> I suppose that, as long as manually adding the route overrides the default
> behavior and makes RSPF work, this is best seen as a documentation issue.
TD> It's really a non-issue for 2.0.* kernels since you were
TD> always supposed to manually add the routing. And for
TD> late-2.1.* kernels the route is added automatically. The
TD> issue is only one of having rspfd modified to read the other
TD> /proc/net/ table in addition to /proc/net/route to ensure it
TD> is aware of the implicit routes.
The problem is what to do with concurrent alternative routes to the same place,
which is a principal issue in dynamic routing. The whole concept of fault
tolerance depends upon this.
As I think I tried to address in an earlier message to this list, I don't view
it as appropriate to treat a radio interface like an Ethernet interface,
because some number of hosts "within the local subnet" may not, in fact, be
directly reachable. One might argue that such an entity is not a subnet, but
the problem is really inherent in the radio medium and cannot be gotten around.
Typically, a radio LAN consists of a group of hosts with an address block which
all communicate with the outside world through a router/gateway/switch. Since
the outside world can treat them as a subnet accessible in the traditional
manner, we use the conventions of subnets such as contiguous addresses, a
netmask, and so on. However, this thing which looks like a subnet to the
outside world does not operate internally as such, because the nature of radio
makes that effectively impossible.
Because of this inherent behavior of radio, RSPF can and should be used to
bring order to the chaos of the internal routing within the subnet. Hosts can
and should be picked up dynamically and relayed by other hosts on the air, and
it is a principal goal of RSPF to provide the framework to do this.
In my view, it is a mistake for the Linux kernel to enshrine assumptions about
how subnets work internally, since those assumptions are demonstrably wrong
when applied to certain types of interfaces, particularly radio. There is no
question that this is going to make life difficult for RSPF, perhaps much more
so than we realize at this time.
-- Mike, N1BEE