ipn/wg: use conn age for status calc

Author: ignoramous

intra/backend/ipn_proxies.go

@@ -414,6 +414,8 @@ type RouterStats struct {
 	LastOK int64
 	// last refresh time in millis
 	LastRefresh int64
+	// last re-connection opened in unix millis
+	LastOpen int64
 	// uptime in millis
 	Since int64
 	// Current proxy status

intra/ipn/proxy.go

@@ -724,7 +724,7 @@ func healthy(p Proxy) error {
 
 	stat := p.Router().Stat()
 	now := now()
-	age := now - stat.Since
+	age := now - stat.LastOpen
 
 	oldEnough := age > ageThreshold.Milliseconds()
 	lastOK := stat.LastOK

intra/ipn/wg/wgconn.go

@@ -358,7 +358,7 @@ again:
 		tries++
 		goto again
 	}
-	if err != nil && !no6 {
+	if err != nil && !no6 { // err and supports v6
 		clos(ipv4)
 		return nil, 0, err
 	}
@@ -522,9 +522,9 @@ func (s *StdNetBind) quit() {
 }
 
 func (s *StdNetBind) makeReceiveFn(fd int, uc net.PacketConn) conn.ReceiveFunc {
+	who := strconv.Itoa(fd)
 	// github.com/WireGuard/wireguard-go/blob/469159ecf/device/device.go#L531
 	return func(bufs [][]byte, sizes []int, eps []conn.Endpoint) (n int, err error) {
-		who := strconv.Itoa(fd)
 		defer core.Recover(core.Exit11, "wgconn.recv."+who+"."+s.id)
 
 		anyProcessed := false // true when numMsgs > 0 (ex: no error)
@@ -619,9 +619,10 @@ func (s *StdNetBind) Send(buf [][]byte, peer conn.Endpoint) (err error) {
 	hasAmnezia := amnezia.Set()
 
 	for _, data := range buf {
-		bufok := len(data) > 0
+		datalen := len(data) // grab the length before amnezia overwrites it
+		bufok := datalen > 0
 
-		if false && log.Debug {
+		if false && log.Verbose {
 			log.VV("wg: bind: send: %s (%d) addr(%v) floodwg? %t, blackhole? %t; noconn? %t; hasbuf? %t",
 				s.id, fd, dstIpp, floodWg, blackhole, noconn, bufok)
 		}
@@ -636,8 +637,6 @@ func (s *StdNetBind) Send(buf [][]byte, peer conn.Endpoint) (err error) {
 		anyProcessed = true
 		anyTransportTyp = anyTransportTyp || transportType(data)
 
-		datalen := len(data) // grab the length before we overwrite it
-
 		overwritten = amnezia.send(&data)
 
 		if !flooded && (floodWg || hasAmnezia) {

intra/ipn/wgproxy.go

@@ -150,6 +150,7 @@ type wgtun struct {
 	// TODO: move status to a state-machine for all proxies
 	status          *core.Volatile[int]   // status of this interface
 	statusReason    core.Volatile[string] // last state transition reason
+	latestOpen      atomic.Int64          // last open time in unix millis
 	latestRefresh   atomic.Int64          // last refresh time in unix millis
 	latestPing      atomic.Int64          // last ping time in unix millis
 	latestErr       core.Volatile[error]  // last open/dial err
@@ -1265,6 +1266,7 @@ func (w *wgproxy) Stat() (out *x.RouterStats) {
 	out.LastGoodRx = w.latestGoodRx.Load()
 	out.LastGoodTx = w.latestGoodTx.Load()
 	out.LastRefresh = w.latestRefresh.Load()
+	out.LastOpen = w.latestOpen.Load()
 	out.Since = w.since
 	out.Status = pxstatus(w.status.Load()).String()
 	out.StatusReason = w.statusReason.Load()
@@ -1720,10 +1722,12 @@ func (h *wgtun) listener(op wg.PktDir, err error) (ended bool) {
 		why = "TNT: closed; prev: " + pxstatus(s).String()
 		s = TNT
 		return
+	} else if op == wg.Opn {
+		h.latestOpen.Store(now())
 	}
 
 	now := now()
-	age := now - h.since
+	age := now - h.latestOpen.Load()
 	if err != nil { // failing
 		s = TKO
 		why = "TKO: " + err.Error()
@@ -1738,6 +1742,8 @@ func (h *wgtun) listener(op wg.PktDir, err error) (ended bool) {
 			// on net.ErrClosed, wg stops recieving routine for all peers; this among
 			// other things mean that the wg.Device is effectively down and would not
 			// recieve any incoming messages (nor outgoing as those use the same socket)
+			// Note that, there could be multiple receive functions (not just one) and
+			// the other ones (one each per ip family) may be running just fine.
 			s = TNT
 			why = "TNT: closed " + string(op)
 		}
@@ -1791,7 +1797,7 @@ func (h *wgtun) listener(op wg.PktDir, err error) (ended bool) {
 		hasNewReads := lastRead > age
 
 		// too much time since last good write and good reads
-		readWriteDeviation := (hasNewReads || hasNewWrites) && deviationMs > markTNTAfterMillis
+		readWriteDeviation := (hasNewReads || hasNewWrites) && deviationMs > 2*markTNTAfterMillis
 		// too much time since last attempted read was good
 		readThres := hasNewReads && readElapsedMs > markTNTAfterMillis
 		// too much time since last attempted write was good