daemon.go 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497
  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 comms network
  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.ClientID = ncutils.MakeRandomString(23) // helps avoid id duplication on broker
  174. opts.SetDefaultPublishHandler(All)
  175. opts.SetAutoReconnect(true)
  176. opts.SetConnectRetry(true)
  177. opts.SetConnectRetryInterval(time.Second << 2)
  178. opts.SetKeepAlive(time.Minute >> 1)
  179. opts.SetWriteTimeout(time.Minute)
  180. opts.SetOnConnectHandler(func(client mqtt.Client) {
  181. networks, err := ncutils.GetSystemNetworks()
  182. if err != nil {
  183. logger.Log(0, "error retriving networks ", err.Error())
  184. }
  185. for _, network := range networks {
  186. var currNodeCfg config.ClientConfig
  187. currNodeCfg.Network = network
  188. currNodeCfg.ReadConfig()
  189. if currNodeCfg.Server.Server == server {
  190. setSubscriptions(client, &currNodeCfg)
  191. }
  192. }
  193. })
  194. opts.SetOrderMatters(true)
  195. opts.SetResumeSubs(true)
  196. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  197. logger.Log(0, "detected broker connection lost, running pull for all nodes")
  198. networks, err := ncutils.GetSystemNetworks()
  199. if err != nil {
  200. logger.Log(0, "error retriving networks ", err.Error())
  201. }
  202. for _, network := range networks {
  203. var cfg config.ClientConfig
  204. cfg.Network = network
  205. cfg.ReadConfig()
  206. _, err := Pull(cfg.Node.Network, true)
  207. if err != nil {
  208. logger.Log(0, "could not run pull, server unreachable: ", err.Error())
  209. logger.Log(0, "waiting to retry...")
  210. }
  211. }
  212. // don't think following log message is accurate
  213. //logger.Log(0, "connection re-established with mqtt server")
  214. })
  215. client := mqtt.NewClient(opts)
  216. tperiod := time.Now().Add(12 * time.Second)
  217. for {
  218. //if after 12 seconds, try a gRPC pull on the last try
  219. if time.Now().After(tperiod) {
  220. networks, err := ncutils.GetSystemNetworks()
  221. if err != nil {
  222. logger.Log(0, "error retriving networks ", err.Error())
  223. }
  224. for _, network := range networks {
  225. var cfg config.ClientConfig
  226. cfg.Network = network
  227. cfg.ReadConfig()
  228. if cfg.Server.Server == server {
  229. _, err := Pull(cfg.Node.Network, true)
  230. if err != nil {
  231. logger.Log(0, "could not run pull, exiting ", cfg.Node.Network, " setup: ", err.Error())
  232. return client
  233. }
  234. }
  235. }
  236. time.Sleep(time.Second)
  237. }
  238. if token := client.Connect(); token.Wait() && token.Error() != nil {
  239. logger.Log(0, "unable to connect to broker, retrying ...")
