rtnetmon/main.go

//go:build linux
// +build linux

// netmon – dual-interface network dashboard (curses)
// WTFPL – 2025-05-16  sneak@sneak.berlin
package main

import (
	"context"
	"encoding/json"
	"flag"
	"fmt"
	"math"
	"net"
	"os"
	"os/exec"
	"os/signal"
	"strconv"
	"strings"
	"sync"
	"syscall"
	"time"

	tcell "github.com/gdamore/tcell/v2"
)

/*────────────────── CLI flags ──────────────────*/

var (
	ifaceA = flag.String("ifaceA", "gu0", "primary network interface")
	labelA = flag.String("labelA", "gu LAN - VPN outbound", "label for ifaceA")
	ifaceB = flag.String("ifaceB", "backhaul0", "secondary network interface")
	labelB = flag.String("labelB", "Cox cable direct", "label for ifaceB")

	hostCSV = flag.String("hosts",
		"1.1.1.1,8.8.8.8,8.8.4.4,google.com,github.com,"+
			"console.aws.amazon.com,console.cloud.google.com,"+
			"fast.com,cloudflare.com,datavi.be",
		"comma-separated reachability hosts")

	logFile = flag.String("logfile", "/tmp/rtnetmon.log", "path to log file")
)

/*────────────────── constants ──────────────────*/

const (
	icmpTimeout      = 500 * time.Millisecond
	tcpTimeout       = 500 * time.Millisecond
	packetLossPings  = 20
	packetLossPeriod = 5 * time.Second
	statsHistory     = 300
	screenRefresh    = 500 * time.Millisecond // Still used as backup refresh rate

	// Unicode characters for the meter
	greenDot      = '🟢' // Unicode green circle
	redDot        = '🔴' // Unicode red circle
	emptyDot      = '⚪' // Unicode white circle
	meterWidth    = 6   // Fixed width of the meter
	maxMeterValue = 10  // Maximum value for the meter
)

/*────────────────── runtime struct ─────────────*/

type InterfaceStatus struct {
	Name, Label, IPInfo        string
	Reachable                  map[string]bool
	Loss                       map[string]float64
	TCP                        map[string][]float64
	TotalICMPReq, TotalICMPRep int
	DroppedCount               int
	LastDrop, LastPing         time.Time
	SpinFrame                  int
	MeterValue                 int // Current value for the packet loss meter
	mu                         sync.RWMutex
}

/*────────────────── colour styles ──────────────*/

var (
	cBrightGreen = tcell.StyleDefault.Foreground(tcell.ColorGreen).Bold(true)
	cGreen       = tcell.StyleDefault.Foreground(tcell.ColorGreen)
	cYellow      = tcell.StyleDefault.Foreground(tcell.ColorYellow)
	cRed         = tcell.StyleDefault.Foreground(tcell.ColorRed)
	cBrightRed   = tcell.StyleDefault.Foreground(tcell.ColorRed).Bold(true)
	cDefault     = tcell.StyleDefault
)

func styleLatency(ms float64) tcell.Style {
	switch {
	case ms < 50:
		return cBrightGreen
	case ms < 100:
		return cGreen
	case ms < 200:
		return cYellow
	default:
		return cRed
	}
}
func styleLoss(p float64) tcell.Style {
	switch {
	case p == 0:
		return cBrightGreen
	case p < 5:
		return cYellow
	default:
		return cBrightRed
	}
}

/*────────────────── math helpers ───────────────*/

func minMaxAvgStd(xs []float64) (min, max, avg, std float64) {
	if len(xs) == 0 {
		return
	}
	min, max = xs[0], xs[0]
	var sum float64
	for _, v := range xs {
		if v < min {
			min = v
		}
		if v > max {
			max = v
		}
		sum += v
	}
	avg = sum / float64(len(xs))
	var vs float64
	for _, v := range xs {
		d := v - avg
		vs += d * d
	}
	std = math.Sqrt(vs / float64(len(xs)))
	return
}

