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.curoverse.com/arvados.git/sdk/go/arvados"
  24         "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
  25         "git.curoverse.com/arvados.git/sdk/go/config"
  26         "git.curoverse.com/arvados.git/sdk/go/health"
  27         "git.curoverse.com/arvados.git/sdk/go/httpserver"
  28         "git.curoverse.com/arvados.git/sdk/go/keepclient"
  29         log "github.com/Sirupsen/logrus"
  30         "github.com/coreos/go-systemd/daemon"
  31         "github.com/ghodss/yaml"
  32         "github.com/gorilla/mux"
  33 )
  34
  35 var version = "dev"
  36
  37 type Config struct {
  38         Client          arvados.Client
  39         Listen          string
  40         DisableGet      bool
  41         DisablePut      bool
  42         DefaultReplicas int
  43         Timeout         arvados.Duration
  44         PIDFile         string
  45         Debug           bool
  46         ManagementToken string
  47 }
  48
  49 func DefaultConfig() *Config {
  50         return &Config{
  51                 Listen:  ":25107",
  52                 Timeout: arvados.Duration(15 * time.Second),
  53         }
  54 }
  55
  56 var (
  57         listener net.Listener
  58         router   http.Handler
  59 )
  60
  61 const rfc3339NanoFixed = "2006-01-02T15:04:05.000000000Z07:00"
  62
  63 func main() {
  64         log.SetFormatter(&log.JSONFormatter{
  65                 TimestampFormat: rfc3339NanoFixed,
  66         })
  67
  68         cfg := DefaultConfig()
  69
  70         flagset := flag.NewFlagSet("keepproxy", flag.ExitOnError)
  71         flagset.Usage = usage
  72
  73         const deprecated = " (DEPRECATED -- use config file instead)"
  74         flagset.StringVar(&cfg.Listen, "listen", cfg.Listen, "Local port to listen on."+deprecated)
  75         flagset.BoolVar(&cfg.DisableGet, "no-get", cfg.DisableGet, "Disable GET operations."+deprecated)
  76         flagset.BoolVar(&cfg.DisablePut, "no-put", cfg.DisablePut, "Disable PUT operations."+deprecated)
  77         flagset.IntVar(&cfg.DefaultReplicas, "default-replicas", cfg.DefaultReplicas, "Default number of replicas to write if not specified by the client. If 0, use site default."+deprecated)
  78         flagset.StringVar(&cfg.PIDFile, "pid", cfg.PIDFile, "Path to write pid file."+deprecated)
  79         timeoutSeconds := flagset.Int("timeout", int(time.Duration(cfg.Timeout)/time.Second), "Timeout (in seconds) on requests to internal Keep services."+deprecated)
  80         flagset.StringVar(&cfg.ManagementToken, "management-token", cfg.ManagementToken, "Authorization token to be included in all health check requests.")
  81
  82         var cfgPath string
  83         const defaultCfgPath = "/etc/arvados/keepproxy/keepproxy.yml"
  84         flagset.StringVar(&cfgPath, "config", defaultCfgPath, "Configuration file `path`")
  85         dumpConfig := flagset.Bool("dump-config", false, "write current configuration to stdout and exit")
  86         getVersion := flagset.Bool("version", false, "Print version information and exit.")
  87         flagset.Parse(os.Args[1:])
  88
  89         // Print version information if requested
  90         if *getVersion {
  91                 fmt.Printf("keepproxy %s\n", version)
  92                 return
  93         }
  94
  95         err := config.LoadFile(cfg, cfgPath)
  96         if err != nil {
  97                 h := os.Getenv("ARVADOS_API_HOST")
  98                 t := os.Getenv("ARVADOS_API_TOKEN")
  99                 if h == "" || t == "" || !os.IsNotExist(err) || cfgPath != defaultCfgPath {
 100                         log.Fatal(err)
 101                 }
 102                 log.Print("DEPRECATED: No config file found, but ARVADOS_API_HOST and ARVADOS_API_TOKEN environment variables are set. Please use a config file instead.")
