Re: [PATCH 2/5] scsi: introduce SHOST_BLOCKED state to support blocking IO

[Bart Van Assche]

On 3/24/23 18:17, Ye Bin wrote:
> diff --git a/drivers/scsi/hosts.c b/drivers/scsi/hosts.c
> index 9b6fbbe15d92..3b497fd4d329 100644
> --- a/drivers/scsi/hosts.c
> +++ b/drivers/scsi/hosts.c
> @@ -90,6 +90,7 @@ int scsi_host_set_state(struct Scsi_Host *shost, enum scsi_host_state state)
>   		switch (oldstate) {
>   		case SHOST_CREATED:
>   		case SHOST_RECOVERY:
> +		case SHOST_BLOCKED:
>   			break;
>   		default:
>   			goto illegal;
> @@ -99,6 +100,7 @@ int scsi_host_set_state(struct Scsi_Host *shost, enum scsi_host_state state)
>   	case SHOST_RECOVERY:
>   		switch (oldstate) {
>   		case SHOST_RUNNING:
> +		case SHOST_BLOCKED:
>   			break;
>   		default:
>   			goto illegal;
> @@ -109,6 +111,7 @@ int scsi_host_set_state(struct Scsi_Host *shost, enum scsi_host_state state)
>   		switch (oldstate) {
>   		case SHOST_CREATED:
>   		case SHOST_RUNNING:
> +		case SHOST_BLOCKED:
>   		case SHOST_CANCEL_RECOVERY:
>   			break;
>   		default:
> @@ -144,6 +147,14 @@ int scsi_host_set_state(struct Scsi_Host *shost, enum scsi_host_state state)
>   			goto illegal;
>   		}
>   		break;
> +	case SHOST_BLOCKED:
> +		switch (oldstate) {
> +		case SHOST_RUNNING:
> +			break;
> +		default:
> +			goto illegal;
> +		}
> +		break;
>   	}

If a host is blocked, error recovery happens and completes, the host 
will be unblocked. I don't think that is acceptable.

The "blocked" property is orthogonal to the host state so a new boolean 
member variable should be introduced in struct Scsi_Host instead of 
introducing a new SCSI host state.

Thanks,

Bart.