daemon.go 18 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561
  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/models"
  17. "github.com/gravitl/netmaker/netclient/auth"
  18. "github.com/gravitl/netmaker/netclient/config"
  19. "github.com/gravitl/netmaker/netclient/local"
  20. "github.com/gravitl/netmaker/netclient/ncutils"
  21. "github.com/gravitl/netmaker/netclient/wireguard"
  22. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  23. )
  24. // ServerKeepalive - stores time of last server keepalive message
  25. var keepalive = make(map[string]time.Time, 3)
  26. var messageCache = make(map[string]string, 20)
  27. const lastNodeUpdate = "lnu"
  28. const lastPeerUpdate = "lpu"
  29. func insert(network, which, cache string) {
  30. var mu sync.Mutex
  31. mu.Lock()
  32. defer mu.Unlock()
  33. messageCache[fmt.Sprintf("%s%s", network, which)] = cache
  34. }
  35. func read(network, which string) string {
  36. return messageCache[fmt.Sprintf("%s%s", network, which)]
  37. }
  38. // Daemon runs netclient daemon from command line
  39. func Daemon() error {
  40. ctx, cancel := context.WithCancel(context.Background())
  41. networks, err := ncutils.GetSystemNetworks()
  42. if err != nil {
  43. cancel()
  44. return err
  45. }
  46. for _, network := range networks {
  47. go MessageQueue(ctx, network)
  48. }
  49. quit := make(chan os.Signal, 1)
  50. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  51. <-quit
  52. cancel()
  53. ncutils.Log("all done")
  54. return nil
  55. }
  56. // SetupMQTT creates a connection to broker and return client
  57. func SetupMQTT(cfg *config.ClientConfig) mqtt.Client {
  58. opts := mqtt.NewClientOptions()
  59. for _, server := range cfg.Node.NetworkSettings.DefaultServerAddrs {
  60. if server.Address != "" && server.IsLeader {
  61. ncutils.Log(fmt.Sprintf("adding server (%s) to listen on network %s", server.Address, cfg.Node.Network))
  62. opts.AddBroker(server.Address + ":1883")
  63. break
  64. }
  65. }
  66. opts.SetDefaultPublishHandler(All)
  67. client := mqtt.NewClient(opts)
  68. tperiod := time.Now().Add(12 * time.Second)
  69. for {
  70. //if after 12 seconds, try a gRPC pull on the last try
  71. if time.Now().After(tperiod) {
  72. ncutils.Log("running pull for " + cfg.Node.Network)
  73. _, err := Pull(cfg.Node.Network, true)
  74. if err != nil {
  75. ncutils.Log("could not run pull, exiting " + cfg.Node.Network + " setup: " + err.Error())
  76. return client
  77. }
  78. time.Sleep(2 * time.Second)
  79. }
  80. if token := client.Connect(); token.Wait() && token.Error() != nil {
  81. ncutils.Log("unable to connect to broker, retrying ...")
  82. if time.Now().After(tperiod) {
  83. ncutils.Log("could not connect to broker, exiting " + cfg.Node.Network + " setup: " + token.Error().Error())
  84. return client
  85. }
  86. } else {
  87. break
  88. }
  89. time.Sleep(2 * time.Second)
  90. }
  91. return client
  92. }
  93. // MessageQueue sets up Message Queue and subsribes/publishes updates to/from server
  94. func MessageQueue(ctx context.Context, network string) {
  95. ncutils.Log("netclient go routine started for " + network)
  96. var cfg config.ClientConfig
  97. cfg.Network = network
  98. ncutils.Log("pulling latest config for " + cfg.Network)
  99. _, err := Pull(network, true)
  100. if err != nil {
  101. ncutils.Log(err.Error())
  102. return
  103. }
  104. time.Sleep(time.Second << 1)
  105. cfg.ReadConfig()
  106. ncutils.Log("daemon started for network: " + network)
  107. client := SetupMQTT(&cfg)
  108. if cfg.DebugOn {
  109. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  110. ncutils.Log(token.Error().Error())
  111. return
  112. }
  113. ncutils.Log("subscribed to all topics for debugging purposes")
  114. }
  115. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  116. ncutils.Log(token.Error().Error())
