Re: nvme-tcp: fix a possible UAF when failing to send request

[Sagi Grimberg]

On 10/02/2025 9:41, zhang.guanghui@xxxxxxxx wrote:
> Hello
>
>
>
>      When using the nvme-tcp driver in a storage cluster, the driver may trigger a null pointer causing the host to crash several times.
>
>
>
> By analyzing the vmcore, we know the direct cause is that  the request->mq_hctx was used after free.
>
>
>
>
>
> CPU1                                                                   CPU2
>
>
>
> nvme_tcp_poll                                                          nvme_tcp_try_send  --failed to send reqrest 13
>
>
>
>      nvme_tcp_try_recv                                                      nvme_tcp_fail_request
>
>
>
>          nvme_tcp_recv_skb                                                      nvme_tcp_end_request
>
>
>
>              nvme_tcp_recv_pdu                                                      nvme_complete_rq
>
>
>
>                  nvme_tcp_handle_comp                                                   nvme_retry_req -- request->mq_hctx have been freed, is NULL.
>
>
>
>                      nvme_tcp_process_nvme_cqe
>
>
>
>                          nvme_complete_rq
>
>
>
>                              nvme_end_req
>
>
>
>                                    blk_mq_end_request
>
>
>
>
>
>
>
> when nvme_tcp_try_send failed to send reqrest 13, it maybe be resulted by selinux or other reasons, this is a problem. then  the nvme_tcp_fail_request would execute。
>
>
>
> but the nvme_tcp_recv_pdu may have received the responding pdu and the nvme_tcp_process_nvme_cqe would have completed the request.  request->mq_hctx was used after free.
>
>
>
> the follow patch is to solve it.

Zhang, your email client needs fixing - it is impossible to follow your 
emails.

> can you give  some suggestions?  thanks!

The problem is the C2HTerm that the host is unable to handle correctly.
And it also appears that nvme_tcp_poll() does not signal correctly to 
blk-mq to stop
calling poll.

One thing to do is to handle C2HTerm PDU correctly, and, here is a 
possible fix to try for the UAF issue:
--
diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
index c637ff04a052..0e390e98aaf9 100644
--- a/drivers/nvme/host/tcp.c
+++ b/drivers/nvme/host/tcp.c
@@ -2673,6 +2673,7 @@ static int nvme_tcp_poll(struct blk_mq_hw_ctx 
*hctx, struct io_comp_batch *iob)
 {
        struct nvme_tcp_queue *queue = hctx->driver_data;
        struct sock *sk = queue->sock->sk;
+       int ret;

        if (!test_bit(NVME_TCP_Q_LIVE, &queue->flags))
                return 0;
@@ -2680,9 +2681,9 @@ static int nvme_tcp_poll(struct blk_mq_hw_ctx 
*hctx, struct io_comp_batch *iob)
        set_bit(NVME_TCP_Q_POLLING, &queue->flags);
        if (sk_can_busy_loop(sk) && 
skb_queue_empty_lockless(&sk->sk_receive_queue))
                sk_busy_loop(sk, true);
-       nvme_tcp_try_recv(queue);
+       ret = nvme_tcp_try_recv(queue);
        clear_bit(NVME_TCP_Q_POLLING, &queue->flags);
-       return queue->nr_cqe;
+       return ret < 0 ? ret : queue->nr_cqe;
 }

 static int nvme_tcp_get_address(struct nvme_ctrl *ctrl, char *buf, int 
size)
--

Does this help?