services/keepproxy/keepproxy.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package main
   6
   7 import (
   8         "errors"
   9         "flag"
  10         "fmt"
  11         "io"
  12         "io/ioutil"
  13         "net"
  14         "net/http"
  15         "os"
  16         "os/signal"
  17         "regexp"
  18         "strings"
  19         "sync"
  20         "syscall"
  21         "time"
  22
  23         "git.arvados.org/arvados.git/lib/config"
  24         "git.arvados.org/arvados.git/sdk/go/arvados"
  25         "git.arvados.org/arvados.git/sdk/go/arvadosclient"
  26         "git.arvados.org/arvados.git/sdk/go/health"
  27         "git.arvados.org/arvados.git/sdk/go/httpserver"
  28         "git.arvados.org/arvados.git/sdk/go/keepclient"
  29         "github.com/coreos/go-systemd/daemon"
  30         "github.com/ghodss/yaml"
  31         "github.com/gorilla/mux"
  32         log "github.com/sirupsen/logrus"
  33 )
  34
  35 var version = "dev"
  36
  37 var (
  38         listener net.Listener
  39         router   http.Handler
  40 )
  41
  42 const rfc3339NanoFixed = "2006-01-02T15:04:05.000000000Z07:00"
  43
  44 func configure(logger log.FieldLogger, args []string) (*arvados.Cluster, error) {
  45         flags := flag.NewFlagSet(args[0], flag.ExitOnError)
  46
  47         dumpConfig := flags.Bool("dump-config", false, "write current configuration to stdout and exit")
  48         getVersion := flags.Bool("version", false, "Print version information and exit.")
  49
  50         loader := config.NewLoader(os.Stdin, logger)
  51         loader.SetupFlags(flags)
  52
  53         args = loader.MungeLegacyConfigArgs(logger, args[1:], "-legacy-keepproxy-config")
  54         flags.Parse(args)
  55
  56         // Print version information if requested
  57         if *getVersion {
  58                 fmt.Printf("keepproxy %s\n", version)
  59                 return nil, nil
  60         }
  61
  62         cfg, err := loader.Load()
  63         if err != nil {
  64                 return nil, err
  65         }
  66         cluster, err := cfg.GetCluster("")
  67         if err != nil {
  68                 return nil, err
  69         }
  70
  71         if *dumpConfig {
  72                 out, err := yaml.Marshal(cfg)
  73                 if err != nil {
  74                         return nil, err
  75                 }
  76                 if _, err := os.Stdout.Write(out); err != nil {
  77                         return nil, err
  78                 }
  79                 return nil, nil
  80         }
  81         return cluster, nil
  82 }
  83
  84 func main() {
  85         logger := log.New()
  86         logger.Formatter = &log.JSONFormatter{
  87                 TimestampFormat: rfc3339NanoFixed,
  88         }
  89
  90         cluster, err := configure(logger, os.Args)
  91         if err != nil {
  92                 log.Fatal(err)
  93         }
  94         if cluster == nil {
  95                 return
  96         }
  97
  98         log.Printf("keepproxy %s started", version)
  99
 100         if err := run(logger, cluster); err != nil {
 101                 log.Fatal(err)
 102         }
 103
 104         log.Println("shutting down")
 105 }
 106
 107 func run(logger log.FieldLogger, cluster *arvados.Cluster) error {
 108         client, err := arvados.NewClientFromConfig(cluster)
 109         if err != nil {
 110                 return err
 111         }
 112         client.AuthToken = cluster.SystemRootToken
 113
 114         arv, err := arvadosclient.New(client)
 115         if err != nil {
 116                 return fmt.Errorf("Error setting up arvados client %v", err)
 117         }
 118
 119         if cluster.SystemLogs.LogLevel == "debug" {
 120                 keepclient.DebugPrintf = log.Printf
 121         }
 122         kc, err := keepclient.MakeKeepClient(arv)
 123         if err != nil {
 124                 return fmt.Errorf("Error setting up keep client %v", err)
 125         }
 126         keepclient.RefreshServiceDiscoveryOnSIGHUP()
 127
 128         if cluster.Collections.DefaultReplication > 0 {
 129                 kc.Want_replicas = cluster.Collections.DefaultReplication
 130         }
 131
 132         var listen arvados.URL
 133         for listen = range cluster.Services.Keepproxy.InternalURLs {
 134                 break
 135         }
 136
 137         var lErr error
 138         listener, lErr = net.Listen("tcp", listen.Host)
 139         if lErr != nil {
 140                 return fmt.Errorf("listen(%s): %v", listen.Host, lErr)
 141         }
 142
 143         if _, err := daemon.SdNotify(false, "READY=1"); err != nil {
 144                 log.Printf("Error notifying init daemon: %v", err)
 145         }
 146         log.Println("listening at", listener.Addr())
 147
 148         // Shut down the server gracefully (by closing the listener)
 149         // if SIGTERM is received.
