inp2p/internal/client/client.go

// Package client implements the inp2pc P2P client.
package client

import (
	"crypto/tls"
	"fmt"
	"log"
	"net"
	"net/netip"
	"net/url"
	"os"
	"os/exec"
	"runtime"
	"strings"
	"sync"
	"time"

	"golang.org/x/sys/unix"

	"github.com/gorilla/websocket"
	"github.com/openp2p-cn/inp2p/pkg/auth"
	"github.com/openp2p-cn/inp2p/pkg/config"
	"github.com/openp2p-cn/inp2p/pkg/nat"
	"github.com/openp2p-cn/inp2p/pkg/protocol"
	"github.com/openp2p-cn/inp2p/pkg/punch"
	"github.com/openp2p-cn/inp2p/pkg/relay"
	"github.com/openp2p-cn/inp2p/pkg/signal"
	"github.com/openp2p-cn/inp2p/pkg/tunnel"
)

// Client is the INP2P client node.
type Client struct {
	cfg        config.ClientConfig
	conn       *signal.Conn
	natType    protocol.NATType
	publicIP   string
	publicPort int
	localPort  int
	tunnels    map[string]*tunnel.Tunnel // peerNode → tunnel
	tMu        sync.RWMutex
	relayMgr   *relay.Manager
	sdwanMu    sync.RWMutex
	sdwan      protocol.SDWANConfig
	sdwanIP    string
	sdwanStop  chan struct{}
	tunMu      sync.Mutex
	tunFile    *os.File
	quit       chan struct{}
	wg         sync.WaitGroup
}

// New creates a new client.
func New(cfg config.ClientConfig) *Client {
	c := &Client{
		cfg:        cfg,
		natType:    protocol.NATUnknown,
		tunnels:    make(map[string]*tunnel.Tunnel),
		sdwanStop:  make(chan struct{}),
		quit:       make(chan struct{}),
		publicPort: 0,
		localPort:  0,
	}

	if cfg.RelayEnabled {
		c.relayMgr = relay.NewManager(cfg.RelayPort, true, cfg.SuperRelay, cfg.MaxRelayLoad, cfg.Token)
	}

	return c
}

// Run is the main client loop. Connects, authenticates, and maintains the connection.
func (c *Client) Run() error {
	for {
		if err := c.connectAndRun(); err != nil {
			log.Printf("[client] disconnected: %v, reconnecting in 5s...", err)
		}

		select {
		case <-c.quit:
			return nil
		case <-time.After(5 * time.Second):
		}
	}
}

func (c *Client) connectAndRun() error {
	// 1. NAT Detection
	log.Printf("[client] detecting NAT type via %s...", c.cfg.ServerHost)
	natResult := nat.Detect(
		c.cfg.ServerHost,
		c.cfg.STUNUDP1, c.cfg.STUNUDP2,
		c.cfg.STUNTCP1, c.cfg.STUNTCP2,
	)
	c.natType = natResult.Type
	c.publicIP = natResult.PublicIP
	c.publicPort = natResult.Port1
	c.localPort = natResult.LocalPort
	log.Printf("[client] NAT type=%s, publicIP=%s, publicPort=%d, localPort=%d", c.natType, c.publicIP, c.publicPort, c.localPort)

	// 2. WSS Connect
	scheme := "ws"
	if !c.cfg.Insecure {
		scheme = "wss"
	}
	u := url.URL{Scheme: scheme, Host: fmt.Sprintf("%s:%d", c.cfg.ServerHost, c.cfg.ServerPort), Path: "/ws"}

	dialer := websocket.Dialer{
		TLSClientConfig: &tls.Config{InsecureSkipVerify: c.cfg.Insecure},
	}
	ws, _, err := dialer.Dial(u.String(), nil)
	if err != nil {
		return fmt.Errorf("ws connect: %w", err)
	}

	c.conn = signal.NewConn(ws)
	defer c.conn.Close()

	// Register handlers BEFORE ReadLoop so server-pushed messages
	// (e.g. SDWANConfig sent right after LoginRsp) are not dropped.
	c.registerHandlers()

