zfcp_erp.c

linux-2.6.15.6

			}
		} else {
			debug_text_event(adapter->erp_dbf, 3, "a_ca_gonereq");
			/*
			 * even if this fsf_req has gone, forget about
			 * association between erp_action and fsf_req
			 */
			erp_action->fsf_req = NULL;
		}
		spin_unlock(&adapter->fsf_req_list_lock);
	} else
		debug_text_event(adapter->erp_dbf, 3, "a_ca_noreq");

	return retval;
}

/*
 * purpose:	generic handler for asynchronous events related to erp_action events
 *		(normal completion, time-out, dismissing, retry after
 *		low memory condition)
 *
 * note:	deletion of timer is not required (e.g. in case of a time-out),
 *		but a second try does no harm,
 *		we leave it in here to allow for greater simplification
 *
 * returns:	0 - there was an action to handle
 *		!0 - otherwise
 */
static int
zfcp_erp_async_handler_nolock(struct zfcp_erp_action *erp_action,
			      unsigned long set_mask)
{
	int retval;
	struct zfcp_adapter *adapter = erp_action->adapter;

	if (zfcp_erp_action_exists(erp_action) == ZFCP_ERP_ACTION_RUNNING) {
		debug_text_event(adapter->erp_dbf, 2, "a_asyh_ex");
		debug_event(adapter->erp_dbf, 2, &erp_action->action,
			    sizeof (int));
		if (!(set_mask & ZFCP_STATUS_ERP_TIMEDOUT))
			del_timer(&erp_action->timer);
		erp_action->status |= set_mask;
		zfcp_erp_action_ready(erp_action);
		retval = 0;
	} else {
		/* action is ready or gone - nothing to do */
		debug_text_event(adapter->erp_dbf, 3, "a_asyh_gone");
		debug_event(adapter->erp_dbf, 3, &erp_action->action,
			    sizeof (int));
		retval = 1;
	}

	return retval;
}

/*
 * purpose:	generic handler for asynchronous events related to erp_action
 *               events	(normal completion, time-out, dismissing, retry after
 *		low memory condition)
 *
 * note:	deletion of timer is not required (e.g. in case of a time-out),
 *		but a second try does no harm,
 *		we leave it in here to allow for greater simplification
 *
 * returns:	0 - there was an action to handle
 *		!0 - otherwise
 */
int
zfcp_erp_async_handler(struct zfcp_erp_action *erp_action,
		       unsigned long set_mask)
{
	struct zfcp_adapter *adapter = erp_action->adapter;
	unsigned long flags;
	int retval;

	write_lock_irqsave(&adapter->erp_lock, flags);
	retval = zfcp_erp_async_handler_nolock(erp_action, set_mask);
	write_unlock_irqrestore(&adapter->erp_lock, flags);

	return retval;
}

/*
 * purpose:	is called for erp_action which was slept waiting for
 *		memory becoming avaliable,
 *		will trigger that this action will be continued
 */
static void
zfcp_erp_memwait_handler(unsigned long data)
{
	struct zfcp_erp_action *erp_action = (struct zfcp_erp_action *) data;
	struct zfcp_adapter *adapter = erp_action->adapter;

	debug_text_event(adapter->erp_dbf, 2, "a_mwh");
	debug_event(adapter->erp_dbf, 2, &erp_action->action, sizeof (int));

	zfcp_erp_async_handler(erp_action, 0);
}

/*
 * purpose:	is called if an asynchronous erp step timed out,
 *		action gets an appropriate flag and will be processed
 *		accordingly
 */
static void
zfcp_erp_timeout_handler(unsigned long data)
{
	struct zfcp_erp_action *erp_action = (struct zfcp_erp_action *) data;
	struct zfcp_adapter *adapter = erp_action->adapter;

	debug_text_event(adapter->erp_dbf, 2, "a_th");
	debug_event(adapter->erp_dbf, 2, &erp_action->action, sizeof (int));

	zfcp_erp_async_handler(erp_action, ZFCP_STATUS_ERP_TIMEDOUT);
}

/*
 * purpose:	is called for an erp_action which needs to be ended
 *		though not being done,
 *		this is usually required if an higher is generated,
 *		action gets an appropriate flag and will be processed
 *		accordingly
 *
 * locks:	erp_lock held (thus we need to call another handler variant)
 */
static int
zfcp_erp_action_dismiss(struct zfcp_erp_action *erp_action)
{
	struct zfcp_adapter *adapter = erp_action->adapter;