  117. return
  118. }
  119. if cfg.DebugOn {
  120. ncutils.Log(fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", cfg.Node.Name, cfg.Node.Network, cfg.Node.ID))
  121. }
  122. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  123. ncutils.Log(token.Error().Error())
  124. return
  125. }
  126. if cfg.DebugOn {
  127. ncutils.Log(fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", cfg.Node.Name, cfg.Node.Network, cfg.Node.ID))
  128. }
  129. var id string
  130. var found bool
  131. for _, server := range cfg.NetworkSettings.DefaultServerAddrs {
  132. if server.IsLeader {
  133. id = server.ID
  134. }
  135. if server.Address != "" {
  136. if token := client.Subscribe("serverkeepalive/"+id, 0, mqtt.MessageHandler(ServerKeepAlive)); token.Wait() && token.Error() != nil {
  137. ncutils.Log(token.Error().Error())
  138. return
  139. }
  140. found = true
  141. if cfg.DebugOn {
  142. ncutils.Log("subscribed to server keepalives for server " + id)
  143. }
  144. }
  145. }
  146. if !found {
  147. ncutils.Log("leader not defined for network " + cfg.Network)
  148. }
  149. defer client.Disconnect(250)
  150. go MonitorKeepalive(ctx, client, &cfg)
  151. go Checkin(ctx, &cfg, network)
  152. <-ctx.Done()
  153. ncutils.Log("shutting down daemon")
  154. }
  155. // All -- mqtt message hander for all ('#') topics
  156. var All mqtt.MessageHandler = func(client mqtt.Client, msg mqtt.Message) {
  157. ncutils.Log("default message handler -- received message but not handling")
  158. ncutils.Log("Topic: " + string(msg.Topic()))
  159. //ncutils.Log("Message: " + string(msg.Payload()))
  160. }
  161. // NodeUpdate -- mqtt message handler for /update/<NodeID> topic
  162. func NodeUpdate(client mqtt.Client, msg mqtt.Message) {
  163. //potentiall blocking i/o so do this in a go routine
  164. go func() {
  165. var newNode models.Node
  166. var cfg config.ClientConfig
  167. var network = parseNetworkFromTopic(msg.Topic())
  168. cfg.Network = network
  169. cfg.ReadConfig()
  170. data, dataErr := decryptMsg(&cfg, msg.Payload())
  171. if dataErr != nil {
  172. return
  173. }
  174. err := json.Unmarshal([]byte(data), &newNode)
  175. if err != nil {
  176. ncutils.Log("error unmarshalling node update data" + err.Error())
  177. return
  178. }
  179. ncutils.Log("received message to update node " + newNode.Name)
  180. // see if cache hit, if so skip
  181. var currentMessage = read(newNode.Network, lastNodeUpdate)
  182. if currentMessage == string(data) {
  183. return
  184. }
  185. insert(newNode.Network, lastNodeUpdate, string(data))
  186. //check if interface name has changed if so delete.
  187. if cfg.Node.Interface != newNode.Interface {
  188. if err = wireguard.RemoveConf(cfg.Node.Interface, true); err != nil {
  189. ncutils.PrintLog("could not delete old interface "+cfg.Node.Interface+": "+err.Error(), 1)
  190. }
  191. }
  192. newNode.PullChanges = "no"
  193. //ensure that OS never changes
  194. newNode.OS = runtime.GOOS
  195. // check if interface needs to delta
  196. ifaceDelta := ncutils.IfaceDelta(&cfg.Node, &newNode)
  197. cfg.Node = newNode
  198. switch newNode.Action {
  199. case models.NODE_DELETE:
  200. if err := RemoveLocalInstance(&cfg, cfg.Network); err != nil {
  201. ncutils.PrintLog("error deleting local instance: "+err.Error(), 1)
  202. return
  203. }
  204. 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 {
  205. ncutils.PrintLog("error unsubscribing during node deletion", 1)
  206. }
  207. return
  208. case models.NODE_UPDATE_KEY:
  209. if err := UpdateKeys(&cfg, client); err != nil {
  210. ncutils.PrintLog("err updating wireguard keys: "+err.Error(), 1)
  211. }
  212. case models.NODE_NOOP:
  213. default:
  214. }
  215. //Save new config
