1 files changed, 123 insertions, 134 deletions

diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
index c30cf5307ce3..26de0dab60bb 100644
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -428,19 +428,6 @@ static int cma_comp_exch(struct rdma_id_private *id_priv,
 	return ret;
 }
 
-static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
-				   enum rdma_cm_state exch)
-{
-	unsigned long flags;
-	enum rdma_cm_state old;
-
-	spin_lock_irqsave(&id_priv->lock, flags);
-	old = id_priv->state;
-	id_priv->state = exch;
-	spin_unlock_irqrestore(&id_priv->lock, flags);
-	return old;
-}
-
 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 {
 	return hdr->ip_version >> 4;
@@ -1829,23 +1816,11 @@ static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
 	}
 }
 
-void rdma_destroy_id(struct rdma_cm_id *id)
+static void _destroy_id(struct rdma_id_private *id_priv,
+			enum rdma_cm_state state)
 {
-	struct rdma_id_private *id_priv;
-	enum rdma_cm_state state;
-
-	id_priv = container_of(id, struct rdma_id_private, id);
-	trace_cm_id_destroy(id_priv);
-	state = cma_exch(id_priv, RDMA_CM_DESTROYING);
 	cma_cancel_operation(id_priv, state);
 
-	/*
-	 * Wait for any active callback to finish.  New callbacks will find
-	 * the id_priv state set to destroying and abort.
-	 */
-	mutex_lock(&id_priv->handler_mutex);
-	mutex_unlock(&id_priv->handler_mutex);
-
 	rdma_restrack_del(&id_priv->res);
 	if (id_priv->cma_dev) {
 		if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
@@ -1874,6 +1849,42 @@ void rdma_destroy_id(struct rdma_cm_id *id)
 	put_net(id_priv->id.route.addr.dev_addr.net);
 	kfree(id_priv);
 }
+
+/*
+ * destroy an ID from within the handler_mutex. This ensures that no other
+ * handlers can start running concurrently.
+ */
+static void destroy_id_handler_unlock(struct rdma_id_private *id_priv)
+	__releases(&idprv->handler_mutex)
+{
+	enum rdma_cm_state state;
+	unsigned long flags;
+
+	trace_cm_id_destroy(id_priv);
+
+	/*
+	 * Setting the state to destroyed under the handler mutex provides a
+	 * fence against calling handler callbacks. If this is invoked due to
+	 * the failure of a handler callback then it guarentees that no future
+	 * handlers will be called.
+	 */
+	lockdep_assert_held(&id_priv->handler_mutex);
+	spin_lock_irqsave(&id_priv->lock, flags);
+	state = id_priv->state;
+	id_priv->state = RDMA_CM_DESTROYING;
+	spin_unlock_irqrestore(&id_priv->lock, flags);
+	mutex_unlock(&id_priv->handler_mutex);
+	_destroy_id(id_priv, state);
+}
+
+void rdma_destroy_id(struct rdma_cm_id *id)
+{
+	struct rdma_id_private *id_priv =
+		container_of(id, struct rdma_id_private, id);
+
+	mutex_lock(&id_priv->handler_mutex);
+	destroy_id_handler_unlock(id_priv);
+}
 EXPORT_SYMBOL(rdma_destroy_id);
 
 static int cma_rep_recv(struct rdma_id_private *id_priv)
@@ -1925,6 +1936,8 @@ static int cma_cm_event_handler(struct rdma_id_private *id_priv,
 {
 	int ret;
 
+	lockdep_assert_held(&id_priv->handler_mutex);
+
 	trace_cm_event_handler(id_priv, event);
 	ret = id_priv->id.event_handler(&id_priv->id, event);
 	trace_cm_event_done(id_priv, event, ret);
@@ -1936,7 +1949,7 @@ static int cma_ib_handler(struct ib_cm_id *cm_id,
 {
 	struct rdma_id_private *id_priv = cm_id->context;
 	struct rdma_cm_event event = {};
-	int ret = 0;
+	int ret;
 
