RDMA/cma: Add multicast communication support

Extend rdma_cm to support multicast communication. Multicast support
is added to the existing RDMA_PS_UDP port space, as well as a new
RDMA_PS_IPOIB port space. The latter port space allows joining the
multicast groups used by IPoIB, which enables offloading IPoIB traffic
to a separate QP. The port space determines the signature used in the
MGID when joining the group. The newly added RDMA_PS_IPOIB also
allows for unicast operations, similar to RDMA_PS_UDP.

Supporting the RDMA_PS_IPOIB requires changing how UD QPs are initialized,
since we can no longer assume that the qkey is constant. This requires
saving the Q_Key to use when attaching to a device, so that it is
available when creating the QP. The Q_Key information is exported to
the user through the existing rdma_init_qp_attr() interface.

Multicast support is also exported to userspace through the rdma_ucm.

Signed-off-by: Roland Dreier <[e-mail]>

drivers/infiniband/core/cma.c

···
 static DEFINE_IDR(sdp_ps);
 static DEFINE_IDR(tcp_ps);
 static DEFINE_IDR(udp_ps);
+static DEFINE_IDR(ipoib_ps);
 static int next_port;
 
 struct cma_device {
···
 	struct list_head	list;
 	struct list_head	listen_list;
 	struct cma_device	*cma_dev;
+	struct list_head	mc_list;
 
 	enum cma_state		state;
 	spinlock_t		lock;
···
 	} cm_id;
 
 	u32			seq_num;
+	u32			qkey;
 	u32			qp_num;
 	u8			srq;
+};
+
+struct cma_multicast {
+	struct rdma_id_private *id_priv;
+	union {
+		struct ib_sa_multicast *ib;
+	} multicast;
+	struct list_head	list;
+	void			*context;
+	struct sockaddr		addr;
+	u8			pad[sizeof(struct sockaddr_in6) -
+				    sizeof(struct sockaddr)];
 };
 
 struct cma_work {
···
 	hh->ip_version = (ip_ver << 4) | (hh->ip_version & 0xF);
 }
 
+static inline int cma_is_ud_ps(enum rdma_port_space ps)
+{
+	return (ps == RDMA_PS_UDP || ps == RDMA_PS_IPOIB);
+}
+
 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 			      struct cma_device *cma_dev)
 {
···
 	id_priv->cma_dev = NULL;
 }
 
+static int cma_set_qkey(struct ib_device *device, u8 port_num,
+			enum rdma_port_space ps,
+			struct rdma_dev_addr *dev_addr, u32 *qkey)
+{
+	struct ib_sa_mcmember_rec rec;
+	int ret = 0;
+
+	switch (ps) {
+	case RDMA_PS_UDP:
+		*qkey = RDMA_UDP_QKEY;
+		break;
+	case RDMA_PS_IPOIB:
+		ib_addr_get_mgid(dev_addr, &rec.mgid);
+		ret = ib_sa_get_mcmember_rec(device, port_num, &rec.mgid, &rec);
+		*qkey = be32_to_cpu(rec.qkey);
+		break;
+	default:
+		break;
+	}
+	return ret;
+}
+
 static int cma_acquire_dev(struct rdma_id_private *id_priv)
 {
-	enum rdma_node_type dev_type = id_priv->id.route.addr.dev_addr.dev_type;
+	struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 	struct cma_device *cma_dev;
 	union ib_gid gid;
 	int ret = -ENODEV;
 
-	switch (rdma_node_get_transport(dev_type)) {
+	switch (rdma_node_get_transport(dev_addr->dev_type)) {
 	case RDMA_TRANSPORT_IB:
-		ib_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
+		ib_addr_get_sgid(dev_addr, &gid);
 		break;
 	case RDMA_TRANSPORT_IWARP:
-		iw_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
+		iw_addr_get_sgid(dev_addr, &gid);
 		break;
 	default:
 		return -ENODEV;
···
 		ret = ib_find_cached_gid(cma_dev->device, &gid,
 					 &id_priv->id.port_num, NULL);
 		if (!ret) {
-			cma_attach_to_dev(id_priv, cma_dev);
+			ret = cma_set_qkey(cma_dev->device,
+					   id_priv->id.port_num,
+					   id_priv->id.ps, dev_addr,
+					   &id_priv->qkey);
+			if (!ret)
+				cma_attach_to_dev(id_priv, cma_dev);
 			break;
 		}
 	}
···
 	init_waitqueue_head(&id_priv->wait_remove);
 	atomic_set(&id_priv->dev_remove, 0);
 	INIT_LIST_HEAD(&id_priv->listen_list);
+	INIT_LIST_HEAD(&id_priv->mc_list);
 	get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 
 	return &id_priv->id;
 }
 EXPORT_SYMBOL(rdma_create_id);
 
