On 3/29/11 4:53 AM, Dave CROCKER wrote: > Jim, > > I found that I got seriously bogged down on some parts of your note, > as you and everyone else will surely see. I am glad to try to set up > a phone call to get a better channel for discussing this. Beyond the > obvious timezone challenges, this week, I've got quite a bit of > flexibility in time and am glad to find a slot where I can call you. > > Also, everyone else is hereby invited to jump in and straighten me out. > > That said, here's where I got in responding: > > > On 3/28/2011 11:27 PM, Jim Fenton wrote: >>> 1. "authors and their organizations" could be misinterpreted to mean >>> that the >>> conjunction defines a single identity. >> >> But the current text says "...examples include the author, ..." so that >> misinterpretation exists there as well. I'd be fine with just "authors' >> organizations". > > How is the "examples" list a misinterpretation? > > The list was crafted carefully to draw some distinctions that can be > significant. Your wording loses the distinction between author and > author's organization. > > I think the distinction is worth maintaining. > > Just to be clear: A domain name is capable of being author-specific. > I recognize that it's not typical, but the construct of 'author' is so > fundamental in this game, it's worth acknowledging that it is (still) > permitted.
With the output of DKIM being the SDID, the identity associated with the signature is of course that of the domain. But when my author-specific domain signs a message for me, it's the domain that does that -- it doesn't matter that it's an organization of one. Putting "author" here just hints that authors might sign messages as well, and I don't think that's intended. I suggest removing "author" to make that clear. > > >>> 3. One form of assessment service -- of which the late Goodmail was >>> an example >>> -- can give a priori assessment and then indicate tghe assessment by >>> providing >>> the signature to the message before it is sent. That is, the authoring >>> organization passes the message to the assessment service and the >>> assessment >>> service hands back the signature to be included in the message. (The >>> details >>> can vary, of course, but this describes the basic model.) Hence the >>> signature >>> is somewhat akin to a capability token. [I thought I had explained this >>> processing option a number of times over the years, specifically >>> citing the >>> Goodmail example.] >> >> That's a specific example of an ISP along the handling path. > > Goodmail was not an ISP and it was not along the path. > > > ................. Goodmail .............. > . . > V V > Client -> Mail -> Transfer -> Service -> Receiver -> Recipient > > Goodmail interacted with the creator of the document and, separately, > with the receiving mail service, as an adjunct "back office" service. > To repeat: /It was not in the direct handling path./ > > DKIM supports that mode of operation quite nicely and it is a > particularly powerful operational mode, so it is worth keeping that > "configuration" in mind explicitly. Given how persistent this > confusion seems to be it might even be worth more language, though I'm > not coming up with a suggestion, offhand. This still seems to me to be too specialized a use case to list in the specification, but would look to WG consensus on which way to go here. > > >> The potential for >> misinterpretation of this is greater than the benefit of explaining this >> potential usage scenario, especially since "assessment" has a very >> specific >> definition in the DKIM context. > > I think we've just seen a good example that indeed it is easily > misunderstood. That begs explicit reference, not potentially confusing > conflation. Can you offer alternate text that avoids the overloading of the word "assessment"? > >>>> 6.3 paragraph 5 has changed "signing identity" to SDID. The signing >>>> identity >>>> really corresponds to the AUID. >>> >>> That has not been correct for any version of rfc4871bis. The term >>> Signing >>> Identity has no normative value and is now only used in the >>> introductory text. >>> >>> Also note that the Update removed any meaningful semantics for AUID: >>> >>> The AUID comprises a domain name and an optional >>> <Local-part>. The domain name is the same as that used for the >>> SDID or is a sub-domain of it. For DKIM processing, the domain >>> name portion of the AUID has only basic domain name semantics; any >>> possible owner-specific semantics are outside the scope of DKIM. >>> >>> In fact, the AUID is not part of DKIM's formal output. So the formal >>> specification cannot then direct it be supplied to the assessment >>> engine. >> >> Nevertheless, suppose a message with From address >> <[email protected]> >> was properly signed with i=marketing.example.com and d=example.com. >> What the > > Your example has d= using a 'parent' domain, not a sub-domain. Your > following discussion refers to aspects of the spec that concern > sub-domains and I am not understanding how the