	debug_text_event(adapter->erp_dbf, 2, "a_adis");
	debug_event(adapter->erp_dbf, 2, &erp_action->action, sizeof (int));

	zfcp_erp_async_handler_nolock(erp_action, ZFCP_STATUS_ERP_DISMISSED);

	return 0;
}

int
zfcp_erp_thread_setup(struct zfcp_adapter *adapter)
{
	int retval = 0;

	atomic_clear_mask(ZFCP_STATUS_ADAPTER_ERP_THREAD_UP, &adapter->status);

	retval = kernel_thread(zfcp_erp_thread, adapter, SIGCHLD);
	if (retval < 0) {
		ZFCP_LOG_NORMAL("error: creation of erp thread failed for "
				"adapter %s\n",
				zfcp_get_busid_by_adapter(adapter));
		debug_text_event(adapter->erp_dbf, 5, "a_thset_fail");
	} else {
		wait_event(adapter->erp_thread_wqh,
			   atomic_test_mask(ZFCP_STATUS_ADAPTER_ERP_THREAD_UP,
					    &adapter->status));
		debug_text_event(adapter->erp_dbf, 5, "a_thset_ok");
	}

	return (retval < 0);
}

/*
 * function:	
 *
 * purpose:	
 *
 * returns:
 *
 * context:	process (i.e. proc-fs or rmmod/insmod)
 *
 * note:	The caller of this routine ensures that the specified
 *		adapter has been shut down and that this operation
 *		has been completed. Thus, there are no pending erp_actions
 *		which would need to be handled here.
 */
int
zfcp_erp_thread_kill(struct zfcp_adapter *adapter)
{
	int retval = 0;

	atomic_set_mask(ZFCP_STATUS_ADAPTER_ERP_THREAD_KILL, &adapter->status);
	up(&adapter->erp_ready_sem);

	wait_event(adapter->erp_thread_wqh,
		   !atomic_test_mask(ZFCP_STATUS_ADAPTER_ERP_THREAD_UP,
				     &adapter->status));

	atomic_clear_mask(ZFCP_STATUS_ADAPTER_ERP_THREAD_KILL,
			  &adapter->status);

	debug_text_event(adapter->erp_dbf, 5, "a_thki_ok");

	return retval;
}

/*
 * purpose:	is run as a kernel thread,
 *		goes through list of error recovery actions of associated adapter
 *		and delegates single action to execution
 *
 * returns:	0
 */
static int
zfcp_erp_thread(void *data)
{
	struct zfcp_adapter *adapter = (struct zfcp_adapter *) data;
	struct list_head *next;
	struct zfcp_erp_action *erp_action;
	unsigned long flags;

	daemonize("zfcperp%s", zfcp_get_busid_by_adapter(adapter));
	/* Block all signals */
	siginitsetinv(&current->blocked, 0);
	atomic_set_mask(ZFCP_STATUS_ADAPTER_ERP_THREAD_UP, &adapter->status);
	debug_text_event(adapter->erp_dbf, 5, "a_th_run");
	wake_up(&adapter->erp_thread_wqh);

	while (!atomic_test_mask(ZFCP_STATUS_ADAPTER_ERP_THREAD_KILL,
				 &adapter->status)) {

		write_lock_irqsave(&adapter->erp_lock, flags);
		next = adapter->erp_ready_head.prev;
		write_unlock_irqrestore(&adapter->erp_lock, flags);

		if (next != &adapter->erp_ready_head) {
			erp_action =
			    list_entry(next, struct zfcp_erp_action, list);
			/*
			 * process action (incl. [re]moving it
			 * from 'ready' queue)
			 */
			zfcp_erp_strategy(erp_action);
		}

		/*
		 * sleep as long as there is nothing to do, i.e.
		 * no action in 'ready' queue to be processed and
		 * thread is not to be killed
		 */
		down_interruptible(&adapter->erp_ready_sem);
		debug_text_event(adapter->erp_dbf, 5, "a_th_woken");
	}

	atomic_clear_mask(ZFCP_STATUS_ADAPTER_ERP_THREAD_UP, &adapter->status);
	debug_text_event(adapter->erp_dbf, 5, "a_th_stop");
	wake_up(&adapter->erp_thread_wqh);

