daemon.go 17 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554
  1. package functions
  2. import (
  3. "context"
  4. "encoding/json"
  5. "errors"
  6. "fmt"
  7. "os"
  8. "os/signal"
  9. "runtime"
  10. "strings"
  11. "sync"
  12. "syscall"
  13. "time"
  14. mqtt "github.com/eclipse/paho.mqtt.golang"
  15. "github.com/gravitl/netmaker/models"
  16. "github.com/gravitl/netmaker/netclient/auth"
  17. "github.com/gravitl/netmaker/netclient/config"
  18. "github.com/gravitl/netmaker/netclient/local"
  19. "github.com/gravitl/netmaker/netclient/ncutils"
  20. "github.com/gravitl/netmaker/netclient/wireguard"
  21. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  22. )
  23. // ServerKeepalive - stores time of last server keepalive message
  24. var keepalive = make(map[string]time.Time, 3)
  25. var messageCache = make(map[string]string, 20)
  26. const lastNodeUpdate = "lnu"
  27. const lastPeerUpdate = "lpu"
  28. func insert(network, which, cache string) {
  29. var mu sync.Mutex
  30. mu.Lock()
  31. defer mu.Unlock()
  32. messageCache[fmt.Sprintf("%s%s", network, which)] = cache
  33. }
  34. func read(network, which string) string {
  35. return messageCache[fmt.Sprintf("%s%s", network, which)]
  36. }
  37. // Daemon runs netclient daemon from command line
  38. func Daemon() error {
  39. ctx, cancel := context.WithCancel(context.Background())
  40. networks, err := ncutils.GetSystemNetworks()
  41. if err != nil {
  42. cancel()
  43. return err
  44. }
  45. for _, network := range networks {
  46. go MessageQueue(ctx, network)
  47. }
  48. quit := make(chan os.Signal, 1)
  49. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  50. <-quit
  51. cancel()
  52. ncutils.Log("all done")
  53. return nil
  54. }
  55. // SetupMQTT creates a connection to broker and return client
  56. func SetupMQTT(cfg *config.ClientConfig) mqtt.Client {
  57. opts := mqtt.NewClientOptions()
  58. for _, server := range cfg.Node.NetworkSettings.DefaultServerAddrs {
  59. if server.Address != "" && server.IsLeader {
  60. ncutils.Log(fmt.Sprintf("adding server (%s) to listen on network %s", server.Address, cfg.Node.Network))
  61. opts.AddBroker(server.Address + ":1883")
  62. break
  63. }
  64. }
  65. opts.SetDefaultPublishHandler(All)
  66. client := mqtt.NewClient(opts)
  67. tperiod := time.Now().Add(12 * time.Second)
  68. for {
  69. //if after 12 seconds, try a gRPC pull on the last try
  70. if time.Now().After(tperiod) {
  71. ncutils.Log("running pull for " + cfg.Node.Network)
  72. _, err := Pull(cfg.Node.Network, true)
  73. if err != nil {
  74. ncutils.Log("could not run pull, exiting " + cfg.Node.Network + " setup: " + err.Error())
  75. return client
  76. }
  77. time.Sleep(2 * time.Second)
  78. }
  79. if token := client.Connect(); token.Wait() && token.Error() != nil {
  80. ncutils.Log("unable to connect to broker, retrying ...")
