daemon.go 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354
  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]config.ClientConfig)
  40. // == initial pull of all networks ==
  41. networks, _ := ncutils.GetSystemNetworks()
  42. if len(networks) == 0 {
  43. return errors.New("no networks")
  44. }
  45. for _, network := range networks {
  46. logger.Log(3, "initializing network", network)
  47. cfg := config.ClientConfig{}
  48. cfg.Network = network
  49. cfg.ReadConfig()
  50. serverSet[cfg.Server.Server] = cfg
  51. if err := wireguard.ApplyConf(&cfg.Node, cfg.Node.Interface, ncutils.GetNetclientPathSpecific()+cfg.Node.Interface+".conf"); err != nil {
  52. logger.Log(0, "failed to start ", cfg.Node.Interface, "wg interface", err.Error())
  53. }
  54. //initialPull(cfg.Network)
  55. }
  56. // set ipforwarding on startup
  57. err := local.SetIPForwarding()
  58. if err != nil {
  59. logger.Log(0, err.Error())
  60. }
  61. // == subscribe to all nodes for each on machine ==
  62. for server := range serverSet {
  63. logger.Log(1, "started daemon for server ", server)
  64. ctx, cancel := context.WithCancel(context.Background())
  65. networkcontext.Store(server, cancel)
  66. config := serverSet[server]
  67. go messageQueue(ctx, &config)
  68. }
  69. // == add waitgroup and cancel for checkin routine ==
  70. wg := sync.WaitGroup{}
  71. ctx, cancel := context.WithCancel(context.Background())
  72. wg.Add(1)
  73. go Checkin(ctx, &wg)
  74. quit := make(chan os.Signal, 1)
  75. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  76. <-quit
  77. for server := range serverSet {
  78. if cancel, ok := networkcontext.Load(server); ok {
  79. cancel.(context.CancelFunc)()
  80. }
  81. }
  82. cancel()
  83. logger.Log(0, "shutting down netclient daemon")
  84. wg.Wait()
  85. logger.Log(0, "shutdown complete")
  86. return nil
  87. }
  88. // UpdateKeys -- updates private key and returns new publickey
  89. func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
  90. logger.Log(0, "received message to update wireguard keys for network ", nodeCfg.Network)
  91. key, err := wgtypes.GeneratePrivateKey()
  92. if err != nil {
  93. logger.Log(0, "error generating privatekey ", err.Error())
  94. return err
  95. }
  96. file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
  97. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  98. logger.Log(0, "error updating wireguard key ", err.Error())
  99. return err
  100. }
  101. if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
  102. logger.Log(0, "failed to save private key", storeErr.Error())
  103. return storeErr
  104. }
  105. nodeCfg.Node.PublicKey = key.PublicKey().String()
  106. PublishNodeUpdate(nodeCfg)
  107. return nil
  108. }
  109. // == Private ==
  110. // sets MQ client subscriptions for a specific node config
  111. // should be called for each node belonging to a given server
  112. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  113. 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 {
  114. if token.Error() == nil {
  115. logger.Log(0, "connection timeout")
  116. } else {
  117. logger.Log(0, token.Error().Error())
  118. }
  119. return
  120. }
  121. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  122. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  123. logger.Log(0, token.Error().Error())
  124. return
  125. }
  126. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  127. }
  128. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  129. // for the node in nodeCfg
  130. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  131. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  132. var ok = true
  133. 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 {
  134. if token.Error() == nil {
  135. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", "connection timeout")
