ipn/{ipnlocal,localapi},wgengine{,/magicsock}: plumb health.Tracker

Down to 25 health.Global users. After this remains controlclient &
net/dns & wgengine/router.

Updates #11874
Updates #4136

Change-Id: I6dd1856e3d9bf523bdd44b60fb3b8f7501d5dc0d
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>

wgengine/magicsock/derp.go

@@ -165,7 +165,7 @@ func (c *Conn) maybeSetNearestDERP(report *netcheck.Report) (preferredDERP int)
 	if testenv.InTest() && !checkControlHealthDuringNearestDERPInTests {
 		connectedToControl = true
 	} else {
-		connectedToControl = health.Global.GetInPollNetMap()
+		connectedToControl = c.health.GetInPollNetMap()
 	}
 	if !connectedToControl {
 		c.mu.Lock()
@@ -201,12 +201,12 @@ func (c *Conn) setNearestDERP(derpNum int) (wantDERP bool) {
 	defer c.mu.Unlock()
 	if !c.wantDerpLocked() {
 		c.myDerp = 0
-		health.Global.SetMagicSockDERPHome(0, c.homeless)
+		c.health.SetMagicSockDERPHome(0, c.homeless)
 		return false
 	}
 	if c.homeless {
 		c.myDerp = 0
-		health.Global.SetMagicSockDERPHome(0, c.homeless)
+		c.health.SetMagicSockDERPHome(0, c.homeless)
 		return false
 	}
 	if derpNum == c.myDerp {
@@ -217,7 +217,7 @@ func (c *Conn) setNearestDERP(derpNum int) (wantDERP bool) {
 		metricDERPHomeChange.Add(1)
 	}
 	c.myDerp = derpNum
-	health.Global.SetMagicSockDERPHome(derpNum, c.homeless)
+	c.health.SetMagicSockDERPHome(derpNum, c.homeless)
 
 	if c.privateKey.IsZero() {
 		// No private key yet, so DERP connections won't come up anyway.
@@ -400,7 +400,7 @@ func (c *Conn) derpWriteChanOfAddr(addr netip.AddrPort, peer key.NodePublic) cha
 		}
 		return derpMap.Regions[regionID]
 	})
-	dc.HealthTracker = health.Global
+	dc.HealthTracker = c.health
 
 	dc.SetCanAckPings(true)
 	dc.NotePreferred(c.myDerp == regionID)
@@ -526,8 +526,8 @@ func (c *Conn) runDerpReader(ctx context.Context, derpFakeAddr netip.AddrPort, d
 		return n
 	}
 
-	defer health.Global.SetDERPRegionConnectedState(regionID, false)
-	defer health.Global.SetDERPRegionHealth(regionID, "")
+	defer c.health.SetDERPRegionConnectedState(regionID, false)
+	defer c.health.SetDERPRegionHealth(regionID, "")
 
 	// peerPresent is the set of senders we know are present on this
 	// connection, based on messages we've received from the server.
@@ -539,7 +539,7 @@ func (c *Conn) runDerpReader(ctx context.Context, derpFakeAddr netip.AddrPort, d
 	for {
 		msg, connGen, err := dc.RecvDetail()
 		if err != nil {
-			health.Global.SetDERPRegionConnectedState(regionID, false)
+			c.health.SetDERPRegionConnectedState(regionID, false)
 			// Forget that all these peers have routes.
 			for peer := range peerPresent {
 				delete(peerPresent, peer)
@@ -577,14 +577,14 @@ func (c *Conn) runDerpReader(ctx context.Context, derpFakeAddr netip.AddrPort, d
 
 		now := time.Now()
 		if lastPacketTime.IsZero() || now.Sub(lastPacketTime) > frameReceiveRecordRate {
-			health.Global.NoteDERPRegionReceivedFrame(regionID)
+			c.health.NoteDERPRegionReceivedFrame(regionID)
 			lastPacketTime = now
 		}
 
