daemon.go 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475
  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/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. var exists = struct{}{}
  38. serverSet := make(map[string]struct{})
  39. // == initial pull of all networks ==
  40. networks, _ := ncutils.GetSystemNetworks()
  41. if len(networks) == 0 {
  42. return errors.New("no networks")
  43. }
  44. for _, network := range networks {
  45. logger.Log(3, "initializing network", network)
  46. cfg := config.ClientConfig{}
  47. cfg.Network = network
  48. cfg.ReadConfig()
  49. serverSet[cfg.Server.Server] = exists
  50. //temporary code --- remove in version v0.13.0
  51. removeHostDNS(network, ncutils.IsWindows())
  52. // end of code to be removed in version v0.13.0
  53. initialPull(cfg.Network)
  54. }
  55. // == subscribe to all nodes for each on machine ==
  56. for server := range serverSet {
  57. logger.Log(1, "started daemon for server ", server)
  58. ctx, cancel := context.WithCancel(context.Background())
  59. networkcontext.Store(server, cancel)
  60. go messageQueue(ctx, server)
  61. }
  62. // == add waitgroup and cancel for checkin routine ==
  63. wg := sync.WaitGroup{}
  64. ctx, cancel := context.WithCancel(context.Background())
  65. wg.Add(1)
  66. go Checkin(ctx, &wg)
  67. quit := make(chan os.Signal, 1)
  68. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  69. <-quit
  70. for server := range serverSet {
  71. if cancel, ok := networkcontext.Load(server); ok {
  72. cancel.(context.CancelFunc)()
  73. }
  74. }
  75. cancel()
  76. logger.Log(0, "shutting down netclient daemon")
  77. wg.Wait()
  78. logger.Log(0, "shutdown complete")
  79. return nil
  80. }
  81. // UpdateKeys -- updates private key and returns new publickey
  82. func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
  83. logger.Log(0, "received message to update wireguard keys for network ", nodeCfg.Network)
  84. key, err := wgtypes.GeneratePrivateKey()
  85. if err != nil {
  86. logger.Log(0, "error generating privatekey ", err.Error())
  87. return err
  88. }
  89. file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
  90. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  91. logger.Log(0, "error updating wireguard key ", err.Error())
  92. return err
  93. }
  94. if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
  95. logger.Log(0, "failed to save private key", storeErr.Error())
  96. return storeErr
  97. }
  98. nodeCfg.Node.PublicKey = key.PublicKey().String()
  99. PublishNodeUpdate(nodeCfg)
  100. return nil
  101. }
  102. // PingServer -- checks if server is reachable
  103. func PingServer(cfg *config.ClientConfig) error {
  104. pinger, err := ping.NewPinger(cfg.Server.Server)
  105. if err != nil {
  106. return err
  107. }
  108. pinger.Timeout = 2 * time.Second
  109. pinger.Run()
  110. stats := pinger.Statistics()
  111. if stats.PacketLoss == 100 {
  112. return errors.New("ping error")
  113. }
  114. logger.Log(3, "ping of server", cfg.Server.Server, "was successful")
  115. return nil
  116. }
  117. // == Private ==
  118. // sets MQ client subscriptions for a specific node config
  119. // should be called for each node belonging to a given server
  120. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  121. if nodeCfg.DebugOn {
  122. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  123. logger.Log(0, token.Error().Error())
  124. return
  125. }
  126. logger.Log(0, "subscribed to all topics for debugging purposes")
  127. }
  128. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  129. logger.Log(0, token.Error().Error())
  130. return
  131. }
  132. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  133. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  134. logger.Log(0, token.Error().Error())
  135. return
  136. }
