service_test.go 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488
  1. package service
  2. import (
  3. "bytes"
  4. "context"
  5. "errors"
  6. "fmt"
  7. "net"
  8. "net/netip"
  9. "os"
  10. "testing"
  11. "time"
  12. "dario.cat/mergo"
  13. "github.com/slackhq/nebula"
  14. "github.com/slackhq/nebula/cert"
  15. "github.com/slackhq/nebula/cert_test"
  16. "github.com/slackhq/nebula/config"
  17. "github.com/slackhq/nebula/logging"
  18. "github.com/slackhq/nebula/overlay"
  19. "go.yaml.in/yaml/v3"
  20. "golang.org/x/sync/errgroup"
  21. )
  22. type m = map[string]any
  23. func newSimpleService(caCrt cert.Certificate, caKey []byte, name string, udpIp netip.Addr, overrides m) *Service {
  24. _, _, myPrivKey, myPEM := cert_test.NewTestCert(cert.Version2, cert.Curve_CURVE25519, caCrt, caKey, "a", time.Now(), time.Now().Add(5*time.Minute), []netip.Prefix{netip.PrefixFrom(udpIp, 24)}, nil, []string{})
  25. caB, err := caCrt.MarshalPEM()
  26. if err != nil {
  27. panic(err)
  28. }
  29. mc := m{
  30. "pki": m{
  31. "ca": string(caB),
  32. "cert": string(myPEM),
  33. "key": string(myPrivKey),
  34. },
  35. //"tun": m{"disabled": true},
  36. "firewall": m{
  37. "outbound": []m{{
  38. "proto": "any",
  39. "port": "any",
  40. "host": "any",
  41. }},
  42. "inbound": []m{{
  43. "proto": "any",
  44. "port": "any",
  45. "host": "any",
  46. }},
  47. },
  48. "timers": m{
  49. "pending_deletion_interval": 2,
  50. "connection_alive_interval": 2,
  51. },
  52. "handshakes": m{
  53. "try_interval": "200ms",
  54. },
  55. }
  56. if overrides != nil {
  57. err = mergo.Merge(&overrides, mc, mergo.WithAppendSlice)
  58. if err != nil {
  59. panic(err)
  60. }
  61. mc = overrides
  62. }
  63. cb, err := yaml.Marshal(mc)
  64. if err != nil {
  65. panic(err)
  66. }
  67. var c config.C
  68. if err := c.LoadString(string(cb)); err != nil {
  69. panic(err)
  70. }
  71. logger := logging.NewLogger(os.Stdout)
  72. control, err := nebula.Main(&c, false, "custom-app", logger, overlay.NewUserDeviceFromConfig)
  73. if err != nil {
  74. panic(err)
  75. }
  76. s, err := New(control)
  77. if err != nil {
  78. panic(err)
  79. }
  80. return s
  81. }
  82. // ephemeralUDPPort reserves a free UDP port by binding port 0, then releases
  83. // it for the caller to use. A fixed port would collide with concurrent test
  84. // runs or an unrelated process (say, a real nebula) already listening on it.
