Re: [PATCH] null_blk: fix warning in blk_mq_start_request

[Bart Van Assche <bvanassche@acm.org>]

On 11/7/23 18:46, Edward Adam Davis wrote:
> Before call blk_mq_start_request() in null_queue_rq(), initialize rq->state to
> MQ_RQ_IDLE.
> 
> Reported-and-tested-by: syzbot+fcc47ba2476570cbbeb0@syzkaller.appspotmail.com
> Signed-off-by: Edward Adam Davis <eadavis@qq.com>
> ---
>   drivers/block/null_blk/main.c | 2 ++
>   1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/block/null_blk/main.c b/drivers/block/null_blk/main.c
> index 22a3cf7f32e2..0726534a5a24 100644
> --- a/drivers/block/null_blk/main.c
> +++ b/drivers/block/null_blk/main.c
> @@ -1724,6 +1724,8 @@ static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
>   	cmd->fake_timeout = should_timeout_request(rq) ||
>   		blk_should_fake_timeout(rq->q);
>   
> +	if (READ_ONCE(rq->state))
> +		WRITE_ONCE(rq->state, MQ_RQ_IDLE);
>   	blk_mq_start_request(rq);
>   
>   	if (should_requeue_request(rq)) {

All code that changes rq->state should occur in the block layer
core. Block drivers must not modify rq->state directly.

Bart.