FreeBSD RESOLVE_BENEATH support and more

[valpackett]

This is everything required to implement support for FreeBSD's (O|AT)_RESOLVE_BENEATH based lookup sandboxing in cap-std :) While here, also expose more Linux-compatible stuff that FreeBSD's libc exposes, also improve kqueue a bit.

The chmodat signature change is an API break :/ but the omission of flags feels like a mistake to me and I think it's better to fix it earlier rather than later. What do you think?

Waiting for: a libc release with rust-lang/libc#3114 rust-lang/libc#3122 rust-lang/libc#3124 wheeee
Includes: #540

[sunfishcode]

Cc @notgull in case you're interested in the kqueue changes.

For chmodat, it feels like it might be error-prone to have the flag on Linux where it isn't supported. What would you think about adding a chmodat_with function with a flags argument, that's only available on platforms that support flags?

[valpackett]

Hm, that can be done for not breaking backwards compatibility I guess.
But AtFlags is already not something that guarantees that every flag will be used. A lot of the AtFlags are function specific, for example AT_REMOVEDIR is for unlinkat and would do nothing for, say, fstatat. I guess the difference here is the kernel would return EINVAL for unsupported flags, but the ignoring implementation won't—but we can just make it do that.

[sunfishcode]

Good point, having it fail with EINVAL sounds good. Or maybe ENOTSUP would be better?

The other concern here is that rustix recently had a semver bump with 0.37, and we hopefully won't need another for quite a while. Would you be ok introducing a different name for now, with a plan to change it once we start thinking about the next semver bump?

[valpackett]

Nah, EINVAL is the correct one for unsupported arguments, that's what all the libcs return in that case.

Fun discovery btw: musl implements AT_SYMLINK_NOFOLLOW there even though glibc doesn't.

Changed to chmodat_with, hopefully the doc stuff looks okay…

[sunfishcode]

This looks good! I made a note about the naming of chmodat_with in #578.

[valpackett]

Yaay :3 I'll have a little bit more coming today (adding more procctl's right now), hopefully it'd be possible to get that into the next minor release