daemon.go 17 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544
  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. log.Fatal("[netclient] could not run pull, exiting ...", err.Error())
  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. log.Fatal("[netclient] 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("pulling latest config for " + cfg.Network)
  98. _, err := Pull(cfg.Network, true)
  99. if err != nil {
  100. log.Fatal("[netclient] ", err.Error())
  101. }
  102. time.Sleep(2 * time.Second)
  103. ncutils.Log("daemon started for network: " + network)
  104. client := SetupMQTT(&cfg)
  105. if cfg.DebugOn {
  106. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  107. log.Fatal("[netclient] ", token.Error())
  108. }
  109. ncutils.Log("subscribed to all topics for debugging purposes")
  110. }
  111. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  112. log.Fatal("[netclient] ", token.Error())
  113. }
  114. if cfg.DebugOn {
  115. ncutils.Log(fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", cfg.Node.Name, cfg.Node.Network, cfg.Node.ID))
  116. }
  117. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  118. log.Fatal("[netclient] ", token.Error())
  119. }
  120. if cfg.DebugOn {
  121. ncutils.Log(fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", cfg.Node.Name, cfg.Node.Network, cfg.Node.ID))
  122. }
  123. var id string
  124. for _, server := range cfg.NetworkSettings.DefaultServerAddrs {
  125. if server.IsLeader {
  126. id = server.ID
  127. }
  128. if server.Address != "" {
  129. if token := client.Subscribe("serverkeepalive/"+id, 0, mqtt.MessageHandler(ServerKeepAlive)); token.Wait() && token.Error() != nil {
  130. log.Fatal("[netclient] ", token.Error())
  131. }
  132. if cfg.DebugOn {
  133. ncutils.Log("subscribed to server keepalives for server " + id)
  134. }
  135. } else {
  136. ncutils.Log("leader not defined for network" + cfg.Network)
  137. }
  138. }
  139. defer client.Disconnect(250)
  140. go MonitorKeepalive(ctx, client, &cfg)
  141. go Checkin(ctx, &cfg, network)
  142. <-ctx.Done()
  143. ncutils.Log("shutting down daemon")
  144. }
  145. // All -- mqtt message hander for all ('#') topics
  146. var All mqtt.MessageHandler = func(client mqtt.Client, msg mqtt.Message) {
  147. ncutils.Log("default message handler -- received message but not handling")
  148. ncutils.Log("Topic: " + string(msg.Topic()))
  149. //ncutils.Log("Message: " + string(msg.Payload()))
  150. }
  151. // NodeUpdate -- mqtt message handler for /update/<NodeID> topic
  152. func NodeUpdate(client mqtt.Client, msg mqtt.Message) {
  153. //potentiall blocking i/o so do this in a go routine
  154. go func() {
  155. var newNode models.Node
  156. var cfg config.ClientConfig
  157. var network = parseNetworkFromTopic(msg.Topic())
  158. cfg.Network = network
  159. cfg.ReadConfig()
  160. data, dataErr := decryptMsg(&cfg, msg.Payload())
  161. if dataErr != nil {
  162. return
  163. }
  164. err := json.Unmarshal([]byte(data), &newNode)
  165. if err != nil {
  166. ncutils.Log("error unmarshalling node update data" + err.Error())
  167. return
  168. }
  169. ncutils.Log("received message to update node " + newNode.Name)
  170. // see if cache hit, if so skip
  171. var currentMessage = read(newNode.Network, lastNodeUpdate)
  172. if currentMessage == string(data) {
  173. return
  174. }
  175. insert(newNode.Network, lastNodeUpdate, string(data))
  176. //check if interface name has changed if so delete.