 		switch m := msg.(type) {
 		case derp.ServerInfoMessage:
-			health.Global.SetDERPRegionConnectedState(regionID, true)
-			health.Global.SetDERPRegionHealth(regionID, "") // until declared otherwise
+			c.health.SetDERPRegionConnectedState(regionID, true)
+			c.health.SetDERPRegionHealth(regionID, "") // until declared otherwise
 			c.logf("magicsock: derp-%d connected; connGen=%v", regionID, connGen)
 			continue
 		case derp.ReceivedPacket:
@@ -624,7 +624,7 @@ func (c *Conn) runDerpReader(ctx context.Context, derpFakeAddr netip.AddrPort, d
 			}()
 			continue
 		case derp.HealthMessage:
-			health.Global.SetDERPRegionHealth(regionID, m.Problem)
+			c.health.SetDERPRegionHealth(regionID, m.Problem)
 			continue
 		case derp.PeerGoneMessage:
 			switch m.Reason {

wgengine/magicsock/magicsock.go

@@ -91,6 +91,7 @@ type Conn struct {
 	testOnlyPacketListener nettype.PacketListener
 	noteRecvActivity       func(key.NodePublic) // or nil, see Options.NoteRecvActivity
 	netMon                 *netmon.Monitor      // or nil
+	health                 *health.Tracker      // or nil
 	controlKnobs           *controlknobs.Knobs  // or nil
 
 	// ================================================================
@@ -369,9 +370,13 @@ type Options struct {
 	NoteRecvActivity func(key.NodePublic)
 
 	// NetMon is the network monitor to use.
-	// With one, the portmapper won't be used.
+	// If nil, the portmapper won't be used.
 	NetMon *netmon.Monitor
 
+	// HealthTracker optionally specifies the health tracker to
+	// report errors and warnings to.
+	HealthTracker *health.Tracker
+
 	// ControlKnobs are the set of control knobs to use.
 	// If nil, they're ignored and not updated.
 	ControlKnobs *controlknobs.Knobs
@@ -463,6 +468,7 @@ func NewConn(opts Options) (*Conn, error) {
 		c.portMapper.SetGatewayLookupFunc(opts.NetMon.GatewayAndSelfIP)
 	}
 	c.netMon = opts.NetMon
+	c.health = opts.HealthTracker
 	c.onPortUpdate = opts.OnPortUpdate
 	c.getPeerByKey = opts.PeerByKeyFunc
 
@@ -666,7 +672,7 @@ func (c *Conn) updateNetInfo(ctx context.Context) (*netcheck.Report, error) {
 		// NOTE(andrew-d): I don't love that we're depending on the
 		// health package here, but I'd rather do that and not store
 		// the exact same state in two different places.
-		GetLastDERPActivity: health.Global.GetDERPRegionReceivedTime,
+		GetLastDERPActivity: c.health.GetDERPRegionReceivedTime,
 	})
 	if err != nil {
 		return nil, err
@@ -2471,7 +2477,7 @@ func (c *Conn) bindSocket(ruc *RebindingUDPConn, network string, curPortFate cur
 		}
 		ruc.setConnLocked(pconn, network, c.bind.BatchSize())
 		if network == "udp4" {
-			health.Global.SetUDP4Unbound(false)
+			c.health.SetUDP4Unbound(false)
 		}
 		return nil
 	}
@@ -2482,7 +2488,7 @@ func (c *Conn) bindSocket(ruc *RebindingUDPConn, network string, curPortFate cur
 	// we get a link change and we can try binding again.
 	ruc.setConnLocked(newBlockForeverConn(), "", c.bind.BatchSize())
 	if network == "udp4" {
-		health.Global.SetUDP4Unbound(true)
+		c.health.SetUDP4Unbound(true)
 	}
 	return fmt.Errorf("failed to bind any ports (tried %v)", ports)
 }

wgengine/magicsock/magicsock_test.go

@@ -3113,21 +3113,23 @@ func TestMaybeSetNearestDERP(t *testing.T) {
 	}
 	for _, tt := range testCases {
 		t.Run(tt.name, func(t *testing.T) {
+			ht := new(health.Tracker)
 			c := newConn()
 			c.logf = t.Logf
 			c.myDerp = tt.old
 			c.derpMap = derpMap
+			c.health = ht
 
