daemon.go 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467
  1. package functions
  2. import (
  3. "context"
  4. "crypto/tls"
  5. "crypto/x509"
  6. "errors"
  7. "fmt"
  8. "io/ioutil"
  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/models"
  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. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  25. )
  26. var messageCache = new(sync.Map)
  27. var networkcontext = new(sync.Map)
  28. const lastNodeUpdate = "lnu"
  29. const lastPeerUpdate = "lpu"
  30. type cachedMessage struct {
  31. Message string
  32. LastSeen time.Time
  33. }
  34. // Daemon runs netclient daemon from command line
  35. func Daemon() error {
  36. var exists = struct{}{}
  37. serverSet := make(map[string]struct{})
  38. // == initial pull of all networks ==
  39. networks, _ := ncutils.GetSystemNetworks()
  40. for _, network := range networks {
  41. cfg := config.ClientConfig{}
  42. cfg.Network = network
  43. cfg.ReadConfig()
  44. serverSet[cfg.Server.Server] = exists
  45. //temporary code --- remove in version v0.13.0
  46. removeHostDNS(network, ncutils.IsWindows())
  47. // end of code to be removed in version v0.13.0
  48. initialPull(cfg.Network)
  49. }
  50. // == subscribe to all nodes for each on machine ==
  51. for server := range serverSet {
  52. logger.Log(1, "started daemon for server ", server)
  53. ctx, cancel := context.WithCancel(context.Background())
  54. networkcontext.Store(server, cancel)
  55. go messageQueue(ctx, server)
  56. }
  57. // == add waitgroup and cancel for checkin routine ==
  58. wg := sync.WaitGroup{}
  59. ctx, cancel := context.WithCancel(context.Background())
  60. wg.Add(1)
  61. go Checkin(ctx, &wg, serverSet)
  62. quit := make(chan os.Signal, 1)
  63. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  64. <-quit
  65. for server := range serverSet {
  66. if cancel, ok := networkcontext.Load(server); ok {
  67. cancel.(context.CancelFunc)()
  68. }
  69. }
  70. cancel()
  71. logger.Log(0, "shutting down netclient daemon")
  72. wg.Wait()
  73. logger.Log(0, "shutdown complete")
  74. return nil
  75. }
  76. // UpdateKeys -- updates private key and returns new publickey
  77. func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
  78. logger.Log(0, "received message to update wireguard keys for network ", nodeCfg.Network)
  79. key, err := wgtypes.GeneratePrivateKey()
  80. if err != nil {
  81. logger.Log(0, "error generating privatekey ", err.Error())
  82. return err
  83. }
  84. file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
  85. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  86. logger.Log(0, "error updating wireguard key ", err.Error())
  87. return err
  88. }
  89. if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
  90. logger.Log(0, "failed to save private key", storeErr.Error())
  91. return storeErr
  92. }
  93. nodeCfg.Node.PublicKey = key.PublicKey().String()
  94. PublishNodeUpdate(nodeCfg)
  95. return nil
  96. }
  97. // PingServer -- checks if server is reachable
  98. // use commsCfg only*
  99. func PingServer(cfg *config.ClientConfig) error {
  100. pinger, err := ping.NewPinger(cfg.Server.Server)
  101. if err != nil {
  102. return err
  103. }
  104. pinger.Timeout = 2 * time.Second
  105. pinger.Run()
  106. stats := pinger.Statistics()
  107. if stats.PacketLoss == 100 {
  108. return errors.New("ping error")
  109. }
  110. logger.Log(3, "ping of server", cfg.Server.Server, "was successful")
  111. return nil
  112. }
  113. // == Private ==
  114. // sets MQ client subscriptions for a specific node config
  115. // should be called for each node belonging to a given comms network
  116. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  117. if nodeCfg.DebugOn {
  118. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  119. logger.Log(0, token.Error().Error())
  120. return
  121. }
  122. logger.Log(0, "subscribed to all topics for debugging purposes")
  123. }
  124. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  125. logger.Log(0, token.Error().Error())
  126. return
  127. }
  128. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  129. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  130. logger.Log(0, token.Error().Error())
