From sys/ioctl.h

Pull sockaddr_nl up a layer as it is present for Android

The addition of sockaddr_nl seems to be a fairly recent
addition to bionic, although it would appear that its lack
of presence in previous versions is more of an oversight than
anything else.

https://github.com/android/platform_bionic/blob/831c8a52498d3c7be204a00847275f6e9163626f/libc/kernel/uapi/linux/netlink.h#L54

The addition of sockaddr_nl seems to be a fairly recent
addition to bionic, although it would appear that its lack
of presence in previous versions is more of an oversight than
anything else.

https://github.com/android/platform_bionic/blob/831c8a52498d3c7be204a00847275f6e9163626f/libc/kernel/uapi/linux/netlink.h#L54



This change is motivated by failures to build nix under Android.

Signed-off-by: Paul Osborne <osbpau@gmail.com>

Make cfg_if uses more explicit and consistent

This commit changes most uses of cfg_if as follows:
- fallthrough `else` usage is avoided for architecture or OS specific
  items
- a comment is added in the final `else` clause to signal intent someone
  modifying

It is safer to omit items than include ones for the wrong platform or
architecture.

This commit changes most uses of cfg_if as follows:
- fallthrough `else` usage is avoided for architecture or OS specific
  items
- a comment is added in the final `else` clause to signal intent someone
  modifying

It is safer to omit items than include ones for the wrong platform or
architecture.

Fix DragonFly: clock_gettime() expects CLOCK_* constants as ulong.

While there add all CLOCK_ constants as found on DragonFly.

Run CI on all branches

Only running them against auto-libc unfortunately means that PRs don't run CI.

Only running them against auto-libc unfortunately means that PRs don't run CI.

Prep repo to be bors-compatible

In... theory this is what we need? Let's try it out.

While there add all CLOCK_ constants as found on DragonFly.

linux: Add SYS_perf_event_open syscall numbers

The associated `perf_event_attr` struct is not included as it contains
unions and a bitfield making it unrepresentable in Rust. Additionally,
it is not entirely stable, and has changed size several times.

apple: Add openpty(3) and forkpty(3)

Bump to 0.2.8

Add MAP_STACK and MAP_HUGETLB entries to MIPS so that the nix crate can be built

Merging more PRs in the queue

When building with musl for an arm target link with a shared libc (like mips
already does).

Merge a few PRs together

Conflicts:
	src/unix/notbsd/linux/musl/b32/arm.rs
	src/unix/notbsd/linux/musl/b32/asmjs.rs
	src/unix/notbsd/linux/musl/b32/x86.rs
	src/unix/notbsd/linux/musl/b64/mod.rs
	src/unix/notbsd/linux/other/b32/x86.rs
	src/unix/notbsd/linux/other/b64/aarch64.rs
	src/unix/notbsd/linux/other/b64/powerpc64.rs
	src/unix/notbsd/linux/other/b64/x86_64.rs

The function is defined in POSIX [0], but according to Gnulib docs [1],
it is missing or not declared on at least some versions of OS X and
FreeBSD. On Solaris, it is not a system call but is present as
fdatasync(3) [2].

[0] http://www.opengroup.org/onlinepubs/9699919799/functions/fdatasync.html
[1] https://www.gnu.org/software/gnulib/manual/html_node/fdatasync.html
[2] http://docs.oracle.com/cd/E36784_01/html/E36874/fdatasync-3c.html

Add SYS_gettid

This is how MetadaExt expects these fields to be named.

See https://github.com/rust-lang/rust/blob/c116ae35cf49b55bd8d82e31f1ba030cf7e63867/src/libstd/os/linux/fs.rs#L107-L121

Use nightlies for iOS and rumprun

iOS was fixed awhile back and we should have rumprun nightlies now, so let's use
them!