daemon.go 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493
  1. package functions
  2. import (
  3. "context"
  4. "crypto/ed25519"
  5. "crypto/tls"
  6. "crypto/x509"
  7. "errors"
  8. "fmt"
  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. 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, serverSet)
  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. // use commsCfg only*
  104. func PingServer(cfg *config.ClientConfig) error {
  105. pinger, err := ping.NewPinger(cfg.Server.Server)
  106. if err != nil {
  107. return err
  108. }
  109. pinger.Timeout = 2 * time.Second
  110. pinger.Run()
  111. stats := pinger.Statistics()
  112. if stats.PacketLoss == 100 {
  113. return errors.New("ping error")
  114. }
  115. logger.Log(3, "ping of server", cfg.Server.Server, "was successful")
  116. return nil
  117. }
  118. // == Private ==
  119. // sets MQ client subscriptions for a specific node config
  120. // should be called for each node belonging to a given comms network
  121. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  122. if nodeCfg.DebugOn {
  123. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  124. logger.Log(0, token.Error().Error())
  125. return
  126. }
  127. logger.Log(0, "subscribed to all topics for debugging purposes")
  128. }
  129. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  130. logger.Log(0, token.Error().Error())
  131. return
  132. }
  133. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  134. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  135. logger.Log(0, token.Error().Error())
  136. return
  137. }
  138. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  139. }
  140. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  141. // for the node in nodeCfg
  142. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  143. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  144. var ok = true
  145. if token := client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  146. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  147. ok = false
  148. }
  149. if token := client.Unsubscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  150. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  151. ok = false
  152. }
  153. if ok {
  154. logger.Log(1, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  155. }
  156. }
  157. // sets up Message Queue and subsribes/publishes updates to/from server
  158. // the client should subscribe to ALL nodes that exist on server locally
  159. func messageQueue(ctx context.Context, server string) {
  160. logger.Log(0, "netclient daemon started for server: ", server)
  161. client := setupMQTTSub(server)
  162. defer client.Disconnect(250)
  163. <-ctx.Done()
  164. logger.Log(0, "shutting down daemon for server ", server)
  165. }
  166. // setupMQTTSub creates a connection to broker and subscribes to topic
  167. // utilizes comms client configs to setup connections
  168. func setupMQTTSub(server string) mqtt.Client {
  169. opts := mqtt.NewClientOptions()
  170. opts.AddBroker("ssl://" + server + ":8883") // TODO get the appropriate port of the comms mq server
  171. opts.TLSConfig = NewTLSConfig(nil, server)
  172. opts.ClientID = ncutils.MakeRandomString(23) // helps avoid id duplication on broker
  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 to connect to broker
  255. func NewTLSConfig(cfg *config.ClientConfig, server string) *tls.Config {
  256. var file string
  257. if cfg != nil {
  258. server = cfg.Server.Server
  259. file = "/etc/netclient/" + cfg.Server.Server + "/root.pem"
  260. } else {
  261. file = "/etc/netclient/" + server + "/root.pem"
  262. }
  263. certpool := x509.NewCertPool()
  264. ca, err := os.ReadFile(file)
  265. if err != nil {
  266. logger.Log(0, "could not read CA file %v\n", err.Error())
  267. }
  268. ok := certpool.AppendCertsFromPEM(ca)
  269. if !ok {
  270. logger.Log(0, "failed to append cert")
  271. }
  272. //clientKeyPair, err := tls.LoadX509KeyPair("/etc/netclient/"+cfg.Server.Server+"/client.pem", "/etc/netclient/client.key")
  273. //if err != nil {
  274. // log.Fatalf("could not read client cert/key %v \n", err)
  275. //}
  276. return &tls.Config{
  277. RootCAs: certpool,
  278. ClientAuth: tls.NoClientCert,
  279. //ClientAuth: tls.VerifyClientCertIfGiven,
  280. ClientCAs: nil,
  281. //InsecureSkipVerify: false fails ---- so need to use VerifyConnection
  282. InsecureSkipVerify: true,
  283. VerifyConnection: func(cs tls.ConnectionState) error {
  284. if cs.ServerName != server {
  285. logger.Log(0, "VerifyConnection - certifiate mismatch")
  286. return errors.New("certificate doesn't match server")
  287. }
  288. ca, err := ssl.ReadCert("/etc/netclient/" + cs.ServerName + "/root.pem")
  289. if err != nil {
  290. logger.Log(0, "VerifyConnection - unable to read ca", err.Error())
  291. return errors.New("unable to read ca")
  292. }
  293. for _, cert := range cs.PeerCertificates {
  294. if cert.IsCA {
  295. if string(cert.PublicKey.(ed25519.PublicKey)) != string(ca.PublicKey.(ed25519.PublicKey)) {
  296. logger.Log(0, "VerifyConnection - public key mismatch")
  297. return errors.New("cert public key does not match ca public key")
  298. }
  299. }
  300. }
  301. return nil
  302. },
  303. }
  304. }
  305. // setupMQTT creates a connection to broker and return client
  306. // utilizes comms client configs to setup connections
  307. func setupMQTT(cfg *config.ClientConfig, publish bool) mqtt.Client {
  308. opts := mqtt.NewClientOptions()
  309. server := cfg.Server.Server
  310. opts.AddBroker("ssl://" + server + ":8883") // TODO get the appropriate port of the comms mq server
  311. opts.TLSConfig = NewTLSConfig(cfg, "")
  312. opts.ClientID = ncutils.MakeRandomString(23) // helps avoid id duplication on broker
  313. opts.SetDefaultPublishHandler(All)
  314. opts.SetAutoReconnect(true)
  315. opts.SetConnectRetry(true)
  316. opts.SetConnectRetryInterval(time.Second << 2)
  317. opts.SetKeepAlive(time.Minute >> 1)
  318. opts.SetWriteTimeout(time.Minute)
  319. opts.SetOnConnectHandler(func(client mqtt.Client) {
  320. if !publish {
  321. networks, err := ncutils.GetSystemNetworks()
  322. if err != nil {
  323. logger.Log(0, "error retriving networks ", err.Error())
  324. }
  325. for _, network := range networks {
  326. var currNodeCfg config.ClientConfig
  327. currNodeCfg.Network = network
  328. currNodeCfg.ReadConfig()
  329. setSubscriptions(client, &currNodeCfg)
  330. }
  331. }
  332. })
  333. opts.SetOrderMatters(true)
  334. opts.SetResumeSubs(true)
  335. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  336. logger.Log(0, "detected broker connection lost, running pull for ", cfg.Node.Network)
  337. _, err := Pull(cfg.Node.Network, true)
  338. if err != nil {
  339. logger.Log(0, "could not run pull, server unreachable: ", err.Error())
  340. logger.Log(0, "waiting to retry...")
