1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
23 "git.curoverse.com/arvados.git/sdk/go/arvados"
24 "git.curoverse.com/arvados.git/sdk/go/auth"
25 "git.curoverse.com/arvados.git/sdk/go/httpserver"
26 "git.curoverse.com/arvados.git/sdk/go/keepclient"
29 var pathPattern = `^/arvados/v1/%s(/([0-9a-z]{5})-%s-[0-9a-z]{15})?(.*)$`
30 var wfRe = regexp.MustCompile(fmt.Sprintf(pathPattern, "workflows", "7fd4e"))
31 var containersRe = regexp.MustCompile(fmt.Sprintf(pathPattern, "containers", "dz642"))
32 var containerRequestsRe = regexp.MustCompile(fmt.Sprintf(pathPattern, "container_requests", "xvhdp"))
33 var collectionRe = regexp.MustCompile(fmt.Sprintf(pathPattern, "collections", "4zz18"))
34 var collectionByPDHRe = regexp.MustCompile(`^/arvados/v1/collections/([0-9a-fA-F]{32}\+[0-9]+)+$`)
36 type genericFederatedRequestHandler struct {
39 matcher *regexp.Regexp
42 type collectionFederatedRequestHandler struct {
47 func (h *Handler) remoteClusterRequest(remoteID string, w http.ResponseWriter, req *http.Request, filter ResponseFilter) {
48 remote, ok := h.Cluster.RemoteClusters[remoteID]
50 httpserver.Error(w, "no proxy available for cluster "+remoteID, http.StatusNotFound)
53 scheme := remote.Scheme
57 err := h.saltAuthToken(req, remoteID)
59 httpserver.Error(w, err.Error(), http.StatusBadRequest)
66 RawPath: req.URL.RawPath,
67 RawQuery: req.URL.RawQuery,
69 client := h.secureClient
71 client = h.insecureClient
73 h.proxy.Do(w, req, urlOut, client, filter)
76 // loadParamsFromForm expects a request with
77 // application/x-www-form-urlencoded body. It parses the query, adds
78 // the query parameters to "params", and replaces the request body
79 // with a buffer holding the original body contents so it can be
80 // re-read by downstream proxy steps.
81 func loadParamsFromForm(req *http.Request, params url.Values) error {
82 body, err := ioutil.ReadAll(req.Body)
86 req.Body = ioutil.NopCloser(bytes.NewBuffer(body))
88 if v2, err = url.ParseQuery(string(body)); err != nil {
91 for k, v := range v2 {
92 params[k] = append(params[k], v...)
97 // loadParamsFromForm expects a request with application/json body.
98 // It parses the body, populates "loadInto", and replaces the request
99 // body with a buffer holding the original body contents so it can be
100 // re-read by downstream proxy steps.
101 func loadParamsFromJson(req *http.Request, loadInto interface{}) error {
103 if req.ContentLength > 0 {
104 cl = req.ContentLength
106 postBody := bytes.NewBuffer(make([]byte, 0, cl))
107 defer req.Body.Close()
109 rdr := io.TeeReader(req.Body, postBody)
111 err := json.NewDecoder(rdr).Decode(loadInto)
115 req.Body = ioutil.NopCloser(postBody)
119 type multiClusterQueryResponseCollector struct {
121 responses []interface{}
126 func (c *multiClusterQueryResponseCollector) collectResponse(resp *http.Response,
127 requestError error) (newResponse *http.Response, err error) {
128 if requestError != nil {
131 c.errors = append(c.errors, requestError)
134 defer resp.Body.Close()
135 loadInto := make(map[string]interface{})
136 err = json.NewDecoder(resp.Body).Decode(&loadInto)
142 if resp.StatusCode != http.StatusOK {
143 c.errors = append(c.errors, fmt.Errorf("error %v", loadInto["errors"]))
145 c.responses = append(c.responses, loadInto["items"].([]interface{})...)