  240. if time.Now().After(tperiod) {
  241. logger.Log(0, "could not connect to broker, exiting ", server, " setup: ", token.Error().Error())
  242. if strings.Contains(token.Error().Error(), "connectex") || strings.Contains(token.Error().Error(), "i/o timeout") {
  243. logger.Log(0, "connection issue detected.. restarting daemon")
  244. daemon.Restart()
  245. }
  246. return client
  247. }
  248. } else {
  249. break
  250. }
  251. time.Sleep(2 * time.Second)
  252. }
  253. return client
  254. }
  255. // NewTLSConf sets up tls to connect to broker
  256. func NewTLSConfig(cfg *config.ClientConfig, server string) *tls.Config {
  257. var file string
  258. if cfg != nil {
  259. server = cfg.Server.Server
  260. }
  261. file = "/etc/netclient/" + server + "/server.pem"
  262. certpool := x509.NewCertPool()
  263. ca, err := os.ReadFile(file)
  264. if err != nil {
  265. logger.Log(0, "could not read CA file ", err.Error())
  266. }
  267. ok := certpool.AppendCertsFromPEM(ca)
  268. if !ok {
  269. logger.Log(0, "failed to append cert")
  270. }
  271. //mycert, err := ssl.ReadCert("/etc/netclient/" + server + "/client.pem")
  272. clientKeyPair, err := tls.LoadX509KeyPair("/etc/netclient/"+server+"/client.pem", "/etc/netclient/client.key")
  273. //clientKeyPair, err := tls.LoadX509KeyPair("/home/mkasun/tmp/client.pem", "/home/mkasun/tmp/client.key")
  274. if err != nil {
  275. log.Fatalf("could not read client cert/key %v \n", err)
  276. }
  277. certs := []tls.Certificate{clientKeyPair}
  278. //certs = append(certs, tls.Certificate(&mycert))
  279. return &tls.Config{
  280. RootCAs: certpool,
  281. ClientAuth: tls.NoClientCert,
  282. //ClientAuth: tls.VerifyClientCertIfGiven,
  283. ClientCAs: nil,
  284. Certificates: certs,
  285. //InsecureSkipVerify: false fails ---- so need to use VerifyConnection
  286. InsecureSkipVerify: true,
  287. VerifyConnection: func(cs tls.ConnectionState) error {
  288. if cs.ServerName != server {
  289. logger.Log(0, "VerifyConnection - certifiate mismatch")
  290. return errors.New("certificate doesn't match server")
  291. }
  292. ca, err := ssl.ReadCert("/etc/netclient/" + cs.ServerName + "/root.pem")
  293. if err != nil {
  294. logger.Log(0, "VerifyConnection - unable to read ca", err.Error())
  295. return errors.New("unable to read ca")
  296. }
  297. for _, cert := range cs.PeerCertificates {
  298. if cert.IsCA {
  299. if string(cert.PublicKey.(ed25519.PublicKey)) != string(ca.PublicKey.(ed25519.PublicKey)) {
  300. logger.Log(0, "VerifyConnection - public key mismatch")
  301. return errors.New("cert public key does not match ca public key")
  302. }
  303. }
  304. }
  305. return nil
  306. },
  307. }
  308. }
  309. // setupMQTT creates a connection to broker and return client
  310. // utilizes comms client configs to setup connections
  311. func setupMQTT(cfg *config.ClientConfig, publish bool) mqtt.Client {
  312. opts := mqtt.NewClientOptions()
  313. server := cfg.Server.Server
  314. opts.AddBroker("ssl://" + server + ":8883") // TODO get the appropriate port of the comms mq server
  315. opts.TLSConfig = NewTLSConfig(cfg, "")
  316. opts.ClientID = ncutils.MakeRandomString(23) // helps avoid id duplication on broker
  317. opts.SetDefaultPublishHandler(All)
  318. opts.SetAutoReconnect(true)
  319. opts.SetConnectRetry(true)
  320. opts.SetConnectRetryInterval(time.Second << 2)
  321. opts.SetKeepAlive(time.Minute >> 1)
  322. opts.SetWriteTimeout(time.Minute)
  323. opts.SetOnConnectHandler(func(client mqtt.Client) {
  324. if !publish {
  325. networks, err := ncutils.GetSystemNetworks()
  326. if err != nil {
  327. logger.Log(0, "error retriving networks ", err.Error())
  328. }
  329. for _, network := range networks {
  330. var currNodeCfg config.ClientConfig
  331. currNodeCfg.Network = network
  332. currNodeCfg.ReadConfig()
  333. setSubscriptions(client, &currNodeCfg)
  334. }
  335. }
  336. })
  337. opts.SetOrderMatters(true)
  338. opts.SetResumeSubs(true)
  339. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  340. logger.Log(0, "detected broker connection lost, running pull for ", cfg.Node.Network)
  341. _, err := Pull(cfg.Node.Network, true)
  342. if err != nil {
  343. logger.Log(0, "could not run pull, server unreachable: ", err.Error())
  344. logger.Log(0, "waiting to retry...")
