Fix aio system call wrappers in librt.

[jhb]

Summary

Fix aio system call wrappers in librt.

• Update aio_return/waitcomplete wrappers for ssize_t.
• Fix aio_return() wrapper to fail with EINVAL on a pending job.

Test Plan

• I haven't written test cases for these, though you almost want to have a version of 'aio_test' that links against librt and a version that doesn't?

[kib]

I am curious about the aio_return() change. My reading is that EINPROGRESS is acceptable outcome for not yet completed iocb.

POSIX says "If the asynchronous I/O operation has not yet completed, the results of aio_return( ) are undefined."

And EINVAL is defined as "The aiocbp argument does not refer to an asynchronous operation whose return status has not yet been retrieved."

I do not object against the change, but wonder for the reasoning behind it.

lib/librt/aio.c

143 ↗

(On Diff #18162)

You might fix the style while changing the line (no initialization in declaration).

[jhb]

In D7120#148480, @kib wrote:

I am curious about the aio_return() change. My reading is that EINPROGRESS is acceptable outcome for not yet completed iocb.

POSIX says "If the asynchronous I/O operation has not yet completed, the results of aio_return( ) are undefined."

And EINVAL is defined as "The aiocbp argument does not refer to an asynchronous operation whose return status has not yet been retrieved."

I do not object against the change, but wonder for the reasoning behind it.

It matches what the kernel does for aio_return(), so it is consistent with using aio_return() if you don't link against librt.

[kib]

In D7120#148639, @jhb wrote:

It matches what the kernel does for aio_return(), so it is consistent with using aio_return() if you don't link against librt.

This sounds as a good one-line comment addition.