 150         term := make(chan os.Signal, 1)
 151         go func(sig <-chan os.Signal) {
 152                 s := <-sig
 153                 log.Println("caught signal:", s)
 154                 listener.Close()
 155         }(term)
 156         signal.Notify(term, syscall.SIGTERM)
 157         signal.Notify(term, syscall.SIGINT)
 158
 159         // Start serving requests.
 160         router = MakeRESTRouter(kc, time.Duration(cluster.API.KeepServiceRequestTimeout), cluster.ManagementToken)
 161         return http.Serve(listener, httpserver.AddRequestIDs(httpserver.LogRequests(router)))
 162 }
 163
 164 type ApiTokenCache struct {
 165         tokens     map[string]int64
 166         lock       sync.Mutex
 167         expireTime int64
 168 }
 169
 170 // Cache the token and set an expire time.  If we already have an expire time
 171 // on the token, it is not updated.
 172 func (this *ApiTokenCache) RememberToken(token string) {
 173         this.lock.Lock()
 174         defer this.lock.Unlock()
 175
 176         now := time.Now().Unix()
 177         if this.tokens[token] == 0 {
 178                 this.tokens[token] = now + this.expireTime
 179         }
 180 }
 181
 182 // Check if the cached token is known and still believed to be valid.
 183 func (this *ApiTokenCache) RecallToken(token string) bool {
 184         this.lock.Lock()
 185         defer this.lock.Unlock()
 186
 187         now := time.Now().Unix()
 188         if this.tokens[token] == 0 {
 189                 // Unknown token
 190                 return false
 191         } else if now < this.tokens[token] {
 192                 // Token is known and still valid
 193                 return true
 194         } else {
 195                 // Token is expired
 196                 this.tokens[token] = 0
 197                 return false
 198         }
 199 }
 200
 201 func GetRemoteAddress(req *http.Request) string {
 202         if xff := req.Header.Get("X-Forwarded-For"); xff != "" {
 203                 return xff + "," + req.RemoteAddr
 204         }
 205         return req.RemoteAddr
 206 }
 207
 208 func CheckAuthorizationHeader(kc *keepclient.KeepClient, cache *ApiTokenCache, req *http.Request) (pass bool, tok string) {
 209         parts := strings.SplitN(req.Header.Get("Authorization"), " ", 2)
 210         if len(parts) < 2 || !(parts[0] == "OAuth2" || parts[0] == "Bearer") || len(parts[1]) == 0 {
 211                 return false, ""
 212         }
 213         tok = parts[1]
 214
 215         // Tokens are validated differently depending on what kind of
 216         // operation is being performed. For example, tokens in
 217         // collection-sharing links permit GET requests, but not
 218         // PUT requests.
