[PATCH] io_uring: fix accounting of dropped CQEs in backlog flush

[PATCH] io_uring: fix accounting of dropped CQEs in backlog flush

[Jens Axboe]

We really should increment the overflow count if we drop events from
the overflow backlog, even if we only do this from the path where
either the task or the ring is going way.

Fixes: c28f37817b7c ("io_uring: add support for backlogged CQ ring")
Reported-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

---

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 321ccc38397c..f2769f5f9662 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -628,6 +628,9 @@ static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
 			WRITE_ONCE(cqe->user_data, req->user_data);
 			WRITE_ONCE(cqe->res, req->result);
 			WRITE_ONCE(cqe->flags, 0);
+		} else {
+			WRITE_ONCE(ctx->rings->cq_overflow,
+				atomic_inc_return(&ctx->cached_cq_overflow));
 		}
 	}
 
-- 
Jens Axboe