RE: [linux-iscsi-devel] [PATCH]cleanup of xmit_task

RE: [linux-iscsi-devel] [PATCH]cleanup of xmit_task

[Smitha Narayanaswamy (smithan)]

<..snip>

> @@ -720,17 +720,6 @@ fill_task_immediate_data(struct iscsi_ta
>   	struct crypto_tfm *tfm= NULL;
>   	struct scatterlist tmpsg;
>
> -	/* make sure we have data to send when we expect to */
> -	if ((iscsi_expected_data_length(sc) == 0) &&
> -	    ((sc->request_bufflen == 0) || (sc->request_buffer
> == NULL))) {
> -		printk("iSCSI: xmit_task for itt %u, sc 0x%x,
> expected %u, no "
> -		       "data in buffer. request_buffer %p len
> %u, buffer %p "
> -		       "len %u\n", task->itt, sc->cmnd[0],
> -		       iscsi_expected_data_length(sc),
> sc->request_buffer,
> -		       sc->request_bufflen, sc->buffer, sc->bufflen);
> -		scsi_print_command(sc);
> -		return 0;
> -	}
>
> What protects this function from being called with a scsi command
> that has no data?  Are you sure we don't need this check with maybe
> just a return here?
>

Is it possible to have scsi write command with no data being sent?

<snip..>

Thanks,
Smitha

Re: [linux-iscsi-devel] [PATCH]cleanup of xmit_task

[David Wysochanski]

Smitha Narayanaswamy (smithan) wrote:
> <..snip>
> 
>  > @@ -720,17 +720,6 @@ fill_task_immediate_data(struct iscsi_ta
>  >       struct crypto_tfm *tfm= NULL;
>  >       struct scatterlist tmpsg;
>  >
>  > -     /* make sure we have data to send when we expect to */
>  > -     if ((iscsi_expected_data_length(sc) == 0) &&
>  > -         ((sc->request_bufflen == 0) || (sc->request_buffer
>  > == NULL))) {
>  > -             printk("iSCSI: xmit_task for itt %u, sc 0x%x,
>  > expected %u, no "
>  > -                    "data in buffer. request_buffer %p len
>  > %u, buffer %p "
>  > -                    "len %u\n", task->itt, sc->cmnd[0],
>  > -                    iscsi_expected_data_length(sc),
>  > sc->request_buffer,
>  > -                    sc->request_bufflen, sc->buffer, sc->bufflen);
>  > -             scsi_print_command(sc);
>  > -             return 0;
>  > -     }
>  >
>  > What protects this function from being called with a scsi command
>  > that has no data?  Are you sure we don't need this check with maybe
>  > just a return here?
>  >
> 
> Is it possible to have scsi write command with no data being sent?
> 
I don't know.

Since nobody replied with a definate 'yes', I'd guess
it's probably a pathalogical case that we don't
need to waste time checking for in the main IO path.