Test suite about af_ktls

[nmav]

I'm bikeshedding here but Is there any advantage over cmocka? The latter does not require any C++ and is already used for unit testing the dtls sliding window.
https://cmocka.org/

[fridex]

Awesome, Lance! Thank you.

While I was implementing and doing benchmarks of AF_KTLS, I used [1]. It is not the best possible solution, but it was/is useful (note some errors are not handled now - e.g. when a client does not connect to a TCP server, ...). I would like to reorganize this git repo and make the repo easier for development/testing/benchmarks, as suggested by @nmav in https://github.com/fridex/af_ktls/issues/17 Could we introduce one tool for automated tests and benchmarks not to diverge?

I don't have time to make a bigger review now, but few notes:

• I see that you use OpenSSL, that's great, since I use GnuTLS in [1]. We can cross-test the implementation to be sure that AF_KTLS works with OpenSSL and GnuTLS as well.
• I see that you copied parts of AF_KLTS. I would like to avoid it.
• Since we are trying to make TLS/DTLS faster in AF_KTLS in some scenarios, I would like to have benchmarks evaluated automatically during the implementation to see if we are actually making our solution faster.
• Each time there will be a bug/enhancement, we can introduce a test case in one tool to cover examples/exceptions/...

Take a look at [1] to save some time and efforts. Some parts could be reused.

[1] https://github.com/fridex/af_ktls-tool

[nmav]

Apart from the comment above on the framework that can certainly be the foundation of unit testing af_ktls (see Frido's comments as well), but good work. Something that caught my eye:
On gen_random() Is it intentional to test only with alphanumeric data? If not you could use RAND_bytes() and avoid that function.

About //TODO: Not checking file content?... A wild idea could be instead of checking the file content to transmit the hash of the sent data (or send oob), so that the verifier can check the contents even of simple sends with random data.

[lancerchao]

The server basically echos the data from the client. Each client sends one (or more) strings and makes sure that the same string(s) was sent back. That way we can make sure the data is sent/transmitted correctly. It also helps that the reply is human-readable.

I think it is perfectly fine for benchmarking and correctness tests to be separate; that way we can tackle the two independently. I have taken a look at your tool, and I think we can work on combining the two.

I found C++ to be preferable in testing purposes because of the abundance of libraries and tools available, which was my main motivation to using Google Test.

Ignore the TODO. I fixed that by having the server reply.

[nmav]

I found C++ to be preferable in testing purposes because of the abundance of libraries and tools available, which was my main motivation to using Google Test.

Could you be more specific on that?

[lancerchao]

Could you be more specific on that?

For example, the boost libraries?
Also the fact that C code can be smoothly embedded in C++, so we are not limiting ourselves in any way from choosing C++, apart from the dependency on a C++ compiler