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. "github.com/gravitl/netmaker/servercfg"
  27. ssl "github.com/gravitl/netmaker/tls"
  28. "golang.zx2c4.com/wireguard/wgctrl/wgtypes"
  29. )
  30. var messageCache = new(sync.Map)
  31. var networkcontext = new(sync.Map)
  32. const lastNodeUpdate = "lnu"
  33. const lastPeerUpdate = "lpu"
  34. type cachedMessage struct {
  35. Message string
  36. LastSeen time.Time
  37. }
  38. // Daemon runs netclient daemon from command line
  39. func Daemon() error {
  40. serverSet := make(map[string]bool)
  41. // == initial pull of all networks ==
  42. networks, _ := ncutils.GetSystemNetworks()
  43. if len(networks) == 0 {
  44. return errors.New("no networks")
  45. }
  46. pubNetworks = append(pubNetworks, networks...)
  47. // set ipforwarding on startup
  48. err := local.SetIPForwarding()
  49. if err != nil {
  50. logger.Log(0, err.Error())
  51. }
  52. for _, network := range networks {
  53. logger.Log(3, "initializing network", network)
  54. cfg := config.ClientConfig{}
  55. cfg.Network = network
  56. cfg.ReadConfig()
  57. if err := wireguard.ApplyConf(&cfg.Node, cfg.Node.Interface, ncutils.GetNetclientPathSpecific()+cfg.Node.Interface+".conf"); err != nil {
  58. logger.Log(0, "failed to start ", cfg.Node.Interface, "wg interface", err.Error())
  59. }
  60. server := cfg.Server.Server
  61. if !serverSet[server] {
  62. // == subscribe to all nodes for each on machine ==
  63. serverSet[server] = true
  64. logger.Log(1, "started daemon for server ", server)
  65. ctx, cancel := context.WithCancel(context.Background())
  66. networkcontext.Store(server, cancel)
  67. go messageQueue(ctx, &cfg)
  68. }
  69. }
  70. // == add waitgroup and cancel for checkin routine ==
  71. wg := sync.WaitGroup{}
  72. ctx, cancel := context.WithCancel(context.Background())
  73. wg.Add(1)
  74. go Checkin(ctx, &wg)
  75. quit := make(chan os.Signal, 1)
  76. signal.Notify(quit, syscall.SIGTERM, os.Interrupt)
  77. <-quit
  78. for server := range serverSet {
  79. if cancel, ok := networkcontext.Load(server); ok {
  80. cancel.(context.CancelFunc)()
  81. }
  82. }
  83. cancel()
  84. logger.Log(0, "shutting down netclient daemon")
  85. wg.Wait()
  86. logger.Log(0, "shutdown complete")
  87. return nil
  88. }
  89. // UpdateKeys -- updates private key and returns new publickey
  90. func UpdateKeys(nodeCfg *config.ClientConfig, client mqtt.Client) error {
  91. logger.Log(0, "received message to update wireguard keys for network ", nodeCfg.Network)
  92. key, err := wgtypes.GeneratePrivateKey()
  93. if err != nil {
  94. logger.Log(0, "error generating privatekey ", err.Error())
  95. return err
  96. }
  97. file := ncutils.GetNetclientPathSpecific() + nodeCfg.Node.Interface + ".conf"
  98. if err := wireguard.UpdatePrivateKey(file, key.String()); err != nil {
  99. logger.Log(0, "error updating wireguard key ", err.Error())
  100. return err
  101. }
  102. if storeErr := wireguard.StorePrivKey(key.String(), nodeCfg.Network); storeErr != nil {
  103. logger.Log(0, "failed to save private key", storeErr.Error())
  104. return storeErr
  105. }
  106. nodeCfg.Node.PublicKey = key.PublicKey().String()
  107. PublishNodeUpdate(nodeCfg)
  108. return nil
  109. }
  110. // == Private ==
  111. // sets MQ client subscriptions for a specific node config
  112. // should be called for each node belonging to a given server
  113. func setSubscriptions(client mqtt.Client, nodeCfg *config.ClientConfig) {
  114. 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 {
  115. if token.Error() == nil {
  116. logger.Log(0, "connection timeout")
  117. } else {
  118. logger.Log(0, token.Error().Error())
  119. }
  120. return
  121. }
  122. logger.Log(3, fmt.Sprintf("subscribed to node updates for node %s update/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  123. if token := client.Subscribe(fmt.Sprintf("peers/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID), 0, mqtt.MessageHandler(UpdatePeers)); token.Wait() && token.Error() != nil {
  124. logger.Log(0, token.Error().Error())
  125. return
  126. }
  127. logger.Log(3, fmt.Sprintf("subscribed to peer updates for node %s peers/%s/%s", nodeCfg.Node.Name, nodeCfg.Node.Network, nodeCfg.Node.ID))
  128. }
  129. // on a delete usually, pass in the nodecfg to unsubscribe client broker communications
  130. // for the node in nodeCfg
  131. func unsubscribeNode(client mqtt.Client, nodeCfg *config.ClientConfig) {
  132. client.Unsubscribe(fmt.Sprintf("update/%s/%s", nodeCfg.Node.Network, nodeCfg.Node.ID))