  137. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  138. }
  139. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  140. // for the node in nodeCfg
  141. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  142. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  143. var ok = true
  144. if token := client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  145. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  146. ok = false
  147. }
  148. if token := client.Unsubscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  149. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  150. ok = false
  151. }
  152. if ok {
  153. logger.Log(1, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  154. }
  155. }
  156. // sets up Message Queue and subsribes/publishes updates to/from server
  157. // the client should subscribe to ALL nodes that exist on server locally
  158. func messageQueue(ctx context.Context, server string) {
  159. logger.Log(0, "netclient daemon started for server: ", server)
  160. client := setupMQTTSub(server)
  161. defer client.Disconnect(250)
  162. <-ctx.Done()
  163. logger.Log(0, "shutting down daemon for server ", server)
  164. }
  165. // setupMQTTSub creates a connection to broker and subscribes to topic
  166. func setupMQTTSub(server string) mqtt.Client {
  167. opts := mqtt.NewClientOptions()
  168. opts.AddBroker("ssl://" + server + ":8883")
  169. opts.TLSConfig = NewTLSConfig(nil, server)
  170. opts.SetDefaultPublishHandler(All)
  171. opts.SetAutoReconnect(true)
  172. opts.SetConnectRetry(true)
  173. opts.SetConnectRetryInterval(time.Second << 2)
  174. opts.SetKeepAlive(time.Minute >> 1)
  175. opts.SetWriteTimeout(time.Minute)
  176. opts.SetOnConnectHandler(func(client mqtt.Client) {
  177. networks, err := ncutils.GetSystemNetworks()
  178. if err != nil {
  179. logger.Log(0, "error retriving networks ", err.Error())
  180. }
  181. for _, network := range networks {
  182. var currNodeCfg config.ClientConfig
  183. currNodeCfg.Network = network
  184. currNodeCfg.ReadConfig()
  185. if currNodeCfg.Server.Server == server {
  186. setSubscriptions(client, &currNodeCfg)
  187. }
  188. }
  189. })
  190. opts.SetOrderMatters(true)
  191. opts.SetResumeSubs(true)
  192. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  193. logger.Log(0, "detected broker connection lost, running pull for all nodes")
  194. networks, err := ncutils.GetSystemNetworks()
  195. if err != nil {
  196. logger.Log(0, "error retriving networks ", err.Error())
  197. }
  198. for _, network := range networks {
  199. var cfg config.ClientConfig
  200. cfg.Network = network
  201. cfg.ReadConfig()
  202. _, err := Pull(cfg.Node.Network, true)
  203. if err != nil {
  204. logger.Log(0, "could not run pull, server unreachable: ", err.Error())
  205. logger.Log(0, "waiting to retry...")
  206. }
  207. }
  208. // don't think following log message is accurate
  209. //logger.Log(0, "connection re-established with mqtt server")
  210. })
  211. client := mqtt.NewClient(opts)
  212. tperiod := time.Now().Add(12 * time.Second)
  213. for {
  214. //if after 12 seconds, try a pull on the last try
  215. if time.Now().After(tperiod) {
  216. networks, err := ncutils.GetSystemNetworks()
  217. if err != nil {
  218. logger.Log(0, "error retriving networks ", err.Error())
  219. }
  220. for _, network := range networks {
  221. var cfg config.ClientConfig
  222. cfg.Network = network
  223. cfg.ReadConfig()
  224. if cfg.Server.Server == server {
  225. _, err := Pull(cfg.Node.Network, true)
  226. if err != nil {
  227. logger.Log(0, "could not run pull, exiting ", cfg.Node.Network, " setup: ", err.Error())
  228. return client
  229. }
  230. }
  231. }
  232. time.Sleep(time.Second)
  233. }
  234. if token := client.Connect(); token.Wait() && token.Error() != nil {
  235. logger.Log(0, "unable to connect to broker, retrying ...")
