netmaker/netclient/functions/daemon.go

package functions

import (
	"context"
	"encoding/json"
	"errors"
	"fmt"
	"io"
	"log"
	"net"
	"net/http"
	"os"
	"os/signal"
	"strings"
	"sync"
	"syscall"
	"time"

	mqtt "github.com/eclipse/paho.mqtt.golang"
	"github.com/gravitl/netmaker/logger"
	"github.com/gravitl/netmaker/models"
	"github.com/gravitl/netmaker/mq"
	"github.com/gravitl/netmaker/netclient/auth"
	"github.com/gravitl/netmaker/netclient/config"
	"github.com/gravitl/netmaker/netclient/global_settings"
	"github.com/gravitl/netmaker/netclient/local"
	"github.com/gravitl/netmaker/netclient/ncutils"
	"github.com/gravitl/netmaker/netclient/wireguard"

	nmproxy "github.com/gravitl/netmaker/nm-proxy"
	"github.com/gravitl/netmaker/nm-proxy/manager"
	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
)

var ProxyMgmChan = make(chan *manager.ManagerAction, 100)
var messageCache = new(sync.Map)
var ProxyStatus = "OFF"
var serverSet map[string]bool

var mqclient mqtt.Client

const lastNodeUpdate = "lnu"
const lastPeerUpdate = "lpu"

type cachedMessage struct {
	Message  string
	LastSeen time.Time
}

// Daemon runs netclient daemon from command line
func Daemon() error {
	logger.Log(0, "netclient daemon started -- version:", ncutils.Version)
	UpdateClientConfig()
	if err := ncutils.SavePID(); err != nil {
		return err
	}
	// reference required to eliminate unused statticcheck
	serverSet = make(map[string]bool)
	serverSet["dummy"] = false
	// set ipforwarding on startup
	err := local.SetIPForwarding()
	if err != nil {
		logger.Log(0, err.Error())
	}

	// == add waitgroup and cancel for checkin routine ==
	wg := sync.WaitGroup{}
	quit := make(chan os.Signal, 1)
	reset := make(chan os.Signal, 1)
	signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
	signal.Notify(reset, syscall.SIGHUP)
	cancel := startGoRoutines(&wg)
	for {
		select {
		case <-quit:
			cancel()
			logger.Log(0, "shutting down netclient daemon")
			wg.Wait()
			if mqclient != nil {
				mqclient.Disconnect(250)
			}
			logger.Log(0, "shutdown complete")
			return nil
		case <-reset:
			logger.Log(0, "received reset")
			cancel()
			wg.Wait()
			if mqclient != nil {
				mqclient.Disconnect(250)
			}
			logger.Log(0, "restarting daemon")
			cancel = startGoRoutines(&wg)
		}
	}
}

func startGoRoutines(wg *sync.WaitGroup) context.CancelFunc {
	ctx, cancel := context.WithCancel(context.Background())
	serverSet := make(map[string]bool)
	networks, _ := ncutils.GetSystemNetworks()
	for _, network := range networks {
		logger.Log(3, "initializing network", network)
		cfg := config.ClientConfig{}
		cfg.Network = network
		cfg.ReadConfig()
		if cfg.Node.Connected == "yes" {
			if err := wireguard.ApplyConf(&cfg.Node, cfg.Node.Interface, ncutils.GetNetclientPathSpecific()+cfg.Node.Interface+".conf"); err != nil {
				logger.Log(0, "failed to start ", cfg.Node.Interface, "wg interface", err.Error())
			}
			if cfg.PublicIPService != "" {
				global_settings.PublicIPServices[network] = cfg.PublicIPService
			}
		}

		server := cfg.Server.Server
		if !serverSet[server] {
			// == subscribe to all nodes for each on machine ==
			serverSet[server] = true
			logger.Log(1, "started daemon for server ", server)
			local.SetNetmakerDomainRoute(cfg.Server.API)
			wg.Add(1)
			go messageQueue(ctx, wg, &cfg)
		}
	}
	wg.Add(1)
	go Checkin(ctx, wg)

	if len(networks) != 0 {
		go func() {
			cfg := config.ClientConfig{}
			cfg.Network = networks[0]
			cfg.ReadConfig()
			apiHost, _, err := net.SplitHostPort(cfg.Server.API)
			if err == nil {
				if ProxyStatus != "ON" {
					ProxyStatus = "ON"
					pCtx, pCancel := context.WithCancel(context.Background())
					go nmproxy.Start(pCtx, ProxyMgmChan, apiHost)
					quit := make(chan os.Signal, 1)
					signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
					<-quit
					pCancel()
					logger.Log(0, "Proxy Shutting down....")
				}

			}
		}()

