Queries are sometimes sent from all-zeroes source MAC address

[wkz]

When starting mcd on a newly created interface, the MAC address can not always be determined. When that happens, mcd soldiers on and uses an SA of all zeroes.

Preferably, we would not allow any queries to be sent on an interface until a proper SA has been selected.

[troglobit]

Indeed, with no SA mcd should go back to sleep and wake up on the next startup-query-interval, see #5.

[wkz]

I looked a bit at the source. Making some notes here.

It seems like we only get one shot, one opportunity, to capture the interface SA - and we unfortunately let it slip:

mcd/src/config.c

Lines 133 to 134 in 23cba46

	if (getmac(ifname, ifi->ifi_hwaddr, sizeof(ifi->ifi_hwaddr)))
	warn("failed finding hw address for iface %s", ifname);

This is the only caller of getmac() AFAIK, so it seems like if the config parser runs before the interface exists, then we're never going to fill in the SA.

Unless we want to do some fancy netlink parsing, I think we need to call getmac() on every RTM_{NEW,SET}LINK to make sure we keep mcd in sync with the interface config.

While we're there I think we need to consider the condition for starting the interface. At the moment, we settle for IFF_UP, i.e. admin up:

mcd/src/iface.c

Lines 249 to 252 in 23cba46

	if (flags & IFF_UP) {
	ifi->ifi_flags &= ~IFIF_DOWN;
	start_iface(ifi);
	}

Should we maybe also require that the link is oper up, i.e. IFF_RUNNING?

If so, that might require us to handle RTM_SETLINK notifications as well:

mcd/src/netlink.c

Lines 48 to 58 in 23cba46

	if (nlh->nlmsg_type == RTM_NEWLINK || nlh->nlmsg_type == RTM_DELLINK) {
	struct ifinfomsg *ifi = (struct ifinfomsg *)NLMSG_DATA(nlh);
	if (nlh->nlmsg_type == RTM_NEWLINK) {
	if (!config_find_iface(ifi->ifi_index, 0))
	iface_add(ifi->ifi_index, ifi->ifi_flags);
	else
	iface_check(ifi->ifi_index, ifi->ifi_flags);
	} else
	iface_del(ifi->ifi_index, ifi->ifi_flags);
	}