 103                 cfg.Client.APIHost = h
 104                 cfg.Client.AuthToken = t
 105                 if regexp.MustCompile("^(?i:1|yes|true)$").MatchString(os.Getenv("ARVADOS_API_HOST_INSECURE")) {
 106                         cfg.Client.Insecure = true
 107                 }
 108                 if y, err := yaml.Marshal(cfg); err == nil && !*dumpConfig {
 109                         log.Print("Current configuration:\n", string(y))
 110                 }
 111                 cfg.Timeout = arvados.Duration(time.Duration(*timeoutSeconds) * time.Second)
 112         }
 113
 114         if *dumpConfig {
 115                 log.Fatal(config.DumpAndExit(cfg))
 116         }
 117
 118         log.Printf("keepproxy %s started", version)
 119
 120         arv, err := arvadosclient.New(&cfg.Client)
 121         if err != nil {
 122                 log.Fatalf("Error setting up arvados client %s", err.Error())
 123         }
 124
 125         if cfg.Debug {
 126                 keepclient.DebugPrintf = log.Printf
 127         }
 128         kc, err := keepclient.MakeKeepClient(arv)
 129         if err != nil {
 130                 log.Fatalf("Error setting up keep client %s", err.Error())
 131         }
 132         keepclient.RefreshServiceDiscoveryOnSIGHUP()
 133
 134         if cfg.PIDFile != "" {
 135                 f, err := os.Create(cfg.PIDFile)
 136                 if err != nil {
 137                         log.Fatal(err)
 138                 }
 139                 defer f.Close()
 140                 err = syscall.Flock(int(f.Fd()), syscall.LOCK_EX|syscall.LOCK_NB)
 141                 if err != nil {
 142                         log.Fatalf("flock(%s): %s", cfg.PIDFile, err)
 143                 }
 144                 defer os.Remove(cfg.PIDFile)
 145                 err = f.Truncate(0)
 146                 if err != nil {
 147                         log.Fatalf("truncate(%s): %s", cfg.PIDFile, err)
 148                 }
 149                 _, err = fmt.Fprint(f, os.Getpid())
 150                 if err != nil {
 151                         log.Fatalf("write(%s): %s", cfg.PIDFile, err)
 152                 }
 153                 err = f.Sync()
 154                 if err != nil {
 155                         log.Fatal("sync(%s): %s", cfg.PIDFile, err)
 156                 }
 157         }
 158
 159         if cfg.DefaultReplicas > 0 {
 160                 kc.Want_replicas = cfg.DefaultReplicas
 161         }
 162
 163         listener, err = net.Listen("tcp", cfg.Listen)
 164         if err != nil {
 165                 log.Fatalf("listen(%s): %s", cfg.Listen, err)
 166         }
 167         if _, err := daemon.SdNotify(false, "READY=1"); err != nil {
 168                 log.Printf("Error notifying init daemon: %v", err)
 169         }
 170         log.Println("Listening at", listener.Addr())
 171
 172         // Shut down the server gracefully (by closing the listener)
 173         // if SIGTERM is received.
 174         term := make(chan os.Signal, 1)
 175         go func(sig <-chan os.Signal) {
 176                 s := <-sig
 177                 log.Println("caught signal:", s)
 178                 listener.Close()
 179         }(term)
 180         signal.Notify(term, syscall.SIGTERM)
 181         signal.Notify(term, syscall.SIGINT)
 182
 183         // Start serving requests.
 184         router = MakeRESTRouter(!cfg.DisableGet, !cfg.DisablePut, kc, time.Duration(cfg.Timeout), cfg.ManagementToken)
 185         http.Serve(listener, httpserver.AddRequestIDs(httpserver.LogRequests(router)))
 186
 187         log.Println("shutting down")
 188 }
 189
 190 type ApiTokenCache struct {
 191         tokens     map[string]int64
 192         lock       sync.Mutex
 193         expireTime int64
 194 }
 195
 196 // Cache the token and set an expire time.  If we already have an expire time
 197 // on the token, it is not updated.
