rtnetmon/internal/monitor/monitor.go

// Package monitor implements the real-time network monitoring dashboard:
// ICMP reachability, packet loss, TCP latency, and the terminal UI that
// renders them. It monitors one or two interfaces.
package monitor

import (
	"context"
	"encoding/json"
	"errors"
	"fmt"
	"math"
	"net"
	"os"
	"os/exec"
	"runtime"
	"strconv"
	"strings"
	"sync"
	"time"

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

// Meter glyphs used to render the ASCII loss meter.
const (
	MeterStart = '['
	MeterEnd   = ']'
	MeterFill  = '='
	MeterEmpty = ' '
)

// Default monitor timing configuration.
const (
	defaultICMPTimeout      = 500 * time.Millisecond
	defaultTCPTimeout       = 500 * time.Millisecond
	defaultPacketLossPings  = 20
	defaultPacketLossPeriod = 5 * time.Second
	defaultStatsHistory     = 300
	defaultScreenRefresh    = 500 * time.Millisecond
)

// Default display geometry (column widths and meter sizing).
const (
	defaultMeterWidth     = 7
	defaultMeterFillWidth = 5
	defaultMaxMeterValue  = 10
	defaultHostWidth      = 30
	defaultNumWidth       = 7
	defaultStdWidth       = 8
	defaultNWidth         = 6
	defaultLostWidth      = 7
)

// Miscellaneous timing constants.
const (
	ipInfoTimeout    = 2 * time.Second
	lossQueryTimeout = 3 * time.Second
	uiUpdateBuffer   = 100
	logFileMode      = 0o644
)

// errNoIPv4 is returned when an interface has no usable IPv4 address.
var errNoIPv4 = errors.New("no IPv4 address on interface")

// Monitor represents the network monitoring system.
type Monitor struct {
	// Configuration
	ICMPTimeout      time.Duration
	TCPTimeout       time.Duration
	PacketLossPings  int
	PacketLossPeriod time.Duration
	StatsHistory     int
	ScreenRefresh    time.Duration
	MeterWidth       int
	MeterFillWidth   int
	MaxMeterValue    int

	// Column widths
	HostWidth int
	NumWidth  int
	StdWidth  int
	NWidth    int
	LostWidth int

	// Host lists
	reachabilityHosts []string
	packetLossHosts   []string
	tcpHosts          []string

	// Interfaces to monitor (one or two)
	interfaces []*InterfaceStatus

	// Logging
	logFile string

	// Runtime state
	screen    tcell.Screen
	startTime time.Time
	uiUpdate  chan struct{}

	// Synchronization
	mu sync.RWMutex
}

// IfaceSpec names one interface to monitor and its display label.
type IfaceSpec struct {
	Name  string
	Label string
}

// NewMonitor creates a new Monitor instance with default settings,
// monitoring the given interfaces (one or two).
func NewMonitor(ifaces []IfaceSpec, logFile string) *Monitor {
	m := &Monitor{
		ICMPTimeout:      defaultICMPTimeout,
		TCPTimeout:       defaultTCPTimeout,
		PacketLossPings:  defaultPacketLossPings,
		PacketLossPeriod: defaultPacketLossPeriod,
		StatsHistory:     defaultStatsHistory,
		ScreenRefresh:    defaultScreenRefresh,

		MeterWidth:     defaultMeterWidth,
		MeterFillWidth: defaultMeterFillWidth,
		MaxMeterValue:  defaultMaxMeterValue,
		HostWidth:      defaultHostWidth,
		NumWidth:       defaultNumWidth,
		StdWidth:       defaultStdWidth,
		NWidth:         defaultNWidth,
		LostWidth:      defaultLostWidth,

		reachabilityHosts: []string{},
		packetLossHosts:   []string{},
		tcpHosts:          []string{},

		logFile:   logFile,
		startTime: time.Now(),
		uiUpdate:  make(chan struct{}, uiUpdateBuffer),
	}

	for _, spec := range ifaces {
		m.interfaces = append(m.interfaces, NewInterfaceStatus(spec.Name, spec.Label))
	}

	return m
}

// AddReachabilityHost adds a host for reachability monitoring.
func (m *Monitor) AddReachabilityHost(host string) {
	m.mu.Lock()
	defer m.mu.Unlock()

	m.reachabilityHosts = append(m.reachabilityHosts, host)
}

// AddPacketLossHost adds a host for packet loss monitoring.
func (m *Monitor) AddPacketLossHost(host string) {
	m.mu.Lock()
	defer m.mu.Unlock()