  177. if cfg.Node.Interface != newNode.Interface {
  178. if err = wireguard.RemoveConf(cfg.Node.Interface, true); err != nil {
  179. ncutils.PrintLog("could not delete old interface "+cfg.Node.Interface+": "+err.Error(), 1)
  180. }
  181. }
  182. newNode.PullChanges = "no"
  183. //ensure that OS never changes
  184. newNode.OS = runtime.GOOS
  185. // check if interface needs to delta
  186. ifaceDelta := ncutils.IfaceDelta(&cfg.Node, &newNode)
  187. cfg.Node = newNode
  188. switch newNode.Action {
  189. case models.NODE_DELETE:
  190. if err := RemoveLocalInstance(&cfg, cfg.Network); err != nil {
  191. ncutils.PrintLog("error deleting local instance: "+err.Error(), 1)
  192. return
  193. }
  194. 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 {
  195. ncutils.PrintLog("error unsubscribing during node deletion", 1)
  196. }
  197. return
  198. case models.NODE_UPDATE_KEY:
  199. if err := UpdateKeys(&cfg, client); err != nil {
  200. ncutils.PrintLog("err updating wireguard keys: "+err.Error(), 1)
  201. }
  202. case models.NODE_NOOP:
  203. default:
  204. }
  205. //Save new config
  206. if err := config.Write(&cfg, cfg.Network); err != nil {
  207. ncutils.PrintLog("error updating node configuration: "+err.Error(), 1)
  208. }
  209. nameserver := cfg.Server.CoreDNSAddr
  210. privateKey, err := wireguard.RetrievePrivKey(newNode.Network)
  211. if err != nil {
  212. ncutils.Log("error reading PrivateKey " + err.Error())
  213. return
  214. }
  215. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  216. if err := wireguard.UpdateWgInterface(file, privateKey, nameserver, newNode); err != nil {
  217. ncutils.Log("error updating wireguard config " + err.Error())
  218. return
  219. }
  220. if ifaceDelta {
  221. ncutils.Log("applying WG conf to " + file)
  222. err = wireguard.ApplyWGQuickConf(file)
  223. if err != nil {
  224. ncutils.Log("error restarting wg after node update " + err.Error())
  225. return
  226. }
  227. time.Sleep(time.Second >> 1)
  228. if err = Resubscribe(client, &cfg); err != nil {
  229. ncutils.Log("error resubscribing after interface change " + err.Error())
  230. return
  231. }
  232. }
  233. /*
  234. else {
  235. ncutils.Log("syncing conf to " + file)
  236. err = wireguard.SyncWGQuickConf(cfg.Node.Interface, file)
  237. if err != nil {
  238. ncutils.Log("error syncing wg after peer update " + err.Error())
  239. return
  240. }
  241. }
  242. */
  243. //deal with DNS
  244. if newNode.DNSOn == "yes" {
  245. ncutils.Log("setting up DNS")
  246. if err = local.UpdateDNS(cfg.Node.Interface, cfg.Network, cfg.Server.CoreDNSAddr); err != nil {
  247. ncutils.Log("error applying dns" + err.Error())
  248. }
  249. } else {
  250. ncutils.Log("settng DNS off")
  251. _, err := ncutils.RunCmd("/usr/bin/resolvectl revert "+cfg.Node.Interface, true)
  252. if err != nil {
  253. ncutils.Log("error applying dns" + err.Error())
  254. }
  255. }
  256. }()
  257. }
  258. // UpdatePeers -- mqtt message handler for peers/<Network>/<NodeID> topic
  259. func UpdatePeers(client mqtt.Client, msg mqtt.Message) {
  260. go func() {
  261. var peerUpdate models.PeerUpdate
  262. var network = parseNetworkFromTopic(msg.Topic())
  263. var cfg = config.ClientConfig{}
  264. cfg.Network = network
  265. cfg.ReadConfig()
  266. data, dataErr := decryptMsg(&cfg, msg.Payload())
  267. if dataErr != nil {
  268. return
  269. }
  270. err := json.Unmarshal([]byte(data), &peerUpdate)
  271. if err != nil {
  272. ncutils.Log("error unmarshalling peer data")
  273. return
  274. }
  275. // see if cache hit, if so skip
  276. var currentMessage = read(peerUpdate.Network, lastPeerUpdate)
  277. if currentMessage == string(data) {
  278. return
  279. }
  280. insert(peerUpdate.Network, lastPeerUpdate, string(data))
  281. ncutils.Log("update peer handler")
  282. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  283. err = wireguard.UpdateWgPeers(file, peerUpdate.Peers)
  284. if err != nil {
  285. ncutils.Log("error updating wireguard peers" + err.Error())
  286. return
  287. }
  288. ncutils.Log("syncing conf to " + file)
  289. //err = wireguard.SyncWGQuickConf(cfg.Node.Interface, file)
  290. err = wireguard.SetPeers(cfg.Node.Interface, cfg.Node.PersistentKeepalive, peerUpdate.Peers)