 219         var op string
 220         if req.Method == "GET" || req.Method == "HEAD" {
 221                 op = "read"
 222         } else {
 223                 op = "write"
 224         }
 225
 226         if cache.RecallToken(op + ":" + tok) {
 227                 // Valid in the cache, short circuit
 228                 return true, tok
 229         }
 230
 231         var err error
 232         arv := *kc.Arvados
 233         arv.ApiToken = tok
 234         arv.RequestID = req.Header.Get("X-Request-Id")
 235         if op == "read" {
 236                 err = arv.Call("HEAD", "keep_services", "", "accessible", nil, nil)
 237         } else {
 238                 err = arv.Call("HEAD", "users", "", "current", nil, nil)
 239         }
 240         if err != nil {
 241                 log.Printf("%s: CheckAuthorizationHeader error: %v", GetRemoteAddress(req), err)
 242                 return false, ""
 243         }
 244
 245         // Success!  Update cache
 246         cache.RememberToken(op + ":" + tok)
 247
 248         return true, tok
 249 }
 250
 251 // We need to make a private copy of the default http transport early
 252 // in initialization, then make copies of our private copy later. It
 253 // won't be safe to copy http.DefaultTransport itself later, because
 254 // its private mutexes might have already been used. (Without this,
 255 // the test suite sometimes panics "concurrent map writes" in
 256 // net/http.(*Transport).removeIdleConnLocked().)
 257 var defaultTransport = *(http.DefaultTransport.(*http.Transport))
 258
 259 type proxyHandler struct {
 260         http.Handler
 261         *keepclient.KeepClient
 262         *ApiTokenCache
 263         timeout   time.Duration
 264         transport *http.Transport
 265 }
 266
 267 // MakeRESTRouter returns an http.Handler that passes GET and PUT
 268 // requests to the appropriate handlers.
 269 func MakeRESTRouter(kc *keepclient.KeepClient, timeout time.Duration, mgmtToken string) http.Handler {
 270         rest := mux.NewRouter()
 271
 272         transport := defaultTransport
 273         transport.DialContext = (&net.Dialer{
 274                 Timeout:   keepclient.DefaultConnectTimeout,
 275                 KeepAlive: keepclient.DefaultKeepAlive,
 276                 DualStack: true,
 277         }).DialContext
 278         transport.TLSClientConfig = arvadosclient.MakeTLSConfig(kc.Arvados.ApiInsecure)
 279         transport.TLSHandshakeTimeout = keepclient.DefaultTLSHandshakeTimeout
 280
 281         h := &proxyHandler{
 282                 Handler:    rest,
 283                 KeepClient: kc,
 284                 timeout:    timeout,
 285                 transport:  &transport,
 286                 ApiTokenCache: &ApiTokenCache{
 287                         tokens:     make(map[string]int64),
 288                         expireTime: 300,
 289                 },
 290         }
 291
 292         rest.HandleFunc(`/{locator:[0-9a-f]{32}\+.*}`, h.Get).Methods("GET", "HEAD")
 293         rest.HandleFunc(`/{locator:[0-9a-f]{32}}`, h.Get).Methods("GET", "HEAD")
 294
 295         // List all blocks
 296         rest.HandleFunc(`/index`, h.Index).Methods("GET")
 297
 298         // List blocks whose hash has the given prefix
 299         rest.HandleFunc(`/index/{prefix:[0-9a-f]{0,32}}`, h.Index).Methods("GET")
 300
 301         rest.HandleFunc(`/{locator:[0-9a-f]{32}\+.*}`, h.Put).Methods("PUT")
 302         rest.HandleFunc(`/{locator:[0-9a-f]{32}}`, h.Put).Methods("PUT")
 303         rest.HandleFunc(`/`, h.Put).Methods("POST")
 304         rest.HandleFunc(`/{any}`, h.Options).Methods("OPTIONS")
 305         rest.HandleFunc(`/`, h.Options).Methods("OPTIONS")
 306
 307         rest.Handle("/_health/{check}", &health.Handler{
 308                 Token:  mgmtToken,
 309                 Prefix: "/_health/",
 310         }).Methods("GET")
 311
 312         rest.NotFoundHandler = InvalidPathHandler{}
 313         return h
 314 }
 315
 316 var errLoopDetected = errors.New("loop detected")
 317
 318 func (*proxyHandler) checkLoop(resp http.ResponseWriter, req *http.Request) error {
 319         if via := req.Header.Get("Via"); strings.Index(via, " "+viaAlias) >= 0 {
 320                 log.Printf("proxy loop detected (request has Via: %q): perhaps keepproxy is misidentified by gateway config as an external client, or its keep_services record does not have service_type=proxy?", via)
 321                 http.Error(resp, errLoopDetected.Error(), http.StatusInternalServerError)
 322                 return errLoopDetected
 323         }
 324         return nil