  131. return
  132. }
  133. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  134. }
  135. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  136. // for the node in nodeCfg
  137. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  138. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  139. var ok = true
  140. if token := client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  141. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  142. ok = false
  143. }
  144. if token := client.Unsubscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  145. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  146. ok = false
  147. }
  148. if ok {
  149. logger.Log(1, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  150. }
  151. }
  152. // sets up Message Queue and subsribes/publishes updates to/from server
  153. // the client should subscribe to ALL nodes that exist on server locally
  154. func messageQueue(ctx context.Context, server string) {
  155. logger.Log(0, "netclient daemon started for server: ", server)
  156. client := setupMQTTSub(server)
  157. defer client.Disconnect(250)
  158. <-ctx.Done()
  159. logger.Log(0, "shutting down daemon for server ", server)
  160. }
  161. // setupMQTTSub creates a connection to broker and subscribes to topic
  162. // utilizes comms client configs to setup connections
  163. func setupMQTTSub(server string) mqtt.Client {
  164. opts := mqtt.NewClientOptions()
  165. opts.AddBroker("ssl://" + server + ":8883") // TODO get the appropriate port of the comms mq server
  166. opts.TLSConfig = NewTLSConfig(nil, server)
  167. opts.ClientID = ncutils.MakeRandomString(23) // helps avoid id duplication on broker
  168. opts.SetDefaultPublishHandler(All)
  169. opts.SetAutoReconnect(true)
  170. opts.SetConnectRetry(true)
  171. opts.SetConnectRetryInterval(time.Second << 2)
  172. opts.SetKeepAlive(time.Minute >> 1)
  173. opts.SetWriteTimeout(time.Minute)
  174. opts.SetOnConnectHandler(func(client mqtt.Client) {
  175. networks, err := ncutils.GetSystemNetworks()
  176. if err != nil {
  177. logger.Log(0, "error retriving networks ", err.Error())
  178. }
  179. for _, network := range networks {
  180. var currNodeCfg config.ClientConfig
  181. currNodeCfg.Network = network
  182. currNodeCfg.ReadConfig()
  183. if currNodeCfg.Server.Server == server {
  184. setSubscriptions(client, &currNodeCfg)
  185. }
  186. }
  187. })
  188. opts.SetOrderMatters(true)
  189. opts.SetResumeSubs(true)
  190. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  191. logger.Log(0, "detected broker connection lost, running pull for all nodes")
  192. networks, err := ncutils.GetSystemNetworks()
  193. if err != nil {
  194. logger.Log(0, "error retriving networks ", err.Error())
  195. }
  196. for _, network := range networks {
  197. var cfg config.ClientConfig
  198. cfg.Network = network
  199. cfg.ReadConfig()
  200. _, err := Pull(cfg.Node.Network, true)
  201. if err != nil {
  202. logger.Log(0, "could not run pull, server unreachable: ", err.Error())
  203. logger.Log(0, "waiting to retry...")
  204. }
  205. }
  206. // don't think following log message is accurate
  207. //logger.Log(0, "connection re-established with mqtt server")
  208. })
  209. client := mqtt.NewClient(opts)
  210. tperiod := time.Now().Add(12 * time.Second)
  211. for {
  212. //if after 12 seconds, try a gRPC pull on the last try
  213. if time.Now().After(tperiod) {
  214. networks, err := ncutils.GetSystemNetworks()
  215. if err != nil {
  216. logger.Log(0, "error retriving networks ", err.Error())
  217. }
  218. for _, network := range networks {
  219. var cfg config.ClientConfig
  220. cfg.Network = network
  221. cfg.ReadConfig()
  222. if cfg.Server.Server == server {
  223. _, err := Pull(cfg.Node.Network, true)
  224. if err != nil {
  225. logger.Log(0, "could not run pull, exiting ", cfg.Node.Network, " setup: ", err.Error())
  226. return client
  227. }
  228. }
  229. }
  230. time.Sleep(time.Second)
  231. }
  232. if token := client.Connect(); token.Wait() && token.Error() != nil {
  233. logger.Log(0, "unable to connect to broker, retrying ...")
