@@ -4728,19 +4728,24 @@ static int ibmvfc_alloc_target(struct ibmvfc_host *vhost,
* and it failed for some reason, such as there being I/O
* pending to the target. In this case, we will have already
* deleted the rport from the FC transport so we do a move
- * login, which works even with I/O pending, as it will cancel
- * any active commands.
+ * login, which works even with I/O pending, however, if
+ * there is still I/O pending, it will stay outstanding, so
+ * we only do this if fast fail is disabled for the rport,
+ * otherwise we let terminate_rport_io clean up the port
+ * before we login at the new location.
*/
if (wtgt->action == IBMVFC_TGT_ACTION_LOGOUT_DELETED_RPORT) {
- /*
- * Do a move login here. The old target is no longer
- * known to the transport layer We don't use the
- * normal ibmvfc_set_tgt_action to set this, as we
- * don't normally want to allow this state change.
- */
- wtgt->new_scsi_id = scsi_id;
- wtgt->action = IBMVFC_TGT_ACTION_INIT;
- ibmvfc_init_tgt(wtgt, ibmvfc_tgt_move_login);
+ if (wtgt->move_login) {
+ /*
+ * Do a move login here. The old target is no longer
+ * known to the transport layer We don't use the
+ * normal ibmvfc_set_tgt_action to set this, as we
+ * don't normally want to allow this state change.
+ */
+ wtgt->new_scsi_id = scsi_id;
+ wtgt->action = IBMVFC_TGT_ACTION_INIT;
+ ibmvfc_init_tgt(wtgt, ibmvfc_tgt_move_login);
+ }
goto unlock_out;
} else {
tgt_err(wtgt, "Unexpected target state: %d, %p\n",
@@ -5486,6 +5491,18 @@ static void ibmvfc_do_work(struct ibmvfc_host *vhost)
rport = tgt->rport;
tgt->rport = NULL;
ibmvfc_set_tgt_action(tgt, IBMVFC_TGT_ACTION_LOGOUT_DELETED_RPORT);
+
+ /*
+ * If fast fail is enabled, we wait for it to fire and then clean up
+ * the old port, since we expect the fast fail timer to clean up the
+ * outstanding I/O faster than waiting for normal command timeouts.
+ * However, if fast fail is disabled, any I/O outstanding to the
+ * rport LUNs will stay outstanding indefinitely, since the EH handlers
+ * won't get invoked for I/O's timing out. If this is a NPIV failover
+ * scenario, the better alternative is to use the move login.
+ */
+ if (rport && rport->fast_io_fail_tmo == -1)
+ tgt->move_login = 1;
spin_unlock_irqrestore(vhost->host->host_lock, flags);
if (rport)
fc_remote_port_delete(rport);
@@ -726,6 +726,7 @@ struct ibmvfc_target {
int add_rport;
int init_retries;
int logo_rcvd;
+ int move_login;
u32 cancel_key;
struct ibmvfc_service_parms service_parms;
struct ibmvfc_service_parms service_parms_change;
If fast fail is enabled and we encounter a WWPN moving from one port id to another port id with I/O outstanding, if we use the move login MAD, although it will work, it will leave any outstanding I/O still outstanding to the old port id. Eventually, the SCSI command timers will fire and we will abort these commands, however, this is generally much longer than the fast fail timeout, which can lead to I/O operations being outstanding for a long time. This patch changes the behavior to avoid the move login if fast fail is enabled. Once terminate_rport_io cleans up the rport, then we force the target back through the delete process, which re-drives the implicit logout, then kicks us back into discovery where we will discover the WWPN at the new location and do a PLOGI to it. Signed-off-by: Brian King <brking@linux.vnet.ibm.com> --- drivers/scsi/ibmvscsi/ibmvfc.c | 39 ++++++++++++++++++++++++++++----------- drivers/scsi/ibmvscsi/ibmvfc.h | 1 + 2 files changed, 29 insertions(+), 11 deletions(-)