daemon.go 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365
  1. package functions
  2. import (
  3. "context"
  4. "crypto/tls"
  5. "crypto/x509"
  6. "errors"
  7. "fmt"
  8. "log"
  9. "os"
  10. "os/signal"
  11. "strings"
  12. "sync"
  13. "syscall"
  14. "time"
  15. mqtt "github.com/eclipse/paho.mqtt.golang"
  16. "github.com/go-ping/ping"
  17. "github.com/gravitl/netmaker/logger"
  18. "github.com/gravitl/netmaker/netclient/auth"
  19. "github.com/gravitl/netmaker/netclient/config"
  20. "github.com/gravitl/netmaker/netclient/daemon"
  21. "github.com/gravitl/netmaker/netclient/ncutils"
  22. "github.com/gravitl/netmaker/netclient/wireguard"
  23. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  24. )
  25. var messageCache = new(sync.Map)
  26. var networkcontext = new(sync.Map)
  27. const lastNodeUpdate = "lnu"
  28. const lastPeerUpdate = "lpu"
  29. type cachedMessage struct {
  30. Message string
  31. LastSeen time.Time
  32. }
  33. // Daemon runs netclient daemon from command line
  34. func Daemon() error {
  35. var exists = struct{}{}
  36. serverSet := make(map[string]struct{})
  37. // == initial pull of all networks ==
  38. networks, _ := ncutils.GetSystemNetworks()
  39. if len(networks) == 0 {
  40. return errors.New("no networks")
  41. }
  42. for _, network := range networks {
  43. logger.Log(3, "initializing network", network)
  44. cfg := config.ClientConfig{}
  45. cfg.Network = network
  46. cfg.ReadConfig()
  47. serverSet[cfg.Server.Server] = exists
  48. //temporary code --- remove in version v0.13.0
  49. removeHostDNS(network, ncutils.IsWindows())
  50. // end of code to be removed in version v0.13.0
  51. initialPull(cfg.Network)
  52. }
  53. // == subscribe to all nodes for each on machine ==
  54. for server := range serverSet {
  55. logger.Log(1, "started daemon for server ", server)
  56. ctx, cancel := context.WithCancel(context.Background())
  57. networkcontext.Store(server, cancel)
  58. go messageQueue(ctx, server)
  59. }
  60. // == add waitgroup and cancel for checkin routine ==
  61. wg := sync.WaitGroup{}
  62. ctx, cancel := context.WithCancel(context.Background())
  63. wg.Add(1)
  64. go Checkin(ctx, &wg)
  65. quit := make(chan os.Signal, 1)
  66. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  67. <-quit
  68. for server := range serverSet {
  69. if cancel, ok := networkcontext.Load(server); ok {
  70. cancel.(context.CancelFunc)()
  71. }
  72. }
  73. cancel()
  74. logger.Log(0, "shutting down netclient daemon")
  75. wg.Wait()
  76. logger.Log(0, "shutdown complete")
  77. return nil
  78. }
  79. // UpdateKeys -- updates private key and returns new publickey
  80. func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
  81. logger.Log(0, "received message to update wireguard keys for network ", nodeCfg.Network)
  82. key, err := wgtypes.GeneratePrivateKey()
  83. if err != nil {
  84. logger.Log(0, "error generating privatekey ", err.Error())
  85. return err
  86. }
  87. file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
  88. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  89. logger.Log(0, "error updating wireguard key ", err.Error())
  90. return err
  91. }
  92. if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
  93. logger.Log(0, "failed to save private key", storeErr.Error())
  94. return storeErr
  95. }
  96. nodeCfg.Node.PublicKey = key.PublicKey().String()
  97. PublishNodeUpdate(nodeCfg)
  98. return nil
  99. }
  100. // PingServer -- checks if server is reachable
  101. func PingServer(cfg *config.ClientConfig) error {
  102. pinger, err := ping.NewPinger(cfg.Server.Server)
  103. if err != nil {
  104. return err
  105. }
  106. pinger.Timeout = 2 * time.Second
  107. pinger.Run()
  108. stats := pinger.Statistics()
  109. if stats.PacketLoss == 100 {
  110. return errors.New("ping error")
  111. }
  112. logger.Log(3, "ping of server", cfg.Server.Server, "was successful")
  113. return nil
  114. }
  115. // == Private ==
  116. // sets MQ client subscriptions for a specific node config
  117. // should be called for each node belonging to a given server
  118. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  119. if nodeCfg.DebugOn {
  120. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  121. logger.Log(0, token.Error().Error())
