Skip to contentSkip to conversation

FC transport: Disable LUN scanning from low level driver

RFC In review Christof Schmitt 2 participants 5 messages Aug 20, 2007 – Aug 21, 2007

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

RFC 1/1 FC transport: Disable LUN scanning from low level driver

The FC transport class calls scsi_scan_target with the SCAN_WILD_CARD flag to automatically scan for logical units. zfcp, on the other hand, only uses the units that are configured via the zfcp sysfs interface. The main reason for this, is that the adapter behind zfcp supports adapter sharing without NPIV: The adapter is logged into the SAN once, and each unit can be used by one Linux system. If one Linux would grab all LUNs, no other one can use them.

If a unit has the LUN 0, then the SCSI midlayer issue a REPORT LUNS, checks the found LUN against the max_lun of the hostadapter. zfcp sets this to 1 to only use its own managed units. If there is a LUN 0 that is used by zfcp, then the SCSI midlayer produces lots of messages like "scsi: host 0 channel 0 id 1 lun2 has a LUN larger than allowed by the host adapter" to indicate the mismatch between the actual LUN and the max_lun setting. These messages only confuse a user in the case of zfcp, since there is no error.

To fix this problem, the LLD (zfcp) has to be able to prevent the automatic scanning from the FC transport class. Attached is a patch that adds a parameter to fc_remote_port_add(), another approach would be an additional flag in the FC transport template.

What do others think? If there is an agreement, i will followup with a new patch.

This can go away, as soon as zfcp does not have to support the non-NPIV adapter sharing anymore, but this won't happen in the foreseeable future.

Christof Schmitt

drivers/s390/scsi/zfcp_erp.c +1 -1
@@ -3175,7 +3175,7 @@ zfcp_erp_action_cleanup(int action, stru
 			ids.port_id = port->d_id;
 			ids.roles = FC_RPORT_ROLE_FCP_TARGET;
 			port->rport =
-				fc_remote_port_add(adapter->scsi_host, 0, &ids);
+				fc_remote_port_add(adapter->scsi_host, 0, &ids, 0);
 			if (!port->rport)
 				ZFCP_LOG_NORMAL("failed registration of rport"
 						"(adapter %s, wwpn=0x%016Lx)\n",
drivers/scsi/scsi_transport_fc.c +12 -4
@@ -2360,7 +2360,7 @@ fc_rport_final_delete(struct work_struct
  **/
 static struct fc_rport *
 fc_rport_create(struct Scsi_Host *shost, int channel,
-	struct fc_rport_identifiers  *ids)
+	struct fc_rport_identifiers  *ids, int scan_target)
 {
 	struct fc_host_attrs *fc_host = shost_to_fc_host(shost);
 	struct fc_internal *fci = to_fc_internal(shost->transportt);
@@ -2424,6 +2424,9 @@ fc_rport_create(struct Scsi_Host *shost,
 	transport_add_device(dev);
 	transport_configure_device(dev);
 
+	if (scan_target)
+		rport->flags |= FC_RPORT_SCAN_TARGET;
+
 	if (rport->roles & FC_PORT_ROLE_FCP_TARGET) {
 		/* initiate a scan of the target */
 		rport->flags |= FC_RPORT_SCAN_PENDING;
@@ -2484,7 +2487,7 @@ delete_rport:
  **/
 struct fc_rport *
 fc_remote_port_add(struct Scsi_Host *shost, int channel,
-	struct fc_rport_identifiers  *ids)
+	struct fc_rport_identifiers  *ids, int scan_target)
 {
 	struct fc_internal *fci = to_fc_internal(shost->transportt);
 	struct fc_host_attrs *fc_host = shost_to_fc_host(shost);
@@ -2574,6 +2577,10 @@ fc_remote_port_add(struct Scsi_Host *sho
 				spin_lock_irqsave(shost->host_lock, flags);
 
 				rport->flags &= ~FC_RPORT_DEVLOSS_PENDING;
+				if (scan_target)
+					rport->flags |= FC_RPORT_SCAN_TARGET;
+				else
+					rport->flags &= ~FC_RPORT_SCAN_TARGET;
 
 				/* if target, initiate a scan */
 				if (rport->scsi_target_id != -1) {
@@ -2657,7 +2664,7 @@ fc_remote_port_add(struct Scsi_Host *sho
 	spin_unlock_irqrestore(shost->host_lock, flags);
 
