rtnetmon/internal/monitor/monitor.go

package monitor

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

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

// Non-configurable constants (meter characters)
const (
	// ASCII characters for the meter
	MeterStart = '['
	MeterEnd   = ']'
	MeterFill  = '='
	MeterEmpty = ' '
)

// 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

	// 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{
		// Default timing configuration
		ICMPTimeout:      500 * time.Millisecond,
		TCPTimeout:       500 * time.Millisecond,
		PacketLossPings:  20,
		PacketLossPeriod: 5 * time.Second,
		StatsHistory:     300,
		ScreenRefresh:    500 * time.Millisecond,

		// Default display configuration
		MeterWidth:     7,
		MeterFillWidth: 5,
		MaxMeterValue:  10,
		HostWidth:      30,
		NumWidth:       7,
		StdWidth:       8,
		NWidth:         6,
		LostWidth:      7,

		// Initialize host lists
		reachabilityHosts: []string{},
		packetLossHosts:   []string{},
		tcpHosts:          []string{},

		// Logging
		logFile: logFile,

		startTime: time.Now(),
	}

	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")

	// Initialize screen
	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)
	}
	if err = scr.Init(); err != nil {
		m.logf("Error initializing screen: %v", err)
		return fmt.Errorf("error initializing screen: %w", err)
	}
	m.screen = scr
	m.logf("Screen initialized")

	// Create cancellable context
	ctx, cancel := context.WithCancel(ctx)
	defer cancel()

	// Handle keyboard events
	go m.keyboardEventLoop(ctx, cancel)

	// Start monitoring goroutines
	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)
	}

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

	return nil
}

// keyboardEventLoop handles keyboard input
func (m *Monitor) keyboardEventLoop(ctx context.Context, cancel context.CancelFunc) {
	m.logf("Starting keyboard event loop")
	for {
		if ev := m.screen.PollEvent(); ev != nil {
			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
				}
			}

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

// logf logs a formatted message
func (m *Monitor) logf(format string, v ...interface{}) {
	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 IP info response
type ipInfoResp struct {
	IP       string `json:"ip"`
	Hostname string `json:"hostname"`
	Org      string `json:"org"`
}

// fetchIPInfo fetches IP information for an interface
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)
}

// 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
func (m *Monitor) pingOnce(iface, host string) bool {
	ctx, cancel := context.WithTimeout(context.Background(), m.ICMPTimeout)
	defer cancel()

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

	return exec.CommandContext(ctx, "ping", args...).Run() == nil
}

// lossPercent calculates packet loss percentage
func (m *Monitor) lossPercent(iface, host string) float64 {
	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
	defer cancel()

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

	out, err := exec.CommandContext(ctx, "ping", args...).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
}

// localAddr gets the local address for an interface
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)
}

// tcpDuration measures TCP connection duration
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 calculates min, max, average and standard deviation
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
}

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

// IsHealthy checks if the interface is healthy
func (st *InterfaceStatus) IsHealthy(tcpTimeout time.Duration) bool {
	st.mu.RLock()
	defer st.mu.RUnlock()

	// Check if there are any currently unreachable hosts
	for _, ok := range st.Reachable {
		if !ok {
			return false
		}
	}

	// Check if there's any current packet loss
	for _, lp := range st.Loss {
		if lp > 0 {
			return false
		}
	}

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

	return true
}

// Spinners and braille characters
var (
	Spins        = []rune{'|', '/', '-', '\\'}
	BrailleSpins = []rune{
		'⠉', '⠘', '⠰', '⠠', '⠄', '⠆', '⠇', '⠋',
	}
)

// Logf is a simple logging function
func Logf(logFile string, format string, v ...interface{}) {
	if logFile == "" {
		return
	}
	f, err := os.OpenFile(logFile, os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0644)
	if err != nil {
		return
	}
	defer f.Close()
	fmt.Fprintf(f, time.Now().Format("2006-01-02 15:04:05.000 ")+format+"\n", v...)
}