  136. } else {
  137. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  138. }
  139. ok = false
  140. }
  141. 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 {
  142. if token.Error() == nil {
  143. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", "connection timeout")
  144. } else {
  145. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  146. }
  147. ok = false
  148. }
  149. if ok {
  150. logger.Log(1, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  151. }
  152. }
  153. // sets up Message Queue and subsribes/publishes updates to/from server
  154. // the client should subscribe to ALL nodes that exist on server locally
  155. func messageQueue(ctx context.Context, cfg *config.ClientConfig) {
  156. logger.Log(0, "netclient daemon started for server: ", cfg.Server.Server)
  157. client, err := setupMQTT(cfg, false)
  158. if err != nil {
  159. logger.Log(0, "unable to connect to broker", err.Error())
  160. return
  161. }
  162. defer client.Disconnect(250)
  163. <-ctx.Done()
  164. logger.Log(0, "shutting down daemon for server ", cfg.Server.Server)
  165. }
  166. // NewTLSConf sets up tls configuration to connect to broker securely
  167. func NewTLSConfig(server string) *tls.Config {
  168. file := ncutils.GetNetclientServerPath(server) + ncutils.GetSeparator() + "root.pem"
  169. certpool := x509.NewCertPool()
  170. ca, err := os.ReadFile(file)
  171. if err != nil {
  172. logger.Log(0, "could not read CA file ", err.Error())
  173. }
  174. ok := certpool.AppendCertsFromPEM(ca)
  175. if !ok {
  176. logger.Log(0, "failed to append cert")
  177. }
  178. clientKeyPair, err := tls.LoadX509KeyPair(ncutils.GetNetclientServerPath(server)+ncutils.GetSeparator()+"client.pem", ncutils.GetNetclientPath()+ncutils.GetSeparator()+"client.key")
  179. if err != nil {
  180. log.Fatalf("could not read client cert/key %v \n", err)
  181. }
  182. certs := []tls.Certificate{clientKeyPair}
  183. return &tls.Config{
  184. RootCAs: certpool,
  185. ClientAuth: tls.NoClientCert,
  186. ClientCAs: nil,
  187. Certificates: certs,
  188. InsecureSkipVerify: false,
  189. }
  190. }
  191. // setupMQTT creates a connection to broker and returns client
  192. // this function is primarily used to create a connection to publish to the broker
  193. func setupMQTT(cfg *config.ClientConfig, publish bool) (mqtt.Client, error) {
  194. opts := mqtt.NewClientOptions()
  195. server := cfg.Server.Server
  196. opts.AddBroker("ssl://" + server + ":8883") // TODO get the appropriate port of the comms mq server
  197. opts.SetTLSConfig(NewTLSConfig(server))
  198. opts.SetClientID(ncutils.MakeRandomString(23))
  199. opts.SetDefaultPublishHandler(All)
  200. opts.SetAutoReconnect(true)
  201. opts.SetConnectRetry(true)
  202. opts.SetConnectRetryInterval(time.Second << 2)
  203. opts.SetKeepAlive(time.Minute >> 1)
  204. opts.SetWriteTimeout(time.Minute)
  205. opts.SetOnConnectHandler(func(client mqtt.Client) {
  206. if !publish {
  207. networks, err := ncutils.GetSystemNetworks()
  208. if err != nil {
  209. logger.Log(0, "error retriving networks ", err.Error())
  210. }
  211. for _, network := range networks {
  212. var currNodeCfg config.ClientConfig
  213. currNodeCfg.Network = network
  214. currNodeCfg.ReadConfig()
  215. setSubscriptions(client, &currNodeCfg)
  216. }
  217. }
  218. })
  219. opts.SetOrderMatters(true)
  220. opts.SetResumeSubs(true)
  221. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  222. logger.Log(0, "detected broker connection lost for", cfg.Server.Server)
  223. })
  224. client := mqtt.NewClient(opts)
  225. for token := client.Connect(); !token.WaitTimeout(30*time.Second) || token.Error() != nil; token = client.Connect() {
  226. logger.Log(0, "unable to connect to broker, retrying ...")
