daemon.go 17 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533
  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(10 * time.Second)
  70. for {
  71. if token := client.Connect(); token.Wait() && token.Error() != nil {
  72. logger.Log(2, "unable to connect to broker, retrying ...")
  73. if time.Now().After(tperiod) {
  74. log.Fatal(0, "could not connect to broker, exiting ...", token.Error())
  75. }
  76. } else {
  77. break
  78. }
  79. time.Sleep(2 * time.Second)
  80. }
  81. return client
  82. }
  83. // MessageQueue sets up Message Queue and subsribes/publishes updates to/from server
  84. func MessageQueue(ctx context.Context, network string) {
  85. ncutils.Log("netclient go routine started for " + network)
  86. var cfg config.ClientConfig
  87. cfg.Network = network
  88. cfg.ReadConfig()
  89. ncutils.Log("daemon started for network:" + network)
  90. client := SetupMQTT(&cfg)
  91. if cfg.DebugOn {
  92. if token := client.Subscribe("#", 0, nil); token.Wait() && token.Error() != nil {
  93. log.Fatal(token.Error())
  94. }
  95. ncutils.Log("subscribed to all topics for debugging purposes")
  96. }
  97. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.Wait() && token.Error() != nil {
  98. log.Fatal(token.Error())
  99. }
  100. if cfg.DebugOn {
  101. ncutils.Log(fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", cfg.Node.Name, cfg.Node.Network, cfg.Node.ID))
  102. }
  103. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  104. log.Fatal(token.Error())
  105. }
  106. if cfg.DebugOn {
  107. ncutils.Log(fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", cfg.Node.Name, cfg.Node.Network, cfg.Node.ID))
  108. }
  109. var id string
  110. for _, server := range cfg.NetworkSettings.DefaultServerAddrs {
  111. if server.IsLeader {
  112. id = server.ID
  113. }
  114. if server.Address != "" {
  115. if token := client.Subscribe("serverkeepalive/"+id, 0, mqtt.MessageHandler(ServerKeepAlive)); token.Wait() && token.Error() != nil {
  116. log.Fatal(token.Error())
  117. }
  118. if cfg.DebugOn {
  119. ncutils.Log("subscribed to server keepalives for server " + id)
  120. }
  121. } else {
  122. ncutils.Log("leader not defined for network" + cfg.Network)
  123. }
  124. }
  125. defer client.Disconnect(250)
  126. go MonitorKeepalive(ctx, client, &cfg)
  127. go Checkin(ctx, &cfg, network)
  128. <-ctx.Done()
  129. ncutils.Log("shutting down daemon")
  130. }
  131. // All -- mqtt message hander for all ('#') topics
  132. var All mqtt.MessageHandler = func(client mqtt.Client, msg mqtt.Message) {
  133. ncutils.Log("default message handler -- received message but not handling")
  134. ncutils.Log("Topic: " + string(msg.Topic()))
  135. //ncutils.Log("Message: " + string(msg.Payload()))
  136. }
  137. // NodeUpdate -- mqtt message handler for /update/<NodeID> topic
  138. func NodeUpdate(client mqtt.Client, msg mqtt.Message) {
  139. //potentiall blocking i/o so do this in a go routine
  140. go func() {
  141. var newNode models.Node
  142. var cfg config.ClientConfig
  143. var network = parseNetworkFromTopic(msg.Topic())
  144. cfg.Network = network
  145. cfg.ReadConfig()
  146. data, dataErr := decryptMsg(&cfg, msg.Payload())
  147. if dataErr != nil {
  148. return
  149. }
  150. err := json.Unmarshal([]byte(data), &newNode)
  151. if err != nil {
  152. ncutils.Log("error unmarshalling node update data" + err.Error())
  153. return
  154. }
  155. log.Println("DELETE: new node addres: " + newNode.Address)
  156. log.Println("DELETE: current node addres: " + cfg.Node.Address)
  157. ncutils.Log("received message to update node " + newNode.Name)
  158. // see if cache hit, if so skip
  159. var currentMessage = read(newNode.Network, lastNodeUpdate)
  160. if currentMessage == string(data) {
  161. return
  162. }
  163. insert(newNode.Network, lastNodeUpdate, string(data))
  164. //check if interface name has changed if so delete.
