suggest rephrasing this, other than I think the
> scope may need to forward-reference a subsequent section that defines the
> technical means for that scope. I suspect you were trying to avoid this,
> but I think that to avoid ambiguity as to what the scope is, you'll want to
> ensure a precise technical definition is linked to the prosaic goal.
> 
>   Item 3: Similar to above, this allows excluding the anyEKU from scope
> (notwithstanding Section 5.3.1)

Are these issues also present in 2.4?

> 3
>   Item 2: I realize the intent is to match the current wording, but it may
> be worth considering clarifying here, in the event of an RA who performs
> validation/verification functions, but does not press the so-called "Big
> Red Button" to issue the cert. Imagine a process where CA receives a
> request, RA does domain validation (... incorrectly), RA tells Subscriber,
> Subscriber then asks CA to issue, CA issues now that RA has fulfilled the
> DCV - this is absolutely something for which multi-factor authentication is
> intended, in the current Mozilla policy, but which ambiguity regarding
> "directly" leads to uncertainty. Perhaps this is for a separate policy
> modification (and I'll let you track that work as appropriate), but perhaps
> "all accounts capable of causing certificate issuance or perform validation
> functions" can clarify this?

https://github.com/mozilla/pkipolicy/issues/60 .

>   Item 3: "verify certificate signing requests" may also lead to ambiguity
> as to whether this applies only to CSRs (which are but one way of
> manifesting a certificate request) or you mean "requests for certificates"
> or simply "certificate request" (omitting signing to avoid the CSR
> ambiguity)

https://github.com/mozilla/pkipolicy/issues/61 .

> 3
>   - Your markdown formatting for a-c is off :)

Hmm. This worked in the Markdown renderer in my editor. </me looks at
GFMD docs>

I did this because I didn't want to have two "bullet 1s" in the same
section. I have resolved this instead, in the two places it occurred, by
adding extra section headings.

> 3
>   - As you reformat this, perhaps it's worth borrowing the Microsoft of
> approach of mapping trust bits to criteria

Can you link to an example?

> 4.1.2
>   - You link to the "Baseline Requirements" document, but don't define what
> a BR audit is. While 4.1.1 lists audit criteria, this ambiguity may be
> undesirable. As with my immediately preceeding section, it may be worth
> mapping "trust bits" to "accepted audits", e.g. "For CA certificates which
> have the SSL trust bit set, we expect the following audits ..."
>    - Similarly, when two audit schemes are interchangable, it may be worth
> clarifying. For example, would Mozilla accept an ETSI TS 102 042 audit to
> the DVCP profile along with a WebTrust for cAs - 2.0 audit? My hope would
> be 'no', but the proposal leaves this ambiguous. https://aka.ms/auditreqs
> gives a clearer idea of what I'm thinking.

I've added lists of acceptable criteria beside each audit requirement.

Should we simply say that a given root (and I say root, as opposed to
'CA') has to be covered by all-WebTrust or all-ETSI auditing?

> 5.2
>   - There's a thread in CA/Browser Forum regarding what an "ASN.1 DER
> encoding error" is. Given 5280/X.509 describe the signature as over the DER
> encoding (but that the certificate doesn't necessarily match - and see the
> IETF discussion), perhaps its' worth clarifying that CAs must not _sign_
> such certificates.

So you are saying we should change the word "issue" to "sign"?

> 6
>   - Is this a subset, superset, or replacement for the Baseline
> Requirements?

Good question. Checking now, it's a mostly-complete subset although,
notably, our version doesn't say "within 24 hours". It certainly wasn't
intended to be a replacement (otherwise it would be called out in the
section which lists all the places where we modify the BRs).

I think we should remove this list of items as redundant, keeping the
other revocation-related stuff.

Gerv
_______________________________________________
dev-security-policy mailing list
[email protected]
https://lists.mozilla.org/listinfo/dev-security-policy

Reply via email to