lib/crunchrun/container_gateway.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package crunchrun
   6
   7 import (
   8         "crypto/hmac"
   9         "crypto/rand"
  10         "crypto/rsa"
  11         "crypto/sha256"
  12         "crypto/tls"
  13         "fmt"
  14         "io"
  15         "net"
  16         "net/http"
  17         "net/url"
  18         "os"
  19         "os/exec"
  20         "sync"
  21         "syscall"
  22         "time"
  23
  24         "git.arvados.org/arvados.git/lib/controller/rpc"
  25         "git.arvados.org/arvados.git/lib/selfsigned"
  26         "git.arvados.org/arvados.git/sdk/go/arvados"
  27         "git.arvados.org/arvados.git/sdk/go/auth"
  28         "git.arvados.org/arvados.git/sdk/go/ctxlog"
  29         "git.arvados.org/arvados.git/sdk/go/httpserver"
  30         "github.com/creack/pty"
  31         "github.com/google/shlex"
  32         "github.com/hashicorp/yamux"
  33         "golang.org/x/crypto/ssh"
  34         "golang.org/x/net/context"
  35 )
  36
  37 type GatewayTarget interface {
  38         // Command that will execute cmd inside the container
  39         InjectCommand(ctx context.Context, detachKeys, username string, usingTTY bool, cmd []string) (*exec.Cmd, error)
  40
  41         // IP address inside container
  42         IPAddress() (string, error)
  43 }
  44
  45 type GatewayTargetStub struct{}
  46
  47 func (GatewayTargetStub) IPAddress() (string, error) { return "127.0.0.1", nil }
  48 func (GatewayTargetStub) InjectCommand(ctx context.Context, detachKeys, username string, usingTTY bool, cmd []string) (*exec.Cmd, error) {
  49         return exec.CommandContext(ctx, cmd[0], cmd[1:]...), nil
  50 }
  51
  52 type Gateway struct {
  53         ContainerUUID string
  54         // Caller should set Address to "", or "host:0" or "host:port"
  55         // where host is a known external IP address; port is a
  56         // desired port number to listen on; and ":0" chooses an
  57         // available dynamic port.
  58         //
  59         // If Address is "", Start() listens only on the loopback
  60         // interface (and changes Address to "127.0.0.1:port").
  61         // Otherwise it listens on all interfaces.
  62         //
  63         // If Address is "host:0", Start() updates Address to
  64         // "host:port".
  65         Address    string
  66         AuthSecret string
  67         Target     GatewayTarget
  68         Log        interface {
  69                 Printf(fmt string, args ...interface{})
  70         }
  71         // If non-nil, set up a ContainerGatewayTunnel, so that the
  72         // controller can connect to us even if our external IP
  73         // address is unknown or not routable from controller.
  74         ArvadosClient *arvados.Client
  75
  76         // When a tunnel is connected or reconnected, this func (if
  77         // not nil) will be called with the InternalURL of the
  78         // controller process at the other end of the tunnel.
  79         UpdateTunnelURL func(url string)
  80
  81         sshConfig   ssh.ServerConfig
  82         requestAuth string
  83         respondAuth string
  84 }
  85
  86 // Start starts an http server that allows authenticated clients to open an
  87 // interactive "docker exec" session and (in future) connect to tcp ports
  88 // inside the docker container.
  89 func (gw *Gateway) Start() error {
  90         gw.sshConfig = ssh.ServerConfig{
  91                 NoClientAuth: true,
  92                 PasswordCallback: func(c ssh.ConnMetadata, pass []byte) (*ssh.Permissions, error) {
  93                         if c.User() == "_" {
  94                                 return nil, nil
  95                         }
  96                         return nil, fmt.Errorf("cannot specify user %q via ssh client", c.User())
  97                 },
  98                 PublicKeyCallback: func(c ssh.ConnMetadata, pubKey ssh.PublicKey) (*ssh.Permissions, error) {
  99                         if c.User() == "_" {
 100                                 return &ssh.Permissions{
 101                                         Extensions: map[string]string{
 102                                                 "pubkey-fp": ssh.FingerprintSHA256(pubKey),
 103                                         },
 104                                 }, nil
 105                         }
 106                         return nil, fmt.Errorf("cannot specify user %q via ssh client", c.User())
 107                 },
 108         }
 109         pvt, err := rsa.GenerateKey(rand.Reader, 2048)
 110         if err != nil {
 111                 return err
 112         }
 113         err = pvt.Validate()
 114         if err != nil {
 115                 return err
 116         }
 117         signer, err := ssh.NewSignerFromKey(pvt)
 118         if err != nil {
 119                 return err
 120         }
 121         gw.sshConfig.AddHostKey(signer)
 122
 123         // Address (typically provided by arvados-dispatch-cloud) is
 124         // HOST:PORT where HOST is our IP address or hostname as seen
 125         // from arvados-controller, and PORT is either the desired
 126         // port where we should run our gateway server, or "0" if we
 127         // should choose an available port.