	m.packetLossHosts = append(m.packetLossHosts, host)
}

// AddTCPHost adds a host:port for TCP connectivity monitoring.
func (m *Monitor) AddTCPHost(hostPort string) {
	m.mu.Lock()
	defer m.mu.Unlock()

	m.tcpHosts = append(m.tcpHosts, hostPort)
}

// Run starts the monitoring system.
func (m *Monitor) Run(ctx context.Context) error {
	m.logf("Starting monitor run")
	m.logf("Initializing screen")

	scr, err := tcell.NewScreen()
	if err != nil {
		m.logf("Error creating screen: %v", err)

		return fmt.Errorf("error creating screen: %w", err)
	}

	err = scr.Init()
	if err != nil {
		m.logf("Error initializing screen: %v", err)

		return fmt.Errorf("error initializing screen: %w", err)
	}

	m.screen = scr
	m.logf("Screen initialized")

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

	go m.keyboardEventLoop(ctx, cancel)

	m.logf("Starting monitoring goroutines")
	m.mu.RLock()
	reachHosts := append([]string{}, m.reachabilityHosts...)
	lossHosts := append([]string{}, m.packetLossHosts...)
	tcpHosts := append([]string{}, m.tcpHosts...)
	m.mu.RUnlock()

	for _, st := range m.interfaces {
		go m.reachLoop(ctx, st, reachHosts)
		go m.lossLoop(ctx, st, lossHosts)
		go m.tcpLoop(ctx, st, tcpHosts)
	}

	m.logf("Starting UI loop")
	m.uiLoop(ctx)
	m.logf("UI loop exited, monitor ending")

	return nil
}

// notifyUI requests a screen redraw without blocking if one is pending.
func (m *Monitor) notifyUI() {
	select {
	case m.uiUpdate <- struct{}{}:
	default:
	}
}

// keyboardEventLoop handles keyboard input. The context is accepted for a
// uniform loop signature; shutdown is driven by cancel on the quit key and
// by the screen being finalized (PollEvent then returns nil).
func (m *Monitor) keyboardEventLoop(_ context.Context, cancel context.CancelFunc) {
	m.logf("Starting keyboard event loop")

	for {
		ev := m.screen.PollEvent()
		if ev == nil {
			continue
		}

		m.logf("Event received: %T", ev)

		if ke, ok := ev.(*tcell.EventKey); ok {
			m.logf("Key event: %v, rune: %c", ke.Key(), ke.Rune())

			if ke.Key() == tcell.KeyCtrlC || ke.Rune() == 'q' {
				m.logf("Quit key detected")
				cancel()

				return
			}
		}

		m.notifyUI()
	}
}

// logf logs a formatted message.
func (m *Monitor) logf(format string, v ...any) {
	Logf(m.logFile, format, v...)
}

// InterfaceStatus holds the runtime status for a network interface.
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
	LostPackets                map[string]int
	mu                         sync.RWMutex
}

// NewInterfaceStatus creates a new interface status.
func NewInterfaceStatus(name, label string) *InterfaceStatus {
	return &InterfaceStatus{
		Name:        name,
		Label:       label,
		IPInfo:      fetchIPInfo(name),
		Reachable:   map[string]bool{},
		Loss:        map[string]float64{},
		TCP:         map[string][]float64{},
		MeterValue:  0,
		LostPackets: map[string]int{},
	}
}

// ipInfoResp holds the ipinfo.io response.
type ipInfoResp struct {
	IP       string `json:"ip"`
	Hostname string `json:"hostname"`
	Org      string `json:"org"`
}

// fetchIPInfo fetches public IP information as seen from an interface.
func fetchIPInfo(iface string) string {
	ctx, cancel := context.WithTimeout(context.Background(), ipInfoTimeout)
	defer cancel()

	cmd := exec.CommandContext( //nolint:gosec // G204: fixed argv, operator CLI input
		ctx, "curl", "-s", "--interface", iface, "--max-time", "2", "ipinfo.io")

	out, _ := cmd.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)
}

// pingArgs builds the arguments for a single reachability ping on the given
// OS. Linux binds the interface with -I and takes -W in seconds; macOS binds
// with -b and takes -W in milliseconds.
func pingArgs(goos, iface, host string) []string {
	if goos == "darwin" {
		return []string{"-b", iface, "-c1", "-W1000", host}
	}

	return []string{"-I", iface, "-c1", "-W1", host}
}

// lossArgs builds the arguments for a packet-loss ping burst of count pings.
func lossArgs(goos, iface, host string, count int) []string {
	c := strconv.Itoa(count)
	if goos == "darwin" {
		return []string{"-q", "-i", "0.05", "-c", c, "-W1000", "-b", iface, host}
	}

	return []string{"-q", "-i", "0.05", "-c", c, "-W1", "-I", iface, host}
}

// pingOnce performs a single ping over the named interface.
func (m *Monitor) pingOnce(ctx context.Context, iface, host string) bool {
	ctx, cancel := context.WithTimeout(ctx, m.ICMPTimeout)
	defer cancel()

	args := pingArgs(runtime.GOOS, iface, host)

	cmd := exec.CommandContext( //nolint:gosec // G204: fixed argv, operator CLI input
		ctx, "ping", args...)