example is relevant to > it. Yes, I see that i= has a subdomain but, again, I don't see how > that relates to your comments. Yes, d= is a parent domain, signing for a subdomain. The point is that the choice of i= had determined whether something ought to be flagged to the recipient. Now it doesn't. That is a behavioral change that is potentially incompatible with the RFC 4871 behavior. > > With obvious trepidation, I am going to raise a concern: On reviewing > the text, I find, under the Section 3.5 text for i= includes: > > "The Signer MAY choose to use the same namespace for its AUIDs as > its users' email addresses or MAY choose other means of representing > its users. However, the signer SHOULD use the same AUID for each > message intended to be evaluated as being within the same sphere of > responsibility, if it wishes to offer receivers the option of using > the AUID as a stable identifier that is finer grained than the SDID." > > I suggest that the first sentence change MAY to "might" in order to > make it non-normative. I really don't think changing MAY to "might" here is going to make things any clearer for implementers. Much the opposite. > > I further suggest removing the second sentence "However...". It is > giving (normative) usage guidance for something that it has already > made out of scope. It's only normative if the "if" clause of that sentence is satisfied, so I don't see a problem. > > >> text is telling us is that the mail system SHOULD take pains to >> ensure that >> example.com is visible to the user. This is counter to all of the >> text in the > > The closest I can come to what you describe in Section 6.3 is: > > "If the SDID is not the same as the address in the From: header > field, the mail system SHOULD take pains to ensure that the actual > SDID is clear to the reader." > > As always, anything DKIM says in the space of human factors varies > between problematic and bad. "clear to the reader" nicely falls > within that range. I know something about user interface design and > don't know how to satisfy this guidance, or what it's supposed benefit > is. > > Offhand, it appears to reflect the original misunderstanding that DKIM > "validates" the message. > > That said, I don't understand how making the SDID clear to the > 'reader' is relevant to your concern about AUID. I expected an objection on the grounds that this is a human factors issue. This probably does belong in the deployment document if anywhere. > > > >> DKIM specification that permits keys for a subdomain to be managed in >> a parent > > In what way is any of this counter (per your above assertion)? > > >> domain. If these is consensus to eliminate signing for subdomains, >> there is a > > You've jumped to a specific conclusion that implies a significant, > unstated logic sequence and I'm not yet understanding the premise, > never mind the rest. > > Please clarify. Let's discuss over the phone or something. I don't think I'm going to do any better in a typed description. > > >> lot of other stuff that needs to be removed from the spec, including >> the i= tag >> itself, the "s" flag in the key record, the text in section 3.9, and the >> security consideration in section 8.13. >> >> The Update removed semantics associated with the local part of the >> AUID, and not >> the domain-part. > > Whereas I think it made changes to the semantics for the entire value. > > Please point to the current version's specification of AUID "semantics". > > >> If there is not consensus to remove subdomain signing, the wording >> described >> here makes it meaningless. This goes to the heart of why I have been >> arguing >> that the output of DKIM should be the AUID (or its default value, >> which is the >> SDID), and not the SDID itself. > > I am not understanding how this language affects the use of AUID at all. (radical idea alert) The point is that if the AUID does not affect the result of the verification at all, why even have it? IN RFC 4871, we had very carefully put together the specification so that a parent domain could sign on behalf of one of its subdomains, and have the same signature semantics as if the subdomain had signed it (including the signature being considered an Author Domain Signature in ADSP). Over time, AUID has been watered down and caveated to the point that it is basically meaningless. If AUID doesn't have any normative effect on DKIM verification, we could eliminate it altogether and make the specification considerably easier to understand and implement, provided there is WG consensus to do so. ADSP already ignores the AUID, and as far as I can tell the only described use of AUID is to perhaps provide some more fine-grained reputation support, which is where you have proposed changing MAY to "might". I'm not really in favor of getting rid of the AUID, but I'm not in favor of keeping features in the spec that have no purpose either. -Jim _______________________________________________ NOTE WELL: This list operates according to http://mipassoc.org/dkim/ietf-list-rules.html
