[PATCH v5 5/6] nvme-tcp: Support KeyUpdate

[Sagi Grimberg]

On 12/11/2025 6:27, alistair23 at gmail.com wrote:
> From: Alistair Francis <alistair.francis at wdc.com>
>
> If the nvme_tcp_try_send() or nvme_tcp_try_recv() functions return
> EKEYEXPIRED then the underlying TLS keys need to be updated. This occurs
> on an KeyUpdate event as described in RFC8446
> https://datatracker.ietf.org/doc/html/rfc8446#section-4.6.3.
>
> If the NVMe Target (TLS server) initiates a KeyUpdate this patch will
> allow the NVMe layer to process the KeyUpdate request and forward the
> request to userspace. Userspace must then update the key to keep the
> connection alive.
>
> This patch allows us to handle the NVMe target sending a KeyUpdate
> request without aborting the connection. At this time we don't support
> initiating a KeyUpdate.
>
> Signed-off-by: Alistair Francis <alistair.francis at wdc.com>
> ---

I see this is on top of Hannes recvmsg patch. Worth noting in the patch 
here at least.

> v5:
>   - Cleanup code flow
>   - Check for MSG_CTRUNC in the msg_flags return from recvmsg
>     and use that to determine if it's a control message
> v4:
>   - Remove all support for initiating KeyUpdate
>   - Don't call cancel_work() when updating keys
> v3:
>   - Don't cancel existing handshake requests
> v2:
>   - Don't change the state
>   - Use a helper function for KeyUpdates
>   - Continue sending in nvme_tcp_send_all() after a KeyUpdate
>   - Remove command message using recvmsg
>
>   drivers/nvme/host/tcp.c | 85 +++++++++++++++++++++++++++++++++--------
>   1 file changed, 70 insertions(+), 15 deletions(-)
>
> diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
> index 4797a4532b0d..5cec5a974bbf 100644
> --- a/drivers/nvme/host/tcp.c
> +++ b/drivers/nvme/host/tcp.c
> @@ -172,6 +172,7 @@ struct nvme_tcp_queue {
>   	bool			tls_enabled;
>   	u32			rcv_crc;
>   	u32			snd_crc;
> +	key_serial_t		handshake_session_id;
>   	__le32			exp_ddgst;
>   	__le32			recv_ddgst;
>   	struct completion       tls_complete;
> @@ -858,7 +859,10 @@ static void nvme_tcp_handle_c2h_term(struct nvme_tcp_queue *queue,
>   static int nvme_tcp_recvmsg_pdu(struct nvme_tcp_queue *queue)
>   {
>   	char *pdu = queue->pdu;
> +	char cbuf[CMSG_LEN(sizeof(char))] = {};
>   	struct msghdr msg = {
> +		.msg_control = cbuf,
> +		.msg_controllen = sizeof(cbuf),
>   		.msg_flags = MSG_DONTWAIT,
>   	};
>   	struct kvec iov = {
> @@ -873,12 +877,17 @@ static int nvme_tcp_recvmsg_pdu(struct nvme_tcp_queue *queue)
>   	if (ret <= 0)
>   		return ret;
>   
> +	hdr = queue->pdu;
> +	if (hdr->type == TLS_HANDSHAKE_KEYUPDATE) {
> +		dev_err(queue->ctrl->ctrl.device, "KeyUpdate message\n");
> +		return 1;
> +	}
> +
>   	queue->pdu_remaining -= ret;
>   	queue->pdu_offset += ret;
>   	if (queue->pdu_remaining)
>   		return 0;
>   
> -	hdr = queue->pdu;
>   	if (unlikely(hdr->hlen != sizeof(struct nvme_tcp_rsp_pdu))) {
>   		if (!nvme_tcp_recv_pdu_supported(hdr->type))
>   			goto unsupported_pdu;
> @@ -944,6 +953,7 @@ static int nvme_tcp_recvmsg_data(struct nvme_tcp_queue *queue)
>   	struct request *rq =
>   		nvme_cid_to_rq(nvme_tcp_tagset(queue), pdu->command_id);
>   	struct nvme_tcp_request *req = blk_mq_rq_to_pdu(rq);
> +	char cbuf[CMSG_LEN(sizeof(char))] = {};
>   
>   	if (nvme_tcp_recv_state(queue) != NVME_TCP_RECV_DATA)
>   		return 0;
> @@ -976,10 +986,26 @@ static int nvme_tcp_recvmsg_data(struct nvme_tcp_queue *queue)
>   
>   		ret = sock_recvmsg(queue->sock, &msg, msg.msg_flags);
>   		if (ret < 0) {
> -			dev_err(queue->ctrl->ctrl.device,
> -				"queue %d failed to receive request %#x data",
> -				nvme_tcp_queue_id(queue), rq->tag);
> -			return ret;
> +			/* If MSG_CTRUNC is set, it's a control message,
> +			 * so let's read the control message.
> +			 */
> +			if (msg.msg_flags & MSG_CTRUNC) {
> +				memset(&msg, 0, sizeof(msg));
> +				msg.msg_flags = MSG_DONTWAIT;
> +				msg.msg_control = cbuf;
> +				msg.msg_controllen = sizeof(cbuf);
> +
> +				ret = sock_recvmsg(queue->sock, &msg, msg.msg_flags);
> +			}
> +
> +			if (ret < 0) {
> +				dev_dbg(queue->ctrl->ctrl.device,
> +					"queue %d failed to receive request %#x data, %d",
> +					nvme_tcp_queue_id(queue), rq->tag, ret);
> +				return ret;
> +			}
> +
> +			return 0;
>   		}
>   		if (queue->data_digest)
>   			nvme_tcp_ddgst_calc(req, &queue->rcv_crc, ret);
> @@ -1384,15 +1410,39 @@ static int nvme_tcp_try_recvmsg(struct nvme_tcp_queue *queue)
>   		}
>   	} while (result >= 0);
>   
> -	if (result < 0 && result != -EAGAIN) {
> -		dev_err(queue->ctrl->ctrl.device,
> -			"receive failed:  %d\n", result);
> -		queue->rd_enabled = false;
> -		nvme_tcp_error_recovery(&queue->ctrl->ctrl);
> -	} else if (result == -EAGAIN)
> -		result = 0;
> +	if (result < 0) {
> +		if (result != -EKEYEXPIRED && result != -EAGAIN) {
> +			dev_err(queue->ctrl->ctrl.device,
> +				"receive failed:  %d\n", result);
> +			queue->rd_enabled = false;
> +			nvme_tcp_error_recovery(&queue->ctrl->ctrl);
> +		}
> +		return result;
> +	}
> +
> +	queue->nr_cqe = nr_cqe;
> +	return nr_cqe;
> +}
> +
> +static void update_tls_keys(struct nvme_tcp_queue *queue)
> +{
> +	int qid = nvme_tcp_queue_id(queue);
> +	int ret;
> +
> +	dev_dbg(queue->ctrl->ctrl.device,
> +		"updating key for queue %d\n", qid);
>   
> -	return result < 0 ? result : (queue->nr_cqe = nr_cqe);
> +	flush_work(&(queue->ctrl->ctrl).async_event_work);
> +
> +	ret = nvme_tcp_start_tls(&(queue->ctrl->ctrl),
> +				 queue, queue->ctrl->ctrl.tls_pskid,
> +				 HANDSHAKE_KEY_UPDATE_TYPE_RECEIVED);

No need to quiesce the queue or anything like that? everything continues 
as usual?
Why are you flushing async_event_work?