daemon.go 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353
  1. package functions
  2. import (
  3. "context"
  4. "crypto/ed25519"
  5. "crypto/rand"
  6. "crypto/tls"
  7. "crypto/x509"
  8. "errors"
  9. "fmt"
  10. "log"
  11. "os"
  12. "os/signal"
  13. "strings"
  14. "sync"
  15. "syscall"
  16. "time"
  17. mqtt "github.com/eclipse/paho.mqtt.golang"
  18. "github.com/gravitl/netmaker/logger"
  19. "github.com/gravitl/netmaker/mq"
  20. "github.com/gravitl/netmaker/netclient/auth"
  21. "github.com/gravitl/netmaker/netclient/config"
  22. "github.com/gravitl/netmaker/netclient/daemon"
  23. "github.com/gravitl/netmaker/netclient/local"
  24. "github.com/gravitl/netmaker/netclient/ncutils"
  25. "github.com/gravitl/netmaker/netclient/wireguard"
  26. ssl "github.com/gravitl/netmaker/tls"
  27. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  28. )
  29. var messageCache = new(sync.Map)
  30. var networkcontext = new(sync.Map)
  31. const lastNodeUpdate = "lnu"
  32. const lastPeerUpdate = "lpu"
  33. type cachedMessage struct {
  34. Message string
  35. LastSeen time.Time
  36. }
  37. // Daemon runs netclient daemon from command line
  38. func Daemon() error {
  39. serverSet := make(map[string]bool)
  40. // == initial pull of all networks ==
  41. networks, _ := ncutils.GetSystemNetworks()
  42. if len(networks) == 0 {
  43. return errors.New("no networks")
  44. }
  45. // set ipforwarding on startup
  46. err := local.SetIPForwarding()
  47. if err != nil {
  48. logger.Log(0, err.Error())
  49. }
  50. for _, network := range networks {
  51. logger.Log(3, "initializing network", network)
  52. cfg := config.ClientConfig{}
  53. cfg.Network = network
  54. cfg.ReadConfig()
  55. if err := wireguard.ApplyConf(&cfg.Node, cfg.Node.Interface, ncutils.GetNetclientPathSpecific()+cfg.Node.Interface+".conf"); err != nil {
  56. logger.Log(0, "failed to start ", cfg.Node.Interface, "wg interface", err.Error())
  57. }
  58. server := cfg.Server.Server
  59. if !serverSet[server] {
  60. // == subscribe to all nodes for each on machine ==
  61. serverSet[server] = true
  62. logger.Log(1, "started daemon for server ", server)
  63. ctx, cancel := context.WithCancel(context.Background())
  64. networkcontext.Store(server, cancel)
  65. go messageQueue(ctx, &cfg)
  66. }
  67. }
  68. // == add waitgroup and cancel for checkin routine ==
  69. wg := sync.WaitGroup{}
  70. ctx, cancel := context.WithCancel(context.Background())
  71. wg.Add(1)
  72. go Checkin(ctx, &wg)
  73. quit := make(chan os.Signal, 1)
  74. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  75. <-quit
  76. for server := range serverSet {
  77. if cancel, ok := networkcontext.Load(server); ok {
  78. cancel.(context.CancelFunc)()
  79. }
  80. }
  81. cancel()
  82. logger.Log(0, "shutting down netclient daemon")
  83. wg.Wait()
  84. logger.Log(0, "shutdown complete")
  85. return nil
  86. }
  87. // UpdateKeys -- updates private key and returns new publickey
  88. func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
  89. logger.Log(0, "received message to update wireguard keys for network ", nodeCfg.Network)
  90. key, err := wgtypes.GeneratePrivateKey()
  91. if err != nil {
  92. logger.Log(0, "error generating privatekey ", err.Error())
  93. return err
  94. }
  95. file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
  96. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  97. logger.Log(0, "error updating wireguard key ", err.Error())
  98. return err
  99. }
  100. if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
  101. logger.Log(0, "failed to save private key", storeErr.Error())
  102. return storeErr
  103. }
  104. nodeCfg.Node.PublicKey = key.PublicKey().String()
  105. PublishNodeUpdate(nodeCfg)
  106. return nil
  107. }
  108. // == Private ==
  109. // sets MQ client subscriptions for a specific node config
  110. // should be called for each node belonging to a given server
  111. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  112. if token := client.Subscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(NodeUpdate)); token.WaitTimeout(mq.MQ_TIMEOUT*time.Second) && token.Error() != nil {
  113. if token.Error() == nil {
  114. logger.Log(0, "connection timeout")
  115. } else {
  116. logger.Log(0, token.Error().Error())
  117. }
  118. return
  119. }
  120. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  121. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  122. logger.Log(0, token.Error().Error())
