Add support of TCP_CONGESTION for setsockopt

[Fensteer]

Implementation proposal for support of TCP_CONGESTION param for setsockoptand getsockopt with the CString type.

[Fensteer]

I use this code to test the feature but I don't know how to integrate into the CI (I don't know if reno is in tcp_available_congestion_control)

extern crate nix;

use nix::sys::socket::sockopt;
use nix::sys::socket::{getsockopt, setsockopt, socket, AddressFamily, SockFlag, SockType};
use std::ffi::CString;

fn main() {
    let socket = socket(
        AddressFamily::Inet,
        SockType::Stream,
        SockFlag::empty(),
        None,
    ).unwrap();

    let algo = CString::new("reno").expect("Cannot create CString");

    setsockopt(socket, sockopt::TcpCongestion, &algo).expect("setsockopt");

    assert_eq!(algo, CString::new("reno").unwrap());

    assert_eq!(
        getsockopt(socket, sockopt::TcpCongestion).unwrap(),
        CString::new("reno").unwrap()
    );
}

[Fensteer]

I don't understand why the buildbot is failing because libc::TCP_CONGESTION exists in latest version of libc : https://docs.rs/libc/0.2.44/libc/constant.TCP_CONGESTION.html

[asomers]

Looks like libc has only defined TCP_CONGESTION for Linux and uclibc. You'll have to submit a PR to libc to add the definition for FreeBSD. You should also check if it's defined on any other OS.

[Fensteer]

Done : rust-lang/libc#1151

[Fensteer]

rust-lang/libc#1151 was merged in libc.
What do you think of this version?

[asomers]

Ok, this PR is starting to look pretty good. I think there are only two issues left:

1. Could you add a test? On FreeBSD, I believe getting the option should work fine. Setting it is a little harder to test, because there's only one valid option unless you load a kernel module. But you should be able to set it back to the same value as the default. I'm not sure what will happen on Linux; you'll have to try to find out.
2. You'll need to rebase and fix the changelog, since Nix just released 0.12.0.

[Fensteer]

CI fails but actually the code is OK. I tried CI on a full emulated MIPS or PowerPC system and the test run well. I think that the CI fails because only the binary is run on a emulated processor but the kernel is still the same than the host.

[asomers]

CI fails but actually the code is OK. I tried CI on a full emulated MIPS or PowerPC system and the test run well. I think that the CI fails because only the binary is run on a emulated processor but the kernel is still the same than the host.

QEMU does emulate the full system. It's not just a binary translator. However, it's got a lot of bugs. What did you use for a "full emulated MIPS or PowerPC system"?

[Fensteer]

Rebase done and I squashed some commits.
CHANGELOG fixed.

[asomers]

Could you please squash all commits together?

[asomers]

Could you please squash all commits together?

[asomers]

Could you please squash all commits together?

[asomers]

Could you please squash all commits together?

[Fensteer]

Done

[asomers]

Sorry for the comment dupes; github seems to have a partial outage today.

bors r+

[bors]

Build succeeded

• continuous-integration/travis-ci/push
• buildbot/nix-rust/nix i386_fbsd11
• buildbot/nix-rust/nix amd64_fbsd11

[Fensteer]

Thanks for the reviews and advices !