	return cmd.Run() == nil
}

// lossPercent measures the packet loss fraction (0..1) for a host.
func (m *Monitor) lossPercent(ctx context.Context, iface, host string) float64 {
	ctx, cancel := context.WithTimeout(ctx, lossQueryTimeout)
	defer cancel()

	args := lossArgs(runtime.GOOS, iface, host, m.PacketLossPings)

	cmd := exec.CommandContext( //nolint:gosec // G204: fixed argv, operator CLI input
		ctx, "ping", args...)

	out, err := cmd.CombinedOutput()
	if err != nil {
		return 1.0
	}

	for _, ln := range strings.Split(string(out), "\n") {
		if !strings.Contains(ln, "packet loss") {
			continue
		}

		for _, f := range strings.Fields(ln) {
			before, ok := strings.CutSuffix(f, "%")
			if !ok {
				continue
			}

			p, _ := strconv.ParseFloat(before, 64)

			return p / percentFull
		}
	}

	return 1.0
}

// localAddr returns the first IPv4 address bound to an interface.
func localAddr(iface string) (net.Addr, error) {
	ifi, err := net.InterfaceByName(iface)
	if err != nil {
		return nil, fmt.Errorf("interface %s: %w", iface, 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("%w: %s", errNoIPv4, iface)
}

// tcpDuration measures how long a TCP connection to hp takes over iface.
func (m *Monitor) tcpDuration(iface, hp string) time.Duration {
	la, err := localAddr(iface)
	if err != nil {
		return m.TCPTimeout
	}

	d := net.Dialer{Timeout: m.TCPTimeout, LocalAddr: la, Control: bindControl(iface)}
	st := time.Now()

	c, err := d.Dial("tcp", hp)
	if err != nil {
		return m.TCPTimeout
	}

	_ = c.Close()

	return time.Since(st)
}

// MinMaxAvgStd returns the minimum, maximum, mean and standard deviation.
func MinMaxAvgStd(xs []float64) (float64, float64, float64, float64) {
	if len(xs) == 0 {
		return 0, 0, 0, 0
	}

	mn, mx := xs[0], xs[0]

	var sum float64

	for _, v := range xs {
		mn = min(mn, v)
		mx = max(mx, v)
		sum += v
	}

	avg := sum / float64(len(xs))

	var variance float64

	for _, v := range xs {
		variance += (v - avg) * (v - avg)
	}

	return mn, mx, avg, math.Sqrt(variance / float64(len(xs)))
}

// Spin advances the spinner frame.
func (st *InterfaceStatus) Spin() {
	st.SpinFrame = (st.SpinFrame + 1) % len(spins())
}

// IsHealthy reports whether the interface currently looks healthy.
func (st *InterfaceStatus) IsHealthy(tcpTimeout time.Duration) bool {
	st.mu.RLock()
	defer st.mu.RUnlock()

	for _, ok := range st.Reachable {
		if !ok {
			return false
		}
	}

	for _, lp := range st.Loss {
		if lp > 0 {
			return false
		}
	}

	for _, hist := range st.TCP {
		if len(hist) == 0 || hist[len(hist)-1] >= float64(tcpTimeout.Milliseconds()) {
			return false
		}
	}

	return true
}

// spins returns the ASCII spinner frames.
func spins() []rune {
	return []rune{'|', '/', '-', '\\'}
}

// brailleSpins returns the braille clock spinner frames.
func brailleSpins() []rune {
	return []rune{'⠉', '⠘', '⠰', '⠠', '⠄', '⠆', '⠇', '⠋'}
}

// Logf appends a timestamped formatted message to logFile, if set.
func Logf(logFile, format string, v ...any) {
	if logFile == "" {
		return
	}

	f, err := os.OpenFile( //nolint:gosec // G304: operator --logfile path
		logFile, os.O_APPEND|os.O_CREATE|os.O_WRONLY, logFileMode)
	if err != nil {
		return
	}
	defer func() { _ = f.Close() }()

	_, _ = fmt.Fprintf(f,
		time.Now().Format("2006-01-02 15:04:05.000 ")+format+"\n", v...)
}