 	mutex_lock(&id_priv->handler_mutex);
 	if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
@@ -2005,14 +2018,12 @@ static int cma_ib_handler(struct ib_cm_id *cm_id,
 	if (ret) {
 		/* Destroy the CM ID by returning a non-zero value. */
 		id_priv->cm_id.ib = NULL;
-		cma_exch(id_priv, RDMA_CM_DESTROYING);
-		mutex_unlock(&id_priv->handler_mutex);
-		rdma_destroy_id(&id_priv->id);
+		destroy_id_handler_unlock(id_priv);
 		return ret;
 	}
 out:
 	mutex_unlock(&id_priv->handler_mutex);
-	return ret;
+	return 0;
 }
 
 static struct rdma_id_private *
@@ -2174,7 +2185,7 @@ static int cma_ib_req_handler(struct ib_cm_id *cm_id,
 	mutex_lock(&listen_id->handler_mutex);
 	if (listen_id->state != RDMA_CM_LISTEN) {
 		ret = -ECONNABORTED;
-		goto err1;
+		goto err_unlock;
 	}
 
 	offset = cma_user_data_offset(listen_id);
@@ -2191,55 +2202,38 @@ static int cma_ib_req_handler(struct ib_cm_id *cm_id,
 	}
 	if (!conn_id) {
 		ret = -ENOMEM;
-		goto err1;
+		goto err_unlock;
 	}
 
 	mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
 	ret = cma_ib_acquire_dev(conn_id, listen_id, &req);
-	if (ret)
-		goto err2;
+	if (ret) {
+		destroy_id_handler_unlock(conn_id);
+		goto err_unlock;
+	}
 
 	conn_id->cm_id.ib = cm_id;
 	cm_id->context = conn_id;
 	cm_id->cm_handler = cma_ib_handler;
 
-	/*
-	 * Protect against the user destroying conn_id from another thread
-	 * until we're done accessing it.
-	 */
-	cma_id_get(conn_id);
 	ret = cma_cm_event_handler(conn_id, &event);
-	if (ret)
-		goto err3;
-	/*
-	 * Acquire mutex to prevent user executing rdma_destroy_id()
-	 * while we're accessing the cm_id.
-	 */
-	mutex_lock(&lock);
+	if (ret) {
+		/* Destroy the CM ID by returning a non-zero value. */
+		conn_id->cm_id.ib = NULL;
+		mutex_unlock(&listen_id->handler_mutex);
+		destroy_id_handler_unlock(conn_id);
+		goto net_dev_put;
+	}
+
 	if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
 	    (conn_id->id.qp_type != IB_QPT_UD)) {
 		trace_cm_send_mra(cm_id->context);
 		ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
 	}
-	mutex_unlock(&lock);
 	mutex_unlock(&conn_id->handler_mutex);
-	mutex_unlock(&listen_id->handler_mutex);
-	cma_id_put(conn_id);
-	if (net_dev)
-		dev_put(net_dev);
-	return 0;
 
-err3:
-	cma_id_put(conn_id);
-	/* Destroy the CM ID by returning a non-zero value. */
-	conn_id->cm_id.ib = NULL;
-err2:
-	cma_exch(conn_id, RDMA_CM_DESTROYING);
-	mutex_unlock(&conn_id->handler_mutex);
-err1:
+err_unlock:
 	mutex_unlock(&listen_id->handler_mutex);
-	if (conn_id)
-		rdma_destroy_id(&conn_id->id);
 
 net_dev_put:
 	if (net_dev)
@@ -2339,9 +2333,7 @@ static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
 	if (ret) {
 		/* Destroy the CM ID by returning a non-zero value. */
 		id_priv->cm_id.iw = NULL;
-		cma_exch(id_priv, RDMA_CM_DESTROYING);
-		mutex_unlock(&id_priv->handler_mutex);
-		rdma_destroy_id(&id_priv->id);
+		destroy_id_handler_unlock(id_priv);
 		return ret;
 	}
 
@@ -2388,16 +2380,16 @@ static int iw_conn_req_handler(struct iw_cm_id *cm_id,
 
 	ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr);
 	if (ret) {
-		mutex_unlock(&conn_id->handler_mutex);
-		rdma_destroy_id(new_cm_id);
-		goto out;
+		mutex_unlock(&listen_id->handler_mutex);
+		destroy_id_handler_unlock(conn_id);
+		return ret;
 	}
 
 	ret = cma_iw_acquire_dev(conn_id, listen_id);
 	if (ret) {
-		mutex_unlock(&conn_id->handler_mutex);
-		rdma_destroy_id(new_cm_id);
-		goto out;
+		mutex_unlock(&listen_id->handler_mutex);
+		destroy_id_handler_unlock(conn_id);
+		return ret;
 	}
 
 	conn_id->cm_id.iw = cm_id;
@@ -2407,25 +2399,16 @@ static int iw_conn_req_handler(struct iw_cm_id *cm_id,
 	memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
 	memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
 
