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         "log"
  14         "net"
  15         "net/http"
  16         "os"
  17         "os/signal"
  18         "regexp"
  19         "strings"
  20         "sync"
  21         "syscall"
  22         "time"
  23
  24         "git.curoverse.com/arvados.git/sdk/go/arvados"
  25         "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
  26         "git.curoverse.com/arvados.git/sdk/go/config"
  27         "git.curoverse.com/arvados.git/sdk/go/health"
  28         "git.curoverse.com/arvados.git/sdk/go/keepclient"
  29         "github.com/coreos/go-systemd/daemon"
  30         "github.com/ghodss/yaml"
  31         "github.com/gorilla/mux"
  32 )
  33
  34 type Config struct {
  35         Client          arvados.Client
  36         Listen          string
  37         DisableGet      bool
  38         DisablePut      bool
  39         DefaultReplicas int
  40         Timeout         arvados.Duration
  41         PIDFile         string
  42         Debug           bool
  43         ManagementToken string
  44 }
  45
  46 func DefaultConfig() *Config {
  47         return &Config{
  48                 Listen:  ":25107",
  49                 Timeout: arvados.Duration(15 * time.Second),
  50         }
  51 }
  52
  53 var (
  54         listener net.Listener
  55         router   http.Handler
  56 )
  57
  58 func main() {
  59         cfg := DefaultConfig()
  60
  61         flagset := flag.NewFlagSet("keepproxy", flag.ExitOnError)
  62         flagset.Usage = usage
  63
  64         const deprecated = " (DEPRECATED -- use config file instead)"
  65         flagset.StringVar(&cfg.Listen, "listen", cfg.Listen, "Local port to listen on."+deprecated)
  66         flagset.BoolVar(&cfg.DisableGet, "no-get", cfg.DisableGet, "Disable GET operations."+deprecated)
  67         flagset.BoolVar(&cfg.DisablePut, "no-put", cfg.DisablePut, "Disable PUT operations."+deprecated)
  68         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)
  69         flagset.StringVar(&cfg.PIDFile, "pid", cfg.PIDFile, "Path to write pid file."+deprecated)
  70         timeoutSeconds := flagset.Int("timeout", int(time.Duration(cfg.Timeout)/time.Second), "Timeout (in seconds) on requests to internal Keep services."+deprecated)
  71         flagset.StringVar(&cfg.ManagementToken, "management-token", cfg.ManagementToken, "Authorization token to be included in all health check requests.")
  72
  73         var cfgPath string
  74         const defaultCfgPath = "/etc/arvados/keepproxy/keepproxy.yml"
  75         flagset.StringVar(&cfgPath, "config", defaultCfgPath, "Configuration file `path`")
  76         dumpConfig := flagset.Bool("dump-config", false, "write current configuration to stdout and exit")
  77         flagset.Parse(os.Args[1:])
  78
  79         err := config.LoadFile(cfg, cfgPath)
  80         if err != nil {
  81                 h := os.Getenv("ARVADOS_API_HOST")
  82                 t := os.Getenv("ARVADOS_API_TOKEN")
  83                 if h == "" || t == "" || !os.IsNotExist(err) || cfgPath != defaultCfgPath {
  84                         log.Fatal(err)
  85                 }
  86                 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.")
  87                 cfg.Client.APIHost = h
  88                 cfg.Client.AuthToken = t
  89                 if regexp.MustCompile("^(?i:1|yes|true)$").MatchString(os.Getenv("ARVADOS_API_HOST_INSECURE")) {
  90                         cfg.Client.Insecure = true
  91                 }
  92                 if y, err := yaml.Marshal(cfg); err == nil && !*dumpConfig {
  93                         log.Print("Current configuration:\n", string(y))
  94                 }
  95                 cfg.Timeout = arvados.Duration(time.Duration(*timeoutSeconds) * time.Second)
  96         }
  97
  98         if *dumpConfig {
  99                 log.Fatal(config.DumpAndExit(cfg))
 100         }
 101
 102         arv, err := arvadosclient.New(&cfg.Client)
 103         if err != nil {
 104                 log.Fatalf("Error setting up arvados client %s", err.Error())
 105         }
 106
 107         if cfg.Debug {
 108                 keepclient.DebugPrintf = log.Printf
 109         }
 110         kc, err := keepclient.MakeKeepClient(arv)
 111         if err != nil {
 112                 log.Fatalf("Error setting up keep client %s", err.Error())
 113         }
 114         keepclient.RefreshServiceDiscoveryOnSIGHUP()
 115
 116         if cfg.PIDFile != "" {
 117                 f, err := os.Create(cfg.PIDFile)
 118                 if err != nil {
 119                         log.Fatal(err)
 120                 }
 121                 defer f.Close()
 122                 err = syscall.Flock(int(f.Fd()), syscall.LOCK_EX|syscall.LOCK_NB)
 123                 if err != nil {
 124                         log.Fatalf("flock(%s): %s", cfg.PIDFile, err)
 125                 }
 126                 defer os.Remove(cfg.PIDFile)
 127                 err = f.Truncate(0)
 128                 if err != nil {
 129                         log.Fatalf("truncate(%s): %s", cfg.PIDFile, err)
 130                 }
 131                 _, err = fmt.Fprint(f, os.Getpid())
 132                 if err != nil {
 133                         log.Fatalf("write(%s): %s", cfg.PIDFile, err)
 134                 }
 135                 err = f.Sync()
 136                 if err != nil {
 137                         log.Fatal("sync(%s): %s", cfg.PIDFile, err)
 138                 }
 139         }
 140
 141         if cfg.DefaultReplicas > 0 {
 142                 kc.Want_replicas = cfg.DefaultReplicas
 143         }
 144
 145         listener, err = net.Listen("tcp", cfg.Listen)
 146         if err != nil {
 147                 log.Fatalf("listen(%s): %s", cfg.Listen, err)
 148         }
 149         if _, err := daemon.SdNotify(false, "READY=1"); err != nil {
 150                 log.Printf("Error notifying init daemon: %v", err)
 151         }
 152         log.Println("Listening at", listener.Addr())
 153
 154         // Shut down the server gracefully (by closing the listener)
 155         // if SIGTERM is received.
 156         term := make(chan os.Signal, 1)
 157         go func(sig <-chan os.Signal) {
 158                 s := <-sig
 159                 log.Println("caught signal:", s)
 160                 listener.Close()
 161         }(term)
 162         signal.Notify(term, syscall.SIGTERM)
 163         signal.Notify(term, syscall.SIGINT)
 164
 165         // Start serving requests.
 166         router = MakeRESTRouter(!cfg.DisableGet, !cfg.DisablePut, kc, time.Duration(cfg.Timeout), cfg.ManagementToken)
 167         http.Serve(listener, router)
 168
 169         log.Println("shutting down")
 170 }
 171
 172 type ApiTokenCache struct {
 173         tokens     map[string]int64
 174         lock       sync.Mutex
 175         expireTime int64
 176 }
 177
 178 // Cache the token and set an expire time.  If we already have an expire time
 179 // on the token, it is not updated.
 180 func (this *ApiTokenCache) RememberToken(token string) {
 181         this.lock.Lock()
 182         defer this.lock.Unlock()
 183
 184         now := time.Now().Unix()
 185         if this.tokens[token] == 0 {
 186                 this.tokens[token] = now + this.expireTime
 187         }
 188 }
 189
 190 // Check if the cached token is known and still believed to be valid.
 191 func (this *ApiTokenCache) RecallToken(token string) bool {
 192         this.lock.Lock()
 193         defer this.lock.Unlock()
 194
 195         now := time.Now().Unix()
 196         if this.tokens[token] == 0 {
 197                 // Unknown token
 198                 return false
 199         } else if now < this.tokens[token] {
 200                 // Token is known and still valid
 201                 return true
 202         } else {
 203                 // Token is expired
 204                 this.tokens[token] = 0
 205                 return false
 206         }
 207 }
 208
 209 func GetRemoteAddress(req *http.Request) string {
 210         if xff := req.Header.Get("X-Forwarded-For"); xff != "" {
 211                 return xff + "," + req.RemoteAddr
 212         }
 213         return req.RemoteAddr
 214 }
 215
 216 func CheckAuthorizationHeader(kc *keepclient.KeepClient, cache *ApiTokenCache, req *http.Request) (pass bool, tok string) {
 217         var auth string
 218         if auth = req.Header.Get("Authorization"); auth == "" {
 219                 return false, ""
 220         }
 221
 222         _, err := fmt.Sscanf(auth, "OAuth2 %s", &tok)
 223         if err != nil {
 224                 // Scanning error
 225                 return false, ""
 226         }
 227
 228         if cache.RecallToken(tok) {
 229                 // Valid in the cache, short circuit
 230                 return true, tok
 231         }
 232
 233         arv := *kc.Arvados
 234         arv.ApiToken = tok
 235         if err := arv.Call("HEAD", "users", "", "current", nil, nil); err != nil {
 236                 log.Printf("%s: CheckAuthorizationHeader error: %v", GetRemoteAddress(req), err)
 237                 return false, ""
 238         }
 239
 240         // Success!  Update cache
 241         cache.RememberToken(tok)
 242
 243         return true, tok
 244 }
 245
 246 type proxyHandler struct {
 247         http.Handler
 248         *keepclient.KeepClient
 249         *ApiTokenCache
 250         timeout   time.Duration
 251         transport *http.Transport
 252 }
 253
 254 // MakeRESTRouter returns an http.Handler that passes GET and PUT
 255 // requests to the appropriate handlers.
 256 func MakeRESTRouter(enable_get bool, enable_put bool, kc *keepclient.KeepClient, timeout time.Duration, mgmtToken string) http.Handler {
 257         rest := mux.NewRouter()
 258
 259         transport := *(http.DefaultTransport.(*http.Transport))
 260         transport.DialContext = (&net.Dialer{
 261                 Timeout:   keepclient.DefaultConnectTimeout,
 262                 KeepAlive: keepclient.DefaultKeepAlive,
 263                 DualStack: true,
 264         }).DialContext
 265         transport.TLSClientConfig = arvadosclient.MakeTLSConfig(kc.Arvados.ApiInsecure)
 266         transport.TLSHandshakeTimeout = keepclient.DefaultTLSHandshakeTimeout
 267
 268         h := &proxyHandler{
 269                 Handler:    rest,
 270                 KeepClient: kc,
 271                 timeout:    timeout,
 272                 transport:  &transport,
 273                 ApiTokenCache: &ApiTokenCache{
 274                         tokens:     make(map[string]int64),
 275                         expireTime: 300,
 276                 },
 277         }
 278
 279         if enable_get {
 280                 rest.HandleFunc(`/{locator:[0-9a-f]{32}\+.*}`, h.Get).Methods("GET", "HEAD")
 281                 rest.HandleFunc(`/{locator:[0-9a-f]{32}}`, h.Get).Methods("GET", "HEAD")
 282
 283                 // List all blocks
 284                 rest.HandleFunc(`/index`, h.Index).Methods("GET")
 285
 286                 // List blocks whose hash has the given prefix
 287                 rest.HandleFunc(`/index/{prefix:[0-9a-f]{0,32}}`, h.Index).Methods("GET")
 288         }
 289
 290         if enable_put {
 291                 rest.HandleFunc(`/{locator:[0-9a-f]{32}\+.*}`, h.Put).Methods("PUT")
 292                 rest.HandleFunc(`/{locator:[0-9a-f]{32}}`, h.Put).Methods("PUT")
 293                 rest.HandleFunc(`/`, h.Put).Methods("POST")
 294                 rest.HandleFunc(`/{any}`, h.Options).Methods("OPTIONS")
 295                 rest.HandleFunc(`/`, h.Options).Methods("OPTIONS")
 296         }
 297
 298         rest.Handle("/_health/{check}", &health.Handler{
 299                 Token:  mgmtToken,
 300                 Prefix: "/_health/",
 301         }).Methods("GET")
 302
 303         rest.NotFoundHandler = InvalidPathHandler{}
 304         return h
 305 }
 306
 307 var errLoopDetected = errors.New("loop detected")
 308
 309 func (*proxyHandler) checkLoop(resp http.ResponseWriter, req *http.Request) error {
 310         if via := req.Header.Get("Via"); strings.Index(via, " "+viaAlias) >= 0 {
 311                 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)
 312                 http.Error(resp, errLoopDetected.Error(), http.StatusInternalServerError)
 313                 return errLoopDetected
 314         }
 315         return nil
 316 }
 317
 318 func SetCorsHeaders(resp http.ResponseWriter) {
 319         resp.Header().Set("Access-Control-Allow-Methods", "GET, HEAD, POST, PUT, OPTIONS")
 320         resp.Header().Set("Access-Control-Allow-Origin", "*")
 321         resp.Header().Set("Access-Control-Allow-Headers", "Authorization, Content-Length, Content-Type, X-Keep-Desired-Replicas")
 322         resp.Header().Set("Access-Control-Max-Age", "86486400")
 323 }
 324
 325 type InvalidPathHandler struct{}
 326
 327 func (InvalidPathHandler) ServeHTTP(resp http.ResponseWriter, req *http.Request) {
 328         log.Printf("%s: %s %s unroutable", GetRemoteAddress(req), req.Method, req.URL.Path)
 329         http.Error(resp, "Bad request", http.StatusBadRequest)
 330 }
 331
 332 func (h *proxyHandler) Options(resp http.ResponseWriter, req *http.Request) {
 333         log.Printf("%s: %s %s", GetRemoteAddress(req), req.Method, req.URL.Path)
 334         SetCorsHeaders(resp)
 335 }
 336
 337 var BadAuthorizationHeader = errors.New("Missing or invalid Authorization header")
 338 var ContentLengthMismatch = errors.New("Actual length != expected content length")
 339 var MethodNotSupported = errors.New("Method not supported")
 340
 341 var removeHint, _ = regexp.Compile("\\+K@[a-z0-9]{5}(\\+|$)")
 342
 343 func (h *proxyHandler) Get(resp http.ResponseWriter, req *http.Request) {
 344         if err := h.checkLoop(resp, req); err != nil {
 345                 return
 346         }
 347         SetCorsHeaders(resp)
 348         resp.Header().Set("Via", req.Proto+" "+viaAlias)
 349
 350         locator := mux.Vars(req)["locator"]
 351         var err error
 352         var status int
 353         var expectLength, responseLength int64
 354         var proxiedURI = "-"
 355
 356         defer func() {
 357                 log.Println(GetRemoteAddress(req), req.Method, req.URL.Path, status, expectLength, responseLength, proxiedURI, err)
 358                 if status != http.StatusOK {
 359                         http.Error(resp, err.Error(), status)
 360                 }
 361         }()
 362
 363         kc := h.makeKeepClient(req)
 364
 365         var pass bool
 366         var tok string
 367         if pass, tok = CheckAuthorizationHeader(kc, h.ApiTokenCache, req); !pass {
 368                 status, err = http.StatusForbidden, BadAuthorizationHeader
 369                 return
 370         }
 371
 372         // Copy ArvadosClient struct and use the client's API token
 373         arvclient := *kc.Arvados
 374         arvclient.ApiToken = tok
 375         kc.Arvados = &arvclient
 376
 377         var reader io.ReadCloser
 378
 379         locator = removeHint.ReplaceAllString(locator, "$1")
 380
 381         switch req.Method {
 382         case "HEAD":
 383                 expectLength, proxiedURI, err = kc.Ask(locator)
 384         case "GET":
 385                 reader, expectLength, proxiedURI, err = kc.Get(locator)
 386                 if reader != nil {
 387                         defer reader.Close()
 388                 }
 389         default:
 390                 status, err = http.StatusNotImplemented, MethodNotSupported
 391                 return
 392         }
 393
 394         if expectLength == -1 {
 395                 log.Println("Warning:", GetRemoteAddress(req), req.Method, proxiedURI, "Content-Length not provided")
 396         }
 397
 398         switch respErr := err.(type) {
 399         case nil:
 400                 status = http.StatusOK
 401                 resp.Header().Set("Content-Length", fmt.Sprint(expectLength))
 402                 switch req.Method {
 403                 case "HEAD":
 404                         responseLength = 0
 405                 case "GET":
 406                         responseLength, err = io.Copy(resp, reader)
 407                         if err == nil && expectLength > -1 && responseLength != expectLength {
 408                                 err = ContentLengthMismatch
 409                         }
 410                 }
 411         case keepclient.Error:
 412                 if respErr == keepclient.BlockNotFound {
 413                         status = http.StatusNotFound
 414                 } else if respErr.Temporary() {
 415                         status = http.StatusBadGateway
 416                 } else {
 417                         status = 422
 418                 }
 419         default:
 420                 status = http.StatusInternalServerError
 421         }
 422 }
 423
 424 var LengthRequiredError = errors.New(http.StatusText(http.StatusLengthRequired))
 425 var LengthMismatchError = errors.New("Locator size hint does not match Content-Length header")
 426
 427 func (h *proxyHandler) Put(resp http.ResponseWriter, req *http.Request) {
 428         if err := h.checkLoop(resp, req); err != nil {
 429                 return
 430         }
 431         SetCorsHeaders(resp)
 432         resp.Header().Set("Via", "HTTP/1.1 "+viaAlias)
 433
 434         kc := h.makeKeepClient(req)
 435
 436         var err error
 437         var expectLength int64
 438         var status = http.StatusInternalServerError
 439         var wroteReplicas int
 440         var locatorOut string = "-"
 441
 442         defer func() {
 443                 log.Println(GetRemoteAddress(req), req.Method, req.URL.Path, status, expectLength, kc.Want_replicas, wroteReplicas, locatorOut, err)
 444                 if status != http.StatusOK {
 445                         http.Error(resp, err.Error(), status)
 446                 }
 447         }()
 448
 449         locatorIn := mux.Vars(req)["locator"]
 450
 451         _, err = fmt.Sscanf(req.Header.Get("Content-Length"), "%d", &expectLength)
 452         if err != nil || expectLength < 0 {
 453                 err = LengthRequiredError
 454                 status = http.StatusLengthRequired
 455                 return
 456         }
 457
 458         if locatorIn != "" {
 459                 var loc *keepclient.Locator
 460                 if loc, err = keepclient.MakeLocator(locatorIn); err != nil {
 461                         status = http.StatusBadRequest
 462                         return
 463                 } else if loc.Size > 0 && int64(loc.Size) != expectLength {
 464                         err = LengthMismatchError
 465                         status = http.StatusBadRequest
 466                         return
 467                 }
 468         }
 469
 470         var pass bool
 471         var tok string
 472         if pass, tok = CheckAuthorizationHeader(kc, h.ApiTokenCache, req); !pass {
 473                 err = BadAuthorizationHeader
 474                 status = http.StatusForbidden
 475                 return
 476         }
 477
 478         // Copy ArvadosClient struct and use the client's API token
 479         arvclient := *kc.Arvados
 480         arvclient.ApiToken = tok
 481         kc.Arvados = &arvclient
 482
 483         // Check if the client specified the number of replicas
 484         if req.Header.Get("X-Keep-Desired-Replicas") != "" {
 485                 var r int
 486                 _, err := fmt.Sscanf(req.Header.Get(keepclient.X_Keep_Desired_Replicas), "%d", &r)
 487                 if err == nil {
 488                         kc.Want_replicas = r
 489                 }
 490         }
 491
 492         // Now try to put the block through
 493         if locatorIn == "" {
 494                 if bytes, err := ioutil.ReadAll(req.Body); err != nil {
 495                         err = errors.New(fmt.Sprintf("Error reading request body: %s", err))