146 c.kind = loadInto["kind"].(string)
149 c.errors = append(c.errors, err)
155 func (h *genericFederatedRequestHandler) handleMultiClusterQuery(w http.ResponseWriter, req *http.Request,
156 params url.Values, clusterId *string) bool {
158 if !(len(params["count"]) == 1 && (params["count"][0] == `none` ||
159 params["count"][0] == `"none"`)) {
160 // don't federate unless params has count=none
164 var filters [][]interface{}
165 err := json.Unmarshal([]byte(params["filters"][0]), &filters)
167 httpserver.Error(w, err.Error(), http.StatusBadRequest)
171 // Split the list of uuids by prefix
172 queryClusters := make(map[string][]string)
173 if len(filters) == 1 && len(filters[0]) == 3 {
175 lhs := f1[0].(string)
177 rhs := f1[2].([]interface{})
178 if lhs == "uuid" && op == "in" {
179 for _, i := range rhs {
182 queryClusters[u[0:5]] = append(queryClusters[u[0:5]], u)
187 if len(queryClusters) == 0 {
188 // Didn't find any ["uuid", "in", ...] filters
192 wg := sync.WaitGroup{}
194 // use channel as a semaphore to limit it to 4
195 // parallel requests at a time
196 sem := make(chan bool, 4)
198 req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
200 rc := multiClusterQueryResponseCollector{}
201 for k, v := range queryClusters {
202 // blocks until it can put a value into the
203 // channel (which has a max queue capacity)
206 go func(k string, v []string) {
211 var remoteReq http.Request
212 remoteReq.Header = req.Header
213 remoteReq.Method = "POST"
214 remoteReq.URL = &url.URL{Path: req.URL.Path}
215 remoteParams := make(url.Values)
216 remoteParams["_method"] = []string{"GET"}
217 remoteParams["count"] = []string{"none"}
218 content, err := json.Marshal(v)
221 defer rc.mtx.Unlock()
222 rc.errors = append(rc.errors, err)
225 remoteParams["filters"] = []string{fmt.Sprintf(`[["uuid", "in", %s]]`, content)}
226 enc := remoteParams.Encode()
227 remoteReq.Body = ioutil.NopCloser(bytes.NewBufferString(enc))
229 if k == h.handler.Cluster.ClusterID {
230 h.handler.localClusterRequest(w, &remoteReq,
233 h.handler.remoteClusterRequest(k, w, &remoteReq,
240 if len(rc.errors) > 0 {
243 for _, e := range rc.errors {
244 strerr = append(strerr, e.Error())
246 httpserver.Errors(w, strerr, http.StatusBadGateway)
248 w.Header().Set("Content-Type", "application/json")
249 w.WriteHeader(http.StatusOK)
250 itemList := make(map[string]interface{})
251 itemList["items"] = rc.responses
252 itemList["kind"] = rc.kind
253 json.NewEncoder(w).Encode(itemList)
259 func (h *genericFederatedRequestHandler) ServeHTTP(w http.ResponseWriter, req *http.Request) {
260 m := h.matcher.FindStringSubmatch(req.URL.Path)
263 if len(m) > 0 && m[2] != "" {
267 // First, parse the query portion of the URL.
268 var params url.Values
270 if params, err = url.ParseQuery(req.URL.RawQuery); err != nil {
271 httpserver.Error(w, err.Error(), http.StatusBadRequest)
275 // Next, if appropriate, merge in parameters from the form POST body.
276 if req.Method == "POST" && req.Header.Get("Content-Type") == "application/x-www-form-urlencoded" {
277 if err = loadParamsFromForm(req, params); err != nil {
278 httpserver.Error(w, err.Error(), http.StatusBadRequest)
283 // Check if the parameters have an explicit cluster_id
284 if len(params["cluster_id"]) == 1 {
285 clusterId = params["cluster_id"][0]
288 // TODO: decide if this actually makes sense...
