On Mon, Mar 14, 2022 at 16:58, Vladimir Oltean [off-list ref] wrote:

On Mon, Mar 14, 2022 at 10:52:20AM +0100, Tobias Waldekranz wrote:

quoted

+int br_mst_fill_info(struct sk_buff *skb, struct net_bridge_vlan_group *vg)
+{
+	struct net_bridge_vlan *v;
+	struct nlattr *nest;
+	unsigned long *seen;
+	int err = 0;
+
+	seen = bitmap_zalloc(VLAN_N_VID, 0);

I see there is precedent in the bridge driver for using dynamic
allocation as opposed to on-stack declaration using DECLARE_BITMAP().
I imagine this isn't just to be "heapsters", but why?

I don't have a very good sense of how much on-stack memory is too much
(a lot probably depends on the expected depth of the call stack too, and here it
doesn't appear to be too deep), but I see that mlxsw_sp_bridge_vxlan_vlan_is_valid()
has a DECLARE_BITMAP(vlans, VLAN_N_VID) too.

The comment applies for callers of br_mst_get_info() too.

In v4, things become even worse, as I need to allocate the bitmap in a
context where I can't return an error. So if it's ok to keep it on the
stack, then that would be great.

Here's the code in question:

size_t br_mst_info_size(const struct net_bridge_vlan_group *vg)
{
	const struct net_bridge_vlan *v;
	unsigned long *seen;
	size_t sz;

	seen = bitmap_zalloc(VLAN_N_VID, 0);
	if (WARN_ON(!seen))
		return 0;

	/* IFLA_BRIDGE_MST */
	sz = nla_total_size(0);

	list_for_each_entry(v, &vg->vlan_list, vlist) {
		if (test_bit(v->brvlan->msti, seen))
			continue;

		/* IFLA_BRIDGE_MST_ENTRY */
		sz += nla_total_size(0) +
			/* IFLA_BRIDGE_MST_ENTRY_MSTI */
			nla_total_size(sizeof(u16)) +
			/* IFLA_BRIDGE_MST_ENTRY_STATE */
			nla_total_size(sizeof(u8));

		__set_bit(v->brvlan->msti, seen);
	}

	kfree(seen);
	return sz;
}