Various changes from Chromium. #50
Conversation
From Chromium.
From Chromium.
From Chromium.
SSL 3.0 requires minimal padding. (TLS doesn't require it but accepts it, so apply uniformly to both.) From Chromium.
From Chromium.
Adapted from a Chromium patch.
Client auth now participates in signature algorithms. From Chromium.
From Chromium.
|
In general: tests for those changes? (more comments coming later) |
|
ok, I think that's all BTW, I have created some unit tests for DHE key exchange here: https://github.com/tomato42/tlslite-1/tree/dhe-kex-2 you may want to reuse them. |
Oh, handy. It looks like you have some bits of DHE and ECDHE already? I haven't looked too closely yet. Perhaps it'd be better to pull that one out [edit: remove it from my branch that is, not yours] for the time being until we sort out how to merge them? ECDHE would be handy for Chromium's test server at some point anyway; the motivation for most of these was that we couldn't easily tighten our False Start requirements until the test server was able to simulate them. (I'll try to get to your comments later today or this weekend. Depends how busy I end up with other stuff.) |
The DHE implementation I've made is less complete than what you have provided here, so I don't think pulling mine over yours is worth it. The ECDHE also depends on external libraries so will need a lot of changes to make it worthy of a pull request. That branch is just a rough draft to workout what tests/infrastructure are needed to support PFS. Transplanting the tests should be easier.
well, I'm working on it so that NSS and OpenSSL (among others) have better test suite (in the form of tlsfuzzer), so I'll be working to the bring tlslite to feature parity with them |
|
tlsfuzzer: Neat. If you haven't seen them, we have some fairly extensive tests in BoringSSL based using Go's TLS implementation as a hackable base. It's not quite a fuzzer, but I'm quite happy with what we've been able to do with it. (It helps that it was basically already feature-complete to start with. And Go's concurrency model means it's all straight-line code with nothing funny, which is nice.) (I'd actually like to move Chromium's test server to Go someday so I don't have to maintain two different test-only SSL stacks, but that'll take more work and politics than I have the bandwidth for right now. :-) In the meantime, feature parity in tlslite is quite useful.) |
|
Thanks David, this is good stuff, I agree with most of Hubert's comments. Tests would be great, both client-server and unit tests (Hubert added a framework for that). Without tests, in a weakly-typed language, things will break over time. The DHE seems like it could use some refactoring too. Once you update this I'd like to absorb these changes, and Hubert's encrypt-then-MAC, then do a release soon (week or two) so there's finally some secure ciphersuites. |
|
@davidben well, sqlite has 3 independent test suites, I don't think we're overdoing it with just two test servers ;) |
Here are a bunch of patches to upstream from Chromium. As with previous ones, these fall under the "The Chromium Authors" section of the LICENSE file.
They include: