lib/controller/localdb/container_gateway.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package localdb
   6
   7 import (
   8         "bufio"
   9         "bytes"
  10         "context"
  11         "crypto/hmac"
  12         "crypto/sha256"
  13         "crypto/subtle"
  14         "crypto/tls"
  15         "crypto/x509"
  16         "errors"
  17         "fmt"
  18         "io"
  19         "io/ioutil"
  20         "net"
  21         "net/http"
  22         "net/http/httputil"
  23         "net/url"
  24         "os"
  25         "strings"
  26
  27         "git.arvados.org/arvados.git/lib/controller/rpc"
  28         "git.arvados.org/arvados.git/lib/service"
  29         "git.arvados.org/arvados.git/lib/webdavfs"
  30         "git.arvados.org/arvados.git/sdk/go/arvados"
  31         "git.arvados.org/arvados.git/sdk/go/auth"
  32         "git.arvados.org/arvados.git/sdk/go/ctxlog"
  33         "git.arvados.org/arvados.git/sdk/go/httpserver"
  34         keepweb "git.arvados.org/arvados.git/services/keep-web"
  35         "github.com/hashicorp/yamux"
  36         "golang.org/x/net/webdav"
  37 )
  38
  39 var (
  40         forceProxyForTest       = false
  41         forceInternalURLForTest *arvados.URL
  42 )
  43
  44 // ContainerRequestLog returns a WebDAV handler that reads logs from
  45 // the indicated container request. It works by proxying the incoming
  46 // HTTP request to
  47 //
  48 //   - the container gateway, if there is an associated container that
  49 //     is running
  50 //
  51 //   - a different controller process, if there is a running container
  52 //     whose gateway is accessible through a tunnel to a different
  53 //     controller process
  54 //
  55 //   - keep-web, if saved logs exist and there is no gateway (or the
  56 //     associated container is finished)
  57 //
  58 //   - an empty-collection stub, if there is no gateway and no saved
  59 //     log
  60 //
  61 // For an incoming request
  62 //
  63 //      GET /arvados/v1/container_requests/{cr_uuid}/log/{c_uuid}{/c_log_path}
  64 //
  65 // The upstream request may be to {c_uuid}'s container gateway
  66 //
  67 //      GET /arvados/v1/container_requests/{cr_uuid}/log/{c_uuid}{/c_log_path}
  68 //      X-Webdav-Prefix: /arvados/v1/container_requests/{cr_uuid}/log/{c_uuid}
  69 //      X-Webdav-Source: /log
  70 //
  71 // ...or the upstream request may be to keep-web (where {cr_log_uuid}
  72 // is the container request log collection UUID)
  73 //
  74 //      GET /arvados/v1/container_requests/{cr_uuid}/log/{c_uuid}{/c_log_path}
  75 //      Host: {cr_log_uuid}.internal
  76 //      X-Webdav-Prefix: /arvados/v1/container_requests/{cr_uuid}/log
  77 //      X-Arvados-Container-Uuid: {c_uuid}
  78 //
  79 // ...or the request may be handled locally using an empty-collection
  80 // stub.
  81 func (conn *Conn) ContainerRequestLog(ctx context.Context, opts arvados.ContainerLogOptions) (http.Handler, error) {
  82         if opts.Method == "OPTIONS" && opts.Header.Get("Access-Control-Request-Method") != "" {
  83                 return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
  84                         if !keepweb.ServeCORSPreflight(w, opts.Header) {
  85                                 // Inconceivable.  We already checked
  86                                 // for the only condition where
  87                                 // ServeCORSPreflight returns false.
  88                                 httpserver.Error(w, "unhandled CORS preflight request", http.StatusInternalServerError)
  89                         }
  90                 }), nil
  91         }
  92         cr, err := conn.railsProxy.ContainerRequestGet(ctx, arvados.GetOptions{UUID: opts.UUID, Select: []string{"uuid", "container_uuid", "log_uuid"}})
  93         if err != nil {
  94                 if se := httpserver.HTTPStatusError(nil); errors.As(err, &se) && se.HTTPStatus() == http.StatusUnauthorized {
  95                         // Hint to WebDAV client that we accept HTTP basic auth.
