io_uring/rw: fix downgraded mshot read

The io-wq path can downgrade a multishot request to oneshot mode,
however io_read_mshot() doesn't handle that and would still post
multiple CQEs. That's not allowed, because io_req_post_cqe() requires
stricter context requirements.

The described can only happen with pollable files that don't support
FMODE_NOWAIT, which is an odd combination, so if even allowed it should
be fairly rare.

Cc: stable@vger.kernel.org
Reported-by: chase xd <sl1589472800@gmail.com>
Fixes: bee1d5becdf5b ("io_uring: disable io-wq execution of multishot NOWAIT requests")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/c5c8c4a50a882fd581257b81bf52eee260ac29fd.1735407848.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/io_uring/rw.c b/io_uring/rw.c
index 0bcb83e4ce3cf9b0d02da3f6cd504a377931ccde..29bb3010f9c06d67066cdc74ab818b6b039c9b7a 100644 (file)
--- a/io_uring/rw.c
+++ b/io_uring/rw.c
@@ -983,6 +983,8 @@ int io_read_mshot(struct io_kiocb *req, unsigned int issue_flags)
                io_kbuf_recycle(req, issue_flags);
                if (ret < 0)
                        req_set_fail(req);
+       } else if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
+               cflags = io_put_kbuf(req, ret, issue_flags);
        } else {
                /*
                 * Any successful return value will keep the multishot read