Hi Everyone,

This is just a reminder email for tomorrow’s crypto-lib meeting (at the usual time).

Some agenda items:

1.       Discuss the “tiering” process for implementations and any guidelines we want to have for this going forward.  Dan Middleton sent out an email about this earlier.

2.       Naming.  Are we happy with Ursa?  We need to straighten everything out with the marketing committee no matter what we decide.

3.       Final proposal edits/suggestions.  Is there anything we still need to change in light of recent discussion?

Thanks a lot, and I hope to hear from many of you tomorrow.

Thanks,

Hart

Hi Dan,

Sounds great!  On your points:

1.       I think there will also be some grey area for tiering.  We will probably have to have maintainers collectively decide these things through votes.  We can have a standardized criteria, but I bet I could find a protocol where we would disagree on tier even given the standardized critieria.  This is a good thing to discuss though and something we will definitely have to consider more going forward.

2.       Is anyone not in favor of Ursa?  It sounded like everyone was pretty happy with it, provided we had a sufficiently scary logo.  If we’re happy with it, we should ask the marketing committee to approve.  Hopefully this shouldn’t be a long discussion tomorrow.

Thanks,

Hart

Tomorrow I’d like to use the blake PR to have some discussion about standards and tiers per

https://lists.hyperledger.org/g/labs/topic/sharedcrypto_3rd_party/27631934?p=,,,20,0,0,0::recentpostdate%2Fsticky,,,20,2,0,27631934

Time permitting maybe we could also settle on a name. I don’t like to burn a lot of time on name-the-baby activities but it sounded like we were all in favor of Ursa. If so, we could formalize that.

--Dan

Yes, that criteria list was heavily biased towards standard implementations. To accommodate the other 2 tiers it should be expanded to quantify the level of review of the algorithm.

When it comes to government required implementations, I’m willing to take the position we declare that as out of scope.

On the one hand, if someone wants to contribute something that’s great. On the other hand, each thing we add costs more overhead in a variety of areas including maintaining the build and CI much less security review. The implications, both good and bad, of government implementations are probably significant but they are beyond what I have time to consider right now. I think the fail-safe is to declare them out of scope for the time being and re-evaluate in the future.

Thanks,

Dan

This is a good point.  However, there are many other criteria that we might use to assess the confidence we have in standards or libraries we are using—the ones Dan lists here are only related to the practical issues around code implementation.  Other questions to consider include things like how much the algorithms being implemented have been studied and/or peer reviewed, whether the library implements a “sketchy” government-designed algorithm that has the potential for back doors, and what cryptographic assumptions the implementations are based upon.

I’m generally in favor of being more permissive in terms of implementations we add to the project (if someone wants to contribute them and they are useful and seemingly secure, then why not).  However, whatever build processes we use should heavily flag nonstandard or nontraditional implementations, and it should be impossible for a user to “build” crypto-lib with such algorithms unintentionally.

Exactly how we want to rank or rate code dependencies (including ones that we potentially write!) is a good thing for open discussion.  We can discuss our approach for something like this at our meeting next week if people like.  I doubt we’ll get universal agreement, but that’s OK—our goal should again be to just inform people using the less standard stuff and make sure they are aware rather than dictate exactly what they should use.

Thanks,

Hart

I’m fine with the updated maintainers concept.

We can enforce n maintainer reviews with github. We already do that in sawtooth. I’m still planning to get the contributor rules writeup in as a PR before our next meeting.

--Dan

Hi Everyone,

I just wanted to give a brief update.  At the TSC meeting today, we discussed the crypto-lib proposal for a little bit and got some good feedback.

The main suggestion (from Chris Ferris) was that it would be simpler (and better for expanding the project in the future) if, instead of having “stewards” and maintainers, we just classified everyone as maintainers and separated them into lists.  This would mean we have a list of “theoretical maintainers” which would currently be our stewards, and “base signature/Zmix maintainers” which would be our current maintainers.  This would simplify our review process, since we could just require people on the “theoretical maintainer” list to sign off on changes that are algorithmic in nature, and this could be done natively in, say, Gerrit.  It would also mean our project structure would be much more like established projects, which is almost certainly a good thing.