/*────────────────── external lookup ────────────*/

type ipInfoResp struct{ IP, Hostname, Org string }

func fetchIPInfo(iface string) string {
	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
	defer cancel()
	out, _ := exec.CommandContext(ctx, "curl", "-s", "--interface", iface, "--max-time", "2", "ipinfo.io").Output()
	var r ipInfoResp
	_ = json.Unmarshal(out, &r)
	if r.IP == "" {
		return "(ipinfo error)"
	}
	return fmt.Sprintf("%s [%s] %s", r.IP, r.Hostname, r.Org)
}

/*────────────────── ICMP helpers ───────────────*/

func pingOnce(iface, host string) bool {
	ctx, cancel := context.WithTimeout(context.Background(), icmpTimeout)
	defer cancel()
	return exec.CommandContext(ctx, "ping", "-I", iface, "-c1", "-W1", host).Run() == nil
}
func lossPercent(iface, host string) float64 {
	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
	defer cancel()
	out, err := exec.CommandContext(ctx, "ping", "-q", "-i", "0.05",
		"-c", fmt.Sprint(packetLossPings), "-W1", "-I", iface, host).CombinedOutput()
	if err != nil {
		return 1.0
	}
	for _, ln := range strings.Split(string(out), "\n") {
		if strings.Contains(ln, "packet loss") {
			for _, f := range strings.Fields(ln) {
				if strings.HasSuffix(f, "%") {
					p, _ := strconv.ParseFloat(strings.TrimSuffix(f, "%"), 64)
					return p / 100.0
				}
			}
		}
	}
	return 1.0
}

/*────────────────── TCP helpers ───────────────*/

func localAddr(iface string) (net.Addr, error) {
	ifi, err := net.InterfaceByName(iface)
	if err != nil {
		return nil, err
	}
	add, _ := ifi.Addrs()
	for _, a := range add {
		if ipnet, ok := a.(*net.IPNet); ok && ipnet.IP.To4() != nil {
			return &net.TCPAddr{IP: ipnet.IP}, nil
		}
	}
	return nil, fmt.Errorf("no IPv4 on %s", iface)
}
func tcpDuration(iface, hp string) time.Duration {
	la, err := localAddr(iface)
	if err != nil {
		return tcpTimeout
	}
	d := net.Dialer{Timeout: tcpTimeout, LocalAddr: la}
	st := time.Now()
	c, err := d.Dial("tcp", hp)
	if err != nil {
		return tcpTimeout
	}
	c.Close()
	return time.Since(st)
}

/*────────────────── spinner ────────────────────*/

var spins = []rune{'|', '/', '-', '\\'}

// Update to add a channel for signaling UI updates
var uiUpdateChan = make(chan struct{}, 100) // Buffered channel to avoid blocking

// Advances the spinner frame and signals a UI update
// This is called only when packets are successfully received,
// so the spinner only moves when there's actual network activity
func (st *InterfaceStatus) spin() {
	st.SpinFrame = (st.SpinFrame + 1) % len(spins)
	// Signal UI update after spinner changes
	select {
	case uiUpdateChan <- struct{}{}:
	default:
		// Non-blocking send - if channel is full, just continue
	}
}

/*────────────────── screen helpers ─────────────*/

func put(scr tcell.Screen, x, y int, txt string, st tcell.Style) {
	for i, r := range txt {
		scr.SetContent(x+i, y, r, nil, st)
	}
}
func hline(w int) string { return strings.Repeat("=", w) }

/*────────────────── UI drawing ─────────────────*/

func ifaceHealthy(st *InterfaceStatus) bool {
	st.mu.RLock()
	defer st.mu.RUnlock()
	if st.DroppedCount > 0 {
		return false
	}
	for _, ok := range st.Reachable {
		if !ok {
			return false
		}
	}
	for _, lp := range st.Loss {
		if lp > 0 {
			return false
		}
	}
	for _, hp := range tcpTestHosts {
		h := st.TCP[hp]
		if len(h) == 0 || h[len(h)-1] >= float64(tcpTimeout.Milliseconds()) {
			return false
		}
	}
	return true
}

const (
	hostW = 30 // Increased width for host column
	numW  = 7
	stdW  = 8
	nW    = 6
)

func drawIface(scr tcell.Screen, y, w int, st *InterfaceStatus) int {
	put(scr, 0, y, hline(w), cDefault)