  122. return
  123. }
  124. logger.Log(0, "subscribed to all topics for debugging purposes")
  125. }
  126. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  127. logger.Log(0, token.Error().Error())
  128. return
  129. }
  130. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  131. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  132. logger.Log(0, token.Error().Error())
  133. return
  134. }
  135. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  136. }
  137. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  138. // for the node in nodeCfg
  139. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  140. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  141. var ok = true
  142. if token := client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  143. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  144. ok = false
  145. }
  146. if token := client.Unsubscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  147. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  148. ok = false
  149. }
  150. if ok {
  151. logger.Log(1, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  152. }
  153. }
  154. // sets up Message Queue and subsribes/publishes updates to/from server
  155. // the client should subscribe to ALL nodes that exist on server locally
  156. func messageQueue(ctx context.Context, server string) {
  157. logger.Log(0, "netclient daemon started for server: ", server)
  158. client := setupMQTT(nil, server, false)
  159. defer client.Disconnect(250)
  160. <-ctx.Done()
  161. logger.Log(0, "shutting down daemon for server ", server)
  162. }
  163. // NewTLSConf sets up tls configuration to connect to broker securely
  164. func NewTLSConfig(server string) *tls.Config {
  165. file := ncutils.GetNetclientServerPath(server) + ncutils.GetSeparator() + "root.pem"
  166. certpool := x509.NewCertPool()
  167. ca, err := os.ReadFile(file)
  168. if err != nil {
  169. logger.Log(0, "could not read CA file ", err.Error())
  170. }
  171. ok := certpool.AppendCertsFromPEM(ca)
  172. if !ok {
  173. logger.Log(0, "failed to append cert")
  174. }
  175. clientKeyPair, err := tls.LoadX509KeyPair(ncutils.GetNetclientServerPath(server)+ncutils.GetSeparator()+"client.pem", ncutils.GetNetclientPath()+"/client.key")
  176. if err != nil {
  177. log.Fatalf("could not read client cert/key %v \n", err)
  178. }
  179. certs := []tls.Certificate{clientKeyPair}
  180. return &tls.Config{
  181. RootCAs: certpool,
  182. ClientAuth: tls.NoClientCert,
  183. ClientCAs: nil,
  184. Certificates: certs,
  185. InsecureSkipVerify: false,
  186. }
  187. }
  188. // setupMQTT creates a connection to broker and returns client
  189. // this function is primarily used to create a connection to publish to the broker
  190. func setupMQTT(cfg *config.ClientConfig, server string, publish bool) mqtt.Client {
  191. opts := mqtt.NewClientOptions()
  192. if cfg != nil {
  193. server = cfg.Server.Server
  194. }
  195. opts.AddBroker("ssl://" + server + ":8883") // TODO get the appropriate port of the comms mq server
  196. opts.SetTLSConfig(NewTLSConfig(server))
  197. opts.SetClientID(ncutils.MakeRandomString(23))
  198. opts.SetDefaultPublishHandler(All)
  199. opts.SetAutoReconnect(true)
  200. opts.SetConnectRetry(true)
  201. opts.SetConnectRetryInterval(time.Second << 2)
  202. opts.SetKeepAlive(time.Minute >> 1)
  203. opts.SetWriteTimeout(time.Minute)
  204. opts.SetOnConnectHandler(func(client mqtt.Client) {
  205. if !publish {
  206. networks, err := ncutils.GetSystemNetworks()
  207. if err != nil {
  208. logger.Log(0, "error retriving networks ", err.Error())
  209. }
  210. for _, network := range networks {
  211. var currNodeCfg config.ClientConfig
  212. currNodeCfg.Network = network
  213. currNodeCfg.ReadConfig()
  214. setSubscriptions(client, &currNodeCfg)
  215. }
  216. }
  217. })
  218. opts.SetOrderMatters(true)
  219. opts.SetResumeSubs(true)
  220. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  221. logger.Log(0, "detected broker connection lost, running pull for ", cfg.Node.Network)
  222. _, err := Pull(cfg.Node.Network, true)
  223. if err != nil {
  224. logger.Log(0, "could not run pull, server unreachable: ", err.Error())
  225. logger.Log(0, "waiting to retry...")