	// Start ReadLoop in background BEFORE sending login
	// (so waiter can receive the LoginRsp)
	readErr := make(chan error, 1)
	go func() {
		readErr <- c.conn.ReadLoop()
	}()

	// 3. Login
	loginReq := protocol.LoginReq{
		Node:           c.cfg.Node,
		Token:          c.cfg.Token,
		User:           c.cfg.User,
		Version:        config.Version,
		NATType:        c.natType,
		ShareBandwidth: c.cfg.ShareBandwidth,
		RelayEnabled:   c.cfg.RelayEnabled,
		SuperRelay:     c.cfg.SuperRelay,
		PublicIP:       c.publicIP,
		PublicPort:     c.publicPort,
	}

	rspData, err := c.conn.Request(
		protocol.MsgLogin, protocol.SubLoginReq, loginReq,
		protocol.MsgLogin, protocol.SubLoginRsp,
		10*time.Second,
	)
	if err != nil {
		return fmt.Errorf("login: %w", err)
	}

	var loginRsp protocol.LoginRsp
	if err := protocol.DecodePayload(rspData, &loginRsp); err != nil {
		return fmt.Errorf("decode login rsp: %w", err)
	}
	if loginRsp.Error != 0 {
		return fmt.Errorf("login rejected: %s", loginRsp.Detail)
	}

	log.Printf("[client] login ok: node=%s, user=%s", loginRsp.Node, loginRsp.User)

	// 4. Send ReportBasic
	c.sendReportBasic()

	// 5. Start heartbeat
	c.wg.Add(1)
	go c.heartbeatLoop()

	// 7. Start relay if enabled
	if c.relayMgr != nil {
		if err := c.relayMgr.Start(); err != nil {
			log.Printf("[client] relay start failed: %v", err)
		}
	}

	// 8. Auto-run configured apps
	for _, app := range c.cfg.Apps {
		if app.Enabled {
			go c.connectApp(app)
		}
	}

	// 9. Wait for disconnect
	return <-readErr
}

func (c *Client) sendReportBasic() {
	hostname, _ := os.Hostname()
	report := protocol.ReportBasic{
		OS:         runtime.GOOS,
		LanIP:      getLocalIP(),
		Version:    config.Version,
		HasIPv4:    1,
		PublicIP:   c.publicIP,
		PublicPort: c.publicPort,
	}
	_ = hostname // for future use
	c.conn.Write(protocol.MsgReport, protocol.SubReportBasic, report)
}

func (c *Client) registerHandlers() {
	// Handle connection coordination from server
	c.conn.OnMessage(protocol.MsgPush, protocol.SubPushConnectReq, func(data []byte) error {
		var req protocol.ConnectReq
		if err := protocol.DecodePayload(data, &req); err != nil {
			return err
		}
		log.Printf("[client] connect request: %s → %s (punch)", req.From, req.To)
		go c.handlePunchRequest(req)
		return nil
	})

	// Handle peer online notification
	c.conn.OnMessage(protocol.MsgPush, protocol.SubPushNodeOnline, func(data []byte) error {
		var msg struct {
			Node string `json:"node"`
		}
		protocol.DecodePayload(data, &msg)
		log.Printf("[client] peer online: %s, retrying apps", msg.Node)
		// Retry apps targeting this node
		for _, app := range c.cfg.Apps {
			if app.Enabled && app.PeerNode == msg.Node {
				go c.connectApp(app)
			}
		}
		return nil
	})

	// Handle SDWAN config push
	c.conn.OnMessage(protocol.MsgPush, protocol.SubPushSDWANConfig, func(data []byte) error {
		var cfg protocol.SDWANConfig
		if err := protocol.DecodePayload(data, &cfg); err != nil {
			return err
		}
		if cfg.GatewayCIDR == "" {
			return nil
		}
		log.Printf("[client] sdwan config received: gateway=%s nodes=%d mode=%s", cfg.GatewayCIDR, len(cfg.Nodes), cfg.Mode)
		_ = os.WriteFile("sdwan.json", data[protocol.HeaderSize:], 0644)

		// apply control+data plane
		if err := c.applySDWAN(cfg); err != nil {
			log.Printf("[client] sdwan apply failed: %v", err)
		}
		return nil
	})