  96                         return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
  97                                 w.Header().Set("Www-Authenticate", "Basic realm=\"collections\"")
  98                                 w.WriteHeader(http.StatusUnauthorized)
  99                         }), nil
 100                 }
 101                 return nil, err
 102         }
 103         ctr, err := conn.railsProxy.ContainerGet(ctx, arvados.GetOptions{UUID: cr.ContainerUUID, Select: []string{"uuid", "state", "gateway_address"}})
 104         if err != nil {
 105                 return nil, err
 106         }
 107         // .../log/{ctr.UUID} is a directory where the currently
 108         // assigned container's log data [will] appear (as opposed to
 109         // previous attempts in .../log/{previous_ctr_uuid}). Requests
 110         // that are outside that directory, and requests on a
 111         // non-running container, are proxied to keep-web instead of
 112         // going through the container gateway system.
 113         //
 114         // Side note: a depth>1 directory tree listing starting at
 115         // .../{cr_uuid}/log will only include subdirectories for
 116         // finished containers, i.e., will not include a subdirectory
 117         // with log data for a current (unfinished) container UUID.
 118         // In order to access live logs, a client must look up the
 119         // container_uuid field of the container request record, and
 120         // explicitly request a path under .../{cr_uuid}/log/{c_uuid}.
 121         if ctr.GatewayAddress == "" ||
 122                 (ctr.State != arvados.ContainerStateLocked && ctr.State != arvados.ContainerStateRunning) ||
 123                 !(opts.Path == "/"+ctr.UUID || strings.HasPrefix(opts.Path, "/"+ctr.UUID+"/")) {
 124                 return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 125                         conn.serveContainerRequestLogViaKeepWeb(opts, cr, w, r)
 126                 }), nil
 127         }
 128         dial, arpc, err := conn.findGateway(ctx, ctr, opts.NoForward)
 129         if err != nil {
 130                 return nil, err
 131         }
 132         if arpc != nil {
 133                 opts.NoForward = true
 134                 return arpc.ContainerRequestLog(ctx, opts)
 135         }
 136         return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 137                 r = r.WithContext(ctx)
 138                 var proxyReq *http.Request
 139                 var proxyErr error
 140                 var expectRespondAuth string
 141                 proxy := &httputil.ReverseProxy{
 142                         // Our custom Transport:
 143                         //
 144                         // - Uses a custom dialer to connect to the
 145                         // gateway (either directly or through a
 146                         // tunnel set up though ContainerTunnel)
 147                         //
 148                         // - Verifies the gateway's TLS certificate
 149                         // using X-Arvados-Authorization headers.
 150                         //
 151                         // This involves modifying the outgoing
 152                         // request header in DialTLSContext.
 153                         // (ReverseProxy certainly doesn't expect us
 154                         // to do this, but it works.)
 155                         Transport: &http.Transport{
 156                                 DialTLSContext: func(ctx context.Context, network, addr string) (net.Conn, error) {
 157                                         tlsconn, requestAuth, respondAuth, err := dial()
 158                                         if err != nil {
 159                                                 return nil, err
 160                                         }
 161                                         proxyReq.Header.Set("X-Arvados-Authorization", requestAuth)
 162                                         expectRespondAuth = respondAuth
 163                                         return tlsconn, nil
 164                                 },
 165                         },
 166                         Director: func(r *http.Request) {
 167                                 // Scheme/host of incoming r.URL are
 168                                 // irrelevant now, and may even be
 169                                 // missing. Host is ignored by our
 170                                 // DialTLSContext, but we need a
 171                                 // generic syntactically correct URL
 172                                 // for net/http to work with.
 173                                 r.URL.Scheme = "https"
 174                                 r.URL.Host = "0.0.0.0:0"
 175                                 r.Header.Set("X-Arvados-Container-Gateway-Uuid", ctr.UUID)
 176                                 r.Header.Set("X-Webdav-Prefix", "/arvados/v1/container_requests/"+cr.UUID+"/log/"+ctr.UUID)
 177                                 r.Header.Set("X-Webdav-Source", "/log")
 178                                 proxyReq = r
 179                         },
 180                         ModifyResponse: func(resp *http.Response) error {
 181                                 if resp.Header.Get("X-Arvados-Authorization-Response") != expectRespondAuth {
 182                                         // Note this is how we detect
 183                                         // an attacker-in-the-middle.