 			report := &netcheck.Report{PreferredDERP: tt.reportDERP}
 
-			oldConnected := health.Global.GetInPollNetMap()
+			oldConnected := ht.GetInPollNetMap()
 			if tt.connectedToControl != oldConnected {
 				if tt.connectedToControl {
-					health.Global.GotStreamedMapResponse()
-					t.Cleanup(health.Global.SetOutOfPollNetMap)
+					ht.GotStreamedMapResponse()
+					t.Cleanup(ht.SetOutOfPollNetMap)
 				} else {
-					health.Global.SetOutOfPollNetMap()
-					t.Cleanup(health.Global.GotStreamedMapResponse)
+					ht.SetOutOfPollNetMap()
+					t.Cleanup(ht.GotStreamedMapResponse)
 				}
 			}
 

wgengine/userspace.go

@@ -98,6 +98,7 @@ type userspaceEngine struct {
 	dns              *dns.Manager
 	magicConn        *magicsock.Conn
 	netMon           *netmon.Monitor
+	health           *health.Tracker
 	netMonOwned      bool                // whether we created netMon (and thus need to close it)
 	netMonUnregister func()              // unsubscribes from changes; used regardless of netMonOwned
 	birdClient       BIRDClient          // or nil
@@ -188,6 +189,9 @@ type Config struct {
 	// If nil, a new network monitor is created.
 	NetMon *netmon.Monitor
 
+	// HealthTracker, if non-nil, is the health tracker to use.
+	HealthTracker *health.Tracker
+
 	// Dialer is the dialer to use for outbound connections.
 	// If nil, a new Dialer is created
 	Dialer *tsdial.Dialer
@@ -310,6 +314,7 @@ func NewUserspaceEngine(logf logger.Logf, conf Config) (_ Engine, reterr error)
 		birdClient:     conf.BIRDClient,
 		controlKnobs:   conf.ControlKnobs,
 		reconfigureVPN: conf.ReconfigureVPN,
+		health:         conf.HealthTracker,
 	}
 
 	if e.birdClient != nil {
@@ -372,6 +377,7 @@ func NewUserspaceEngine(logf logger.Logf, conf Config) (_ Engine, reterr error)
 		IdleFunc:         e.tundev.IdleDuration,
 		NoteRecvActivity: e.noteRecvActivity,
 		NetMon:           e.netMon,
+		HealthTracker:    e.health,
 		ControlKnobs:     conf.ControlKnobs,
 		OnPortUpdate:     onPortUpdate,
 		PeerByKeyFunc:    e.PeerByKey,
@@ -970,7 +976,7 @@ func (e *userspaceEngine) Reconfig(cfg *wgcfg.Config, routerCfg *router.Config,
 		e.logf("wgengine: Reconfig: configuring router")
 		e.networkLogger.ReconfigRoutes(routerCfg)
 		err := e.router.Set(routerCfg)
-		health.Global.SetRouterHealth(err)
+		e.health.SetRouterHealth(err)
 		if err != nil {
 			return err
 		}
@@ -979,7 +985,7 @@ func (e *userspaceEngine) Reconfig(cfg *wgcfg.Config, routerCfg *router.Config,
 		// assigned address.
 		e.logf("wgengine: Reconfig: configuring DNS")
 		err = e.dns.Set(*dnsCfg)
-		health.Global.SetDNSHealth(err)
+		e.health.SetDNSHealth(err)
 		if err != nil {
 			return err
 		}
@@ -1183,7 +1189,7 @@ func (e *userspaceEngine) linkChange(delta *netmon.ChangeDelta) {
 		e.logf("[v1] LinkChange: minor")
 	}
 
-	health.Global.SetAnyInterfaceUp(up)
+	e.health.SetAnyInterfaceUp(up)
 	e.magicConn.SetNetworkUp(up)
 	if !up || changed {
 		if err := e.dns.FlushCaches(); err != nil {