zfcp_erp.c

来自「优龙2410linux2.6.8内核源代码」· C语言 代码 · 共 2,360 行 · 第 1/5 页

	/* action must be dequeued (here to allow for further ones) */
	zfcp_erp_action_dequeue(erp_action);

	/*
	 * put this target through the erp mill again if someone has
	 * requested to change the status of a target being online 
	 * to offline or the other way around
	 * (old retval is preserved if nothing has to be done here)
	 */
	retval = zfcp_erp_strategy_statechange(action, status, adapter,
					       port, unit, retval);

	/*
	 * leave if target is in permanent error state or if
	 * action is repeated in order to process state change
	 */
	if (retval == ZFCP_ERP_EXIT) {
		debug_text_event(adapter->erp_dbf, 2, "a_st_exit");
		goto unlock;
	}

	/* trigger follow up actions */
	zfcp_erp_strategy_followup_actions(action, adapter, port, unit, retval);

 unlock:
	write_unlock(&adapter->erp_lock);
	read_unlock_irqrestore(&zfcp_data.config_lock, flags);
	
	if (retval != ZFCP_ERP_CONTINUES)
		zfcp_erp_action_cleanup(action, adapter, port, unit, retval);

	/*
	 * a few tasks remain when the erp queues are empty
	 * (don't do that if the last action evaluated was dismissed
	 * since this clearly indicates that there is more to come) :
	 * - close the name server port if it is open yet
	 *   (enqueues another [probably] final action)
	 * - otherwise, wake up whoever wants to be woken when we are
	 *   done with erp
	 */
	if (retval != ZFCP_ERP_DISMISSED)
		zfcp_erp_strategy_check_queues(adapter);

	debug_text_event(adapter->erp_dbf, 6, "a_st_done");

	return retval;
}

/*
 * function:	
 *
 * purpose:	
 *
 * returns:	ZFCP_ERP_DISMISSED	- if action has been dismissed
 *		retval			- otherwise
 */
static int
zfcp_erp_strategy_check_action(struct zfcp_erp_action *erp_action, int retval)
{
	struct zfcp_adapter *adapter = erp_action->adapter;

	zfcp_erp_strategy_check_fsfreq(erp_action);

	debug_event(adapter->erp_dbf, 5, &erp_action->action, sizeof (int));
	if (erp_action->status & ZFCP_STATUS_ERP_DISMISSED) {
		debug_text_event(adapter->erp_dbf, 3, "a_stcd_dis");
		zfcp_erp_action_dequeue(erp_action);
		retval = ZFCP_ERP_DISMISSED;
	} else
		debug_text_event(adapter->erp_dbf, 5, "a_stcd_nodis");

	return retval;
}

/*
 * function:	
 *
 * purpose:	
 *
 * returns:
 */
static int
zfcp_erp_strategy_do_action(struct zfcp_erp_action *erp_action)
{
	int retval = ZFCP_ERP_FAILED;
	struct zfcp_adapter *adapter = erp_action->adapter;

	/*
	 * try to execute/continue action as far as possible,
	 * note: no lock in subsequent strategy routines
	 * (this allows these routine to call schedule, e.g.
	 * kmalloc with such flags or qdio_initialize & friends)
	 * Note: in case of timeout, the seperate strategies will fail
	 * anyhow. No need for a special action. Even worse, a nameserver
	 * failure would not wake up waiting ports without the call.
	 */
	switch (erp_action->action) {

	case ZFCP_ERP_ACTION_REOPEN_ADAPTER:
		retval = zfcp_erp_adapter_strategy(erp_action);
		break;

	case ZFCP_ERP_ACTION_REOPEN_PORT_FORCED:
		retval = zfcp_erp_port_forced_strategy(erp_action);
		break;

	case ZFCP_ERP_ACTION_REOPEN_PORT:
		retval = zfcp_erp_port_strategy(erp_action);
		break;

	case ZFCP_ERP_ACTION_REOPEN_UNIT:
		retval = zfcp_erp_unit_strategy(erp_action);
		break;