 184                                         return httpserver.ErrorWithStatus(errors.New("bad X-Arvados-Authorization-Response header"), http.StatusBadGateway)
 185                                 }
 186                                 resp.Header.Del("X-Arvados-Authorization-Response")
 187                                 for hdr := range resp.Header {
 188                                         // proxy.ServeHTTP adds each
 189                                         // resp.Header to w.Header,
 190                                         // which causes duplicate CORS
 191                                         // and request-id headers,
 192                                         // unless we do this.
 193                                         w.Header().Del(hdr)
 194                                 }
 195                                 return nil
 196                         },
 197                         ErrorHandler: func(w http.ResponseWriter, r *http.Request, err error) {
 198                                 proxyErr = err
 199                         },
 200                 }
 201                 proxy.ServeHTTP(w, r)
 202                 if proxyErr == nil {
 203                         // proxy succeeded
 204                         return
 205                 }
 206                 // If proxying to the container gateway fails, it
 207                 // might be caused by a race where crunch-run exited
 208                 // after we decided (above) the log was not final.
 209                 // In that case we should proxy to keep-web.
 210                 ctr, err := conn.railsProxy.ContainerGet(ctx, arvados.GetOptions{
 211                         UUID:   ctr.UUID,
 212                         Select: []string{"uuid", "state", "gateway_address", "log"},
 213                 })
 214                 if err != nil {
 215                         // Lost access to the container record?
 216                         httpserver.Error(w, "error re-fetching container record: "+err.Error(), http.StatusServiceUnavailable)
 217                 } else if ctr.State == arvados.ContainerStateLocked || ctr.State == arvados.ContainerStateRunning {
 218                         // No race, proxyErr was the best we can do
 219                         httpserver.Error(w, "proxy error: "+proxyErr.Error(), http.StatusServiceUnavailable)
 220                 } else {
 221                         conn.serveContainerRequestLogViaKeepWeb(opts, cr, w, r)
 222                 }
 223         }), nil
 224 }
 225
 226 // serveContainerLogViaKeepWeb handles a request for saved container
 227 // log content by proxying to one of the configured keep-web servers.
 228 //
 229 // It tries to choose a keep-web server that is running on this host.
 230 func (conn *Conn) serveContainerRequestLogViaKeepWeb(opts arvados.ContainerLogOptions, cr arvados.ContainerRequest, w http.ResponseWriter, r *http.Request) {
 231         if cr.LogUUID == "" {
 232                 // Special case: if no log data exists yet, we serve
 233                 // an empty collection by ourselves instead of
 234                 // proxying to keep-web.
 235                 conn.serveEmptyDir("/arvados/v1/container_requests/"+cr.UUID+"/log", w, r)
 236                 return
 237         }
 238         myURL, _ := service.URLFromContext(r.Context())
 239         u := url.URL(myURL)
 240         myHostname := u.Hostname()
 241         var webdavBase arvados.URL
 242         var ok bool
 243         for webdavBase = range conn.cluster.Services.WebDAV.InternalURLs {
 244                 ok = true
 245                 u := url.URL(webdavBase)
 246                 if h := u.Hostname(); h == "127.0.0.1" || h == "0.0.0.0" || h == "::1" || h == myHostname {
 247                         // Prefer a keep-web service running on the
 248                         // same host as us. (If we don't find one, we
 249                         // pick one arbitrarily.)
 250                         break
 251                 }
 252         }
 253         if !ok {
 254                 httpserver.Error(w, "no internalURLs configured for WebDAV service", http.StatusInternalServerError)
 255                 return
 256         }
 257         proxy := &httputil.ReverseProxy{
 258                 Director: func(r *http.Request) {
 259                         r.URL.Scheme = webdavBase.Scheme
 260                         r.URL.Host = webdavBase.Host
 261                         // Outgoing Host header specifies the
 262                         // collection ID.