-	/*
-	 * Protect against the user destroying conn_id from another thread
-	 * until we're done accessing it.
-	 */
-	cma_id_get(conn_id);
 	ret = cma_cm_event_handler(conn_id, &event);
 	if (ret) {
 		/* User wants to destroy the CM ID */
 		conn_id->cm_id.iw = NULL;
-		cma_exch(conn_id, RDMA_CM_DESTROYING);
-		mutex_unlock(&conn_id->handler_mutex);
 		mutex_unlock(&listen_id->handler_mutex);
-		cma_id_put(conn_id);
-		rdma_destroy_id(&conn_id->id);
+		destroy_id_handler_unlock(conn_id);
 		return ret;
 	}
 
 	mutex_unlock(&conn_id->handler_mutex);
-	cma_id_put(conn_id);
 
 out:
 	mutex_unlock(&listen_id->handler_mutex);
@@ -2482,6 +2465,10 @@ static int cma_listen_handler(struct rdma_cm_id *id,
 {
 	struct rdma_id_private *id_priv = id->context;
 
+	/* Listening IDs are always destroyed on removal */
+	if (event->event == RDMA_CM_EVENT_DEVICE_REMOVAL)
+		return -1;
+
 	id->context = id_priv->id.context;
 	id->event_handler = id_priv->id.event_handler;
 	trace_cm_event_handler(id_priv, event);
@@ -2657,21 +2644,21 @@ static void cma_work_handler(struct work_struct *_work)
 {
 	struct cma_work *work = container_of(_work, struct cma_work, work);
 	struct rdma_id_private *id_priv = work->id;
-	int destroy = 0;
 
 	mutex_lock(&id_priv->handler_mutex);
 	if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
-		goto out;
+		goto out_unlock;
 
 	if (cma_cm_event_handler(id_priv, &work->event)) {
-		cma_exch(id_priv, RDMA_CM_DESTROYING);
-		destroy = 1;
+		cma_id_put(id_priv);
+		destroy_id_handler_unlock(id_priv);
+		goto out_free;
 	}
-out:
+
+out_unlock:
 	mutex_unlock(&id_priv->handler_mutex);
 	cma_id_put(id_priv);
-	if (destroy)
-		rdma_destroy_id(&id_priv->id);
+out_free:
 	kfree(work);
 }
 
@@ -2679,23 +2666,22 @@ static void cma_ndev_work_handler(struct work_struct *_work)
 {
 	struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
 	struct rdma_id_private *id_priv = work->id;
-	int destroy = 0;
 
 	mutex_lock(&id_priv->handler_mutex);
 	if (id_priv->state == RDMA_CM_DESTROYING ||
 	    id_priv->state == RDMA_CM_DEVICE_REMOVAL)
-		goto out;
+		goto out_unlock;
 
 	if (cma_cm_event_handler(id_priv, &work->event)) {
-		cma_exch(id_priv, RDMA_CM_DESTROYING);
-		destroy = 1;
+		cma_id_put(id_priv);
+		destroy_id_handler_unlock(id_priv);
+		goto out_free;
 	}
 
-out:
+out_unlock:
 	mutex_unlock(&id_priv->handler_mutex);
 	cma_id_put(id_priv);
-	if (destroy)
-		rdma_destroy_id(&id_priv->id);
+out_free:
 	kfree(work);
 }
 
