Should LMS Verify be included in OpenSSL 3.5 release scheduled for April?
All, the BAC has been asked to reach out to the community and see what the thoughts were around including LMS verification in the OpenSSL3.5 release. This release is scheduled for April 2025. It appears the dev is mostly done. However delivering this could delay the inclusion of SLH-DSA in the OpenSSL3.5 release. So as Tim Hudson put it... "So the question really is would half-of-LMS be a higher priority to all of SLH-DSA for OpenSSL-3.5?"
What say you all? If I knew how to do a poll here I would but I haven't a lot of experience with this tool. I will learn :). For now it will be manual.
Additionally I would like to start regular meetings for the large business community. Working thru what that might look like now. Stay tuned. Thanks!
I'm against inclusion of LMS. It's an algorithm that next to nobody will be able to deploy securely.
I've completely opposed to inclusion of signing code of LMS: software implementation of LMS cannot be used safely.
SLH-DSA is far more useful with none of the pitfalls of LMS.
Red Hat is much more interested in SLH-DSA and not interested in LMS signing (and considers it totally unsafe)
We'd definitely prefer availability of SLH-DSA over LMS verification. We're looking at how to sign software packages in the future, and the current choices are essentially:
ML-DSA — reasonably small and fast, but new math
SLH-DSA — larger and slower, but based on hashes, which we understand very well
LMS or XMSS — can catastrophically fail if the state of the private key is not carefully managed, something that should probably only be done in hardware (and the entire backup and restore discussion with NIST in the Cryptographic Module User Forum shows that even then, there will be possible problems).
CNSA 2.0 forces us to chose from (ML-DSA, LMS, XMSS), but we'd rather not rely on new math for software package signatures only, and the pitfalls of LMS and XMSS sound scary. We'd prefer being able to use both ML-DSA and SLH-DSA.
Given a choice between one or the other, I would prefer to see SLH-DSA.
We also vastly prefer ML-DSA and SLH-DSA over LMS.
I'm one of the three people working full time on PQC for OpenSSL 3.5, the others being Shane Lontis (Oracle) and Viktor Dukhovni (OpenSSL). Neither Shane nor I believe that LMS's inclusion will materially impact SLH-DSA's. The LMS work is essentially complete, the remaining effort is in reviewing which can not be done by us (so it's not taking effort from the other algorithms). Review feedback might create more work, but it will be relatively minor in all likelihood.
Some notes:
OpenSSL 3.5 is slated to be the next LTS version.
ML-DSA is the current top priority for us.
This proposal is only about LMS verify.
Verify is the only part of LMS that can be FIPS validated for software modules.
As noted several times, LMS signing has lots of pitfalls, this is not being proposed here.
CNSA 2.0 is very forward looking. At the time of publication, and still, majority of software / hardware / firmware have no support for things it demands. Thus any amount of time here or there, will not have any material impact.
In the now de-published first version of the CNSA 2.0 document, both "Web Browsers and Cloud Services" (whatever that actually means in practice, possibly key exchange using ML-KEM, possibly ML-DSA signatures in TLS, it didn't say) and "Software Signatures" had timelines that expected change in 2025. For a timeline this tight, the choices we make now matter.
The recently published FAQ document no longer lists this explicitly, or I couldn't find it. It does however say "no requirement to transition will be enforced prior to December 31, 2025". The same paragraph says "CNSSP 15 states that by January 1, 2027, all new acquisitions for NSS will be required to be CNSA 2.0 compliant unless otherwise noted."
To me, this sounds like new purchases that have to comply with CNSA will no longer be able to buy products that don't support PQC in 2027. That's not a very long time for LTS distributions, for example.
@Clemens Lang note that procurement that requires CNSA is very small, compared to FIPS. And FIPS PQC timelines are currently in draft but are for 2030+ timeframe; plus certification on top; thus realistically roughly 2031. Still not much time, but it is closer to 2 - 2.5 OS LTS cycles.
My understanding of PQC in FIPS is that ACVP testing for ML-KEM and ML-DSA is available already, and the ACVP docs also already mention SLH-DSA and LMS. Atsec has obtained CAVP certificates for PQC already, so it seems like you can FIPS-certify PQC now if you want.
Can you point me to a source for your statement on FIPS PQC timelines?