 128         extAddr := gw.Address
 129         // Generally we can't know which local interface corresponds
 130         // to an externally reachable IP address, so if we expect to
 131         // be reachable by external hosts, we listen on all
 132         // interfaces.
 133         listenHost := ""
 134         if extAddr == "" {
 135                 // If the dispatcher doesn't tell us our external IP
 136                 // address, controller will only be able to connect
 137                 // through the tunnel (see runTunnel), so our gateway
 138                 // server only needs to listen on the loopback
 139                 // interface.
 140                 extAddr = "127.0.0.1:0"
 141                 listenHost = "127.0.0.1"
 142         }
 143         extHost, extPort, err := net.SplitHostPort(extAddr)
 144         if err != nil {
 145                 return err
 146         }
 147         cert, err := selfsigned.CertGenerator{}.Generate()
 148         if err != nil {
 149                 return err
 150         }
 151         h := hmac.New(sha256.New, []byte(gw.AuthSecret))
 152         h.Write(cert.Certificate[0])
 153         gw.requestAuth = fmt.Sprintf("%x", h.Sum(nil))
 154         h.Reset()
 155         h.Write([]byte(gw.requestAuth))
 156         gw.respondAuth = fmt.Sprintf("%x", h.Sum(nil))
 157
 158         srv := &httpserver.Server{
 159                 Server: http.Server{
 160                         Handler: http.HandlerFunc(gw.handleSSH),
 161                         TLSConfig: &tls.Config{
 162                                 Certificates: []tls.Certificate{cert},
 163                         },
 164                 },
 165                 Addr: net.JoinHostPort(listenHost, extPort),
 166         }
 167         err = srv.Start()
 168         if err != nil {
 169                 return err
 170         }
 171         go func() {
 172                 err := srv.Wait()
 173                 gw.Log.Printf("gateway server stopped: %s", err)
 174         }()
 175         // Get the port number we are listening on (extPort might be
 176         // "0" or a port name, in which case this will be different).
 177         _, listenPort, err := net.SplitHostPort(srv.Addr)
 178         if err != nil {
 179                 return err
 180         }
 181         // When changing state to Running, the caller will want to set
 182         // gateway_address to a "HOST:PORT" that, if controller
 183         // connects to it, will reach this gateway server.
 184         //
 185         // The most likely thing to work is: HOST is our external
 186         // hostname/IP as provided by the caller
 187         // (arvados-dispatch-cloud) or 127.0.0.1 to indicate
 188         // non-tunnel connections aren't available; and PORT is the
 189         // port number we are listening on.
 190         gw.Address = net.JoinHostPort(extHost, listenPort)
 191         gw.Log.Printf("gateway server listening at %s", gw.Address)
 192         if gw.ArvadosClient != nil {
 193                 go gw.maintainTunnel(gw.Address)
 194         }
 195         return nil
 196 }
 197
 198 func (gw *Gateway) maintainTunnel(addr string) {
 199         for ; ; time.Sleep(5 * time.Second) {
 200                 err := gw.runTunnel(addr)
 201                 gw.Log.Printf("runTunnel: %s", err)
 202         }
 203 }
 204
 205 // runTunnel connects to controller and sets up a tunnel through
 206 // which controller can connect to the gateway server at the given
 207 // addr.