  291. if err != nil {
  292. ncutils.Log("error syncing wg after peer update " + err.Error())
  293. return
  294. }
  295. }()
  296. }
  297. // MonitorKeepalive - checks time last server keepalive received. If more than 3+ minutes, notify and resubscribe
  298. func MonitorKeepalive(ctx context.Context, client mqtt.Client, cfg *config.ClientConfig) {
  299. var id string
  300. for _, servAddr := range cfg.NetworkSettings.DefaultServerAddrs {
  301. if servAddr.IsLeader {
  302. id = servAddr.ID
  303. }
  304. }
  305. for {
  306. select {
  307. case <-ctx.Done():
  308. return
  309. case <-time.After(time.Second * 150):
  310. if time.Since(keepalive[id]) > time.Second*200 { // more than 3+ minutes
  311. ncutils.Log("server keepalive not recieved in more than minutes, resubscribe to message queue")
  312. Resubscribe(client, cfg)
  313. }
  314. }
  315. }
  316. }
  317. // ServerKeepAlive -- handler to react to keepalive messages published by server
  318. func ServerKeepAlive(client mqtt.Client, msg mqtt.Message) {
  319. var mu sync.Mutex
  320. mu.Lock()
  321. defer mu.Unlock()
  322. serverid, err := getID(msg.Topic())
  323. if err != nil {
  324. ncutils.Log("invalid ID in serverkeepalive topic")
  325. }
  326. keepalive[serverid] = time.Now()
  327. // ncutils.Log("keepalive from server")
  328. }
  329. // Resubscribe --- handles resubscribing if needed
  330. func Resubscribe(client mqtt.Client, cfg *config.ClientConfig) error {
  331. if err := config.ModConfig(&cfg.Node); err == nil {
  332. ncutils.Log("resubbing on network " + cfg.Node.Network)
  333. client.Disconnect(250)
  334. client = SetupMQTT(cfg)
  335. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, NodeUpdate); token.Wait() && token.Error() != nil {
  336. log.Fatal("[netclient] ", token.Error())
  337. }
  338. if cfg.DebugOn {
  339. ncutils.Log("subscribed to node updates for node " + cfg.Node.Name + " update/" + cfg.Node.ID)
  340. }
  341. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, UpdatePeers); token.Wait() && token.Error() != nil {
  342. log.Fatal("[netclient] ", token.Error())
  343. }
  344. var id string
  345. for _, server := range cfg.NetworkSettings.DefaultServerAddrs {
  346. if server.IsLeader {
  347. id = server.ID
  348. }
  349. if server.Address != "" {
  350. if token := client.Subscribe("serverkeepalive/"+id, 0, mqtt.MessageHandler(ServerKeepAlive)); token.Wait() && token.Error() != nil {
  351. log.Fatal("[netclient] ", token.Error())
  352. }
  353. if cfg.DebugOn {
  354. ncutils.Log("subscribed to server keepalives for server " + id)
  355. }
  356. } else {
  357. ncutils.Log("leader not defined for network" + cfg.Network)
  358. }
  359. }
  360. ncutils.Log("finished re subbing")
  361. return nil
  362. } else {
  363. ncutils.Log("could not mod config when re-subbing")
  364. return err
  365. }
  366. }
  367. // UpdateKeys -- updates private key and returns new publickey
  368. func UpdateKeys(cfg *config.ClientConfig, client mqtt.Client) error {
  369. ncutils.Log("received message to update keys")
  370. //potentiall blocking i/o so do this in a go routine
  371. key, err := wgtypes.GeneratePrivateKey()
  372. if err != nil {
  373. ncutils.Log("error generating privatekey " + err.Error())
  374. return err
  375. }
  376. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  377. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  378. ncutils.Log("error updating wireguard key " + err.Error())
  379. return err
  380. }
  381. cfg.Node.PublicKey = key.PublicKey().String()
  382. PublishNodeUpdate(cfg)
  383. if err := config.ModConfig(&cfg.Node); err != nil {
  384. ncutils.Log("error updating local config " + err.Error())
  385. }
  386. return nil
  387. }
  388. // Checkin -- go routine that checks for public or local ip changes, publishes changes
  389. // if there are no updates, simply "pings" the server as a checkin
  390. func Checkin(ctx context.Context, cfg *config.ClientConfig, network string) {
  391. for {
  392. select {
  393. case <-ctx.Done():
  394. ncutils.Log("Checkin cancelled")
  395. return
  396. //delay should be configuraable -> use cfg.Node.NetworkSettings.DefaultCheckInInterval ??