  236. if time.Now().After(tperiod) {
  237. logger.Log(0, "could not connect to broker, exiting ", server, " setup: ", token.Error().Error())
  238. if strings.Contains(token.Error().Error(), "connectex") || strings.Contains(token.Error().Error(), "i/o timeout") {
  239. logger.Log(0, "connection issue detected.. restarting daemon")
  240. daemon.Restart()
  241. }
  242. return client
  243. }
  244. } else {
  245. break
  246. }
  247. time.Sleep(2 * time.Second)
  248. }
  249. return client
  250. }
  251. // NewTLSConf sets up tls configuration to connect to broker securely
  252. func NewTLSConfig(cfg *config.ClientConfig, server string) *tls.Config {
  253. var file string
  254. if cfg != nil {
  255. server = cfg.Server.Server
  256. }
  257. file = ncutils.GetNetclientServerPath(server) + "/root.pem"
  258. certpool := x509.NewCertPool()
  259. ca, err := os.ReadFile(file)
  260. if err != nil {
  261. logger.Log(0, "could not read CA file ", err.Error())
  262. }
  263. ok := certpool.AppendCertsFromPEM(ca)
  264. if !ok {
  265. logger.Log(0, "failed to append cert")
  266. }
  267. clientKeyPair, err := tls.LoadX509KeyPair(ncutils.GetNetclientServerPath(server)+"/client.pem", ncutils.GetNetclientPath()+"/client.key")
  268. if err != nil {
  269. log.Fatalf("could not read client cert/key %v \n", err)
  270. }
  271. certs := []tls.Certificate{clientKeyPair}
  272. return &tls.Config{
  273. RootCAs: certpool,
  274. ClientAuth: tls.NoClientCert,
  275. ClientCAs: nil,
  276. Certificates: certs,
  277. //InsecureSkipVerify: false fails ---- so need to use VerifyConnection
  278. InsecureSkipVerify: true,
  279. VerifyConnection: func(cs tls.ConnectionState) error {
  280. if cs.ServerName != server {
  281. logger.Log(0, "VerifyConnection - certifiate mismatch")
  282. return errors.New("certificate doesn't match server")
  283. }
  284. ca, err := ssl.ReadCert(ncutils.GetNetclientServerPath(cs.ServerName) + "/root.pem")
  285. if err != nil {
  286. logger.Log(0, "VerifyConnection - unable to read ca", err.Error())
  287. return errors.New("unable to read ca")
  288. }
  289. for _, cert := range cs.PeerCertificates {
  290. if cert.IsCA {
  291. if string(cert.PublicKey.(ed25519.PublicKey)) != string(ca.PublicKey.(ed25519.PublicKey)) {
  292. logger.Log(0, "VerifyConnection - public key mismatch")
  293. return errors.New("cert public key does not match ca public key")
  294. }
  295. }
  296. }
  297. return nil
  298. },
  299. }
  300. }
  301. // setupMQTT creates a connection to broker and return client
  302. // this function is primarily used to create a connection to publish to the broker
  303. func setupMQTT(cfg *config.ClientConfig, publish bool) mqtt.Client {
  304. opts := mqtt.NewClientOptions()
  305. server := cfg.Server.Server
  306. opts.AddBroker("ssl://" + server + ":8883")
  307. opts.TLSConfig = NewTLSConfig(cfg, "")
  308. opts.SetDefaultPublishHandler(All)
  309. opts.SetAutoReconnect(true)
  310. opts.SetConnectRetry(true)
  311. opts.SetConnectRetryInterval(time.Second << 2)
  312. opts.SetKeepAlive(time.Minute >> 1)
  313. opts.SetWriteTimeout(time.Minute)
  314. opts.SetOnConnectHandler(func(client mqtt.Client) {
  315. if !publish {
  316. networks, err := ncutils.GetSystemNetworks()
  317. if err != nil {
  318. logger.Log(0, "error retriving networks ", err.Error())
  319. }
  320. for _, network := range networks {
  321. var currNodeCfg config.ClientConfig
  322. currNodeCfg.Network = network
  323. currNodeCfg.ReadConfig()
  324. setSubscriptions(client, &currNodeCfg)
  325. }
  326. }
  327. })
  328. opts.SetOrderMatters(true)
  329. opts.SetResumeSubs(true)
  330. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  331. logger.Log(0, "detected broker connection lost, running pull for ", cfg.Node.Network)
  332. _, err := Pull(cfg.Node.Network, true)
  333. if err != nil {
  334. logger.Log(0, "could not run pull, server unreachable: ", err.Error())
  335. logger.Log(0, "waiting to retry...")