 208 func (gw *Gateway) runTunnel(addr string) error {
 209         ctx := auth.NewContext(context.Background(), auth.NewCredentials(gw.ArvadosClient.AuthToken))
 210         arpc := rpc.NewConn("", &url.URL{Scheme: "https", Host: gw.ArvadosClient.APIHost}, gw.ArvadosClient.Insecure, rpc.PassthroughTokenProvider)
 211         tun, err := arpc.ContainerGatewayTunnel(ctx, arvados.ContainerGatewayTunnelOptions{
 212                 UUID:       gw.ContainerUUID,
 213                 AuthSecret: gw.AuthSecret,
 214         })
 215         if err != nil {
 216                 return fmt.Errorf("error creating gateway tunnel: %s", err)
 217         }
 218         mux, err := yamux.Client(tun.Conn, nil)
 219         if err != nil {
 220                 return fmt.Errorf("error setting up mux client end: %s", err)
 221         }
 222         if url := tun.Header.Get("X-Arvados-Internal-Url"); url != "" && gw.UpdateTunnelURL != nil {
 223                 gw.UpdateTunnelURL(url)
 224         }
 225         for {
 226                 muxconn, err := mux.AcceptStream()
 227                 if err != nil {
 228                         return err
 229                 }
 230                 gw.Log.Printf("tunnel connection %d started", muxconn.StreamID())
 231                 go func() {
 232                         defer muxconn.Close()
 233                         gwconn, err := net.Dial("tcp", addr)
 234                         if err != nil {
 235                                 gw.Log.Printf("tunnel connection %d: error connecting to %s: %s", muxconn.StreamID(), addr, err)
 236                                 return
 237                         }
 238                         defer gwconn.Close()
 239                         var wg sync.WaitGroup
 240                         wg.Add(2)
 241                         go func() {
 242                                 defer wg.Done()
 243                                 _, err := io.Copy(gwconn, muxconn)
 244                                 if err != nil {
 245                                         gw.Log.Printf("tunnel connection %d: tunnel: %s", muxconn.StreamID(), err)
 246                                 }
 247                                 muxconn.Close()
 248                                 gwconn.Close()
 249                         }()
 250                         go func() {
 251                                 defer wg.Done()
 252                                 _, err := io.Copy(muxconn, gwconn)
 253                                 if err != nil {
 254                                         gw.Log.Printf("tunnel connection %d: gateway: %s", muxconn.StreamID(), err)
 255                                 }
 256                                 gwconn.Close()
 257                                 muxconn.Close()
 258                         }()
 259                         wg.Wait()
 260                         gw.Log.Printf("tunnel connection %d finished", muxconn.StreamID())
 261                 }()
 262         }
 263 }
 264
 265 // handleSSH connects to an SSH server that allows the caller to run
 266 // interactive commands as root (or any other desired user) inside the
 267 // container. The tunnel itself can only be created by an
 268 // authenticated caller, so the SSH server itself is wide open (any
 269 // password or key will be accepted).
 270 //
 271 // Requests must have path "/ssh" and the following headers:
 272 //
 273 // Connection: upgrade
 274 // Upgrade: ssh
 275 // X-Arvados-Target-Uuid: uuid of container
 276 // X-Arvados-Authorization: must match
 277 // hmac(AuthSecret,certfingerprint) (this prevents other containers
 278 // and shell nodes from connecting directly)
 279 //
 280 // Optional headers:
 281 //
 282 // X-Arvados-Detach-Keys: argument to "docker exec --detach-keys",
 283 // e.g., "ctrl-p,ctrl-q"
 284 // X-Arvados-Login-Username: argument to "docker exec --user": account
 285 // used to run command(s) inside the container.
 286 func (gw *Gateway) handleSSH(w http.ResponseWriter, req *http.Request) {
 287         // In future we'll handle browser traffic too, but for now the
 288         // only traffic we expect is an SSH tunnel from
 289         // (*lib/controller/localdb.Conn)ContainerSSH()
 290         if req.Method != "POST" || req.Header.Get("Upgrade") != "ssh" {
 291                 http.Error(w, "path not found", http.StatusNotFound)
 292                 return
 293         }
 294         req.ParseForm()
 295         if want := req.Form.Get("uuid"); want != gw.ContainerUUID {
 296                 http.Error(w, fmt.Sprintf("misdirected request: meant for %q but received by crunch-run %q", want, gw.ContainerUUID), http.StatusBadGateway)
 297                 return
 298         }
 299         if req.Header.Get("X-Arvados-Authorization") != gw.requestAuth {
 300                 http.Error(w, "bad X-Arvados-Authorization header", http.StatusUnauthorized)
 301                 return
 302         }
 303         detachKeys := req.Form.Get("detach_keys")
 304         username := req.Form.Get("login_username")
 305         if username == "" {
 306                 username = "root"
 307         }
 308         hj, ok := w.(http.Hijacker)
 309         if !ok {
 310                 http.Error(w, "ResponseWriter does not support connection upgrade", http.StatusInternalServerError)
 311                 return
 312         }
 313         netconn, _, err := hj.Hijack()
 314         if !ok {
 315                 http.Error(w, err.Error(), http.StatusInternalServerError)
 316                 return
 317         }
 318         defer netconn.Close()
 319         w.Header().Set("Connection", "upgrade")
 320         w.Header().Set("Upgrade", "ssh")
 321         w.Header().Set("X-Arvados-Authorization-Response", gw.respondAuth)
 322         netconn.Write([]byte("HTTP/1.1 101 Switching Protocols\r\n"))
 323         w.Header().Write(netconn)
 324         netconn.Write([]byte("\r\n"))
 325
 326         ctx := req.Context()
 327
 328         conn, newchans, reqs, err := ssh.NewServerConn(netconn, &gw.sshConfig)
 329         if err == io.EOF {
 330                 return
 331         } else if err != nil {
