[PATCH V2 1/2] nvmet-tcp: propagate nvmet_tcp_build_pdu_iovec() errors to its callers

[Hannes Reinecke]

On 3/12/26 14:40, Maurizio Lombardi wrote:
> Currently, when nvmet_tcp_build_pdu_iovec() detects an out-of-bounds
> PDU length or offset, it triggers nvmet_tcp_fatal_error(cmd->queue)
> and returns early. However, because the function returns void, the
> callers are entirely unaware that a fatal error has occurred and
> that the cmd->recv_msg.msg_iter was left uninitialized.
> 
> Callers such as nvmet_tcp_handle_h2c_data_pdu() proceed to blindly
> overwrite the queue state with queue->rcv_state = NVMET_TCP_RECV_DATA
> Consequently, the socket receiving loop may attempt to read incoming
> network data into the uninitialized iterator.
> 
> Fix this by shifting the error handling responsibility to the callers.
> 
> Fixes: 52a0a9854934 ("nvmet-tcp: add bounds checks in nvmet_tcp_build_pdu_iovec")
> Signed-off-by: Maurizio Lombardi <mlombard at redhat.com>
> ---
>   drivers/nvme/target/tcp.c | 51 ++++++++++++++++++++++-----------------
>   1 file changed, 29 insertions(+), 22 deletions(-)
> 
> diff --git a/drivers/nvme/target/tcp.c b/drivers/nvme/target/tcp.c
> index acc71a26733f..1fbf12df1183 100644
> --- a/drivers/nvme/target/tcp.c
> +++ b/drivers/nvme/target/tcp.c
> @@ -351,7 +351,7 @@ static void nvmet_tcp_free_cmd_buffers(struct nvmet_tcp_cmd *cmd)
>   
>   static void nvmet_tcp_fatal_error(struct nvmet_tcp_queue *queue);
>   
> -static void nvmet_tcp_build_pdu_iovec(struct nvmet_tcp_cmd *cmd)
> +static int nvmet_tcp_build_pdu_iovec(struct nvmet_tcp_cmd *cmd)
>   {
>   	struct bio_vec *iov = cmd->iov;
>   	struct scatterlist *sg;
> @@ -364,22 +364,19 @@ static void nvmet_tcp_build_pdu_iovec(struct nvmet_tcp_cmd *cmd)
>   	offset = cmd->rbytes_done;
>   	cmd->sg_idx = offset / PAGE_SIZE;
>   	sg_offset = offset % PAGE_SIZE;
> -	if (!cmd->req.sg_cnt || cmd->sg_idx >= cmd->req.sg_cnt) {
> -		nvmet_tcp_fatal_error(cmd->queue);
> -		return;
> -	}
> +	if (!cmd->req.sg_cnt || cmd->sg_idx >= cmd->req.sg_cnt)
> +		return -EPROTO;
> +
>   	sg = &cmd->req.sg[cmd->sg_idx];
>   	sg_remaining = cmd->req.sg_cnt - cmd->sg_idx;
>   
>   	while (length) {
> -		if (!sg_remaining) {
> -			nvmet_tcp_fatal_error(cmd->queue);
> -			return;
> -		}
> -		if (!sg->length || sg->length <= sg_offset) {
> -			nvmet_tcp_fatal_error(cmd->queue);
> -			return;
> -		}
> +		if (!sg_remaining)
> +			return -EPROTO;
> +
> +		if (!sg->length || sg->length <= sg_offset)
> +			return -EPROTO;
> +
>   		u32 iov_len = min_t(u32, length, sg->length - sg_offset);
>   
>   		bvec_set_page(iov, sg_page(sg), iov_len,
> @@ -394,6 +391,7 @@ static void nvmet_tcp_build_pdu_iovec(struct nvmet_tcp_cmd *cmd)
>   
>   	iov_iter_bvec(&cmd->recv_msg.msg_iter, ITER_DEST, cmd->iov,
>   		      nr_pages, cmd->pdu_len);
> +	return 0;
>   }
>   
>   static void nvmet_tcp_fatal_error(struct nvmet_tcp_queue *queue)
> @@ -931,7 +929,7 @@ static int nvmet_tcp_handle_icreq(struct nvmet_tcp_queue *queue)
>   	return 0;
>   }
>   
> -static void nvmet_tcp_handle_req_failure(struct nvmet_tcp_queue *queue,
> +static int nvmet_tcp_handle_req_failure(struct nvmet_tcp_queue *queue,
>   		struct nvmet_tcp_cmd *cmd, struct nvmet_req *req)
>   {
>   	size_t data_len = le32_to_cpu(req->cmd->common.dptr.sgl.length);
> @@ -947,19 +945,23 @@ static void nvmet_tcp_handle_req_failure(struct nvmet_tcp_queue *queue,
>   	if (!nvme_is_write(cmd->req.cmd) || !data_len ||
>   	    data_len > cmd->req.port->inline_data_size) {
>   		nvmet_prepare_receive_pdu(queue);
> -		return;
> +		return 0;
>   	}
>   
>   	ret = nvmet_tcp_map_data(cmd);
>   	if (unlikely(ret)) {
>   		pr_err("queue %d: failed to map data\n", queue->idx);
>   		nvmet_tcp_fatal_error(queue);
> -		return;
> +		return -EPROTO;
>   	}
>   
>   	queue->rcv_state = NVMET_TCP_RECV_DATA;
> -	nvmet_tcp_build_pdu_iovec(cmd);
>   	cmd->flags |= NVMET_TCP_F_INIT_FAILED;
> +	ret = nvmet_tcp_build_pdu_iovec(cmd);
> +	if (unlikely(ret))
> +		pr_err("queue %d: failed to build PDU iovec\n", queue->idx);

