-----Original Message-----
From: ietf-dkim-bounces(_at_)mipassoc(_dot_)org [mailto:ietf-dkim-
bounces(_at_)mipassoc(_dot_)org] On Behalf Of Steve Atkins
Sent: Wednesday, March 11, 2009 4:36 PM
To: ietf-dkim WG
Subject: Re: [ietf-dkim] Another take on "all email from us is dkim
signed"
On Mar 11, 2009, at 1:20 PM, MH Michael Hammer (5304) wrote:
It seems to me that the domains likely to benefit from the ability
to
state "All email we send is DKIM signed" share a few things in
common.
1. They're concerned about other people sending email claiming to
be "from" the domain.
2. They send email using the domain to, typically, a large number
of B2C recipients (excluding the null assertion "we send no email",
which can be handled in other ways)
I would not exclude B2B out of hand. Many financial institutions
communicate with business customers. I would also point to this
example
http://www.webpronews.com/topnews/2007/10/31/grocery-store-falls-for-10-
million-phishing-scam as to why businesses might want to assert that
all
their B2B mail is signed.
The same reasoning applies to B2B email.
Which other ways would you assert "we send no mail" to protect the
RFC2822 From email address?
SPF is the obvious solution. MX 0 . is another.
There are a number of issues in using SPF in the manner you suggest. The
first is that SPF is path based. It basically says "here are the hosts
authorized to send mail for this domain" (transport layer). The RFC2822
From is data, not transport layer. This is exactly the complaint against
Microsoft/Sender-ID when they decided to misuse SPFv1 records for PRA
evaluation.
SPF protects the transport layer Mail From and EHLO. For most
implementations I'm aware of the intent and practice is to reject at
EHLO or reject on Mail From. At this point the RFC2822 From is not
available for checking. If a receiver has already checked
However, given the supposed threat is phishing, "we send no email"
is mostly a red herring.
3. All the email they send is DKIM signed.
4. They primarily care about mail appearing to be "from" their
domain being sent to users who also legitimately receive real email
from them.
It also seems that the number of domains who want this will likely
be
a small fraction of the total number of domains, and likely a small
fraction of the number of emails sent.
That may be true today but may not be true tomorrow.
The combination of 2, 3 and 4 means that any receiving ISP that
receives "forged" email that the domain cares about will also
receive
legitimate email from that domain.
Perhaps, perhaps not. There is also the case of an ISP receiving
fraudulent email prior to receiving legitimate email (race
condition).
If the threat is phishing, this is very unlikely. Phishing is
pretending
to be an existing brand. If the existing brand is already in use then
it's very likely that there's going to be at least one existing user
at
an ISP. It would also be very easy to setup your business model
such that you can ensure that every user has received at least
one legitimate email from you before they are in the situation
where they are at any risk of phishing. Sending them an email
with their username in it, for example.
A brand can have brand recognition without an individual (or receiving
domain) having received an email from that domain before. Take the
domain ibm.com. How many people are familiar with IBM as a
company/brand? How many people know that IBM never sends email from the
primary domain ibm.com? So how many receivers have received an email
from ibm.com? (the correct answer is none).
Consider the cases of Facebook, MySpace, and LinkedIn. Many
people/receivers were aware of the brands long before they received an
email from those brands. So, how would individuals/receivers know a
priori what the characteristics of a legitimate email from those brands
are or what a legitimate email from those brands should look like?
If the threat is not phishing, then we need to be explicit about what
the threat is before judging how different approaches affect it.
I've been talking about phishing. What other threats would you like to
throw into the fire?
If there were another field in the DKIM-Signature header, or an
entirely separate email header covered by the DKIM signature, that
stated "all email sent using this domain in the From field will be
DKIM signed" then any receiving MTA or MTA cluster could keep track
of
that state (probably using their existing reputation tracking
system
in the case of large receivers, and using a fairly trivial
extension
to their DKIM plugins in the case of smaller ones).
That would provide all the benefit of ADSP to senders who want it,
without adding a per-email latency overhead for receivers who want
to
support it, at the cost of a keeping a fairly small amount of state
at
the receiving MTA.
(Other information could be communicated in-band in the same way -
including "we're no longer dkim signing every email sent").
Why not include both options (trying to be flexible here)? If one
looks
at Daves affilias proposal, some receivers might choose to check for
ADSP records against some arbitrary list of domains (all registered
financial institutions for example).
If there's a third party list then this, and ADSP, are entirely
irrelevant.
Not at all. Before DKIM signing becomes universal (Mike starts singing
"that'll be the day"), receivers may choose to use various approaches to
authenticating domains and validating the nature of those domains. For
example, BITS has stated (and indicated specific time frames) for
members (financial institutions) to implement SPF and DKIM signing.
Therefore, a receiver may choose to look at a list of members provided
by BITS after the compliance date as a means of knowing that mail
purporting to be from these member institutions should have an ADSP
record. In this case ADSP is NOT irrelevant. Determining that a domain
SHOULD have an ADSP record does not tell us what that ADSP record
contains.
This, and ADSP, address the specific use case where senders
believe there should be no such third party list and that they should
be able to self-publish the information.
That doesn't mean that they couldn't both be used, just that they
shouldn't be conflated.
No conflation here.
_______________________________________________
NOTE WELL: This list operates according to
http://mipassoc.org/dkim/ietf-list-rules.html