  133. var ok = true
  134. 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 {
  135. if token.Error() == nil {
  136. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", "connection timeout")
  137. } else {
  138. logger.Log(1, "unable to unsubscribe from updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  139. }
  140. ok = false
  141. }
  142. 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 {
  143. if token.Error() == nil {
  144. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", "connection timeout")
  145. } else {
  146. logger.Log(1, "unable to unsubscribe from peer updates for node ", nodeCfg.Node.Name, "\n", token.Error().Error())
  147. }
  148. ok = false
  149. }
  150. if ok {
  151. logger.Log(1, "successfully unsubscribed node ", nodeCfg.Node.ID, " : ", nodeCfg.Node.Name)
  152. }
  153. }
  154. // sets up Message Queue and subsribes/publishes updates to/from server
  155. // the client should subscribe to ALL nodes that exist on server locally
  156. func messageQueue(ctx context.Context, cfg *config.ClientConfig) {
  157. logger.Log(0, "netclient daemon started for server: ", cfg.Server.Server)
  158. client, err := setupMQTT(cfg, false)
  159. if err != nil {
  160. logger.Log(0, "unable to connect to broker", err.Error())
  161. return
  162. }
  163. defer client.Disconnect(250)
  164. <-ctx.Done()
  165. logger.Log(0, "shutting down daemon for server ", cfg.Server.Server)
  166. }
  167. // NewTLSConf sets up tls configuration to connect to broker securely
  168. func NewTLSConfig(server string) *tls.Config {
  169. file := ncutils.GetNetclientServerPath(server) + ncutils.GetSeparator() + "root.pem"
  170. certpool := x509.NewCertPool()
  171. ca, err := os.ReadFile(file)
  172. if err != nil {
  173. logger.Log(0, "could not read CA file ", err.Error())
  174. }
  175. ok := certpool.AppendCertsFromPEM(ca)
  176. if !ok {
  177. logger.Log(0, "failed to append cert")
  178. }
  179. clientKeyPair, err := tls.LoadX509KeyPair(ncutils.GetNetclientServerPath(server)+ncutils.GetSeparator()+"client.pem", ncutils.GetNetclientPath()+ncutils.GetSeparator()+"client.key")
  180. if err != nil {
  181. log.Fatalf("could not read client cert/key %v \n", err)
  182. }
  183. certs := []tls.Certificate{clientKeyPair}
  184. return &tls.Config{
  185. RootCAs: certpool,
  186. ClientAuth: tls.NoClientCert,
  187. ClientCAs: nil,
  188. Certificates: certs,
  189. InsecureSkipVerify: false,
  190. }
  191. }
  192. // setupMQTT creates a connection to broker and returns client
  193. // this function is primarily used to create a connection to publish to the broker
  194. func setupMQTT(cfg *config.ClientConfig, publish bool) (mqtt.Client, error) {
  195. opts := mqtt.NewClientOptions()
  196. server := cfg.Server.Server
  197. opts.AddBroker("ssl://" + server + ":" + servercfg.GetMQPort())
  198. opts.SetTLSConfig(NewTLSConfig(server))
  199. opts.SetClientID(ncutils.MakeRandomString(23))
  200. opts.SetDefaultPublishHandler(All)
  201. opts.SetAutoReconnect(true)
  202. opts.SetConnectRetry(true)
  203. opts.SetConnectRetryInterval(time.Second << 2)
  204. opts.SetKeepAlive(time.Minute >> 1)
  205. opts.SetWriteTimeout(time.Minute)
  206. opts.SetOnConnectHandler(func(client mqtt.Client) {
  207. if !publish {
  208. networks, err := ncutils.GetSystemNetworks()
  209. if err != nil {
  210. logger.Log(0, "error retriving networks ", err.Error())
  211. }
  212. for _, network := range networks {
  213. var currNodeCfg config.ClientConfig
  214. currNodeCfg.Network = network
  215. currNodeCfg.ReadConfig()
  216. setSubscriptions(client, &currNodeCfg)
  217. }
  218. }
  219. })
  220. opts.SetOrderMatters(true)
  221. opts.SetResumeSubs(true)
  222. opts.SetConnectionLostHandler(func(c mqtt.Client, e error) {
  223. logger.Log(0, "detected broker connection lost for", cfg.Server.Server)
  224. })
  225. client := mqtt.NewClient(opts)
  226. for token := client.Connect(); !token.WaitTimeout(30*time.Second) || token.Error() != nil; token = client.Connect() {
  227. logger.Log(0, "unable to connect to broker, retrying ...")
  228. var err error
  229. if token.Error() == nil {
  230. err = errors.New("connect timeout")
  231. } else {
  232. err = token.Error()
  233. }
  234. if err = checkBroker(cfg.Server.Server, cfg.Server.BrokerPort); err != nil &&
  235. (strings.Contains(err.Error(), "connectex") ||
  236. strings.Contains(err.Error(), "connect timeout")) ||
  237. strings.Contains(err.Error(), EMPTY_BROKER_ERR) {
  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. }