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         "os"
  18         "os/exec"
  19         "sync"
  20         "syscall"
  21
  22         "git.arvados.org/arvados.git/lib/selfsigned"
  23         "git.arvados.org/arvados.git/sdk/go/httpserver"
  24         "github.com/creack/pty"
  25         "github.com/google/shlex"
  26         "golang.org/x/crypto/ssh"
  27 )
  28
  29 type Gateway struct {
  30         DockerContainerID *string
  31         ContainerUUID     string
  32         Address           string // listen host:port; if port=0, Start() will change it to the selected port
  33         AuthSecret        string
  34         Log               interface {
  35                 Printf(fmt string, args ...interface{})
  36         }
  37
  38         sshConfig   ssh.ServerConfig
  39         requestAuth string
  40         respondAuth string
  41 }
  42
  43 // startGatewayServer starts an http server that allows authenticated
  44 // clients to open an interactive "docker exec" session and (in
  45 // future) connect to tcp ports inside the docker container.
  46 func (gw *Gateway) Start() error {
  47         gw.sshConfig = ssh.ServerConfig{
  48                 NoClientAuth: true,
  49                 PasswordCallback: func(c ssh.ConnMetadata, pass []byte) (*ssh.Permissions, error) {
  50                         if c.User() == "_" {
  51                                 return nil, nil
  52                         } else {
  53                                 return nil, fmt.Errorf("cannot specify user %q via ssh client", c.User())
  54                         }
  55                 },
  56                 PublicKeyCallback: func(c ssh.ConnMetadata, pubKey ssh.PublicKey) (*ssh.Permissions, error) {
  57                         if c.User() == "_" {
  58                                 return &ssh.Permissions{
  59                                         Extensions: map[string]string{
  60                                                 "pubkey-fp": ssh.FingerprintSHA256(pubKey),
  61                                         },
  62                                 }, nil
  63                         } else {
  64                                 return nil, fmt.Errorf("cannot specify user %q via ssh client", c.User())
  65                         }
  66                 },
  67         }
  68         pvt, err := rsa.GenerateKey(rand.Reader, 2048)
  69         if err != nil {
  70                 return err
  71         }
  72         err = pvt.Validate()
  73         if err != nil {
  74                 return err
  75         }
  76         signer, err := ssh.NewSignerFromKey(pvt)
  77         if err != nil {
  78                 return err
  79         }
  80         gw.sshConfig.AddHostKey(signer)
  81
  82         // Address (typically provided by arvados-dispatch-cloud) is
  83         // HOST:PORT where HOST is our IP address or hostname as seen
  84         // from arvados-controller, and PORT is either the desired
  85         // port where we should run our gateway server, or "0" if we
  86         // should choose an available port.
  87         host, port, err := net.SplitHostPort(gw.Address)
  88         if err != nil {
  89                 return err
  90         }
  91         cert, err := selfsigned.CertGenerator{}.Generate()
  92         if err != nil {
  93                 return err
  94         }
  95         h := hmac.New(sha256.New, []byte(gw.AuthSecret))
  96         h.Write(cert.Certificate[0])
  97         gw.requestAuth = fmt.Sprintf("%x", h.Sum(nil))
  98         h.Reset()
  99         h.Write([]byte(gw.requestAuth))
 100         gw.respondAuth = fmt.Sprintf("%x", h.Sum(nil))
 101
 102         srv := &httpserver.Server{
 103                 Server: http.Server{
 104                         Handler: http.HandlerFunc(gw.handleSSH),
 105                         TLSConfig: &tls.Config{
 106                                 Certificates: []tls.Certificate{cert},
 107                         },
 108                 },
 109                 Addr: ":" + port,
 110         }
 111         err = srv.Start()
 112         if err != nil {
 113                 return err
 114         }
 115         // Get the port number we are listening on (the port might be
 116         // "0" or a port name, in which case this will be different).
 117         _, port, err = net.SplitHostPort(srv.Addr)
 118         if err != nil {
 119                 return err
 120         }
 121         // When changing state to Running, we will set
 122         // gateway_address to "HOST:PORT" where HOST is our
 123         // external hostname/IP as provided by arvados-dispatch-cloud,
 124         // and PORT is the port number we ended up listening on.
 125         gw.Address = net.JoinHostPort(host, port)
 126         return nil
 127 }
 128
 129 // handleSSH connects to an SSH server that runs commands as root in
 130 // the container. The tunnel itself can only be created by an
 131 // authenticated caller, so the SSH server itself is wide open (any
 132 // password or key will be accepted).
