multicast.c

来自「linux 内核源代码」· C语言 代码 · 共 837 行 · 第 1/2 页

		if (group->state == MCAST_ERROR) {
			spin_unlock_irq(&group->lock);
			process_group_error(group);
			goto retest;
		}

		member = list_entry(group->pending_list.next,
				    struct mcast_member, list);
		multicast = &member->multicast;
		join_state = multicast->rec.join_state;
		atomic_inc(&member->refcount);

		if (join_state == (group->rec.join_state & join_state)) {
			status = cmp_rec(&group->rec, &multicast->rec,
					 multicast->comp_mask);
			if (!status)
				join_group(group, member, join_state);
			else
				list_del_init(&member->list);
			spin_unlock_irq(&group->lock);
			ret = multicast->callback(status, multicast);
		} else {
			spin_unlock_irq(&group->lock);
			status = send_join(group, member);
			if (!status) {
				deref_member(member);
				return;
			}
			ret = fail_join(group, member, status);
		}

		deref_member(member);
		if (ret)
			ib_sa_free_multicast(&member->multicast);
		spin_lock_irq(&group->lock);
	}

	join_state = get_leave_state(group);
	if (join_state) {
		group->rec.join_state &= ~join_state;
		spin_unlock_irq(&group->lock);
		if (send_leave(group, join_state))
			goto retest;
	} else {
		group->state = MCAST_IDLE;
		spin_unlock_irq(&group->lock);
		release_group(group);
	}
}

/*
 * Fail a join request if it is still active - at the head of the pending queue.
 */
static void process_join_error(struct mcast_group *group, int status)
{
	struct mcast_member *member;
	int ret;

	spin_lock_irq(&group->lock);
	member = list_entry(group->pending_list.next,
			    struct mcast_member, list);
	if (group->last_join == member) {
		atomic_inc(&member->refcount);
		list_del_init(&member->list);
		spin_unlock_irq(&group->lock);
		ret = member->multicast.callback(status, &member->multicast);
		deref_member(member);
		if (ret)
			ib_sa_free_multicast(&member->multicast);
	} else
		spin_unlock_irq(&group->lock);
}

static void join_handler(int status, struct ib_sa_mcmember_rec *rec,
			 void *context)
{
	struct mcast_group *group = context;

	if (status)
		process_join_error(group, status);
	else {
		spin_lock_irq(&group->port->lock);
		group->rec = *rec;
		if (!memcmp(&mgid0, &group->rec.mgid, sizeof mgid0)) {
			rb_erase(&group->node, &group->port->table);
			mcast_insert(group->port, group, 1);
		}
		spin_unlock_irq(&group->port->lock);
	}
	mcast_work_handler(&group->work);
}

static void leave_handler(int status, struct ib_sa_mcmember_rec *rec,
			  void *context)
{
	struct mcast_group *group = context;

	mcast_work_handler(&group->work);
}

static struct mcast_group *acquire_group(struct mcast_port *port,
					 union ib_gid *mgid, gfp_t gfp_mask)
{
	struct mcast_group *group, *cur_group;
	unsigned long flags;
	int is_mgid0;

	is_mgid0 = !memcmp(&mgid0, mgid, sizeof mgid0);
	if (!is_mgid0) {
		spin_lock_irqsave(&port->lock, flags);
		group = mcast_find(port, mgid);
		if (group)
			goto found;
		spin_unlock_irqrestore(&port->lock, flags);
	}

	group = kzalloc(sizeof *group, gfp_mask);
	if (!group)
		return NULL;

	group->port = port;
	group->rec.mgid = *mgid;
	INIT_LIST_HEAD(&group->pending_list);
	INIT_LIST_HEAD(&group->active_list);
	INIT_WORK(&group->work, mcast_work_handler);
	spin_lock_init(&group->lock);

	spin_lock_irqsave(&port->lock, flags);
	cur_group = mcast_insert(port, group, is_mgid0);
	if (cur_group) {
		kfree(group);
		group = cur_group;
	} else
		atomic_inc(&port->refcount);
found:
	atomic_inc(&group->refcount);
	spin_unlock_irqrestore(&port->lock, flags);
	return group;
}