 496                         status = http.StatusInternalServerError
 497                         return
 498                 } else {
 499                         locatorOut, wroteReplicas, err = kc.PutB(bytes)
 500                 }
 501         } else {
 502                 locatorOut, wroteReplicas, err = kc.PutHR(locatorIn, req.Body, expectLength)
 503         }
 504
 505         // Tell the client how many successful PUTs we accomplished
 506         resp.Header().Set(keepclient.X_Keep_Replicas_Stored, fmt.Sprintf("%d", wroteReplicas))
 507
 508         switch err.(type) {
 509         case nil:
 510                 status = http.StatusOK
 511                 _, err = io.WriteString(resp, locatorOut)
 512
 513         case keepclient.OversizeBlockError:
 514                 // Too much data
 515                 status = http.StatusRequestEntityTooLarge
 516
 517         case keepclient.InsufficientReplicasError:
 518                 if wroteReplicas > 0 {
 519                         // At least one write is considered success.  The
 520                         // client can decide if getting less than the number of
 521                         // replications it asked for is a fatal error.
 522                         status = http.StatusOK
 523                         _, err = io.WriteString(resp, locatorOut)
 524                 } else {
 525                         status = http.StatusServiceUnavailable
 526                 }
 527
 528         default:
 529                 status = http.StatusBadGateway
 530         }
 531 }
 532
 533 // ServeHTTP implementation for IndexHandler
 534 // Supports only GET requests for /index/{prefix:[0-9a-f]{0,32}}
 535 // For each keep server found in LocalRoots:
 536 //   Invokes GetIndex using keepclient
 537 //   Expects "complete" response (terminating with blank new line)
 538 //   Aborts on any errors
 539 // Concatenates responses from all those keep servers and returns
 540 func (h *proxyHandler) Index(resp http.ResponseWriter, req *http.Request) {
 541         SetCorsHeaders(resp)
 542
 543         prefix := mux.Vars(req)["prefix"]
 544         var err error
 545         var status int
 546
 547         defer func() {
 548                 if status != http.StatusOK {
 549                         http.Error(resp, err.Error(), status)
 550                 }
 551         }()
 552
 553         kc := h.makeKeepClient(req)
 554         ok, token := CheckAuthorizationHeader(kc, h.ApiTokenCache, req)
 555         if !ok {
 556                 status, err = http.StatusForbidden, BadAuthorizationHeader
 557                 return
 558         }
 559
 560         // Copy ArvadosClient struct and use the client's API token
 561         arvclient := *kc.Arvados
 562         arvclient.ApiToken = token
 563         kc.Arvados = &arvclient
 564
 565         // Only GET method is supported
 566         if req.Method != "GET" {
 567                 status, err = http.StatusNotImplemented, MethodNotSupported
 568                 return
 569         }
 570
 571         // Get index from all LocalRoots and write to resp
 572         var reader io.Reader
 573         for uuid := range kc.LocalRoots() {
 574                 reader, err = kc.GetIndex(uuid, prefix)
 575                 if err != nil {
 576                         status = http.StatusBadGateway
 577                         return
 578                 }
 579
 580                 _, err = io.Copy(resp, reader)
 581                 if err != nil {
 582                         status = http.StatusBadGateway
 583                         return
 584                 }
 585         }
 586
 587         // Got index from all the keep servers and wrote to resp
 588         status = http.StatusOK
 589         resp.Write([]byte("\n"))
 590 }
 591
 592 func (h *proxyHandler) makeKeepClient(req *http.Request) *keepclient.KeepClient {
 593         kc := *h.KeepClient
 594         kc.HTTPClient = &proxyClient{
 595                 client: &http.Client{
 596                         Timeout:   h.timeout,
 597                         Transport: h.transport,
 598                 },
 599                 proto: req.Proto,
 600         }
 601         return &kc
 602 }