-static int cma_init_ib_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
+static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 {
 	struct ib_qp_attr qp_attr;
-	struct rdma_dev_addr *dev_addr;
-	int ret;
+	int qp_attr_mask, ret;
 
-	dev_addr = &id_priv->id.route.addr.dev_addr;
-	ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
-				  ib_addr_get_pkey(dev_addr),
-				  &qp_attr.pkey_index);
+	qp_attr.qp_state = IB_QPS_INIT;
+	ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 	if (ret)
 		return ret;
 
-	qp_attr.qp_state = IB_QPS_INIT;
-	qp_attr.qp_access_flags = 0;
-	qp_attr.port_num = id_priv->id.port_num;
-	return ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_ACCESS_FLAGS |
-					  IB_QP_PKEY_INDEX | IB_QP_PORT);
+	ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
+	if (ret)
+		return ret;
+
+	qp_attr.qp_state = IB_QPS_RTR;
+	ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
+	if (ret)
+		return ret;
+
+	qp_attr.qp_state = IB_QPS_RTS;
+	qp_attr.sq_psn = 0;
+	ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
+
+	return ret;
 }
 
-static int cma_init_iw_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
+static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 {
 	struct ib_qp_attr qp_attr;
+	int qp_attr_mask, ret;
 
 	qp_attr.qp_state = IB_QPS_INIT;
-	qp_attr.qp_access_flags = IB_ACCESS_LOCAL_WRITE;
+	ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
+	if (ret)
+		return ret;
 
-	return ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_ACCESS_FLAGS);
+	return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 }
 
 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
···
 	if (IS_ERR(qp))
 		return PTR_ERR(qp);
 
-	switch (rdma_node_get_transport(id->device->node_type)) {
-	case RDMA_TRANSPORT_IB:
-		ret = cma_init_ib_qp(id_priv, qp);
-		break;
-	case RDMA_TRANSPORT_IWARP:
-		ret = cma_init_iw_qp(id_priv, qp);
-		break;
-	default:
-		ret = -ENOSYS;
-		break;
-	}
-
+	if (cma_is_ud_ps(id_priv->id.ps))
+		ret = cma_init_ud_qp(id_priv, qp);
+	else
+		ret = cma_init_conn_qp(id_priv, qp);
 	if (ret)
 		goto err;
 
···
 	return ib_modify_qp(id->qp, &qp_attr, IB_QP_STATE);
 }
 
+static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
+			       struct ib_qp_attr *qp_attr, int *qp_attr_mask)
+{
+	struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
+	int ret;
+
+	ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
+				  ib_addr_get_pkey(dev_addr),
+				  &qp_attr->pkey_index);
+	if (ret)
+		return ret;
+
+	qp_attr->port_num = id_priv->id.port_num;
+	*qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
+
+	if (cma_is_ud_ps(id_priv->id.ps)) {
+		qp_attr->qkey = id_priv->qkey;
+		*qp_attr_mask |= IB_QP_QKEY;
+	} else {
+		qp_attr->qp_access_flags = 0;
+		*qp_attr_mask |= IB_QP_ACCESS_FLAGS;
+	}
+	return 0;
+}
+
 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
 		       int *qp_attr_mask)
 {
 	struct rdma_id_private *id_priv;
-	int ret;
+	int ret = 0;
 
 	id_priv = container_of(id, struct rdma_id_private, id);
 	switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
 	case RDMA_TRANSPORT_IB:
-		ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
-					 qp_attr_mask);
+		if (!id_priv->cm_id.ib || cma_is_ud_ps(id_priv->id.ps))
+			ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
+		else
+			ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
+						 qp_attr_mask);
 		if (qp_attr->qp_state == IB_QPS_RTR)
 			qp_attr->rq_psn = id_priv->seq_num;
 		break;
 	case RDMA_TRANSPORT_IWARP:
-		ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
-					qp_attr_mask);
+		if (!id_priv->cm_id.iw) {
+			qp_attr->qp_access_flags = IB_ACCESS_LOCAL_WRITE;
+			*qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
+		} else
+			ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
+						 qp_attr_mask);
 		break;
 	default:
 		ret = -ENOSYS;
···
 	mutex_unlock(&lock);
 }
 