	default:
		debug_text_exception(adapter->erp_dbf, 1, "a_stda_bug");
		debug_event(adapter->erp_dbf, 1, &erp_action->action,
			    sizeof (int));
		ZFCP_LOG_NORMAL("bug: unknown erp action requested on "
				"adapter %s (action=%d)\n",
				zfcp_get_busid_by_adapter(erp_action->adapter),
				erp_action->action);
	}

	return retval;
}

/*
 * function:	
 *
 * purpose:	triggers retry of this action after a certain amount of time
 *		by means of timer provided by erp_action
 *
 * returns:	ZFCP_ERP_CONTINUES - erp_action sleeps in erp running queue
 */
static int
zfcp_erp_strategy_memwait(struct zfcp_erp_action *erp_action)
{
	int retval = ZFCP_ERP_CONTINUES;
	struct zfcp_adapter *adapter = erp_action->adapter;

	debug_text_event(adapter->erp_dbf, 6, "a_mwinit");
	debug_event(adapter->erp_dbf, 6, &erp_action->action, sizeof (int));
	init_timer(&erp_action->timer);
	erp_action->timer.function = zfcp_erp_memwait_handler;
	erp_action->timer.data = (unsigned long) erp_action;
	erp_action->timer.expires = jiffies + ZFCP_ERP_MEMWAIT_TIMEOUT;
	add_timer(&erp_action->timer);

	return retval;
}

/* 
 * function:    zfcp_erp_adapter_failed
 *
 * purpose:     sets the adapter and all underlying devices to ERP_FAILED
 *
 */
void
zfcp_erp_adapter_failed(struct zfcp_adapter *adapter)
{
	zfcp_erp_modify_adapter_status(adapter,
				       ZFCP_STATUS_COMMON_ERP_FAILED, ZFCP_SET);
	ZFCP_LOG_NORMAL("adapter erp failed on adapter %s\n",
			zfcp_get_busid_by_adapter(adapter));
	debug_text_event(adapter->erp_dbf, 2, "a_afail");
}

/* 
 * function:    zfcp_erp_port_failed
 *
 * purpose:     sets the port and all underlying devices to ERP_FAILED
 *
 */
void
zfcp_erp_port_failed(struct zfcp_port *port)
{
	zfcp_erp_modify_port_status(port,
				    ZFCP_STATUS_COMMON_ERP_FAILED, ZFCP_SET);

	ZFCP_LOG_NORMAL("port erp failed on port 0x%016Lx on adapter %s\n",
			port->wwpn, zfcp_get_busid_by_port(port));
	debug_text_event(port->adapter->erp_dbf, 2, "p_pfail");
	debug_event(port->adapter->erp_dbf, 2, &port->wwpn, sizeof (wwn_t));
}

/* 
 * function:    zfcp_erp_unit_failed
 *
 * purpose:     sets the unit to ERP_FAILED
 *
 */
void
zfcp_erp_unit_failed(struct zfcp_unit *unit)
{
	zfcp_erp_modify_unit_status(unit,
				    ZFCP_STATUS_COMMON_ERP_FAILED, ZFCP_SET);

	ZFCP_LOG_NORMAL("unit erp failed on unit 0x%016Lx on port 0x%016Lx "
			" on adapter %s\n", unit->fcp_lun,
			unit->port->wwpn, zfcp_get_busid_by_unit(unit));
	debug_text_event(unit->port->adapter->erp_dbf, 2, "u_ufail");
	debug_event(unit->port->adapter->erp_dbf, 2,
		    &unit->fcp_lun, sizeof (fcp_lun_t));
}

/*
 * function:	zfcp_erp_strategy_check_target
 *
 * purpose:	increments the erp action count on the device currently in
 *              recovery if the action failed or resets the count in case of
 *              success. If a maximum count is exceeded the device is marked
 *              as ERP_FAILED.
 *		The 'blocked' state of a target which has been recovered
 *              successfully is reset.
 *
 * returns:	ZFCP_ERP_CONTINUES	- action continues (not considered)
 *		ZFCP_ERP_SUCCEEDED	- action finished successfully 
 *		ZFCP_ERP_EXIT		- action failed and will not continue
 */
static int
zfcp_erp_strategy_check_target(struct zfcp_erp_action *erp_action, int result)
{
	struct zfcp_adapter *adapter = erp_action->adapter;
	struct zfcp_port *port = erp_action->port;
	struct zfcp_unit *unit = erp_action->unit;