 	/* No consistent binding found - create new remote port entry */
-	rport = fc_rport_create(shost, channel, ids);
+	rport = fc_rport_create(shost, channel, ids, scan_target);
 
 	return rport;
 }
@@ -2991,7 +2998,8 @@ fc_scsi_scan_rport(struct work_struct *w
 	unsigned long flags;
 
 	if ((rport->port_state == FC_PORTSTATE_ONLINE) &&
-	    (rport->roles & FC_PORT_ROLE_FCP_TARGET)) {
+	    (rport->roles & FC_PORT_ROLE_FCP_TARGET) &&
+	    (rport->flags & FC_RPORT_SCAN_TARGET)) {
 		scsi_scan_target(&rport->dev, rport->channel,
 			rport->scsi_target_id, SCAN_WILD_CARD, 1);
 	}
include/scsi/scsi_transport_fc.h +3 -1
@@ -338,6 +338,7 @@ struct fc_rport {	/* aka fc_starget_attr
 /* bit field values for struct fc_rport "flags" field: */
 #define FC_RPORT_DEVLOSS_PENDING	0x01
 #define FC_RPORT_SCAN_PENDING		0x02
+#define FC_RPORT_SCAN_TARGET		0x04
 
 #define	dev_to_rport(d)				\
 	container_of(d, struct fc_rport, dev)
@@ -713,7 +714,8 @@ struct scsi_transport_template *fc_attac
 void fc_release_transport(struct scsi_transport_template *);
 void fc_remove_host(struct Scsi_Host *);
 struct fc_rport *fc_remote_port_add(struct Scsi_Host *shost,
-			int channel, struct fc_rport_identifiers  *ids);
+			int channel, struct fc_rport_identifiers  *ids,
+			int scan_target);
 void fc_remote_port_delete(struct fc_rport  *rport);
 void fc_remote_port_rolechg(struct fc_rport  *rport, u32 roles);
 int scsi_is_fc_rport(const struct device *);
#1 lore raw

I'd prefer a flag/bit in the fc template to indicate no scanning by the transport. Thus, it's up to the driver to call __scsi_add_device, or perform the appropriate scans...

Note: this doesn't stop the problem in absolute. as long as there's scan interfaces via sysfs, any tool/admin could request a scan and hit your issues. To truly solve it, you need, within the LLDD, to munge what the midlayer sees for LUN lists.

-- james s

Quoted 127 lines · Christof Schmitt

The FC transport class calls scsi_scan_target with the SCAN_WILD_CARD flag to automatically scan for logical units. zfcp, on the other hand, only uses the units that are configured via the zfcp sysfs interface. The main reason for this, is that the adapter behind zfcp supports adapter sharing without NPIV: The adapter is logged into the SAN once, and each unit can be used by one Linux system. If one Linux would grab all LUNs, no other one can use them.

If a unit has the LUN 0, then the SCSI midlayer issue a REPORT LUNS, checks the found LUN against the max_lun of the hostadapter. zfcp sets this to 1 to only use its own managed units. If there is a LUN 0 that is used by zfcp, then the SCSI midlayer produces lots of messages like "scsi: host 0 channel 0 id 1 lun2 has a LUN larger than allowed by the host adapter" to indicate the mismatch between the actual LUN and the max_lun setting. These messages only confuse a user in the case of zfcp, since there is no error.

To fix this problem, the LLD (zfcp) has to be able to prevent the automatic scanning from the FC transport class. Attached is a patch that adds a parameter to fc_remote_port_add(), another approach would be an additional flag in the FC transport template.

What do others think? If there is an agreement, i will followup with a new patch.

This can go away, as soon as zfcp does not have to support the non-NPIV adapter sharing anymore, but this won't happen in the foreseeable future.

Christof Schmitt

drivers/s390/scsi/zfcp_erp.c +1 -1
@@ -3175,7 +3175,7 @@ zfcp_erp_action_cleanup(int action, stru
 			ids.port_id = port->d_id;
 			ids.roles = FC_RPORT_ROLE_FCP_TARGET;
 			port->rport =
-				fc_remote_port_add(adapter->scsi_host, 0, &ids);
+				fc_remote_port_add(adapter->scsi_host, 0, &ids, 0);
 			if (!port->rport)
 				ZFCP_LOG_NORMAL("failed registration of rport"
 						"(adapter %s, wwpn=0x%016Lx)\n",
drivers/scsi/scsi_transport_fc.c +12 -4
@@ -2360,7 +2360,7 @@ fc_rport_final_delete(struct work_struct
  **/
 static struct fc_rport *
 fc_rport_create(struct Scsi_Host *shost, int channel,
-	struct fc_rport_identifiers  *ids)
+	struct fc_rport_identifiers  *ids, int scan_target)
 {
 	struct fc_host_attrs *fc_host = shost_to_fc_host(shost);
 	struct fc_internal *fci = to_fc_internal(shost->transportt);
@@ -2424,6 +2424,9 @@ fc_rport_create(struct Scsi_Host *shost,
 	transport_add_device(dev);
 	transport_configure_device(dev);
 
+	if (scan_target)
+		rport->flags |= FC_RPORT_SCAN_TARGET;
+
 	if (rport->roles & FC_PORT_ROLE_FCP_TARGET) {
 		/* initiate a scan of the target */
 		rport->flags |= FC_RPORT_SCAN_PENDING;
@@ -2484,7 +2487,7 @@ delete_rport:
  **/
 struct fc_rport *
 fc_remote_port_add(struct Scsi_Host *shost, int channel,
-	struct fc_rport_identifiers  *ids)
+	struct fc_rport_identifiers  *ids, int scan_target)
 {
 	struct fc_internal *fci = to_fc_internal(shost->transportt);
 	struct fc_host_attrs *fc_host = shost_to_fc_host(shost);
@@ -2574,6 +2577,10 @@ fc_remote_port_add(struct Scsi_Host *sho
 				spin_lock_irqsave(shost->host_lock, flags);
 
 				rport->flags &= ~FC_RPORT_DEVLOSS_PENDING;
+				if (scan_target)
+					rport->flags |= FC_RPORT_SCAN_TARGET;
+				else
+					rport->flags &= ~FC_RPORT_SCAN_TARGET;
 
 				/* if target, initiate a scan */
 				if (rport->scsi_target_id != -1) {
@@ -2657,7 +2664,7 @@ fc_remote_port_add(struct Scsi_Host *sho
 	spin_unlock_irqrestore(shost->host_lock, flags);
 
 	/* No consistent binding found - create new remote port entry */
-	rport = fc_rport_create(shost, channel, ids);
+	rport = fc_rport_create(shost, channel, ids, scan_target);
 
 	return rport;
 }
@@ -2991,7 +2998,8 @@ fc_scsi_scan_rport(struct work_struct *w
 	unsigned long flags;
 
 	if ((rport->port_state == FC_PORTSTATE_ONLINE) &&
-	    (rport->roles & FC_PORT_ROLE_FCP_TARGET)) {
+	    (rport->roles & FC_PORT_ROLE_FCP_TARGET) &&
+	    (rport->flags & FC_RPORT_SCAN_TARGET)) {
 		scsi_scan_target(&rport->dev, rport->channel,
 			rport->scsi_target_id, SCAN_WILD_CARD, 1);
 	}
include/scsi/scsi_transport_fc.h +3 -2
@@ -338,6 +338,7 @@ struct fc_rport {	/* aka fc_starget_attr
 /* bit field values for struct fc_rport "flags" field: */
 #define FC_RPORT_DEVLOSS_PENDING	0x01
 #define FC_RPORT_SCAN_PENDING		0x02
+#define FC_RPORT_SCAN_TARGET		0x04
 
 #define	dev_to_rport(d)				\
 	container_of(d, struct fc_rport, dev)
@@ -713,7 +714,8 @@ struct scsi_transport_template *fc_attac
 void fc_release_transport(struct scsi_transport_template *);
 void fc_remove_host(struct Scsi_Host *);
 struct fc_rport *fc_remote_port_add(struct Scsi_Host *shost,
-			int channel, struct fc_rport_identifiers  *ids);
+			int channel, struct fc_rport_identifiers  *ids,
+			int scan_target);
 void fc_remote_port_delete(struct fc_rport  *rport);
 void fc_remote_port_rolechg(struct fc_rport  *rport, u32 roles);
 int scsi_is_fc_rport(const struct device *);
-

To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#2 lore raw

James,

thanks for the feedback.