  85. func ephemeralUDPPort(t *testing.T) int {
  86. t.Helper()
  87. pc, err := net.ListenPacket("udp4", "127.0.0.1:0")
  88. if err != nil {
  89. t.Fatal(err)
  90. }
  91. port := pc.LocalAddr().(*net.UDPAddr).Port
  92. _ = pc.Close()
  93. return port
  94. }
  95. func TestService(t *testing.T) {
  96. lighthousePort := ephemeralUDPPort(t)
  97. ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
  98. a := newSimpleService(ca, caKey, "a", netip.MustParseAddr("10.0.0.1"), m{
  99. "static_host_map": m{},
  100. "lighthouse": m{
  101. "am_lighthouse": true,
  102. },
  103. "listen": m{
  104. "host": "0.0.0.0",
  105. "port": lighthousePort,
  106. },
  107. })
  108. b := newSimpleService(ca, caKey, "b", netip.MustParseAddr("10.0.0.2"), m{
  109. "static_host_map": m{
  110. "10.0.0.1": []string{fmt.Sprintf("localhost:%d", lighthousePort)},
  111. },
  112. "lighthouse": m{
  113. "hosts": []string{"10.0.0.1"},
  114. "interval": 1,
  115. },
  116. })
  117. ln, err := a.Listen("tcp", ":1234")
  118. if err != nil {
  119. t.Fatal(err)
  120. }
  121. var eg errgroup.Group
  122. eg.Go(func() error {
  123. conn, err := ln.Accept()
  124. if err != nil {
  125. return err
  126. }
  127. defer conn.Close()
  128. t.Log("accepted connection")
  129. if _, err := conn.Write([]byte("server msg")); err != nil {
  130. return err
  131. }
  132. t.Log("server: wrote message")
  133. data := make([]byte, 100)
  134. n, err := conn.Read(data)
  135. if err != nil {
  136. return err
  137. }
  138. data = data[:n]
  139. if !bytes.Equal(data, []byte("client msg")) {
  140. return errors.New("got invalid message from client")
  141. }
  142. t.Log("server: read message")
  143. return conn.Close()
  144. })
  145. c, err := b.DialContext(context.Background(), "tcp", "10.0.0.1:1234")
  146. if err != nil {
  147. t.Fatal(err)
  148. }
  149. if _, err := c.Write([]byte("client msg")); err != nil {
  150. t.Fatal(err)
  151. }
  152. data := make([]byte, 100)
  153. n, err := c.Read(data)
  154. if err != nil {
  155. t.Fatal(err)
  156. }
  157. data = data[:n]
  158. if !bytes.Equal(data, []byte("server msg")) {
  159. t.Fatal("got invalid message from client")
  160. }
  161. if err := c.Close(); err != nil {
  162. t.Fatal(err)
  163. }
  164. if err := eg.Wait(); err != nil {
  165. t.Fatal(err)
  166. }
  167. }
  168. // A listener that is never accepted from must not stall connections to other
  169. // ports, and closing it must not block.
  170. func TestServiceUnacceptedListener(t *testing.T) {
  171. lighthousePort := ephemeralUDPPort(t)
  172. ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
  173. a := newSimpleService(ca, caKey, "a", netip.MustParseAddr("10.0.0.1"), m{
  174. "static_host_map": m{},
  175. "lighthouse": m{
  176. "am_lighthouse": true,
  177. },
  178. "listen": m{
  179. "host": "0.0.0.0",
  180. "port": lighthousePort,
  181. },
  182. })
  183. t.Cleanup(func() { _ = a.Close() })
  184. b := newSimpleService(ca, caKey, "b", netip.MustParseAddr("10.0.0.2"), m{
  185. "static_host_map": m{
  186. "10.0.0.1": []string{fmt.Sprintf("127.0.0.1:%d", lighthousePort)},
  187. },
  188. "lighthouse": m{
  189. "hosts": []string{"10.0.0.1"},
  190. "interval": 1,
  191. },
  192. })
  193. t.Cleanup(func() { _ = b.Close() })
  194. idle, err := a.Listen("tcp", ":1234")
  195. if err != nil {
  196. t.Fatal(err)
  197. }
  198. active, err := a.Listen("tcp", ":1235")
  199. if err != nil {
  200. t.Fatal(err)
  201. }
  202. ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
  203. defer cancel()
  204. idleConn, err := b.DialContext(ctx, "tcp", "10.0.0.1:1234")
  205. if err != nil {
  206. t.Fatal(err)
  207. }
  208. defer idleConn.Close()
  209. accepted := make(chan error, 1)
  210. go func() {
  211. conn, err := active.Accept()
  212. if err == nil {
  213. err = conn.Close()
  214. }
  215. accepted <- err
  216. }()
  217. ctx, cancel = context.WithTimeout(context.Background(), 5*time.Second)