  216. if err := config.Write(&cfg, cfg.Network); err != nil {
  217. ncutils.PrintLog("error updating node configuration: "+err.Error(), 1)
  218. }
  219. nameserver := cfg.Server.CoreDNSAddr
  220. privateKey, err := wireguard.RetrievePrivKey(newNode.Network)
  221. if err != nil {
  222. ncutils.Log("error reading PrivateKey " + err.Error())
  223. return
  224. }
  225. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  226. if err := wireguard.UpdateWgInterface(file, privateKey, nameserver, newNode); err != nil {
  227. ncutils.Log("error updating wireguard config " + err.Error())
  228. return
  229. }
  230. if ifaceDelta {
  231. ncutils.Log("applying WG conf to " + file)
  232. err = wireguard.ApplyWGQuickConf(file, cfg.Node.Interface)
  233. if err != nil {
  234. ncutils.Log("error restarting wg after node update " + err.Error())
  235. return
  236. }
  237. time.Sleep(time.Second >> 1)
  238. if err = Resubscribe(client, &cfg); err != nil {
  239. ncutils.Log("error resubscribing after interface change " + err.Error())
  240. return
  241. }
  242. }
  243. /*
  244. else {
  245. ncutils.Log("syncing conf to " + file)
  246. err = wireguard.SyncWGQuickConf(cfg.Node.Interface, file)
  247. if err != nil {
  248. ncutils.Log("error syncing wg after peer update " + err.Error())
  249. return
  250. }
  251. }
  252. */
  253. //deal with DNS
  254. if newNode.DNSOn == "yes" {
  255. ncutils.Log("setting up DNS")
  256. if err = local.UpdateDNS(cfg.Node.Interface, cfg.Network, cfg.Server.CoreDNSAddr); err != nil {
  257. ncutils.Log("error applying dns" + err.Error())
  258. }
  259. } else {
  260. ncutils.Log("settng DNS off")
  261. _, err := ncutils.RunCmd("/usr/bin/resolvectl revert "+cfg.Node.Interface, true)
  262. if err != nil {
  263. ncutils.Log("error applying dns" + err.Error())
  264. }
  265. }
  266. }()
  267. }
  268. // UpdatePeers -- mqtt message handler for peers/<Network>/<NodeID> topic
  269. func UpdatePeers(client mqtt.Client, msg mqtt.Message) {
  270. go func() {
  271. var peerUpdate models.PeerUpdate
  272. var network = parseNetworkFromTopic(msg.Topic())
  273. var cfg = config.ClientConfig{}
  274. cfg.Network = network
  275. cfg.ReadConfig()
  276. data, dataErr := decryptMsg(&cfg, msg.Payload())
  277. if dataErr != nil {
  278. return
  279. }
  280. err := json.Unmarshal([]byte(data), &peerUpdate)
  281. if err != nil {
  282. ncutils.Log("error unmarshalling peer data")
  283. return
  284. }
  285. // see if cache hit, if so skip
  286. var currentMessage = read(peerUpdate.Network, lastPeerUpdate)
  287. if currentMessage == string(data) {
  288. return
  289. }
  290. insert(peerUpdate.Network, lastPeerUpdate, string(data))
  291. ncutils.Log("update peer handler")
  292. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  293. err = wireguard.UpdateWgPeers(file, peerUpdate.Peers)
  294. if err != nil {
  295. ncutils.Log("error updating wireguard peers" + err.Error())
  296. return
  297. }
  298. ncutils.Log("syncing conf to " + file)
  299. //err = wireguard.SyncWGQuickConf(cfg.Node.Interface, file)
  300. err = wireguard.SetPeers(cfg.Node.Interface, cfg.Node.PersistentKeepalive, peerUpdate.Peers)
  301. if err != nil {
  302. ncutils.Log("error syncing wg after peer update " + err.Error())
  303. return
  304. }
  305. }()
  306. }
  307. // MonitorKeepalive - checks time last server keepalive received. If more than 3+ minutes, notify and resubscribe
  308. func MonitorKeepalive(ctx context.Context, client mqtt.Client, cfg *config.ClientConfig) {
  309. var id string
  310. for _, servAddr := range cfg.NetworkSettings.DefaultServerAddrs {
  311. if servAddr.IsLeader {
  312. id = servAddr.ID
  313. }
  314. }
  315. for {
  316. select {
  317. case <-ctx.Done():
  318. return
  319. case <-time.After(time.Second * 150):
  320. if time.Since(keepalive[id]) > time.Second*200 { // more than 3+ minutes
