On Thu, 2012-09-27 at 13:43 -0400, Martin K. Petersen wrote:

quoted

quoted

quoted

quoted

quoted

"Li" == Li Zhong [off-list ref] writes:

quoted

@@ -845,8 +844,11 @@ static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,

	scsi_eh_restore_cmnd(scmd, &ses);

-	if (sdrv && sdrv->eh_action)
-		rtn = sdrv->eh_action(scmd, cmnd, cmnd_size, rtn);
+	if (scmd->request->cmd_type == REQ_TYPE_FS) {
+		struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
+		if (sdrv->eh_action)
+			rtn = sdrv->eh_action(scmd, cmnd, cmnd_size, rtn);
+	}

	return rtn;
}

My only concern is whether our device lifetime rules guarantee that the
ULD is always attached when we service an error handling command?

Yes, they are.  We can only get REQ_TYPE_FS through a filesystem, which
must be mounted on a block device, which is provided by the ULD.  You
can't unmount with outstanding I/O (which people complain about when it
goes into error handling, I'll admit), so the ULD has to stay bound.

James