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         // Get the port number we are listening on (extPort might be
 172         // "0" or a port name, in which case this will be different).
 173         _, listenPort, err := net.SplitHostPort(srv.Addr)
 174         if err != nil {
 175                 return err
 176         }
 177         // When changing state to Running, the caller will want to set
 178         // gateway_address to a "HOST:PORT" that, if controller
 179         // connects to it, will reach this gateway server.
 180         //
 181         // The most likely thing to work is: HOST is our external
 182         // hostname/IP as provided by the caller
 183         // (arvados-dispatch-cloud) or 127.0.0.1 to indicate
 184         // non-tunnel connections aren't available; and PORT is the
 185         // port number we are listening on.
 186         gw.Address = net.JoinHostPort(extHost, listenPort)
 187         if gw.ArvadosClient != nil {
 188                 go gw.maintainTunnel(gw.Address)
 189         }
 190         return nil
 191 }
 192
 193 func (gw *Gateway) maintainTunnel(addr string) {
 194         for ; ; time.Sleep(5 * time.Second) {
 195                 err := gw.runTunnel(addr)
 196                 gw.Log.Printf("runTunnel: %s", err)
 197         }
 198 }
 199
 200 // runTunnel connects to controller and sets up a tunnel through
 201 // which controller can connect to the gateway server at the given
 202 // addr.
 203 func (gw *Gateway) runTunnel(addr string) error {
 204         ctx := auth.NewContext(context.Background(), auth.NewCredentials(gw.ArvadosClient.AuthToken))
 205         arpc := rpc.NewConn("", &url.URL{Scheme: "https", Host: gw.ArvadosClient.APIHost}, gw.ArvadosClient.Insecure, rpc.PassthroughTokenProvider)
 206         tun, err := arpc.ContainerGatewayTunnel(ctx, arvados.ContainerGatewayTunnelOptions{
 207                 UUID:       gw.ContainerUUID,
 208                 AuthSecret: gw.AuthSecret,
 209         })
 210         if err != nil {
 211                 return fmt.Errorf("error creating gateway tunnel: %s", err)
 212         }
 213         mux, err := yamux.Client(tun.Conn, nil)
 214         if err != nil {
 215                 return fmt.Errorf("error setting up mux client end: %s", err)
 216         }
 217         if url := tun.Header.Get("X-Arvados-Internal-Url"); url != "" && gw.UpdateTunnelURL != nil {
 218                 gw.UpdateTunnelURL(url)
 219         }
 220         for {
 221                 muxconn, err := mux.Accept()
 222                 if err != nil {
 223                         return err
 224                 }
 225                 gw.Log.Printf("receiving connection from tunnel, remoteAddr %s", muxconn.RemoteAddr().String())
 226                 go func() {
 227                         defer muxconn.Close()
 228                         gwconn, err := net.Dial("tcp", addr)
 229                         if err != nil {
 230                                 gw.Log.Printf("error connecting to %s on behalf of tunnel connection: %s", addr, err)
 231                                 return
 232                         }
 233                         defer gwconn.Close()
 234                         var wg sync.WaitGroup
 235                         wg.Add(2)
 236                         go func() {
 237                                 defer wg.Done()
 238                                 io.Copy(gwconn, muxconn)
 239                         }()
 240                         go func() {
 241                                 defer wg.Done()
 242                                 io.Copy(muxconn, gwconn)
 243                         }()
 244                         wg.Wait()
 245                 }()
 246         }
 247 }
 248
 249 // handleSSH connects to an SSH server that allows the caller to run
 250 // interactive commands as root (or any other desired user) inside the
 251 // container. The tunnel itself can only be created by an
 252 // authenticated caller, so the SSH server itself is wide open (any
 253 // password or key will be accepted).
 254 //
 255 // Requests must have path "/ssh" and the following headers:
 256 //
 257 // Connection: upgrade
 258 // Upgrade: ssh
 259 // X-Arvados-Target-Uuid: uuid of container
 260 // X-Arvados-Authorization: must match
 261 // hmac(AuthSecret,certfingerprint) (this prevents other containers
 262 // and shell nodes from connecting directly)
 263 //
 264 // Optional headers:
 265 //
 266 // X-Arvados-Detach-Keys: argument to "docker exec --detach-keys",
 267 // e.g., "ctrl-p,ctrl-q"
 268 // X-Arvados-Login-Username: argument to "docker exec --user": account
 269 // used to run command(s) inside the container.
