RDMA/rxe: Close a race after ib_register_device

Since rxe allows unregistration from other threads the rxe pointer can
become invalid any moment after ib_register_driver returns. This could
cause a user triggered use after free.

Add another driver callback to be called right after the device becomes
registered to complete any device setup required post-registration.  This
callback has enough core locking to prevent the device from becoming
unregistered.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

drivers/infiniband/sw/rxe/rxe_net.c

@@ -517,24 +517,24 @@ enum rdma_link_layer rxe_link_layer(struct rxe_dev *rxe, unsigned int port_num)
 	return IB_LINK_LAYER_ETHERNET;
 }
 
-struct rxe_dev *rxe_net_add(struct net_device *ndev)
+int rxe_net_add(struct net_device *ndev)
 {
 	int err;
 	struct rxe_dev *rxe = NULL;
 
 	rxe = ib_alloc_device(rxe_dev, ib_dev);
 	if (!rxe)
-		return NULL;
+		return -ENOMEM;
 
 	rxe->ndev = ndev;
 
 	err = rxe_add(rxe, ndev->mtu);
 	if (err) {
 		ib_dealloc_device(&rxe->ib_dev);
-		return NULL;
+		return err;
 	}
 
-	return rxe;
+	return 0;
 }
 
 static void rxe_port_event(struct rxe_dev *rxe,

drivers/infiniband/sw/rxe/rxe_net.h

@@ -43,7 +43,7 @@ struct rxe_recv_sockets {
 	struct socket *sk6;
 };
 
-struct rxe_dev *rxe_net_add(struct net_device *ndev);
+int rxe_net_add(struct net_device *ndev);
 
 int rxe_net_init(void);
 void rxe_net_exit(void);

drivers/infiniband/sw/rxe/rxe_sysfs.c

@@ -60,7 +60,6 @@ static int rxe_param_set_add(const char *val, const struct kernel_param *kp)
 	char intf[32];
 	struct net_device *ndev;
 	struct rxe_dev *exists;
-	struct rxe_dev *rxe;
 
 	len = sanitize_arg(val, intf, sizeof(intf));
 	if (!len) {
@@ -82,16 +81,12 @@ static int rxe_param_set_add(const char *val, const struct kernel_param *kp)
 		goto err;
 	}
 
-	rxe = rxe_net_add(ndev);
-	if (!rxe) {
+	err = rxe_net_add(ndev);
+	if (err) {
 		pr_err("failed to add %s\n", intf);
-		err = -EINVAL;
 		goto err;
 	}
 
-	rxe_set_port_state(rxe);
-	dev_info(&rxe->ib_dev.dev, "added %s\n", intf);
-
 err:
 	dev_put(ndev);
 	return err;

drivers/infiniband/sw/rxe/rxe_verbs.c

@@ -1125,6 +1125,15 @@ static const struct attribute_group rxe_attr_group = {
 	.attrs = rxe_dev_attributes,
 };
 
+static int rxe_enable_driver(struct ib_device *ib_dev)
+{
+	struct rxe_dev *rxe = container_of(ib_dev, struct rxe_dev, ib_dev);
+
+	rxe_set_port_state(rxe);
+	dev_info(&rxe->ib_dev.dev, "added %s\n", netdev_name(rxe->ndev));
+	return 0;
+}
+
 static const struct ib_device_ops rxe_dev_ops = {
 	.alloc_hw_stats = rxe_ib_alloc_hw_stats,
 	.alloc_mr = rxe_alloc_mr,
@@ -1144,6 +1153,7 @@ static const struct ib_device_ops rxe_dev_ops = {
 	.destroy_qp = rxe_destroy_qp,
 	.destroy_srq = rxe_destroy_srq,
 	.detach_mcast = rxe_detach_mcast,
+	.enable_driver = rxe_enable_driver,
 	.get_dma_mr = rxe_get_dma_mr,
 	.get_hw_stats = rxe_ib_get_hw_stats,
 	.get_link_layer = rxe_get_link_layer,
@@ -1245,5 +1255,9 @@ int rxe_register_device(struct rxe_dev *rxe)
 	if (err)
 		pr_warn("%s failed with error %d\n", __func__, err);
 
+	/*
+	 * Note that rxe may be invalid at this point if another thread
+	 * unregistered it.
+	 */
 	return err;
 }