	// SDWAN peer online/update event
	c.conn.OnMessage(protocol.MsgPush, protocol.SubPushSDWANPeer, func(data []byte) error {
		var p protocol.SDWANPeer
		if err := protocol.DecodePayload(data, &p); err != nil {
			return err
		}
		if p.Node == "" || p.Node == c.cfg.Node || p.IP == "" {
			return nil
		}
		_ = runCmd("ip", "route", "replace", p.IP+"/32", "dev", "optun")
		return nil
	})

	// SDWAN peer offline/delete event
	c.conn.OnMessage(protocol.MsgPush, protocol.SubPushSDWANDel, func(data []byte) error {
		var p protocol.SDWANPeer
		if err := protocol.DecodePayload(data, &p); err != nil {
			return err
		}
		if p.IP != "" {
			_ = runCmd("ip", "route", "del", p.IP+"/32", "dev", "optun")
		}
		if p.Node != "" {
			c.tMu.Lock()
			if t, ok := c.tunnels[p.Node]; ok {
				t.Close()
				delete(c.tunnels, p.Node)
			}
			c.tMu.Unlock()
		}
		return nil
	})

	// SDWAN packet from server, inject to local TUN
	c.conn.OnMessage(protocol.MsgTunnel, protocol.SubTunnelSDWANData, func(data []byte) error {
		var pkt protocol.SDWANPacket
		if err := protocol.DecodePayload(data, &pkt); err != nil {
			return err
		}
		if len(pkt.Payload) == 0 {
			return nil
		}
		return c.writeTUN(pkt.Payload)
	})

	// SDWAN raw packet (binary payload) from server
	c.conn.OnMessage(protocol.MsgTunnel, protocol.SubTunnelSDWANRaw, func(data []byte) error {
		if len(data) <= protocol.HeaderSize {
			return nil
		}
		payload := data[protocol.HeaderSize:]
		if len(payload) == 0 {
			return nil
		}
		return c.writeTUN(payload)
	})

	// Handle edit app push
	c.conn.OnMessage(protocol.MsgPush, protocol.SubPushEditApp, func(data []byte) error {
		var app protocol.AppConfig
		if err := protocol.DecodePayload(data, &app); err != nil {
			return err
		}
		log.Printf("[client] edit app push: %s → %s:%d", app.AppName, app.PeerNode, app.DstPort)
		go c.connectApp(config.AppConfig{
			AppName:  app.AppName,
			Protocol: app.Protocol,
			SrcPort:  app.SrcPort,
			PeerNode: app.PeerNode,
			DstHost:  app.DstHost,
			DstPort:  app.DstPort,
			Enabled:  true,
		})
		return nil
	})

	// Handle relay connect request (when this node acts as relay)
	if c.relayMgr != nil {
		c.conn.OnMessage(protocol.MsgPush, protocol.SubPushRelayOffer, func(data []byte) error {
			var req struct {
				From  string `json:"from"`
				To    string `json:"to"`
				Token uint64 `json:"token"`
			}
			if err := protocol.DecodePayload(data, &req); err != nil {
				return err
			}

			// Verify TOTP
			if !auth.VerifyTOTP(req.Token, c.cfg.Token, time.Now().Unix()) {
				log.Printf("[client] relay request from %s denied: TOTP mismatch", req.From)
				return nil
			}

			log.Printf("[client] accepting relay: %s → %s", req.From, req.To)
			return nil
		})
	}
}

func (c *Client) heartbeatLoop() {
	defer c.wg.Done()
	ticker := time.NewTicker(time.Duration(config.HeartbeatInterval) * time.Second)
	defer ticker.Stop()

	for {
		select {
		case <-ticker.C:
			if err := c.conn.Write(protocol.MsgHeartbeat, protocol.SubHeartbeatPing, nil); err != nil {
				log.Printf("[client] heartbeat send failed: %v", err)
				return
			}
		case <-c.quit:
			return
		}
	}
}

// connectApp establishes a tunnel for an app config.
func (c *Client) connectApp(app config.AppConfig) {
	log.Printf("[client] connecting app %s: :%d → %s:%d", app.AppName, app.SrcPort, app.PeerNode, app.DstPort)