 198 func (this *ApiTokenCache) RememberToken(token string) {
 199         this.lock.Lock()
 200         defer this.lock.Unlock()
 201
 202         now := time.Now().Unix()
 203         if this.tokens[token] == 0 {
 204                 this.tokens[token] = now + this.expireTime
 205         }
 206 }
 207
 208 // Check if the cached token is known and still believed to be valid.
 209 func (this *ApiTokenCache) RecallToken(token string) bool {
 210         this.lock.Lock()
 211         defer this.lock.Unlock()
 212
 213         now := time.Now().Unix()
 214         if this.tokens[token] == 0 {
 215                 // Unknown token
 216                 return false
 217         } else if now < this.tokens[token] {
 218                 // Token is known and still valid
 219                 return true
 220         } else {
 221                 // Token is expired
 222                 this.tokens[token] = 0
 223                 return false
 224         }
 225 }
 226
 227 func GetRemoteAddress(req *http.Request) string {
 228         if xff := req.Header.Get("X-Forwarded-For"); xff != "" {
 229                 return xff + "," + req.RemoteAddr
 230         }
 231         return req.RemoteAddr
 232 }
 233
 234 func CheckAuthorizationHeader(kc *keepclient.KeepClient, cache *ApiTokenCache, req *http.Request) (pass bool, tok string) {
 235         var auth string
 236         if auth = req.Header.Get("Authorization"); auth == "" {
 237                 return false, ""
 238         }
 239
 240         _, err := fmt.Sscanf(auth, "OAuth2 %s", &tok)
 241         if err != nil {
 242                 // Scanning error
 243                 return false, ""
 244         }
 245
 246         if cache.RecallToken(tok) {
 247                 // Valid in the cache, short circuit
 248                 return true, tok
 249         }
 250
 251         arv := *kc.Arvados
 252         arv.ApiToken = tok
 253         if err := arv.Call("HEAD", "users", "", "current", nil, nil); err != nil {
 254                 log.Printf("%s: CheckAuthorizationHeader error: %v", GetRemoteAddress(req), err)
 255                 return false, ""
 256         }
 257
 258         // Success!  Update cache
 259         cache.RememberToken(tok)
 260
 261         return true, tok
 262 }
 263
 264 type proxyHandler struct {
 265         http.Handler
 266         *keepclient.KeepClient
 267         *ApiTokenCache
 268         timeout   time.Duration
 269         transport *http.Transport
 270 }
 271
 272 // MakeRESTRouter returns an http.Handler that passes GET and PUT
 273 // requests to the appropriate handlers.
 274 func MakeRESTRouter(enable_get bool, enable_put bool, kc *keepclient.KeepClient, timeout time.Duration, mgmtToken string) http.Handler {
 275         rest := mux.NewRouter()
 276
 277         transport := *(http.DefaultTransport.(*http.Transport))
 278         transport.DialContext = (&net.Dialer{
 279                 Timeout:   keepclient.DefaultConnectTimeout,
 280                 KeepAlive: keepclient.DefaultKeepAlive,
 281                 DualStack: true,
 282         }).DialContext
 283         transport.TLSClientConfig = arvadosclient.MakeTLSConfig(kc.Arvados.ApiInsecure)
 284         transport.TLSHandshakeTimeout = keepclient.DefaultTLSHandshakeTimeout
 285
 286         h := &proxyHandler{
 287                 Handler:    rest,
 288                 KeepClient: kc,
 289                 timeout:    timeout,
 290                 transport:  &transport,
 291                 ApiTokenCache: &ApiTokenCache{
 292                         tokens:     make(map[string]int64),
 293                         expireTime: 300,
 294                 },
 295         }
 296
 297         if enable_get {
 298                 rest.HandleFunc(`/{locator:[0-9a-f]{32}\+.*}`, h.Get).Methods("GET", "HEAD")
 299                 rest.HandleFunc(`/{locator:[0-9a-f]{32}}`, h.Get).Methods("GET", "HEAD")
 300
 301                 // List all blocks