289 if clusterId == "" && req.Method == "POST" && req.Header.Get("Content-Type") == "application/json" {
290 var hasClusterId struct {
291 ClusterID string `json:"cluster_id"`
293 if err = loadParamsFromJson(req, &hasClusterId); err != nil {
294 httpserver.Error(w, err.Error(), http.StatusBadRequest)
297 clusterId = hasClusterId.ClusterID
300 effectiveMethod := req.Method
301 if req.Method == "POST" && len(params["_method"]) == 1 {
302 effectiveMethod = params["_method"][0]
305 if effectiveMethod == "GET" && clusterId == "" && len(params["filters"]) == 1 {
306 if h.handleMultiClusterQuery(w, req, params, &clusterId) {
311 if clusterId == "" || clusterId == h.handler.Cluster.ClusterID {
312 h.next.ServeHTTP(w, req)
314 h.handler.remoteClusterRequest(clusterId, w, req, nil)
318 type rewriteSignaturesClusterId struct {
323 func (rw rewriteSignaturesClusterId) rewriteSignatures(resp *http.Response, requestError error) (newResponse *http.Response, err error) {
324 if requestError != nil {
325 return resp, requestError
328 if resp.StatusCode != 200 {
332 originalBody := resp.Body
333 defer originalBody.Close()
335 var col arvados.Collection
336 err = json.NewDecoder(resp.Body).Decode(&col)
341 // rewriting signatures will make manifest text 5-10% bigger so calculate
342 // capacity accordingly
343 updatedManifest := bytes.NewBuffer(make([]byte, 0, int(float64(len(col.ManifestText))*1.1)))
346 mw := io.MultiWriter(hasher, updatedManifest)
349 scanner := bufio.NewScanner(strings.NewReader(col.ManifestText))
350 scanner.Buffer(make([]byte, 1048576), len(col.ManifestText))
352 line := scanner.Text()
353 tokens := strings.Split(line, " ")
355 return nil, fmt.Errorf("Invalid stream (<3 tokens): %q", line)
358 n, err := mw.Write([]byte(tokens[0]))
360 return nil, fmt.Errorf("Error updating manifest: %v", err)
363 for _, token := range tokens[1:] {
364 n, err = mw.Write([]byte(" "))
366 return nil, fmt.Errorf("Error updating manifest: %v", err)
370 m := keepclient.SignedLocatorRe.FindStringSubmatch(token)
372 // Rewrite the block signature to be a remote signature
373 _, err = fmt.Fprintf(updatedManifest, "%s%s%s+R%s-%s%s", m[1], m[2], m[3], rw.clusterID, m[5][2:], m[8])
375 return nil, fmt.Errorf("Error updating manifest: %v", err)
378 // for hash checking, ignore signatures
379 n, err = fmt.Fprintf(hasher, "%s%s", m[1], m[2])
381 return nil, fmt.Errorf("Error updating manifest: %v", err)
385 n, err = mw.Write([]byte(token))
387 return nil, fmt.Errorf("Error updating manifest: %v", err)
392 n, err = mw.Write([]byte("\n"))
394 return nil, fmt.Errorf("Error updating manifest: %v", err)
399 // Check that expected hash is consistent with
400 // portable_data_hash field of the returned record
401 if rw.expectHash == "" {
402 rw.expectHash = col.PortableDataHash
403 } else if rw.expectHash != col.PortableDataHash {
404 return nil, fmt.Errorf("portable_data_hash %q on returned record did not match expected hash %q ", rw.expectHash, col.PortableDataHash)
407 // Certify that the computed hash of the manifest_text matches our expectation
408 sum := hasher.Sum(nil)
409 computedHash := fmt.Sprintf("%x+%v", sum, sz)
410 if computedHash != rw.expectHash {
411 return nil, fmt.Errorf("Computed manifest_text hash %q did not match expected hash %q", computedHash, rw.expectHash)
414 col.ManifestText = updatedManifest.String()
416 newbody, err := json.Marshal(col)
421 buf := bytes.NewBuffer(newbody)
422 resp.Body = ioutil.NopCloser(buf)
423 resp.ContentLength = int64(buf.Len())
424 resp.Header.Set("Content-Length", fmt.Sprintf("%v", buf.Len()))
429 func filterLocalClusterResponse(resp *http.Response, requestError error) (newResponse *http.Response, err error) {
430 if requestError != nil {
431 return resp, requestError
434 if resp.StatusCode == 404 {
435 // Suppress returning this result, because we want to
436 // search the federation.
