daemon.go 17 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538
  1. package functions
  2. import (
  3. "context"
  4. "encoding/json"
  5. "errors"
  6. "fmt"
  7. "log"
  8. "os"
  9. "os/signal"
  10. "runtime"
  11. "strings"
  12. "sync"
  13. "syscall"
  14. "time"
  15. mqtt "github.com/eclipse/paho.mqtt.golang"
  16. "github.com/gravitl/netmaker/logger"
  17. "github.com/gravitl/netmaker/models"
  18. "github.com/gravitl/netmaker/netclient/auth"
  19. "github.com/gravitl/netmaker/netclient/config"
  20. "github.com/gravitl/netmaker/netclient/local"
  21. "github.com/gravitl/netmaker/netclient/ncutils"
  22. "github.com/gravitl/netmaker/netclient/wireguard"
  23. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  24. )
  25. // ServerKeepalive - stores time of last server keepalive message
  26. var keepalive = make(map[string]time.Time, 3)
  27. var messageCache = make(map[string]string, 20)
  28. const lastNodeUpdate = "lnu"
  29. const lastPeerUpdate = "lpu"
  30. func insert(network, which, cache string) {
  31. var mu sync.Mutex
  32. mu.Lock()
  33. defer mu.Unlock()
  34. messageCache[fmt.Sprintf("%s%s", network, which)] = cache
  35. }
  36. func read(network, which string) string {
  37. return messageCache[fmt.Sprintf("%s%s", network, which)]
  38. }
  39. // Daemon runs netclient daemon from command line
  40. func Daemon() error {
  41. ctx, cancel := context.WithCancel(context.Background())
  42. networks, err := ncutils.GetSystemNetworks()
  43. if err != nil {
  44. cancel()
  45. return err
  46. }
  47. for _, network := range networks {
  48. go MessageQueue(ctx, network)
  49. }
  50. quit := make(chan os.Signal, 1)
  51. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  52. <-quit
  53. cancel()
  54. ncutils.Log("all done")
  55. return nil
  56. }
  57. // SetupMQTT creates a connection to broker and return client
  58. func SetupMQTT(cfg *config.ClientConfig) mqtt.Client {
  59. opts := mqtt.NewClientOptions()
  60. for _, server := range cfg.Node.NetworkSettings.DefaultServerAddrs {
  61. if server.Address != "" && server.IsLeader {
  62. ncutils.Log(fmt.Sprintf("adding server (%s) to listen on network %s", server.Address, cfg.Node.Network))
  63. opts.AddBroker(server.Address + ":1883")
  64. break
  65. }
  66. }
  67. opts.SetDefaultPublishHandler(All)
  68. client := mqtt.NewClient(opts)
  69. tperiod := time.Now().Add(12 * time.Second)
  70. for {
  71. //if after 12 seconds, try a gRPC pull on the last try
  72. if time.Now().After(tperiod) {
  73. _, err := Pull(cfg.Node.Network, true)
  74. if err != nil {
  75. log.Fatal(0, "could not connect to broker, exiting ...", err.Error())
  76. }
  77. time.Sleep(2 * time.Second)
  78. }
  79. if token := client.Connect(); token.Wait() && token.Error() != nil {
  80. logger.Log(1, "unable to connect to broker, retrying ...")