 302                 rest.HandleFunc(`/index`, h.Index).Methods("GET")
 303
 304                 // List blocks whose hash has the given prefix
 305                 rest.HandleFunc(`/index/{prefix:[0-9a-f]{0,32}}`, h.Index).Methods("GET")
 306         }
 307
 308         if enable_put {
 309                 rest.HandleFunc(`/{locator:[0-9a-f]{32}\+.*}`, h.Put).Methods("PUT")
 310                 rest.HandleFunc(`/{locator:[0-9a-f]{32}}`, h.Put).Methods("PUT")
 311                 rest.HandleFunc(`/`, h.Put).Methods("POST")
 312                 rest.HandleFunc(`/{any}`, h.Options).Methods("OPTIONS")
 313                 rest.HandleFunc(`/`, h.Options).Methods("OPTIONS")
 314         }
 315
 316         rest.Handle("/_health/{check}", &health.Handler{
 317                 Token:  mgmtToken,
 318                 Prefix: "/_health/",
 319         }).Methods("GET")
 320
 321         rest.NotFoundHandler = InvalidPathHandler{}
 322         return h
 323 }
 324
 325 var errLoopDetected = errors.New("loop detected")
 326
 327 func (*proxyHandler) checkLoop(resp http.ResponseWriter, req *http.Request) error {
 328         if via := req.Header.Get("Via"); strings.Index(via, " "+viaAlias) >= 0 {
 329                 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)
 330                 http.Error(resp, errLoopDetected.Error(), http.StatusInternalServerError)
 331                 return errLoopDetected
 332         }
 333         return nil
 334 }
 335
 336 func SetCorsHeaders(resp http.ResponseWriter) {
 337         resp.Header().Set("Access-Control-Allow-Methods", "GET, HEAD, POST, PUT, OPTIONS")
 338         resp.Header().Set("Access-Control-Allow-Origin", "*")
 339         resp.Header().Set("Access-Control-Allow-Headers", "Authorization, Content-Length, Content-Type, X-Keep-Desired-Replicas")
 340         resp.Header().Set("Access-Control-Max-Age", "86486400")
 341 }
 342
 343 type InvalidPathHandler struct{}
 344
 345 func (InvalidPathHandler) ServeHTTP(resp http.ResponseWriter, req *http.Request) {
 346         log.Printf("%s: %s %s unroutable", GetRemoteAddress(req), req.Method, req.URL.Path)
 347         http.Error(resp, "Bad request", http.StatusBadRequest)
 348 }
 349
 350 func (h *proxyHandler) Options(resp http.ResponseWriter, req *http.Request) {
 351         log.Printf("%s: %s %s", GetRemoteAddress(req), req.Method, req.URL.Path)
 352         SetCorsHeaders(resp)
 353 }
 354
 355 var BadAuthorizationHeader = errors.New("Missing or invalid Authorization header")
 356 var ContentLengthMismatch = errors.New("Actual length != expected content length")
 357 var MethodNotSupported = errors.New("Method not supported")
 358
 359 var removeHint, _ = regexp.Compile("\\+K@[a-z0-9]{5}(\\+|$)")
 360
 361 func (h *proxyHandler) Get(resp http.ResponseWriter, req *http.Request) {
 362         if err := h.checkLoop(resp, req); err != nil {
 363                 return
 364         }
 365         SetCorsHeaders(resp)
 366         resp.Header().Set("Via", req.Proto+" "+viaAlias)
 367
 368         locator := mux.Vars(req)["locator"]
 369         var err error
 370         var status int
 371         var expectLength, responseLength int64
 372         var proxiedURI = "-"
 373
 374         defer func() {
 375                 log.Println(GetRemoteAddress(req), req.Method, req.URL.Path, status, expectLength, responseLength, proxiedURI, err)
 376                 if status != http.StatusOK {
 377                         http.Error(resp, err.Error(), status)
 378                 }
 379         }()
 380
 381         kc := h.makeKeepClient(req)
 382
 383         var pass bool
 384         var tok string
 385         if pass, tok = CheckAuthorizationHeader(kc, h.ApiTokenCache, req); !pass {
 386                 status, err = http.StatusForbidden, BadAuthorizationHeader
 387                 return
 388         }
 389
 390         // Copy ArvadosClient struct and use the client's API token
 391         arvclient := *kc.Arvados