	return 0;
}

/*
 * function:	
 *
 * purpose:	drives single error recovery action and schedules higher and
 *		subordinate actions, if necessary
 *
 * returns:	ZFCP_ERP_CONTINUES	- action continues (asynchronously)
 *		ZFCP_ERP_SUCCEEDED	- action finished successfully (deqd)
 *		ZFCP_ERP_FAILED		- action finished unsuccessfully (deqd)
 *		ZFCP_ERP_EXIT		- action finished (dequeued), offline
 *		ZFCP_ERP_DISMISSED	- action canceled (dequeued)
 */
static int
zfcp_erp_strategy(struct zfcp_erp_action *erp_action)
{
	int retval = 0;
	struct zfcp_adapter *adapter = erp_action->adapter;
	struct zfcp_port *port = erp_action->port;
	struct zfcp_unit *unit = erp_action->unit;
	int action = erp_action->action;
	u32 status = erp_action->status;
	unsigned long flags;

	/* serialise dismissing, timing out, moving, enqueueing */
	read_lock_irqsave(&zfcp_data.config_lock, flags);
	write_lock(&adapter->erp_lock);

	/* dequeue dismissed action and leave, if required */
	retval = zfcp_erp_strategy_check_action(erp_action, retval);
	if (retval == ZFCP_ERP_DISMISSED) {
		debug_text_event(adapter->erp_dbf, 4, "a_st_dis1");
		goto unlock;
	}

	/*
	 * move action to 'running' queue before processing it
	 * (to avoid a race condition regarding moving the
	 * action to the 'running' queue and back)
	 */
	zfcp_erp_action_to_running(erp_action);

	/*
	 * try to process action as far as possible,
	 * no lock to allow for blocking operations (kmalloc, qdio, ...),
	 * afterwards the lock is required again for the following reasons:
	 * - dequeueing of finished action and enqueueing of
	 *   follow-up actions must be atomic so that any other
	 *   reopen-routine does not believe there is nothing to do
	 *   and that it is safe to enqueue something else,
	 * - we want to force any control thread which is dismissing
	 *   actions to finish this before we decide about
	 *   necessary steps to be taken here further
	 */
	write_unlock(&adapter->erp_lock);
	read_unlock_irqrestore(&zfcp_data.config_lock, flags);
	retval = zfcp_erp_strategy_do_action(erp_action);
	read_lock_irqsave(&zfcp_data.config_lock, flags);
	write_lock(&adapter->erp_lock);

	/*
	 * check for dismissed status again to avoid follow-up actions,
	 * failing of targets and so on for dismissed actions
	 */
	retval = zfcp_erp_strategy_check_action(erp_action, retval);

	switch (retval) {
	case ZFCP_ERP_DISMISSED:
		/* leave since this action has ridden to its ancestors */
		debug_text_event(adapter->erp_dbf, 6, "a_st_dis2");
		goto unlock;
	case ZFCP_ERP_NOMEM:
		/* no memory to continue immediately, let it sleep */
		if (!(erp_action->status & ZFCP_STATUS_ERP_LOWMEM)) {
			++adapter->erp_low_mem_count;
			erp_action->status |= ZFCP_STATUS_ERP_LOWMEM;
		}
		/* This condition is true if there is no memory available
		   for any erp_action on this adapter. This implies that there
		   are no elements in the memory pool(s) left for erp_actions.
		   This might happen if an erp_action that used a memory pool
		   element was timed out.
		 */
		if (adapter->erp_total_count == adapter->erp_low_mem_count) {
			debug_text_event(adapter->erp_dbf, 3, "a_st_lowmem");
			ZFCP_LOG_NORMAL("error: no mempool elements available, "
					"restarting I/O on adapter %s "
					"to free mempool\n",
					zfcp_get_busid_by_adapter(adapter));
			zfcp_erp_adapter_reopen_internal(adapter, 0);
		} else {
		debug_text_event(adapter->erp_dbf, 2, "a_st_memw");
		retval = zfcp_erp_strategy_memwait(erp_action);
		}
		goto unlock;
	case ZFCP_ERP_CONTINUES:
		/* leave since this action runs asynchronously */
		debug_text_event(adapter->erp_dbf, 6, "a_st_cont");
		if (erp_action->status & ZFCP_STATUS_ERP_LOWMEM) {
			--adapter->erp_low_mem_count;
			erp_action->status &= ~ZFCP_STATUS_ERP_LOWMEM;
		}
		goto unlock;
	}
	/* ok, finished action (whatever its result is) */

	/* check for unrecoverable targets */
	retval = zfcp_erp_strategy_check_target(erp_action, retval);

	/* 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;