  165. if cfg.Node.Interface != newNode.Interface {
  166. if err = wireguard.RemoveConf(cfg.Node.Interface, true); err != nil {
  167. ncutils.PrintLog("could not delete old interface "+cfg.Node.Interface+": "+err.Error(), 1)
  168. }
  169. }
  170. newNode.PullChanges = "no"
  171. //ensure that OS never changes
  172. newNode.OS = runtime.GOOS
  173. // check if interface needs to delta
  174. log.Println("DELETE: running ifacedelta")
  175. ifaceDelta := ncutils.IfaceDelta(&cfg.Node, &newNode)
  176. cfg.Node = newNode
  177. switch newNode.Action {
  178. case models.NODE_DELETE:
  179. if err := RemoveLocalInstance(&cfg, cfg.Network); err != nil {
  180. ncutils.PrintLog("error deleting local instance: "+err.Error(), 1)
  181. return
  182. }
  183. 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 {
  184. ncutils.PrintLog("error unsubscribing during node deletion", 1)
  185. }
  186. return
  187. case models.NODE_UPDATE_KEY:
  188. if err := UpdateKeys(&cfg, client); err != nil {
  189. ncutils.PrintLog("err updating wireguard keys: "+err.Error(), 1)
  190. }
  191. case models.NODE_NOOP:
  192. default:
  193. }
  194. //Save new config
  195. if err := config.Write(&cfg, cfg.Network); err != nil {
  196. ncutils.PrintLog("error updating node configuration: "+err.Error(), 1)
  197. }
  198. nameserver := cfg.Server.CoreDNSAddr
  199. privateKey, err := wireguard.RetrievePrivKey(newNode.Network)
  200. if err != nil {
  201. ncutils.Log("error reading PrivateKey " + err.Error())
  202. return
  203. }
  204. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  205. if err := wireguard.UpdateWgInterface(file, privateKey, nameserver, newNode); err != nil {
  206. ncutils.Log("error updating wireguard config " + err.Error())
  207. return
  208. }
  209. if ifaceDelta {
  210. ncutils.Log("applying WG conf to " + file)
  211. err = wireguard.ApplyWGQuickConf(file)
  212. if err != nil {
  213. ncutils.Log("error restarting wg after node update " + err.Error())
  214. return
  215. }
  216. log.Println("DELETE: sleeepy time")
  217. log.Println("DELETE: sleeepy time")
  218. log.Println("DELETE: sleeepy time")
  219. log.Println("DELETE: sleeepy time")
  220. time.Sleep(time.Second * 5)
  221. if err = Resubscribe(client, &cfg); err != nil {
  222. ncutils.Log("error resubscribing after interface change " + err.Error())
  223. return
  224. }
  225. } else {
  226. ncutils.Log("syncing conf to " + file)
  227. err = wireguard.SyncWGQuickConf(cfg.Node.Interface, file)
  228. if err != nil {
  229. ncutils.Log("error syncing wg after peer update " + err.Error())
  230. return
  231. }
  232. }
  233. //deal with DNS
  234. if newNode.DNSOn == "yes" {
  235. ncutils.Log("setting up DNS")
  236. if err = local.UpdateDNS(cfg.Node.Interface, cfg.Network, cfg.Server.CoreDNSAddr); err != nil {
  237. ncutils.Log("error applying dns" + err.Error())
  238. }
  239. } else {
  240. ncutils.Log("settng DNS off")
  241. _, err := ncutils.RunCmd("/usr/bin/resolvectl revert "+cfg.Node.Interface, true)
  242. if err != nil {
  243. ncutils.Log("error applying dns" + err.Error())
  244. }
  245. }
  246. }()
  247. }
  248. // UpdatePeers -- mqtt message handler for peers/<Network>/<NodeID> topic
  249. func UpdatePeers(client mqtt.Client, msg mqtt.Message) {
  250. go func() {
  251. var peerUpdate models.PeerUpdate
  252. var network = parseNetworkFromTopic(msg.Topic())
  253. var cfg = config.ClientConfig{}
  254. cfg.Network = network
  255. cfg.ReadConfig()
  256. data, dataErr := decryptMsg(&cfg, msg.Payload())
  257. if dataErr != nil {
  258. return
  259. }
  260. err := json.Unmarshal([]byte(data), &peerUpdate)
  261. if err != nil {
  262. ncutils.Log("error unmarshalling peer data")
  263. return
  264. }
  265. // see if cache hit, if so skip
  266. var currentMessage = read(peerUpdate.Network, lastPeerUpdate)
  267. if currentMessage == string(data) {
  268. return
  269. }
  270. insert(peerUpdate.Network, lastPeerUpdate, string(data))
  271. ncutils.Log("update peer handler")
  272. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  273. err = wireguard.UpdateWgPeers(file, peerUpdate.Peers)
  274. if err != nil {
  275. ncutils.Log("error updating wireguard peers" + err.Error())
  276. return