 392         arvclient.ApiToken = tok
 393         kc.Arvados = &arvclient
 394
 395         var reader io.ReadCloser
 396
 397         locator = removeHint.ReplaceAllString(locator, "$1")
 398
 399         switch req.Method {
 400         case "HEAD":
 401                 expectLength, proxiedURI, err = kc.Ask(locator)
 402         case "GET":
 403                 reader, expectLength, proxiedURI, err = kc.Get(locator)
 404                 if reader != nil {
 405                         defer reader.Close()
 406                 }
 407         default:
 408                 status, err = http.StatusNotImplemented, MethodNotSupported
 409                 return
 410         }
 411
 412         if expectLength == -1 {
 413                 log.Println("Warning:", GetRemoteAddress(req), req.Method, proxiedURI, "Content-Length not provided")
 414         }
 415
 416         switch respErr := err.(type) {
 417         case nil:
 418                 status = http.StatusOK
 419                 resp.Header().Set("Content-Length", fmt.Sprint(expectLength))
 420                 switch req.Method {
 421                 case "HEAD":
 422                         responseLength = 0
 423                 case "GET":
 424                         responseLength, err = io.Copy(resp, reader)
 425                         if err == nil && expectLength > -1 && responseLength != expectLength {
 426                                 err = ContentLengthMismatch
 427                         }
 428                 }
 429         case keepclient.Error:
 430                 if respErr == keepclient.BlockNotFound {
 431                         status = http.StatusNotFound
 432                 } else if respErr.Temporary() {
 433                         status = http.StatusBadGateway
 434                 } else {
 435                         status = 422
 436                 }
 437         default:
 438                 status = http.StatusInternalServerError
 439         }
 440 }
 441
 442 var LengthRequiredError = errors.New(http.StatusText(http.StatusLengthRequired))
 443 var LengthMismatchError = errors.New("Locator size hint does not match Content-Length header")
 444
 445 func (h *proxyHandler) Put(resp http.ResponseWriter, req *http.Request) {
 446         if err := h.checkLoop(resp, req); err != nil {
 447                 return
 448         }
 449         SetCorsHeaders(resp)
 450         resp.Header().Set("Via", "HTTP/1.1 "+viaAlias)
 451
 452         kc := h.makeKeepClient(req)
 453
 454         var err error
 455         var expectLength int64
 456         var status = http.StatusInternalServerError
 457         var wroteReplicas int
 458         var locatorOut string = "-"
 459
 460         defer func() {
 461                 log.Println(GetRemoteAddress(req), req.Method, req.URL.Path, status, expectLength, kc.Want_replicas, wroteReplicas, locatorOut, err)
 462                 if status != http.StatusOK {
 463                         http.Error(resp, err.Error(), status)
 464                 }
 465         }()
 466
 467         locatorIn := mux.Vars(req)["locator"]
 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                 if bytes, err := ioutil.ReadAll(req.Body); err != nil {
 513                         err = errors.New(fmt.Sprintf("Error reading request body: %s", err))
 514                         status = http.StatusInternalServerError
 515                         return
 516                 } else {
 517                         locatorOut, wroteReplicas, err = kc.PutB(bytes)
 518                 }
 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.HTTPClient = &proxyClient{
 613                 client: &http.Client{
 614                         Timeout:   h.timeout,
 615                         Transport: h.transport,
 616                 },
 617                 proto:     req.Proto,
 618                 requestID: req.Header.Get("X-Request-Id"),
 619         }
 620         return &kc
 621 }