tree cbd387754a123a6400b088fefb2234cda005fcfa
parent 04aa88b0459760b974aa09e59ba11ea6ecfab770
author Jens Axboe <axboe@kernel.dk> 1754163295 -0600
committer Jens Axboe <axboe@kernel.dk> 1754163295 -0600

test/poll-race-mshot: update for kernels supporting msg_inq FOR UNIX sock

The upstream kernel commit:

ommit df30285b3670bf52e1e5512e4d4482bec5e93c16
Author: Kuniyuki Iwashima <kuniyu@google.com>
Date:   Wed Jul 2 22:35:18 2025 +0000

    af_unix: Introduce SO_INQ.

added support for msg_inq querying for unix sockets. io_uring sets
->msg_get_inq by default, as it can use this information to do retries
more efficiently for bundles and multishot. However, this test case
assumed that we'd always get an empty posting without CQE_F_MORE set
for multishot. And that isn't true after unix sockets got support for
returning data left in the socket. If the socket type passes back that
there's zero bytes left in the socket, then io_uring won't retry as it
knows there's nothing there. Hence no final CQE gets posted terminating
the multishot, and the test case will wait forever.

Just fix this to break when NREQ multishot completions have been seen.

Signed-off-by: Jens Axboe <axboe@kernel.dk>