 325 }
 326
 327 func SetCorsHeaders(resp http.ResponseWriter) {
 328         resp.Header().Set("Access-Control-Allow-Methods", "GET, HEAD, POST, PUT, OPTIONS")
 329         resp.Header().Set("Access-Control-Allow-Origin", "*")
 330         resp.Header().Set("Access-Control-Allow-Headers", "Authorization, Content-Length, Content-Type, X-Keep-Desired-Replicas")
 331         resp.Header().Set("Access-Control-Max-Age", "86486400")
 332 }
 333
 334 type InvalidPathHandler struct{}
 335
 336 func (InvalidPathHandler) ServeHTTP(resp http.ResponseWriter, req *http.Request) {
 337         log.Printf("%s: %s %s unroutable", GetRemoteAddress(req), req.Method, req.URL.Path)
 338         http.Error(resp, "Bad request", http.StatusBadRequest)
 339 }
 340
 341 func (h *proxyHandler) Options(resp http.ResponseWriter, req *http.Request) {
 342         log.Printf("%s: %s %s", GetRemoteAddress(req), req.Method, req.URL.Path)
 343         SetCorsHeaders(resp)
 344 }
 345
 346 var BadAuthorizationHeader = errors.New("Missing or invalid Authorization header")
 347 var ContentLengthMismatch = errors.New("Actual length != expected content length")
 348 var MethodNotSupported = errors.New("Method not supported")
 349
 350 var removeHint, _ = regexp.Compile("\\+K@[a-z0-9]{5}(\\+|$)")
 351
 352 func (h *proxyHandler) Get(resp http.ResponseWriter, req *http.Request) {
 353         if err := h.checkLoop(resp, req); err != nil {
 354                 return
 355         }
 356         SetCorsHeaders(resp)
 357         resp.Header().Set("Via", req.Proto+" "+viaAlias)
 358
 359         locator := mux.Vars(req)["locator"]
 360         var err error
 361         var status int
 362         var expectLength, responseLength int64
 363         var proxiedURI = "-"
 364
 365         defer func() {
 366                 log.Println(GetRemoteAddress(req), req.Method, req.URL.Path, status, expectLength, responseLength, proxiedURI, err)
 367                 if status != http.StatusOK {
 368                         http.Error(resp, err.Error(), status)
 369                 }
 370         }()
 371
 372         kc := h.makeKeepClient(req)
 373
 374         var pass bool
 375         var tok string
 376         if pass, tok = CheckAuthorizationHeader(kc, h.ApiTokenCache, req); !pass {
 377                 status, err = http.StatusForbidden, BadAuthorizationHeader
 378                 return
 379         }
 380
 381         // Copy ArvadosClient struct and use the client's API token
 382         arvclient := *kc.Arvados
 383         arvclient.ApiToken = tok
 384         kc.Arvados = &arvclient
 385
 386         var reader io.ReadCloser
 387
 388         locator = removeHint.ReplaceAllString(locator, "$1")
 389
 390         switch req.Method {
 391         case "HEAD":
 392                 expectLength, proxiedURI, err = kc.Ask(locator)
 393         case "GET":
 394                 reader, expectLength, proxiedURI, err = kc.Get(locator)
 395                 if reader != nil {
 396                         defer reader.Close()
 397                 }
 398         default:
 399                 status, err = http.StatusNotImplemented, MethodNotSupported
 400                 return
 401         }
 402
 403         if expectLength == -1 {
 404                 log.Println("Warning:", GetRemoteAddress(req), req.Method, proxiedURI, "Content-Length not provided")
 405         }
 406
 407         switch respErr := err.(type) {
 408         case nil:
 409                 status = http.StatusOK
 410                 resp.Header().Set("Content-Length", fmt.Sprint(expectLength))
 411                 switch req.Method {
 412                 case "HEAD":
 413                         responseLength = 0
 414                 case "GET":
 415                         responseLength, err = io.Copy(resp, reader)
 416                         if err == nil && expectLength > -1 && responseLength != expectLength {
 417                                 err = ContentLengthMismatch
 418                         }
 419                 }
 420         case keepclient.Error:
 421                 if respErr == keepclient.BlockNotFound {
 422                         status = http.StatusNotFound
 423                 } else if respErr.Temporary() {
 424                         status = http.StatusBadGateway
 425                 } else {
 426                         status = 422
 427                 }
 428         default:
 429                 status = http.StatusInternalServerError
 430         }
 431 }
 432
 433 var LengthRequiredError = errors.New(http.StatusText(http.StatusLengthRequired))
