Re: [PATCH] rdma: Fix drbd_transport_rdma module reference count exception

Applied as
https://github.com/LINBIT/drbd/commit/a3807781a29ff488346eed47bd457bbd2cc78c91

Thanks!

Best regards,
 Philipp

On Wed, Feb 19, 2025 at 4:13 AM zhengbing.huang <
[email protected]> wrote:

> In testing, we find drbd_transport_rdma module reference count is abnormal:
> drbd_transport_rdma 262144 28293
>
> we don't have that many drbd devices.
>
> If the XXX_ADDR_ERROR/XXX_ROUTE_ERROR events occurs
> and the DSB_CONNECTING flag bit is not set,
> the dtr_cma_event_handler() returns 0 directly.
> The cm structure cannot be destroyed,
> and the drbd_transport_rdma module reference count is abnormal.
>
> So, for XXX_ADDR_ERROR/XXX_ROUTE_ERROR events,
> we do not need to judge the DSB_CONNECTING flag,
> and we need to kref_put of cm structure.
>
> Signed-off-by: zhengbing.huang <[email protected]>
> ---
>  drbd/drbd_transport_rdma.c | 5 +++++
>  1 file changed, 5 insertions(+)
>
> diff --git a/drbd/drbd_transport_rdma.c b/drbd/drbd_transport_rdma.c
> index ba4f1baa7..bb59e6501 100644
> --- a/drbd/drbd_transport_rdma.c
> +++ b/drbd/drbd_transport_rdma.c
> @@ -1292,6 +1292,11 @@ static int dtr_cma_event_handler(struct rdma_cm_id
> *cm_id, struct rdma_cm_event
>                 // pr_info("%s: RDMA_CM_EVENT_ADDR_ERROR\n", cm->name);
>         case RDMA_CM_EVENT_ROUTE_ERROR:
>                 // pr_info("%s: RDMA_CM_EVENT_ROUTE_ERROR\n", cm->name);
> +               set_bit(DSB_ERROR, &cm->state);
> +
> +               dtr_cma_retry_connect(cm->path, cm);
> +               break;
> +
>         case RDMA_CM_EVENT_CONNECT_ERROR:
>                 // pr_info("%s: RDMA_CM_EVENT_CONNECT_ERROR\n", cm->name);
>         case RDMA_CM_EVENT_UNREACHABLE:
> --
> 2.43.0
>
>