  123. return
  124. }
  125. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  126. }
  127. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  128. // for the node in nodeCfg
  129. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  130. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  131. var ok = true
  132. if token := client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.WaitTimeout(mq.MQ_TIMEOUT*time.Second) && token.Error() != nil {
  133. if token.Error() == nil {
  134. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", "connection timeout")
  135. } else {
  136. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  137. }
  138. ok = false
  139. }
  140. if token := client.Unsubscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID)); token.WaitTimeout(mq.MQ_TIMEOUT*time.Second) && token.Error() != nil {
  141. if token.Error() == nil {
  142. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", "connection timeout")
  143. } else {
  144. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  145. }
  146. ok = false
  147. }
  148. if ok {
  149. logger.Log(1, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  150. }
  151. }
  152. // sets up Message Queue and subsribes/publishes updates to/from server
  153. // the client should subscribe to ALL nodes that exist on server locally
  154. func messageQueue(ctx context.Context, cfg *config.ClientConfig) {
  155. logger.Log(0, "netclient daemon started for server: ", cfg.Server.Server)
  156. client, err := setupMQTT(cfg, false)
  157. if err != nil {
  158. logger.Log(0, "unable to connect to broker", err.Error())
  159. return
  160. }
  161. defer client.Disconnect(250)
  162. <-ctx.Done()
  163. logger.Log(0, "shutting down daemon for server ", cfg.Server.Server)
  164. }
  165. // NewTLSConf sets up tls configuration to connect to broker securely
  166. func NewTLSConfig(server string) *tls.Config {
  167. file := ncutils.GetNetclientServerPath(server) + ncutils.GetSeparator() + "root.pem"
  168. certpool := x509.NewCertPool()
  169. ca, err := os.ReadFile(file)
  170. if err != nil {
  171. logger.Log(0, "could not read CA file ", err.Error())
  172. }
  173. ok := certpool.AppendCertsFromPEM(ca)
  174. if !ok {
  175. logger.Log(0, "failed to append cert")
  176. }
  177. clientKeyPair, err := tls.LoadX509KeyPair(ncutils.GetNetclientServerPath(server)+ncutils.GetSeparator()+"client.pem", ncutils.GetNetclientPath()+ncutils.GetSeparator()+"client.key")
  178. if err != nil {
  179. log.Fatalf("could not read client cert/key %v \n", err)
  180. }
  181. certs := []tls.Certificate{clientKeyPair}
  182. return &tls.Config{
  183. RootCAs: certpool,
  184. ClientAuth: tls.NoClientCert,
  185. ClientCAs: nil,
  186. Certificates: certs,
  187. InsecureSkipVerify: false,
  188. }
  189. }
  190. // setupMQTT creates a connection to broker and returns client
  191. // this function is primarily used to create a connection to publish to the broker
  192. func setupMQTT(cfg *config.ClientConfig, publish bool) (mqtt.Client, error) {
  193. opts := mqtt.NewClientOptions()
  194. server := cfg.Server.Server
  195. opts.AddBroker("ssl://" + server + ":8883") // TODO get the appropriate port of the comms mq server
  196. opts.SetTLSConfig(NewTLSConfig(server))
  197. opts.SetClientID(ncutils.MakeRandomString(23))
  198. opts.SetDefaultPublishHandler(All)
  199. opts.SetAutoReconnect(true)
  200. opts.SetConnectRetry(true)
  201. opts.SetConnectRetryInterval(time.Second << 2)
  202. opts.SetKeepAlive(time.Minute >> 1)
  203. opts.SetWriteTimeout(time.Minute)
  204. opts.SetOnConnectHandler(func(client mqtt.Client) {
  205. if !publish {
  206. networks, err := ncutils.GetSystemNetworks()
  207. if err != nil {
  208. logger.Log(0, "error retriving networks ", err.Error())
  209. }
  210. for _, network := range networks {
  211. var currNodeCfg config.ClientConfig
  212. currNodeCfg.Network = network
  213. currNodeCfg.ReadConfig()
  214. setSubscriptions(client, &currNodeCfg)
  215. }
  216. }
  217. })
  218. opts.SetOrderMatters(true)
  219. opts.SetResumeSubs(true)
  220. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  221. logger.Log(0, "detected broker connection lost for", cfg.Server.Server)
  222. })
  223. client := mqtt.NewClient(opts)
  224. for token := client.Connect(); !token.WaitTimeout(30*time.Second) || token.Error() != nil; token = client.Connect() {
  225. logger.Log(0, "unable to connect to broker, retrying ...")