 270 func (gw *Gateway) handleSSH(w http.ResponseWriter, req *http.Request) {
 271         // In future we'll handle browser traffic too, but for now the
 272         // only traffic we expect is an SSH tunnel from
 273         // (*lib/controller/localdb.Conn)ContainerSSH()
 274         if req.Method != "POST" || req.Header.Get("Upgrade") != "ssh" {
 275                 http.Error(w, "path not found", http.StatusNotFound)
 276                 return
 277         }
 278         req.ParseForm()
 279         if want := req.Form.Get("uuid"); want != gw.ContainerUUID {
 280                 http.Error(w, fmt.Sprintf("misdirected request: meant for %q but received by crunch-run %q", want, gw.ContainerUUID), http.StatusBadGateway)
 281                 return
 282         }
 283         if req.Header.Get("X-Arvados-Authorization") != gw.requestAuth {
 284                 http.Error(w, "bad X-Arvados-Authorization header", http.StatusUnauthorized)
 285                 return
 286         }
 287         detachKeys := req.Form.Get("detach_keys")
 288         username := req.Form.Get("login_username")
 289         if username == "" {
 290                 username = "root"
 291         }
 292         hj, ok := w.(http.Hijacker)
 293         if !ok {
 294                 http.Error(w, "ResponseWriter does not support connection upgrade", http.StatusInternalServerError)
 295                 return
 296         }
 297         netconn, _, err := hj.Hijack()
 298         if !ok {
 299                 http.Error(w, err.Error(), http.StatusInternalServerError)
 300                 return
 301         }
 302         defer netconn.Close()
 303         w.Header().Set("Connection", "upgrade")
 304         w.Header().Set("Upgrade", "ssh")
 305         w.Header().Set("X-Arvados-Authorization-Response", gw.respondAuth)
 306         netconn.Write([]byte("HTTP/1.1 101 Switching Protocols\r\n"))
 307         w.Header().Write(netconn)
 308         netconn.Write([]byte("\r\n"))
 309
 310         ctx := req.Context()
 311
 312         conn, newchans, reqs, err := ssh.NewServerConn(netconn, &gw.sshConfig)
 313         if err == io.EOF {
 314                 return
 315         } else if err != nil {
 316                 gw.Log.Printf("ssh.NewServerConn: %s", err)
 317                 return
 318         }
 319         defer conn.Close()
 320         go ssh.DiscardRequests(reqs)
 321         for newch := range newchans {
 322                 switch newch.ChannelType() {
 323                 case "direct-tcpip":
 324                         go gw.handleDirectTCPIP(ctx, newch)
 325                 case "session":
 326                         go gw.handleSession(ctx, newch, detachKeys, username)
 327                 default:
 328                         go newch.Reject(ssh.UnknownChannelType, fmt.Sprintf("unsupported channel type %q", newch.ChannelType()))
 329                 }
 330         }
 331 }
 332
 333 func (gw *Gateway) handleDirectTCPIP(ctx context.Context, newch ssh.NewChannel) {
 334         ch, reqs, err := newch.Accept()
 335         if err != nil {
 336                 gw.Log.Printf("accept direct-tcpip channel: %s", err)
 337                 return
 338         }
 339         defer ch.Close()
 340         go ssh.DiscardRequests(reqs)
 341
 342         // RFC 4254 7.2 (copy of channelOpenDirectMsg in
 343         // golang.org/x/crypto/ssh)
 344         var msg struct {
 345                 Raddr string
 346                 Rport uint32
 347                 Laddr string
 348                 Lport uint32
 349         }
 350         err = ssh.Unmarshal(newch.ExtraData(), &msg)
 351         if err != nil {
 352                 fmt.Fprintf(ch.Stderr(), "unmarshal direct-tcpip extradata: %s\n", err)
 353                 return
 354         }
 355         switch msg.Raddr {
 356         case "localhost", "0.0.0.0", "127.0.0.1", "::1", "::":
 357         default:
 358                 fmt.Fprintf(ch.Stderr(), "cannot forward to ports on %q, only localhost\n", msg.Raddr)
 359                 return
 360         }
 361
 362         dstaddr, err := gw.Target.IPAddress()
 363         if err != nil {
 364                 fmt.Fprintf(ch.Stderr(), "container has no IP address: %s\n", err)
 365                 return
 366         } else if dstaddr == "" {
 367                 fmt.Fprintf(ch.Stderr(), "container has no IP address\n")
 368                 return
 369         }
 370
 371         dst := net.JoinHostPort(dstaddr, fmt.Sprintf("%d", msg.Rport))
 372         tcpconn, err := net.Dial("tcp", dst)
 373         if err != nil {
 374                 fmt.Fprintf(ch.Stderr(), "%s: %s\n", dst, err)
 375                 return
 376         }
 377         go func() {
 378                 n, _ := io.Copy(ch, tcpconn)
 379                 ctxlog.FromContext(ctx).Debugf("tcpip: sent %d bytes\n", n)
 380                 ch.CloseWrite()
 381         }()
 382         n, _ := io.Copy(tcpconn, ch)
 383         ctxlog.FromContext(ctx).Debugf("tcpip: received %d bytes\n", n)
 384 }
 385
 386 func (gw *Gateway) handleSession(ctx context.Context, newch ssh.NewChannel, detachKeys, username string) {
 387         ch, reqs, err := newch.Accept()
 388         if err != nil {
 389                 gw.Log.Printf("accept session channel: %s", err)
 390                 return
 391         }
 392         var pty0, tty0 *os.File
 393         // Where to send errors/messages for the client to see
 394         logw := io.Writer(ch.Stderr())
 395         // How to end lines when sending errors/messages to the client
 396         // (changes to \r\n when using a pty)
 397         eol := "\n"
 398         // Env vars to add to child process
 399         termEnv := []string(nil)
 400         for req := range reqs {
 401                 ok := false
 402                 switch req.Type {
 403                 case "shell", "exec":
 404                         ok = true
 405                         var payload struct {
 406                                 Command string
 407                         }
 408                         ssh.Unmarshal(req.Payload, &payload)
 409                         execargs, err := shlex.Split(payload.Command)
 410                         if err != nil {
 411                                 fmt.Fprintf(logw, "error parsing supplied command: %s"+eol, err)