  321. ncutils.Log("server keepalive not recieved in more than minutes, resubscribe to message queue")
  322. err := Resubscribe(client, cfg)
  323. if err != nil {
  324. ncutils.Log("closing " + err.Error())
  325. }
  326. }
  327. }
  328. }
  329. }
  330. // ServerKeepAlive -- handler to react to keepalive messages published by server
  331. func ServerKeepAlive(client mqtt.Client, msg mqtt.Message) {
  332. var mu sync.Mutex
  333. mu.Lock()
  334. defer mu.Unlock()
  335. serverid, err := getID(msg.Topic())
  336. if err != nil {
  337. ncutils.Log("invalid ID in serverkeepalive topic")
  338. }
  339. keepalive[serverid] = time.Now()
  340. // ncutils.Log("keepalive from server")
  341. }
  342. // Resubscribe --- handles resubscribing if needed
  343. func Resubscribe(client mqtt.Client, cfg *config.ClientConfig) error {
  344. if err := config.ModConfig(&cfg.Node); err == nil {
  345. ncutils.Log("resubbing on network " + cfg.Node.Network)
  346. client.Disconnect(250)
  347. client = SetupMQTT(cfg)
  348. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, NodeUpdate); token.Wait() && token.Error() != nil {
  349. ncutils.Log("error resubscribing to updates for " + cfg.Node.Network)
  350. return token.Error()
  351. }
  352. if cfg.DebugOn {
  353. ncutils.Log("subscribed to node updates for node " + cfg.Node.Name + " update/" + cfg.Node.ID)
  354. }
  355. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, UpdatePeers); token.Wait() && token.Error() != nil {
  356. ncutils.Log("error resubscribing to peers for " + cfg.Node.Network)
  357. return token.Error()
  358. }
  359. var id string
  360. for _, server := range cfg.NetworkSettings.DefaultServerAddrs {
  361. if server.IsLeader {
  362. id = server.ID
  363. }
  364. if server.Address != "" {
  365. if token := client.Subscribe("serverkeepalive/"+id, 0, mqtt.MessageHandler(ServerKeepAlive)); token.Wait() && token.Error() != nil {
  366. ncutils.Log("error resubscribing to serverkeepalive for " + cfg.Node.Network)
  367. return token.Error()
  368. }
  369. if cfg.DebugOn {
  370. ncutils.Log("subscribed to server keepalives for server " + id)
  371. }
  372. } else {
  373. log.Println(cfg.NetworkSettings.DefaultServerAddrs)
  374. ncutils.Log("leader not defined for network " + cfg.Network)
  375. }
  376. }
  377. ncutils.Log("finished re subbing")
  378. return nil
  379. } else {
  380. ncutils.Log("could not mod config when re-subbing")
  381. return err
  382. }
  383. }
  384. // UpdateKeys -- updates private key and returns new publickey
  385. func UpdateKeys(cfg *config.ClientConfig, client mqtt.Client) error {
  386. ncutils.Log("received message to update keys")
  387. //potentiall blocking i/o so do this in a go routine
  388. key, err := wgtypes.GeneratePrivateKey()
  389. if err != nil {
  390. ncutils.Log("error generating privatekey " + err.Error())
  391. return err
  392. }
  393. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  394. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  395. ncutils.Log("error updating wireguard key " + err.Error())
  396. return err
  397. }
  398. cfg.Node.PublicKey = key.PublicKey().String()
  399. PublishNodeUpdate(cfg)
  400. if err := config.ModConfig(&cfg.Node); err != nil {
  401. ncutils.Log("error updating local config " + err.Error())
  402. }
  403. return nil
  404. }
  405. // Checkin -- go routine that checks for public or local ip changes, publishes changes
  406. // if there are no updates, simply "pings" the server as a checkin
  407. func Checkin(ctx context.Context, cfg *config.ClientConfig, network string) {
  408. for {
  409. select {
  410. case <-ctx.Done():
  411. ncutils.Log("Checkin cancelled")
  412. return
  413. //delay should be configuraable -> use cfg.Node.NetworkSettings.DefaultCheckInInterval ??