	debug_text_event(adapter->erp_dbf, 5, "a_stct_norm");
	debug_event(adapter->erp_dbf, 5, &erp_action->action, sizeof (int));
	debug_event(adapter->erp_dbf, 5, &result, sizeof (int));

	switch (erp_action->action) {

	case ZFCP_ERP_ACTION_REOPEN_UNIT:
		result = zfcp_erp_strategy_check_unit(unit, result);
		break;

	case ZFCP_ERP_ACTION_REOPEN_PORT_FORCED:
	case ZFCP_ERP_ACTION_REOPEN_PORT:
		result = zfcp_erp_strategy_check_port(port, result);
		break;

	case ZFCP_ERP_ACTION_REOPEN_ADAPTER:
		result = zfcp_erp_strategy_check_adapter(adapter, result);
		break;
	}

	return result;
}

/*
 * function:	
 *
 * purpose:	
 *
 * returns:
 */
static int
zfcp_erp_strategy_statechange(int action,
			      u32 status,
			      struct zfcp_adapter *adapter,
			      struct zfcp_port *port,
			      struct zfcp_unit *unit, int retval)
{
	debug_text_event(adapter->erp_dbf, 3, "a_stsc");
	debug_event(adapter->erp_dbf, 3, &action, sizeof (int));

	switch (action) {

	case ZFCP_ERP_ACTION_REOPEN_ADAPTER:
		if (zfcp_erp_strategy_statechange_detected(&adapter->status,
							   status)) {
			zfcp_erp_adapter_reopen_internal(adapter, ZFCP_STATUS_COMMON_ERP_FAILED);
			retval = ZFCP_ERP_EXIT;
		}
		break;

	case ZFCP_ERP_ACTION_REOPEN_PORT_FORCED:
	case ZFCP_ERP_ACTION_REOPEN_PORT:
		if (zfcp_erp_strategy_statechange_detected(&port->status,
							   status)) {
			zfcp_erp_port_reopen_internal(port, ZFCP_STATUS_COMMON_ERP_FAILED);
			retval = ZFCP_ERP_EXIT;
		}
		break;

	case ZFCP_ERP_ACTION_REOPEN_UNIT:
		if (zfcp_erp_strategy_statechange_detected(&unit->status,
							   status)) {
			zfcp_erp_unit_reopen_internal(unit, ZFCP_STATUS_COMMON_ERP_FAILED);
			retval = ZFCP_ERP_EXIT;
		}
		break;
	}

	return retval;
}

/*
 * function:	
 *
 * purpose:	
 *
 * returns:
 */
static inline int
zfcp_erp_strategy_statechange_detected(atomic_t * target_status, u32 erp_status)
{
	return
	    /* take it online */
	    (atomic_test_mask(ZFCP_STATUS_COMMON_RUNNING, target_status) &&
	     (ZFCP_STATUS_ERP_CLOSE_ONLY & erp_status)) ||
	    /* take it offline */
	    (!atomic_test_mask(ZFCP_STATUS_COMMON_RUNNING, target_status) &&
	     !(ZFCP_STATUS_ERP_CLOSE_ONLY & erp_status));
}

/**
 * zfcp_erp_scsi_add_device
 * @data: pointer to a struct zfcp_unit
 *
 * Registers a logical unit with the SCSI stack.
 */
static void
zfcp_erp_scsi_add_device(void *data)
{
	struct {
		struct zfcp_unit  *unit;
		struct work_struct work;
	} *p;

	p = data;
	scsi_add_device(p->unit->port->adapter->scsi_host,
			0, p->unit->port->scsi_id, p->unit->scsi_lun);
	atomic_set(&p->unit->scsi_add_work, 0);
	wake_up(&p->unit->scsi_add_wq);
	zfcp_unit_put(p->unit);
	kfree(p);
}

