Based on discussion with the authors, the -05 version of this draft resolves the issues raised in the Gen-ART review of the -04 version. An important element of the approach taken to issue [1] has been to explain why the security requirements for redaction are significantly weaker than the strength of the secure hashes that are suggested by the draft. Thanks, --David > -----Original Message----- > From: Black, David > Sent: Tuesday, January 10, 2012 9:44 PM > To: ietf at cybernothing.org; Murray S. Kucherawy; gen-art at ietf.org; ietf at ietf.org > Cc: Black, David; marf at ietf.org; presnick at qualcomm.com > Subject: Gen-ART review of draft-ietf-marf-redaction-04 > > I am the assigned Gen-ART reviewer for this draft. For background on Gen-ART, please > see the FAQ at . > > Please resolve these comments along with any other Last Call comments you may receive. > > Document: draft-ietf-marf-redaction-04 > Reviewer: David L. Black > Review Date: January 10, 2012 > IETF LC End Date: January 18, 2011 > IESG Telechat Date: January 19, 2011 > > Summary: This draft is on the right track but has open issues, described in the review. > > This draft specifies a method for redacting information from email abuse reports > (e.g., hiding the local part [user] of an email address), while still allowing > correlation of the redacted information across related abuse reports from the same > source. The draft is short, clear, and well written. > > There are two open issues: > > [1] The first open issue is the absence of security guidance to ensure that this > redaction technique effectively hides the redacted information. The redaction > technique is to concatenate a secret string (called the "redaction key") to the > information to be redacted, apply "any hashing/digest algorithm", convert the output > to base64 and use that base64 string to replace the redacted information. > > There are two important ways in which this technique could fail to effectively hide > the redacted information: > - The secret string may inject insufficient entropy. > - The hashing/digest algorithm may be weak. > > To take an extreme example, if the secret string ("redaction key") consists of a > single ASCII character, and a short email local part is being redacted, then the > output is highly vulnerable to dictionary and brute force attacks because only 6 bits > of entropy are added (the result may look secure, but it's not). Beyond this extreme > example, this is a potentially real concern - e.g., applying the rule of thumb that > ASCII text contains 4-5 bits of entropy per character, the example in Appendix A > uses a "redaction key" of "potatoes" that injects at most 40 bits of entropy - > is that sufficient for email redaction purposes? > > To take a silly example, if a CRC is used as the hash with that sort of short input, > the result is not particularly difficult to invert. > > I suggest a couple of changes: > 1) Change "any hashing/digest algorithm" to require use of a secure hash, and > explain what is meant by "secure hash" in the security considerations section. > 2) Require a minimum length of the "redaction key" string, and strongly suggest > (SHOULD) that it be randomly generated (e.g., by running sufficient output > of an entropy-rich random number generator through a base64 converter). > > For the latter change, figure out the amount of entropy that should be used > for redaction - the recommended string length will be larger because printable > ASCII is not entropy-dense (at best it's good for 6 bits of entropy in each > 8-bit character, and human-written text such as this message has significantly > less). > > From a pure security perspective, use of HMAC with specified secure hashes > (SHA2-family) and an approach of hashing the "redaction key" down to a binary > key for HMAC would be a stronger approach. I suggest that authors consider > approach, but there may be practical usage concerns that suggest not adopting it. > > [2] The second open issue is absence of security considerations for the redaction > key. The security considerations section needs to caution that the redaction key > is a secret key that must be managed and protected as a secret key. Disclosure > of a redaction key removes the redaction from all reports that used that key. > As part of this, guidance should be provided on when and how to change the > redaction key in order to limit the effects of loss of secrecy for a single > redaction key. > > Editorial Nit: I believe that "anonymization" is a better description of what > this draft is doing (as opposed to "redaction"), particularly as the result is > intended to be correlatable via string match across reports from the same source. > > idnits 2.12.13 didn't find any nits. > > Thanks, > --David > ---------------------------------------------------- > David L. Black, Distinguished Engineer > EMC Corporation, 176 South St., Hopkinton, MA  01748 > +1 (508) 293-7953             FAX: +1 (508) 293-7786 > david.black at emc.com        Mobile: +1 (978) 394-7754 > ----------------------------------------------------