daemon.go 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334
  1. package functions
  2. import (
  3. "context"
  4. "crypto/ed25519"
  5. "crypto/rand"
  6. "crypto/tls"
  7. "crypto/x509"
  8. "errors"
  9. "fmt"
  10. "log"
  11. "os"
  12. "os/signal"
  13. "strings"
  14. "sync"
  15. "syscall"
  16. "time"
  17. mqtt "github.com/eclipse/paho.mqtt.golang"
  18. "github.com/gravitl/netmaker/logger"
  19. "github.com/gravitl/netmaker/netclient/auth"
  20. "github.com/gravitl/netmaker/netclient/config"
  21. "github.com/gravitl/netmaker/netclient/daemon"
  22. "github.com/gravitl/netmaker/netclient/ncutils"
  23. "github.com/gravitl/netmaker/netclient/wireguard"
  24. ssl "github.com/gravitl/netmaker/tls"
  25. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  26. )
  27. var messageCache = new(sync.Map)
  28. var networkcontext = new(sync.Map)
  29. const lastNodeUpdate = "lnu"
  30. const lastPeerUpdate = "lpu"
  31. type cachedMessage struct {
  32. Message string
  33. LastSeen time.Time
  34. }
  35. // Daemon runs netclient daemon from command line
  36. func Daemon() error {
  37. serverSet := make(map[string]config.ClientConfig)
  38. // == initial pull of all networks ==
  39. networks, _ := ncutils.GetSystemNetworks()
  40. if len(networks) == 0 {
  41. return errors.New("no networks")
  42. }
  43. for _, network := range networks {
  44. logger.Log(3, "initializing network", network)
  45. cfg := config.ClientConfig{}
  46. cfg.Network = network
  47. cfg.ReadConfig()
  48. serverSet[cfg.Server.Server] = cfg
  49. if err := wireguard.ApplyConf(&cfg.Node, cfg.Node.Interface, ncutils.GetNetclientPathSpecific()+cfg.Node.Interface+".conf"); err != nil {
  50. logger.Log(0, "failed to start ", cfg.Node.Interface, "wg interface", err.Error())
  51. }
  52. //initialPull(cfg.Network)
  53. }
  54. // == subscribe to all nodes for each on machine ==
  55. for server, config := range serverSet {
  56. logger.Log(1, "started daemon for server ", server)
  57. ctx, cancel := context.WithCancel(context.Background())
  58. networkcontext.Store(server, cancel)
  59. go messageQueue(ctx, &config)
  60. }
  61. // == add waitgroup and cancel for checkin routine ==
  62. wg := sync.WaitGroup{}
  63. ctx, cancel := context.WithCancel(context.Background())
  64. wg.Add(1)
  65. go Checkin(ctx, &wg)
  66. quit := make(chan os.Signal, 1)
  67. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  68. <-quit
  69. for server := range serverSet {
  70. if cancel, ok := networkcontext.Load(server); ok {
  71. cancel.(context.CancelFunc)()
  72. }
  73. }
  74. cancel()
  75. logger.Log(0, "shutting down netclient daemon")
  76. wg.Wait()
  77. logger.Log(0, "shutdown complete")
  78. return nil
  79. }
  80. // UpdateKeys -- updates private key and returns new publickey
  81. func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
  82. logger.Log(0, "received message to update wireguard keys for network ", nodeCfg.Network)
  83. key, err := wgtypes.GeneratePrivateKey()
  84. if err != nil {
  85. logger.Log(0, "error generating privatekey ", err.Error())
  86. return err
  87. }
  88. file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
  89. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  90. logger.Log(0, "error updating wireguard key ", err.Error())
  91. return err
  92. }
  93. if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
  94. logger.Log(0, "failed to save private key", storeErr.Error())
  95. return storeErr
  96. }
  97. nodeCfg.Node.PublicKey = key.PublicKey().String()
  98. PublishNodeUpdate(nodeCfg)
  99. return nil
  100. }
  101. // == Private ==
  102. // sets MQ client subscriptions for a specific node config
  103. // should be called for each node belonging to a given server
  104. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  105. if nodeCfg.DebugOn {
  106. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  107. logger.Log(0, token.Error().Error())
  108. return
  109. }
  110. logger.Log(0, "subscribed to all topics for debugging purposes")
  111. }
  112. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  113. logger.Log(0, token.Error().Error())
