Fix CMSG_DATA(3) and friends on BSD

[asomers]

PR #1098 added the CMSG_DATA(3) family of functions into libc. Because
they're defined as macros in C, they had to be rewritten as Rust
functions for libc. Also, they can't be tested in CI for the same
reason. But that PR erroneously used the same definitions in BSD as in
Linux.

This commit corrects the definitions for OSX, FreeBSD, DragonflyBSD,
OpenBSD, and NetBSD. I renamed a few variables and collapsed a few
macros in order to combine the definitions where possible.

Fixes #1210

[rust-highfive]

r? @alexcrichton

(rust_highfive has picked a reviewer for you, use r? to override)

[asomers]

I'm not 100% sure about the _ALIGNBYTES definitions for NetBSD on arm and arm64, because NetBSD defines its platforms differently than Rust. And I'm not 100% sure that I got the OSX definitions correct, because I don't have a system to look at. I rely on Google for OSX headers. But using these definitions I did get a good test suite to pass on FreeBSD, NetBSD, OpenBSD, and Linux, all on amd64.

[asomers]

One question. The CMSG_DATA, CMSG_LEN, and CMSG_SPACE functions are always safe to use (the other two are not). But the f! macro marks them as unsafe extern, even though they are neither. Should I bypass the macro to declare them as safe Rust functions? I hesitate because I don't see any precedent for doing that, even for other obviously safe functions like WEXITSTATUS.

[alexcrichton]

@bors: r+

Thanks!

I think we covered the safety in another issue where we're hoping for consistency here, but we'll see how that pays off over time!

[bors]

📌 Commit 8067378 has been approved by alexcrichton

[bors]

⌛ Testing commit 8067378 with merge 70284d7...

[bors]

☀️ Test successful - checks-cirrus, checks-travis, status-appveyor
Approved by: alexcrichton
Pushing 70284d7 to master...