/**
 * zfcp_erp_schedule_work
 * @unit: pointer to unit which should be registered with SCSI stack
 *
 * Schedules work which registers a unit with the SCSI stack
 */
static int
zfcp_erp_schedule_work(struct zfcp_unit *unit)
{
	struct {
		struct zfcp_unit * unit;
		struct work_struct work;
	} *p;

	if (atomic_compare_and_swap(0, 1, &unit->scsi_add_work))
		return 0;

	if ((p = kmalloc(sizeof(*p), GFP_KERNEL)) == NULL) {
		ZFCP_LOG_NORMAL("error: registration at SCSI stack failed for "
				"unit 0x%016Lx on port 0x%016Lx on "
				"adapter %s\n", unit->fcp_lun, unit->port->wwpn,
				zfcp_get_busid_by_unit(unit));
		atomic_set(&unit->scsi_add_work, 0);
		return -ENOMEM;
	}

	zfcp_unit_get(unit);
	memset(p, 0, sizeof(*p));
	INIT_WORK(&p->work, zfcp_erp_scsi_add_device, p);
	p->unit = unit;
	schedule_work(&p->work);
	return 0;
}

/*
 * function:	
 *
 * purpose:	
 *
 * returns:
 */
static int
zfcp_erp_strategy_check_unit(struct zfcp_unit *unit, int result)
{
	debug_text_event(unit->port->adapter->erp_dbf, 5, "u_stct");
	debug_event(unit->port->adapter->erp_dbf, 5, &unit->fcp_lun,
		    sizeof (fcp_lun_t));

	switch (result) {
	case ZFCP_ERP_SUCCEEDED :
		atomic_set(&unit->erp_counter, 0);
		zfcp_erp_unit_unblock(unit);
		break;
	case ZFCP_ERP_FAILED :
		atomic_inc(&unit->erp_counter);
		if (atomic_read(&unit->erp_counter) > ZFCP_MAX_ERPS)
			zfcp_erp_unit_failed(unit);
		break;
	case ZFCP_ERP_EXIT :
		/* nothing */
		break;
	}

	if (atomic_test_mask(ZFCP_STATUS_COMMON_ERP_FAILED, &unit->status)) {
		zfcp_erp_unit_block(unit, 0); /* for ZFCP_ERP_SUCCEEDED */
		result = ZFCP_ERP_EXIT;
	}

	return result;
}

/*
 * function:	
 *
 * purpose:	
 *
 * returns:
 */
static int
zfcp_erp_strategy_check_port(struct zfcp_port *port, int result)
{
	debug_text_event(port->adapter->erp_dbf, 5, "p_stct");
	debug_event(port->adapter->erp_dbf, 5, &port->wwpn, sizeof (wwn_t));

	switch (result) {
	case ZFCP_ERP_SUCCEEDED :
		atomic_set(&port->erp_counter, 0);
		zfcp_erp_port_unblock(port);
		break;
	case ZFCP_ERP_FAILED :
		atomic_inc(&port->erp_counter);
		if (atomic_read(&port->erp_counter) > ZFCP_MAX_ERPS)
			zfcp_erp_port_failed(port);
		break;
	case ZFCP_ERP_EXIT :
		/* nothing */
		break;
	}

	if (atomic_test_mask(ZFCP_STATUS_COMMON_ERP_FAILED, &port->status)) {
		zfcp_erp_port_block(port, 0); /* for ZFCP_ERP_SUCCEEDED */
		result = ZFCP_ERP_EXIT;
	}

	return result;
}

/*
 * function:	
 *
 * purpose:	
 *
 * returns:
 */
static int
zfcp_erp_strategy_check_adapter(struct zfcp_adapter *adapter, int result)
{
	debug_text_event(adapter->erp_dbf, 5, "a_stct");

	switch (result) {
	case ZFCP_ERP_SUCCEEDED :
		atomic_set(&adapter->erp_counter, 0);
		zfcp_erp_adapter_unblock(adapter);
		break;
	case ZFCP_ERP_FAILED :
		atomic_inc(&adapter->erp_counter);
		if (atomic_read(&adapter->erp_counter) > ZFCP_MAX_ERPS)
			zfcp_erp_adapter_failed(adapter);
		break;
	case ZFCP_ERP_EXIT :
		/* nothing */
		break;