/*
 * We serialize all join requests to a single group to make our lives much
 * easier.  Otherwise, two users could try to join the same group
 * simultaneously, with different configurations, one could leave while the
 * join is in progress, etc., which makes locking around error recovery
 * difficult.
 */
struct ib_sa_multicast *
ib_sa_join_multicast(struct ib_sa_client *client,
		     struct ib_device *device, u8 port_num,
		     struct ib_sa_mcmember_rec *rec,
		     ib_sa_comp_mask comp_mask, gfp_t gfp_mask,
		     int (*callback)(int status,
				     struct ib_sa_multicast *multicast),
		     void *context)
{
	struct mcast_device *dev;
	struct mcast_member *member;
	struct ib_sa_multicast *multicast;
	int ret;

	dev = ib_get_client_data(device, &mcast_client);
	if (!dev)
		return ERR_PTR(-ENODEV);

	member = kmalloc(sizeof *member, gfp_mask);
	if (!member)
		return ERR_PTR(-ENOMEM);

	ib_sa_client_get(client);
	member->client = client;
	member->multicast.rec = *rec;
	member->multicast.comp_mask = comp_mask;
	member->multicast.callback = callback;
	member->multicast.context = context;
	init_completion(&member->comp);
	atomic_set(&member->refcount, 1);
	member->state = MCAST_JOINING;

	member->group = acquire_group(&dev->port[port_num - dev->start_port],
				      &rec->mgid, gfp_mask);
	if (!member->group) {
		ret = -ENOMEM;
		goto err;
	}

	/*
	 * The user will get the multicast structure in their callback.  They
	 * could then free the multicast structure before we can return from
	 * this routine.  So we save the pointer to return before queuing
	 * any callback.
	 */
	multicast = &member->multicast;
	queue_join(member);
	return multicast;

err:
	ib_sa_client_put(client);
	kfree(member);
	return ERR_PTR(ret);
}
EXPORT_SYMBOL(ib_sa_join_multicast);

void ib_sa_free_multicast(struct ib_sa_multicast *multicast)
{
	struct mcast_member *member;
	struct mcast_group *group;

	member = container_of(multicast, struct mcast_member, multicast);
	group = member->group;

	spin_lock_irq(&group->lock);
	if (member->state == MCAST_MEMBER)
		adjust_membership(group, multicast->rec.join_state, -1);

	list_del_init(&member->list);

	if (group->state == MCAST_IDLE) {
		group->state = MCAST_BUSY;
		spin_unlock_irq(&group->lock);
		/* Continue to hold reference on group until callback */
		queue_work(mcast_wq, &group->work);
	} else {
		spin_unlock_irq(&group->lock);
		release_group(group);
	}

	deref_member(member);
	wait_for_completion(&member->comp);
	ib_sa_client_put(member->client);
	kfree(member);
}
EXPORT_SYMBOL(ib_sa_free_multicast);

int ib_sa_get_mcmember_rec(struct ib_device *device, u8 port_num,
			   union ib_gid *mgid, struct ib_sa_mcmember_rec *rec)
{
	struct mcast_device *dev;
	struct mcast_port *port;
	struct mcast_group *group;
	unsigned long flags;
	int ret = 0;

	dev = ib_get_client_data(device, &mcast_client);
	if (!dev)
		return -ENODEV;

	port = &dev->port[port_num - dev->start_port];
	spin_lock_irqsave(&port->lock, flags);
	group = mcast_find(port, mgid);
	if (group)
		*rec = group->rec;
	else
		ret = -EADDRNOTAVAIL;
	spin_unlock_irqrestore(&port->lock, flags);

	return ret;
}
EXPORT_SYMBOL(ib_sa_get_mcmember_rec);

int ib_init_ah_from_mcmember(struct ib_device *device, u8 port_num,
			     struct ib_sa_mcmember_rec *rec,
			     struct ib_ah_attr *ah_attr)
{
	int ret;
	u16 gid_index;
	u8 p;

	ret = ib_find_cached_gid(device, &rec->port_gid, &p, &gid_index);
	if (ret)
		return ret;