  277. }
  278. ncutils.Log("syncing conf to " + file)
  279. err = wireguard.SyncWGQuickConf(cfg.Node.Interface, file)
  280. if err != nil {
  281. ncutils.Log("error syncing wg after peer update " + err.Error())
  282. return
  283. }
  284. }()
  285. }
  286. // MonitorKeepalive - checks time last server keepalive received. If more than 3+ minutes, notify and resubscribe
  287. func MonitorKeepalive(ctx context.Context, client mqtt.Client, cfg *config.ClientConfig) {
  288. var id string
  289. for _, servAddr := range cfg.NetworkSettings.DefaultServerAddrs {
  290. if servAddr.IsLeader {
  291. id = servAddr.ID
  292. }
  293. }
  294. for {
  295. select {
  296. case <-ctx.Done():
  297. return
  298. case <-time.After(time.Second * 150):
  299. if time.Since(keepalive[id]) > time.Second*200 { // more than 3+ minutes
  300. ncutils.Log("server keepalive not recieved in more than minutes, resubscribe to message queue")
  301. Resubscribe(client, cfg)
  302. }
  303. }
  304. }
  305. }
  306. // ServerKeepAlive -- handler to react to keepalive messages published by server
  307. func ServerKeepAlive(client mqtt.Client, msg mqtt.Message) {
  308. serverid, err := getID(msg.Topic())
  309. if err != nil {
  310. ncutils.Log("invalid ID in serverkeepalive topic")
  311. }
  312. keepalive[serverid] = time.Now()
  313. ncutils.Log("keepalive from server")
  314. }
  315. // Resubscribe --- handles resubscribing if needed
  316. func Resubscribe(client mqtt.Client, cfg *config.ClientConfig) error {
  317. if err := config.ModConfig(&cfg.Node); err == nil {
  318. ncutils.Log("resubbing on network " + cfg.Node.Network)
  319. client.Disconnect(250)
  320. client = SetupMQTT(cfg)
  321. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, NodeUpdate); token.Wait() && token.Error() != nil {
  322. log.Fatal(token.Error())
  323. }
  324. if cfg.DebugOn {
  325. ncutils.Log("subscribed to node updates for node " + cfg.Node.Name + " update/" + cfg.Node.ID)
  326. }
  327. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", cfg.Node.Network, cfg.Node.ID), 0, UpdatePeers); token.Wait() && token.Error() != nil {
  328. log.Fatal(token.Error())
  329. }
  330. var id string
  331. for _, server := range cfg.NetworkSettings.DefaultServerAddrs {
  332. if server.IsLeader {
  333. id = server.ID
  334. }
  335. if server.Address != "" {
  336. if token := client.Subscribe("serverkeepalive/"+id, 0, mqtt.MessageHandler(ServerKeepAlive)); token.Wait() && token.Error() != nil {
  337. log.Fatal(token.Error())
  338. }
  339. if cfg.DebugOn {
  340. ncutils.Log("subscribed to server keepalives for server " + id)
  341. }
  342. } else {
  343. ncutils.Log("leader not defined for network" + cfg.Network)
  344. }
  345. }
  346. ncutils.Log("finished re subbing")
  347. return nil
  348. } else {
  349. ncutils.Log("could not mod config when re-subbing")
  350. return err
  351. }
  352. }
  353. // UpdateKeys -- updates private key and returns new publickey
  354. func UpdateKeys(cfg *config.ClientConfig, client mqtt.Client) error {
  355. ncutils.Log("received message to update keys")
  356. //potentiall blocking i/o so do this in a go routine
  357. key, err := wgtypes.GeneratePrivateKey()
  358. if err != nil {
  359. ncutils.Log("error generating privatekey " + err.Error())
  360. return err
  361. }
  362. file := ncutils.GetNetclientPathSpecific() + cfg.Node.Interface + ".conf"
  363. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  364. ncutils.Log("error updating wireguard key " + err.Error())
  365. return err
  366. }
  367. cfg.Node.PublicKey = key.PublicKey().String()
  368. PublishNodeUpdate(cfg)
  369. if err := config.ModConfig(&cfg.Node); err != nil {
  370. ncutils.Log("error updating local config " + err.Error())
  371. }
  372. return nil
  373. }
  374. // Checkin -- go routine that checks for public or local ip changes, publishes changes
  375. // if there are no updates, simply "pings" the server as a checkin
  376. func Checkin(ctx context.Context, cfg *config.ClientConfig, network string) {
  377. for {
  378. select {
  379. case <-ctx.Done():
  380. ncutils.Log("Checkin cancelled")
  381. return
  382. //delay should be configuraable -> use cfg.Node.NetworkSettings.DefaultCheckInInterval ??
