ietf-mailsig
[Top] [All Lists]

DoS and Replay protection for message signatures

2005-08-01 00:58:43

Retraining a history of behavior based upon authenticated identifiers offers a means to protect systems from abuse that could otherwise lead to a denial of service. However, message signatures offer _no_ authenticated identifier prior to resources being committed. While a signature's history could accrue against the domain name used to obtain the key, acting upon a bad signature's history by refusing service against the domain name comes too late to preserve resources. This lack of resource protection becomes even more significant when confronting replay attacks.

One of the strengths of message signatures is having an ability to be independent of the IP address. A majority of domain owners share their address space, which makes a protection scheme based upon the domain name rather than the IP address desirable. This goal can never be reached when IP address abuse history must be applied first in order to preserver resources. The use of IP address history means either high levels of abuse must be tolerated, or a high level of collateral blocking may occur.

The HELO authentication process is essential for providing resource protect when domain names are used as a basis for history accrual. Suggestions have been made that a history of a key itself could be used. While this may help identify new and possibly poisoned keys, this type of enhancement is better done with ancillary checks when new keys are discovered. While retaining a key hash may be one method to track when a key changes, using non-recursive lookups could perhaps be another method. The types of ancillary checks that could be used are many, but none of these schemes provide any resource protection.

Authenticating the HELO offers a solution to allow the feeding forward of a domain name's abuse history to a point in the process where resources can be protected. Should it then be determined that the Signature's domain name is a parent of HELO, there should be no need to do further history or anti-replay checks. This is where message signatures would benefit by a convention where there is a signing domain's relationship with the HELO in most cases.

For larger domains, the use of an opaque revocation-identifier is common. DKIM should establish conventions for the placement of this optional information, where it is included within the signature. This convention could also establish a means to self-publish "bad- identifier" list. This "bad-identifier" checking also can work together with the HELO authentication to eliminate a need to perform the check.

There is another efficiency that could be obtained by authenticating the HELO prior to checking the signature. Assertions of the types of mechanisms supported can be established concurrently with the HELO authentication. It seems a holistic approach to message signatures should include a prior host authentication operation.

-Doug

<Prev in Thread] Current Thread [Next in Thread>