daemon.go 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489
  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. for _, network := range networks {
  42. cfg := config.ClientConfig{}
  43. cfg.Network = network
  44. cfg.ReadConfig()
  45. serverSet[cfg.Server.Server] = exists
  46. //temporary code --- remove in version v0.13.0
  47. removeHostDNS(network, ncutils.IsWindows())
  48. // end of code to be removed in version v0.13.0
  49. initialPull(cfg.Network)
  50. }
  51. // == subscribe to all nodes for each on machine ==
  52. for server := range serverSet {
  53. logger.Log(1, "started daemon for server ", server)
  54. ctx, cancel := context.WithCancel(context.Background())
  55. networkcontext.Store(server, cancel)
  56. go messageQueue(ctx, server)
  57. }
  58. // == add waitgroup and cancel for checkin routine ==
  59. wg := sync.WaitGroup{}
  60. ctx, cancel := context.WithCancel(context.Background())
  61. wg.Add(1)
  62. go Checkin(ctx, &wg, serverSet)
  63. quit := make(chan os.Signal, 1)
  64. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  65. <-quit
  66. for server := range serverSet {
  67. if cancel, ok := networkcontext.Load(server); ok {
  68. cancel.(context.CancelFunc)()
  69. }
  70. }
  71. cancel()
  72. logger.Log(0, "shutting down netclient daemon")
  73. wg.Wait()
  74. logger.Log(0, "shutdown complete")
  75. return nil
  76. }
  77. // UpdateKeys -- updates private key and returns new publickey
  78. func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
  79. logger.Log(0, "received message to update wireguard keys for network ", nodeCfg.Network)
  80. key, err := wgtypes.GeneratePrivateKey()
  81. if err != nil {
  82. logger.Log(0, "error generating privatekey ", err.Error())
  83. return err
  84. }
  85. file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
  86. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  87. logger.Log(0, "error updating wireguard key ", err.Error())
  88. return err
  89. }
  90. if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
  91. logger.Log(0, "failed to save private key", storeErr.Error())
  92. return storeErr
  93. }
  94. nodeCfg.Node.PublicKey = key.PublicKey().String()
  95. PublishNodeUpdate(nodeCfg)
  96. return nil
  97. }
  98. // PingServer -- checks if server is reachable
  99. // use commsCfg only*
  100. func PingServer(cfg *config.ClientConfig) error {
  101. pinger, err := ping.NewPinger(cfg.Server.Server)
  102. if err != nil {
  103. return err
  104. }
  105. pinger.Timeout = 2 * time.Second
  106. pinger.Run()
  107. stats := pinger.Statistics()
  108. if stats.PacketLoss == 100 {
  109. return errors.New("ping error")
  110. }
  111. logger.Log(3, "ping of server", cfg.Server.Server, "was successful")
  112. return nil
  113. }
  114. // == Private ==
  115. // sets MQ client subscriptions for a specific node config
  116. // should be called for each node belonging to a given comms network
  117. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  118. if nodeCfg.DebugOn {
  119. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  120. logger.Log(0, token.Error().Error())
  121. return
  122. }
  123. logger.Log(0, "subscribed to all topics for debugging purposes")
  124. }
  125. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  126. logger.Log(0, token.Error().Error())
  127. return
  128. }
  129. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  130. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  131. logger.Log(0, token.Error().Error())
  132. return
  133. }
  134. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  135. }
  136. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  137. // for the node in nodeCfg
  138. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  139. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  140. var ok = true
  141. if token := client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  142. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  143. ok = false
  144. }
  145. if token := client.Unsubscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.Wait() && token.Error() != nil {
  146. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  147. ok = false
  148. }
  149. if ok {
  150. logger.Log(1, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  151. }
  152. }
  153. // sets up Message Queue and subsribes/publishes updates to/from server
  154. // the client should subscribe to ALL nodes that exist on server locally
  155. func messageQueue(ctx context.Context, server string) {
  156. logger.Log(0, "netclient daemon started for server: ", server)
  157. client := setupMQTTSub(server)
  158. defer client.Disconnect(250)
  159. <-ctx.Done()
  160. logger.Log(0, "shutting down daemon for server ", server)
  161. }
  162. // setupMQTTSub creates a connection to broker and subscribes to topic
  163. // utilizes comms client configs to setup connections
  164. func setupMQTTSub(server string) mqtt.Client {
  165. opts := mqtt.NewClientOptions()
  166. opts.AddBroker("ssl://" + server + ":8883") // TODO get the appropriate port of the comms mq server
  167. opts.TLSConfig = NewTLSConfig(nil, server)
  168. opts.ClientID = ncutils.MakeRandomString(23) // helps avoid id duplication on broker
  169. opts.SetDefaultPublishHandler(All)
  170. opts.SetAutoReconnect(true)
  171. opts.SetConnectRetry(true)
  172. opts.SetConnectRetryInterval(time.Second << 2)
  173. opts.SetKeepAlive(time.Minute >> 1)
  174. opts.SetWriteTimeout(time.Minute)
  175. opts.SetOnConnectHandler(func(client mqtt.Client) {
  176. networks, err := ncutils.GetSystemNetworks()
  177. if err != nil {
  178. logger.Log(0, "error retriving networks ", err.Error())
  179. }
  180. for _, network := range networks {
  181. var currNodeCfg config.ClientConfig
  182. currNodeCfg.Network = network
  183. currNodeCfg.ReadConfig()
  184. if currNodeCfg.Server.Server == server {
  185. setSubscriptions(client, &currNodeCfg)
  186. }
  187. }
  188. })
  189. opts.SetOrderMatters(true)
  190. opts.SetResumeSubs(true)
  191. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  192. logger.Log(0, "detected broker connection lost, running pull for all nodes")
  193. networks, err := ncutils.GetSystemNetworks()
  194. if err != nil {
  195. logger.Log(0, "error retriving networks ", err.Error())
  196. }
  197. for _, network := range networks {
  198. var cfg config.ClientConfig
  199. cfg.Network = network
  200. cfg.ReadConfig()
  201. _, err := Pull(cfg.Node.Network, true)
  202. if err != nil {
  203. logger.Log(0, "could not run pull, server unreachable: ", err.Error())
  204. logger.Log(0, "waiting to retry...")
