There is a report that iowq executed getsockopt never completes. The reason being that io_uring_cmd_sock() can return a positive result, and io_uring_cmd() propagates it back to core io_uring, instead of IOU_OK. In case of io_wq_submit_work(), the request will be dropped without completing it. The offending code was introduced by a hack in a9c3eda7eada9 ("io_uring: fix submission-failure handling for uring-cmd"), however it was fine until getsockopt was introduced and started returning positive results. The right solution is to always return IOU_OK, since e0b23d9953b0c ("io_uring: optimise ltimeout for inline execution"), we should be able to do it without problems, however for the sake of backporting and minimising side effects, let's keep returning negative return codes and otherwise do IOU_OK. Link: https://github.com/axboe/liburing/issues/1181 Cc: stable@xxxxxxxxxxxxxxx Fixes: 8e9fad0e70b7b ("io_uring: Add io_uring command support for sockets") Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx> Reviewed-by: Breno Leitao <leitao@xxxxxxxxxx> Link: https://lore.kernel.org/r/ff349cf0654018189b6077e85feed935f0f8839e.1721149870.git.asml.silence@xxxxxxxxx Signed-off-by: Jens Axboe <axboe@xxxxxxxxx> diff --git a/io_uring/uring_cmd.c b/io_uring/uring_cmd.c index 21ac5fb2d5f0..a54163a83968 100644 --- a/io_uring/uring_cmd.c +++ b/io_uring/uring_cmd.c @@ -265,7 +265,7 @@ int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags) req_set_fail(req); io_req_uring_cleanup(req, issue_flags); io_req_set_res(req, ret, 0); - return ret; + return ret < 0 ? ret : IOU_OK; } int io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw,