Skip to contentSkip to conversation

target: fix the pgr/alua_support_store functions

PATCH Superseded Maurizio Lombardi 3 participants 4 messages Sep 3, 2021 – Sep 6, 2021

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 target: fix the pgr/alua_support_store functions

Commit 356ba2a8bc8d ("scsi: target: tcmu: Make pgr_support and alua_support attributes writable")
introduced support for changeable alua_support and pgr_support target attributes. They can only be changed
if the backstore is user-backed, otherwise the kernel returns -EINVAL.

This caused a regression in targetcli/rtslib because now a warning is triggered when performing a target restore that includes non-userbacked backstores, even if rtslib is not trying to change the attributes' values:

$ targetctl restore
Storage Object block/storage1: Cannot set attribute alua_support:
[Errno 22] Invalid argument, skipped
Storage Object block/storage1: Cannot set attribute pgr_support:
[Errno 22] Invalid argument, skipped

Fix this warning by returning an error only if we are really going to flip the PGR/ALUA bit in the transport_flags field, otherwise we'll do nothing and return success.

Return EOPNOTSUPP instead of EINVAL if the pgr/alua attributes can't be changed, this way it'll be possible for userspace to understand if the operation failed because an invalid value has been passed to strtobool() or because the attributes are fixed.

Fixes 356ba2a8bc8d ("scsi: target: tcmu: Make pgr_support and alua_support attributes writable")
---
 drivers/target/target_core_configfs.c | 32 +++++++++++++++++----------
 1 file changed, 20 insertions(+), 12 deletions(-)
drivers/target/target_core_configfs.c +20 -12
index 102ec644bc8a..72d750f753bf 100644
@@ -1110,20 +1110,24 @@ static ssize_t alua_support_store(struct config_item *item,
 {
 	struct se_dev_attrib *da = to_attrib(item);
 	struct se_device *dev = da->da_dev;
-	bool flag;
+	bool flag, oldflag;
 	int ret;
 
+	ret = strtobool(page, &flag);
+	if (ret < 0)
+		return ret;
+
+	oldflag = !(dev->transport_flags & TRANSPORT_FLAG_PASSTHROUGH_ALUA);
+	if (flag == oldflag)
+		return count;
+
 	if (!(dev->transport->transport_flags_changeable &
 	      TRANSPORT_FLAG_PASSTHROUGH_ALUA)) {
 		pr_err("dev[%p]: Unable to change SE Device alua_support:"
 			" alua_support has fixed value\n", dev);
-		return -EINVAL;
+		return -EOPNOTSUPP;
 	}
 
-	ret = strtobool(page, &flag);
-	if (ret < 0)
-		return ret;
-
 	if (flag)
 		dev->transport_flags &= ~TRANSPORT_FLAG_PASSTHROUGH_ALUA;
 	else
@@ -1145,20 +1149,24 @@ static ssize_t pgr_support_store(struct config_item *item,
 {
 	struct se_dev_attrib *da = to_attrib(item);
 	struct se_device *dev = da->da_dev;
-	bool flag;
+	bool flag, oldflag;
 	int ret;
 
+	ret = strtobool(page, &flag);
+	if (ret < 0)
+		return ret;
+
+	oldflag = !(dev->transport_flags & TRANSPORT_FLAG_PASSTHROUGH_PGR);
+	if (flag == oldflag)
+		return count;
+
 	if (!(dev->transport->transport_flags_changeable &
 	      TRANSPORT_FLAG_PASSTHROUGH_PGR)) {
 		pr_err("dev[%p]: Unable to change SE Device pgr_support:"
 			" pgr_support has fixed value\n", dev);
-		return -EINVAL;
+		return -EOPNOTSUPP;
 	}
 
-	ret = strtobool(page, &flag);
-	if (ret < 0)
-		return ret;
-
 	if (flag)
 		dev->transport_flags &= ~TRANSPORT_FLAG_PASSTHROUGH_PGR;
 	else
Signature · 1 line
2.27.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 26 lines · Maurizio Lombardi

Commit 356ba2a8bc8d ("scsi: target: tcmu: Make pgr_support and alua_support attributes writable")
introduced support for changeable alua_support and pgr_support target attributes. They can only be changed
if the backstore is user-backed, otherwise the kernel returns -EINVAL.

This caused a regression in targetcli/rtslib because now a warning is triggered when performing a target restore that includes non-userbacked backstores, even if rtslib is not trying to change the attributes' values:

$ targetctl restore
Storage Object block/storage1: Cannot set attribute alua_support:
[Errno 22] Invalid argument, skipped
Storage Object block/storage1: Cannot set attribute pgr_support:
[Errno 22] Invalid argument, skipped

Fix this warning by returning an error only if we are really going to flip the PGR/ALUA bit in the transport_flags field, otherwise we'll do nothing and return success.

Return EOPNOTSUPP instead of EINVAL if the pgr/alua attributes can't be changed, this way it'll be possible for userspace to understand if the operation failed because an invalid value has been passed to strtobool() or because the attributes are fixed.

Quoted 4 lines
Fixes 356ba2a8bc8d ("scsi: target: tcmu: Make pgr_support and alua_support attributes writable")

---

#2 lore raw
In reply to Maurizio Lombardi
Quoted 17 lines · Maurizio Lombardi

Commit 356ba2a8bc8d ("scsi: target: tcmu: Make pgr_support and alua_support attributes writable")
introduced support for changeable alua_support and pgr_support target attributes. They can only be changed
if the backstore is user-backed, otherwise the kernel returns -EINVAL.

This caused a regression in targetcli/rtslib because now a warning is triggered when performing a target restore that includes non-userbacked backstores, even if rtslib is not trying to change the attributes' values:

$ targetctl restore
Storage Object block/storage1: Cannot set attribute alua_support:
[Errno 22] Invalid argument, skipped
Storage Object block/storage1: Cannot set attribute pgr_support:
[Errno 22] Invalid argument, skipped
Quoted 8 lines

Fix this warning by returning an error only if we are really going to flip the PGR/ALUA bit in the transport_flags field, otherwise we'll do nothing and return success.

Return EOPNOTSUPP instead of EINVAL if the pgr/alua attributes can't be changed, this way it'll be possible for userspace to understand if the operation failed because an invalid value has been passed to strtobool() or because the attributes are fixed.

Maybe you want to use -ENOSYS. Other lio sysfs files return that when the operation is not supported. You could then handle all the warnings with the same check.

#3 lore raw

so 4. 9. 2021 v 22:05 odesílatel <[email protected]> napsal:

Quoted 5 lines

Maybe you want to use -ENOSYS. Other lio sysfs files return that when the operation is not supported. You could then handle all the warnings with the same check.

Ok, better to be consistent with other LIO sysfs functions. I will submit a V2 later today.

Thanks,
Maurizio