  336. }
  337. logger.Log(0, "connection re-established with mqtt server")
  338. })
  339. client := mqtt.NewClient(opts)
  340. tperiod := time.Now().Add(12 * time.Second)
  341. for {
  342. //if after 12 seconds, try a pull on the last try
  343. if time.Now().After(tperiod) {
  344. logger.Log(0, "running pull for ", cfg.Node.Network)
  345. _, err := Pull(cfg.Node.Network, true)
  346. if err != nil {
  347. logger.Log(0, "could not run pull, exiting ", cfg.Node.Network, " setup: ", err.Error())
  348. return client
  349. }
  350. time.Sleep(time.Second)
  351. }
  352. if token := client.Connect(); token.Wait() && token.Error() != nil {
  353. logger.Log(0, "unable to connect to broker, retrying ...")
  354. if time.Now().After(tperiod) {
  355. logger.Log(0, "could not connect to broker, exiting ", cfg.Node.Network, " setup: ", token.Error().Error())
  356. if strings.Contains(token.Error().Error(), "connectex") || strings.Contains(token.Error().Error(), "i/o timeout") {
  357. logger.Log(0, "connection issue detected.. pulling and restarting daemon")
  358. Pull(cfg.Node.Network, true)
  359. daemon.Restart()
  360. }
  361. return client
  362. }
  363. } else {
  364. break
  365. }
  366. time.Sleep(2 * time.Second)
  367. }
  368. return client
  369. }
  370. // publishes a message to server to update peers on this peer's behalf
  371. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  372. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  373. return err
  374. }
  375. return nil
  376. }
  377. func initialPull(network string) {
  378. logger.Log(0, "pulling latest config for ", network)
  379. var configPath = fmt.Sprintf("%snetconfig-%s", ncutils.GetNetclientPathSpecific(), network)
  380. fileInfo, err := os.Stat(configPath)
  381. if err != nil {
  382. logger.Log(0, "could not stat config file: ", configPath)
  383. return
  384. }
  385. // speed up UDP rest
  386. if !fileInfo.ModTime().IsZero() && time.Now().After(fileInfo.ModTime().Add(time.Minute)) {
  387. sleepTime := 2
  388. for {
  389. _, err := Pull(network, true)
  390. if err == nil {
  391. break
  392. }
  393. if sleepTime > 3600 {
  394. sleepTime = 3600
  395. }
  396. logger.Log(0, "failed to pull for network ", network)
  397. logger.Log(0, fmt.Sprintf("waiting %d seconds to retry...", sleepTime))
  398. time.Sleep(time.Second * time.Duration(sleepTime))
  399. sleepTime = sleepTime * 2
  400. }
  401. time.Sleep(time.Second << 1)
  402. }
  403. }
  404. func parseNetworkFromTopic(topic string) string {
  405. return strings.Split(topic, "/")[1]
  406. }
  407. // should only ever use node client configs
  408. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  409. if len(msg) <= 24 { // make sure message is of appropriate length
  410. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  411. }
  412. // setup the keys
  413. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  414. if keyErr != nil {
  415. return nil, keyErr
  416. }
  417. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  418. if err != nil {
  419. return nil, err
  420. }
  421. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  422. }
  423. // == Message Caches ==
  424. func insert(network, which, cache string) {
  425. var newMessage = cachedMessage{
  426. Message: cache,
  427. LastSeen: time.Now(),
  428. }
  429. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  430. }
  431. func read(network, which string) string {
  432. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  433. if isok {
  434. var readMessage = val.(cachedMessage) // fetch current cached message
  435. if readMessage.LastSeen.IsZero() {
  436. return ""
  437. }
  438. if time.Now().After(readMessage.LastSeen.Add(time.Minute * 10)) { // check if message has been there over a minute
  439. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  440. return ""
  441. }
  442. return readMessage.Message // return current message if not expired
  443. }
  444. return ""
  445. }