  114. return
  115. }
  116. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  117. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  118. logger.Log(0, token.Error().Error())
  119. return
  120. }
  121. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  122. }
  123. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  124. // for the node in nodeCfg
  125. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  126. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  127. var ok = true
  128. if token := client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  129. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  130. ok = false
  131. }
  132. if token := client.Unsubscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  133. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  134. ok = false
  135. }
  136. if ok {
  137. logger.Log(1, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  138. }
  139. }
  140. // sets up Message Queue and subsribes/publishes updates to/from server
  141. // the client should subscribe to ALL nodes that exist on server locally
  142. func messageQueue(ctx context.Context, cfg *config.ClientConfig) {
  143. logger.Log(0, "netclient daemon started for server: ", cfg.Server.Server)
  144. client := setupMQTT(cfg, false)
  145. defer client.Disconnect(250)
  146. <-ctx.Done()
  147. logger.Log(0, "shutting down daemon for server ", cfg.Server.Server)
  148. }
  149. // NewTLSConf sets up tls configuration to connect to broker securely
  150. func NewTLSConfig(server string) *tls.Config {
  151. file := ncutils.GetNetclientServerPath(server) + ncutils.GetSeparator() + "root.pem"
  152. certpool := x509.NewCertPool()
  153. ca, err := os.ReadFile(file)
  154. if err != nil {
  155. logger.Log(0, "could not read CA file ", err.Error())
  156. }
  157. ok := certpool.AppendCertsFromPEM(ca)
  158. if !ok {
  159. logger.Log(0, "failed to append cert")
  160. }
  161. clientKeyPair, err := tls.LoadX509KeyPair(ncutils.GetNetclientServerPath(server)+ncutils.GetSeparator()+"client.pem", ncutils.GetNetclientPath()+ncutils.GetSeparator()+"client.key")
  162. if err != nil {
  163. log.Fatalf("could not read client cert/key %v \n", err)
  164. }
  165. certs := []tls.Certificate{clientKeyPair}
  166. return &tls.Config{
  167. RootCAs: certpool,
  168. ClientAuth: tls.NoClientCert,
  169. ClientCAs: nil,
  170. Certificates: certs,
  171. InsecureSkipVerify: false,
  172. }
  173. }
  174. // setupMQTT creates a connection to broker and returns client
  175. // this function is primarily used to create a connection to publish to the broker
  176. func setupMQTT(cfg *config.ClientConfig, publish bool) mqtt.Client {
  177. opts := mqtt.NewClientOptions()
  178. server := cfg.Server.Server
  179. opts.AddBroker("ssl://" + server + ":8883") // TODO get the appropriate port of the comms mq server
  180. opts.SetTLSConfig(NewTLSConfig(server))
  181. opts.SetClientID(ncutils.MakeRandomString(23))
  182. opts.SetDefaultPublishHandler(All)
  183. opts.SetAutoReconnect(true)
  184. opts.SetConnectRetry(true)
  185. opts.SetConnectRetryInterval(time.Second << 2)
  186. opts.SetKeepAlive(time.Minute >> 1)
  187. opts.SetWriteTimeout(time.Minute)
  188. opts.SetOnConnectHandler(func(client mqtt.Client) {
  189. if !publish {
  190. networks, err := ncutils.GetSystemNetworks()
  191. if err != nil {
  192. logger.Log(0, "error retriving networks ", err.Error())
  193. }
  194. for _, network := range networks {
  195. var currNodeCfg config.ClientConfig
  196. currNodeCfg.Network = network
  197. currNodeCfg.ReadConfig()
  198. setSubscriptions(client, &currNodeCfg)
  199. }
  200. }
  201. })
  202. opts.SetOrderMatters(true)
  203. opts.SetResumeSubs(true)
  204. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  205. logger.Log(0, "detected broker connection lost for", cfg.Server.Server)
  206. })
  207. client := mqtt.NewClient(opts)
  208. for token := client.Connect(); !token.WaitTimeout(30*time.Second) || token.Error() != nil; token = client.Connect() {
  209. logger.Log(0, "unable to connect to broker, retrying ...")
