iscsi_initiator and tag opening problem

Daniel Braniss danny at cs.huji.ac.il
Wed May 25 07:20:46 UTC 2011


> >
> > it's a CAM thing not PDU/BHS, or at least that is my undertanding.
> > myabe it's time to read the RFC again
> >
> 
> No, what I meant can be seen in the following from iscsi_subr.c:
> 
>       /*
>        | map tag option, default is UNTAGGED
>        */
>       switch(csio->tag_action) {
>       case MSG_SIMPLE_Q_TAG:     cmd->attr = iSCSI_TASK_SIMPLE;  break;
>       case MSG_HEAD_OF_Q_TAG:    cmd->attr = iSCSI_TASK_HOFQ;    break;
>       case MSG_ORDERED_Q_TAG:    cmd->attr = iSCSI_TASK_ORDER;   break;
>       case MSG_ACA_TASK:         cmd->attr = iSCSI_TASK_ACA;     break;
>       }
> 

how emabarrassing! in my defense I wrote this back in 2004 and never looked
at it again :-)

> 
> That is, you're translating from CAM to iSCSI task attributes. Or, rather, 
> iSCSI transport related task attributes. The same thing can be seen in the 
> isp FibreChannel driver:
> 
>                  if (ttype == REQFLAG_OTAG) {
>                          ttype = FCP_CMND_TASK_ATTR_ORDERED;
>                  } else if (ttype == REQFLAG_HTAG) {
>                          ttype = FCP_CMND_TASK_ATTR_HEAD;
>                  } else {
>                          ttype = FCP_CMND_TASK_ATTR_SIMPLE;
>                  }
>                  ((ispreqt7_t *)reqp)->req_task_attribute = ttype;
> 
> (yes, mine is skipping ACA, oops)
> 
> These are attributes of SAM task management. CAM, like any other SCSI 
> midlayer, is quite right to attach the specified attribute. Whether the 
> underlying transport can faithfully support those attributes is another 
> matter.

is this somehow detectable? or a list of '[non]complaint' targets is needed?

> 
> The RFC has the following:
> 
> 3.5.1.1.  SCSI-Command
> 
>     This request carries the SCSI CDB and all the other SCSI execute
>     command procedure call (see [SAM2]) IN arguments such as task
>     attributes, Expected Data Transfer Length for one or both transfer
>     directions
> 
> This would imply that iSCSI is supposed to support task attributes.
> 
> Now, it turns out that the linux openiscsi implementation only uses the 
> iSCSI_TASK_SIMPLE attribute, so perhaps the simplest fix here is to just 
> do that for the FreeBSD initiator driver.

any side effects? or should I make it tunable?

Thanks Matthew, you made my day!

	danny





More information about the freebsd-scsi mailing list