	// Check if we already have a tunnel
	c.tMu.RLock()
	if t, ok := c.tunnels[app.PeerNode]; ok && t.IsAlive() {
		c.tMu.RUnlock()
		// Tunnel exists, just add the port forward
		if err := t.ListenAndForward(app.Protocol, app.SrcPort, app.DstHost, app.DstPort); err != nil {
			log.Printf("[client] listen error for %s: %v", app.AppName, err)
		}
		return
	}
	c.tMu.RUnlock()

	// Request connection coordination from server
	req := protocol.ConnectReq{
		From:     c.cfg.Node,
		To:       app.PeerNode,
		Protocol: app.Protocol,
		SrcPort:  app.SrcPort,
		DstHost:  app.DstHost,
		DstPort:  app.DstPort,
	}

	rspData, err := c.conn.Request(
		protocol.MsgPush, protocol.SubPushConnectReq, req,
		protocol.MsgPush, protocol.SubPushConnectRsp,
		15*time.Second,
	)
	if err != nil {
		log.Printf("[client] connect coordination failed for %s: %v", app.PeerNode, err)
		c.tryRelay(app)
		return
	}

	var rsp protocol.ConnectRsp
	protocol.DecodePayload(rspData, &rsp)
	if rsp.Error != 0 {
		log.Printf("[client] connect denied: %s", rsp.Detail)
		c.tryRelay(app)
		return
	}

	// Attempt punch
	result := punch.Connect(punch.Config{
		PeerIP:      rsp.Peer.IP,
		PeerPort:    rsp.Peer.Port,
		PeerNAT:     rsp.Peer.NATType,
		SelfNAT:     c.natType,
		SelfPort:    c.localPort,
		IsInitiator: true,
	})

	if result.Error != nil {
		log.Printf("[client] punch failed for %s: %v", app.PeerNode, result.Error)
		c.tryRelay(app)
		c.reportConnect(app, protocol.ReportConnect{
			PeerNode: app.PeerNode, Error: result.Error.Error(),
			NATType: c.natType, PeerNATType: rsp.Peer.NATType,
		})
		return
	}

	// Punch success — create tunnel
	t := tunnel.New(app.PeerNode, result.Conn, result.Mode, result.RTT, true)
	c.tMu.Lock()
	c.tunnels[app.PeerNode] = t
	c.tMu.Unlock()

	if err := t.ListenAndForward(app.Protocol, app.SrcPort, app.DstHost, app.DstPort); err != nil {
		log.Printf("[client] listen error: %v", err)
	}

	c.reportConnect(app, protocol.ReportConnect{
		PeerNode: app.PeerNode, LinkMode: result.Mode,
		RTT:     int(result.RTT.Milliseconds()),
		NATType: c.natType, PeerNATType: rsp.Peer.NATType,
	})

	log.Printf("[client] tunnel established: %s via %s (rtt=%s)", app.PeerNode, result.Mode, result.RTT)
}

// tryRelay attempts to use a relay node.
func (c *Client) tryRelay(app config.AppConfig) {
	log.Printf("[client] trying relay for %s", app.PeerNode)

	rspData, err := c.conn.Request(
		protocol.MsgRelay, protocol.SubRelayNodeReq,
		protocol.RelayNodeReq{PeerNode: app.PeerNode},
		protocol.MsgRelay, protocol.SubRelayNodeRsp,
		10*time.Second,
	)
	if err != nil {
		log.Printf("[client] relay request failed: %v", err)
		return
	}

	var rsp protocol.RelayNodeRsp
	protocol.DecodePayload(rspData, &rsp)
	if rsp.Error != 0 {
		log.Printf("[client] no relay available for %s", app.PeerNode)
		return
	}

	log.Printf("[client] relay via %s (%s mode), connecting...", rsp.RelayName, rsp.Mode)