 133 //
 134 // Requests must have path "/ssh" and the following headers:
 135 //
 136 // Connection: upgrade
 137 // Upgrade: ssh
 138 // X-Arvados-Target-Uuid: uuid of container
 139 // X-Arvados-Authorization: must match
 140 // hmac(AuthSecret,certfingerprint) (this prevents other containers
 141 // and shell nodes from connecting directly)
 142 //
 143 // Optional header:
 144 //
 145 // X-Arvados-Detach-Keys: argument to "docker attach --detach-keys",
 146 // e.g., "ctrl-p,ctrl-q"
 147 func (gw *Gateway) handleSSH(w http.ResponseWriter, req *http.Request) {
 148         // In future we'll handle browser traffic too, but for now the
 149         // only traffic we expect is an SSH tunnel from
 150         // (*lib/controller/localdb.Conn)ContainerSSH()
 151         if req.Method != "GET" || req.Header.Get("Upgrade") != "ssh" {
 152                 http.Error(w, "path not found", http.StatusNotFound)
 153                 return
 154         }
 155         if want := req.Header.Get("X-Arvados-Target-Uuid"); want != gw.ContainerUUID {
 156                 http.Error(w, fmt.Sprintf("misdirected request: meant for %q but received by crunch-run %q", want, gw.ContainerUUID), http.StatusBadGateway)
 157                 return
 158         }
 159         if req.Header.Get("X-Arvados-Authorization") != gw.requestAuth {
 160                 http.Error(w, "bad X-Arvados-Authorization header", http.StatusUnauthorized)
 161                 return
 162         }
 163         detachKeys := req.Header.Get("X-Arvados-Detach-Keys")
 164         username := req.Header.Get("X-Arvados-Login-Username")
 165         if username == "" {
 166                 username = "root"
 167         }
 168         hj, ok := w.(http.Hijacker)
 169         if !ok {
 170                 http.Error(w, "ResponseWriter does not support connection upgrade", http.StatusInternalServerError)
 171                 return
 172         }
 173         netconn, _, err := hj.Hijack()
 174         if !ok {
 175                 http.Error(w, err.Error(), http.StatusInternalServerError)
 176                 return
 177         }
 178         defer netconn.Close()
 179         w.Header().Set("Connection", "upgrade")
 180         w.Header().Set("Upgrade", "ssh")
 181         w.Header().Set("X-Arvados-Authorization-Response", gw.respondAuth)
 182         netconn.Write([]byte("HTTP/1.1 101 Switching Protocols\r\n"))
 183         w.Header().Write(netconn)
 184         netconn.Write([]byte("\r\n"))
 185
 186         ctx := req.Context()
 187
 188         conn, newchans, reqs, err := ssh.NewServerConn(netconn, &gw.sshConfig)
 189         if err != nil {
 190                 gw.Log.Printf("ssh.NewServerConn: %s", err)
 191                 return
 192         }
 193         defer conn.Close()
 194         go ssh.DiscardRequests(reqs)
 195         for newch := range newchans {
 196                 if newch.ChannelType() != "session" {
 197                         newch.Reject(ssh.UnknownChannelType, "unknown channel type")
 198                         continue
 199                 }
 200                 ch, reqs, err := newch.Accept()
 201                 if err != nil {
 202                         gw.Log.Printf("accept channel: %s", err)
 203                         return
 204                 }
 205                 var pty0, tty0 *os.File
 206                 go func() {
 207                         defer pty0.Close()
 208                         defer tty0.Close()
 209                         // Where to send errors/messages for the
 210                         // client to see
 211                         logw := io.Writer(ch.Stderr())
 212                         // How to end lines when sending
 213                         // errors/messages to the client (changes to
 214                         // \r\n when using a pty)
 215                         eol := "\n"
 216                         // Env vars to add to child process
 217                         termEnv := []string(nil)
 218                         for req := range reqs {
 219                                 ok := false
 220                                 switch req.Type {
 221                                 case "shell", "exec":
 222                                         ok = true
 223                                         var payload struct {
 224                                                 Command string
 225                                         }
 226                                         ssh.Unmarshal(req.Payload, &payload)
 227                                         execargs, err := shlex.Split(payload.Command)
 228                                         if err != nil {
 229                                                 fmt.Fprintf(logw, "error parsing supplied command: %s"+eol, err)
 230                                                 return
 231                                         }
 232                                         if len(execargs) == 0 {
 233                                                 execargs = []string{"/bin/bash", "-login"}
 234                                         }
 235                                         go func() {
 236                                                 cmd := exec.CommandContext(ctx, "docker", "exec", "-i", "--detach-keys="+detachKeys, "--user="+username)
 237                                                 cmd.Stdin = ch
 238                                                 cmd.Stdout = ch
 239                                                 cmd.Stderr = ch.Stderr()
 240                                                 if tty0 != nil {
 241                                                         cmd.Args = append(cmd.Args, "-t")
 242                                                         cmd.Stdin = tty0
 243                                                         cmd.Stdout = tty0
 244                                                         cmd.Stderr = tty0
 245                                                         var wg sync.WaitGroup
 246                                                         defer wg.Wait()
 247                                                         wg.Add(2)
 248                                                         go func() { io.Copy(ch, pty0); wg.Done() }()
 249                                                         go func() { io.Copy(pty0, ch); wg.Done() }()
 250                                                         // Send our own debug messages to tty as well.