+static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
+{
+	struct cma_multicast *mc;
+
+	while (!list_empty(&id_priv->mc_list)) {
+		mc = container_of(id_priv->mc_list.next,
+				  struct cma_multicast, list);
+		list_del(&mc->list);
+		ib_sa_free_multicast(mc->multicast.ib);
+		kfree(mc);
+	}
+}
+
 void rdma_destroy_id(struct rdma_cm_id *id)
 {
 	struct rdma_id_private *id_priv;
···
 		default:
 			break;
 		}
+		cma_leave_mc_groups(id_priv);
 		mutex_lock(&lock);
 		cma_detach_from_dev(id_priv);
 	}
···
 	memset(&event, 0, sizeof event);
 	offset = cma_user_data_offset(listen_id->id.ps);
 	event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
-	if (listen_id->id.ps == RDMA_PS_UDP) {
+	if (cma_is_ud_ps(listen_id->id.ps)) {
 		conn_id = cma_new_udp_id(&listen_id->id, ib_event);
 		event.param.ud.private_data = ib_event->private_data + offset;
 		event.param.ud.private_data_len =
···
 	struct rdma_bind_list *bind_list;
 	int port, ret;
 
-	bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
+	bind_list = kmalloc(sizeof *bind_list, GFP_KERNEL);
 	if (!bind_list)
 		return -ENOMEM;
 
···
 	case RDMA_PS_UDP:
 		ps = &udp_ps;
 		break;
+	case RDMA_PS_IPOIB:
+		ps = &ipoib_ps;
+		break;
 	default:
 		return -EPROTONOSUPPORT;
 	}
···
 			event.status = ib_event->param.sidr_rep_rcvd.status;
 			break;
 		}
-		if (rep->qkey != RDMA_UD_QKEY) {
+		if (id_priv->qkey != rep->qkey) {
 			event.event = RDMA_CM_EVENT_UNREACHABLE;
 			event.status = -EINVAL;
 			break;
···
 
 	switch (rdma_node_get_transport(id->device->node_type)) {
 	case RDMA_TRANSPORT_IB:
-		if (id->ps == RDMA_PS_UDP)
+		if (cma_is_ud_ps(id->ps))
 			ret = cma_resolve_ib_udp(id_priv, conn_param);
 		else
 			ret = cma_connect_ib(id_priv, conn_param);
···
 	rep.status = status;
 	if (status == IB_SIDR_SUCCESS) {
 		rep.qp_num = id_priv->qp_num;
-		rep.qkey = RDMA_UD_QKEY;
+		rep.qkey = id_priv->qkey;
 	}
 	rep.private_data = private_data;
 	rep.private_data_len = private_data_len;
···
 
 	switch (rdma_node_get_transport(id->device->node_type)) {
 	case RDMA_TRANSPORT_IB:
-		if (id->ps == RDMA_PS_UDP)
+		if (cma_is_ud_ps(id->ps))
 			ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
 						conn_param->private_data,
 						conn_param->private_data_len);
···
 