  345. }
  346. logger.Log(0, "connection re-established with mqtt server")
  347. })
  348. client := mqtt.NewClient(opts)
  349. tperiod := time.Now().Add(12 * time.Second)
  350. for {
  351. //if after 12 seconds, try a gRPC pull on the last try
  352. if time.Now().After(tperiod) {
  353. logger.Log(0, "running pull for ", cfg.Node.Network)
  354. _, err := Pull(cfg.Node.Network, true)
  355. if err != nil {
  356. logger.Log(0, "could not run pull, exiting ", cfg.Node.Network, " setup: ", err.Error())
  357. return client
  358. }
  359. time.Sleep(time.Second)
  360. }
  361. if token := client.Connect(); token.Wait() && token.Error() != nil {
  362. logger.Log(0, "unable to connect to broker, retrying ...")
  363. if time.Now().After(tperiod) {
  364. logger.Log(0, "could not connect to broker, exiting ", cfg.Node.Network, " setup: ", token.Error().Error())
  365. if strings.Contains(token.Error().Error(), "connectex") || strings.Contains(token.Error().Error(), "i/o timeout") {
  366. logger.Log(0, "connection issue detected.. pulling and restarting daemon")
  367. Pull(cfg.Node.Network, true)
  368. daemon.Restart()
  369. }
  370. return client
  371. }
  372. } else {
  373. break
  374. }
  375. time.Sleep(2 * time.Second)
  376. }
  377. return client
  378. }
  379. // publishes a message to server to update peers on this peer's behalf
  380. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  381. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  382. return err
  383. }
  384. return nil
  385. }
  386. func initialPull(network string) {
  387. logger.Log(0, "pulling latest config for ", network)
  388. var configPath = fmt.Sprintf("%snetconfig-%s", ncutils.GetNetclientPathSpecific(), network)
  389. fileInfo, err := os.Stat(configPath)
  390. if err != nil {
  391. logger.Log(0, "could not stat config file: ", configPath)
  392. return
  393. }
  394. // speed up UDP rest
  395. if !fileInfo.ModTime().IsZero() && time.Now().After(fileInfo.ModTime().Add(time.Minute)) {
  396. sleepTime := 2
  397. for {
  398. _, err := Pull(network, true)
  399. if err == nil {
  400. break
  401. }
  402. if sleepTime > 3600 {
  403. sleepTime = 3600
  404. }
  405. logger.Log(0, "failed to pull for network ", network)
  406. logger.Log(0, fmt.Sprintf("waiting %d seconds to retry...", sleepTime))
  407. time.Sleep(time.Second * time.Duration(sleepTime))
  408. sleepTime = sleepTime * 2
  409. }
  410. time.Sleep(time.Second << 1)
  411. }
  412. }
  413. func parseNetworkFromTopic(topic string) string {
  414. return strings.Split(topic, "/")[1]
  415. }
  416. // should only ever use node client configs
  417. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  418. if len(msg) <= 24 { // make sure message is of appropriate length
  419. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  420. }
  421. // setup the keys
  422. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  423. if keyErr != nil {
  424. return nil, keyErr
  425. }
  426. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  427. if err != nil {
  428. return nil, err
  429. }
  430. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  431. }
  432. func getServerAddress(cfg *config.ClientConfig) string {
  433. var server models.ServerAddr
  434. for _, server = range cfg.Node.NetworkSettings.DefaultServerAddrs {
  435. if server.Address != "" && server.IsLeader {
  436. break
  437. }
  438. }
  439. return server.Address
  440. }
  441. // == Message Caches ==
  442. func insert(network, which, cache string) {
  443. var newMessage = cachedMessage{
  444. Message: cache,
  445. LastSeen: time.Now(),
  446. }
  447. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  448. }
  449. func read(network, which string) string {
  450. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  451. if isok {
  452. var readMessage = val.(cachedMessage) // fetch current cached message
  453. if readMessage.LastSeen.IsZero() {
  454. return ""
  455. }
  456. if time.Now().After(readMessage.LastSeen.Add(time.Minute * 10)) { // check if message has been there over a minute
  457. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  458. return ""
  459. }
  460. return readMessage.Message // return current message if not expired
  461. }
  462. return ""
  463. }
  464. // == End Message Caches ==