  81. if time.Now().After(tperiod) {
  82. log.Fatal(0, "could not connect to broker, exiting ...", token.Error())
  83. }
  84. } else {
  85. break
  86. }
  87. time.Sleep(2 * time.Second)
  88. }
  89. return client
  90. }
  91. // MessageQueue sets up Message Queue and subsribes/publishes updates to/from server
  92. func MessageQueue(ctx context.Context, network string) {
  93. ncutils.Log("netclient go routine started for " + network)
  94. var cfg config.ClientConfig
  95. cfg.Network = network
  96. cfg.ReadConfig()
  97. ncutils.Log("daemon started for network:" + network)
  98. client := SetupMQTT(&cfg)
  99. if cfg.DebugOn {
  100. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  101. log.Fatal(token.Error())
  102. }
  103. ncutils.Log("subscribed to all topics for debugging purposes")
  104. }
  105. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  106. log.Fatal(token.Error())
  107. }
  108. if cfg.DebugOn {
  109. ncutils.Log(fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", cfg.Node.Name, cfg.Node.Network, cfg.Node.ID))
  110. }
  111. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  112. log.Fatal(token.Error())
  113. }
  114. if cfg.DebugOn {
  115. ncutils.Log(fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", cfg.Node.Name, cfg.Node.Network, cfg.Node.ID))
  116. }
  117. var id string
  118. for _, server := range cfg.NetworkSettings.DefaultServerAddrs {
  119. if server.IsLeader {
  120. id = server.ID
  121. }
  122. if server.Address != "" {
  123. if token := client.Subscribe("serverkeepalive/"+id, 0, mqtt.MessageHandler(ServerKeepAlive)); token.Wait() && token.Error() != nil {
  124. log.Fatal(token.Error())
  125. }
  126. if cfg.DebugOn {
  127. ncutils.Log("subscribed to server keepalives for server " + id)
  128. }
  129. } else {
  130. ncutils.Log("leader not defined for network" + cfg.Network)
  131. }
  132. }
  133. defer client.Disconnect(250)
  134. go MonitorKeepalive(ctx, client, &cfg)
  135. go Checkin(ctx, &cfg, network)
  136. <-ctx.Done()
  137. ncutils.Log("shutting down daemon")
  138. }
  139. // All -- mqtt message hander for all ('#') topics
  140. var All mqtt.MessageHandler = func(client mqtt.Client, msg mqtt.Message) {
  141. ncutils.Log("default message handler -- received message but not handling")
  142. ncutils.Log("Topic: " + string(msg.Topic()))
  143. //ncutils.Log("Message: " + string(msg.Payload()))
  144. }
  145. // NodeUpdate -- mqtt message handler for /update/<NodeID> topic
  146. func NodeUpdate(client mqtt.Client, msg mqtt.Message) {
  147. //potentiall blocking i/o so do this in a go routine
  148. go func() {
  149. var newNode models.Node
  150. var cfg config.ClientConfig
  151. var network = parseNetworkFromTopic(msg.Topic())
  152. cfg.Network = network
  153. cfg.ReadConfig()
  154. data, dataErr := decryptMsg(&cfg, msg.Payload())
  155. if dataErr != nil {
  156. return
  157. }
  158. err := json.Unmarshal([]byte(data), &newNode)
  159. if err != nil {
  160. ncutils.Log("error unmarshalling node update data" + err.Error())
  161. return
  162. }
  163. ncutils.Log("received message to update node " + newNode.Name)
  164. // see if cache hit, if so skip
  165. var currentMessage = read(newNode.Network, lastNodeUpdate)
  166. if currentMessage == string(data) {
  167. return
  168. }
  169. insert(newNode.Network, lastNodeUpdate, string(data))
  170. //check if interface name has changed if so delete.
  171. if cfg.Node.Interface != newNode.Interface {
  172. if err = wireguard.RemoveConf(cfg.Node.Interface, true); err != nil {
  173. ncutils.PrintLog("could not delete old interface "+cfg.Node.Interface+": "+err.Error(), 1)
  174. }
  175. }
  176. newNode.PullChanges = "no"
  177. //ensure that OS never changes
  178. newNode.OS = runtime.GOOS
  179. // check if interface needs to delta
  180. ifaceDelta := ncutils.IfaceDelta(&cfg.Node, &newNode)
  181. cfg.Node = newNode
  182. switch newNode.Action {
  183. case models.NODE_DELETE:
  184. if err := RemoveLocalInstance(&cfg, cfg.Network); err != nil {
  185. ncutils.PrintLog("error deleting local instance: "+err.Error(), 1)
  186. return
  187. }
  188. if token := client.Unsubscribe(fmt.Sprintf("update/%s/%s", newNode.Network, newNode.ID), fmt.Sprintf("peers/%s/%s", newNode.Network, newNode.ID)); token.Wait() && token.Error() != nil {
  189. ncutils.PrintLog("error unsubscribing during node deletion", 1)
  190. }
  191. return
  192. case models.NODE_UPDATE_KEY:
  193. if err := UpdateKeys(&cfg, client); err != nil {
  194. ncutils.PrintLog("err updating wireguard keys: "+err.Error(), 1)
  195. }
  196. case models.NODE_NOOP:
  197. default:
  198. }
  199. //Save new config
  200. if err := config.Write(&cfg, cfg.Network); err != nil {
  201. ncutils.PrintLog("error updating node configuration: "+err.Error(), 1)
  202. }
  203. nameserver := cfg.Server.CoreDNSAddr
  204. privateKey, err := wireguard.RetrievePrivKey(newNode.Network)
  205. if err != nil {
  206. ncutils.Log("error reading PrivateKey " + err.Error())
  207. return
  208. }
  209. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  210. if err := wireguard.UpdateWgInterface(file, privateKey, nameserver, newNode); err != nil {
  211. ncutils.Log("error updating wireguard config " + err.Error())
  212. return
  213. }
  214. if ifaceDelta {
  215. ncutils.Log("applying WG conf to " + file)
  216. err = wireguard.ApplyWGQuickConf(file)
  217. if err != nil {
  218. ncutils.Log("error restarting wg after node update " + err.Error())
  219. return
  220. }
  221. time.Sleep(time.Second >> 1)
  222. if err = Resubscribe(client, &cfg); err != nil {
  223. ncutils.Log("error resubscribing after interface change " + err.Error())
  224. return
  225. }
  226. }
  227. /*
  228. else {
  229. ncutils.Log("syncing conf to " + file)
  230. err = wireguard.SyncWGQuickConf(cfg.Node.Interface, file)
  231. if err != nil {
  232. ncutils.Log("error syncing wg after peer update " + err.Error())
  233. return
  234. }
  235. }
  236. */
  237. //deal with DNS
  238. if newNode.DNSOn == "yes" {
  239. ncutils.Log("setting up DNS")
  240. if err = local.UpdateDNS(cfg.Node.Interface, cfg.Network, cfg.Server.CoreDNSAddr); err != nil {
  241. ncutils.Log("error applying dns" + err.Error())
  242. }
  243. } else {
  244. ncutils.Log("settng DNS off")
  245. _, err := ncutils.RunCmd("/usr/bin/resolvectl revert "+cfg.Node.Interface, true)
  246. if err != nil {
  247. ncutils.Log("error applying dns" + err.Error())
  248. }
  249. }
  250. }()
  251. }
  252. // UpdatePeers -- mqtt message handler for peers/<Network>/<NodeID> topic
  253. func UpdatePeers(client mqtt.Client, msg mqtt.Message) {
  254. go func() {
  255. var peerUpdate models.PeerUpdate
  256. var network = parseNetworkFromTopic(msg.Topic())
  257. var cfg = config.ClientConfig{}
  258. cfg.Network = network
  259. cfg.ReadConfig()
  260. data, dataErr := decryptMsg(&cfg, msg.Payload())
  261. if dataErr != nil {
  262. return
  263. }
  264. err := json.Unmarshal([]byte(data), &peerUpdate)
  265. if err != nil {
  266. ncutils.Log("error unmarshalling peer data")
  267. return
  268. }
  269. // see if cache hit, if so skip
  270. var currentMessage = read(peerUpdate.Network, lastPeerUpdate)
  271. if currentMessage == string(data) {
  272. return
  273. }
  274. insert(peerUpdate.Network, lastPeerUpdate, string(data))
  275. ncutils.Log("update peer handler")
  276. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  277. err = wireguard.UpdateWgPeers(file, peerUpdate.Peers)
  278. if err != nil {
  279. ncutils.Log("error updating wireguard peers" + err.Error())
  280. return
  281. }
  282. ncutils.Log("syncing conf to " + file)
  283. //err = wireguard.SyncWGQuickConf(cfg.Node.Interface, file)
  284. err = wireguard.SetPeers(cfg.Node.Interface, cfg.Node.PersistentKeepalive, peerUpdate.Peers)
  285. if err != nil {
  286. ncutils.Log("error syncing wg after peer update " + err.Error())
  287. return
  288. }
  289. }()
  290. }
