[PATCH] secur32: Split push/pull schannel transports

[Anton Romanov]

Jacek,


> The code looks generally good and it matches what MSDN states about
> thread safety of those functions. Still, having little trust to MSDN, I
> wonder if it would make sense to make them more generally thread safe.
> We could, for example, store critical section inside schan_context and
> lock it in all functions accessing it. Did you consider that? It would
> make all those functions thread safe (with little overhead). It just
> seems more straightforward solution to me.

Well, I'd say that would just introduce additional complexity without
much/any gain. I mean - if MSDN does not state other use is supposed
to be safe - why would we try and make it?
Unless there is an app that would actually benefit that.

> As for the approach you took, it may be correct. According to [1] that's
> also how gnutls works. Do you know if Secure Transport on Mac is
> similar? I couldn't find any documentation about it.
>
> Thanks,
> Jacek
>
> [1] https://gnutls.org/manual/html_node/Thread-safety.html

Unfortunately I am not entirely sure if SecureTransport is (supposed
to be) thread-safe/unsafe either... Yeah, documentation does not
really clarify that. Source code [2] is not too clear either.
However, libcurl (which I would trust) states that it is using Secure
Transport in an "entirely thread safe way" [3], and I don't see it
making any locks when doing sslread/write [4].

[2] https://opensource.apple.com/source/Security/Security-55179.13/libsecurity_ssl/
[3] https://curl.haxx.se/libcurl/c/threadsafe.html
[4] https://github.com/curl/curl/blob/master/lib/vtls/darwinssl.c