 263                         r.Host = cr.LogUUID + ".internal"
 264                         // We already checked permission on the
 265                         // container, so we can use a root token here
 266                         // instead of counting on the "access to log
 267                         // via container request and container"
 268                         // permission check, which can be racy when a
 269                         // request gets retried with a new container.
 270                         r.Header.Set("Authorization", "Bearer "+conn.cluster.SystemRootToken)
 271                         // We can't change r.URL.Path without
 272                         // confusing WebDAV (request body and response
 273                         // headers refer to the same paths) so we tell
 274                         // keep-web to map the log collection onto the
 275                         // containers/X/log/ namespace.
 276                         r.Header.Set("X-Webdav-Prefix", "/arvados/v1/container_requests/"+cr.UUID+"/log")
 277                         if len(opts.Path) >= 28 && opts.Path[6:13] == "-dz642-" {
 278                                 // "/arvados/v1/container_requests/{crUUID}/log/{cUUID}..."
 279                                 // proxies to
 280                                 // "/log for container {cUUID}..."
 281                                 r.Header.Set("X-Webdav-Prefix", "/arvados/v1/container_requests/"+cr.UUID+"/log/"+opts.Path[1:28])
 282                                 r.Header.Set("X-Webdav-Source", "/log for container "+opts.Path[1:28]+"/")
 283                         }
 284                 },
 285                 ModifyResponse: func(resp *http.Response) error {
 286                         for hdr := range resp.Header {
 287                                 // proxy.ServeHTTP adds each
 288                                 // resp.Header to w.Header, which
 289                                 // causes duplicate CORS and
 290                                 // request-id headers, unless we do
 291                                 // this.
 292                                 w.Header().Del(hdr)
 293                         }
 294                         return nil
 295                 },
 296         }
 297         if conn.cluster.TLS.Insecure {
 298                 proxy.Transport = &http.Transport{
 299                         TLSClientConfig: &tls.Config{
 300                                 InsecureSkipVerify: conn.cluster.TLS.Insecure,
 301                         },
 302                 }
 303         }
 304         proxy.ServeHTTP(w, r)
 305 }
 306
 307 // serveEmptyDir handles read-only webdav requests as if there was an
 308 // empty collection rooted at the given path. It's equivalent to
 309 // proxying to an empty collection in keep-web, but avoids the extra
 310 // hop.
 311 func (conn *Conn) serveEmptyDir(path string, w http.ResponseWriter, r *http.Request) {
 312         wh := webdav.Handler{
 313                 Prefix:     path,
 314                 FileSystem: webdav.NewMemFS(),
 315                 LockSystem: webdavfs.NoLockSystem,
 316                 Logger: func(r *http.Request, err error) {
 317                         if err != nil && !os.IsNotExist(err) {
 318                                 ctxlog.FromContext(r.Context()).WithError(err).Info("webdav error on empty collection fs")
 319                         }
 320                 },
 321         }
 322         wh.ServeHTTP(w, r)
 323 }
 324
 325 // ContainerSSH returns a connection to the SSH server in the
 326 // appropriate crunch-run process on the worker node where the
 327 // specified container is running.
 328 //
 329 // If the returned error is nil, the caller is responsible for closing
 330 // sshconn.Conn.