 434 var LengthMismatchError = errors.New("Locator size hint does not match Content-Length header")
 435
 436 func (h *proxyHandler) Put(resp http.ResponseWriter, req *http.Request) {
 437         if err := h.checkLoop(resp, req); err != nil {
 438                 return
 439         }
 440         SetCorsHeaders(resp)
 441         resp.Header().Set("Via", "HTTP/1.1 "+viaAlias)
 442
 443         kc := h.makeKeepClient(req)
 444
 445         var err error
 446         var expectLength int64
 447         var status = http.StatusInternalServerError
 448         var wroteReplicas int
 449         var locatorOut string = "-"
 450
 451         defer func() {
 452                 log.Println(GetRemoteAddress(req), req.Method, req.URL.Path, status, expectLength, kc.Want_replicas, wroteReplicas, locatorOut, err)
 453                 if status != http.StatusOK {
 454                         http.Error(resp, err.Error(), status)
 455                 }
 456         }()
 457
 458         locatorIn := mux.Vars(req)["locator"]
 459
 460         // Check if the client specified storage classes
 461         if req.Header.Get("X-Keep-Storage-Classes") != "" {
 462                 var scl []string
 463                 for _, sc := range strings.Split(req.Header.Get("X-Keep-Storage-Classes"), ",") {
 464                         scl = append(scl, strings.Trim(sc, " "))
 465                 }
 466                 kc.StorageClasses = scl
 467         }
 468
 469         _, err = fmt.Sscanf(req.Header.Get("Content-Length"), "%d", &expectLength)
 470         if err != nil || expectLength < 0 {
 471                 err = LengthRequiredError
 472                 status = http.StatusLengthRequired
 473                 return
 474         }
 475
 476         if locatorIn != "" {
 477                 var loc *keepclient.Locator
 478                 if loc, err = keepclient.MakeLocator(locatorIn); err != nil {
 479                         status = http.StatusBadRequest
 480                         return
 481                 } else if loc.Size > 0 && int64(loc.Size) != expectLength {
 482                         err = LengthMismatchError
 483                         status = http.StatusBadRequest
 484                         return
 485                 }
 486         }
 487
 488         var pass bool
 489         var tok string
 490         if pass, tok = CheckAuthorizationHeader(kc, h.ApiTokenCache, req); !pass {
 491                 err = BadAuthorizationHeader
 492                 status = http.StatusForbidden
 493                 return
 494         }
 495
 496         // Copy ArvadosClient struct and use the client's API token
 497         arvclient := *kc.Arvados
 498         arvclient.ApiToken = tok
 499         kc.Arvados = &arvclient
 500
 501         // Check if the client specified the number of replicas
 502         if req.Header.Get("X-Keep-Desired-Replicas") != "" {
 503                 var r int
 504                 _, err := fmt.Sscanf(req.Header.Get(keepclient.X_Keep_Desired_Replicas), "%d", &r)
 505                 if err == nil {
 506                         kc.Want_replicas = r
 507                 }
 508         }
 509
 510         // Now try to put the block through
 511         if locatorIn == "" {
 512                 bytes, err2 := ioutil.ReadAll(req.Body)
 513                 if err2 != nil {
 514                         err = fmt.Errorf("Error reading request body: %s", err2)
 515                         status = http.StatusInternalServerError
 516                         return
 517                 }
 518                 locatorOut, wroteReplicas, err = kc.PutB(bytes)
 519         } else {
 520                 locatorOut, wroteReplicas, err = kc.PutHR(locatorIn, req.Body, expectLength)
 521         }
 522
 523         // Tell the client how many successful PUTs we accomplished
 524         resp.Header().Set(keepclient.X_Keep_Replicas_Stored, fmt.Sprintf("%d", wroteReplicas))
 525
 526         switch err.(type) {
 527         case nil:
 528                 status = http.StatusOK
 529                 _, err = io.WriteString(resp, locatorOut)
 530
 531         case keepclient.OversizeBlockError:
 532                 // Too much data
 533                 status = http.StatusRequestEntityTooLarge
 534
 535         case keepclient.InsufficientReplicasError:
 536                 if wroteReplicas > 0 {
 537                         // At least one write is considered success.  The
 538                         // client can decide if getting less than the number of
 539                         // replications it asked for is a fatal error.