	// header line
	healthy := ifaceHealthy(st)
	style := cBrightGreen
	if !healthy {
		style = cBrightRed
	}
	st.mu.RLock()
	header := fmt.Sprintf("%s: %s —   %s", st.Name, st.Label, st.IPInfo)
	spin := spins[st.SpinFrame]
	meterValue := st.MeterValue
	st.mu.RUnlock()

	// Create the meter - fixed width of 6 characters
	meter := createMeter(meterValue, meterWidth)

	put(scr, 0, y+1, "== ", cDefault)
	put(scr, 3, y+1, string(spin)+" ", cDefault)
	put(scr, 5, y+1, meter+" ", cDefault)        // Add the meter after the spinner
	put(scr, 5+meterWidth+1, y+1, header, style) // Move the header after the meter
	put(scr, 0, y+2, hline(w), cDefault)
	y += 4

	/* reachability */
	st.mu.RLock()
	total := len(st.Reachable)
	good := 0
	for _, ok := range st.Reachable {
		if ok {
			good++
		}
	}
	age := time.Since(st.LastPing).Round(time.Second)
	reachStyle := cBrightGreen
	if good != total {
		reachStyle = cBrightRed
	}
	put(scr, 0, y, fmt.Sprintf("Reachable: %d/%d (at %s, age %s)",
		good, total, st.LastPing.Format("15:04:05"), age), reachStyle)
	y++
	if good == total {
		put(scr, 0, y, "Unreachable: none", cDefault)
	} else {
		var down []string
		for h, ok := range st.Reachable {
			if !ok {
				down = append(down, h)
			}
		}
		put(scr, 0, y, "Unreachable: "+strings.Join(down, ", "), cBrightRed)
	}
	y += 2

	/* packet loss */
	put(scr, 0, y, "Packet Loss:", cDefault)
	y++
	for _, h := range packetLossHosts {
		p := st.Loss[h] * 100
		put(scr, 0, y, fmt.Sprintf("%-16s %5.0f%%", h+":", p), styleLoss(p))
		y++
	}
	dAge := "N/A"
	if !st.LastDrop.IsZero() {
		dAge = time.Since(st.LastDrop).Round(time.Second).String()
	}
	put(scr, 0, y, fmt.Sprintf("Dropped: %d (last at %s, age %s)",
		st.DroppedCount, st.LastDrop.Format("15:04:05"), dAge), cDefault)
	y += 2

	/* TCP table */
	put(scr, 0, y, "TCP Connect Stats:", cDefault)
	y++
	// header row
	headerRow := fmt.Sprintf("%-*s %*s %*s %*s %*s %*s %*s",
		hostW, "Host", numW, "last", numW, "min", numW, "avg",
		numW, "max", stdW, "stddev", nW, "n")
	put(scr, 0, y, headerRow, cDefault)
	y++

	for _, hp := range tcpTestHosts {
		hist := st.TCP[hp]
		if len(hist) == 0 {
			continue
		}
		last := hist[len(hist)-1]
		mi, ma, av, sd := minMaxAvgStd(hist)

		row := fmt.Sprintf("%-*s %*s %*s %*s %*s %*s %*d",
			hostW, hp,
			numW, fmt.Sprintf("%.0fms", last),
			numW, fmt.Sprintf("%.0fms", mi),
			numW, fmt.Sprintf("%.0fms", av),
			numW, fmt.Sprintf("%.0fms", ma),
			stdW, fmt.Sprintf("%.0fms", sd),
			nW, len(hist),
		)
		put(scr, 0, y, row, cDefault)
		// colourise individual numbers
		put(scr, hostW+1, y, fmt.Sprintf("%*s", numW, fmt.Sprintf("%.0fms", last)), styleLatency(last))
		put(scr, hostW+1+numW+1, y, fmt.Sprintf("%*s", numW, fmt.Sprintf("%.0fms", mi)), styleLatency(mi))
		put(scr, hostW+1+numW*2+2, y, fmt.Sprintf("%*s", numW, fmt.Sprintf("%.0fms", av)), styleLatency(av))
		put(scr, hostW+1+numW*3+3, y, fmt.Sprintf("%*s", numW, fmt.Sprintf("%.0fms", ma)), styleLatency(ma))
		put(scr, hostW+1+numW*4+4, y, fmt.Sprintf("%*s", stdW, fmt.Sprintf("%.0fms", sd)), cDefault)
		y++
	}
	y++