Due to severe usability issues with LMS, I strongly encourage not to include it all, or have it off by default. If it is on by default, may distributions are likely to turn it off on security grounds. Or only offer it as opt-in / as part of separate builds. As usage of LMS is likely have no practical need. It is sort of dead on arrival.
Oracle has discussed internally and would like to see support for PQC in the following order of priority: ML-KEM, ML-DSA, LMS Verify, SLH-DSA.
Regarding the order of SLH-DSA & LMS Verify:
SLH-DSA does not have a known use case within Oracle currently, but Oracle would still like it included if possible.
Some companies have been following NSA and adding HSS/LMS for signing in HSMs, like Luna https://cpl.thalesgroup.com/sites/default/files/content/solution_briefs/field_document/2023-08/thales-luna-post-quantum-crypto-functionality-module-sb.pdf. The algorithm for verification needs to be available in the crypto provider used in the firmware.
Oracle is optimistic that both LMS verify and SLH-DSA can be included for the following reasons:
Shane Lontis from Oracle has been responsible for dev work for both of these algorithms so Oracle has insight into the status of this work.
LMS verify development work is complete and has high testing coverage. The work has had 1 review and only requires 1 more review to complete.
Shane Lontis and Paul Dale are currently working on ML-DSA & SLH-DSA.
The 1 outstanding LMS verify review must be from someone other than Shane or Paul, which means it should not affect their progress on SLH-DSA.
>>Tim Chevalier: Given a choice between one or the other, I would prefer to see SLH-DSA.
As Chris points out, that choice is not a given. So I think the question is simply whether to include LMS Verify since it is nearly done.
>> Alicja Kario: ...opposed to inclusion of signing code of LMS:
>> software implementation of LMS cannot be used safely.
>> Dmitry Belyavsky: ...not interested in LMS signing (and considers it totally unsafe)
I have concerns about LMS signing as well, but that's not what this is about. We want to know whether to include LMS verification (no signing).
This is an opportunity to include LMS Verify in the next FIPS validation (FIPS covers verify but not signing in software) and CNSA 2.0 includes it, so why not add it?
This is an opportunity to include LMS Verify in the next FIPS validation (FIPS covers verify but not signing in software) and CNSA 2.0 includes it, so why not add it?
Because next to nobody will be able to generate those signatures. Having code that's there but unused is just increasing the attack surface for no benefit. We don't trust NSA's insistence on LMS, XMSS and explicitly excluding SLH-DSA. Same for their exclusion of HSS and XMSSMT.
Pretty use case dependent. Lower in the stack LMS is being used today by HSM vendors for firmware/image signing and on-prem equipment vendors. Obviously this is not a use case for everyone. LMS signature verification does have value IMO. For other use cases further up the stack, sure, LMS is not ideal nor even a good choice. I'll leave the crypto discussion to cryptographers (that's not me). :)
Since CNSA 2.0 includes LMS and some HSM vendors do support it, let us include LMS verify so that the implementation is complete by all respects.
What key encoding format is used for LMS verify?
Hi!
AWS and Amazon are not opposed to LMS Verify but we would prefer to prioritize SLH-DSA if there is a resource constraint. We consider ML-DSA to be the highest priority.
We would also ask that the SLH-DSA implementation makes sure to take https://eprint.iacr.org/2024/018.pdf in consideration and that it be be parametrizable to support future smaller SLH-DSA parameters coming from NIST.
Thanks!
Benjamin
Hi,
Even though OpenSSL 3.5 was already released last week without LMS Verify support, I just want to point out that NIAP certification currently supports only LMS/XMSS in its published protection profiles. And as per the latest CNSA 2.0 FAQ, CAVP validation (not CMVP) of LMS Verify should be enough for NIAP certification.
As the deadline for CNSA 2.0 SW/FW signing is more urgent, Versa Networks would prefer to include LMV Verify in OpenSSL as soon as possible.
Thanks!
Hi Jeff,
I don't think the availability of LMS is so urgent, that half of an implementation is better than the full scope implementation. As it is part of the SP 800-208 recommendation it needs to be available eventually, however.
From my perspective an LTS version would have much higher priority, as we need to support the shipped OpenSSL versions for multiple years and the current LTS version is nearing its public end of support date already.
Just my 2 cents ...
-- Matthias