Re: [PATCH 1/2] nbd: replace wait_for_reconnect with non-blocking retry

Friendly ping.

On 7/7/2026 6:27 PM, Yun Zhou wrote:
> wait_for_reconnect() sleeps in the block dispatch path holding the
> SRCU read lock.  This blocks blk_mq_quiesce_queue() (which needs
> synchronize_srcu) for the entire dead_conn_timeout duration, triggering
> hung task warnings.
> 
> Replace it with a non-blocking nbd_reconnect_possible() check that
> returns BLK_STS_DEV_RESOURCE to keep the request on the dispatch list,
> with a 1-second delayed queue run to re-evaluate.  Once the timeout
> expires or the device is disconnected, fail the I/O immediately.
> 
> Fixes: 560bc4b39952 ("nbd: handle dead connections")
> Reported-by: [email protected]
> Closes: https://syzkaller.appspot.com/bug?extid=30c16035531e3248dcbc
> Signed-off-by: Yun Zhou <[email protected]>
> ---
>   drivers/block/nbd.c | 61 +++++++++++++++++++++++++++------------------
>   1 file changed, 37 insertions(+), 24 deletions(-)
> 
> diff --git a/drivers/block/nbd.c b/drivers/block/nbd.c
> index 70a04d541ea4..43aa4121f0c5 100644
> --- a/drivers/block/nbd.c
> +++ b/drivers/block/nbd.c
> @@ -94,11 +94,11 @@ struct nbd_config {
>   	u32 flags;
>   	unsigned long runtime_flags;
>   	u64 dead_conn_timeout;
> +	unsigned long dead_conn_start; /* jiffies, 0 = not waiting */
>   
>   	struct nbd_sock **socks;
>   	int num_connections;
>   	atomic_t live_connections;
> -	wait_queue_head_t conn_wait;
>   
>   	atomic_t recv_threads;
>   	wait_queue_head_t recv_wq;
> @@ -1124,20 +1124,25 @@ static int find_fallback(struct nbd_device *nbd, int index)
>   	return new_index;
>   }
>   
> -static int wait_for_reconnect(struct nbd_device *nbd)
> +static bool nbd_reconnect_possible(struct nbd_device *nbd)
>   {
>   	struct nbd_config *config = nbd->config;
> +
>   	if (!config->dead_conn_timeout)
> -		return 0;
> +		return false;
> +	if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
> +		return false;
>   
> -	if (!wait_event_timeout(config->conn_wait,
> -				test_bit(NBD_RT_DISCONNECTED,
> -					 &config->runtime_flags) ||
> -				atomic_read(&config->live_connections) > 0,
> -				config->dead_conn_timeout))
> -		return 0;
> +	/* Record when all connections first went dead */
> +	if (!READ_ONCE(config->dead_conn_start))
> +		WRITE_ONCE(config->dead_conn_start, jiffies ? : 1);
> +
> +	/* Check if we've exceeded the reconnect timeout */
> +	if (time_after(jiffies, READ_ONCE(config->dead_conn_start) +
> +		       (unsigned long)config->dead_conn_timeout))
> +		return false;
>   
> -	return !test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
> +	return true;
>   }
>   
>   static blk_status_t nbd_handle_cmd(struct nbd_cmd *cmd, int index)
> @@ -1168,23 +1173,24 @@ static blk_status_t nbd_handle_cmd(struct nbd_cmd *cmd, int index)
>   	nsock = config->socks[index];
>   	mutex_lock(&nsock->tx_lock);
>   	if (nsock->dead) {
> -		int old_index = index;
>   		index = find_fallback(nbd, index);
>   		mutex_unlock(&nsock->tx_lock);
>   		if (index < 0) {
> -			if (wait_for_reconnect(nbd)) {
> -				index = old_index;
> -				goto again;
> +			if (!nbd_reconnect_possible(nbd)) {
> +				sock_shutdown(nbd);
> +				nbd_config_put(nbd);
> +				return BLK_STS_IOERR;
>   			}
> -			/* All the sockets should already be down at this point,
> -			 * we just want to make sure that DISCONNECTED is set so
> -			 * any requests that come in that were queue'ed waiting
> -			 * for the reconnect timer don't trigger the timer again
> -			 * and instead just error out.
> +			/*
> +			 * All connections are dead but reconnect timeout
> +			 * has not expired.  Return BLK_STS_DEV_RESOURCE
> +			 * so the request stays on the dispatch list, and
> +			 * schedule a delayed queue run after 1 second to
> +			 * re-evaluate.
>   			 */
> -			sock_shutdown(nbd);
> +			blk_mq_delay_run_hw_queues(nbd->disk->queue, 1000);
>   			nbd_config_put(nbd);
> -			return BLK_STS_IOERR;
> +			return BLK_STS_DEV_RESOURCE;
>   		}
>   		goto again;
>   	}
> @@ -1437,7 +1443,9 @@ static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
>   		queue_work(nbd->recv_workq, &args->work);
>   
>   		atomic_inc(&config->live_connections);
> -		wake_up(&config->conn_wait);
> +		/* Reconnected -- stop waiting for dead_conn_timeout */
> +		WRITE_ONCE(config->dead_conn_start, 0);
> +		blk_mq_run_hw_queues(nbd->disk->queue, true);
>   		return 0;
>   	}
>   	sk_clear_memalloc(sock->sk);
> @@ -1499,6 +1507,13 @@ static int nbd_disconnect(struct nbd_device *nbd)
>   	dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
>   	set_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
>   	set_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags);
> +	/*
> +	 * If all connections are already dead, no sock_shutdown callback
> +	 * will fire to set NBD_RT_DISCONNECTED.  Set it here so
> +	 * nbd_reconnect_possible() stops waiting immediately.
> +	 */
> +	if (atomic_read(&config->live_connections) == 0)
> +		set_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
>   	send_disconnects(nbd);
>   	return 0;
>   }
> @@ -1766,7 +1781,6 @@ static int nbd_alloc_and_init_config(struct nbd_device *nbd)
>   
>   	atomic_set(&config->recv_threads, 0);
>   	init_waitqueue_head(&config->recv_wq);
> -	init_waitqueue_head(&config->conn_wait);
>   	config->blksize_bits = NBD_DEF_BLKSIZE_BITS;
>   	atomic_set(&config->live_connections, 0);
>   
> @@ -2337,7 +2351,6 @@ static void nbd_disconnect_and_put(struct nbd_device *nbd)
>   	mutex_lock(&nbd->config_lock);
>   	nbd_disconnect(nbd);
>   	sock_shutdown(nbd);
> -	wake_up(&nbd->config->conn_wait);
>   	/*
>   	 * Clear NBD_RT_BOUND before releasing config_lock so that
>   	 * nbd_genl_reconfigure() won't queue new recv_work between