  81. if time.Now().After(tperiod) {
  82. ncutils.Log("could not connect to broker, exiting " + cfg.Node.Network + " setup: " + token.Error().Error())
  83. return client
  84. }
  85. } else {
  86. break
  87. }
  88. time.Sleep(2 * time.Second)
  89. }
  90. return client
  91. }
  92. // MessageQueue sets up Message Queue and subsribes/publishes updates to/from server
  93. func MessageQueue(ctx context.Context, network string) {
  94. ncutils.Log("netclient go routine started for " + network)
  95. var cfg config.ClientConfig
  96. cfg.Network = network
  97. cfg.ReadConfig()
  98. ncutils.Log("pulling latest config for " + cfg.Network)
  99. _, err := Pull(cfg.Network, true)
  100. if err != nil {
  101. ncutils.Log(err.Error())
  102. return
  103. }
  104. time.Sleep(2 * time.Second)
  105. ncutils.Log("daemon started for network: " + network)
  106. client := SetupMQTT(&cfg)
  107. if cfg.DebugOn {
  108. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  109. ncutils.Log(token.Error().Error())
  110. return
  111. }
  112. ncutils.Log("subscribed to all topics for debugging purposes")
  113. }
  114. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  115. ncutils.Log(token.Error().Error())
  116. return
  117. }
  118. if cfg.DebugOn {
  119. ncutils.Log(fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", cfg.Node.Name, cfg.Node.Network, cfg.Node.ID))
  120. }
  121. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  122. ncutils.Log(token.Error().Error())
  123. return
  124. }
  125. if cfg.DebugOn {
  126. ncutils.Log(fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", cfg.Node.Name, cfg.Node.Network, cfg.Node.ID))
  127. }
  128. var id string
  129. for _, server := range cfg.NetworkSettings.DefaultServerAddrs {
  130. if server.IsLeader {
  131. id = server.ID
  132. }
  133. if server.Address != "" {
  134. if token := client.Subscribe("serverkeepalive/"+id, 0, mqtt.MessageHandler(ServerKeepAlive)); token.Wait() && token.Error() != nil {
  135. ncutils.Log(token.Error().Error())
  136. return
  137. }
  138. if cfg.DebugOn {
  139. ncutils.Log("subscribed to server keepalives for server " + id)
  140. }
  141. } else {
  142. ncutils.Log("leader not defined for network" + cfg.Network)
  143. }
  144. }
  145. defer client.Disconnect(250)
  146. go MonitorKeepalive(ctx, client, &cfg)
  147. go Checkin(ctx, &cfg, network)
  148. <-ctx.Done()
  149. ncutils.Log("shutting down daemon")
  150. }
  151. // All -- mqtt message hander for all ('#') topics
  152. var All mqtt.MessageHandler = func(client mqtt.Client, msg mqtt.Message) {
  153. ncutils.Log("default message handler -- received message but not handling")
  154. ncutils.Log("Topic: " + string(msg.Topic()))
  155. //ncutils.Log("Message: " + string(msg.Payload()))
  156. }
  157. // NodeUpdate -- mqtt message handler for /update/<NodeID> topic
  158. func NodeUpdate(client mqtt.Client, msg mqtt.Message) {
  159. //potentiall blocking i/o so do this in a go routine
  160. go func() {
  161. var newNode models.Node
  162. var cfg config.ClientConfig
  163. var network = parseNetworkFromTopic(msg.Topic())
  164. cfg.Network = network
  165. cfg.ReadConfig()
  166. data, dataErr := decryptMsg(&cfg, msg.Payload())
  167. if dataErr != nil {
  168. return
  169. }
  170. err := json.Unmarshal([]byte(data), &newNode)
  171. if err != nil {
  172. ncutils.Log("error unmarshalling node update data" + err.Error())
  173. return
  174. }
  175. ncutils.Log("received message to update node " + newNode.Name)
  176. // see if cache hit, if so skip
  177. var currentMessage = read(newNode.Network, lastNodeUpdate)
  178. if currentMessage == string(data) {
  179. return
  180. }
  181. insert(newNode.Network, lastNodeUpdate, string(data))
  182. //check if interface name has changed if so delete.
  183. if cfg.Node.Interface != newNode.Interface {
  184. if err = wireguard.RemoveConf(cfg.Node.Interface, true); err != nil {
  185. ncutils.PrintLog("could not delete old interface "+cfg.Node.Interface+": "+err.Error(), 1)
  186. }
  187. }
  188. newNode.PullChanges = "no"
  189. //ensure that OS never changes
  190. newNode.OS = runtime.GOOS