	// Connect to relay node
	result := punch.AttemptDirect(punch.Config{
		PeerIP:   rsp.RelayIP,
		PeerPort: rsp.RelayPort,
	})
	if result.Error != nil {
		log.Printf("[client] relay connect failed: %v", result.Error)
		return
	}

	t := tunnel.New(app.PeerNode, result.Conn, "relay-"+rsp.Mode, result.RTT, true)
	c.tMu.Lock()
	c.tunnels[app.PeerNode] = t
	c.tMu.Unlock()

	if err := t.ListenAndForward(app.Protocol, app.SrcPort, app.DstHost, app.DstPort); err != nil {
		log.Printf("[client] relay listen error: %v", err)
	}

	c.reportConnect(app, protocol.ReportConnect{
		PeerNode: app.PeerNode, LinkMode: "relay", RelayNode: rsp.RelayName,
	})

	log.Printf("[client] relay tunnel established: %s via %s", app.PeerNode, rsp.RelayName)
}

func (c *Client) handlePunchRequest(req protocol.ConnectReq) {
	log.Printf("[client] handling punch from %s, NAT=%s", req.From, req.Peer.NATType)

	result := punch.Connect(punch.Config{
		PeerIP:      req.Peer.IP,
		PeerPort:    req.Peer.Port,
		PeerNAT:     req.Peer.NATType,
		SelfNAT:     c.natType,
		SelfPort:    c.localPort,
		IsInitiator: false,
	})

	rsp := protocol.ConnectRsp{
		From: c.cfg.Node,
		To:   req.From,
	}

	if result.Error != nil {
		rsp.Error = 1
		rsp.Detail = result.Error.Error()
		log.Printf("[client] punch from %s failed: %v", req.From, result.Error)
	} else {
		rsp.Peer = protocol.PunchParams{
			IP:      c.publicIP,
			NATType: c.natType,
		}
		log.Printf("[client] punch from %s OK via %s", req.From, result.Mode)

		// Create tunnel for the incoming connection
		t := tunnel.New(req.From, result.Conn, result.Mode, result.RTT, false)
		c.tMu.Lock()
		c.tunnels[req.From] = t
		c.tMu.Unlock()
	}

	c.conn.Write(protocol.MsgPush, protocol.SubPushConnectRsp, rsp)
}

func (c *Client) reportConnect(app config.AppConfig, rc protocol.ReportConnect) {
	rc.Protocol = app.Protocol
	rc.SrcPort = app.SrcPort
	rc.DstPort = app.DstPort
	rc.DstHost = app.DstHost
	rc.Version = config.Version
	rc.ShareBandwidth = c.cfg.ShareBandwidth
	c.conn.Write(protocol.MsgReport, protocol.SubReportConnect, rc)
}

func (c *Client) applySDWAN(cfg protocol.SDWANConfig) error {
	selfIP := ""
	for _, n := range cfg.Nodes {
		if n.Node == c.cfg.Node {
			selfIP = strings.TrimSpace(n.IP)
			break
		}
	}
	if selfIP == "" {
		return fmt.Errorf("node %s not found in sdwan nodes", c.cfg.Node)
	}
	// Use ioctl method only - it creates the device if not exists
	// Skip ip tuntap add to avoid conflicts
	_ = runCmd("ip", "tuntap", "add", "dev", "optun", "mode", "tun")
	_ = runCmd("ip", "link", "set", "dev", "optun", "up")
	_ = runCmd("ip", "link", "set", "dev", "optun", "mtu", "1420")
	_ = runCmd("ip", "addr", "add", selfIP+"/32", "dev", "optun")

	pfx, err := netip.ParsePrefix(cfg.GatewayCIDR)
	if err != nil {
		return fmt.Errorf("invalid gateway cidr: %s", cfg.GatewayCIDR)
	}
	// prefer /32 host routes for full-mesh precision
	for _, n := range cfg.Nodes {
		ip := strings.TrimSpace(n.IP)
		if ip == "" || ip == selfIP {
			log.Printf("[client] tun read error: %v", err)
		}
		_ = runCmd("ip", "route", "replace", ip+"/32", "dev", "optun")
	}
	// fallback broad route for hub mode / compatibility
	_ = runCmd("ip", "route", "replace", pfx.String(), "dev", "optun")

	c.sdwanMu.Lock()
	c.sdwan = cfg
	c.sdwanIP = selfIP
	c.sdwanMu.Unlock()