	/* totals */
	put(scr, 0, y, fmt.Sprintf("Total ICMP Requests: %d", st.TotalICMPReq), cDefault)
	y++
	put(scr, 0, y, fmt.Sprintf("Total ICMP Replies:  %d", st.TotalICMPRep), cDefault)
	y += 2
	st.mu.RUnlock()
	return y
}

// Create a visual meter using green and red dots
func createMeter(value, width int) string {
	if value > maxMeterValue {
		value = maxMeterValue
	}
	if value < 0 {
		value = 0
	}

	// Calculate how many dots of each color to show
	redDots := value
	if redDots > width {
		redDots = width
	}
	greenDots := width - redDots

	// Build the meter string
	var b strings.Builder

	// Add red dots for unreplied packets
	for i := 0; i < redDots; i++ {
		b.WriteRune(redDot)
	}

	// Add green dots for available capacity
	for i := 0; i < greenDots; i++ {
		b.WriteRune(greenDot)
	}

	return b.String()
}

/*────────────────── goroutine loops ─────────────*/

func reachLoop(ctx context.Context, st *InterfaceStatus, hosts []string) {
	logf("Starting reachability monitoring for %s with %d hosts", st.Name, len(hosts))
	tk := time.NewTicker(time.Second)
	defer tk.Stop()
	for {
		select {
		case <-ctx.Done():
			logf("Stopping reachability monitoring for %s", st.Name)
			return
		case <-tk.C:
			// logf("Checking reachability for %s", st.Name) // Too verbose
			var wg sync.WaitGroup
			res := make(map[string]bool, len(hosts))
			mu := sync.Mutex{}
			for _, h := range hosts {
				wg.Add(1)
				go func(host string) {
					defer wg.Done()
					st.mu.Lock()
					st.TotalICMPReq++
					// Increase meter value when a packet is sent
					st.MeterValue++
					if st.MeterValue > maxMeterValue {
						st.MeterValue = maxMeterValue
					}
					st.mu.Unlock()

					ok := pingOnce(st.Name, host)
					mu.Lock()
					res[host] = ok
					mu.Unlock()

					st.mu.Lock()
					if ok {
						st.TotalICMPRep++
						// Decrease meter value when a packet is successfully received
						st.MeterValue--
						if st.MeterValue < 0 {
							st.MeterValue = 0
						}
						// Only update spinner when packets are successfully received
						st.spin()
					} else {
						st.DroppedCount++
						st.LastDrop = time.Now()
						// For failed pings, we don't decrease the meter value
						// Trigger UI update on ping failure
						select {
						case uiUpdateChan <- struct{}{}:
						default:
						}
					}
					st.mu.Unlock()
				}(h)
			}
			wg.Wait()

			// Check if reachability status changed
			statusChanged := false
			st.mu.Lock()
			for host, newStatus := range res {
				if oldStatus, ok := st.Reachable[host]; !ok || oldStatus != newStatus {
					statusChanged = true
					break
				}
			}

			st.Reachable = res
			st.LastPing = time.Now()

			// Reset DroppedCount if all hosts are reachable
			allReachable := true
			for _, ok := range res {
				if !ok {
					allReachable = false
					break
				}
			}
			if allReachable {
				st.DroppedCount = 0
				// If all hosts are reachable, gradually decay the meter value
				if st.MeterValue > 0 {
					st.MeterValue--
				}
			}
			st.mu.Unlock()