  191. // check if interface needs to delta
  192. ifaceDelta := ncutils.IfaceDelta(&cfg.Node, &newNode)
  193. cfg.Node = newNode
  194. switch newNode.Action {
  195. case models.NODE_DELETE:
  196. if err := RemoveLocalInstance(&cfg, cfg.Network); err != nil {
  197. ncutils.PrintLog("error deleting local instance: "+err.Error(), 1)
  198. return
  199. }
  200. 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 {
  201. ncutils.PrintLog("error unsubscribing during node deletion", 1)
  202. }
  203. return
  204. case models.NODE_UPDATE_KEY:
  205. if err := UpdateKeys(&cfg, client); err != nil {
  206. ncutils.PrintLog("err updating wireguard keys: "+err.Error(), 1)
  207. }
  208. case models.NODE_NOOP:
  209. default:
  210. }
  211. //Save new config
  212. if err := config.Write(&cfg, cfg.Network); err != nil {
  213. ncutils.PrintLog("error updating node configuration: "+err.Error(), 1)
  214. }
  215. nameserver := cfg.Server.CoreDNSAddr
  216. privateKey, err := wireguard.RetrievePrivKey(newNode.Network)
  217. if err != nil {
  218. ncutils.Log("error reading PrivateKey " + err.Error())
  219. return
  220. }
  221. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  222. if err := wireguard.UpdateWgInterface(file, privateKey, nameserver, newNode); err != nil {
  223. ncutils.Log("error updating wireguard config " + err.Error())
  224. return
  225. }
  226. if ifaceDelta {
  227. ncutils.Log("applying WG conf to " + file)
  228. err = wireguard.ApplyWGQuickConf(file)
  229. if err != nil {
  230. ncutils.Log("error restarting wg after node update " + err.Error())
  231. return
  232. }
  233. time.Sleep(time.Second >> 1)
  234. if err = Resubscribe(client, &cfg); err != nil {
  235. ncutils.Log("error resubscribing after interface change " + err.Error())
  236. return
  237. }
  238. }
  239. /*
  240. else {
  241. ncutils.Log("syncing conf to " + file)
  242. err = wireguard.SyncWGQuickConf(cfg.Node.Interface, file)
  243. if err != nil {
  244. ncutils.Log("error syncing wg after peer update " + err.Error())
  245. return
  246. }
  247. }
  248. */
  249. //deal with DNS
  250. if newNode.DNSOn == "yes" {
  251. ncutils.Log("setting up DNS")
  252. if err = local.UpdateDNS(cfg.Node.Interface, cfg.Network, cfg.Server.CoreDNSAddr); err != nil {
  253. ncutils.Log("error applying dns" + err.Error())
  254. }
  255. } else {
  256. ncutils.Log("settng DNS off")
  257. _, err := ncutils.RunCmd("/usr/bin/resolvectl revert "+cfg.Node.Interface, true)
  258. if err != nil {
  259. ncutils.Log("error applying dns" + err.Error())
  260. }
  261. }
  262. }()
  263. }
  264. // UpdatePeers -- mqtt message handler for peers/<Network>/<NodeID> topic
  265. func UpdatePeers(client mqtt.Client, msg mqtt.Message) {
  266. go func() {
  267. var peerUpdate models.PeerUpdate
  268. var network = parseNetworkFromTopic(msg.Topic())
  269. var cfg = config.ClientConfig{}
  270. cfg.Network = network
  271. cfg.ReadConfig()
  272. data, dataErr := decryptMsg(&cfg, msg.Payload())
  273. if dataErr != nil {
  274. return
  275. }
  276. err := json.Unmarshal([]byte(data), &peerUpdate)
  277. if err != nil {
  278. ncutils.Log("error unmarshalling peer data")
  279. return
  280. }
  281. // see if cache hit, if so skip
  282. var currentMessage = read(peerUpdate.Network, lastPeerUpdate)
  283. if currentMessage == string(data) {
  284. return
  285. }
  286. insert(peerUpdate.Network, lastPeerUpdate, string(data))
  287. ncutils.Log("update peer handler")
  288. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  289. err = wireguard.UpdateWgPeers(file, peerUpdate.Peers)
  290. if err != nil {
  291. ncutils.Log("error updating wireguard peers" + err.Error())
  292. return
  293. }
  294. ncutils.Log("syncing conf to " + file)
  295. //err = wireguard.SyncWGQuickConf(cfg.Node.Interface, file)
  296. err = wireguard.SetPeers(cfg.Node.Interface, cfg.Node.PersistentKeepalive, peerUpdate.Peers)
  297. if err != nil {
  298. ncutils.Log("error syncing wg after peer update " + err.Error())
  299. return
  300. }
  301. }()
  302. }
  303. // MonitorKeepalive - checks time last server keepalive received. If more than 3+ minutes, notify and resubscribe