442 type searchRemoteClusterForPDH struct {
447 sharedContext *context.Context
453 func (s *searchRemoteClusterForPDH) filterRemoteClusterResponse(resp *http.Response, requestError error) (newResponse *http.Response, err error) {
458 // Another request already returned a response
462 if requestError != nil {
463 *s.errors = append(*s.errors, fmt.Sprintf("Request error contacting %q: %v", s.remoteID, requestError))
464 // Record the error and suppress response
468 if resp.StatusCode != 200 {
469 // Suppress returning unsuccessful result. Maybe
470 // another request will find it.
471 // TODO collect and return error responses.
472 *s.errors = append(*s.errors, fmt.Sprintf("Response from %q: %v", s.remoteID, resp.Status))
473 if resp.StatusCode != 404 {
474 // Got a non-404 error response, convert into BadGateway
475 *s.statusCode = http.StatusBadGateway
482 // This reads the response body. We don't want to hold the
483 // lock while doing this because other remote requests could
484 // also have made it to this point, and we don't want a
485 // slow response holding the lock to block a faster response
486 // that is waiting on the lock.
487 newResponse, err = rewriteSignaturesClusterId{s.remoteID, s.pdh}.rewriteSignatures(resp, nil)
492 // Another request already returned a response
497 // Suppress returning unsuccessful result. Maybe
498 // another request will be successful.
499 *s.errors = append(*s.errors, fmt.Sprintf("Error parsing response from %q: %v", s.remoteID, err))
503 // We have a successful response. Suppress/cancel all the
504 // other requests/responses.
505 *s.sentResponse = true
508 return newResponse, nil
511 func (h *collectionFederatedRequestHandler) ServeHTTP(w http.ResponseWriter, req *http.Request) {
512 if req.Method != "GET" {
513 // Only handle GET requests right now
514 h.next.ServeHTTP(w, req)
518 m := collectionByPDHRe.FindStringSubmatch(req.URL.Path)
520 // Not a collection PDH GET request
521 m = collectionRe.FindStringSubmatch(req.URL.Path)
528 if clusterId != "" && clusterId != h.handler.Cluster.ClusterID {
529 // request for remote collection by uuid
530 h.handler.remoteClusterRequest(clusterId, w, req,
531 rewriteSignaturesClusterId{clusterId, ""}.rewriteSignatures)
534 // not a collection UUID request, or it is a request
535 // for a local UUID, either way, continue down the
537 h.next.ServeHTTP(w, req)
541 // Request for collection by PDH. Search the federation.
543 // First, query the local cluster.
544 if h.handler.localClusterRequest(w, req, filterLocalClusterResponse) {
548 sharedContext, cancelFunc := context.WithCancel(req.Context())
550 req = req.WithContext(sharedContext)
552 // Create a goroutine for each cluster in the
553 // RemoteClusters map. The first valid result gets
554 // returned to the client. When that happens, all
555 // other outstanding requests are cancelled or
557 sentResponse := false
559 wg := sync.WaitGroup{}
561 var errorCode int = 404
563 // use channel as a semaphore to limit it to 4
564 // parallel requests at a time
565 sem := make(chan bool, 4)
567 for remoteID := range h.handler.Cluster.RemoteClusters {
568 // blocks until it can put a value into the
569 // channel (which has a max queue capacity)
574 search := &searchRemoteClusterForPDH{m[1], remoteID, &mtx, &sentResponse,
575 &sharedContext, cancelFunc, &errors, &errorCode}
578 h.handler.remoteClusterRequest(search.remoteID, w, req, search.filterRemoteClusterResponse)
589 // No successful responses, so return the error
590 httpserver.Errors(w, errors, errorCode)
593 func (h *Handler) setupProxyRemoteCluster(next http.Handler) http.Handler {
594 mux := http.NewServeMux()
595 mux.Handle("/arvados/v1/workflows", &genericFederatedRequestHandler{next, h, wfRe})
596 mux.Handle("/arvados/v1/workflows/", &genericFederatedRequestHandler{next, h, wfRe})
597 mux.Handle("/arvados/v1/containers", &genericFederatedRequestHandler{next, h, containersRe})
598 mux.Handle("/arvados/v1/containers/", &genericFederatedRequestHandler{next, h, containersRe})
599 mux.Handle("/arvados/v1/container_requests", &genericFederatedRequestHandler{next, h, containerRequestsRe})
600 mux.Handle("/arvados/v1/container_requests/", &genericFederatedRequestHandler{next, h, containerRequestsRe})
601 mux.Handle("/arvados/v1/collections", next)
602 mux.Handle("/arvados/v1/collections/", &collectionFederatedRequestHandler{next, h})
603 mux.Handle("/", next)
605 return http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
606 parts := strings.Split(req.Header.Get("Authorization"), "/")
607 alreadySalted := (len(parts) == 3 && parts[0] == "Bearer v2" && len(parts[2]) == 40)
610 strings.Index(req.Header.Get("Via"), "arvados-controller") != -1 {
611 // The token is already salted, or this is a
612 // request from another instance of
613 // arvados-controller. In either case, we
614 // don't want to proxy this query, so just
615 // continue down the instance handler stack.