 332                 gw.Log.Printf("ssh.NewServerConn: %s", err)
 333                 return
 334         }
 335         defer conn.Close()
 336         go ssh.DiscardRequests(reqs)
 337         for newch := range newchans {
 338                 switch newch.ChannelType() {
 339                 case "direct-tcpip":
 340                         go gw.handleDirectTCPIP(ctx, newch)
 341                 case "session":
 342                         go gw.handleSession(ctx, newch, detachKeys, username)
 343                 default:
 344                         go newch.Reject(ssh.UnknownChannelType, fmt.Sprintf("unsupported channel type %q", newch.ChannelType()))
 345                 }
 346         }
 347 }
 348
 349 func (gw *Gateway) handleDirectTCPIP(ctx context.Context, newch ssh.NewChannel) {
 350         ch, reqs, err := newch.Accept()
 351         if err != nil {
 352                 gw.Log.Printf("accept direct-tcpip channel: %s", err)
 353                 return
 354         }
 355         defer ch.Close()
 356         go ssh.DiscardRequests(reqs)
 357
 358         // RFC 4254 7.2 (copy of channelOpenDirectMsg in
 359         // golang.org/x/crypto/ssh)
 360         var msg struct {
 361                 Raddr string
 362                 Rport uint32
 363                 Laddr string
 364                 Lport uint32
 365         }
 366         err = ssh.Unmarshal(newch.ExtraData(), &msg)
 367         if err != nil {
 368                 fmt.Fprintf(ch.Stderr(), "unmarshal direct-tcpip extradata: %s\n", err)
 369                 return
 370         }
 371         switch msg.Raddr {
 372         case "localhost", "0.0.0.0", "127.0.0.1", "::1", "::":
 373         default:
 374                 fmt.Fprintf(ch.Stderr(), "cannot forward to ports on %q, only localhost\n", msg.Raddr)
 375                 return
 376         }
 377
 378         dstaddr, err := gw.Target.IPAddress()
 379         if err != nil {
 380                 fmt.Fprintf(ch.Stderr(), "container has no IP address: %s\n", err)
 381                 return
 382         } else if dstaddr == "" {
 383                 fmt.Fprintf(ch.Stderr(), "container has no IP address\n")
 384                 return
 385         }
 386
 387         dst := net.JoinHostPort(dstaddr, fmt.Sprintf("%d", msg.Rport))
 388         tcpconn, err := net.Dial("tcp", dst)
 389         if err != nil {
 390                 fmt.Fprintf(ch.Stderr(), "%s: %s\n", dst, err)
 391                 return
 392         }
 393         go func() {
 394                 n, _ := io.Copy(ch, tcpconn)
 395                 ctxlog.FromContext(ctx).Debugf("tcpip: sent %d bytes\n", n)
 396                 ch.CloseWrite()
 397         }()
 398         n, _ := io.Copy(tcpconn, ch)
 399         ctxlog.FromContext(ctx).Debugf("tcpip: received %d bytes\n", n)
 400 }
 401
 402 func (gw *Gateway) handleSession(ctx context.Context, newch ssh.NewChannel, detachKeys, username string) {
 403         ch, reqs, err := newch.Accept()
 404         if err != nil {
 405                 gw.Log.Printf("accept session channel: %s", err)
 406                 return
 407         }
 408         var pty0, tty0 *os.File
 409         // Where to send errors/messages for the client to see
 410         logw := io.Writer(ch.Stderr())
 411         // How to end lines when sending errors/messages to the client
 412         // (changes to \r\n when using a pty)
 413         eol := "\n"
 414         // Env vars to add to child process
 415         termEnv := []string(nil)
 416         for req := range reqs {
 417                 ok := false
 418                 switch req.Type {
 419                 case "shell", "exec":
 420                         ok = true
 421                         var payload struct {
 422                                 Command string
 423                         }
 424                         ssh.Unmarshal(req.Payload, &payload)
 425                         execargs, err := shlex.Split(payload.Command)
 426                         if err != nil {
 427                                 fmt.Fprintf(logw, "error parsing supplied command: %s"+eol, err)
 428                                 return
 429                         }
 430                         if len(execargs) == 0 {
 431                                 execargs = []string{"/bin/bash", "-login"}
 432                         }
 433                         go func() {
 434                                 var resp struct {
 435                                         Status uint32
 436                                 }
 437                                 defer func() {
 438                                         ch.SendRequest("exit-status", false, ssh.Marshal(&resp))
 439                                         ch.Close()
 440                                 }()
 441
 442                                 cmd, err := gw.Target.InjectCommand(ctx, detachKeys, username, tty0 != nil, execargs)
 443                                 if err != nil {
 444                                         fmt.Fprintln(ch.Stderr(), err)
 445                                         ch.CloseWrite()
 446                                         resp.Status = 1
 447                                         return
 448                                 }
 449                                 cmd.Stdin = ch
 450                                 cmd.Stdout = ch
 451                                 cmd.Stderr = ch.Stderr()
 452                                 if tty0 != nil {
 453                                         cmd.Stdin = tty0
 454                                         cmd.Stdout = tty0
 455                                         cmd.Stderr = tty0
 456                                         var wg sync.WaitGroup
 457                                         defer wg.Wait()
 458                                         wg.Add(2)
 459                                         go func() { io.Copy(ch, pty0); wg.Done() }()
 460                                         go func() { io.Copy(pty0, ch); wg.Done() }()
 461                                         // Send our own debug messages to tty as well.