  218. defer cancel()
  219. activeConn, err := b.DialContext(ctx, "tcp", "10.0.0.1:1235")
  220. if err != nil {
  221. t.Fatalf("dial to accepted port stalled behind idle listener: %v", err)
  222. }
  223. defer activeConn.Close()
  224. select {
  225. case err := <-accepted:
  226. if err != nil {
  227. t.Fatal(err)
  228. }
  229. case <-time.After(5 * time.Second):
  230. t.Fatal("accept stalled behind idle listener")
  231. }
  232. closed := make(chan error, 1)
  233. go func() { closed <- idle.Close() }()
  234. select {
  235. case err := <-closed:
  236. if err != nil {
  237. t.Fatal(err)
  238. }
  239. case <-time.After(5 * time.Second):
  240. t.Fatal("closing idle listener blocked")
  241. }
  242. if _, err := idle.Accept(); !errors.Is(err, net.ErrClosed) {
  243. t.Fatalf("expected net.ErrClosed from closed listener, got %v", err)
  244. }
  245. if err := idle.Close(); err != nil {
  246. t.Fatalf("second close: %v", err)
  247. }
  248. }
  249. // Accept must report net.ErrClosed after Close even while a handler is still
  250. // trying to hand off a connection.
  251. func TestListenerAcceptAfterClose(t *testing.T) {
  252. s := &Service{}
  253. s.mu.listeners = map[uint16]*tcpListener{}
  254. l := &tcpListener{
  255. port: 1234,
  256. s: s,
  257. accept: make(chan net.Conn),
  258. closed: make(chan struct{}),
  259. }
  260. s.mu.listeners[l.port] = l
  261. if err := l.Close(); err != nil {
  262. t.Fatal(err)
  263. }
  264. // Keep a send pending so every Accept has a connection ready alongside closed.
  265. stop := make(chan struct{})
  266. done := make(chan struct{})
  267. go func() {
  268. defer close(done)
  269. for {
  270. server, client := net.Pipe()
  271. _ = client.Close()
  272. select {
  273. case l.accept <- server:
  274. case <-stop:
  275. _ = server.Close()
  276. return
  277. }
  278. }
  279. }()
  280. defer func() {
  281. close(stop)
  282. <-done
  283. }()
  284. for range 100 {
  285. time.Sleep(time.Millisecond)
  286. conn, err := l.Accept()
  287. if conn != nil {
  288. _ = conn.Close()
  289. }
  290. if !errors.Is(err, net.ErrClosed) {
  291. t.Fatalf("Accept after Close: got err %v, want net.ErrClosed", err)
  292. }
  293. }
  294. }
  295. // Close must release blocked Accept calls and let Wait report a clean shutdown.
  296. func TestServiceClose(t *testing.T) {
  297. ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
  298. a := newSimpleService(ca, caKey, "a", netip.MustParseAddr("10.0.0.1"), m{
  299. "static_host_map": m{},
  300. "listen": m{
  301. "host": "127.0.0.1",
  302. "port": 0,
  303. },
  304. })
  305. ln, err := a.Listen("tcp", ":1234")
  306. if err != nil {
  307. t.Fatal(err)
  308. }
  309. accepted := make(chan error, 1)
  310. go func() {
  311. _, err := ln.Accept()
  312. accepted <- err
  313. }()
  314. if err := a.Close(); err != nil {
  315. t.Fatal(err)
  316. }
  317. select {
  318. case err := <-accepted:
  319. if !errors.Is(err, net.ErrClosed) {
  320. t.Fatalf("expected net.ErrClosed from Accept, got %v", err)
  321. }
  322. case <-time.After(5 * time.Second):
  323. t.Fatal("Accept still blocked after Close")
  324. }
  325. waited := make(chan error, 1)
  326. go func() { waited <- a.Wait() }()
  327. select {
  328. case err := <-waited:
  329. if err != nil {
  330. t.Fatalf("expected clean shutdown from Wait, got %v", err)
  331. }
  332. case <-time.After(5 * time.Second):
  333. t.Fatal("Wait still blocked after Close")
  334. }
  335. if a.ipstack.HasNIC(nicID) {
  336. t.Fatal("netstack not torn down after Wait")
  337. }
  338. if err := a.Close(); err != nil {
  339. t.Fatalf("second close: %v", err)
  340. }
  341. if _, err := a.Listen("tcp", ":1235"); !errors.Is(err, net.ErrClosed) {
  342. t.Fatalf("expected net.ErrClosed from Listen after Close, got %v", err)
  343. }
  344. }
  345. // If nebula stops without Close being called, the service must still release
  346. // Accept and tear down the netstack.
