io_uring: allow multishot recv CQEs to overflow
authorDylan Yudaken <dylany@meta.com>
Mon, 7 Nov 2022 12:52:35 +0000 (04:52 -0800)
committerJens Axboe <axboe@kernel.dk>
Mon, 7 Nov 2022 20:17:17 +0000 (13:17 -0700)
With commit aa1df3a360a0 ("io_uring: fix CQE reordering"), there are
stronger guarantees for overflow ordering. Specifically ensuring that
userspace will not receive out of order receive CQEs. Therefore this is
not needed any more for recv/recvmsg.

Signed-off-by: Dylan Yudaken <dylany@meta.com>
Link: https://lore.kernel.org/r/20221107125236.260132-4-dylany@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_uring/net.c

index 0d77ddcce0af3f7172a04d555dd4bed4f880dd2a..4b79b61f5597073897faf0672de3bad76f1df79a 100644 (file)
@@ -603,15 +603,11 @@ static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
 
        if (!mshot_finished) {
                if (io_post_aux_cqe(req->ctx, req->cqe.user_data, *ret,
-                                   cflags | IORING_CQE_F_MORE, false)) {
+                                   cflags | IORING_CQE_F_MORE, true)) {
                        io_recv_prep_retry(req);
                        return false;
                }
-               /*
-                * Otherwise stop multishot but use the current result.
-                * Probably will end up going into overflow, but this means
-                * we cannot trust the ordering anymore
-                */
+               /* Otherwise stop multishot but use the current result. */
        }
 
        io_req_set_res(req, *ret, cflags);