 331 func (conn *Conn) ContainerSSH(ctx context.Context, opts arvados.ContainerSSHOptions) (sshconn arvados.ConnectionResponse, err error) {
 332         user, err := conn.railsProxy.UserGetCurrent(ctx, arvados.GetOptions{})
 333         if err != nil {
 334                 return sshconn, err
 335         }
 336         ctr, err := conn.railsProxy.ContainerGet(ctx, arvados.GetOptions{UUID: opts.UUID, Select: []string{"uuid", "state", "gateway_address", "interactive_session_started"}})
 337         if err != nil {
 338                 return sshconn, err
 339         }
 340         ctxRoot := auth.NewContext(ctx, &auth.Credentials{Tokens: []string{conn.cluster.SystemRootToken}})
 341         if !user.IsAdmin || !conn.cluster.Containers.ShellAccess.Admin {
 342                 if !conn.cluster.Containers.ShellAccess.User {
 343                         return sshconn, httpserver.ErrorWithStatus(errors.New("shell access is disabled in config"), http.StatusServiceUnavailable)
 344                 }
 345                 crs, err := conn.railsProxy.ContainerRequestList(ctxRoot, arvados.ListOptions{Limit: -1, Filters: []arvados.Filter{{"container_uuid", "=", opts.UUID}}})
 346                 if err != nil {
 347                         return sshconn, err
 348                 }
 349                 for _, cr := range crs.Items {
 350                         if cr.ModifiedByUserUUID != user.UUID {
 351                                 return sshconn, httpserver.ErrorWithStatus(errors.New("permission denied: container is associated with requests submitted by other users"), http.StatusForbidden)
 352                         }
 353                 }
 354                 if crs.ItemsAvailable != len(crs.Items) {
 355                         return sshconn, httpserver.ErrorWithStatus(errors.New("incomplete response while checking permission"), http.StatusInternalServerError)
 356                 }
 357         }
 358
 359         if ctr.State == arvados.ContainerStateQueued || ctr.State == arvados.ContainerStateLocked {
 360                 return sshconn, httpserver.ErrorWithStatus(fmt.Errorf("container is not running yet (state is %q)", ctr.State), http.StatusServiceUnavailable)
 361         } else if ctr.State != arvados.ContainerStateRunning {
 362                 return sshconn, httpserver.ErrorWithStatus(fmt.Errorf("container has ended (state is %q)", ctr.State), http.StatusGone)
 363         }
 364
 365         dial, arpc, err := conn.findGateway(ctx, ctr, opts.NoForward)
 366         if err != nil {
 367                 return sshconn, err
 368         }
 369         if arpc != nil {
 370                 opts.NoForward = true
 371                 return arpc.ContainerSSH(ctx, opts)
 372         }
 373
 374         tlsconn, requestAuth, respondAuth, err := dial()
 375         if err != nil {
 376                 return sshconn, err
 377         }
 378         bufr := bufio.NewReader(tlsconn)
 379         bufw := bufio.NewWriter(tlsconn)
 380
 381         u := url.URL{
 382                 Scheme: "http",
 383                 Host:   tlsconn.RemoteAddr().String(),
 384                 Path:   "/ssh",
 385         }
 386         postform := url.Values{
 387                 // uuid is only needed for older crunch-run versions
 388                 // (current version uses X-Arvados-* header below)
 389                 "uuid":           {opts.UUID},
 390                 "detach_keys":    {opts.DetachKeys},
 391                 "login_username": {opts.LoginUsername},
 392                 "no_forward":     {fmt.Sprintf("%v", opts.NoForward)},
 393         }
 394         postdata := postform.Encode()
 395         bufw.WriteString("POST " + u.String() + " HTTP/1.1\r\n")
 396         bufw.WriteString("Host: " + u.Host + "\r\n")
 397         bufw.WriteString("Upgrade: ssh\r\n")
 398         bufw.WriteString("X-Arvados-Container-Gateway-Uuid: " + opts.UUID + "\r\n")
 399         bufw.WriteString("X-Arvados-Authorization: " + requestAuth + "\r\n")
 400         bufw.WriteString("Content-Type: application/x-www-form-urlencoded\r\n")
 401         fmt.Fprintf(bufw, "Content-Length: %d\r\n", len(postdata))
 402         bufw.WriteString("\r\n")
 403         bufw.WriteString(postdata)