  397. case <-time.After(time.Second * 60):
  398. // ncutils.Log("Checkin running")
  399. //read latest config
  400. cfg.ReadConfig()
  401. if cfg.Node.Roaming == "yes" && cfg.Node.IsStatic != "yes" {
  402. extIP, err := ncutils.GetPublicIP()
  403. if err != nil {
  404. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  405. }
  406. if cfg.Node.Endpoint != extIP && extIP != "" {
  407. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+extIP, 1)
  408. cfg.Node.Endpoint = extIP
  409. PublishNodeUpdate(cfg)
  410. }
  411. intIP, err := getPrivateAddr()
  412. if err != nil {
  413. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  414. }
  415. if cfg.Node.LocalAddress != intIP && intIP != "" {
  416. ncutils.PrintLog("local Address has changed from "+cfg.Node.LocalAddress+" to "+intIP, 1)
  417. cfg.Node.LocalAddress = intIP
  418. PublishNodeUpdate(cfg)
  419. }
  420. } else {
  421. localIP, err := ncutils.GetLocalIP(cfg.Node.LocalRange)
  422. if err != nil {
  423. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  424. }
  425. if cfg.Node.Endpoint != localIP && localIP != "" {
  426. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+localIP, 1)
  427. cfg.Node.Endpoint = localIP
  428. PublishNodeUpdate(cfg)
  429. }
  430. }
  431. Hello(cfg, network)
  432. // ncutils.Log("Checkin complete")
  433. }
  434. }
  435. }
  436. // PublishNodeUpdates -- saves node and pushes changes to broker
  437. func PublishNodeUpdate(cfg *config.ClientConfig) {
  438. if err := config.Write(cfg, cfg.Network); err != nil {
  439. ncutils.Log("error saving configuration" + err.Error())
  440. }
  441. data, err := json.Marshal(cfg.Node)
  442. if err != nil {
  443. ncutils.Log("error marshling node update " + err.Error())
  444. }
  445. if err = publish(cfg, fmt.Sprintf("update/%s", cfg.Node.ID), data); err != nil {
  446. ncutils.Log(fmt.Sprintf("error publishing endpoint update, %v", err))
  447. }
  448. }
  449. // Hello -- ping the broker to let server know node is alive and doing fine
  450. func Hello(cfg *config.ClientConfig, network string) {
  451. if err := publish(cfg, fmt.Sprintf("ping/%s", cfg.Node.ID), []byte("hello world!")); err != nil {
  452. ncutils.Log(fmt.Sprintf("error publishing ping, %v", err))
  453. }
  454. }
  455. func publish(cfg *config.ClientConfig, dest string, msg []byte) error {
  456. // setup the keys
  457. trafficPrivKey, err := auth.RetrieveTrafficKey(cfg.Node.Network)
  458. if err != nil {
  459. return err
  460. }
  461. serverPubKey, err := ncutils.ConvertBytesToKey(cfg.Node.TrafficKeys.Server)
  462. if err != nil {
  463. return err
  464. }
  465. client := SetupMQTT(cfg)
  466. defer client.Disconnect(250)
  467. encrypted, err := ncutils.BoxEncrypt(msg, serverPubKey, trafficPrivKey)
  468. if err != nil {
  469. return err
  470. }
  471. if token := client.Publish(dest, 0, false, encrypted); token.Wait() && token.Error() != nil {
  472. return token.Error()
  473. }
  474. return nil
  475. }
  476. func parseNetworkFromTopic(topic string) string {
  477. return strings.Split(topic, "/")[1]
  478. }
  479. func decryptMsg(cfg *config.ClientConfig, msg []byte) ([]byte, error) {
  480. // setup the keys
  481. diskKey, keyErr := auth.RetrieveTrafficKey(cfg.Node.Network)
  482. if keyErr != nil {
  483. return nil, keyErr
  484. }
  485. serverPubKey, err := ncutils.ConvertBytesToKey(cfg.Node.TrafficKeys.Server)
  486. if err != nil {
  487. return nil, err
  488. }
  489. return ncutils.BoxDecrypt(msg, serverPubKey, diskKey)
  490. }
  491. func shouldResub(currentServers, newServers []models.ServerAddr) bool {
  492. if len(currentServers) != len(newServers) {
  493. return true
  494. }
  495. for _, srv := range currentServers {
  496. if !ncutils.ServerAddrSliceContains(newServers, srv) {
  497. return true
  498. }
  499. }
  500. return false
  501. }
  502. func getID(topic string) (string, error) {
  503. parts := strings.Split(topic, "/")
  504. count := len(parts)
  505. if count == 1 {
  506. return "", errors.New("invalid topic")
  507. }
  508. //the last part of the topic will be the network.ID
  509. return parts[count-1], nil
  510. }