  304. func MonitorKeepalive(ctx context.Context, client mqtt.Client, cfg *config.ClientConfig) {
  305. var id string
  306. for _, servAddr := range cfg.NetworkSettings.DefaultServerAddrs {
  307. if servAddr.IsLeader {
  308. id = servAddr.ID
  309. }
  310. }
  311. for {
  312. select {
  313. case <-ctx.Done():
  314. return
  315. case <-time.After(time.Second * 150):
  316. if time.Since(keepalive[id]) > time.Second*200 { // more than 3+ minutes
  317. ncutils.Log("server keepalive not recieved in more than minutes, resubscribe to message queue")
  318. err := Resubscribe(client, cfg)
  319. ncutils.Log("closing " + err.Error())
  320. }
  321. }
  322. }
  323. }
  324. // ServerKeepAlive -- handler to react to keepalive messages published by server
  325. func ServerKeepAlive(client mqtt.Client, msg mqtt.Message) {
  326. var mu sync.Mutex
  327. mu.Lock()
  328. defer mu.Unlock()
  329. serverid, err := getID(msg.Topic())
  330. if err != nil {
  331. ncutils.Log("invalid ID in serverkeepalive topic")
  332. }
  333. keepalive[serverid] = time.Now()
  334. // ncutils.Log("keepalive from server")
  335. }
  336. // Resubscribe --- handles resubscribing if needed
  337. func Resubscribe(client mqtt.Client, cfg *config.ClientConfig) error {
  338. if err := config.ModConfig(&cfg.Node); err == nil {
  339. ncutils.Log("resubbing on network " + cfg.Node.Network)
  340. client.Disconnect(250)
  341. client = SetupMQTT(cfg)
  342. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, NodeUpdate); token.Wait() && token.Error() != nil {
  343. ncutils.Log("error resubscribing to updates for " + cfg.Node.Network)
  344. return token.Error()
  345. }
  346. if cfg.DebugOn {
  347. ncutils.Log("subscribed to node updates for node " + cfg.Node.Name + " update/" + cfg.Node.ID)
  348. }
  349. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, UpdatePeers); token.Wait() && token.Error() != nil {
  350. ncutils.Log("error resubscribing to peers for " + cfg.Node.Network)
  351. return token.Error()
  352. }
  353. var id string
  354. for _, server := range cfg.NetworkSettings.DefaultServerAddrs {
  355. if server.IsLeader {
  356. id = server.ID
  357. }
  358. if server.Address != "" {
  359. if token := client.Subscribe("serverkeepalive/"+id, 0, mqtt.MessageHandler(ServerKeepAlive)); token.Wait() && token.Error() != nil {
  360. ncutils.Log("error resubscribing to serverkeepalive for " + cfg.Node.Network)
  361. return token.Error()
  362. }
  363. if cfg.DebugOn {
  364. ncutils.Log("subscribed to server keepalives for server " + id)
  365. }
  366. } else {
  367. ncutils.Log("leader not defined for network" + cfg.Network)
  368. }
  369. }
  370. ncutils.Log("finished re subbing")
  371. return nil
  372. } else {
  373. ncutils.Log("could not mod config when re-subbing")
  374. return err
  375. }
  376. }
  377. // UpdateKeys -- updates private key and returns new publickey
  378. func UpdateKeys(cfg *config.ClientConfig, client mqtt.Client) error {
  379. ncutils.Log("received message to update keys")
  380. //potentiall blocking i/o so do this in a go routine
  381. key, err := wgtypes.GeneratePrivateKey()
  382. if err != nil {
  383. ncutils.Log("error generating privatekey " + err.Error())
  384. return err
  385. }
  386. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  387. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  388. ncutils.Log("error updating wireguard key " + err.Error())
  389. return err
  390. }
  391. cfg.Node.PublicKey = key.PublicKey().String()
  392. PublishNodeUpdate(cfg)
  393. if err := config.ModConfig(&cfg.Node); err != nil {
  394. ncutils.Log("error updating local config " + err.Error())
  395. }
  396. return nil
  397. }
  398. // Checkin -- go routine that checks for public or local ip changes, publishes changes
  399. // if there are no updates, simply "pings" the server as a checkin
  400. func Checkin(ctx context.Context, cfg *config.ClientConfig, network string) {
  401. for {
  402. select {
  403. case <-ctx.Done():
  404. ncutils.Log("Checkin cancelled")
  405. return
  406. //delay should be configuraable -> use cfg.Node.NetworkSettings.DefaultCheckInInterval ??
