ietf-dkim
[Top] [All Lists]

Re: [ietf-dkim] Re: draft-ietf-dkim-threats-00 Unreasonable estimate of impact from a highly probable exploit

2006-01-29 12:07:00

On Jan 26, 2006, at 4:45 PM, Jim Fenton wrote:

Here's my rationale for the values I put in the document regarding Signed Message Replay.

There should perhaps be a third column in addition to "impact" and "likelihood"; let's call it "utility". The Signed Message Replay attack in section 4.1.5 differs from nearly all of the other attacks in that the attacker can't do everything they'd perhaps like. If they steal the private key for the domain, they can sign any message they like and send out advertising or objectionable material, etc. With the SMR attack they can't do that -- they can only inappropriately distribute things that are already signed. This can be bad for the domain's reputation, but if the attacker is looking to sell pharmecuticals, this isn't going to help.

So I marked this attack as "low impact", in part because of its limited utility, and in part because (in accordance with the definitions in section 1.2) this attack deals with individual messages -- it doesn't affect the majority of the messages from the domain, only those selected by the attacker for replay. Perhaps the definitions need adjusting but I think we need some objective means for scoring the threats.

I do recognize my role on this is as a document editor, and I will adjust the document to group consensus, but I wanted you to know where I was coming from.

The Low/High impact of replay abuse depends upon the expected role of a DKIM signature, and how a "good" actor can be differentiated from that of a "bad." Challenging this "Low" impact rating is questioning whether acceptance of the _message_ or the _signing-domain_ will be modified as a defensive measure. To suggest the impact from replay- abuse would be "Low" also suggests an expectation of isolating assessments on either a matching email-address or perhaps the signature, but not on the signing-domain. It seems that one should consider what this approach entails when resolving accountability beyond that of the signing-domain.

With tens of millions of domains, each domain may also support hundreds of millions of email-addresses. There is virtually no cost associated with an email-address, nor is there a limit to the number of email-address that might be created. Using the signature itself would further increase the assessment database by orders of magnitude. If the recipient finds that resolving beyond the domain to be problematic, there would be two choices:

 1) hold the signing-domain accountable for the abuse

 2) ignoring the signature altogether


When a newsletter or promotional ad is distributed well beyond a list of subscribers, is this a case of:

 1) an unintended replay by a miscreant

2) the marketing department relying upon an inability of an assessment to differentiate good from bad


When dealing with large organizations, these assessments must be defendable. Many companies already use third-parties to distribute these types of messages. When enough of this abuse becomes common place, any such abuse may result in the signing-domain being blocked, perhaps out of frustration.

Many domains leak rate restricted abuse from perhaps tens of thousands of compromised systems offered access. Monitoring or outbound filters will not be effective control measures. There are any number of messages between friends and colleagues that take the form:

Give me your opinion on this.
http://tinyurl.com/1234


Is it practical for a recipient to assess accountability for individual messages? Should the recipient filter on signatures? Should the recipient attempt to hold each email-address separately accountable? The unlimited supply of email-addresses would seem to suggest that this strategy may be futile.

It seems from the Low impact assessment, you expect the recipient to make significant accommodations and to pick and choose messages from within a signing-domain. It would also seem that by even attempting to make this type of accommodation, a signing-domain would need to constrain the use of email-addresses. Domains that offer email- addresses like list-master will be in greater peril unless each posting changes this email-address.

-Doug



-Doug _______________________________________________
ietf-dkim mailing list
http://dkim.org