	memset(ah_attr, 0, sizeof *ah_attr);
	ah_attr->dlid = be16_to_cpu(rec->mlid);
	ah_attr->sl = rec->sl;
	ah_attr->port_num = port_num;
	ah_attr->static_rate = rec->rate;

	ah_attr->ah_flags = IB_AH_GRH;
	ah_attr->grh.dgid = rec->mgid;

	ah_attr->grh.sgid_index = (u8) gid_index;
	ah_attr->grh.flow_label = be32_to_cpu(rec->flow_label);
	ah_attr->grh.hop_limit = rec->hop_limit;
	ah_attr->grh.traffic_class = rec->traffic_class;

	return 0;
}
EXPORT_SYMBOL(ib_init_ah_from_mcmember);

static void mcast_groups_lost(struct mcast_port *port)
{
	struct mcast_group *group;
	struct rb_node *node;
	unsigned long flags;

	spin_lock_irqsave(&port->lock, flags);
	for (node = rb_first(&port->table); node; node = rb_next(node)) {
		group = rb_entry(node, struct mcast_group, node);
		spin_lock(&group->lock);
		if (group->state == MCAST_IDLE) {
			atomic_inc(&group->refcount);
			queue_work(mcast_wq, &group->work);
		}
		group->state = MCAST_ERROR;
		spin_unlock(&group->lock);
	}
	spin_unlock_irqrestore(&port->lock, flags);
}

static void mcast_event_handler(struct ib_event_handler *handler,
				struct ib_event *event)
{
	struct mcast_device *dev;

	dev = container_of(handler, struct mcast_device, event_handler);

	switch (event->event) {
	case IB_EVENT_PORT_ERR:
	case IB_EVENT_LID_CHANGE:
	case IB_EVENT_SM_CHANGE:
	case IB_EVENT_CLIENT_REREGISTER:
		mcast_groups_lost(&dev->port[event->element.port_num -
					     dev->start_port]);
		break;
	default:
		break;
	}
}

static void mcast_add_one(struct ib_device *device)
{
	struct mcast_device *dev;
	struct mcast_port *port;
	int i;

	if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
		return;

	dev = kmalloc(sizeof *dev + device->phys_port_cnt * sizeof *port,
		      GFP_KERNEL);
	if (!dev)
		return;

	if (device->node_type == RDMA_NODE_IB_SWITCH)
		dev->start_port = dev->end_port = 0;
	else {
		dev->start_port = 1;
		dev->end_port = device->phys_port_cnt;
	}

	for (i = 0; i <= dev->end_port - dev->start_port; i++) {
		port = &dev->port[i];
		port->dev = dev;
		port->port_num = dev->start_port + i;
		spin_lock_init(&port->lock);
		port->table = RB_ROOT;
		init_completion(&port->comp);
		atomic_set(&port->refcount, 1);
	}

	dev->device = device;
	ib_set_client_data(device, &mcast_client, dev);

	INIT_IB_EVENT_HANDLER(&dev->event_handler, device, mcast_event_handler);
	ib_register_event_handler(&dev->event_handler);
}

static void mcast_remove_one(struct ib_device *device)
{
	struct mcast_device *dev;
	struct mcast_port *port;
	int i;

	dev = ib_get_client_data(device, &mcast_client);
	if (!dev)
		return;

	ib_unregister_event_handler(&dev->event_handler);
	flush_workqueue(mcast_wq);

	for (i = 0; i <= dev->end_port - dev->start_port; i++) {
		port = &dev->port[i];
		deref_port(port);
		wait_for_completion(&port->comp);
	}

	kfree(dev);
}

int mcast_init(void)
{
	int ret;

	mcast_wq = create_singlethread_workqueue("ib_mcast");
	if (!mcast_wq)
		return -ENOMEM;

	ib_sa_register_client(&sa_client);

	ret = ib_register_client(&mcast_client);
	if (ret)
		goto err;
	return 0;

err:
	ib_sa_unregister_client(&sa_client);
	destroy_workqueue(mcast_wq);
	return ret;
}

void mcast_cleanup(void)
{
	ib_unregister_client(&mcast_client);
	ib_sa_unregister_client(&sa_client);
	destroy_workqueue(mcast_wq);
}