  407. case <-time.After(time.Second * 60):
  408. // ncutils.Log("Checkin running")
  409. //read latest config
  410. cfg.ReadConfig()
  411. if cfg.Node.Roaming == "yes" && cfg.Node.IsStatic != "yes" {
  412. extIP, err := ncutils.GetPublicIP()
  413. if err != nil {
  414. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  415. }
  416. if cfg.Node.Endpoint != extIP && extIP != "" {
  417. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+extIP, 1)
  418. cfg.Node.Endpoint = extIP
  419. PublishNodeUpdate(cfg)
  420. }
  421. intIP, err := getPrivateAddr()
  422. if err != nil {
  423. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  424. }
  425. if cfg.Node.LocalAddress != intIP && intIP != "" {
  426. ncutils.PrintLog("local Address has changed from "+cfg.Node.LocalAddress+" to "+intIP, 1)
  427. cfg.Node.LocalAddress = intIP
  428. PublishNodeUpdate(cfg)
  429. }
  430. } else {
  431. localIP, err := ncutils.GetLocalIP(cfg.Node.LocalRange)
  432. if err != nil {
  433. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  434. }
  435. if cfg.Node.Endpoint != localIP && localIP != "" {
  436. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+localIP, 1)
  437. cfg.Node.Endpoint = localIP
  438. PublishNodeUpdate(cfg)
  439. }
  440. }
  441. Hello(cfg, network)
  442. // ncutils.Log("Checkin complete")
  443. }
  444. }
  445. }
  446. // PublishNodeUpdates -- saves node and pushes changes to broker
  447. func PublishNodeUpdate(cfg *config.ClientConfig) {
  448. if err := config.Write(cfg, cfg.Network); err != nil {
  449. ncutils.Log("error saving configuration" + err.Error())
  450. }
  451. data, err := json.Marshal(cfg.Node)
  452. if err != nil {
  453. ncutils.Log("error marshling node update " + err.Error())
  454. }
  455. if err = publish(cfg, fmt.Sprintf("update/%s", cfg.Node.ID), data); err != nil {
  456. ncutils.Log(fmt.Sprintf("error publishing endpoint update, %v", err))
  457. }
  458. }
  459. // Hello -- ping the broker to let server know node is alive and doing fine
  460. func Hello(cfg *config.ClientConfig, network string) {
  461. if err := publish(cfg, fmt.Sprintf("ping/%s", cfg.Node.ID), []byte("hello world!")); err != nil {
  462. ncutils.Log(fmt.Sprintf("error publishing ping, %v", err))
  463. }
  464. }
  465. func publish(cfg *config.ClientConfig, dest string, msg []byte) error {
  466. // setup the keys
  467. trafficPrivKey, err := auth.RetrieveTrafficKey(cfg.Node.Network)
  468. if err != nil {
  469. return err
  470. }
  471. serverPubKey, err := ncutils.ConvertBytesToKey(cfg.Node.TrafficKeys.Server)
  472. if err != nil {
  473. return err
  474. }
  475. client := SetupMQTT(cfg)
  476. defer client.Disconnect(250)
  477. encrypted, err := ncutils.BoxEncrypt(msg, serverPubKey, trafficPrivKey)
  478. if err != nil {
  479. return err
  480. }
  481. if token := client.Publish(dest, 0, false, encrypted); token.Wait() && token.Error() != nil {
  482. return token.Error()
  483. }
  484. return nil
  485. }
  486. func parseNetworkFromTopic(topic string) string {
  487. return strings.Split(topic, "/")[1]
  488. }
  489. func decryptMsg(cfg *config.ClientConfig, msg []byte) ([]byte, error) {
  490. // setup the keys
  491. diskKey, keyErr := auth.RetrieveTrafficKey(cfg.Node.Network)
  492. if keyErr != nil {
  493. return nil, keyErr
  494. }
  495. serverPubKey, err := ncutils.ConvertBytesToKey(cfg.Node.TrafficKeys.Server)
  496. if err != nil {
  497. return nil, err
  498. }
  499. return ncutils.BoxDecrypt(msg, serverPubKey, diskKey)
  500. }
  501. func shouldResub(currentServers, newServers []models.ServerAddr) bool {
  502. if len(currentServers) != len(newServers) {
  503. return true
  504. }
  505. for _, srv := range currentServers {
  506. if !ncutils.ServerAddrSliceContains(newServers, srv) {
  507. return true
  508. }
  509. }
  510. return false
  511. }
  512. func getID(topic string) (string, error) {
  513. parts := strings.Split(topic, "/")
  514. count := len(parts)
  515. if count == 1 {
  516. return "", errors.New("invalid topic")
  517. }
  518. //the last part of the topic will be the network.ID
  519. return parts[count-1], nil
  520. }