[PATCH 0/3] avoid race between time out and tear down

[Chao Leng]

On 2020/10/21 2:43, Sagi Grimberg wrote:
> 
>> Avoid race between time out and tear down for rdma and tcp.
> 
> This patchset overall looks good, but we still need the patch that
> avoids double completion:
> 
> -- 
> diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
> index 629b025685d1..46428ff0b0fc 100644
> --- a/drivers/nvme/host/tcp.c
> +++ b/drivers/nvme/host/tcp.c
> @@ -2175,7 +2175,7 @@ static void nvme_tcp_complete_timed_out(struct request *rq)
>          /* fence other contexts that may complete the command */
>          mutex_lock(&to_tcp_ctrl(ctrl)->teardown_lock);
>          nvme_tcp_stop_queue(ctrl, nvme_tcp_queue_id(req->queue));
> -       if (!blk_mq_request_completed(rq)) {
> +       if (blk_mq_request_started(rq) && !blk_mq_request_completed(rq)) {
Yes, this patch is need. and samely for nvme_cancel_request.
This will fix the race with asynchronous completion.

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index e85f6304efd7..1e838d952096 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -338,7 +338,7 @@ bool nvme_cancel_request(struct request *req, void *data, bool reserved)
                                 "Cancelling I/O %d", req->tag);

         /* don't abort one completed request */
-       if (blk_mq_request_completed(req))
+       if (blk_mq_request_completed(req) || !blk_mq_request_started(rq))
                 return true;

         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
-- 
>                  nvme_req(rq)->status = NVME_SC_HOST_ABORTED_CMD;
>                  blk_mq_complete_request_sync(rq);
>          }
> -- 
> .