Once initialized routing restrictions are found in the array of
ipr_res_table structures called "ipr_restrictions".  The index is the
network number as found in the "net" struct (net.n_net -- network
index number), perhaps it should be net.n_int (interface number), but
looking at how these are initialized in gw/src/gate.c lead me to use
n_net.  "ipr_restrictions" is allocated in "initintabs" in
gw/in/inrte.c, and initialized in "init_restrictions" which is called
per net (1 to "nnets") from "initintabs".

A routing protocol when running calls either res_ignore_route or
res_advertise_route with a protocol number (currently either F_RES_RIP
or F_RES_EGP, defined in gw/in/inrte.h), a network address, and an
network index.  The routing looks in "ipr_restrictions[net_index]" to
find the appropriate info and looks up that network using the hash
table stuff, return a take/punt or squawk/shut up decision.

Do to the way "ipr_res_entry" is defined, you cannot have different
restrictions on a given interface for more than one protocol.  E.g. on
interface 1 you cannot listen to 18.80.0.0 via RIP and also not
announce it via EGP.  This could be fixed by having a separate flag
word per-protocol, but for now I wanted to keep it simple.  Or you
could allow more than one entry per network.  This is probably what
you want to do if you want to add an address field so you can
distinguish among protocol speakers on a interface (would proably be
most useful for EGP).  You would have to modify the hash routines not
to do lookups first on creates, and to take some more arguments for
lookups (i.e. you have to skip by wrong protocol/addresses and keep
checking until you come to the end of the hash chain or until you get
a match).

EGP doesn't use restrictions for advertising but does for incoming
routes (though GW won't use it for now).

The metric stuff is really intended to be used if EGP routes ever get
turned into RIP routes.