616 next.ServeHTTP(w, req)
620 mux.ServeHTTP(w, req)
626 type CurrentUser struct {
627 Authorization arvados.APIClientAuthorization
631 func (h *Handler) validateAPItoken(req *http.Request, user *CurrentUser) error {
636 return db.QueryRowContext(req.Context(), `SELECT api_client_authorizations.uuid, users.uuid FROM api_client_authorizations JOIN users on api_client_authorizations.user_id=users.id WHERE api_token=$1 AND (expires_at IS NULL OR expires_at > current_timestamp) LIMIT 1`, user.Authorization.APIToken).Scan(&user.Authorization.UUID, &user.UUID)
639 // Extract the auth token supplied in req, and replace it with a
640 // salted token for the remote cluster.
641 func (h *Handler) saltAuthToken(req *http.Request, remote string) error {
642 creds := auth.NewCredentials()
643 creds.LoadTokensFromHTTPRequest(req)
644 if len(creds.Tokens) == 0 && req.Header.Get("Content-Type") == "application/x-www-form-encoded" {
645 // Override ParseForm's 10MiB limit by ensuring
646 // req.Body is a *http.maxBytesReader.
647 req.Body = http.MaxBytesReader(nil, req.Body, 1<<28) // 256MiB. TODO: use MaxRequestSize from discovery doc or config.
648 if err := creds.LoadTokensFromHTTPRequestBody(req); err != nil {
651 // Replace req.Body with a buffer that re-encodes the
652 // form without api_token, in case we end up
653 // forwarding the request.
654 if req.PostForm != nil {
655 req.PostForm.Del("api_token")
657 req.Body = ioutil.NopCloser(bytes.NewBufferString(req.PostForm.Encode()))
659 if len(creds.Tokens) == 0 {
662 token, err := auth.SaltToken(creds.Tokens[0], remote)
663 if err == auth.ErrObsoleteToken {
664 // If the token exists in our own database, salt it
665 // for the remote. Otherwise, assume it was issued by
666 // the remote, and pass it through unmodified.
667 currentUser := CurrentUser{Authorization: arvados.APIClientAuthorization{APIToken: creds.Tokens[0]}}
668 err = h.validateAPItoken(req, ¤tUser)
669 if err == sql.ErrNoRows {
670 // Not ours; pass through unmodified.
671 token = currentUser.Authorization.APIToken
672 } else if err != nil {
675 // Found; make V2 version and salt it.
676 token, err = auth.SaltToken(currentUser.Authorization.TokenV2(), remote)
681 } else if err != nil {
684 req.Header.Set("Authorization", "Bearer "+token)
686 // Remove api_token=... from the the query string, in case we
687 // end up forwarding the request.
688 if values, err := url.ParseQuery(req.URL.RawQuery); err != nil {
690 } else if _, ok := values["api_token"]; ok {
691 delete(values, "api_token")
692 req.URL.RawQuery = values.Encode()