  291. // MonitorKeepalive - checks time last server keepalive received. If more than 3+ minutes, notify and resubscribe
  292. func MonitorKeepalive(ctx context.Context, client mqtt.Client, cfg *config.ClientConfig) {
  293. var id string
  294. for _, servAddr := range cfg.NetworkSettings.DefaultServerAddrs {
  295. if servAddr.IsLeader {
  296. id = servAddr.ID
  297. }
  298. }
  299. for {
  300. select {
  301. case <-ctx.Done():
  302. return
  303. case <-time.After(time.Second * 150):
  304. if time.Since(keepalive[id]) > time.Second*200 { // more than 3+ minutes
  305. ncutils.Log("server keepalive not recieved in more than minutes, resubscribe to message queue")
  306. Resubscribe(client, cfg)
  307. }
  308. }
  309. }
  310. }
  311. // ServerKeepAlive -- handler to react to keepalive messages published by server
  312. func ServerKeepAlive(client mqtt.Client, msg mqtt.Message) {
  313. var mu sync.Mutex
  314. mu.Lock()
  315. defer mu.Unlock()
  316. serverid, err := getID(msg.Topic())
  317. if err != nil {
  318. ncutils.Log("invalid ID in serverkeepalive topic")
  319. }
  320. keepalive[serverid] = time.Now()
  321. // ncutils.Log("keepalive from server")
  322. }
  323. // Resubscribe --- handles resubscribing if needed
  324. func Resubscribe(client mqtt.Client, cfg *config.ClientConfig) error {
  325. if err := config.ModConfig(&cfg.Node); err == nil {
  326. ncutils.Log("resubbing on network " + cfg.Node.Network)
  327. client.Disconnect(250)
  328. client = SetupMQTT(cfg)
  329. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, NodeUpdate); token.Wait() && token.Error() != nil {
  330. log.Fatal(token.Error())
  331. }
  332. if cfg.DebugOn {
  333. ncutils.Log("subscribed to node updates for node " + cfg.Node.Name + " update/" + cfg.Node.ID)
  334. }
  335. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, UpdatePeers); token.Wait() && token.Error() != nil {
  336. log.Fatal(token.Error())
  337. }
  338. var id string
  339. for _, server := range cfg.NetworkSettings.DefaultServerAddrs {
  340. if server.IsLeader {
  341. id = server.ID
  342. }
  343. if server.Address != "" {
  344. if token := client.Subscribe("serverkeepalive/"+id, 0, mqtt.MessageHandler(ServerKeepAlive)); token.Wait() && token.Error() != nil {
  345. log.Fatal(token.Error())
  346. }
  347. if cfg.DebugOn {
  348. ncutils.Log("subscribed to server keepalives for server " + id)
  349. }
  350. } else {
  351. ncutils.Log("leader not defined for network" + cfg.Network)
  352. }
  353. }
  354. ncutils.Log("finished re subbing")
  355. return nil
  356. } else {
  357. ncutils.Log("could not mod config when re-subbing")
  358. return err
  359. }
  360. }
  361. // UpdateKeys -- updates private key and returns new publickey
  362. func UpdateKeys(cfg *config.ClientConfig, client mqtt.Client) error {
  363. ncutils.Log("received message to update keys")
  364. //potentiall blocking i/o so do this in a go routine
  365. key, err := wgtypes.GeneratePrivateKey()
  366. if err != nil {
  367. ncutils.Log("error generating privatekey " + err.Error())
  368. return err
  369. }
  370. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  371. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  372. ncutils.Log("error updating wireguard key " + err.Error())
  373. return err
  374. }
  375. cfg.Node.PublicKey = key.PublicKey().String()
  376. PublishNodeUpdate(cfg)
  377. if err := config.ModConfig(&cfg.Node); err != nil {
  378. ncutils.Log("error updating local config " + err.Error())
  379. }
  380. return nil
  381. }
  382. // Checkin -- go routine that checks for public or local ip changes, publishes changes
  383. // if there are no updates, simply "pings" the server as a checkin
  384. func Checkin(ctx context.Context, cfg *config.ClientConfig, network string) {
  385. for {
  386. select {
  387. case <-ctx.Done():
  388. ncutils.Log("Checkin cancelled")
  389. return
  390. //delay should be configuraable -> use cfg.Node.NetworkSettings.DefaultCheckInInterval ??