 540                         status = http.StatusOK
 541                         _, err = io.WriteString(resp, locatorOut)
 542                 } else {
 543                         status = http.StatusServiceUnavailable
 544                 }
 545
 546         default:
 547                 status = http.StatusBadGateway
 548         }
 549 }
 550
 551 // ServeHTTP implementation for IndexHandler
 552 // Supports only GET requests for /index/{prefix:[0-9a-f]{0,32}}
 553 // For each keep server found in LocalRoots:
 554 //   Invokes GetIndex using keepclient
 555 //   Expects "complete" response (terminating with blank new line)
 556 //   Aborts on any errors
 557 // Concatenates responses from all those keep servers and returns
 558 func (h *proxyHandler) Index(resp http.ResponseWriter, req *http.Request) {
 559         SetCorsHeaders(resp)
 560
 561         prefix := mux.Vars(req)["prefix"]
 562         var err error
 563         var status int
 564
 565         defer func() {
 566                 if status != http.StatusOK {
 567                         http.Error(resp, err.Error(), status)
 568                 }
 569         }()
 570
 571         kc := h.makeKeepClient(req)
 572         ok, token := CheckAuthorizationHeader(kc, h.ApiTokenCache, req)
 573         if !ok {
 574                 status, err = http.StatusForbidden, BadAuthorizationHeader
 575                 return
 576         }
 577
 578         // Copy ArvadosClient struct and use the client's API token
 579         arvclient := *kc.Arvados
 580         arvclient.ApiToken = token
 581         kc.Arvados = &arvclient
 582
 583         // Only GET method is supported
 584         if req.Method != "GET" {
 585                 status, err = http.StatusNotImplemented, MethodNotSupported
 586                 return
 587         }
 588
 589         // Get index from all LocalRoots and write to resp
 590         var reader io.Reader
 591         for uuid := range kc.LocalRoots() {
 592                 reader, err = kc.GetIndex(uuid, prefix)
 593                 if err != nil {
 594                         status = http.StatusBadGateway
 595                         return
 596                 }
 597
 598                 _, err = io.Copy(resp, reader)
 599                 if err != nil {
 600                         status = http.StatusBadGateway
 601                         return
 602                 }
 603         }
 604
 605         // Got index from all the keep servers and wrote to resp
 606         status = http.StatusOK
 607         resp.Write([]byte("\n"))
 608 }
 609
 610 func (h *proxyHandler) makeKeepClient(req *http.Request) *keepclient.KeepClient {
 611         kc := *h.KeepClient
 612         kc.RequestID = req.Header.Get("X-Request-Id")
 613         kc.HTTPClient = &proxyClient{
 614                 client: &http.Client{
 615                         Timeout:   h.timeout,
 616                         Transport: h.transport,
 617                 },
 618                 proto: req.Proto,
 619         }
 620         return &kc
 621 }