seeker: start doing a channel probe if we see unknown node_announcement msgs.

It usually means we're missing something, but there's no way to ask what.
Simply start a broad scid probe.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

gossipd/gossip_generation.c

@@ -202,7 +202,8 @@ static void update_own_node_announcement(struct daemon *daemon)
 
 	/* This injects it into the routing code in routing.c; it should not
 	 * reject it! */
-	err = handle_node_announcement(daemon->rstate, take(nannounce), NULL);
+	err = handle_node_announcement(daemon->rstate, take(nannounce),
+				       NULL, NULL);
 	if (err)
 		status_failed(STATUS_FAIL_INTERNAL_ERROR,
 			      "rejected own node announcement: %s",

gossipd/gossip_store.c

@@ -721,7 +721,7 @@ u32 gossip_store_load(struct routing_state *rstate, struct gossip_store *gs)
 		case WIRE_NODE_ANNOUNCEMENT:
 			if (!routing_add_node_announcement(rstate,
 							   take(msg), gs->len,
-							   NULL)) {
+							   NULL, NULL)) {
 				bad = "Bad node_announcement";
 				goto badmsg;
 			}

gossipd/gossipd.c

@@ -395,6 +395,18 @@ out:
 	return true;
 }
 
+static u8 *handle_node_announce(struct peer *peer, const u8 *msg)
+{
+	bool was_unknown = false;
+	u8 *err;
+
+	err = handle_node_announcement(peer->daemon->rstate, msg, peer,
+				       &was_unknown);
+	if (was_unknown)
+		query_unknown_node(peer->daemon->seeker, peer);
+	return err;
+}
+
 /*~ This is where the per-peer daemons send us messages.  It's either forwarded
  * gossip, or a request for information.  We deliberately use non-overlapping
  * message types so we can distinguish them. */
@@ -414,7 +426,7 @@ static struct io_plan *peer_msg_in(struct io_conn *conn,
 		err = handle_channel_update_msg(peer, msg);
 		goto handled_relay;
 	case WIRE_NODE_ANNOUNCEMENT:
-		err = handle_node_announcement(peer->daemon->rstate, msg, peer);
+		err = handle_node_announce(peer, msg);
 		goto handled_relay;
 	case WIRE_QUERY_CHANNEL_RANGE:
 		err = handle_query_channel_range(peer, msg);

gossipd/routing.c

@@ -1517,7 +1517,7 @@ static void process_pending_node_announcement(struct routing_state *rstate,
 		if (!routing_add_node_announcement(rstate,
 						   pna->node_announcement,
 						   pna->index,
-						   pna->peer_softref))
+						   pna->peer_softref, NULL))
 			status_unusual("pending node_announcement %s too old?",
 				       tal_hex(tmpctx, pna->node_announcement));
 		/* Never send this again. */
@@ -2390,7 +2390,8 @@ struct wireaddr *read_addresses(const tal_t *ctx, const u8 *ser)
 bool routing_add_node_announcement(struct routing_state *rstate,
 				   const u8 *msg TAKES,
 				   u32 index,
-				   struct peer *peer)
+				   struct peer *peer,
+				   bool *was_unknown)
 {
 	struct node *node;
 	secp256k1_ecdsa_signature signature;
@@ -2400,6 +2401,9 @@ bool routing_add_node_announcement(struct routing_state *rstate,
 	u8 alias[32];
 	u8 *features, *addresses;
 
+	if (was_unknown)
+		*was_unknown = false;
+
 	/* Make sure we own msg, even if we don't save it. */
 	if (taken(msg))
 		tal_steal(tmpctx, msg);
@@ -2442,6 +2446,8 @@ bool routing_add_node_announcement(struct routing_state *rstate,
 		pna = pending_node_map_get(rstate->pending_node_map,
 					   &node_id);
 		if (!pna) {
+			if (was_unknown)
+				*was_unknown = true;
 			bad_gossip_order(msg, peer,
 					 type_to_string(tmpctx, struct node_id,
 							&node_id));
@@ -2525,7 +2531,7 @@ bool routing_add_node_announcement(struct routing_state *rstate,
 }
 
 u8 *handle_node_announcement(struct routing_state *rstate, const u8 *node_ann,
-			     struct peer *peer)
+			     struct peer *peer, bool *was_unknown)
 {
 	u8 *serialized;
 	struct sha256_double hash;
@@ -2538,6 +2544,9 @@ u8 *handle_node_announcement(struct routing_state *rstate, const u8 *node_ann,
 	struct wireaddr *wireaddrs;
 	size_t len = tal_count(node_ann);
 
+	if (was_unknown)
+		*was_unknown = false;
+
 	serialized = tal_dup_arr(tmpctx, u8, node_ann, len, 0);
 	if (!fromwire_node_announcement(tmpctx, serialized,
 					&signature, &features, &timestamp,
@@ -2613,7 +2622,7 @@ u8 *handle_node_announcement(struct routing_state *rstate, const u8 *node_ann,
 	}
 
 	/* May still fail, if we don't know the node. */
-	routing_add_node_announcement(rstate, serialized, 0, peer);
+	routing_add_node_announcement(rstate, serialized, 0, peer, was_unknown);
 	return NULL;
 }
 

gossipd/routing.h

@@ -392,9 +392,11 @@ u8 *handle_channel_update(struct routing_state *rstate, const u8 *update TAKES,
 			  struct peer *peer,
 			  struct short_channel_id *unknown_scid);
 