Quoted 3 lines · James Smart

I'd prefer a flag/bit in the fc template to indicate no scanning by the transport. Thus, it's up to the driver to call __scsi_add_device, or perform the appropriate scans...

I attached a new patch that adds a flag to the fc template. So far, it is untested. What is the better approach? Avoid scanning with the flag being 1 or use the 1 to enable scaning and make this change in all FC drivers?

Quoted 4 lines

Note: this doesn't stop the problem in absolute. as long as there's scan interfaces via sysfs, any tool/admin could request a scan and hit your issues. To truly solve it, you need, within the LLDD, to munge what the midlayer sees for LUN lists.

We had the discussion about zfcp inventing LUNs in the past
(http://www.spinics.net/lists/linux-scsi/msg17125.html).

Disabling the scanning in the FC transport class is the most important step at the moment, since the admin can only prevent the warning messages by changing the LUN 0 to something else on the storage system. I think for the user requested scan, we can now advice the user not to issue it for zfcp, if he does anyway he will simply get the warning message.

Christof Schmitt

---
 drivers/scsi/scsi_transport_fc.c |    4 +++-
 include/scsi/scsi_transport_fc.h |    3 +++
 2 files changed, 6 insertions(+), 1 deletion(-)
drivers/scsi/scsi_transport_fc.c +3 -1
@@ -2988,10 +2988,12 @@ fc_scsi_scan_rport(struct work_struct *w
 	struct fc_rport *rport =
 		container_of(work, struct fc_rport, scan_work);
 	struct Scsi_Host *shost = rport_to_shost(rport);
+	struct fc_internal *i = to_fc_internal(shost->transportt);
 	unsigned long flags;
 
 	if ((rport->port_state == FC_PORTSTATE_ONLINE) &&
-	    (rport->roles & FC_PORT_ROLE_FCP_TARGET)) {
+	    (rport->roles & FC_PORT_ROLE_FCP_TARGET) &&
+	    !(i->f->no_target_scan)) {
 		scsi_scan_target(&rport->dev, rport->channel,
 			rport->scsi_target_id, SCAN_WILD_CARD, 1);
 	}
include/scsi/scsi_transport_fc.h +3 -0
@@ -632,6 +632,9 @@ struct fc_function_template {
 	unsigned long	show_host_fabric_name:1;
 	unsigned long	show_host_symbolic_name:1;
 	unsigned long	show_host_system_hostname:1;
+
+	/* The driver can disable the LUN scanning from the FC transport class */
+	unsigned long	no_target_scan:1;
 };
 
 
---
 drivers/s390/scsi/zfcp_scsi.c |    1 +
 1 file changed, 1 insertion(+)
drivers/s390/scsi/zfcp_scsi.c +1 -0
@@ -800,6 +800,7 @@ struct fc_function_template zfcp_transpo
 	.show_host_port_type = 1,
 	.show_host_speed = 1,
 	.show_host_port_id = 1,
+	.no_target_scan = 1,
 };
 
 /**
#3 lore raw

Nit (but preferred) : change "no_target_scan" to "disable_target_scan". Otherwise, patch is good.

Quoted 4 lines · Christof Schmitt

I attached a new patch that adds a flag to the fc template. So far, it is untested. What is the better approach? Avoid scanning with the flag being 1 or use the 1 to enable scaning and make this change in all FC drivers?

1 to disable. Avoid changing the drivers...

Quoted 2 lines

We had the discussion about zfcp inventing LUNs in the past
(http://www.spinics.net/lists/linux-scsi/msg17125.html).

Yep. However, I don't agree with Hannes in this case. Given that the firmware is already doing funky things to share the real adapter, I believe it (or the driver) should do the right thing in the LUN presentation to keep the system happy. But - I guess that's something for a different time.

-- james s

#4 lore raw
Quoted 2 lines · James Smart

Nit (but preferred) : change "no_target_scan" to "disable_target_scan". Otherwise, patch is good.

Ok. I will send the fixed version in a separate mail.

Quoted 5 lines

Yep. However, I don't agree with Hannes in this case. Given that the firmware is already doing funky things to share the real adapter, I believe it (or the driver) should do the right thing in the LUN presentation to keep the system happy. But - I guess that's something for a different time.

Yes, i think it is better to fix the problems when they occur. Adapter sharing in general works well.

Christof