 462                                         logw = tty0
 463                                 }
 464                                 cmd.SysProcAttr = &syscall.SysProcAttr{
 465                                         Setctty: tty0 != nil,
 466                                         Setsid:  true,
 467                                 }
 468                                 cmd.Env = append(os.Environ(), termEnv...)
 469                                 err = cmd.Run()
 470                                 if exiterr, ok := err.(*exec.ExitError); ok {
 471                                         if status, ok := exiterr.Sys().(syscall.WaitStatus); ok {
 472                                                 resp.Status = uint32(status.ExitStatus())
 473                                         }
 474                                 } else if err != nil {
 475                                         // Propagate errors like `exec: "docker": executable file not found in $PATH`
 476                                         fmt.Fprintln(ch.Stderr(), err)
 477                                 }
 478                                 errClose := ch.CloseWrite()
 479                                 if resp.Status == 0 && (err != nil || errClose != nil) {
 480                                         resp.Status = 1
 481                                 }
 482                         }()
 483                 case "pty-req":
 484                         eol = "\r\n"
 485                         p, t, err := pty.Open()
 486                         if err != nil {
 487                                 fmt.Fprintf(ch.Stderr(), "pty failed: %s"+eol, err)
 488                                 break
 489                         }
 490                         defer p.Close()
 491                         defer t.Close()
 492                         pty0, tty0 = p, t
 493                         ok = true
 494                         var payload struct {
 495                                 Term string
 496                                 Cols uint32
 497                                 Rows uint32
 498                                 X    uint32
 499                                 Y    uint32
 500                         }
 501                         ssh.Unmarshal(req.Payload, &payload)
 502                         termEnv = []string{"TERM=" + payload.Term, "USE_TTY=1"}
 503                         err = pty.Setsize(pty0, &pty.Winsize{Rows: uint16(payload.Rows), Cols: uint16(payload.Cols), X: uint16(payload.X), Y: uint16(payload.Y)})
 504                         if err != nil {
 505                                 fmt.Fprintf(logw, "pty-req: setsize failed: %s"+eol, err)
 506                         }
 507                 case "window-change":
 508                         var payload struct {
 509                                 Cols uint32
 510                                 Rows uint32
 511                                 X    uint32
 512                                 Y    uint32
 513                         }
 514                         ssh.Unmarshal(req.Payload, &payload)
 515                         err := pty.Setsize(pty0, &pty.Winsize{Rows: uint16(payload.Rows), Cols: uint16(payload.Cols), X: uint16(payload.X), Y: uint16(payload.Y)})
 516                         if err != nil {
 517                                 fmt.Fprintf(logw, "window-change: setsize failed: %s"+eol, err)
 518                                 break
 519                         }
 520                         ok = true
 521                 case "env":
 522                         // TODO: implement "env"
 523                         // requests by setting env
 524                         // vars in the docker-exec
 525                         // command (not docker-exec's
 526                         // own environment, which
 527                         // would be a gaping security
 528                         // hole).
 529                 default:
 530                         // fmt.Fprintf(logw, "declining %q req"+eol, req.Type)
 531                 }
 532                 if req.WantReply {
 533                         req.Reply(ok, nil)
 534                 }
 535         }
 536 }