  205. }
  206. }
  207. // don't think following log message is accurate
  208. //logger.Log(0, "connection re-established with mqtt server")
  209. })
  210. client := mqtt.NewClient(opts)
  211. tperiod := time.Now().Add(12 * time.Second)
  212. for {
  213. //if after 12 seconds, try a gRPC pull on the last try
  214. if time.Now().After(tperiod) {
  215. networks, err := ncutils.GetSystemNetworks()
  216. if err != nil {
  217. logger.Log(0, "error retriving networks ", err.Error())
  218. }
  219. for _, network := range networks {
  220. var cfg config.ClientConfig
  221. cfg.Network = network
  222. cfg.ReadConfig()
  223. if cfg.Server.Server == server {
  224. _, err := Pull(cfg.Node.Network, true)
  225. if err != nil {
  226. logger.Log(0, "could not run pull, exiting ", cfg.Node.Network, " setup: ", err.Error())
  227. return client
  228. }
  229. }
  230. }
  231. time.Sleep(time.Second)
  232. }
  233. if token := client.Connect(); token.Wait() && token.Error() != nil {
  234. logger.Log(0, "unable to connect to broker, retrying ...")
  235. if time.Now().After(tperiod) {
  236. logger.Log(0, "could not connect to broker, exiting ", server, " setup: ", token.Error().Error())
  237. if strings.Contains(token.Error().Error(), "connectex") || strings.Contains(token.Error().Error(), "i/o timeout") {
  238. logger.Log(0, "connection issue detected.. restarting daemon")
  239. daemon.Restart()
  240. }
  241. return client
  242. }
  243. } else {
  244. break
  245. }
  246. time.Sleep(2 * time.Second)
  247. }
  248. return client
  249. }
  250. // NewTLSConf sets up tls to connect to broker
  251. func NewTLSConfig(cfg *config.ClientConfig, server string) *tls.Config {
  252. var file string
  253. if cfg != nil {
  254. server = cfg.Server.Server
  255. file = "/etc/netclient/" + cfg.Server.Server + "/root.pem"
  256. } else {
  257. file = "/etc/netclient/" + server + "/root.pem"
  258. }
  259. certpool := x509.NewCertPool()
  260. ca, err := os.ReadFile(file)
  261. if err != nil {
  262. logger.Log(0, "could not read CA file %v\n", err.Error())
  263. }
  264. ok := certpool.AppendCertsFromPEM(ca)
  265. if !ok {
  266. logger.Log(0, "failed to append cert")
  267. }
  268. //clientKeyPair, err := tls.LoadX509KeyPair("/etc/netclient/"+cfg.Server.Server+"/client.pem", "/etc/netclient/client.key")
  269. //if err != nil {
  270. // log.Fatalf("could not read client cert/key %v \n", err)
  271. //}
  272. return &tls.Config{
  273. RootCAs: certpool,
  274. ClientAuth: tls.NoClientCert,
  275. //ClientAuth: tls.VerifyClientCertIfGiven,
  276. ClientCAs: nil,
  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("/etc/netclient/" + 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. // utilizes comms client configs to setup connections
  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") // TODO get the appropriate port of the comms mq server
  307. opts.TLSConfig = NewTLSConfig(cfg, "")
  308. opts.ClientID = ncutils.MakeRandomString(23) // helps avoid id duplication on broker
  309. opts.SetDefaultPublishHandler(All)
  310. opts.SetAutoReconnect(true)
  311. opts.SetConnectRetry(true)
  312. opts.SetConnectRetryInterval(time.Second << 2)
  313. opts.SetKeepAlive(time.Minute >> 1)
  314. opts.SetWriteTimeout(time.Minute)
  315. opts.SetOnConnectHandler(func(client mqtt.Client) {
  316. if !publish {
  317. networks, err := ncutils.GetSystemNetworks()
  318. if err != nil {
  319. logger.Log(0, "error retriving networks ", err.Error())
  320. }
  321. for _, network := range networks {
  322. var currNodeCfg config.ClientConfig
  323. currNodeCfg.Network = network
  324. currNodeCfg.ReadConfig()
  325. setSubscriptions(client, &currNodeCfg)
  326. }
  327. }
  328. })
  329. opts.SetOrderMatters(true)
  330. opts.SetResumeSubs(true)
  331. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  332. logger.Log(0, "detected broker connection lost, running pull for ", cfg.Node.Network)
  333. _, err := Pull(cfg.Node.Network, true)
  334. if err != nil {
  335. logger.Log(0, "could not run pull, server unreachable: ", err.Error())
  336. logger.Log(0, "waiting to retry...")