	}

	go func(networks []string) {

		for _, network := range networks {
			logger.Log(0, "Collecting interface and peers info to configure proxy...")
			cfg := config.ClientConfig{}
			cfg.Network = network
			cfg.ReadConfig()
			node, err := GetNodeInfo(&cfg)
			if err != nil {
				log.Println("Failed to get node info: ", err)
				continue
			}
			ProxyMgmChan <- &manager.ManagerAction{
				Action: manager.AddInterface,
				Payload: manager.ManagerPayload{
					InterfaceName: node.Node.Interface,
					Peers:         node.Peers,
				},
			}

		}

	}(networks)
	return cancel
}
func GetNodeInfo(cfg *config.ClientConfig) (models.NodeGet, error) {
	var nodeGET models.NodeGet
	token, err := Authenticate(cfg)
	if err != nil {
		return nodeGET, err
	}
	url := fmt.Sprintf("https://%s/api/nodes/%s/%s", cfg.Server.API, cfg.Network, cfg.Node.ID)
	response, err := API("", http.MethodGet, url, token)
	if err != nil {
		return nodeGET, err
	}
	if response.StatusCode != http.StatusOK {
		bytes, err := io.ReadAll(response.Body)
		if err != nil {
			fmt.Println(err)
		}
		return nodeGET, (fmt.Errorf("%s %w", string(bytes), err))
	}
	defer response.Body.Close()
	if err := json.NewDecoder(response.Body).Decode(&nodeGET); err != nil {
		return nodeGET, fmt.Errorf("error decoding node %w", err)
	}
	return nodeGET, nil
}

// UpdateKeys -- updates private key and returns new publickey
func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
	logger.Log(0, "interface:", nodeCfg.Node.Interface, "received message to update wireguard keys for network ", nodeCfg.Network)
	key, err := wgtypes.GeneratePrivateKey()
	if err != nil {
		logger.Log(0, "network:", nodeCfg.Node.Network, "error generating privatekey ", err.Error())
		return err
	}
	file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
	if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
		logger.Log(0, "network:", nodeCfg.Node.Network, "error updating wireguard key ", err.Error())
		return err
	}
	if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
		logger.Log(0, "network:", nodeCfg.Network, "failed to save private key", storeErr.Error())
		return storeErr
	}

	nodeCfg.Node.PublicKey = key.PublicKey().String()
	PublishNodeUpdate(nodeCfg)
	return nil
}

// == Private ==

// sets MQ client subscriptions for a specific node config
// should be called for each node belonging to a given server
func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
	if token := client.Subscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.WaitTimeout(mq.MQ_TIMEOUT*time.Second) && token.Error() != nil {
		if token.Error() == nil {
			logger.Log(0, "network:", nodeCfg.Node.Network, "connection timeout")
		} else {
			logger.Log(0, "network:", nodeCfg.Node.Network, token.Error().Error())
		}
		return
	}
	if token := client.Subscribe(fmt.Sprintf("proxy/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(ProxyUpdate)); token.WaitTimeout(mq.MQ_TIMEOUT*time.Second) && token.Error() != nil {
		if token.Error() == nil {
			logger.Log(0, "network:", nodeCfg.Node.Network, "connection timeout")
		} else {
			logger.Log(0, "network:", nodeCfg.Node.Network, token.Error().Error())
		}
		return
	}
	logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
	if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
		logger.Log(0, "network", nodeCfg.Node.Network, token.Error().Error())
		return
	}
	logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
}

// on a delete usually, pass in the nodecfg to unsubscribe client broker communications
// for the node in nodeCfg
func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
	client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
	var ok = true
	if token := client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.WaitTimeout(mq.MQ_TIMEOUT*time.Second) && token.Error() != nil {
		if token.Error() == nil {
			logger.Log(1, "network:", nodeCfg.Node.Network, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", "connection timeout")
		} else {
			logger.Log(1, "network:", nodeCfg.Node.Network, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
		}
		ok = false
	}
	if token := client.Unsubscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.WaitTimeout(mq.MQ_TIMEOUT*time.Second) && token.Error() != nil {
		if token.Error() == nil {
			logger.Log(1, "network:", nodeCfg.Node.Network, "unable to unsubscribe from peer updates for node", nodeCfg.Node.Name, "\n", "connection timeout")
		} else {
			logger.Log(1, "network:", nodeCfg.Node.Network, "unable to unsubscribe from peer updates for node", nodeCfg.Node.Name, "\n", token.Error().Error())
		}
		ok = false
	}
	if ok {
		logger.Log(1, "network:", nodeCfg.Node.Network, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
	}
}

// sets up Message Queue and subsribes/publishes updates to/from server
// the client should subscribe to ALL nodes that exist on server locally
func messageQueue(ctx context.Context, wg *sync.WaitGroup, cfg *config.ClientConfig) {
	defer wg.Done()
	logger.Log(0, "network:", cfg.Node.Network, "netclient message queue started for server:", cfg.Server.Server)
	err := setupMQTT(cfg)
	if err != nil {
		logger.Log(0, "unable to connect to broker", cfg.Server.Server, err.Error())
		return
	}
	//defer mqclient.Disconnect(250)
	<-ctx.Done()
	logger.Log(0, "shutting down message queue for server", cfg.Server.Server)
}