  414. case <-time.After(time.Second * 60):
  415. // ncutils.Log("Checkin running")
  416. //read latest config
  417. cfg.ReadConfig()
  418. if cfg.Node.Roaming == "yes" && cfg.Node.IsStatic != "yes" {
  419. extIP, err := ncutils.GetPublicIP()
  420. if err != nil {
  421. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  422. }
  423. if cfg.Node.Endpoint != extIP && extIP != "" {
  424. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+extIP, 1)
  425. cfg.Node.Endpoint = extIP
  426. PublishNodeUpdate(cfg)
  427. }
  428. intIP, err := getPrivateAddr()
  429. if err != nil {
  430. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  431. }
  432. if cfg.Node.LocalAddress != intIP && intIP != "" {
  433. ncutils.PrintLog("local Address has changed from "+cfg.Node.LocalAddress+" to "+intIP, 1)
  434. cfg.Node.LocalAddress = intIP
  435. PublishNodeUpdate(cfg)
  436. }
  437. } else {
  438. localIP, err := ncutils.GetLocalIP(cfg.Node.LocalRange)
  439. if err != nil {
  440. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  441. }
  442. if cfg.Node.Endpoint != localIP && localIP != "" {
  443. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+localIP, 1)
  444. cfg.Node.Endpoint = localIP
  445. PublishNodeUpdate(cfg)
  446. }
  447. }
  448. Hello(cfg, network)
  449. // ncutils.Log("Checkin complete")
  450. }
  451. }
  452. }
  453. // PublishNodeUpdates -- saves node and pushes changes to broker
  454. func PublishNodeUpdate(cfg *config.ClientConfig) {
  455. if err := config.Write(cfg, cfg.Network); err != nil {
  456. ncutils.Log("error saving configuration: " + err.Error())
  457. }
  458. data, err := json.Marshal(cfg.Node)
  459. if err != nil {
  460. ncutils.Log("error marshling node update: " + err.Error())
  461. }
  462. if err = publish(cfg, fmt.Sprintf("update/%s", cfg.Node.ID), data); err != nil {
  463. ncutils.Log(fmt.Sprintf("error publishing endpoint update, %v", err))
  464. }
  465. }
  466. // Hello -- ping the broker to let server know node is alive and doing fine
  467. func Hello(cfg *config.ClientConfig, network string) {
  468. if err := publish(cfg, fmt.Sprintf("ping/%s", cfg.Node.ID), []byte("hello world!")); err != nil {
  469. ncutils.Log(fmt.Sprintf("error publishing ping, %v", err))
  470. }
  471. }
  472. func publish(cfg *config.ClientConfig, dest string, msg []byte) error {
  473. // setup the keys
  474. trafficPrivKey, err := auth.RetrieveTrafficKey(cfg.Node.Network)
  475. if err != nil {
  476. return err
  477. }
  478. serverPubKey, err := ncutils.ConvertBytesToKey(cfg.Node.TrafficKeys.Server)
  479. if err != nil {
  480. return err
  481. }
  482. client := SetupMQTT(cfg)
  483. defer client.Disconnect(250)
  484. encrypted, err := ncutils.BoxEncrypt(msg, serverPubKey, trafficPrivKey)
  485. if err != nil {
  486. return err
  487. }
  488. if token := client.Publish(dest, 0, false, encrypted); token.Wait() && token.Error() != nil {
  489. return token.Error()
  490. }
  491. return nil
  492. }
  493. func parseNetworkFromTopic(topic string) string {
  494. return strings.Split(topic, "/")[1]
  495. }
  496. func decryptMsg(cfg *config.ClientConfig, msg []byte) ([]byte, error) {
  497. // setup the keys
  498. diskKey, keyErr := auth.RetrieveTrafficKey(cfg.Node.Network)
  499. if keyErr != nil {
  500. return nil, keyErr
  501. }
  502. serverPubKey, err := ncutils.ConvertBytesToKey(cfg.Node.TrafficKeys.Server)
  503. if err != nil {
  504. return nil, err
  505. }
  506. return ncutils.BoxDecrypt(msg, serverPubKey, diskKey)
  507. }
  508. func shouldResub(currentServers, newServers []models.ServerAddr) bool {
  509. if len(currentServers) != len(newServers) {
  510. return true
  511. }
  512. for _, srv := range currentServers {
  513. if !ncutils.ServerAddrSliceContains(newServers, srv) {
  514. return true
  515. }
  516. }
  517. return false
  518. }
  519. func getID(topic string) (string, error) {
  520. parts := strings.Split(topic, "/")
  521. count := len(parts)
  522. if count == 1 {
  523. return "", errors.New("invalid topic")
  524. }
  525. //the last part of the topic will be the network.ID
  526. return parts[count-1], nil
  527. }