zfcp: close window with unblocked rport during rport gone
authorSteffen Maier <maier@linux.vnet.ibm.com>
Wed, 10 Aug 2016 16:30:46 +0000 (18:30 +0200)
committerMartin K. Petersen <martin.petersen@oracle.com>
Fri, 12 Aug 2016 20:17:12 +0000 (16:17 -0400)
On a successful end of reopen port forced,
zfcp_erp_strategy_followup_success() re-uses the port erp_action
and the subsequent zfcp_erp_action_cleanup() now
sees ZFCP_ERP_SUCCEEDED with
erp_action->action==ZFCP_ERP_ACTION_REOPEN_PORT
instead of ZFCP_ERP_ACTION_REOPEN_PORT_FORCED
but must not perform zfcp_scsi_schedule_rport_register().

We can detect this because the fresh port reopen erp_action
is in its very first step ZFCP_ERP_STEP_UNINITIALIZED.

Otherwise this opens a time window with unblocked rport
(until the followup port reopen recovery would block it again).
If a scsi_cmnd timeout occurs during this time window
fc_timed_out() cannot work as desired and such command
would indeed time out and trigger scsi_eh. This prevents
a clean and timely path failover.
This should not happen if the path issue can be recovered
on FC transport layer such as path issues involving RSCNs.

Also, unnecessary and repeated DID_IMM_RETRY for pending and
undesired new requests occur because internally zfcp still
has its zfcp_port blocked.

As follow-on errors with scsi_eh, it can cause,
in the worst case, permanently lost paths due to one of:
sd <scsidev>: [<scsidisk>] Medium access timeout failure. Offlining disk!
sd <scsidev>: Device offlined - not ready after error recovery

For fix validation and to aid future debugging with other recoveries
we now also trace (un)blocking of rports.

Signed-off-by: Steffen Maier <maier@linux.vnet.ibm.com>
Fixes: 5767620c383a ("[SCSI] zfcp: Do not unblock rport from REOPEN_PORT_FORCED")
Fixes: a2fa0aede07c ("[SCSI] zfcp: Block FC transport rports early on errors")
Fixes: 5f852be9e11d ("[SCSI] zfcp: Fix deadlock between zfcp ERP and SCSI")
Fixes: 338151e06608 ("[SCSI] zfcp: make use of fc_remote_port_delete when target port is unavailable")
Fixes: 3859f6a248cb ("[PATCH] zfcp: add rports to enable scsi_add_device to work again")
Cc: <stable@vger.kernel.org> #2.6.32+
Reviewed-by: Benjamin Block <bblock@linux.vnet.ibm.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/s390/scsi/zfcp_dbf.h
drivers/s390/scsi/zfcp_erp.c
drivers/s390/scsi/zfcp_scsi.c

index 0be3d48681aead94466a71ab7b34c6ae7ca098ff..7901deb4ba89a11a20954e18775d24bc2019b035 100644 (file)
@@ -2,7 +2,7 @@
  * zfcp device driver
  * debug feature declarations
  *
- * Copyright IBM Corp. 2008, 2010
+ * Copyright IBM Corp. 2008, 2015
  */
 
 #ifndef ZFCP_DBF_H
 
 #define ZFCP_DBF_INVALID_LUN   0xFFFFFFFFFFFFFFFFull
 
+enum zfcp_dbf_pseudo_erp_act_type {
+       ZFCP_PSEUDO_ERP_ACTION_RPORT_ADD = 0xff,
+       ZFCP_PSEUDO_ERP_ACTION_RPORT_DEL = 0xfe,
+};
+
 /**
  * struct zfcp_dbf_rec_trigger - trace record for triggered recovery action
  * @ready: number of ready recovery actions
index 3fb410977014f81821e5dd16f65eebcd7f43c4d7..a59d678125bd0e0ad0bd1ca74b0d42985abb25d8 100644 (file)
@@ -3,7 +3,7 @@
  *
  * Error Recovery Procedures (ERP).
  *
- * Copyright IBM Corp. 2002, 2010
+ * Copyright IBM Corp. 2002, 2015
  */
 
 #define KMSG_COMPONENT "zfcp"
@@ -1217,8 +1217,14 @@ static void zfcp_erp_action_cleanup(struct zfcp_erp_action *act, int result)
                break;
 
        case ZFCP_ERP_ACTION_REOPEN_PORT:
-               if (result == ZFCP_ERP_SUCCEEDED)
-                       zfcp_scsi_schedule_rport_register(port);
+               /* This switch case might also happen after a forced reopen
+                * was successfully done and thus overwritten with a new
+                * non-forced reopen at `ersfs_2'. In this case, we must not
+                * do the clean-up of the non-forced version.
+                */
+               if (act->step != ZFCP_ERP_STEP_UNINITIALIZED)
+                       if (result == ZFCP_ERP_SUCCEEDED)
+                               zfcp_scsi_schedule_rport_register(port);
                /* fall through */
        case ZFCP_ERP_ACTION_REOPEN_PORT_FORCED:
                put_device(&port->dev);
index b3c6ff49103b851f6467da7ff1fcb361865e2698..9069f98a18172e754c943010654de65e91c2fb7c 100644 (file)
@@ -3,7 +3,7 @@
  *
  * Interface to Linux SCSI midlayer.
  *
- * Copyright IBM Corp. 2002, 2013
+ * Copyright IBM Corp. 2002, 2015
  */
 
 #define KMSG_COMPONENT "zfcp"
@@ -556,6 +556,9 @@ static void zfcp_scsi_rport_register(struct zfcp_port *port)
        ids.port_id = port->d_id;
        ids.roles = FC_RPORT_ROLE_FCP_TARGET;
 
+       zfcp_dbf_rec_trig("scpaddy", port->adapter, port, NULL,
+                         ZFCP_PSEUDO_ERP_ACTION_RPORT_ADD,
+                         ZFCP_PSEUDO_ERP_ACTION_RPORT_ADD);
        rport = fc_remote_port_add(port->adapter->scsi_host, 0, &ids);
        if (!rport) {
                dev_err(&port->adapter->ccw_device->dev,
@@ -577,6 +580,9 @@ static void zfcp_scsi_rport_block(struct zfcp_port *port)
        struct fc_rport *rport = port->rport;
 
        if (rport) {
+               zfcp_dbf_rec_trig("scpdely", port->adapter, port, NULL,
+                                 ZFCP_PSEUDO_ERP_ACTION_RPORT_DEL,
+                                 ZFCP_PSEUDO_ERP_ACTION_RPORT_DEL);
                fc_remote_port_delete(rport);
                port->rport = NULL;
        }