  341. }
  342. logger.Log(0, "connection re-established with mqtt server")
  343. })
  344. client := mqtt.NewClient(opts)
  345. tperiod := time.Now().Add(12 * time.Second)
  346. for {
  347. //if after 12 seconds, try a gRPC pull on the last try
  348. if time.Now().After(tperiod) {
  349. logger.Log(0, "running pull for ", cfg.Node.Network)
  350. _, err := Pull(cfg.Node.Network, true)
  351. if err != nil {
  352. logger.Log(0, "could not run pull, exiting ", cfg.Node.Network, " setup: ", err.Error())
  353. return client
  354. }
  355. time.Sleep(time.Second)
  356. }
  357. if token := client.Connect(); token.Wait() && token.Error() != nil {
  358. logger.Log(0, "unable to connect to broker, retrying ...")
  359. if time.Now().After(tperiod) {
  360. logger.Log(0, "could not connect to broker, exiting ", cfg.Node.Network, " setup: ", token.Error().Error())
  361. if strings.Contains(token.Error().Error(), "connectex") || strings.Contains(token.Error().Error(), "i/o timeout") {
  362. logger.Log(0, "connection issue detected.. pulling and restarting daemon")
  363. Pull(cfg.Node.Network, true)
  364. daemon.Restart()
  365. }
  366. return client
  367. }
  368. } else {
  369. break
  370. }
  371. time.Sleep(2 * time.Second)
  372. }
  373. return client
  374. }
  375. // publishes a message to server to update peers on this peer's behalf
  376. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  377. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  378. return err
  379. }
  380. return nil
  381. }
  382. func initialPull(network string) {
  383. logger.Log(0, "pulling latest config for ", network)
  384. var configPath = fmt.Sprintf("%snetconfig-%s", ncutils.GetNetclientPathSpecific(), network)
  385. fileInfo, err := os.Stat(configPath)
  386. if err != nil {
  387. logger.Log(0, "could not stat config file: ", configPath)
  388. return
  389. }
  390. // speed up UDP rest
  391. if !fileInfo.ModTime().IsZero() && time.Now().After(fileInfo.ModTime().Add(time.Minute)) {
  392. sleepTime := 2
  393. for {
  394. _, err := Pull(network, true)
  395. if err == nil {
  396. break
  397. }
  398. if sleepTime > 3600 {
  399. sleepTime = 3600
  400. }
  401. logger.Log(0, "failed to pull for network ", network)
  402. logger.Log(0, fmt.Sprintf("waiting %d seconds to retry...", sleepTime))
  403. time.Sleep(time.Second * time.Duration(sleepTime))
  404. sleepTime = sleepTime * 2
  405. }
  406. time.Sleep(time.Second << 1)
  407. }
  408. }
  409. func parseNetworkFromTopic(topic string) string {
  410. return strings.Split(topic, "/")[1]
  411. }
  412. // should only ever use node client configs
  413. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  414. if len(msg) <= 24 { // make sure message is of appropriate length
  415. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  416. }
  417. // setup the keys
  418. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  419. if keyErr != nil {
  420. return nil, keyErr
  421. }
  422. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  423. if err != nil {
  424. return nil, err
  425. }
  426. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  427. }
  428. func getServerAddress(cfg *config.ClientConfig) string {
  429. var server models.ServerAddr
  430. for _, server = range cfg.Node.NetworkSettings.DefaultServerAddrs {
  431. if server.Address != "" && server.IsLeader {
  432. break
  433. }
  434. }
  435. return server.Address
  436. }
  437. // == Message Caches ==
  438. func insert(network, which, cache string) {
  439. var newMessage = cachedMessage{
  440. Message: cache,
  441. LastSeen: time.Now(),
  442. }
  443. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  444. }
  445. func read(network, which string) string {
  446. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  447. if isok {
  448. var readMessage = val.(cachedMessage) // fetch current cached message
  449. if readMessage.LastSeen.IsZero() {
  450. return ""
  451. }
  452. if time.Now().After(readMessage.LastSeen.Add(time.Minute * 10)) { // check if message has been there over a minute
  453. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  454. return ""
  455. }
  456. return readMessage.Message // return current message if not expired
  457. }
  458. return ""
  459. }
  460. // == End Message Caches ==