daemon.go 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488
  1. package functions
  2. import (
  3. "context"
  4. "crypto/ed25519"
  5. "crypto/tls"
  6. "crypto/x509"
  7. "errors"
  8. "fmt"
  9. "log"
  10. "os"
  11. "os/signal"
  12. "strings"
  13. "sync"
  14. "syscall"
  15. "time"
  16. mqtt "github.com/eclipse/paho.mqtt.golang"
  17. "github.com/go-ping/ping"
  18. "github.com/gravitl/netmaker/logger"
  19. "github.com/gravitl/netmaker/models"
  20. "github.com/gravitl/netmaker/netclient/auth"
  21. "github.com/gravitl/netmaker/netclient/config"
  22. "github.com/gravitl/netmaker/netclient/daemon"
  23. "github.com/gravitl/netmaker/netclient/ncutils"
  24. "github.com/gravitl/netmaker/netclient/wireguard"
  25. ssl "github.com/gravitl/netmaker/tls"
  26. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  27. )
  28. var messageCache = new(sync.Map)
  29. var networkcontext = new(sync.Map)
  30. const lastNodeUpdate = "lnu"
  31. const lastPeerUpdate = "lpu"
  32. type cachedMessage struct {
  33. Message string
  34. LastSeen time.Time
  35. }
  36. // Daemon runs netclient daemon from command line
  37. func Daemon() error {
  38. var exists = struct{}{}
  39. serverSet := make(map[string]struct{})
  40. // == initial pull of all networks ==
  41. networks, _ := ncutils.GetSystemNetworks()
  42. if len(networks) == 0 {
  43. return errors.New("no networks")
  44. }
  45. for _, network := range networks {
  46. logger.Log(3, "initializing network", network)
  47. cfg := config.ClientConfig{}
  48. cfg.Network = network
  49. cfg.ReadConfig()
  50. serverSet[cfg.Server.Server] = exists
  51. //temporary code --- remove in version v0.13.0
  52. removeHostDNS(network, ncutils.IsWindows())
  53. // end of code to be removed in version v0.13.0
  54. initialPull(cfg.Network)
  55. }
  56. // == subscribe to all nodes for each on machine ==
  57. for server := range serverSet {
  58. logger.Log(1, "started daemon for server ", server)
  59. ctx, cancel := context.WithCancel(context.Background())
  60. networkcontext.Store(server, cancel)
  61. go messageQueue(ctx, server)
  62. }
  63. // == add waitgroup and cancel for checkin routine ==
  64. wg := sync.WaitGroup{}
  65. ctx, cancel := context.WithCancel(context.Background())
  66. wg.Add(1)
  67. go Checkin(ctx, &wg, serverSet)
  68. quit := make(chan os.Signal, 1)
  69. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  70. <-quit
  71. for server := range serverSet {
  72. if cancel, ok := networkcontext.Load(server); ok {
  73. cancel.(context.CancelFunc)()
  74. }
  75. }
  76. cancel()
  77. logger.Log(0, "shutting down netclient daemon")
  78. wg.Wait()
  79. logger.Log(0, "shutdown complete")
  80. return nil
  81. }
  82. // UpdateKeys -- updates private key and returns new publickey
  83. func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
  84. logger.Log(0, "received message to update wireguard keys for network ", nodeCfg.Network)
  85. key, err := wgtypes.GeneratePrivateKey()
  86. if err != nil {
  87. logger.Log(0, "error generating privatekey ", err.Error())
  88. return err
  89. }
  90. file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
  91. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  92. logger.Log(0, "error updating wireguard key ", err.Error())
  93. return err
  94. }
  95. if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
  96. logger.Log(0, "failed to save private key", storeErr.Error())
  97. return storeErr
  98. }
  99. nodeCfg.Node.PublicKey = key.PublicKey().String()
  100. PublishNodeUpdate(nodeCfg)
  101. return nil
  102. }
  103. // PingServer -- checks if server is reachable
  104. // use commsCfg only*
  105. func PingServer(cfg *config.ClientConfig) error {
  106. pinger, err := ping.NewPinger(cfg.Server.Server)
  107. if err != nil {
  108. return err
  109. }
  110. pinger.Timeout = 2 * time.Second
  111. pinger.Run()
  112. stats := pinger.Statistics()
  113. if stats.PacketLoss == 100 {
  114. return errors.New("ping error")
  115. }
  116. logger.Log(3, "ping of server", cfg.Server.Server, "was successful")
  117. return nil
  118. }
  119. // == Private ==
  120. // sets MQ client subscriptions for a specific node config
  121. // should be called for each node belonging to a given server
  122. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  123. if nodeCfg.DebugOn {
  124. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  125. logger.Log(0, token.Error().Error())
