Comments on a Formal Verification of PolarSSL
blog.regehr.org
blog.regehr.org
I would really like to see a crypto library with a good API; I also believe Rust would make a much better language to implement a crypto library in than C, and would serve the community better in the long run. If one agrees that this is the case (a better library needs to be built), then at some point, that project must get underway.
While I understand why the rule "crypto is hard. Don't implement your own." exists, at some point someone must break it, no? How should someone go about doing that? Who should go about doing that, since it seems like no matter how much I learn, "don't implement your own crypto"[1] will always apply? Otherwise, we're stuck here with OpenSSL, unable to do anything about it — no?
Of course, any new crypto library must fall in the "idiosyncratic" category; perhaps asked differently is there anything a new library could do to earn trust of people like yourself, and gain better acceptance? (e.g., tests?, though I do know that unit tests are not a proof of correctness in implementation.)
[1] and by "don't implement your own", I mean exclusively implementing the current state of the art algorithms, such as SHA2, AES, RSA, etc. I do not mean inventing your own algorithms, which is a whole other breed of crazy.
I guess it's what you have to do to make sure less skilled/experienced developers do the least-stupid thing they might do when it comes to crypto, but on the other hand, to not be on an openssl-emergency treadmill for the past few years, there's no alternative to being a very sharp and independent developer who uses something niche but quite well made like polarssl, or finally decides to attack the openssl codebase like the openbsd developers have done (and since you suggest their work, you must have confidence that they won't fk it up.)
Would this be a good use case for Rust? Are there any Rust TLS projects that are credible?
Mostly Rust uses libraries that wrap OpenSSL and attempt to do some verification of "safeness" before passing to OpenSSL
What's good though is that someone writing pure-rust-tls in a year or two can start with implementation that ignores anything below TLS1.2 and save a lot of code and scope for errors that way.