 	switch (rdma_node_get_transport(id->device->node_type)) {
 	case RDMA_TRANSPORT_IB:
-		if (id->ps == RDMA_PS_UDP)
+		if (cma_is_ud_ps(id->ps))
 			ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT,
 						private_data, private_data_len);
 		else
···
 	return ret;
 }
 EXPORT_SYMBOL(rdma_disconnect);
+
+static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
+{
+	struct rdma_id_private *id_priv;
+	struct cma_multicast *mc = multicast->context;
+	struct rdma_cm_event event;
+	int ret;
+
+	id_priv = mc->id_priv;
+	atomic_inc(&id_priv->dev_remove);
+	if (!cma_comp(id_priv, CMA_ADDR_BOUND) &&
+	    !cma_comp(id_priv, CMA_ADDR_RESOLVED))
+		goto out;
+
+	if (!status && id_priv->id.qp)
+		status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
+					 multicast->rec.mlid);
+
+	memset(&event, 0, sizeof event);
+	event.status = status;
+	event.param.ud.private_data = mc->context;
+	if (!status) {
+		event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
+		ib_init_ah_from_mcmember(id_priv->id.device,
+					 id_priv->id.port_num, &multicast->rec,
+					 &event.param.ud.ah_attr);
+		event.param.ud.qp_num = 0xFFFFFF;
+		event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
+	} else
+		event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
+
+	ret = id_priv->id.event_handler(&id_priv->id, &event);
+	if (ret) {
+		cma_exch(id_priv, CMA_DESTROYING);
+		cma_release_remove(id_priv);
+		rdma_destroy_id(&id_priv->id);
+		return 0;
+	}
+out:
+	cma_release_remove(id_priv);
+	return 0;
+}
+
+static void cma_set_mgid(struct rdma_id_private *id_priv,
+			 struct sockaddr *addr, union ib_gid *mgid)
+{
+	unsigned char mc_map[MAX_ADDR_LEN];
+	struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
+	struct sockaddr_in *sin = (struct sockaddr_in *) addr;
+	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
+
+	if (cma_any_addr(addr)) {
+		memset(mgid, 0, sizeof *mgid);
+	} else if ((addr->sa_family == AF_INET6) &&
+		   ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFF10A01B) ==
+								 0xFF10A01B)) {
+		/* IPv6 address is an SA assigned MGID. */
+		memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
+	} else {
+		ip_ib_mc_map(sin->sin_addr.s_addr, mc_map);
+		if (id_priv->id.ps == RDMA_PS_UDP)
+			mc_map[7] = 0x01;	/* Use RDMA CM signature */
+		mc_map[8] = ib_addr_get_pkey(dev_addr) >> 8;
+		mc_map[9] = (unsigned char) ib_addr_get_pkey(dev_addr);
+		*mgid = *(union ib_gid *) (mc_map + 4);
+	}
+}
+
+static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
+				 struct cma_multicast *mc)
+{
+	struct ib_sa_mcmember_rec rec;
+	struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
+	ib_sa_comp_mask comp_mask;
+	int ret;
+
+	ib_addr_get_mgid(dev_addr, &rec.mgid);
+	ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
+				     &rec.mgid, &rec);
+	if (ret)
+		return ret;
+
+	cma_set_mgid(id_priv, &mc->addr, &rec.mgid);
+	if (id_priv->id.ps == RDMA_PS_UDP)
+		rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
+	ib_addr_get_sgid(dev_addr, &rec.port_gid);
+	rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
+	rec.join_state = 1;
+
+	comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
+		    IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
+		    IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
+		    IB_SA_MCMEMBER_REC_FLOW_LABEL |
+		    IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
+
+	mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
+						id_priv->id.port_num, &rec,
+						comp_mask, GFP_KERNEL,
+						cma_ib_mc_handler, mc);
+	if (IS_ERR(mc->multicast.ib))
+		return PTR_ERR(mc->multicast.ib);
+
+	return 0;
+}
+
+int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
+			void *context)
+{
+	struct rdma_id_private *id_priv;
+	struct cma_multicast *mc;
+	int ret;
+
+	id_priv = container_of(id, struct rdma_id_private, id);
+	if (!cma_comp(id_priv, CMA_ADDR_BOUND) &&
+	    !cma_comp(id_priv, CMA_ADDR_RESOLVED))
+		return -EINVAL;
+
+	mc = kmalloc(sizeof *mc, GFP_KERNEL);
+	if (!mc)
+		return -ENOMEM;
+
+	memcpy(&mc->addr, addr, ip_addr_size(addr));
+	mc->context = context;
+	mc->id_priv = id_priv;
+
+	spin_lock(&id_priv->lock);
+	list_add(&mc->list, &id_priv->mc_list);
+	spin_unlock(&id_priv->lock);
+
+	switch (rdma_node_get_transport(id->device->node_type)) {
+	case RDMA_TRANSPORT_IB:
+		ret = cma_join_ib_multicast(id_priv, mc);
+		break;
+	default:
+		ret = -ENOSYS;
+		break;
+	}
+
+	if (ret) {
+		spin_lock_irq(&id_priv->lock);
+		list_del(&mc->list);
+		spin_unlock_irq(&id_priv->lock);
+		kfree(mc);
+	}
+	return ret;
+}
+EXPORT_SYMBOL(rdma_join_multicast);
+
+void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
+{
+	struct rdma_id_private *id_priv;
+	struct cma_multicast *mc;
+
+	id_priv = container_of(id, struct rdma_id_private, id);
+	spin_lock_irq(&id_priv->lock);
+	list_for_each_entry(mc, &id_priv->mc_list, list) {
+		if (!memcmp(&mc->addr, addr, ip_addr_size(addr))) {
+			list_del(&mc->list);
+			spin_unlock_irq(&id_priv->lock);
+
+			if (id->qp)
+				ib_detach_mcast(id->qp,
+						&mc->multicast.ib->rec.mgid,
+						mc->multicast.ib->rec.mlid);
+			ib_sa_free_multicast(mc->multicast.ib);
+			kfree(mc);
+			return;
+		}
+	}
+	spin_unlock_irq(&id_priv->lock);
+}
+EXPORT_SYMBOL(rdma_leave_multicast);
 