  126. return
  127. }
  128. logger.Log(0, "subscribed to all topics for debugging purposes")
  129. }
  130. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  131. logger.Log(0, token.Error().Error())
  132. return
  133. }
  134. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  135. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  136. logger.Log(0, token.Error().Error())
  137. return
  138. }
  139. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  140. }
  141. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  142. // for the node in nodeCfg
  143. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  144. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  145. var ok = true
  146. if token := client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  147. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  148. ok = false
  149. }
  150. if token := client.Unsubscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  151. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  152. ok = false
  153. }
  154. if ok {
  155. logger.Log(1, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  156. }
  157. }
  158. // sets up Message Queue and subsribes/publishes updates to/from server
  159. // the client should subscribe to ALL nodes that exist on server locally
  160. func messageQueue(ctx context.Context, server string) {
  161. logger.Log(0, "netclient daemon started for server: ", server)
  162. client := setupMQTTSub(server)
  163. defer client.Disconnect(250)
  164. <-ctx.Done()
  165. logger.Log(0, "shutting down daemon for server ", server)
  166. }
  167. // setupMQTTSub creates a connection to broker and subscribes to topic
  168. // utilizes comms client configs to setup connections
  169. func setupMQTTSub(server string) mqtt.Client {
  170. opts := mqtt.NewClientOptions()
  171. opts.AddBroker("ssl://" + server + ":8883") // TODO get the appropriate port of the comms mq server
  172. opts.TLSConfig = NewTLSConfig(nil, server)
  173. opts.SetDefaultPublishHandler(All)
  174. opts.SetAutoReconnect(true)
  175. opts.SetConnectRetry(true)
  176. opts.SetConnectRetryInterval(time.Second << 2)
  177. opts.SetKeepAlive(time.Minute >> 1)
  178. opts.SetWriteTimeout(time.Minute)
  179. opts.SetOnConnectHandler(func(client mqtt.Client) {
  180. networks, err := ncutils.GetSystemNetworks()
  181. if err != nil {
  182. logger.Log(0, "error retriving networks ", err.Error())
  183. }
  184. for _, network := range networks {
  185. var currNodeCfg config.ClientConfig
  186. currNodeCfg.Network = network
  187. currNodeCfg.ReadConfig()
  188. if currNodeCfg.Server.Server == server {
  189. setSubscriptions(client, &currNodeCfg)
  190. }
  191. }
  192. })
  193. opts.SetOrderMatters(true)
  194. opts.SetResumeSubs(true)
  195. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  196. logger.Log(0, "detected broker connection lost, running pull for all nodes")
  197. networks, err := ncutils.GetSystemNetworks()
  198. if err != nil {
  199. logger.Log(0, "error retriving networks ", err.Error())
  200. }
  201. for _, network := range networks {
  202. var cfg config.ClientConfig
  203. cfg.Network = network
  204. cfg.ReadConfig()
  205. _, err := Pull(cfg.Node.Network, true)
  206. if err != nil {
  207. logger.Log(0, "could not run pull, server unreachable: ", err.Error())
  208. logger.Log(0, "waiting to retry...")
  209. }
  210. }
  211. // don't think following log message is accurate
  212. //logger.Log(0, "connection re-established with mqtt server")
  213. })
  214. client := mqtt.NewClient(opts)
  215. tperiod := time.Now().Add(12 * time.Second)
  216. for {
  217. //if after 12 seconds, try a gRPC pull on the last try
  218. if time.Now().After(tperiod) {
  219. networks, err := ncutils.GetSystemNetworks()
  220. if err != nil {
  221. logger.Log(0, "error retriving networks ", err.Error())
  222. }
  223. for _, network := range networks {
  224. var cfg config.ClientConfig
  225. cfg.Network = network
  226. cfg.ReadConfig()
  227. if cfg.Server.Server == server {
  228. _, err := Pull(cfg.Node.Network, true)
  229. if err != nil {
  230. logger.Log(0, "could not run pull, exiting ", cfg.Node.Network, " setup: ", err.Error())
  231. return client
  232. }
  233. }
  234. }
  235. time.Sleep(time.Second)
  236. }
  237. if token := client.Connect(); token.Wait() && token.Error() != nil {
  238. logger.Log(0, "unable to connect to broker, retrying ...")