  234. if time.Now().After(tperiod) {
  235. logger.Log(0, "could not connect to broker, exiting ", server, " setup: ", token.Error().Error())
  236. if strings.Contains(token.Error().Error(), "connectex") || strings.Contains(token.Error().Error(), "i/o timeout") {
  237. logger.Log(0, "connection issue detected.. restarting daemon")
  238. daemon.Restart()
  239. }
  240. return client
  241. }
  242. } else {
  243. break
  244. }
  245. time.Sleep(2 * time.Second)
  246. }
  247. return client
  248. }
  249. // NewTLSConf sets up tls to connect to broker
  250. func NewTLSConfig(cfg *config.ClientConfig, server string) *tls.Config {
  251. var ca []byte
  252. var err error
  253. certpool := x509.NewCertPool()
  254. if cfg != nil {
  255. ca, err = ioutil.ReadFile("/etc/netclient/" + cfg.Server.Server + "/root.pem")
  256. if err != nil {
  257. logger.Log(0, "could not read CA file %v\n", err.Error())
  258. }
  259. } else {
  260. ca, err = ioutil.ReadFile("/etc/netclient/" + server + "/root.pem")
  261. if err != nil {
  262. logger.Log(0, "could not read CA file %v\n", err.Error())
  263. }
  264. }
  265. certpool.AppendCertsFromPEM(ca)
  266. //clientKeyPair, err := tls.LoadX509KeyPair("/etc/netclient/"+cfg.Server.Server+"/client.pem", "/etc/netclient/client.key")
  267. //if err != nil {
  268. // log.Fatalf("could not read client cert/key %v \n", err)
  269. //}
  270. return &tls.Config{
  271. RootCAs: certpool,
  272. ClientAuth: tls.NoClientCert,
  273. //ClientAuth: tls.VerifyClientCertIfGiven,
  274. ClientCAs: nil,
  275. InsecureSkipVerify: true,
  276. //Certificates: []tls.Certificate{clientKeyPair},
  277. }
  278. }
  279. // setupMQTT creates a connection to broker and return client
  280. // utilizes comms client configs to setup connections
  281. func setupMQTT(cfg *config.ClientConfig, publish bool) mqtt.Client {
  282. opts := mqtt.NewClientOptions()
  283. server := cfg.Server.Server
  284. opts.AddBroker("ssl://" + server + ":8883") // TODO get the appropriate port of the comms mq server
  285. opts.TLSConfig = NewTLSConfig(cfg, "")
  286. opts.ClientID = ncutils.MakeRandomString(23) // helps avoid id duplication on broker
  287. opts.SetDefaultPublishHandler(All)
  288. opts.SetAutoReconnect(true)
  289. opts.SetConnectRetry(true)
  290. opts.SetConnectRetryInterval(time.Second << 2)
  291. opts.SetKeepAlive(time.Minute >> 1)
  292. opts.SetWriteTimeout(time.Minute)
  293. opts.SetOnConnectHandler(func(client mqtt.Client) {
  294. if !publish {
  295. networks, err := ncutils.GetSystemNetworks()
  296. if err != nil {
  297. logger.Log(0, "error retriving networks ", err.Error())
  298. }
  299. for _, network := range networks {
  300. var currNodeCfg config.ClientConfig
  301. currNodeCfg.Network = network
  302. currNodeCfg.ReadConfig()
  303. setSubscriptions(client, &currNodeCfg)
  304. }
  305. }
  306. })
  307. opts.SetOrderMatters(true)
  308. opts.SetResumeSubs(true)
  309. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  310. logger.Log(0, "detected broker connection lost, running pull for ", cfg.Node.Network)
  311. _, err := Pull(cfg.Node.Network, true)
  312. if err != nil {
  313. logger.Log(0, "could not run pull, server unreachable: ", err.Error())
  314. logger.Log(0, "waiting to retry...")