	// Try to start TUN reader, but don't fail SDWAN apply if it errors
	if err := c.ensureTUNReader(); err != nil {
		log.Printf("[client] ensureTUNReader failed (non-fatal): %v", err)
	}
	log.Printf("[client] sdwan applied: optun=%s route=%s dev optun", selfIP, pfx.String())
	return nil
}

func (c *Client) ensureTUNReader() error {
	c.tunMu.Lock()
	defer c.tunMu.Unlock()
	if c.tunFile != nil {
		return nil
	}
	// Try to open existing TUN device without deleting it
	f, err := os.OpenFile("/dev/net/tun", os.O_RDWR, 0)
	if err != nil {
		log.Printf("[client] open /dev/net/tun: %v", err)
		return err
	}
	ifr, err := unix.NewIfreq("optun")
	if err != nil {
		f.Close()
		log.Printf("[client] new ifreq: %v", err)
		return err
	}
	ifr.SetUint16(unix.IFF_TUN | unix.IFF_NO_PI)
	if err := unix.IoctlIfreq(int(f.Fd()), unix.TUNSETIFF, ifr); err != nil {
		// Device might already exist and be bound to another process
		// Try to use it anyway - maybe we can read from it
		log.Printf("[client] TUNSETIFF: %v (continuing anyway)", err)
	}
	c.tunFile = f
	c.wg.Add(1)
	go c.tunReadLoop()
	log.Printf("[client] tun reader started")
	return nil
}

func (c *Client) tunReadLoop() {
	defer c.wg.Done()
	buf := make([]byte, 65535)
	for {
		select {
		case <-c.quit:
			return
		default:
		}
		c.tunMu.Lock()
		f := c.tunFile
		c.tunMu.Unlock()
		if f == nil {
			return
		}
		n, err := f.Read(buf)
		if err != nil {
			if c.IsStopping() {
				return
			}
			time.Sleep(100 * time.Millisecond)
			log.Printf("[client] tun read error: %v", err)
		}
		if n == 0 || n < 20 {
			log.Printf("[client] tun read error: %v", err)
		}
		pkt := buf[:n]
		version := pkt[0] >> 4
		if version != 4 {
			log.Printf("[client] tun read error: %v", err)
		}
		dstIP := net.IP(pkt[16:20]).String()
		c.sdwanMu.RLock()
		self := c.sdwanIP
		c.sdwanMu.RUnlock()
		if dstIP == self {
			log.Printf("[client] tun read error: %v", err)
		}
		// send raw binary to avoid JSON base64 overhead
		log.Printf("[client] tun: read pkt len=%d dst=%s", n, dstIP)
		frame := protocol.EncodeRaw(protocol.MsgTunnel, protocol.SubTunnelSDWANRaw, pkt)
		_ = c.conn.WriteRaw(frame)
	}
}

func (c *Client) writeTUN(payload []byte) error {
	c.tunMu.Lock()
	f := c.tunFile
	c.tunMu.Unlock()
	if f == nil {
		return nil
	}
	_, err := f.Write(payload)
	return err
}

func (c *Client) IsStopping() bool {
	select {
	case <-c.quit:
		return true
	default:
		return false
	}
}

func runCmd(name string, args ...string) error {
	cmd := exec.Command(name, args...)
	out, err := cmd.CombinedOutput()
	if err != nil {
		return fmt.Errorf("%s %v: %w: %s", name, args, err, strings.TrimSpace(string(out)))
	}
	return nil
}

// Stop shuts down the client.
func (c *Client) Stop() {
	close(c.quit)
	c.tunMu.Lock()
	if c.tunFile != nil {
		_ = c.tunFile.Close()
		c.tunFile = nil
	}
	c.tunMu.Unlock()
	if c.conn != nil {
		c.conn.Close()
	}
	if c.relayMgr != nil {
		c.relayMgr.Stop()
	}
	c.tMu.Lock()
	for _, t := range c.tunnels {
		t.Close()
	}
	c.tMu.Unlock()
	c.wg.Wait()
}

// ─── helpers ───

func getLocalIP() string {
	// Simple heuristic: find the first non-loopback IPv4
	addrs, _ := os.Hostname()
	_ = addrs
	return "0.0.0.0" // placeholder, will be properly implemented
}