 404         bufw.Flush()
 405         resp, err := http.ReadResponse(bufr, &http.Request{Method: "POST"})
 406         if err != nil {
 407                 tlsconn.Close()
 408                 return sshconn, httpserver.ErrorWithStatus(fmt.Errorf("error reading http response from gateway: %w", err), http.StatusBadGateway)
 409         }
 410         defer resp.Body.Close()
 411         if resp.StatusCode != http.StatusSwitchingProtocols {
 412                 body, _ := ioutil.ReadAll(io.LimitReader(resp.Body, 1000))
 413                 tlsconn.Close()
 414                 return sshconn, httpserver.ErrorWithStatus(fmt.Errorf("unexpected status %s %q", resp.Status, body), http.StatusBadGateway)
 415         }
 416         if strings.ToLower(resp.Header.Get("Upgrade")) != "ssh" ||
 417                 strings.ToLower(resp.Header.Get("Connection")) != "upgrade" {
 418                 tlsconn.Close()
 419                 return sshconn, httpserver.ErrorWithStatus(errors.New("bad upgrade"), http.StatusBadGateway)
 420         }
 421         if resp.Header.Get("X-Arvados-Authorization-Response") != respondAuth {
 422                 tlsconn.Close()
 423                 return sshconn, httpserver.ErrorWithStatus(errors.New("bad X-Arvados-Authorization-Response header"), http.StatusBadGateway)
 424         }
 425
 426         if !ctr.InteractiveSessionStarted {
 427                 _, err = conn.railsProxy.ContainerUpdate(ctxRoot, arvados.UpdateOptions{
 428                         UUID: opts.UUID,
 429                         Attrs: map[string]interface{}{
 430                                 "interactive_session_started": true,
 431                         },
 432                 })
 433                 if err != nil {
 434                         tlsconn.Close()
 435                         return sshconn, httpserver.ErrorWithStatus(err, http.StatusInternalServerError)
 436                 }
 437         }
 438
 439         sshconn.Conn = tlsconn
 440         sshconn.Bufrw = &bufio.ReadWriter{Reader: bufr, Writer: bufw}
 441         sshconn.Logger = ctxlog.FromContext(ctx)
 442         sshconn.Header = http.Header{"Upgrade": {"ssh"}}
 443         return sshconn, nil
 444 }
 445
 446 // ContainerGatewayTunnel sets up a tunnel enabling us (controller) to
 447 // connect to the caller's (crunch-run's) gateway server.
 448 func (conn *Conn) ContainerGatewayTunnel(ctx context.Context, opts arvados.ContainerGatewayTunnelOptions) (resp arvados.ConnectionResponse, err error) {
 449         h := hmac.New(sha256.New, []byte(conn.cluster.SystemRootToken))
 450         fmt.Fprint(h, opts.UUID)
 451         authSecret := fmt.Sprintf("%x", h.Sum(nil))
 452         if subtle.ConstantTimeCompare([]byte(authSecret), []byte(opts.AuthSecret)) != 1 {
 453                 ctxlog.FromContext(ctx).Info("received incorrect auth_secret")
 454                 return resp, httpserver.ErrorWithStatus(errors.New("authentication error"), http.StatusUnauthorized)
 455         }
 456
 457         muxconn, clientconn := net.Pipe()
 458         tunnel, err := yamux.Server(muxconn, nil)
 459         if err != nil {
 460                 clientconn.Close()
 461                 return resp, httpserver.ErrorWithStatus(err, http.StatusInternalServerError)
 462         }
 463
 464         conn.gwTunnelsLock.Lock()
 465         if conn.gwTunnels == nil {
 466                 conn.gwTunnels = map[string]*yamux.Session{opts.UUID: tunnel}
 467         } else {
 468                 conn.gwTunnels[opts.UUID] = tunnel
 469         }
 470         conn.gwTunnelsLock.Unlock()
 471
 472         go func() {
 473                 <-tunnel.CloseChan()
 474                 conn.gwTunnelsLock.Lock()
 475                 if conn.gwTunnels[opts.UUID] == tunnel {
 476                         delete(conn.gwTunnels, opts.UUID)
 477                 }
 478                 conn.gwTunnelsLock.Unlock()
 479         }()
 480
 481         // Assuming we're acting as the backend of an http server,
 482         // lib/controller/router will call resp's ServeHTTP handler,
 483         // which upgrades the incoming http connection to a raw socket
 484         // and connects it to our yamux.Server through our net.Pipe().