  315. }
  316. logger.Log(0, "connection re-established with mqtt server")
  317. })
  318. client := mqtt.NewClient(opts)
  319. tperiod := time.Now().Add(12 * time.Second)
  320. for {
  321. //if after 12 seconds, try a gRPC pull on the last try
  322. if time.Now().After(tperiod) {
  323. logger.Log(0, "running pull for ", cfg.Node.Network)
  324. _, err := Pull(cfg.Node.Network, true)
  325. if err != nil {
  326. logger.Log(0, "could not run pull, exiting ", cfg.Node.Network, " setup: ", err.Error())
  327. return client
  328. }
  329. time.Sleep(time.Second)
  330. }
  331. if token := client.Connect(); token.Wait() && token.Error() != nil {
  332. logger.Log(0, "unable to connect to broker, retrying ...")
  333. if time.Now().After(tperiod) {
  334. logger.Log(0, "could not connect to broker, exiting ", cfg.Node.Network, " setup: ", token.Error().Error())
  335. if strings.Contains(token.Error().Error(), "connectex") || strings.Contains(token.Error().Error(), "i/o timeout") {
  336. logger.Log(0, "connection issue detected.. pulling and restarting daemon")
  337. Pull(cfg.Node.Network, true)
  338. daemon.Restart()
  339. }
  340. return client
  341. }
  342. } else {
  343. break
  344. }
  345. time.Sleep(2 * time.Second)
  346. }
  347. return client
  348. }
  349. // publishes a message to server to update peers on this peer's behalf
  350. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  351. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  352. return err
  353. }
  354. return nil
  355. }
  356. func initialPull(network string) {
  357. logger.Log(0, "pulling latest config for ", network)
  358. var configPath = fmt.Sprintf("%snetconfig-%s", ncutils.GetNetclientPathSpecific(), network)
  359. fileInfo, err := os.Stat(configPath)
  360. if err != nil {
  361. logger.Log(0, "could not stat config file: ", configPath)
  362. return
  363. }
  364. // speed up UDP rest
  365. if !fileInfo.ModTime().IsZero() && time.Now().After(fileInfo.ModTime().Add(time.Minute)) {
  366. sleepTime := 2
  367. for {
  368. _, err := Pull(network, true)
  369. if err == nil {
  370. break
  371. }
  372. if sleepTime > 3600 {
  373. sleepTime = 3600
  374. }
  375. logger.Log(0, "failed to pull for network ", network)
  376. logger.Log(0, fmt.Sprintf("waiting %d seconds to retry...", sleepTime))
  377. time.Sleep(time.Second * time.Duration(sleepTime))
  378. sleepTime = sleepTime * 2
  379. }
  380. time.Sleep(time.Second << 1)
  381. }
  382. }
  383. func parseNetworkFromTopic(topic string) string {
  384. return strings.Split(topic, "/")[1]
  385. }
  386. // should only ever use node client configs
  387. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  388. if len(msg) <= 24 { // make sure message is of appropriate length
  389. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  390. }
  391. // setup the keys
  392. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  393. if keyErr != nil {
  394. return nil, keyErr
  395. }
  396. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  397. if err != nil {
  398. return nil, err
  399. }
  400. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  401. }
  402. func getServerAddress(cfg *config.ClientConfig) string {
  403. var server models.ServerAddr
  404. for _, server = range cfg.Node.NetworkSettings.DefaultServerAddrs {
  405. if server.Address != "" && server.IsLeader {
  406. break
  407. }
  408. }
  409. return server.Address
  410. }
  411. // == Message Caches ==
  412. func insert(network, which, cache string) {
  413. var newMessage = cachedMessage{
  414. Message: cache,
  415. LastSeen: time.Now(),
  416. }
  417. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  418. }
  419. func read(network, which string) string {
  420. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  421. if isok {
  422. var readMessage = val.(cachedMessage) // fetch current cached message
  423. if readMessage.LastSeen.IsZero() {
  424. return ""
  425. }
  426. if time.Now().After(readMessage.LastSeen.Add(time.Minute * 10)) { // check if message has been there over a minute
  427. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  428. return ""
  429. }
  430. return readMessage.Message // return current message if not expired
  431. }
  432. return ""
  433. }
  434. // == End Message Caches ==