-/* Returns NULL if all OK, otherwise an error for the peer which sent. */
+/* Returns NULL if all OK, otherwise an error for the peer which sent.
+ * If was_unknown is not NULL, sets it to true if that was the reason for
+ * the error: the node was unknown to us. */
 u8 *handle_node_announcement(struct routing_state *rstate, const u8 *node,
-			     struct peer *peer);
+			     struct peer *peer, bool *was_unknown);
 
 /* Get a node: use this instead of node_map_get() */
 struct node *get_node(struct routing_state *rstate,
@@ -460,7 +462,8 @@ bool routing_add_channel_update(struct routing_state *rstate,
 bool routing_add_node_announcement(struct routing_state *rstate,
 				   const u8 *msg TAKES,
 				   u32 index,
-				   struct peer *peer);
+				   struct peer *peer,
+				   bool *was_unknown);
 
 
 /**

gossipd/seeker.c

@@ -96,6 +96,10 @@ struct seeker {
 	u8 *nannounce_query_flags;
 	size_t nannounce_offset;
 
+	/* Are there any node_ids we didn't know?  Implies we're
+	 * missing channels. */
+	bool unknown_nodes;
+
 	/* Peers we've asked to stream us gossip */
 	struct peer *gossiper_softref[3];
 
@@ -148,6 +152,7 @@ struct seeker *new_seeker(struct daemon *daemon)
 	for (size_t i = 0; i < ARRAY_SIZE(seeker->gossiper_softref); i++)
 		seeker->gossiper_softref[i] = NULL;
 	seeker->preferred_peer_softref = NULL;
+	seeker->unknown_nodes = false;
 	set_state(seeker, STARTING_UP);
 	begin_check_timer(seeker);
 	memleak_add_helper(seeker, memleak_help_seeker);
@@ -186,7 +191,12 @@ static struct peer *random_seeker(struct seeker *seeker,
 {
 	struct peer *peer = seeker->preferred_peer_softref;
 
-	if (peer && check_peer(peer)) {
+	/* 80% chance of immediately choosing a peer who reported the missing
+	 * stuff: they presumably can tell us more about it.  We don't
+	 * *always* choose it because it could be simply spamming us with
+	 * invalid announcements to get chosen, and we don't handle that case
+	 * well yet. */
+	if (peer && check_peer(peer) && pseudorand(5) != 0) {
 		clear_softref(seeker, &seeker->random_peer_softref);
 		return peer;
 	}
@@ -409,6 +419,7 @@ static bool seek_any_stale_scids(struct seeker *seeker)
 	return true;
 }
 
+/* We can't ask for channels by node_id, so probe at random */
 /* Returns true and sets first_blocknum and number_of_blocks if
  * there's more to find. */
 static bool next_block_range(struct seeker *seeker,
@@ -846,6 +857,16 @@ set_gossiper:
 	enable_gossip_stream(seeker, peer);
 }
 
+static bool seek_any_unknown_nodes(struct seeker *seeker)
+{
+	if (!seeker->unknown_nodes)
+		return false;
+
+	seeker->unknown_nodes = false;
+	probe_many_random_scids(seeker);
+	return true;
+}
+
 /* Periodic timer to see how our gossip is going. */
 static void seeker_check(struct seeker *seeker)
 {
@@ -867,8 +888,9 @@ static void seeker_check(struct seeker *seeker)
 		break;
 	case NORMAL:
 		maybe_rotate_gossipers(seeker);
-		if (!seek_any_unknown_scids(seeker))
+		if (!seek_any_unknown_scids(seeker)
-			seek_any_stale_scids(seeker);
+		    && !seek_any_stale_scids(seeker))
+			seek_any_unknown_nodes(seeker);
 		break;
 	}
 
@@ -953,3 +975,10 @@ void query_unknown_channel(struct daemon *daemon,
 	if (!add_unknown_scid(daemon->seeker, id, peer))
 		return;
 }
+
+/* This peer told us about an unknown node.  Start probing it. */
+void query_unknown_node(struct seeker *seeker, struct peer *peer)
+{
+	seeker->unknown_nodes = true;
+	set_preferred_peer(seeker, peer);
+}

gossipd/seeker.h

@@ -12,6 +12,8 @@ void query_unknown_channel(struct daemon *daemon,
 			   struct peer *peer,
 			   const struct short_channel_id *id);
 
+void query_unknown_node(struct seeker *seeker, struct peer *peer);
+
 void seeker_setup_peer_gossip(struct seeker *seeker, struct peer *peer);
 
 bool remove_unknown_scid(struct seeker *seeker,

gossipd/test/run-crc32_of_update.c

@@ -58,7 +58,7 @@ u8 *handle_channel_update(struct routing_state *rstate UNNEEDED, const u8 *updat
 { fprintf(stderr, "handle_channel_update called!\n"); abort(); }
 /* Generated stub for handle_node_announcement */
 u8 *handle_node_announcement(struct routing_state *rstate UNNEEDED, const u8 *node UNNEEDED,
-			     struct peer *peer UNNEEDED)
+			     struct peer *peer UNNEEDED, bool *was_unknown UNNEEDED)
 { fprintf(stderr, "handle_node_announcement called!\n"); abort(); }
 /* Generated stub for master_badmsg */
 void master_badmsg(u32 type_expected UNNEEDED, const u8 *msg)