Re: [PATCH 05/42] scsi: stop using DRIVER_ERROR

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

On 4/21/21 10:47 AM, Hannes Reinecke wrote:
> diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
> index c532f9390ae3..2d9b533ef1ec 100644
> --- a/drivers/scsi/scsi_lib.c
> +++ b/drivers/scsi/scsi_lib.c
> @@ -245,20 +245,23 @@ int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
>   {
>   	struct request *req;
>   	struct scsi_request *rq;
> -	int ret = DRIVER_ERROR << 24;
> +	int ret;
>   
>   	req = blk_get_request(sdev->request_queue,
>   			data_direction == DMA_TO_DEVICE ?
>   			REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN,
>   			rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
>   	if (IS_ERR(req))
> -		return ret;
> -	rq = scsi_req(req);
> +		return PTR_ERR(req);
>   
> -	if (bufflen &&	blk_rq_map_kern(sdev->request_queue, req,
> -					buffer, bufflen, GFP_NOIO))
> -		goto out;
> +	rq = scsi_req(req);
>   
> +	if (bufflen) {
> +		ret = blk_rq_map_kern(sdev->request_queue, req,
> +				      buffer, bufflen, GFP_NOIO);
> +		if (ret)
> +			goto out;
> +	}
>   	rq->cmd_len = COMMAND_SIZE(cmd[0]);
>   	memcpy(rq->cmd, cmd, rq->cmd_len);
>   	rq->retries = retries;

Please mention in the patch description that this change involves a user 
space ABI change. My understanding is that the current behavior of the 
IOC_PR_* ioctls is as follows:
* A value <= 0 is returned for NVMe where 0 represents success and a 
negative value represents failure.
* A value >= 0 is returned for SCSI where 0 represents success and a 
positive value is a four-byte SCSI status code.

This patch changes the behavior for SCSI from returning a value >= 0 
into returning a value that can be negative, zero or positive where only 
the return value 0 represents success.

See also sd_pr_command() in drivers/scsi/sd.c.

Thanks,

Bart.