 251                                                         logw = tty0
 252                                                 }
 253                                                 cmd.Args = append(cmd.Args, *gw.DockerContainerID)
 254                                                 cmd.Args = append(cmd.Args, execargs...)
 255                                                 cmd.SysProcAttr = &syscall.SysProcAttr{
 256                                                         Setctty: tty0 != nil,
 257                                                         Setsid:  true,
 258                                                 }
 259                                                 cmd.Env = append(os.Environ(), termEnv...)
 260                                                 err := cmd.Run()
 261                                                 errClose := ch.CloseWrite()
 262                                                 var resp struct {
 263                                                         Status uint32
 264                                                 }
 265                                                 if err, ok := err.(*exec.ExitError); ok {
 266                                                         if status, ok := err.Sys().(syscall.WaitStatus); ok {
 267                                                                 resp.Status = uint32(status.ExitStatus())
 268                                                         }
 269                                                 }
 270                                                 if resp.Status == 0 && (err != nil || errClose != nil) {
 271                                                         resp.Status = 1
 272                                                 }
 273                                                 ch.SendRequest("exit-status", false, ssh.Marshal(&resp))
 274                                                 ch.Close()
 275                                         }()
 276                                 case "pty-req":
 277                                         eol = "\r\n"
 278                                         p, t, err := pty.Open()
 279                                         if err != nil {
 280                                                 fmt.Fprintf(ch.Stderr(), "pty failed: %s"+eol, err)
 281                                                 break
 282                                         }
 283                                         pty0, tty0 = p, t
 284                                         ok = true
 285                                         var payload struct {
 286                                                 Term string
 287                                                 Cols uint32
 288                                                 Rows uint32
 289                                                 X    uint32
 290                                                 Y    uint32
 291                                         }
 292                                         ssh.Unmarshal(req.Payload, &payload)
 293                                         termEnv = []string{"TERM=" + payload.Term, "USE_TTY=1"}
 294                                         err = pty.Setsize(pty0, &pty.Winsize{Rows: uint16(payload.Rows), Cols: uint16(payload.Cols), X: uint16(payload.X), Y: uint16(payload.Y)})
 295                                         if err != nil {
 296                                                 fmt.Fprintf(logw, "pty-req: setsize failed: %s"+eol, err)
 297                                         }
 298                                 case "window-change":
 299                                         var payload struct {
 300                                                 Cols uint32
 301                                                 Rows uint32
 302                                                 X    uint32
 303                                                 Y    uint32
 304                                         }
 305                                         ssh.Unmarshal(req.Payload, &payload)
 306                                         err := pty.Setsize(pty0, &pty.Winsize{Rows: uint16(payload.Rows), Cols: uint16(payload.Cols), X: uint16(payload.X), Y: uint16(payload.Y)})
 307                                         if err != nil {
 308                                                 fmt.Fprintf(logw, "window-change: setsize failed: %s"+eol, err)
 309                                                 break
 310                                         }
 311                                         ok = true
 312                                 case "env":
 313                                         // TODO: implement "env"
 314                                         // requests by setting env
 315                                         // vars in the docker-exec
 316                                         // command (not docker-exec's
 317                                         // own environment, which
 318                                         // would be a gaping security
 319                                         // hole).
 320                                 default:
 321                                         // fmt.Fprintf(logw, "declining %q req"+eol, req.Type)
 322                                 }
 323                                 if req.WantReply {
 324                                         req.Reply(ok, nil)
 325                                 }
 326                         }
 327                 }()
 328         }
 329 }