  226. var err error
  227. if token.Error() == nil {
  228. err = errors.New("connect timeout")
  229. } else {
  230. err = token.Error()
  231. }
  232. if err := checkBroker(cfg.Server.Server); err != nil {
  233. return nil, err
  234. }
  235. logger.Log(0, "could not connect to broker", cfg.Server.Server, err.Error())
  236. if strings.Contains(err.Error(), "connectex") || strings.Contains(err.Error(), "connect timeout") {
  237. logger.Log(0, "connection issue detected.. attempt connection with new certs")
  238. key, err := ssl.ReadKey(ncutils.GetNetclientPath() + ncutils.GetSeparator() + "client.key")
  239. if err != nil {
  240. _, *key, err = ed25519.GenerateKey(rand.Reader)
  241. if err != nil {
  242. log.Fatal("could not generate new key")
  243. }
  244. }
  245. RegisterWithServer(key, cfg)
  246. daemon.Restart()
  247. }
  248. }
  249. return client, nil
  250. }
  251. // publishes a message to server to update peers on this peer's behalf
  252. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  253. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  254. return err
  255. }
  256. return nil
  257. }
  258. func initialPull(network string) {
  259. logger.Log(0, "pulling latest config for ", network)
  260. var configPath = fmt.Sprintf("%snetconfig-%s", ncutils.GetNetclientPathSpecific(), network)
  261. fileInfo, err := os.Stat(configPath)
  262. if err != nil {
  263. logger.Log(0, "could not stat config file: ", configPath)
  264. return
  265. }
  266. // speed up UDP rest
  267. if !fileInfo.ModTime().IsZero() && time.Now().After(fileInfo.ModTime().Add(time.Minute)) {
  268. sleepTime := 2
  269. for {
  270. _, err := Pull(network, true)
  271. if err == nil {
  272. break
  273. }
  274. if sleepTime > 3600 {
  275. sleepTime = 3600
  276. }
  277. logger.Log(0, "failed to pull for network ", network)
  278. logger.Log(0, fmt.Sprintf("waiting %d seconds to retry...", sleepTime))
  279. time.Sleep(time.Second * time.Duration(sleepTime))
  280. sleepTime = sleepTime * 2
  281. }
  282. time.Sleep(time.Second << 1)
  283. }
  284. }
  285. func parseNetworkFromTopic(topic string) string {
  286. return strings.Split(topic, "/")[1]
  287. }
  288. // should only ever use node client configs
  289. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  290. if len(msg) <= 24 { // make sure message is of appropriate length
  291. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  292. }
  293. // setup the keys
  294. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  295. if keyErr != nil {
  296. return nil, keyErr
  297. }
  298. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  299. if err != nil {
  300. return nil, err
  301. }
  302. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  303. }
  304. // == Message Caches ==
  305. func insert(network, which, cache string) {
  306. var newMessage = cachedMessage{
  307. Message: cache,
  308. LastSeen: time.Now(),
  309. }
  310. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  311. }
  312. func read(network, which string) string {
  313. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  314. if isok {
  315. var readMessage = val.(cachedMessage) // fetch current cached message
  316. if readMessage.LastSeen.IsZero() {
  317. return ""
  318. }
  319. if time.Now().After(readMessage.LastSeen.Add(time.Minute * 10)) { // check if message has been there over a minute
  320. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  321. return ""
  322. }
  323. return readMessage.Message // return current message if not expired
  324. }
  325. return ""
  326. }