  383. case <-time.After(time.Second * 60):
  384. ncutils.Log("Checkin running")
  385. //read latest config
  386. cfg.ReadConfig()
  387. if cfg.Node.Roaming == "yes" && cfg.Node.IsStatic != "yes" {
  388. extIP, err := ncutils.GetPublicIP()
  389. if err != nil {
  390. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  391. }
  392. if cfg.Node.Endpoint != extIP && extIP != "" {
  393. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+extIP, 1)
  394. cfg.Node.Endpoint = extIP
  395. PublishNodeUpdate(cfg)
  396. }
  397. intIP, err := getPrivateAddr()
  398. if err != nil {
  399. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  400. }
  401. if cfg.Node.LocalAddress != intIP && intIP != "" {
  402. ncutils.PrintLog("local Address has changed from "+cfg.Node.LocalAddress+" to "+intIP, 1)
  403. cfg.Node.LocalAddress = intIP
  404. PublishNodeUpdate(cfg)
  405. }
  406. } else {
  407. localIP, err := ncutils.GetLocalIP(cfg.Node.LocalRange)
  408. if err != nil {
  409. ncutils.PrintLog("error encountered checking ip addresses: "+err.Error(), 1)
  410. }
  411. if cfg.Node.Endpoint != localIP && localIP != "" {
  412. ncutils.PrintLog("endpoint has changed from "+cfg.Node.Endpoint+" to "+localIP, 1)
  413. cfg.Node.Endpoint = localIP
  414. PublishNodeUpdate(cfg)
  415. }
  416. }
  417. Hello(cfg, network)
  418. ncutils.Log("Checkin complete")
  419. }
  420. }
  421. }
  422. // PublishNodeUpdates -- saves node and pushes changes to broker
  423. func PublishNodeUpdate(cfg *config.ClientConfig) {
  424. if err := config.Write(cfg, cfg.Network); err != nil {
  425. ncutils.Log("error saving configuration" + err.Error())
  426. }
  427. data, err := json.Marshal(cfg.Node)
  428. if err != nil {
  429. ncutils.Log("error marshling node update " + err.Error())
  430. }
  431. if err = publish(cfg, fmt.Sprintf("update/%s", cfg.Node.ID), data); err != nil {
  432. ncutils.Log(fmt.Sprintf("error publishing endpoint update, %v", err))
  433. }
  434. }
  435. // Hello -- ping the broker to let server know node is alive and doing fine
  436. func Hello(cfg *config.ClientConfig, network string) {
  437. if err := publish(cfg, fmt.Sprintf("ping/%s", cfg.Node.ID), []byte("hello world!")); err != nil {
  438. ncutils.Log(fmt.Sprintf("error publishing ping, %v", err))
  439. }
  440. }
  441. func publish(cfg *config.ClientConfig, dest string, msg []byte) error {
  442. // setup the keys
  443. trafficPrivKey, err := auth.RetrieveTrafficKey(cfg.Node.Network)
  444. if err != nil {
  445. return err
  446. }
  447. serverPubKey, err := ncutils.ConvertBytesToKey(cfg.Node.TrafficKeys.Server)
  448. if err != nil {
  449. return err
  450. }
  451. client := SetupMQTT(cfg)
  452. defer client.Disconnect(250)
  453. encrypted, err := ncutils.BoxEncrypt(msg, serverPubKey, trafficPrivKey)
  454. if err != nil {
  455. return err
  456. }
  457. if token := client.Publish(dest, 0, false, encrypted); token.Wait() && token.Error() != nil {
  458. return token.Error()
  459. }
  460. return nil
  461. }
  462. func parseNetworkFromTopic(topic string) string {
  463. return strings.Split(topic, "/")[1]
  464. }
  465. func decryptMsg(cfg *config.ClientConfig, msg []byte) ([]byte, error) {
  466. // setup the keys
  467. diskKey, keyErr := auth.RetrieveTrafficKey(cfg.Node.Network)
  468. if keyErr != nil {
  469. return nil, keyErr
  470. }
  471. serverPubKey, err := ncutils.ConvertBytesToKey(cfg.Node.TrafficKeys.Server)
  472. if err != nil {
  473. return nil, err
  474. }
  475. return ncutils.BoxDecrypt(msg, serverPubKey, diskKey)
  476. }
  477. func shouldResub(currentServers, newServers []models.ServerAddr) bool {
  478. if len(currentServers) != len(newServers) {
  479. return true
  480. }
  481. for _, srv := range currentServers {
  482. if !ncutils.ServerAddrSliceContains(newServers, srv) {
  483. return true
  484. }
  485. }
  486. return false
  487. }
  488. func getID(topic string) (string, error) {
  489. parts := strings.Split(topic, "/")
  490. count := len(parts)
  491. if count == 1 {
  492. return "", errors.New("invalid topic")
  493. }
  494. //the last part of the topic will be the network.ID
  495. return parts[count-1], nil
  496. }