I don’t think this is a radical change (or even much of a change) from what we had in mind and shouldn’t change the way things are currently working.  The semantics are just a little bit different, and it will help in the future if we want to further define roles (i.e. security code review expert, or post-quantum cryptography expert) to get appropriate reviews.

Does anyone have any thoughts on or objections to this change?  If not, then I’ll modify the project proposal to reflect this.

Thanks,

Hart

This is a good point.  However, there are many other criteria that we might use to assess the confidence we have in standards or libraries we are using—the ones Dan lists here are only related to the practical issues around code implementation.  Other questions to consider include things like how much the algorithms being implemented have been studied and/or peer reviewed, whether the library implements a “sketchy” government-designed algorithm that has the potential for back doors, and what cryptographic assumptions the implementations are based upon.

I’m generally in favor of being more permissive in terms of implementations we add to the project (if someone wants to contribute them and they are useful and seemingly secure, then why not).  However, whatever build processes we use should heavily flag nonstandard or nontraditional implementations, and it should be impossible for a user to “build” crypto-lib with such algorithms unintentionally.

Exactly how we want to rank or rate code dependencies (including ones that we potentially write!) is a good thing for open discussion.  We can discuss our approach for something like this at our meeting next week if people like.  I doubt we’ll get universal agreement, but that’s OK—our goal should again be to just inform people using the less standard stuff and make sure they are aware rather than dictate exactly what they should use.

Thanks,

Hart

I propose we establish some standards for libraries we will incorporate in crypto-lib (or Ursa or whatever we will soon call it :)  )

As a motivating example there’s a PR to add a blake2 library. I’ve not independently verified the performance claims but it looks like it is quite fast. In the risk department, though, the source repo indicates a single contributor and only 2-3 months of history. The latter raises risks that the code is not hardened and the former is a risk that it won’t be maintained.

The different tiers we establish complicate having a single list of criteria. Without being too rigid we could probably make a matrix of what degree applies to which tier. Here’s a starter list of criteria:

• Maturity (how long has this code existed)
• Maintainer count (how likely is the code to be maintained and issues responded to)
• Community size (are there active mail lists and users that indicate it’s in active use)
• Bug reporting (is there a way to submit security bugs)
• What is the maintenance history (regular updates, patches, responsiveness for CVEs)?
• Known issues (due diligence that the code is sound)
• Are there protected releases (can we depend on signed libraries)

Taking `maturity` as a simple example we could set the levels for the 3 tiers as

Standard:            1 year

Semi-Trusted:   3 months

Research:            NA

Interested in feedback on this approach.

Regards,

Dan

Hi Everyone,

This is just a reminder for tomorrow’s crypto-lib meeting (7:00 AM Pacific time, as usual).  Some things to discuss:

1.       My perception is that we’re on the cusp of proposing crypto-lib to the TSC for incubation status.  I’d like to discuss this with everyone at the meeting.  As such, if you haven’t gone over the proposal document recently, please take a look: https://docs.google.com/document/d/1JtFT5L-82egj6shgGXzTsNAg6_UHuMheKfsst6NS_Xo/edit?usp=sharing.  I have made some edits recently after discussions with Dan M and Shawn, among others.  Ideally, we could announce to the TSC this week that we were planning on asking for a vote in the near future.

2.       Hackfest recap:  we will go over the progress made at the hackfest and next steps.

3.       Whatever else people want to talk about.

Thanks a lot for everyone’s time, and I hope to speak with many of you tomorrow.

Thanks,

Hart

I can't promise anything since I haven't confirmed yet, but I might be able to offer space at our Sunnyvale campus either before or after Real World Crypto  (but probably before, since it's a Wed-Fri event) this coming January.  If it's a small group, then I can almost certainly do it.

I doubt RWC is a good venue for a hackathon though--people are usually pretty drained from going to the talks all day.  I've never seen anyone try to do a hackathon in conjunction with an IACR conference before, and I'd be worried about turnout (RWC may be an exception to this rule, but I'd be surprised).  Most people are there to go to talks and socialize, not hack.  It would be difficult to get new people. 

Are lots of people on this list planning on coming to RWC?  If so, at minimum I can try to plan for a crypto-lib workday on the Tuesday before RWC.

Thanks,

Hart