  347. func TestServiceNebulaStopped(t *testing.T) {
  348. ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
  349. a := newSimpleService(ca, caKey, "a", netip.MustParseAddr("10.0.0.1"), m{
  350. "static_host_map": m{},
  351. "listen": m{
  352. "host": "127.0.0.1",
  353. "port": 0,
  354. },
  355. })
  356. ln, err := a.Listen("tcp", ":1234")
  357. if err != nil {
  358. t.Fatal(err)
  359. }
  360. accepted := make(chan error, 1)
  361. go func() {
  362. _, err := ln.Accept()
  363. accepted <- err
  364. }()
  365. a.control.Stop()
  366. waited := make(chan error, 1)
  367. go func() { waited <- a.Wait() }()
  368. select {
  369. case err := <-waited:
  370. if err != nil {
  371. t.Fatalf("expected clean shutdown from Wait, got %v", err)
  372. }
  373. case <-time.After(5 * time.Second):
  374. t.Fatal("Wait still blocked after nebula stopped")
  375. }
  376. select {
  377. case err := <-accepted:
  378. if !errors.Is(err, net.ErrClosed) {
  379. t.Fatalf("expected net.ErrClosed from Accept, got %v", err)
  380. }
  381. case <-time.After(5 * time.Second):
  382. t.Fatal("Accept still blocked after nebula stopped")
  383. }
  384. if a.ipstack.HasNIC(nicID) {
  385. t.Fatal("netstack not torn down after Wait")
  386. }
  387. }
  388. // Dials racing Close must not hold Wait until their connect times out.
  389. func TestServiceDialDuringClose(t *testing.T) {
  390. ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
  391. a := newSimpleService(ca, caKey, "a", netip.MustParseAddr("10.0.0.1"), m{
  392. "static_host_map": m{},
  393. "listen": m{
  394. "host": "127.0.0.1",
  395. "port": 0,
  396. },
  397. })
  398. stop := make(chan struct{})
  399. var eg errgroup.Group
  400. for range 8 {
  401. eg.Go(func() error {
  402. for {
  403. select {
  404. case <-stop:
  405. return nil
  406. default:
  407. }
  408. // 10.0.0.2 has no tunnel, so each dial sits in SYN-SENT until aborted
  409. if c, err := a.Dial("tcp", "10.0.0.2:80"); err == nil {
  410. _ = c.Close()
  411. }
  412. }
  413. })
  414. }
  415. time.Sleep(50 * time.Millisecond)
  416. if err := a.Close(); err != nil {
  417. t.Fatal(err)
  418. }
  419. waited := make(chan error, 1)
  420. go func() { waited <- a.Wait() }()
  421. select {
  422. case err := <-waited:
  423. if err != nil {
  424. t.Fatalf("expected clean shutdown from Wait, got %v", err)
  425. }
  426. case <-time.After(5 * time.Second):
  427. t.Fatal("Wait blocked behind a dial that raced Close")
  428. }
  429. close(stop)
  430. dialersDone := make(chan error, 1)
  431. go func() { dialersDone <- eg.Wait() }()
  432. select {
  433. case <-dialersDone:
  434. case <-time.After(5 * time.Second):
  435. t.Fatal("Dial still blocked after Wait")
  436. }
  437. }