  239. if time.Now().After(tperiod) {
  240. logger.Log(0, "could not connect to broker, exiting ", server, " setup: ", token.Error().Error())
  241. if strings.Contains(token.Error().Error(), "connectex") || strings.Contains(token.Error().Error(), "i/o timeout") {
  242. logger.Log(0, "connection issue detected.. restarting daemon")
  243. daemon.Restart()
  244. }
  245. return client
  246. }
  247. } else {
  248. break
  249. }
  250. time.Sleep(2 * time.Second)
  251. }
  252. return client
  253. }
  254. // NewTLSConf sets up tls configuration to connect to broker securely
  255. func NewTLSConfig(cfg *config.ClientConfig, server string) *tls.Config {
  256. var file string
  257. if cfg != nil {
  258. server = cfg.Server.Server
  259. }
  260. file = ncutils.GetNetclientServerPath(server) + "/root.pem"
  261. certpool := x509.NewCertPool()
  262. ca, err := os.ReadFile(file)
  263. if err != nil {
  264. logger.Log(0, "could not read CA file ", err.Error())
  265. }
  266. ok := certpool.AppendCertsFromPEM(ca)
  267. if !ok {
  268. logger.Log(0, "failed to append cert")
  269. }
  270. clientKeyPair, err := tls.LoadX509KeyPair(ncutils.GetNetclientServerPath(server)+"/client.pem", ncutils.GetNetclientPath()+"/client.key")
  271. if err != nil {
  272. log.Fatalf("could not read client cert/key %v \n", err)
  273. }
  274. certs := []tls.Certificate{clientKeyPair}
  275. return &tls.Config{
  276. RootCAs: certpool,
  277. ClientAuth: tls.NoClientCert,
  278. ClientCAs: nil,
  279. Certificates: certs,
  280. //InsecureSkipVerify: false fails ---- so need to use VerifyConnection
  281. InsecureSkipVerify: true,
  282. VerifyConnection: func(cs tls.ConnectionState) error {
  283. if cs.ServerName != server {
  284. logger.Log(0, "VerifyConnection - certifiate mismatch")
  285. return errors.New("certificate doesn't match server")
  286. }
  287. ca, err := ssl.ReadCert(ncutils.GetNetclientServerPath(cs.ServerName) + "/root.pem")
  288. if err != nil {
  289. logger.Log(0, "VerifyConnection - unable to read ca", err.Error())
  290. return errors.New("unable to read ca")
  291. }
  292. for _, cert := range cs.PeerCertificates {
  293. if cert.IsCA {
  294. if string(cert.PublicKey.(ed25519.PublicKey)) != string(ca.PublicKey.(ed25519.PublicKey)) {
  295. logger.Log(0, "VerifyConnection - public key mismatch")
  296. return errors.New("cert public key does not match ca public key")
  297. }
  298. }
  299. }
  300. return nil
  301. },
  302. }
  303. }
  304. // setupMQTT creates a connection to broker and return client
  305. // this function is primarily used to create a connection to publish to the broker
  306. func setupMQTT(cfg *config.ClientConfig, publish bool) mqtt.Client {
  307. opts := mqtt.NewClientOptions()
  308. server := cfg.Server.Server
  309. opts.AddBroker("ssl://" + server + ":8883") // TODO get the appropriate port of the comms mq server
  310. opts.TLSConfig = NewTLSConfig(cfg, "")
  311. opts.SetDefaultPublishHandler(All)
  312. opts.SetAutoReconnect(true)
  313. opts.SetConnectRetry(true)
  314. opts.SetConnectRetryInterval(time.Second << 2)
  315. opts.SetKeepAlive(time.Minute >> 1)
  316. opts.SetWriteTimeout(time.Minute)
  317. opts.SetOnConnectHandler(func(client mqtt.Client) {
  318. if !publish {
  319. networks, err := ncutils.GetSystemNetworks()
  320. if err != nil {
  321. logger.Log(0, "error retriving networks ", err.Error())
  322. }
  323. for _, network := range networks {
  324. var currNodeCfg config.ClientConfig
  325. currNodeCfg.Network = network
  326. currNodeCfg.ReadConfig()
  327. setSubscriptions(client, &currNodeCfg)
  328. }
  329. }
  330. })
  331. opts.SetOrderMatters(true)
  332. opts.SetResumeSubs(true)
  333. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  334. logger.Log(0, "detected broker connection lost, running pull for ", cfg.Node.Network)
  335. _, err := Pull(cfg.Node.Network, true)
  336. if err != nil {
  337. logger.Log(0, "could not run pull, server unreachable: ", err.Error())
  338. logger.Log(0, "waiting to retry...")