 485         resp.Conn = clientconn
 486         resp.Bufrw = &bufio.ReadWriter{Reader: bufio.NewReader(&bytes.Buffer{}), Writer: bufio.NewWriter(&bytes.Buffer{})}
 487         resp.Logger = ctxlog.FromContext(ctx)
 488         resp.Header = http.Header{"Upgrade": {"tunnel"}}
 489         if u, ok := service.URLFromContext(ctx); ok {
 490                 resp.Header.Set("X-Arvados-Internal-Url", u.String())
 491         } else if forceInternalURLForTest != nil {
 492                 resp.Header.Set("X-Arvados-Internal-Url", forceInternalURLForTest.String())
 493         }
 494         return
 495 }
 496
 497 type gatewayDialer func() (conn net.Conn, requestAuth, respondAuth string, err error)
 498
 499 // findGateway figures out how to connect to ctr's gateway.
 500 //
 501 // If the gateway can be contacted directly or through a tunnel on
 502 // this instance, the first return value is a non-nil dialer.
 503 //
 504 // If the gateway is only accessible through a tunnel through a
 505 // different controller process, the second return value is a non-nil
 506 // *rpc.Conn for that controller.
 507 func (conn *Conn) findGateway(ctx context.Context, ctr arvados.Container, noForward bool) (gatewayDialer, *rpc.Conn, error) {
 508         conn.gwTunnelsLock.Lock()
 509         tunnel := conn.gwTunnels[ctr.UUID]
 510         conn.gwTunnelsLock.Unlock()
 511
 512         myURL, _ := service.URLFromContext(ctx)
 513
 514         if host, _, splitErr := net.SplitHostPort(ctr.GatewayAddress); splitErr == nil && host != "" && host != "127.0.0.1" {
 515                 // If crunch-run provided a GatewayAddress like
 516                 // "ipaddr:port", that means "ipaddr" is one of the
 517                 // external interfaces where the gateway is
 518                 // listening. In that case, it's the most
 519                 // reliable/direct option, so we use it even if a
 520                 // tunnel might also be available.
 521                 return func() (net.Conn, string, string, error) {
 522                         rawconn, err := (&net.Dialer{}).DialContext(ctx, "tcp", ctr.GatewayAddress)
 523                         if err != nil {
 524                                 return nil, "", "", httpserver.ErrorWithStatus(err, http.StatusServiceUnavailable)
 525                         }
 526                         return conn.dialGatewayTLS(ctx, ctr, rawconn)
 527                 }, nil, nil
 528         }
 529         if tunnel != nil && !(forceProxyForTest && !noForward) {
 530                 // If we can't connect directly, and the gateway has
 531                 // established a yamux tunnel with us, connect through
 532                 // the tunnel.
 533                 //
 534                 // ...except: forceProxyForTest means we are emulating
 535                 // a situation where the gateway has established a
 536                 // yamux tunnel with controller B, and the
 537                 // ContainerSSH request arrives at controller A. If
 538                 // noForward==false then we are acting as A, so
 539                 // we pretend not to have a tunnel, and fall through
 540                 // to the "tunurl" case below. If noForward==true
 541                 // then the client is A and we are acting as B, so we
 542                 // connect to our tunnel.
 543                 return func() (net.Conn, string, string, error) {
 544                         rawconn, err := tunnel.Open()
 545                         if err != nil {
 546                                 return nil, "", "", httpserver.ErrorWithStatus(err, http.StatusServiceUnavailable)
 547                         }
 548                         return conn.dialGatewayTLS(ctx, ctr, rawconn)
 549                 }, nil, nil
 550         }
 551         if tunurl := strings.TrimPrefix(ctr.GatewayAddress, "tunnel "); tunurl != ctr.GatewayAddress &&
 552                 tunurl != "" &&
 553                 tunurl != myURL.String() &&
 554                 !noForward {
 555                 // If crunch-run provided a GatewayAddress like
 556                 // "tunnel https://10.0.0.10:1010/", that means the
 557                 // gateway has established a yamux tunnel with the
 558                 // controller process at the indicated InternalURL
 559                 // (which isn't us, otherwise we would have had
 560                 // "tunnel != nil" above). We need to proxy through to
 561                 // the other controller process in order to use the
 562                 // tunnel.
