An odd issue occurs when more than one vxlan over v6 tunnel is configured.

Once a second tunnel is configured, incoming packets for the first are dropped with a "no such tunnel" type of error.

Some tracing shows that after the second hash_set() call, subsequent hash_get() calls for the key of the first tunnel return 0. Because of this it’s then possible to re-add the same tunnel twice – and its key is now findable with hash_get() – until another tunnel is created and them magically it stops again.

The result is that vxlan over ipv6 is currently unreliable when more than one tunnel is configured.