As some of you might recall, in the aftermath of Snowden revelations which cast aspersions on SP 800-90A DRBG’s security and integrity, NIST took the unprecedented step to withdraw SP 800-90A and open up a draft for comments.
This week NIST officially published a revised draft (found here) and has opened up public comments until May 23, 2014. Comments must be sent to RBG_comments@nist.gov with “Comments on SP 800-90A” in the subject line.
Acumen did a diff between this draft and the previously published standard and found the biggest change (no surprises here!) was removal of the Dual EC DRBG. This change was absolutely expected since it was this option of SP 800-90A DRBG that was considered to be weak (some say have a backdoor using skeleton keys)
Apart from this there are a number of other changes that could potentially impact future FIPS validations. These are enumerated below (Appendix F has a high level summary):
1. Definition of entropy source has been updated. Rather than pointing to SP 800-90B, this revision defines the entropy source as:
“A combination of a noise source (e.g., thermal noise or hard drive seek times), health tests, and an optional conditioning component that produce random bitstrings to be used by an RBG”
The interesting part here is that entropy source is defined as including health tests as well as (optional) conditioning. But this remains same as the definition in SP 800-90B so should be no surprises here.
2. In section 7, this revision changes the definition of DRBG. While previously it was “assumed” that DRBG are the mechanisms defined in SP 800-90A, the revision defines the DRBG as the DRBG mechanism and an approved entropy source (more on what is an entropy source is covered below).
3. Section 8.4 has been updated to require (it is a SHALL statement now) the crypto module not to use pseudorandom bits for an application that requires a higher security strength than the DRBG is instantiated to support. Based on the wording (and example provided) if the DRBG is instantiated to 128 bits security strength it cannot be used to generate 256 bit AES key. Previously it was understood that generating 256 bits of AES key would require two calls to the 128 bit based DRBG and as such security strength would be maintained (2 times 128 bits). The authors seem to imply this assumption is not correct.
4. Section 8.6.5 talks about entropy input and requires that the entropy input has to be obtained from within the cryptographic boundary. This could mean that for software modules (especially libraries) the entropy input would need to come from within the module itself. This potentially creates issues evaluating software modules that require HW entropy sources. The draft revision and SP 800-90C does define a concept of distributed DRBG boundary which could allow for hardware entropy sources however the last statement of section 8.5 seemingly closes the door on this by requiring that the sub-boundaries be the same or fully contained within a separate cryptographic module boundary. We believe there needs to be further clarification on DRBG boundary, distributed boundaries and how these relate to cryptographic module boundary.
5. Section 8.6.7 now requires that the nonce must be generated within the cryptographic boundary and defines how the nonce can be constructed (it provides four options). Note that it seems to be acceptable to generate the nonce in a different module as long as the module is FIPS validated and there is a secure channel to transmit the nonce between the two modules
6. Section 11.3.1 now explicitly defines in what order the health tests (KATs for the most part) will be performed. In our experience most modules do run the tests in the order specified however it does affect FIPS requirements and something to consider. Another requirement to note is that internal state needs to be carried forward between the various health tests. This is described in good detail in 11.3.2-11.3.4 sections.
As you can see above there are a few questions that need further clarification especially around the definition of DRBG boundary and it’s relationship to crytographic boundary. Acumen will be reaching out to the authors and CMVP to get a better understanding. We will provide updates on what we find. We will also be submitting comments seeking clarifications on #3 and #4.
It would be a good idea to keep an eye on this draft and track progress to ensure any design/code changes are accounted for in your development plan!
For those interested in looking at the raw diffs, they can be found here (requires Name and Email). Be warned the diffs look ugly!