Re: [PATCH v2 7/7] nvmet-tcp: Support KeyUpdate

["Maurizio Lombardi" <mlombard@bsdbackstore.eu>]

On Fri Sep 5, 2025 at 4:46 AM CEST, alistair23 wrote:
> From: Alistair Francis <alistair.francis@wdc.com>
>
> If the nvmet_tcp_try_recv() function return EKEYEXPIRED or if we receive
> a KeyUpdate handshake type then the underlying TLS keys need to be
> updated.
>
> If the NVMe Host (TLS client) 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 host sending a KeyUpdate
> request without aborting the connection. At this time we don't support
> initiating a KeyUpdate.
>
> Link: https://datatracker.ietf.org/doc/html/rfc8446#section-4.6.3
> Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
> ---
> v2:
>  - Use a helper function for KeyUpdates
>  - Ensure keep alive timer is stopped
>  - Wait for TLS KeyUpdate to complete
>
>  drivers/nvme/target/tcp.c | 90 ++++++++++++++++++++++++++++++++++++---
>  1 file changed, 84 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/nvme/target/tcp.c b/drivers/nvme/target/tcp.c
> index bee0355195f5..dd09940e9635 100644
> --- a/drivers/nvme/target/tcp.c
> +++ b/drivers/nvme/target/tcp.c
> @@ -175,6 +175,7 @@ struct nvmet_tcp_queue {
>  
>  	/* TLS state */
>  	key_serial_t		tls_pskid;
> +	key_serial_t		user_session_id;
>  	struct delayed_work	tls_handshake_tmo_work;
>  
>  	unsigned long           poll_end;
> @@ -186,6 +187,8 @@ struct nvmet_tcp_queue {
>  	struct sockaddr_storage	sockaddr_peer;
>  	struct work_struct	release_work;
>  
> +	struct completion       tls_complete;
> +
>  	int			idx;
>  	struct list_head	queue_list;
>  
> @@ -836,6 +839,11 @@ static int nvmet_tcp_try_send_one(struct nvmet_tcp_queue *queue,
>  	return 1;
>  }
>  
> +#ifdef CONFIG_NVME_TARGET_TCP_TLS
> +static int nvmet_tcp_try_peek_pdu(struct nvmet_tcp_queue *queue);
> +static void nvmet_tcp_tls_handshake_timeout(struct work_struct *w);
> +#endif
> +
>  static int nvmet_tcp_try_send(struct nvmet_tcp_queue *queue,
>  		int budget, int *sends)
>  {
> @@ -844,6 +852,13 @@ static int nvmet_tcp_try_send(struct nvmet_tcp_queue *queue,
>  	for (i = 0; i < budget; i++) {
>  		ret = nvmet_tcp_try_send_one(queue, i == budget - 1);
>  		if (unlikely(ret < 0)) {
> +#ifdef CONFIG_NVME_TARGET_TCP_TLS
> +			if (ret == -EKEYEXPIRED &&
> +				queue->state != NVMET_TCP_Q_DISCONNECTING &&
> +				queue->state != NVMET_TCP_Q_TLS_HANDSHAKE) {
> +					goto done;
> +			}
> +#endif
>  			nvmet_tcp_socket_error(queue, ret);
>  			goto done;
>  		} else if (ret == 0) {
> @@ -1110,11 +1125,52 @@ static inline bool nvmet_tcp_pdu_valid(u8 type)
>  	return false;
>  }
>  
> +#ifdef CONFIG_NVME_TARGET_TCP_TLS
> +static int update_tls_keys(struct nvmet_tcp_queue *queue)
> +{
> +	int ret;
> +
> +	cancel_work(&queue->io_work);
> +	handshake_req_cancel(queue->sock->sk);
> +	handshake_sk_destruct_req(queue->sock->sk);
> +	queue->state = NVMET_TCP_Q_TLS_HANDSHAKE;
> +
> +	/* Restore the default callbacks before starting upcall */
> +	read_lock_bh(&queue->sock->sk->sk_callback_lock);
> +	queue->sock->sk->sk_data_ready =  queue->data_ready;
> +	queue->sock->sk->sk_state_change = queue->state_change;
> +	queue->sock->sk->sk_write_space = queue->write_space;
> +	queue->sock->sk->sk_user_data = NULL;

Shouldn't "sk_user_data = NULL" be protected by a write lock?

Maurizio