Hacker News Reader

Comment by dfc | Hacker News Reader

I had a slightly different reaction to the "high standard" bit. I always thought Prof. Boneh was a high standard. Does anyone know if there were any obvious reasons to exclude SJCL?

View on HN

Disclaimer: I contribute to the core crypto library in Google End-To-End.

I was also a student of Prof. Boneh. I took his CS255, and became a TA for his infamous's Crypto I class on Coursera. So I guess at the end of the day it's still Boneh's teaching that has helped my colleagues and me create this library ;-).

SJCL is a great library, but it didn't quite work for us because:

* It isn't a Closure library. We want to use Closure because it supports types, which in turn make it easier and less error-prone to develop crypto code.

* It doesn't support typed arrays. We don't have typed arrays in End-To-End yet, but we're working on that.

* It doesn't support all the curves we want, and it seems that the main developer isn't interested in adding new curves, e.g., Curve25519.

* It doesn't have ciphers or signature schemes that we need such as RSA, Ed25519, deterministic ECDSA, etc.

Hello,

maybe a stupid question, but any reason why not use OpenPGP.js?

Is it even worse for your case than SJCL?

Not a stupid question at all. We actually considered this option, but OpenPGP.js looked pretty bad back then.

Security-wise the library wasn't in good shape. One of our cryptographers would "classify [OpenPGP.js] as trash". It has been audited recently, but the result doesn't look very good either [1]. I don't know the current status though.

OpenPGP.js didn't actually implement most of the ciphers - it just imported them from various sources. This made the library inconsistent, i.e., some functions expect string, while others expect byte array, which in turn made it harder to use correctly in a language like Javascript. If we chose OpenPGP.js, we needed to change these ciphers anyway, so we thought it's just better to write them from scratch.

[1] https://cure53.de/pentest-report_openpgpjs.pdf.

Thanks.

I considered using OpenPGP.js in one project, I didn't go so deep though. Thanks for the information.

I'm a contributor to OpenPGP.js. I understand your choice as the code was indeed in quite bad shape several months ago. I do want to point out though, that the code has gone through a big refactoring an cleanup since then: https://github.com/openpgpjs/openpgpjs/releases

We also fixed all critical, high and medium issue: https://github.com/openpgpjs/openpgpjs/wiki/Cure53-security-...

Having said that, a consistent rewrite using typed array and native web crypto apis under the hood does indeed sound very reasonable. I saw that native web crypto is not used throughout. What are your plans in regards to web crypto?

Also what is the predicted timeline for getting End-to-End into a production ready state? We would be quite interested in using it as a standalone library in our Chrome Packaged App: https://whiteout.io

Thanks

Thanks for the update on OpenPGP.js.

> What are your plans in regards to web crypto?

The plan is to use WebCrypto if it's available. We've moved RSA to WebCrypto, and the next targets are ECDH and ECDSA.

> Also what is the predicted timeline for getting End-to-End into a production ready state? We would be quite interested in using it as a standalone library in our Chrome Packaged App: https://whiteout.io

I can't tell you about our timeline for the extension. But if you just want to use the crypto library, you may want to wait for a couple of weeks, just to make sure none discovers any serious vulnerabilities.

I like WhiteOut. It's a great product in the right direction. We really want and will support the usage of the library in products like yours.

Thanks for for reply. I'm wondering if you know if it's possible to use the AES-CFB mode from the Web Crypto Apis, since the OpenPGP CFB (resync) mode seems to have special requirements?

> I like WhiteOut. It's a great product in the right direction. We really want and will support the usage of the library in products like yours.

Thanks! Is there a guide somewhere that explains how to build the standalone lib?

> Thanks for for reply. I'm wondering if you know if it's possible to use the AES-CFB mode from the Web Crypto Apis, since the OpenPGP CFB (resync) mode seems to have special requirements?

I haven't looked into it.

> Thanks! Is there a guide somewhere that explains how to build the standalone lib?

No, there isn't. But can you file a bug with us? I'll make sure we have something for you.

PS: how can I contact you?

My email is tankred@whiteout.io

IIRC, openpgp.js is what both Lavaboom and Protonmail use.

Have you guys took a look at https://github.com/indutny/elliptic ?

Oh we haven't. Looks like it's a nice library. One question: it seems that you use the message directly, instead of its hash, in ECDSA? [1].

[1] https://github.com/indutny/elliptic/blob/master/lib/elliptic...

PS: were you the guy that won the CloudFlare's HeartBleed challenge? great work :-).

I do expect users to hash the message before passing it to ECDSA, this way you could use any hashing library with it. Though, elliptic.js does actually depends on hash.js to seed it's PRNG.

Thank you!

> I do expect users to hash the message before passing it to ECDSA, this way you could use any hashing library with it. Though, elliptic.js does actually depends on hash.js to seed it's PRNG.

I think this isn't a good design because most people won't know that they must hash the message before passing it to the ECDSA. People will misuse it, and open themselves to attacks.

What you can do instead is to pick the right hash based on the curve, like what we did in End-To-End: https://code.google.com/p/end-to-end/source/browse/javascrip....

I actually already have a hash in `elliptic.curves` presets, but thanks for the idea!

Anyway, it would be interesting to compare speeds of our implementations. I'm really obsessed with try to reach 1ms verification, but so far only got to 4ms.

> his infamous's Crypto I class

s/infamous/famous.

In the long term, could you extract the underlying crypto library and make it available for others to use in easy form? Crypto.js is pretty difficult to use and I keep seeing people including the files into their projects because of this. I'm mainly looking for:

* Officially hosted on GitHub (or at least official mirror)

* Bower and NPM packet management

* Common.js modules for Node

Yes. This is something we or at least I would do in a few months.

SJCL does not support all of the primitives needed for OpenPGP.

I guess I should have been more clear. Why not add the primitives to SJCL and leverage an existing code base--a code base authored by someone who is considered to set a very high standard in crypto--instead of reinventing everything? Setting a higher standard than Prof. Boneh seems like a tough thing to do. Why not stand on Boneh et. al's shoulders?

It would be useful to have someone from Google elaborate on the summary they posted today. But I do recall that at least one of Prof. Boneh's postdocs mentioned in the original 2009 SJCL whitepaper is now at Google working on encryption...

[edit: Looks like that elaboration did happen elsewhere in the discussion -- just noticed it now.]

Reply on news.ycombinator.com