			// Always trigger UI update when reachability status changes
			if statusChanged {
				select {
				case uiUpdateChan <- struct{}{}:
				default:
				}
			}
		}
	}
}
func lossLoop(ctx context.Context, st *InterfaceStatus, hosts []string) {
	logf("Starting packet loss monitoring for %s with %d hosts", st.Name, len(hosts))
	tk := time.NewTicker(packetLossPeriod)
	defer tk.Stop()
	for {
		select {
		case <-ctx.Done():
			logf("Stopping packet loss monitoring for %s", st.Name)
			return
		case <-tk.C:
			// logf("Checking packet loss for %s", st.Name) // Too verbose
			var wg sync.WaitGroup
			res := make(map[string]float64, len(hosts))
			mu := sync.Mutex{}
			for _, h := range hosts {
				wg.Add(1)
				go func(host string) {
					defer wg.Done()
					lp := lossPercent(st.Name, host)
					mu.Lock()
					res[host] = lp
					mu.Unlock()

					st.mu.Lock()
					if lp == 0 {
						// Only update spinner when there's 0% packet loss (successful receipt)
						st.spin()
					} else {
						// Trigger UI update on packet loss
						select {
						case uiUpdateChan <- struct{}{}:
						default:
						}
					}
					st.mu.Unlock()
				}(h)
			}
			wg.Wait()

			// Check if loss status changed
			statusChanged := false
			st.mu.Lock()
			for host, newLoss := range res {
				if oldLoss, ok := st.Loss[host]; !ok || math.Abs(oldLoss-newLoss) > 0.01 {
					statusChanged = true
					break
				}
			}

			for k, v := range res {
				st.Loss[k] = v
			}
			st.mu.Unlock()

			// Always trigger UI update when loss status changes
			if statusChanged {
				select {
				case uiUpdateChan <- struct{}{}:
				default:
				}
			}
		}
	}
}
func tcpLoop(ctx context.Context, st *InterfaceStatus, hosts []string) {
	logf("Starting TCP monitoring for %s with %d hosts", st.Name, len(hosts))
	tk := time.NewTicker(time.Second)
	defer tk.Stop()
	for {
		select {
		case <-ctx.Done():
			logf("Stopping TCP monitoring for %s", st.Name)
			return
		case <-tk.C:
			// logf("Checking TCP for %s", st.Name) // Too verbose
			statusChanged := false

			for _, hp := range hosts {
				ms := float64(tcpDuration(st.Name, hp).Milliseconds())
				st.mu.Lock()

				// Check if TCP latency significantly changed
				hist := st.TCP[hp]
				if len(hist) > 0 {
					lastMs := hist[len(hist)-1]
					if math.Abs(lastMs-ms) > 20 { // 20ms threshold for significant change
						statusChanged = true
					}
				} else {
					// First measurement
					statusChanged = true
				}

				if ms < float64(tcpTimeout.Milliseconds()) {
					// Only update spinner on successful TCP connections
					st.spin()
				} else {
					// Trigger UI update on TCP timeout
					select {
					case uiUpdateChan <- struct{}{}:
					default:
					}
				}

				if len(hist) >= statsHistory {
					hist = hist[1:]
				}
				st.TCP[hp] = append(hist, ms)
				st.mu.Unlock()
			}

			// Always trigger UI update when TCP status changes significantly
			if statusChanged {
				select {
				case uiUpdateChan <- struct{}{}:
				default:
				}
			}
		}
	}
}

/*────────────────── UI loop ─────────────────────*/

func uiLoop(ctx context.Context, scr tcell.Screen, a, b *InterfaceStatus, start time.Time) {
	logf("UI loop started")
	defer func() {
		logf("UI loop cleanup")
		scr.Clear()
		scr.ShowCursor(0, 0)
		scr.Fini()
		logf("Screen finalized")
	}()

	// Remove the ticker as we'll update on spinner ticks
	// tk := time.NewTicker(screenRefresh)
	// defer tk.Stop()
	spin := 0

	// Function to draw the screen
	drawScreen := func() {
		w, _ := scr.Size()
		scr.Clear()
		put(scr, 0, 0, hline(w), cDefault)
		put(scr, 0, 1, fmt.Sprintf("== %c %s", spins[spin%len(spins)], time.Now().Format(time.RFC1123Z)), cDefault)
		put(scr, 0, 2, hline(w), cDefault)
		spin++
		put(scr, 0, 3, "Runtime: "+time.Since(start).Round(time.Second).String(), cDefault)
		y := 5
		y = drawIface(scr, y, w, a)
		_ = drawIface(scr, y, w, b)
		scr.Show()
	}

	// Initial draw
	drawScreen()