  227. var err error
  228. if token.Error() == nil {
  229. err = errors.New("connect timeout")
  230. } else {
  231. err = token.Error()
  232. }
  233. if err := checkBroker(cfg.Server.Server); err != nil {
  234. return nil, err
  235. }
  236. logger.Log(0, "could not connect to broker", cfg.Server.Server, err.Error())
  237. if strings.Contains(err.Error(), "connectex") || strings.Contains(err.Error(), "connect timeout") {
  238. logger.Log(0, "connection issue detected.. attempt connection with new certs")
  239. key, err := ssl.ReadKey(ncutils.GetNetclientPath() + ncutils.GetSeparator() + "client.key")
  240. if err != nil {
  241. _, *key, err = ed25519.GenerateKey(rand.Reader)
  242. if err != nil {
  243. log.Fatal("could not generate new key")
  244. }
  245. }
  246. RegisterWithServer(key, cfg)
  247. daemon.Restart()
  248. }
  249. }
  250. return client, nil
  251. }
  252. // publishes a message to server to update peers on this peer's behalf
  253. func publishSignal(nodeCfg *config.ClientConfig, signal byte) error {
  254. if err := publish(nodeCfg, fmt.Sprintf("signal/%s", nodeCfg.Node.ID), []byte{signal}, 1); err != nil {
  255. return err
  256. }
  257. return nil
  258. }
  259. func initialPull(network string) {
  260. logger.Log(0, "pulling latest config for ", network)
  261. var configPath = fmt.Sprintf("%snetconfig-%s", ncutils.GetNetclientPathSpecific(), network)
  262. fileInfo, err := os.Stat(configPath)
  263. if err != nil {
  264. logger.Log(0, "could not stat config file: ", configPath)
  265. return
  266. }
  267. // speed up UDP rest
  268. if !fileInfo.ModTime().IsZero() && time.Now().After(fileInfo.ModTime().Add(time.Minute)) {
  269. sleepTime := 2
  270. for {
  271. _, err := Pull(network, true)
  272. if err == nil {
  273. break
  274. }
  275. if sleepTime > 3600 {
  276. sleepTime = 3600
  277. }
  278. logger.Log(0, "failed to pull for network ", network)
  279. logger.Log(0, fmt.Sprintf("waiting %d seconds to retry...", sleepTime))
  280. time.Sleep(time.Second * time.Duration(sleepTime))
  281. sleepTime = sleepTime * 2
  282. }
  283. time.Sleep(time.Second << 1)
  284. }
  285. }
  286. func parseNetworkFromTopic(topic string) string {
  287. return strings.Split(topic, "/")[1]
  288. }
  289. // should only ever use node client configs
  290. func decryptMsg(nodeCfg *config.ClientConfig, msg []byte) ([]byte, error) {
  291. if len(msg) <= 24 { // make sure message is of appropriate length
  292. return nil, fmt.Errorf("recieved invalid message from broker %v", msg)
  293. }
  294. // setup the keys
  295. diskKey, keyErr := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  296. if keyErr != nil {
  297. return nil, keyErr
  298. }
  299. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  300. if err != nil {
  301. return nil, err
  302. }
  303. return ncutils.DeChunk(msg, serverPubKey, diskKey)
  304. }
  305. // == Message Caches ==
  306. func insert(network, which, cache string) {
  307. var newMessage = cachedMessage{
  308. Message: cache,
  309. LastSeen: time.Now(),
  310. }
  311. messageCache.Store(fmt.Sprintf("%s%s", network, which), newMessage)
  312. }
  313. func read(network, which string) string {
  314. val, isok := messageCache.Load(fmt.Sprintf("%s%s", network, which))
  315. if isok {
  316. var readMessage = val.(cachedMessage) // fetch current cached message
  317. if readMessage.LastSeen.IsZero() {
  318. return ""
  319. }
  320. if time.Now().After(readMessage.LastSeen.Add(time.Minute * 10)) { // check if message has been there over a minute
  321. messageCache.Delete(fmt.Sprintf("%s%s", network, which)) // remove old message if expired
  322. return ""
  323. }
  324. return readMessage.Message // return current message if not expired
  325. }
  326. return ""
  327. }