 static void cma_add_one(struct ib_device *device)
 {
···
 	idr_destroy(&sdp_ps);
 	idr_destroy(&tcp_ps);
 	idr_destroy(&udp_ps);
+	idr_destroy(&ipoib_ps);
 }
 
 module_init(cma_init);

drivers/infiniband/core/ucma.c

···
 	u64			uid;
 
 	struct list_head	list;
+	struct list_head	mc_list;
+};
+
+struct ucma_multicast {
+	struct ucma_context	*ctx;
+	int			id;
+	int			events_reported;
+
+	u64			uid;
+	struct list_head	list;
+	struct sockaddr		addr;
+	u8			pad[sizeof(struct sockaddr_in6) -
+				    sizeof(struct sockaddr)];
 };
 
 struct ucma_event {
 	struct ucma_context	*ctx;
+	struct ucma_multicast	*mc;
 	struct list_head	list;
 	struct rdma_cm_id	*cm_id;
 	struct rdma_ucm_event_resp resp;
···
 
 static DEFINE_MUTEX(mut);
 static DEFINE_IDR(ctx_idr);
+static DEFINE_IDR(multicast_idr);
 
 static inline struct ucma_context *_ucma_find_context(int id,
 						      struct ucma_file *file)
···
 
 	atomic_set(&ctx->ref, 1);
 	init_completion(&ctx->comp);
+	INIT_LIST_HEAD(&ctx->mc_list);
 	ctx->file = file;
 
 	do {
···
 
 error:
 	kfree(ctx);
+	return NULL;
+}
+
+static struct ucma_multicast* ucma_alloc_multicast(struct ucma_context *ctx)
+{
+	struct ucma_multicast *mc;
+	int ret;
+
+	mc = kzalloc(sizeof(*mc), GFP_KERNEL);
+	if (!mc)
+		return NULL;
+
+	do {
+		ret = idr_pre_get(&multicast_idr, GFP_KERNEL);
+		if (!ret)
+			goto error;
+
+		mutex_lock(&mut);
+		ret = idr_get_new(&multicast_idr, mc, &mc->id);
+		mutex_unlock(&mut);
+	} while (ret == -EAGAIN);
+
+	if (ret)
+		goto error;
+
+	mc->ctx = ctx;
+	list_add_tail(&mc->list, &ctx->mc_list);
+	return mc;
+
+error:
+	kfree(mc);
 	return NULL;
 }
 
···
 				   struct ucma_event *uevent)
 {
 	uevent->ctx = ctx;
-	uevent->resp.uid = ctx->uid;
-	uevent->resp.id = ctx->id;
+	switch (event->event) {
+	case RDMA_CM_EVENT_MULTICAST_JOIN:
+	case RDMA_CM_EVENT_MULTICAST_ERROR:
+		uevent->mc = (struct ucma_multicast *)
+			     event->param.ud.private_data;
+		uevent->resp.uid = uevent->mc->uid;
+		uevent->resp.id = uevent->mc->id;
+		break;
+	default:
+		uevent->resp.uid = ctx->uid;
+		uevent->resp.id = ctx->id;
+		break;
+	}
 }
 
 static int ucma_event_handler(struct rdma_cm_id *cm_id,
···
 	ucma_set_event_context(ctx, event, uevent);
 	uevent->resp.event = event->event;
 	uevent->resp.status = event->status;
-	if (cm_id->ps == RDMA_PS_UDP)
+	if (cm_id->ps == RDMA_PS_UDP || cm_id->ps == RDMA_PS_IPOIB)
 		ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud);
 	else
 		ucma_copy_conn_event(&uevent->resp.param.conn,
···
 
 	list_del(&uevent->list);
 	uevent->ctx->events_reported++;
+	if (uevent->mc)
+		uevent->mc->events_reported++;
 	kfree(uevent);
 done:
 	mutex_unlock(&file->mut);
···
 	return ret;
 }
 
