Re: [PATCH rdma-next 1/2] RDMA/core: Add an rb_tree that stores cm_ids sorted by ifindex and remote IP

[Jason Gunthorpe <jgg@nvidia.com>]

On Mon, Apr 04, 2022 at 03:27:26PM +0300, Leon Romanovsky wrote:
> +static int compare_netdev_and_ip(int ifindex_a, struct sockaddr *sa,
> +				 int ifindex_b, struct sockaddr *sb)
> +{
> +	if (ifindex_a != ifindex_b)
> +		return ifindex_a - ifindex_b;

These subtraction tricks don't work if the value can overflow
INT_MAX - INT_MIN == undefined


> +static int cma_add_id_to_tree(struct rdma_id_private *node_id_priv)
> +{
> +	struct rb_node **new = &id_table.rb_node, *parent = NULL;

This read of rb_node has to be under the spinlock

> +	struct id_table_entry *this, *node;
> +	struct rdma_id_private *id_priv;
> +	unsigned long flags;
> +	int result;
> +
> +	node = kzalloc(sizeof(*node), GFP_KERNEL);
> +	if (!node)
> +		return -ENOMEM;
> +
> +	spin_lock_irqsave(&id_table_lock, flags);
> +	while (*new) {
> +		this = container_of(*new, struct id_table_entry, rb_node);

Because rebalacing can alter the head

> +		id_priv = list_first_entry(
> +			&this->id_list, struct rdma_id_private, id_list_entry);
> +		result = compare_netdev_and_ip(
> +			node_id_priv->id.route.addr.dev_addr.bound_dev_if,
> +			cma_dst_addr(node_id_priv),
> +			id_priv->id.route.addr.dev_addr.bound_dev_if,
> +			cma_dst_addr(id_priv));

This pattern keeps repeating, one of the arguments to compare should
just be the id_table_entry * and do the list_first/etc inside the
compare.

Jason