  339. }
  340. logger.Log(0, "connection re-established with mqtt server")
  341. })
  342. client := mqtt.NewClient(opts)
  343. tperiod := time.Now().Add(12 * time.Second)
  344. for {
  345. //if after 12 seconds, try a gRPC pull on the last try
  346. if time.Now().After(tperiod) {
  347. logger.Log(0, "running pull for ", cfg.Node.Network)
  348. _, err := Pull(cfg.Node.Network, true)
  349. if err != nil {
  350. logger.Log(0, "could not run pull, exiting ", cfg.Node.Network, " setup: ", err.Error())
  351. return client
  352. }
  353. time.Sleep(time.Second)
  354. }
  355. if token := client.Connect(); token.Wait() && token.Error() != nil {
  356. logger.Log(0, "unable to connect to broker, retrying ...")
  357. if time.Now().After(tperiod) {
  358. logger.Log(0, "could not connect to broker, exiting ", cfg.Node.Network, " setup: ", token.Error().Error())
  359. if strings.Contains(token.Error().Error(), "connectex") || strings.Contains(token.Error().Error(), "i/o timeout") {
  360. logger.Log(0, "connection issue detected.. pulling and restarting daemon")
  361. Pull(cfg.Node.Network, true)
  362. daemon.Restart()
  363. }
  364. return client
  365. }
  366. } else {
  367. break
  368. }
  369. time.Sleep(2 * time.Second)
  370. }
  371. return client
  372. }
  373. // publishes a message to server to update peers on this peer's behalf
  374. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  375. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  376. return err
  377. }
  378. return nil
  379. }
  380. func initialPull(network string) {
  381. logger.Log(0, "pulling latest config for ", network)
  382. var configPath = fmt.Sprintf("%snetconfig-%s", ncutils.GetNetclientPathSpecific(), network)
  383. fileInfo, err := os.Stat(configPath)
  384. if err != nil {
  385. logger.Log(0, "could not stat config file: ", configPath)
  386. return
  387. }
  388. // speed up UDP rest
  389. if !fileInfo.ModTime().IsZero() && time.Now().After(fileInfo.ModTime().Add(time.Minute)) {
  390. sleepTime := 2
  391. for {
  392. _, err := Pull(network, true)
  393. if err == nil {
  394. break
  395. }
  396. if sleepTime > 3600 {
  397. sleepTime = 3600
  398. }
  399. logger.Log(0, "failed to pull for network ", network)
  400. logger.Log(0, fmt.Sprintf("waiting %d seconds to retry...", sleepTime))
  401. time.Sleep(time.Second * time.Duration(sleepTime))
  402. sleepTime = sleepTime * 2
  403. }
  404. time.Sleep(time.Second << 1)
  405. }
  406. }
  407. func parseNetworkFromTopic(topic string) string {
  408. return strings.Split(topic, "/")[1]
  409. }
  410. // should only ever use node client configs
  411. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  412. if len(msg) <= 24 { // make sure message is of appropriate length
  413. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  414. }
  415. // setup the keys
  416. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  417. if keyErr != nil {
  418. return nil, keyErr
  419. }
  420. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  421. if err != nil {
  422. return nil, err
  423. }
  424. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  425. }
  426. func getServerAddress(cfg *config.ClientConfig) string {
  427. var server models.ServerAddr
  428. for _, server = range cfg.Node.NetworkSettings.DefaultServerAddrs {
  429. if server.Address != "" && server.IsLeader {
  430. break
  431. }
  432. }
  433. return server.Address
  434. }
  435. // == Message Caches ==
  436. func insert(network, which, cache string) {
  437. var newMessage = cachedMessage{
  438. Message: cache,
  439. LastSeen: time.Now(),
  440. }
  441. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  442. }
  443. func read(network, which string) string {
  444. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  445. if isok {
  446. var readMessage = val.(cachedMessage) // fetch current cached message
  447. if readMessage.LastSeen.IsZero() {
  448. return ""
  449. }
  450. if time.Now().After(readMessage.LastSeen.Add(time.Minute * 10)) { // check if message has been there over a minute
  451. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  452. return ""
  453. }
  454. return readMessage.Message // return current message if not expired
  455. }
  456. return ""
  457. }