// func setMQTTSingenton creates a connection to broker for single use (ie to publish a message)
// only to be called from cli (eg. connect/disconnect, join, leave) and not from daemon ---
func setupMQTTSingleton(cfg *config.ClientConfig) error {
	opts := mqtt.NewClientOptions()
	server := cfg.Server.Server
	port := cfg.Server.MQPort
	pass, err := os.ReadFile(ncutils.GetNetclientPathSpecific() + "secret-" + cfg.Network)
	if err != nil {
		return fmt.Errorf("could not read secrets file %w", err)
	}
	opts.AddBroker("mqtts://" + server + ":" + port)
	opts.SetUsername(cfg.Node.ID)
	opts.SetPassword(string(pass))
	mqclient = mqtt.NewClient(opts)
	var connecterr error
	opts.SetClientID(ncutils.MakeRandomString(23))
	if token := mqclient.Connect(); !token.WaitTimeout(30*time.Second) || token.Error() != nil {
		logger.Log(0, "unable to connect to broker, retrying ...")
		if token.Error() == nil {
			connecterr = errors.New("connect timeout")
		} else {
			connecterr = token.Error()
		}
	}
	return connecterr
}

// setupMQTT creates a connection to broker and returns client
// this function is primarily used to create a connection to publish to the broker
func setupMQTT(cfg *config.ClientConfig) error {
	opts := mqtt.NewClientOptions()
	server := cfg.Server.Server
	port := cfg.Server.MQPort
	pass, err := os.ReadFile(ncutils.GetNetclientPathSpecific() + "secret-" + cfg.Network)
	if err != nil {
		return fmt.Errorf("could not read secrets file %w", err)
	}
	opts.AddBroker(fmt.Sprintf("mqtts://%s:%s", server, port))
	opts.SetUsername(cfg.Node.ID)
	opts.SetPassword(string(pass))
	opts.SetClientID(ncutils.MakeRandomString(23))
	opts.SetDefaultPublishHandler(All)
	opts.SetAutoReconnect(true)
	opts.SetConnectRetry(true)
	opts.SetConnectRetryInterval(time.Second << 2)
	opts.SetKeepAlive(time.Minute >> 1)
	opts.SetWriteTimeout(time.Minute)

	opts.SetOnConnectHandler(func(client mqtt.Client) {
		networks, err := ncutils.GetSystemNetworks()
		if err != nil {
			logger.Log(0, "error retriving networks", err.Error())
		}
		for _, network := range networks {
			var currNodeCfg config.ClientConfig
			currNodeCfg.Network = network
			currNodeCfg.ReadConfig()
			setSubscriptions(client, &currNodeCfg)
		}
	})
	opts.SetOrderMatters(true)
	opts.SetResumeSubs(true)
	opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
		logger.Log(0, "network:", cfg.Node.Network, "detected broker connection lost for", cfg.Server.Server)
	})
	mqclient = mqtt.NewClient(opts)
	var connecterr error
	for count := 0; count < 3; count++ {
		connecterr = nil
		if token := mqclient.Connect(); !token.WaitTimeout(30*time.Second) || token.Error() != nil {
			logger.Log(0, "unable to connect to broker, retrying ...")
			if token.Error() == nil {
				connecterr = errors.New("connect timeout")
			} else {
				connecterr = token.Error()
			}
			if err := checkBroker(cfg.Server.Server, cfg.Server.MQPort); err != nil {
				logger.Log(0, "could not connect to broker", cfg.Server.Server, err.Error())
			}
		}
	}
	if connecterr != nil {
		logger.Log(0, "failed to establish connection to broker: ", connecterr.Error())
		return connecterr
	}

	return nil
}

// publishes a message to server to update peers on this peer's behalf
func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
	if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
		return err
	}
	return nil
}

func parseNetworkFromTopic(topic string) string {
	return strings.Split(topic, "/")[1]
}

// should only ever use node client configs
func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
	if len(msg) <= 24 { // make sure message is of appropriate length
		return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
	}

	// setup the keys
	diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
	if keyErr != nil {
		return nil, keyErr
	}

	serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
	if err != nil {
		return nil, err
	}

	return ncutils.DeChunk(msg, serverPubKey, diskKey)
}

// == Message Caches ==

func insert(network, which, cache string) {
	var newMessage = cachedMessage{
		Message:  cache,
		LastSeen: time.Now(),
	}
	messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
}

func read(network, which string) string {
	val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
	if isok {
		var readMessage = val.(cachedMessage) // fetch current cached message
		if readMessage.LastSeen.IsZero() {
			return ""
		}
		if time.Now().After(readMessage.LastSeen.Add(time.Hour * 24)) { // check if message has been there over a minute
			messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
			return ""
		}
		return readMessage.Message // return current message if not expired
	}
	return ""
}