[SCSI] zfcp: IO stall after deleting and path checker changes after reenabling zfcp...
authorMichael Loehr <mloehr2@linux.vnet.ibm.com>
Wed, 9 May 2007 09:01:24 +0000 (11:01 +0200)
committerJames Bottomley <jejb@mulgrave.il.steeleye.com>
Wed, 16 May 2007 14:01:38 +0000 (10:01 -0400)
IO stall after deleting and path checker changes after reenabling zfcp device

Setting one zfcp device offline using chccwdev in a multipath
environment and waiting will lead to IO stall on all paths.
After setting the zfcp device back online using chccwdev,
the devices with io stall will have a different path checker.
Devices corresponding to the deleted units are never freed.
This has the effect that 'slave_destroy' is never called and zfcp
still thinks that this unit is registered
(ZFCP_STATUS_UNIT_REGISTERED is still set). Hence the erp
routine is not called correctly and the unit is not enabled properly.

Do not delete rport and the sdev. Just set the host to block on
'offline'. Setting host online again will then remove the blocked status
and everything is fine again.

Signed-off-by: Michael Loehr <mloehr2@linux.vnet.ibm.com>
Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
drivers/s390/scsi/zfcp_aux.c
drivers/s390/scsi/zfcp_ccw.c
drivers/s390/scsi/zfcp_scsi.c

index ddff40c4212c28068dcb3b58e501ba1398d4e336..821cde65e3694ecc3ab6f68cb30c168c276581cf 100644 (file)
@@ -1127,6 +1127,7 @@ zfcp_adapter_dequeue(struct zfcp_adapter *adapter)
        int retval = 0;
        unsigned long flags;
 
+       zfcp_adapter_scsi_unregister(adapter);
        device_unregister(&adapter->generic_services);
        zfcp_sysfs_adapter_remove_files(&adapter->ccw_device->dev);
        dev_set_drvdata(&adapter->ccw_device->dev, NULL);
index 81680efa172138499311cab2789070a115f0c730..1c8f71a59855b26b2123e611b3ce3978ee6d5dd2 100644 (file)
@@ -189,9 +189,7 @@ zfcp_ccw_set_online(struct ccw_device *ccw_device)
  * @ccw_device: pointer to belonging ccw device
  *
  * This function gets called by the common i/o layer and sets an adapter
- * into state offline. Setting an fcp device offline means that it will be
- * unregistered from the SCSI stack and that the adapter will be shut down
- * asynchronously.
+ * into state offline.
  */
 static int
 zfcp_ccw_set_offline(struct ccw_device *ccw_device)
@@ -202,7 +200,6 @@ zfcp_ccw_set_offline(struct ccw_device *ccw_device)
        adapter = dev_get_drvdata(&ccw_device->dev);
        zfcp_erp_adapter_shutdown(adapter, 0);
        zfcp_erp_wait(adapter);
-       zfcp_adapter_scsi_unregister(adapter);
        zfcp_erp_thread_kill(adapter);
        zfcp_adapter_debug_unregister(adapter);
        up(&zfcp_data.config_sema);
index 16e2d64658afea075143230bdcc08d9a492a58a8..0acf6db0a08d7ff3a60d13dc4cccec0e189ae0da 100644 (file)
@@ -569,6 +569,9 @@ zfcp_adapter_scsi_register(struct zfcp_adapter *adapter)
        int retval = 0;
        static unsigned int unique_id = 0;
 
+       if (adapter->scsi_host)
+               goto out;
+
        /* register adapter as SCSI host with mid layer of SCSI stack */
        adapter->scsi_host = scsi_host_alloc(&zfcp_data.scsi_host_template,
                                             sizeof (struct zfcp_adapter *));