	// Even without a ticker, ensure we update at least every second
	// This is a backup in case there are no spinner updates
	backupTicker := time.NewTicker(time.Second)
	defer backupTicker.Stop()

	for {
		select {
		case <-ctx.Done():
			logf("Context cancelled, exiting UI loop")
			return
		case <-uiUpdateChan:
			// Update on spinner ticks (no rate limiting)
			drawScreen()
		case <-backupTicker.C:
			// Fallback to ensure we update at least once per second
			drawScreen()
		}
	}
}

/*────────────────── main ─────────────────────────*/

var (
	reachHosts      []string
	packetLossHosts = []string{"github.com", "google.com", "1.1.1.1", "8.8.8.8"}
	tcpTestHosts    = []string{
		"datavi.be:443", "fast.com:443", "cloudflare.com:443",
		"console.aws.amazon.com:443", "console.cloud.google.com:443",
	}
)

// Simple logging function
func logf(format string, v ...interface{}) {
	if logFile == nil || *logFile == "" {
		return
	}
	f, err := os.OpenFile(*logFile, os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0644)
	if err != nil {
		return // silently fail if we can't log
	}
	defer f.Close()
	fmt.Fprintf(f, time.Now().Format("2006-01-02 15:04:05.000 ")+format+"\n", v...)
}

func main() {
	flag.Parse()
	logf("Starting rtnetmon")
	reachHosts = strings.Split(*hostCSV, ",")
	logf("Monitoring interfaces %s and %s", *ifaceA, *ifaceB)

	// Initialize UI update channel
	uiUpdateChan = make(chan struct{}, 100)
	logf("UI update channel initialized with buffer size 100")

	newStatus := func(name, label string) *InterfaceStatus {
		logf("Initializing status for interface %s", name)
		return &InterfaceStatus{
			Name:       name,
			Label:      label,
			IPInfo:     fetchIPInfo(name),
			Reachable:  map[string]bool{},
			Loss:       map[string]float64{},
			TCP:        map[string][]float64{},
			MeterValue: 0, // Initialize meter value to 0
		}
	}
	a, b := newStatus(*ifaceA, *labelA), newStatus(*ifaceB, *labelB)

	logf("Initializing screen")
	scr, err := tcell.NewScreen()
	if err != nil {
		logf("Error creating screen: %v", err)
		panic(err)
	}
	if err = scr.Init(); err != nil {
		logf("Error initializing screen: %v", err)
		panic(err)
	}
	logf("Screen initialized")

	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()

	sig := make(chan os.Signal, 1)
	signal.Notify(sig, os.Interrupt, syscall.SIGTERM)
	go func() {
		s := <-sig
		logf("Signal received: %v", s)
		cancel()
	}()

	// Event polling loop with logging
	go func() {
		logf("Starting keyboard event loop")
		for {
			if ev := scr.PollEvent(); ev != nil {
				logf("Event received: %T", ev)
				if ke, ok := ev.(*tcell.EventKey); ok {
					logf("Key event: %v, rune: %c", ke.Key(), ke.Rune())
					if ke.Key() == tcell.KeyCtrlC || ke.Rune() == 'q' {
						logf("Quit key detected")
						cancel()
						return
					}
				}

				// Trigger UI update on any event
				select {
				case uiUpdateChan <- struct{}{}:
				default:
				}
			}
		}
	}()

	// Start monitoring goroutines
	logf("Starting monitoring goroutines")
	go reachLoop(ctx, a, reachHosts)
	go reachLoop(ctx, b, reachHosts)
	go lossLoop(ctx, a, packetLossHosts)
	go lossLoop(ctx, b, packetLossHosts)
	go tcpLoop(ctx, a, tcpTestHosts)
	go tcpLoop(ctx, b, tcpTestHosts)

	logf("Starting UI loop")
	uiLoop(ctx, scr, a, b, time.Now())
	logf("UI loop exited, program ending")
}

/*────────────────── Extra packet-loss hosts ─────

To widen geographic and CDN coverage you could add:

  facebook.com
  microsoft.com
  apple.com
  twitter.com
  akamai.com

These are large anycast/CDN endpoints that tend to reveal regional
network quirks.  Add them to `packetLossHosts` (and `reachHosts`
if you also want individual pings every second). */