Why don't we call 'nvmet_tcp_fatal_error()' here?
The original code did ...

> +
> +	return ret;
>   }
>   
>   static int nvmet_tcp_handle_h2c_data_pdu(struct nvmet_tcp_queue *queue)
> @@ -1011,7 +1013,10 @@ static int nvmet_tcp_handle_h2c_data_pdu(struct nvmet_tcp_queue *queue)
>   		goto err_proto;
>   	}
>   	cmd->pdu_recv = 0;
> -	nvmet_tcp_build_pdu_iovec(cmd);
> +	if (unlikely(nvmet_tcp_build_pdu_iovec(cmd))) {
> +		pr_err("queue %d: failed to build PDU iovec\n", queue->idx);
> +		goto err_proto;
> +	}
>   	queue->cmd = cmd;
>   	queue->rcv_state = NVMET_TCP_RECV_DATA;
>   
> @@ -1074,8 +1079,7 @@ static int nvmet_tcp_done_recv_pdu(struct nvmet_tcp_queue *queue)
>   			le32_to_cpu(req->cmd->common.dptr.sgl.length),
>   			le16_to_cpu(req->cqe->status));
>   
> -		nvmet_tcp_handle_req_failure(queue, queue->cmd, req);
> -		return 0;
> +		return nvmet_tcp_handle_req_failure(queue, queue->cmd, req);
>   	}
>   
>   	ret = nvmet_tcp_map_data(queue->cmd);
> @@ -1092,8 +1096,11 @@ static int nvmet_tcp_done_recv_pdu(struct nvmet_tcp_queue *queue)
>   	if (nvmet_tcp_need_data_in(queue->cmd)) {
>   		if (nvmet_tcp_has_inline_data(queue->cmd)) {
>   			queue->rcv_state = NVMET_TCP_RECV_DATA;
> -			nvmet_tcp_build_pdu_iovec(queue->cmd);
> -			return 0;
> +			ret = nvmet_tcp_build_pdu_iovec(queue->cmd);
> +			if (unlikely(ret))
> +				pr_err("queue %d: failed to build PDU iovec\n",
> +					queue->idx);
> +			return ret;
>   		}
>   		/* send back R2T */
>   		nvmet_tcp_queue_response(&queue->cmd->req);

Cheers,

Hannes
-- 
Dr. Hannes Reinecke                  Kernel Storage Architect
hare at suse.de                                +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich