Re: nvme-tcp: fix a possible UAF when failing to send request【请注意，邮件由sagigrim@xxxxxxxxx代发】

[Hannes Reinecke]

On 3/13/25 02:48, zhang.guanghui@xxxxxxxx wrote:
> Yes, the problem here is that,  despite the nvme_tcp_try_send() failure, the target sends a response capsule for the command, leading to a UAF in the host.
>
> Is it more reasonable to disable queue->rd_enabled to prevent receiving. Thanks
>   
> diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
> index be04c5f3856d..17407eb12ad9 100644
> --- a/drivers/nvme/host/tcp.c
> +++ b/drivers/nvme/host/tcp.c
> @@ -1203,8 +1203,9 @@ static int nvme_tcp_try_send(struct nvme_tcp_queue *queue)
>          } else if (ret < 0) {
>                  dev_err(queue->ctrl->ctrl.device,
>                          "failed to send request %d\n", ret);
> -               nvme_tcp_fail_request(queue->request);
>                  nvme_tcp_done_send_req(queue);
> +              queue->rd_enabled = false;
> +              nvme_tcp_error_recovery(&queue->ctrl->ctrl);
>          }
>   out:
>          memalloc_noreclaim_restore(noreclaim_flag);
Hmm. In principle, yes. Problem is that network is a bi-directional 
communication, and a failure on one side doesn't necessarily imply
a failure on the other.
In particular when the send side fails we should _continue_ to read
as we should be flushing the read side buffer before closing.

So I agree with starting error recovery, but not with disabling the 
reading side (as we haven't encountered a read error).

Cheers,

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