  226. }
  227. logger.Log(0, "connection re-established with mqtt server")
  228. })
  229. client := mqtt.NewClient(opts)
  230. tperiod := time.Now().Add(12 * time.Second)
  231. for {
  232. //if after 12 seconds, try a pull on the last try
  233. if time.Now().After(tperiod) {
  234. logger.Log(0, "running pull for ", cfg.Node.Network)
  235. _, err := Pull(cfg.Node.Network, true)
  236. if err != nil {
  237. logger.Log(0, "could not run pull, exiting ", cfg.Node.Network, " setup: ", err.Error())
  238. return client
  239. }
  240. time.Sleep(time.Second)
  241. }
  242. if token := client.Connect(); token.Wait() && token.Error() != nil {
  243. logger.Log(0, "unable to connect to broker, retrying ...")
  244. if time.Now().After(tperiod) {
  245. logger.Log(0, "could not connect to broker, exiting ", cfg.Node.Network, " setup: ", token.Error().Error())
  246. if strings.Contains(token.Error().Error(), "connectex") || strings.Contains(token.Error().Error(), "i/o timeout") {
  247. logger.Log(0, "connection issue detected.. pulling and restarting daemon")
  248. Pull(cfg.Node.Network, true)
  249. daemon.Restart()
  250. }
  251. return client
  252. }
  253. } else {
  254. break
  255. }
  256. time.Sleep(2 * time.Second)
  257. }
  258. return client
  259. }
  260. // publishes a message to server to update peers on this peer's behalf
  261. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  262. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  263. return err
  264. }
  265. return nil
  266. }
  267. func initialPull(network string) {
  268. logger.Log(0, "pulling latest config for ", network)
  269. var configPath = fmt.Sprintf("%snetconfig-%s", ncutils.GetNetclientPathSpecific(), network)
  270. fileInfo, err := os.Stat(configPath)
  271. if err != nil {
  272. logger.Log(0, "could not stat config file: ", configPath)
  273. return
  274. }
  275. // speed up UDP rest
  276. if !fileInfo.ModTime().IsZero() && time.Now().After(fileInfo.ModTime().Add(time.Minute)) {
  277. sleepTime := 2
  278. for {
  279. _, err := Pull(network, true)
  280. if err == nil {
  281. break
  282. }
  283. if sleepTime > 3600 {
  284. sleepTime = 3600
  285. }
  286. logger.Log(0, "failed to pull for network ", network)
  287. logger.Log(0, fmt.Sprintf("waiting %d seconds to retry...", sleepTime))
  288. time.Sleep(time.Second * time.Duration(sleepTime))
  289. sleepTime = sleepTime * 2
  290. }
  291. time.Sleep(time.Second << 1)
  292. }
  293. }
  294. func parseNetworkFromTopic(topic string) string {
  295. return strings.Split(topic, "/")[1]
  296. }
  297. // should only ever use node client configs
  298. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  299. if len(msg) <= 24 { // make sure message is of appropriate length
  300. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  301. }
  302. // setup the keys
  303. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  304. if keyErr != nil {
  305. return nil, keyErr
  306. }
  307. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  308. if err != nil {
  309. return nil, err
  310. }
  311. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  312. }
  313. // == Message Caches ==
  314. func insert(network, which, cache string) {
  315. var newMessage = cachedMessage{
  316. Message: cache,
  317. LastSeen: time.Now(),
  318. }
  319. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  320. }
  321. func read(network, which string) string {
  322. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  323. if isok {
  324. var readMessage = val.(cachedMessage) // fetch current cached message
  325. if readMessage.LastSeen.IsZero() {
  326. return ""
  327. }
  328. if time.Now().After(readMessage.LastSeen.Add(time.Minute * 10)) { // check if message has been there over a minute
  329. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  330. return ""
  331. }
  332. return readMessage.Message // return current message if not expired
  333. }
  334. return ""
  335. }