 412                                 return
 413                         }
 414                         if len(execargs) == 0 {
 415                                 execargs = []string{"/bin/bash", "-login"}
 416                         }
 417                         go func() {
 418                                 var resp struct {
 419                                         Status uint32
 420                                 }
 421                                 defer func() {
 422                                         ch.SendRequest("exit-status", false, ssh.Marshal(&resp))
 423                                         ch.Close()
 424                                 }()
 425
 426                                 cmd, err := gw.Target.InjectCommand(ctx, detachKeys, username, tty0 != nil, execargs)
 427                                 if err != nil {
 428                                         fmt.Fprintln(ch.Stderr(), err)
 429                                         ch.CloseWrite()
 430                                         resp.Status = 1
 431                                         return
 432                                 }
 433                                 cmd.Stdin = ch
 434                                 cmd.Stdout = ch
 435                                 cmd.Stderr = ch.Stderr()
 436                                 if tty0 != nil {
 437                                         cmd.Stdin = tty0
 438                                         cmd.Stdout = tty0
 439                                         cmd.Stderr = tty0
 440                                         var wg sync.WaitGroup
 441                                         defer wg.Wait()
 442                                         wg.Add(2)
 443                                         go func() { io.Copy(ch, pty0); wg.Done() }()
 444                                         go func() { io.Copy(pty0, ch); wg.Done() }()
 445                                         // Send our own debug messages to tty as well.
 446                                         logw = tty0
 447                                 }
 448                                 cmd.SysProcAttr = &syscall.SysProcAttr{
 449                                         Setctty: tty0 != nil,
 450                                         Setsid:  true,
 451                                 }
 452                                 cmd.Env = append(os.Environ(), termEnv...)
 453                                 err = cmd.Run()
 454                                 if exiterr, ok := err.(*exec.ExitError); ok {
 455                                         if status, ok := exiterr.Sys().(syscall.WaitStatus); ok {
 456                                                 resp.Status = uint32(status.ExitStatus())
 457                                         }
 458                                 } else if err != nil {
 459                                         // Propagate errors like `exec: "docker": executable file not found in $PATH`
 460                                         fmt.Fprintln(ch.Stderr(), err)
 461                                 }
 462                                 errClose := ch.CloseWrite()
 463                                 if resp.Status == 0 && (err != nil || errClose != nil) {
 464                                         resp.Status = 1
 465                                 }
 466                         }()
 467                 case "pty-req":
 468                         eol = "\r\n"
 469                         p, t, err := pty.Open()
 470                         if err != nil {
 471                                 fmt.Fprintf(ch.Stderr(), "pty failed: %s"+eol, err)
 472                                 break
 473                         }
 474                         defer p.Close()
 475                         defer t.Close()
 476                         pty0, tty0 = p, t
 477                         ok = true
 478                         var payload struct {
 479                                 Term string
 480                                 Cols uint32
 481                                 Rows uint32
 482                                 X    uint32
 483                                 Y    uint32
 484                         }
 485                         ssh.Unmarshal(req.Payload, &payload)
 486                         termEnv = []string{"TERM=" + payload.Term, "USE_TTY=1"}
 487                         err = pty.Setsize(pty0, &pty.Winsize{Rows: uint16(payload.Rows), Cols: uint16(payload.Cols), X: uint16(payload.X), Y: uint16(payload.Y)})
 488                         if err != nil {
 489                                 fmt.Fprintf(logw, "pty-req: setsize failed: %s"+eol, err)
 490                         }
 491                 case "window-change":
 492                         var payload struct {
 493                                 Cols uint32
 494                                 Rows uint32
 495                                 X    uint32
 496                                 Y    uint32
 497                         }
 498                         ssh.Unmarshal(req.Payload, &payload)
 499                         err := pty.Setsize(pty0, &pty.Winsize{Rows: uint16(payload.Rows), Cols: uint16(payload.Cols), X: uint16(payload.X), Y: uint16(payload.Y)})
 500                         if err != nil {
 501                                 fmt.Fprintf(logw, "window-change: setsize failed: %s"+eol, err)
 502                                 break
 503                         }
 504                         ok = true
 505                 case "env":
 506                         // TODO: implement "env"
 507                         // requests by setting env
 508                         // vars in the docker-exec
 509                         // command (not docker-exec's
 510                         // own environment, which
 511                         // would be a gaping security
 512                         // hole).
 513                 default:
 514                         // fmt.Fprintf(logw, "declining %q req"+eol, req.Type)
 515                 }
 516                 if req.WantReply {
 517                         req.Reply(ok, nil)
 518                 }
 519         }
 520 }