 563                 for u := range conn.cluster.Services.Controller.InternalURLs {
 564                         if u.String() == tunurl {
 565                                 ctxlog.FromContext(ctx).Debugf("connecting to container gateway through other controller at %s", u)
 566                                 u := url.URL(u)
 567                                 return nil, rpc.NewConn(conn.cluster.ClusterID, &u, conn.cluster.TLS.Insecure, rpc.PassthroughTokenProvider), nil
 568                         }
 569                 }
 570                 ctxlog.FromContext(ctx).Warnf("container gateway provided a tunnel endpoint %s that is not one of Services.Controller.InternalURLs", tunurl)
 571                 return nil, nil, httpserver.ErrorWithStatus(errors.New("container gateway is running but tunnel endpoint is invalid"), http.StatusServiceUnavailable)
 572         }
 573         if ctr.GatewayAddress == "" {
 574                 return nil, nil, httpserver.ErrorWithStatus(errors.New("container is running but gateway is not available"), http.StatusServiceUnavailable)
 575         } else {
 576                 return nil, nil, httpserver.ErrorWithStatus(errors.New("container is running but tunnel is down"), http.StatusServiceUnavailable)
 577         }
 578 }
 579
 580 // dialGatewayTLS negotiates a TLS connection to a container gateway
 581 // over the given raw connection.
 582 func (conn *Conn) dialGatewayTLS(ctx context.Context, ctr arvados.Container, rawconn net.Conn) (*tls.Conn, string, string, error) {
 583         // crunch-run uses a self-signed / unverifiable TLS
 584         // certificate, so we use the following scheme to ensure we're
 585         // not talking to an attacker-in-the-middle.
 586         //
 587         // 1. Compute ctrKey = HMAC-SHA256(sysRootToken,ctrUUID) --
 588         // this will be the same ctrKey that a-d-c supplied to
 589         // crunch-run in the GatewayAuthSecret env var.
 590         //
 591         // 2. Compute requestAuth = HMAC-SHA256(ctrKey,serverCert) and
 592         // send it to crunch-run as the X-Arvados-Authorization
 593         // header, proving that we know ctrKey. (Note a MITM cannot
 594         // replay the proof to a real crunch-run server, because the
 595         // real crunch-run server would have a different cert.)
 596         //
 597         // 3. Compute respondAuth = HMAC-SHA256(ctrKey,requestAuth)
 598         // and ensure the server returns it in the
 599         // X-Arvados-Authorization-Response header, proving that the
 600         // server knows ctrKey.
 601         var requestAuth, respondAuth string
 602         tlsconn := tls.Client(rawconn, &tls.Config{
 603                 InsecureSkipVerify: true,
 604                 VerifyPeerCertificate: func(rawCerts [][]byte, verifiedChains [][]*x509.Certificate) error {
 605                         if len(rawCerts) == 0 {
 606                                 return errors.New("no certificate received, cannot compute authorization header")
 607                         }
 608                         h := hmac.New(sha256.New, []byte(conn.cluster.SystemRootToken))
 609                         fmt.Fprint(h, ctr.UUID)
 610                         authKey := fmt.Sprintf("%x", h.Sum(nil))
 611                         h = hmac.New(sha256.New, []byte(authKey))
 612                         h.Write(rawCerts[0])
 613                         requestAuth = fmt.Sprintf("%x", h.Sum(nil))
 614                         h.Reset()
 615                         h.Write([]byte(requestAuth))
 616                         respondAuth = fmt.Sprintf("%x", h.Sum(nil))
 617                         return nil
 618                 },
 619         })
 620         err := tlsconn.HandshakeContext(ctx)
 621         if err != nil {
 622                 return nil, "", "", httpserver.ErrorWithStatus(fmt.Errorf("TLS handshake failed: %w", err), http.StatusBadGateway)
 623         }
 624         if respondAuth == "" {
 625                 tlsconn.Close()
 626                 return nil, "", "", httpserver.ErrorWithStatus(errors.New("BUG: no respondAuth"), http.StatusInternalServerError)
 627         }
 628         return tlsconn, requestAuth, respondAuth, nil
 629 }