  210. var err error
  211. if token.Error() == nil {
  212. err = errors.New("connect timeout")
  213. } else {
  214. err = token.Error()
  215. }
  216. logger.Log(0, "could not connect to broker", cfg.Server.Server, err.Error())
  217. if strings.Contains(err.Error(), "connectex") || strings.Contains(err.Error(), "connect timeout") {
  218. logger.Log(0, "connection issue detected.. attempt connection with new certs")
  219. key, err := ssl.ReadKey(ncutils.GetNetclientPath() + ncutils.GetSeparator() + "client.key")
  220. if err != nil {
  221. _, *key, err = ed25519.GenerateKey(rand.Reader)
  222. if err != nil {
  223. log.Fatal("could not generate new key")
  224. }
  225. }
  226. RegisterWithServer(key, cfg)
  227. daemon.Restart()
  228. }
  229. }
  230. return client
  231. }
  232. // publishes a message to server to update peers on this peer's behalf
  233. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  234. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  235. return err
  236. }
  237. return nil
  238. }
  239. func initialPull(network string) {
  240. logger.Log(0, "pulling latest config for ", network)
  241. var configPath = fmt.Sprintf("%snetconfig-%s", ncutils.GetNetclientPathSpecific(), network)
  242. fileInfo, err := os.Stat(configPath)
  243. if err != nil {
  244. logger.Log(0, "could not stat config file: ", configPath)
  245. return
  246. }
  247. // speed up UDP rest
  248. if !fileInfo.ModTime().IsZero() && time.Now().After(fileInfo.ModTime().Add(time.Minute)) {
  249. sleepTime := 2
  250. for {
  251. _, err := Pull(network, true)
  252. if err == nil {
  253. break
  254. }
  255. if sleepTime > 3600 {
  256. sleepTime = 3600
  257. }
  258. logger.Log(0, "failed to pull for network ", network)
  259. logger.Log(0, fmt.Sprintf("waiting %d seconds to retry...", sleepTime))
  260. time.Sleep(time.Second * time.Duration(sleepTime))
  261. sleepTime = sleepTime * 2
  262. }
  263. time.Sleep(time.Second << 1)
  264. }
  265. }
  266. func parseNetworkFromTopic(topic string) string {
  267. return strings.Split(topic, "/")[1]
  268. }
  269. // should only ever use node client configs
  270. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  271. if len(msg) <= 24 { // make sure message is of appropriate length
  272. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  273. }
  274. // setup the keys
  275. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  276. if keyErr != nil {
  277. return nil, keyErr
  278. }
  279. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  280. if err != nil {
  281. return nil, err
  282. }
  283. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  284. }
  285. // == Message Caches ==
  286. func insert(network, which, cache string) {
  287. var newMessage = cachedMessage{
  288. Message: cache,
  289. LastSeen: time.Now(),
  290. }
  291. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  292. }
  293. func read(network, which string) string {
  294. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  295. if isok {
  296. var readMessage = val.(cachedMessage) // fetch current cached message
  297. if readMessage.LastSeen.IsZero() {
  298. return ""
  299. }
  300. if time.Now().After(readMessage.LastSeen.Add(time.Minute * 10)) { // check if message has been there over a minute
  301. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  302. return ""
  303. }
  304. return readMessage.Message // return current message if not expired
  305. }
  306. return ""
  307. }