[PATCH v3] nvme-tcp: Do not reset transport on data digest errors

[Sagi Grimberg]

On 8/26/21 1:21 AM, Daniel Wagner wrote:
> The spec says
> 
>    7.4.6.1 Digest Error handling
> 
>    When a host detects a data digest error in a C2HData PDU, that host
>    shall continue processing C2HData PDUs associated with the command and
>    when the command processing has completed, if a successful status was
>    returned by the controller, the host shall fail the command with a
>    non-fatal transport error.
> 
> Currently the transport is reseted when a data digest error is
> detected. Instead, when a digest error is detected, mark the final
> status as NVME_SC_DATA_XFER_ERROR and let the upper layer handle
> the error.
> 
> In order to keep track of the final result maintain a status field in
> nvme_tcp_request object and use it to overwrite the completion queue
> status (which might be successful even though a digest error has been
> detected) when completing the request.
> 
> Reviewed-by: Hannes Reinecke <hare at suse.de>
> Signed-off-by: Daniel Wagner <dwagner at suse.de>
> ---
> The status member placed so that it fills up a hole in struct
> nvme_tcp_request:
> 
> struct nvme_tcp_request {
>          struct nvme_request        req;                  /*     0    32 */
>          void *                     pdu;                  /*    32     8 */
>          struct nvme_tcp_queue *    queue;                /*    40     8 */
>          u32                        data_len;             /*    48     4 */
>          u32                        pdu_len;              /*    52     4 */
>          u32                        pdu_sent;             /*    56     4 */
>          u16                        ttag;                 /*    60     2 */
>          u16                        status;               /*    62     2 */
>          /* --- cacheline 1 boundary (64 bytes) --- */
>          struct list_head           entry;                /*    64    16 */
>          struct llist_node          lentry;               /*    80     8 */
>          __le32                     ddgst;                /*    88     4 */
> 
>          /* XXX 4 bytes hole, try to pack */
> 
>          struct bio *               curr_bio;             /*    96     8 */
>          struct iov_iter            iter;                 /*   104    40 */
>          /* --- cacheline 2 boundary (128 bytes) was 16 bytes ago --- */
>          size_t                     offset;               /*   144     8 */
>          size_t                     data_sent;            /*   152     8 */
>          enum nvme_tcp_send_state   state;                /*   160     4 */
> 
>          /* size: 168, cachelines: 3, members: 16 */
>          /* sum members: 160, holes: 1, sum holes: 4 */
>          /* padding: 4 */
>          /* last cacheline: 40 bytes */
> };
> 
> v3:
>   - initialize req->status in nvme_tcp_setup_cmd_pdu()
>   - add rb tag from Hannes
> 
> v2:
>   - https://lore.kernel.org/linux-nvme/20210825124259.28707-1-dwagner@suse.de/
>   - moved 'status' from nvme_tcp_queue to nvme_tcp_request.
> 
> v1:
>   - https://lore.kernel.org/linux-nvme/20210805121541.77613-1-dwagner@suse.de/
> 
> drivers/nvme/host/tcp.c | 23 +++++++++++++++++++----
>   1 file changed, 19 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
> index 645025620154..29ef0f74f620 100644
> --- a/drivers/nvme/host/tcp.c
> +++ b/drivers/nvme/host/tcp.c
> @@ -45,6 +45,7 @@ struct nvme_tcp_request {
>   	u32			pdu_len;
>   	u32			pdu_sent;
>   	u16			ttag;
> +	u16			status;
>   	struct list_head	entry;
>   	struct llist_node	lentry;
>   	__le32			ddgst;
> @@ -485,7 +486,9 @@ static void nvme_tcp_error_recovery(struct nvme_ctrl *ctrl)
>   static int nvme_tcp_process_nvme_cqe(struct nvme_tcp_queue *queue,
>   		struct nvme_completion *cqe)
>   {
> +	struct nvme_tcp_request *req;
>   	struct request *rq;
> +	u16 status;
>   
>   	rq = nvme_find_rq(nvme_tcp_tagset(queue), cqe->command_id);
>   	if (!rq) {
> @@ -496,7 +499,12 @@ static int nvme_tcp_process_nvme_cqe(struct nvme_tcp_queue *queue,
>   		return -EINVAL;
>   	}
>   
> -	if (!nvme_try_complete_req(rq, cqe->status, cqe->result))
> +	req = blk_mq_rq_to_pdu(rq);
> +	status = req->status;
> +	if (status == NVME_SC_SUCCESS)
> +		status = cqe->status;

Maybe more intuitive to skip the local status variable?
	/*  */
	if (req->status == NVME_SC_SUCCESS)
		req->status = cqe->status;

This way it is always consistent completing with req->status.