Restore some non-POSIX functions for AIX

[xingxue-ibm]

Description

These non-POSIX functions were previously removed because they are guarded by the _KERNEL macro in AIX headers. After discussion, we thought they might be useful, so this patch restores them with fixes.

Sources

Checklist

• Relevant tests in libc-test/semver have been updated
• No placeholder or unstable values like *LAST or *MAX are
  included (see #3131)
• Tested locally (cd libc-test && cargo test --target mytarget);
  especially relevant for platforms that may not be checked in CI

[rustbot]

Unknown labels: O-unix, S-waiting-on-review

[tgross35]

Per https://www.ibm.com/docs/en/aix/7.2.0?topic=s-sctp-opt-info-subroutine, opt is sctp_assoc_t. I assume this is defined to unsigned, but is it worth a typedef?

[xingxue-ibm]

Thanks for checking. Yeah, there seems to be some AIX header weirdness. The AIX man page specifies the type sctp_assoc_t for the argument id, but the prototype in the header files has type uint32_t for it.
Man page:

       int sctp_opt_info(sd, id, opt, *arg_size, *size);
       int sd;
       sctp_assoc_t id;
       int opt;
       void *arg_size;
       size_t *size;

net/proto_uipc.h:

int             sctp_opt_info PROTO((int, uint32_t, int, void *, size_t *));

I agree it makes it consistent with the man page if we define its type as sctp_assoc_t on the Rust side. The typedef of sctp_assoc_t is defined in header netinet/sctp.h so I am pulling the header in. sctp_peeloff and sctp_opt_info are also declared in netinet/sctp.h without the _KERNEL guard so I am removing them from the function skip list in build.rs.

typedef uint32_t sctp_assoc_t;
int sctp_peeloff(int s, uint32_t * assoc_id);
int sctp_opt_info(int s, uint32_t id, int opt, void *arg, size_t *size);
int sctp_getladdrs(int sd, sctp_assoc_t id, struct sockaddr **addrs);
int sctp_getpaddrs(int sd, sctp_assoc_t id, struct sockaddr **addrs);

[tgross35]

It looks like this only has three fields at https://www.ibm.com/docs/en/aix/7.3.0?topic=s-sendmmsg-subroutine, is that inaccurate?

[xingxue-ibm]

The online doc does not seem to be accurate. The prototype in AIX header net/proto_uipc.h is as follows. I will open an issue against the AIX doc.

int             sendmmsg PROTO((int, struct mmsghdr *, unsigned int, int));

[tgross35]

Just checking, are these somewhat common to use on AIX even though they aren't defined? mount on other platforms usually has a void *data arg of fs-specific flags, I'm wondering if there's a chance the AIX version changes to take one as well.

sendmmsg/recvmmsg are pretty common, I'm surprised nobody has moved your #ifdef __KERNEL a few lines lower yet 🙂

[xingxue-ibm]

Thanks again for checking. I checked the documentation for the latest AIX release (v7.3), and mount still takes three arguments. By definition, _KERNEL is meant to guard code used for writing kernel extensions, but it seems some of these functions are also used in OSS. Yes, sendmmsg and recvmmsg should be moved out of the block. I will open an issue against the AIX headers for these functions as well as mount.

[tgross35]

Thanks for the PR and the detailed followup, everything sounds reasonable to me.

[xingxue-ibm]

Thank you for checking and pointing out the discrepancies in the man pages.