For the sake of expediency, I'm helping Hilda with some of these edits.

> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On Behalf Of John 
> R Levine
> Sent: Tuesday, December 13, 2011 7:37 PM
> To: ARF mailing list
> Subject: [marf] comments on draft-ietf-marf-authfailure-report-06
> 
> I see there's a new version of authfailure-report.  Overall it's
> getting there, but it still has some severe confusion about RFC 2119
> language. In 2119-ese, MUST means that if you don't do that, it won't
> interoperate at all so don't even try.
> 
> In section 3.1 of the draft:
> 
>     Authentication-Results:  Syntax as specified in [ARF]. Furthermore,
>        [ARF] specifies this field is OPTIONAL and appears at most once;
>        for this extension, this field MUST be present if that value is
>        available.  It MUST be formatted according to [AUTH-RESULTS], and
>        MUST reflect only a single email authentication method failure.
> 
> The ARF RFC doesn't define Authentication-results, RFC 5451 does.
> 
> I don't understand what "MUST be present if that value is available"
> means.

I imagine it's copied from the others.  The point there is that for SPF (for 
example), the client IP is needed to paint the whole picture, but for a DKIM 
report it isn't, and in either case it may not be available.  Same for the 
other parameters; different filtering points in the handling chain have access 
to different subsets of them.

> Referring to 2119, it would mean that if you have cruddy software that
> doesn't record the auth results so the value isn't available, it's OK
> to send a report without it, but if do you have it available but don't
> want to send it, you can't send a report at all, which is silly.  I
> understand that some people really really REALLY want their reports to
> be unredacted, but as we discussed a month or two ago, reporters will
> send what they're willing to send and no more, regardless of what an
> RFC says.  In fact, an auth failure report without this info is often
> entirely usable, so just delete this paragraph.

Do you really mean "delete this paragraph"?  Because we do want the field to be 
there.  RFC5451 gives a standard way to report this result, so it makes sense 
to either require that it be there, or don't send a report at all.

What about:

"Syntax as specified in <xref target="AUTH-RESULTS"/>.  Furthermore, <xref 
target="ARF"/> specifies this field is OPTIONAL and appears at most once; for 
this extension, this field MUST be present, but MUST reflect only a
single authentication method's result."

>        To report multiple failures for a single message, multiple reports
>        MUST be generated.
> 
> That sentence says that if you have multiple failures you have to
> report all of them, or else don't report anything.  Move it to just
> above section 3.1 and change it to:
> 
>        A single report describes a single authentication failure.
>        Multiple reports MAY be used to report multiple failures for a
>        single message.

That works.

> Then there are several sections like this one with the same MUST
> language:
> 
>     Original-Envelope-Id:  Syntax as specified in [ARF].  Furthermore,
>        [ARF] specifies this field is OPTIONAL and appears at most once;
>        for this extension, this field MUST be present if that value is
>        available.
> 
> The next three similar paragraphs cover Original-Mail-From, Source-IP,
> and Reported-Domain.  Same problem, same solution, delete all four of them.
> I suppose you could say something like it is RECOMMENDED to include
> these items to help better diagnose the authentication failure.

I'd prefer that last approach.  I didn't like the MUST either, but I appeared 
to be in the minority.

> In Section 4, the ABNF, the definition of auth-failure both says that a
> token is one of a list of types and that it's imported from [MIME].
> Importing seems wrong since that suggests it can be any tokenish
> string, but there's nothing that says that unknown auth-failure types
> are ignored.
> Suggest removing the references to token and instead enumerate the
> failure types, similar to the list of delivery results just below it.

That works.

> dkim-selector is also defined as a token, but actually it's a sub-
> domain.
> Import it from [DKIM].

Actually it's even easier to just import "selector" from [DKIM].

-MSK
_______________________________________________
marf mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/marf

Reply via email to