  337. }
  338. logger.Log(0, "connection re-established with mqtt server")
  339. })
  340. client := mqtt.NewClient(opts)
  341. tperiod := time.Now().Add(12 * time.Second)
  342. for {
  343. //if after 12 seconds, try a gRPC pull on the last try
  344. if time.Now().After(tperiod) {
  345. logger.Log(0, "running pull for ", cfg.Node.Network)
  346. _, err := Pull(cfg.Node.Network, true)
  347. if err != nil {
  348. logger.Log(0, "could not run pull, exiting ", cfg.Node.Network, " setup: ", err.Error())
  349. return client
  350. }
  351. time.Sleep(time.Second)
  352. }
  353. if token := client.Connect(); token.Wait() && token.Error() != nil {
  354. logger.Log(0, "unable to connect to broker, retrying ...")
  355. if time.Now().After(tperiod) {
  356. logger.Log(0, "could not connect to broker, exiting ", cfg.Node.Network, " setup: ", token.Error().Error())
  357. if strings.Contains(token.Error().Error(), "connectex") || strings.Contains(token.Error().Error(), "i/o timeout") {
  358. logger.Log(0, "connection issue detected.. pulling and restarting daemon")
  359. Pull(cfg.Node.Network, true)
  360. daemon.Restart()
  361. }
  362. return client
  363. }
  364. } else {
  365. break
  366. }
  367. time.Sleep(2 * time.Second)
  368. }
  369. return client
  370. }
  371. // publishes a message to server to update peers on this peer's behalf
  372. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  373. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  374. return err
  375. }
  376. return nil
  377. }
  378. func initialPull(network string) {
  379. logger.Log(0, "pulling latest config for ", network)
  380. var configPath = fmt.Sprintf("%snetconfig-%s", ncutils.GetNetclientPathSpecific(), network)
  381. fileInfo, err := os.Stat(configPath)
  382. if err != nil {
  383. logger.Log(0, "could not stat config file: ", configPath)
  384. return
  385. }
  386. // speed up UDP rest
  387. if !fileInfo.ModTime().IsZero() && time.Now().After(fileInfo.ModTime().Add(time.Minute)) {
  388. sleepTime := 2
  389. for {
  390. _, err := Pull(network, true)
  391. if err == nil {
  392. break
  393. }
  394. if sleepTime > 3600 {
  395. sleepTime = 3600
  396. }
  397. logger.Log(0, "failed to pull for network ", network)
  398. logger.Log(0, fmt.Sprintf("waiting %d seconds to retry...", sleepTime))
  399. time.Sleep(time.Second * time.Duration(sleepTime))
  400. sleepTime = sleepTime * 2
  401. }
  402. time.Sleep(time.Second << 1)
  403. }
  404. }
  405. func parseNetworkFromTopic(topic string) string {
  406. return strings.Split(topic, "/")[1]
  407. }
  408. // should only ever use node client configs
  409. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  410. if len(msg) <= 24 { // make sure message is of appropriate length
  411. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  412. }
  413. // setup the keys
  414. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  415. if keyErr != nil {
  416. return nil, keyErr
  417. }
  418. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  419. if err != nil {
  420. return nil, err
  421. }
  422. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  423. }
  424. func getServerAddress(cfg *config.ClientConfig) string {
  425. var server models.ServerAddr
  426. for _, server = range cfg.Node.NetworkSettings.DefaultServerAddrs {
  427. if server.Address != "" && server.IsLeader {
  428. break
  429. }
  430. }
  431. return server.Address
  432. }
  433. // == Message Caches ==
  434. func insert(network, which, cache string) {
  435. var newMessage = cachedMessage{
  436. Message: cache,
  437. LastSeen: time.Now(),
  438. }
  439. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  440. }
  441. func read(network, which string) string {
  442. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  443. if isok {
  444. var readMessage = val.(cachedMessage) // fetch current cached message
  445. if readMessage.LastSeen.IsZero() {
  446. return ""
  447. }
  448. if time.Now().After(readMessage.LastSeen.Add(time.Minute * 10)) { // check if message has been there over a minute
  449. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  450. return ""
  451. }
  452. return readMessage.Message // return current message if not expired
  453. }
  454. return ""
  455. }
  456. // == End Message Caches ==