Add socket_peerpidfd

[ids1024]

Supported on Linux since 6.5.

I see socket_peercred is only provided on Linux, while OpenBSD (https://man.openbsd.org/getsockopt.2) has some version of SO_PEERCRED, and FreeBSD (https://man.freebsd.org/cgi/man.cgi?unix(4)) has a LOCAL_PEERCRED (libwayland for instance has code to use those on their respective platforms). These APIs do seem to be slightly different though. In any case, SO_PEERPIDFD is definitely Linux-specific for the time being.

[ids1024]

pidfd_getpid would also be good to add. Apparently that's not a system call, so for linux-raw we'd need to parse /proc/self/fdinfo/$pidfd for the Pid: field. But that should be fine, and not too hard.

[sunfishcode]

This looks good to me, with one quick question:

[sunfishcode]

I'm curious why these lines were needed in the tests, that weren't otherwise changed.

[ids1024]

That fixes a compile error with cargo test --features net,process,event. Though it also works to use --features=all-apis like CI uses. (I suppose ideally it shouldn't error for any combination of features; but doesn't matter that much for tests.)

[sunfishcode]

Those imports are unfortunately evoking warnings, which are currently configured to be errors in CI. Could you add conditionals to the imports so that they're not imported if they're not needed?

[ids1024]

Huh, looking at CI, I guess libc doesn't define SO_PEERPIDFD in https://github.com/rust-lang/libc/blob/main/src/unix/linux_like/android/mod.rs.

Not sure why those are defined separately for Android; I'd assume unless the C standard library is doing something weird, those particular definitions should be the same for anything with a Linux kernel.

[zeenix]

@ids1024 Thanks so much for doing this. Do you intend to finish this PR?

[ids1024]

I guess to pass CI this could just be disabled on Android. Presumably libc should be updated to have things like SO_PEERPIDFD. Changed that for now.

Tests seem to be broken with a recent compile, so I rebased this on #1523. Looks like there are a couple other unrelated CI errors? But I think this should be mergeable.

[ids1024]

Or well, one of the errors was related to this change. Fixed that, and added a commit for another CI error. So now everything is passing.

[sunfishcode]

I haven't investigated this yet myself, but CI is failing on a missing SO_PEERPIDFD on nightly.

[ids1024]

I guess the last time I rebased this I did it on a different computer without pulling the latest version of the branch; so it lacked a fix to not build enable the feature on Android.

Presumably Android should support this too, but that will require a PR to libc to add the necessary constants.

This should be mergeable now.