+static void ucma_cleanup_multicast(struct ucma_context *ctx)
+{
+	struct ucma_multicast *mc, *tmp;
+
+	mutex_lock(&mut);
+	list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) {
+		list_del(&mc->list);
+		idr_remove(&multicast_idr, mc->id);
+		kfree(mc);
+	}
+	mutex_unlock(&mut);
+}
+
 static void ucma_cleanup_events(struct ucma_context *ctx)
 {
 	struct ucma_event *uevent, *tmp;
···
 	}
 }
 
+static void ucma_cleanup_mc_events(struct ucma_multicast *mc)
+{
+	struct ucma_event *uevent, *tmp;
+
+	list_for_each_entry_safe(uevent, tmp, &mc->ctx->file->event_list, list) {
+		if (uevent->mc != mc)
+			continue;
+
+		list_del(&uevent->list);
+		kfree(uevent);
+	}
+}
+
 static int ucma_free_ctx(struct ucma_context *ctx)
 {
 	int events_reported;
 
 	/* No new events will be generated after destroying the id. */
 	rdma_destroy_id(ctx->cm_id);
+
+	ucma_cleanup_multicast(ctx);
 
 	/* Cleanup events not yet reported to the user. */
 	mutex_lock(&ctx->file->mut);
···
 	return ret;
 }
 
+static ssize_t ucma_join_multicast(struct ucma_file *file,
+				   const char __user *inbuf,
+				   int in_len, int out_len)
+{
+	struct rdma_ucm_join_mcast cmd;
+	struct rdma_ucm_create_id_resp resp;
+	struct ucma_context *ctx;
+	struct ucma_multicast *mc;
+	int ret;
+
+	if (out_len < sizeof(resp))
+		return -ENOSPC;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	ctx = ucma_get_ctx(file, cmd.id);
+	if (IS_ERR(ctx))
+		return PTR_ERR(ctx);
+
+	mutex_lock(&file->mut);
+	mc = ucma_alloc_multicast(ctx);
+	if (IS_ERR(mc)) {
+		ret = PTR_ERR(mc);
+		goto err1;
+	}
+
+	mc->uid = cmd.uid;
+	memcpy(&mc->addr, &cmd.addr, sizeof cmd.addr);
+	ret = rdma_join_multicast(ctx->cm_id, &mc->addr, mc);
+	if (ret)
+		goto err2;
+
+	resp.id = mc->id;
+	if (copy_to_user((void __user *)(unsigned long)cmd.response,
+			 &resp, sizeof(resp))) {
+		ret = -EFAULT;
+		goto err3;
+	}
+
+	mutex_unlock(&file->mut);
+	ucma_put_ctx(ctx);
+	return 0;
+
+err3:
+	rdma_leave_multicast(ctx->cm_id, &mc->addr);
+	ucma_cleanup_mc_events(mc);
+err2:
+	mutex_lock(&mut);
+	idr_remove(&multicast_idr, mc->id);
+	mutex_unlock(&mut);
+	list_del(&mc->list);
+	kfree(mc);
+err1:
+	mutex_unlock(&file->mut);
+	ucma_put_ctx(ctx);
+	return ret;
+}
+
+static ssize_t ucma_leave_multicast(struct ucma_file *file,
+				    const char __user *inbuf,
+				    int in_len, int out_len)
+{
+	struct rdma_ucm_destroy_id cmd;
+	struct rdma_ucm_destroy_id_resp resp;
+	struct ucma_multicast *mc;
+	int ret = 0;
+
+	if (out_len < sizeof(resp))
+		return -ENOSPC;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	mutex_lock(&mut);
+	mc = idr_find(&multicast_idr, cmd.id);
+	if (!mc)
+		mc = ERR_PTR(-ENOENT);
+	else if (mc->ctx->file != file)
+		mc = ERR_PTR(-EINVAL);
+	else {
+		idr_remove(&multicast_idr, mc->id);
+		atomic_inc(&mc->ctx->ref);
+	}
+	mutex_unlock(&mut);
+
+	if (IS_ERR(mc)) {
+		ret = PTR_ERR(mc);
+		goto out;
+	}
+
+	rdma_leave_multicast(mc->ctx->cm_id, &mc->addr);
+	mutex_lock(&mc->ctx->file->mut);
+	ucma_cleanup_mc_events(mc);
+	list_del(&mc->list);
+	mutex_unlock(&mc->ctx->file->mut);
+
+	ucma_put_ctx(mc->ctx);
+	resp.events_reported = mc->events_reported;
+	kfree(mc);
+
+	if (copy_to_user((void __user *)(unsigned long)cmd.response,
+			 &resp, sizeof(resp)))
+		ret = -EFAULT;
+out:
+	return ret;
+}
+
 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file,
 				   const char __user *inbuf,
 				   int in_len, int out_len) = {
···
 	[RDMA_USER_CM_CMD_GET_OPTION]	= NULL,
 	[RDMA_USER_CM_CMD_SET_OPTION]	= NULL,
 	[RDMA_USER_CM_CMD_NOTIFY]	= ucma_notify,
+	[RDMA_USER_CM_CMD_JOIN_MCAST]	= ucma_join_multicast,
+	[RDMA_USER_CM_CMD_LEAVE_MCAST]	= ucma_leave_multicast,
 };
 
