feature/routecheck,ipn/routecheck: probe reachability in the background

Previously, refreshing the routecheck.Client would probe to generate a
new routecheck.Report, but this method was only wired up to the
LocalAPI and the `tailscale routecheck` command. However, waiting for
a probe to finish before choosing a router would take too long, so we
must keep a regularly updated report to be consulted as necessary.

This patch adds a Start and Close method to the routecheck.Client and
starts it in the background from features/routecheck. To enable this
feature for a given node, set both of the following node attributes:
`client-side-reachability` and `client-side-reachability-routecheck`.

This patch also wires up the RouterTracker.OnRoutersChange hook, which
fires a callback whenever a new network map includes information about
a router node, This signals to the routecheck.Client that it might
need to schedule another probe, if the shape of the routing table has
changed materially.

Updates #17366
Updates tailscale/corp#33033

Signed-off-by: Simon Law <sfllaw@tailscale.com>

feature/routecheck/localapi.go

@@ -37,18 +37,16 @@ func serveRouteCheck(h *localapi.Handler, w http.ResponseWriter, r *http.Request
 		return
 	}
 
-	var err error
-	var report *routecheck.Report
+	report := rc.Report()
 	if def.Bool(r.FormValue("probe"), false) {
 		timeout := def.Duration(r.FormValue("timeout"), routecheck.DefaultTimeout)
-		timeout = min(max(0, timeout), 60*time.Second) // clamp to [0s, 60s]
-		report, err = rc.Refresh(r.Context(), timeout)
-	} else {
-		report = rc.Report()
-	}
-	if err != nil {
-		localapi.WriteErrorJSON(w, err)
-		return
+		timeout = clampRouteCheckTimeout(timeout)
+		rp, err := rc.Refresh(r.Context(), timeout)
+		if err != nil {
+			localapi.WriteErrorJSON(w, err)
+			return
+		}
+		report = rp
 	}
 
 	w.Header().Set("Content-Type", "application/json")
@@ -61,3 +59,10 @@ func serveRouteCheck(h *localapi.Handler, w http.ResponseWriter, r *http.Request
 	// with its default options, marshal with DefaultOptionsV1.
 	jsonv2.MarshalWrite(w, report, jsonv1.DefaultOptionsV1())
 }
+
+func clampRouteCheckTimeout(timeout time.Duration) time.Duration {
+	if timeout < 0 {
+		timeout = routecheck.DefaultTimeout
+	}
+	return min(max(0, timeout), 60*time.Second) // clamp to [0s, 60s]
+}

feature/routecheck/routecheck.go

@@ -22,6 +22,7 @@ import (
 	"tailscale.com/net/routecheck"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/logger"
+	"tailscale.com/util/eventbus"
 )
 
 // FeatureName is the name of the feature implemented by this package.
@@ -43,6 +44,7 @@ type Extension struct {
 
 	logf    logger.Logf
 	backend ipnext.SafeBackend
+	ec      *eventbus.Client
 	nb      nodeBackender
 	nm      routecheck.NetMapper
 	routers *RouterTracker
@@ -83,7 +85,7 @@ func (e *Extension) Init(h ipnext.Host) error {
 
 	pinger := e.backend.Sys().Engine.Get()
 
-	c, err := routecheck.NewClient(e.logf, e.nb, e.nm, pinger)
+	c, err := routecheck.NewClient(context.Background(), e.logf, e.nb, e.nm, pinger)
 	if err != nil {
 		return err
 	}
@@ -91,7 +93,11 @@ func (e *Extension) Init(h ipnext.Host) error {
 
 	e.routers = TrackRouters(context.Background(), e.logf, ipnbus)
 	e.routers.OnNetMapAvailable = e.Client.NotifyNetMapAvailable
-	e.routers.OnRoutersChange = e.incrementalRefresh
+	e.routers.OnRoutersChange = e.Client.NeedsIncrRefresh
+
+	bus := e.backend.Sys().Bus.Get()
+	e.ec = bus.Client("routecheck")
+	eventbus.SubscribeFunc(e.ec, e.Client.WatchForNetMonRebind)
 
 	// Watch for changes to the self node that would toggle the routecheck feature.
 	e.reconcile.args = make(chan tailcfg.NodeView, 1)
@@ -99,6 +105,9 @@ func (e *Extension) Init(h ipnext.Host) error {
 	go e.reconcileLoop()
 	h.Hooks().OnSelfChange.Add(e.reconcileWatcher)
 
+	// Probe for reachable peers.
+	go e.Client.Start()
+
 	return nil
 }
 
@@ -109,22 +118,12 @@ func (e *Extension) Shutdown() error {
 	close(e.reconcile.args) // lock prevents reconcileWatcher from writing to this channel
 	e.reconcile.Unlock()
 
+	e.ec.Close()
 	e.routers.Close() // stop the watcher before waiting for reconcile.done
 	<-e.reconcile.done
 	return e.Client.Close()
 }
 
-func (e *Extension) needsRefresh() {
-	// TODO(sfllaw): Call e.Client.NeedsRefresh() after implementing it.
-}
-
-func (e *Extension) incrementalRefresh(added, modified, removed []tailcfg.NodeID) {
-	// TODO(sfllaw): This refresh should be incremental,
-	// based on the added, modified, and removed nodes.
-	// Currently it refreshes everything.
-	e.needsRefresh()
-}
-
 // reconcileWatcher is called whenever e.routers should start, stop, or restart its watcher.
 // It may trigger a restart when self indicates that we have switched to a different tailnet or user,
 // in order to reset the internal state of e.routers and start tracking from scratch.
@@ -159,7 +158,7 @@ func (e *Extension) reconcileLoop() {
 			continue // can be started by toggling the nodeattr
 		}
 		if started {
-			e.needsRefresh()
+			e.Client.NeedsRefresh()
 		}
 	}
 }