Re: [PATCH] scsi_dh_alua: Requeue another not ready check condition at ML

[Mike Christie <michaelc@cs.wisc.edu>]

On 02/27/2014 06:14 PM, Stewart, Sean wrote:
> This allows the sd driver to retry commands like read capacity until a
> LUN is ready, rather than giving up after three retries.
> 
> In NetApp E-Series, a controller can return not ready like this when it
> quiesces I/O on the controller that just came on the network, during a
> firmware upgrade procedure, and retrying the command at the midlayer
> will allow the discovery to complete, successfully.
> 
> Signed-off-by: Sean Stewart <sean.stewart@netapp.com>
> ---
>  drivers/scsi/device_handler/scsi_dh_alua.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/drivers/scsi/device_handler/scsi_dh_alua.c b/drivers/scsi/device_handler/scsi_dh_alua.c
> index 5248c88..95d87fe 100644
> --- a/drivers/scsi/device_handler/scsi_dh_alua.c
> +++ b/drivers/scsi/device_handler/scsi_dh_alua.c
> @@ -454,6 +454,11 @@ static int alua_check_sense(struct scsi_device *sdev,
>  {
>  	switch (sense_hdr->sense_key) {
>  	case NOT_READY:
> +		if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x01)
> +			/*
> +			 * LUN Not Ready -- In process of becoming ready
> +			 */
> +			return ADD_TO_MLQUEUE;

It seems like the check_sense callout is being used to work around
scsi-ml in a lot of the additions that are not alua specific. If this is
meant for a specific target then it should not be here. If this is
non-alua specific behavior then it should also not be here either.

If the IO was not a REQ_TYPE_BLOCK_PC request, then it would retried by
scsi_io_completion. Same with the other ones like inquiry data changed,
report luns data changed, etc.

Are we sure we don't want to fix the REQ_TYPE_BLOCK_PC/scsi_execute*
users to retry, or to add some new flag that those users can use that
tells scsi-ml to retry like it normally would so callers do not have to
check for all these errors, or just add these to scsi_decide_disposition?