 static ssize_t ucma_write(struct file *filp, const char __user *buf,

include/rdma/rdma_cm.h

···
 	RDMA_CM_EVENT_ESTABLISHED,
 	RDMA_CM_EVENT_DISCONNECTED,
 	RDMA_CM_EVENT_DEVICE_REMOVAL,
+	RDMA_CM_EVENT_MULTICAST_JOIN,
+	RDMA_CM_EVENT_MULTICAST_ERROR
 };
 
 enum rdma_port_space {
 	RDMA_PS_SDP  = 0x0001,
+	RDMA_PS_IPOIB= 0x0002,
 	RDMA_PS_TCP  = 0x0106,
 	RDMA_PS_UDP  = 0x0111,
 	RDMA_PS_SCTP = 0x0183
···
  */
 int rdma_disconnect(struct rdma_cm_id *id);
 
-#endif /* RDMA_CM_H */
+/**
+ * rdma_join_multicast - Join the multicast group specified by the given
+ *   address.
+ * @id: Communication identifier associated with the request.
+ * @addr: Multicast address identifying the group to join.
+ * @context: User-defined context associated with the join request, returned
+ * to the user through the private_data pointer in multicast events.
+ */
+int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
+			void *context);
 
+/**
+ * rdma_leave_multicast - Leave the multicast group specified by the given
+ *   address.
+ */
+void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr);
+
+#endif /* RDMA_CM_H */

include/rdma/rdma_cm_ib.h

···
 int rdma_set_ib_paths(struct rdma_cm_id *id,
 		      struct ib_sa_path_rec *path_rec, int num_paths);
 
-/* Global qkey for UD QPs and multicast groups. */
-#define RDMA_UD_QKEY 0x01234567
+/* Global qkey for UDP QPs and multicast groups. */
+#define RDMA_UDP_QKEY 0x01234567
 
 #endif /* RDMA_CM_IB_H */

include/rdma/rdma_user_cm.h

···
 #include <rdma/ib_user_verbs.h>
 #include <rdma/ib_user_sa.h>
 
-#define RDMA_USER_CM_ABI_VERSION	3
+#define RDMA_USER_CM_ABI_VERSION	4
 
 #define RDMA_MAX_PRIVATE_DATA		256
 
···
 	RDMA_USER_CM_CMD_GET_EVENT,
 	RDMA_USER_CM_CMD_GET_OPTION,
 	RDMA_USER_CM_CMD_SET_OPTION,
-	RDMA_USER_CM_CMD_NOTIFY
+	RDMA_USER_CM_CMD_NOTIFY,
+	RDMA_USER_CM_CMD_JOIN_MCAST,
+	RDMA_USER_CM_CMD_LEAVE_MCAST
 };
 
 /*
···
 struct rdma_ucm_notify {
 	__u32 id;
 	__u32 event;
+};
+
+struct rdma_ucm_join_mcast {
+	__u64 response;		/* rdma_ucm_create_id_resp */
+	__u64 uid;
+	struct sockaddr_in6 addr;
+	__u32 id;
 };
 
 struct rdma_ucm_get_event {