@@ -3171,9 +3157,7 @@ static void addr_handler(int status, struct sockaddr *src_addr,
 		event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
 
 	if (cma_cm_event_handler(id_priv, &event)) {
-		cma_exch(id_priv, RDMA_CM_DESTROYING);
-		mutex_unlock(&id_priv->handler_mutex);
-		rdma_destroy_id(&id_priv->id);
+		destroy_id_handler_unlock(id_priv);
 		return;
 	}
 out:
@@ -3790,7 +3774,7 @@ static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
 	struct rdma_cm_event event = {};
 	const struct ib_cm_sidr_rep_event_param *rep =
 				&ib_event->param.sidr_rep_rcvd;
-	int ret = 0;
+	int ret;
 
 	mutex_lock(&id_priv->handler_mutex);
 	if (id_priv->state != RDMA_CM_CONNECT)
@@ -3840,14 +3824,12 @@ static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
 	if (ret) {
 		/* Destroy the CM ID by returning a non-zero value. */
 		id_priv->cm_id.ib = NULL;
-		cma_exch(id_priv, RDMA_CM_DESTROYING);
-		mutex_unlock(&id_priv->handler_mutex);
-		rdma_destroy_id(&id_priv->id);
+		destroy_id_handler_unlock(id_priv);
 		return ret;
 	}
 out:
 	mutex_unlock(&id_priv->handler_mutex);
-	return ret;
+	return 0;
 }
 
 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
@@ -4372,9 +4354,7 @@ static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
 
 	rdma_destroy_ah_attr(&event.param.ud.ah_attr);
 	if (ret) {
-		cma_exch(id_priv, RDMA_CM_DESTROYING);
-		mutex_unlock(&id_priv->handler_mutex);
-		rdma_destroy_id(&id_priv->id);
+		destroy_id_handler_unlock(id_priv);
 		return 0;
 	}
 
@@ -4789,50 +4769,59 @@ free_cma_dev:
 	return ret;
 }
 
-static int cma_remove_id_dev(struct rdma_id_private *id_priv)
+static void cma_send_device_removal_put(struct rdma_id_private *id_priv)
 {
-	struct rdma_cm_event event = {};
+	struct rdma_cm_event event = { .event = RDMA_CM_EVENT_DEVICE_REMOVAL };
 	enum rdma_cm_state state;
-	int ret = 0;
-
-	/* Record that we want to remove the device */
-	state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
-	if (state == RDMA_CM_DESTROYING)
-		return 0;
+	unsigned long flags;
 
-	cma_cancel_operation(id_priv, state);
 	mutex_lock(&id_priv->handler_mutex);
+	/* Record that we want to remove the device */
+	spin_lock_irqsave(&id_priv->lock, flags);
+	state = id_priv->state;
+	if (state == RDMA_CM_DESTROYING || state == RDMA_CM_DEVICE_REMOVAL) {
+		spin_unlock_irqrestore(&id_priv->lock, flags);
+		mutex_unlock(&id_priv->handler_mutex);
+		cma_id_put(id_priv);
+		return;
+	}
+	id_priv->state = RDMA_CM_DEVICE_REMOVAL;
+	spin_unlock_irqrestore(&id_priv->lock, flags);
 
-	/* Check for destruction from another callback. */
-	if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
-		goto out;
-
-	event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
-	ret = cma_cm_event_handler(id_priv, &event);
-out:
+	if (cma_cm_event_handler(id_priv, &event)) {
+		/*
+		 * At this point the ULP promises it won't call
+		 * rdma_destroy_id() concurrently
+		 */
+		cma_id_put(id_priv);
+		mutex_unlock(&id_priv->handler_mutex);
+		trace_cm_id_destroy(id_priv);
+		_destroy_id(id_priv, state);
+		return;
+	}
 	mutex_unlock(&id_priv->handler_mutex);
-	return ret;
+
+	/*
+	 * If this races with destroy then the thread that first assigns state
+	 * to a destroying does the cancel.
+	 */
+	cma_cancel_operation(id_priv, state);
+	cma_id_put(id_priv);
 }
 
 static void cma_process_remove(struct cma_device *cma_dev)
 {
-	struct rdma_id_private *id_priv;
-	int ret;
-
 	mutex_lock(&lock);
 	while (!list_empty(&cma_dev->id_list)) {
-		id_priv = list_entry(cma_dev->id_list.next,
-				     struct rdma_id_private, list);
+		struct rdma_id_private *id_priv = list_first_entry(
+			&cma_dev->id_list, struct rdma_id_private, list);
 
 		list_del(&id_priv->listen_list);
 		list_del_init(&id_priv->list);
 		cma_id_get(id_priv);
 		mutex_unlock(&lock);
 
-		ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
-		cma_id_put(id_priv);
-		if (ret)
-			rdma_destroy_id(&id_priv->id);
+		cma_send_device_removal_put(id_priv);
 
 		mutex_lock(&lock);
 	}