  391. case <-time.After(time.Second * 60):
  392. // ncutils.Log("Checkin running")
  393. //read latest config
  394. cfg.ReadConfig()
  395. if cfg.Node.Roaming == "yes" && cfg.Node.IsStatic != "yes" {
  396. extIP, err := ncutils.GetPublicIP()
  397. if err != nil {
  398. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  399. }
  400. if cfg.Node.Endpoint != extIP && extIP != "" {
  401. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+extIP, 1)
  402. cfg.Node.Endpoint = extIP
  403. PublishNodeUpdate(cfg)
  404. }
  405. intIP, err := getPrivateAddr()
  406. if err != nil {
  407. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  408. }
  409. if cfg.Node.LocalAddress != intIP && intIP != "" {
  410. ncutils.PrintLog("local Address has changed from "+cfg.Node.LocalAddress+" to "+intIP, 1)
  411. cfg.Node.LocalAddress = intIP
  412. PublishNodeUpdate(cfg)
  413. }
  414. } else {
  415. localIP, err := ncutils.GetLocalIP(cfg.Node.LocalRange)
  416. if err != nil {
  417. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  418. }
  419. if cfg.Node.Endpoint != localIP && localIP != "" {
  420. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+localIP, 1)
  421. cfg.Node.Endpoint = localIP
  422. PublishNodeUpdate(cfg)
  423. }
  424. }
  425. Hello(cfg, network)
  426. // ncutils.Log("Checkin complete")
  427. }
  428. }
  429. }
  430. // PublishNodeUpdates -- saves node and pushes changes to broker
  431. func PublishNodeUpdate(cfg *config.ClientConfig) {
  432. if err := config.Write(cfg, cfg.Network); err != nil {
  433. ncutils.Log("error saving configuration" + err.Error())
  434. }
  435. data, err := json.Marshal(cfg.Node)
  436. if err != nil {
  437. ncutils.Log("error marshling node update " + err.Error())
  438. }
  439. if err = publish(cfg, fmt.Sprintf("update/%s", cfg.Node.ID), data); err != nil {
  440. ncutils.Log(fmt.Sprintf("error publishing endpoint update, %v", err))
  441. }
  442. }
  443. // Hello -- ping the broker to let server know node is alive and doing fine
  444. func Hello(cfg *config.ClientConfig, network string) {
  445. if err := publish(cfg, fmt.Sprintf("ping/%s", cfg.Node.ID), []byte("hello world!")); err != nil {
  446. ncutils.Log(fmt.Sprintf("error publishing ping, %v", err))
  447. }
  448. }
  449. func publish(cfg *config.ClientConfig, dest string, msg []byte) error {
  450. // setup the keys
  451. trafficPrivKey, err := auth.RetrieveTrafficKey(cfg.Node.Network)
  452. if err != nil {
  453. return err
  454. }
  455. serverPubKey, err := ncutils.ConvertBytesToKey(cfg.Node.TrafficKeys.Server)
  456. if err != nil {
  457. return err
  458. }
  459. client := SetupMQTT(cfg)
  460. defer client.Disconnect(250)
  461. encrypted, err := ncutils.BoxEncrypt(msg, serverPubKey, trafficPrivKey)
  462. if err != nil {
  463. return err
  464. }
  465. if token := client.Publish(dest, 0, false, encrypted); token.Wait() && token.Error() != nil {
  466. return token.Error()
  467. }
  468. return nil
  469. }
  470. func parseNetworkFromTopic(topic string) string {
  471. return strings.Split(topic, "/")[1]
  472. }
  473. func decryptMsg(cfg *config.ClientConfig, msg []byte) ([]byte, error) {
  474. // setup the keys
  475. diskKey, keyErr := auth.RetrieveTrafficKey(cfg.Node.Network)
  476. if keyErr != nil {
  477. return nil, keyErr
  478. }
  479. serverPubKey, err := ncutils.ConvertBytesToKey(cfg.Node.TrafficKeys.Server)
  480. if err != nil {
  481. return nil, err
  482. }
  483. return ncutils.BoxDecrypt(msg, serverPubKey, diskKey)
  484. }
  485. func shouldResub(currentServers, newServers []models.ServerAddr) bool {
  486. if len(currentServers) != len(newServers) {
  487. return true
  488. }
  489. for _, srv := range currentServers {
  490. if !ncutils.ServerAddrSliceContains(newServers, srv) {
  491. return true
  492. }
  493. }
  494. return false
  495. }
  496. func getID(topic string) (string, error) {
  497. parts := strings.Split(topic, "/")
  498. count := len(parts)
  499. if count == 1 {
  500. return "", errors.New("invalid topic")
  501. }
  502. //the last part of the topic will be the network.ID
  503. return parts[count-1], nil
  504. }