io_uring/msg-ring: ensure flags passing works for task_work completions

If the target ring is using IORING_SETUP_SINGLE_ISSUER and we're posting
a message from a different thread, then we need to ensure that the
fallback task_work that posts the CQE knwos about the flags passing as
well. If not we'll always be posting 0 as the flags.

Fixes: 3563d7ed58a5 ("io_uring/msg_ring: Pass custom flags to the cqe")
Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/io_uring/msg_ring.c b/io_uring/msg_ring.c
index 27992bd..8803c09 100644 (file)
--- a/io_uring/msg_ring.c
+++ b/io_uring/msg_ring.c
@@ -99,6 +99,11 @@ static void io_msg_tw_complete(struct callback_head *head)
        if (current->flags & PF_EXITING) {
                ret = -EOWNERDEAD;
        } else {
+               u32 flags = 0;
+
+               if (msg->flags & IORING_MSG_RING_FLAGS_PASS)
+                       flags = msg->cqe_flags;
+
                /*
                 * If the target ring is using IOPOLL mode, then we need to be
                 * holding the uring_lock for posting completions. Other ring
@@ -107,7 +112,7 @@ static void io_msg_tw_complete(struct callback_head *head)
                 */
                if (target_ctx->flags & IORING_SETUP_IOPOLL)
                        mutex_lock(&target_ctx->uring_lock);
-               if (!io_post_aux_cqe(target_ctx, msg->user_data, msg->len, 0))
+               if (!io_post_aux_cqe(target_ctx, msg->user_data, msg->len, flags))
                        ret = -EOVERFLOW;
                if (target_ctx->flags & IORING_SETUP_IOPOLL)
                        mutex_unlock(&target_ctx->uring_lock);