Re: [PATCHv5 net-next] vxlan: virtual extensible lan

[David Miller <davem@davemloft.net>]

From: Stephen Hemminger <shemminger@vyatta.com>
Date: Tue, 25 Sep 2012 15:09:57 -0700

> +/* Hash Ethernet address */
> +static u32 eth_hash(const unsigned char *addr)
> +{
> +	/* could be optimized for unaligned access */
> +	u32 a = addr[5] << 8 | addr[4];
> +	u32 b = addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0];
> +
> +	return jhash_2words(a, b, vxlan_salt);
> +}
> +
> +/* Hash chain to use given mac address */
> +static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
> +						const u8 *mac)
> +{
> +	return &vxlan->fdb_head[hash_32(eth_hash(mac), FDB_HASH_BITS)];
> +}

This hashing is way overkill, and in fact is a mis-use of Jenkins.

Jenkins can work fine with power-of-two hash sizes, so using hash_32()
on the Jenkins hash result is nothing short of silly.

I would go one step further and change this to:

	(a ^ b) * vxlan_salt

much like how we hash IP addresses in the ipv4 ARP tables.  That
function is a universal hash, and therefore provides whatever kind
of protection you're trying to obtain using the salt.

But I wonder if this matters at all, the administrator controls
the contents of this table, rather than external entitites.