`if_conf` can be null if the allocation fails in line 520.

Set the version to be "unknown" when the git repository does not contain
any tags.

Only allocate the neighbour after all the basic sanity checks have been
performed.  This is important in the HMAC branch.

This slows down initial convergence, but avoids a broadcast burst
at boot.  Thanks to Teco Boot for pointing this out.

The purpose is to override *locally* RTA_PREFSRC ('src' in ip-route).

This is only a cleanup and not a functional change

Source prefix can be specified only once.

It was removed by accident in commit ce457a79.

DTLS has recently changed to disallow unprotected IHU, so we now
need to send IHU over unicast.

What was I thinking?

Since this causes flushbuf to take a struct interface, we're basically
carrying a struct interface all over the place.  Good.

Avoid the duplication by passing around an interface pointer.

src_plen for IPv4 routes imported from kernel has not been adjusted
to v4mapped encoding. Therefore IPv4 xroutes used an inconsistent encoding
which lead to failed comparisons when sending updates.

Routes received from neighbors with the same prefix as xroutes therefore
have been announced to neighbours instead.

This issue is fixed by converting src_plen on import.
Signed-off-by: Fabian Bläse <fabian@blaese.de>

This was introduced when the old_if table was made dynamic.

As pointed out by Toke Boot, this causes huge bursts of Hellos when we
encounter many new neighbours at a time.

Thanks to Dave Taht.

This removes an arbitrary limit on the number of interfaces.
Thanks to Christof Schulze.

This avoids a network meltdown in extreme cases.
Thanks to Teco Boot.

When adding a network interface that is already known to babeld, an
error message is displayed and the new configuration is ignored. This
message is adjusted to include the name of the interface being added so
it is visible from the logs without context

This removes a duplicate call to local_notify_interface(ifp, LOCAL_FLUSH)
which lead to duplicate output on the monitoring socket.

It was essentially a no-op with an inverted test.

This reverts commit 005f1197.

Thanks to Dave Taht for the report.