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 {
120 responses []interface{}
126 func (c *multiClusterQueryResponseCollector) collectResponse(resp *http.Response,
127 requestError error) (newResponse *http.Response, err error) {
128 if requestError != nil {
129 c.error = requestError
133 defer resp.Body.Close()
134 loadInto := make(map[string]interface{})
135 err = json.NewDecoder(resp.Body).Decode(&loadInto)
138 if resp.StatusCode != http.StatusOK {
139 c.error = fmt.Errorf("error fetching from %v (%v): %v", c.clusterID, resp.Status, loadInto["errors"])
141 c.responses = loadInto["items"].([]interface{})
142 c.kind, _ = loadInto["kind"].(string)
145 c.error = fmt.Errorf("error fetching from %v (%v): %v", c.clusterID, resp.Status, err)
151 func (h *genericFederatedRequestHandler) remoteQueryUUIDs(w http.ResponseWriter,
152 req *http.Request, params url.Values,
153 clusterID string, uuids []string) (rp []interface{}, kind string, err error) {
155 found := make(map[string]bool)
157 var remoteReq http.Request
158 remoteReq.Header = req.Header
159 remoteReq.Method = "POST"
160 remoteReq.URL = &url.URL{Path: req.URL.Path}
161 remoteParams := make(url.Values)
162 remoteParams["_method"] = []string{"GET"}
163 remoteParams["count"] = []string{"none"}
164 if len(params["select"]) != 0 {
165 remoteParams["select"] = params["select"]
167 content, err := json.Marshal(uuids)
171 remoteParams["filters"] = []string{fmt.Sprintf(`[["uuid", "in", %s]]`, content)}
172 enc := remoteParams.Encode()
173 remoteReq.Body = ioutil.NopCloser(bytes.NewBufferString(enc))
175 rc := multiClusterQueryResponseCollector{clusterID: clusterID}
177 if clusterID == h.handler.Cluster.ClusterID {
178 h.handler.localClusterRequest(w, &remoteReq,
181 h.handler.remoteClusterRequest(clusterID, w, &remoteReq,
185 return nil, "", rc.error
190 if len(rc.responses) == 0 {
191 // We got zero responses, no point in doing
196 rp = append(rp, rc.responses...)
198 // Go through the responses and determine what was
199 // returned. If there are remaining items, loop
200 // around and do another request with just the
202 for _, i := range rc.responses {
203 m, ok := i.(map[string]interface{})
205 uuid, ok := m["uuid"].(string)
213 for _, u := range uuids {
224 func (h *genericFederatedRequestHandler) handleMultiClusterQuery(w http.ResponseWriter, req *http.Request,
225 params url.Values, clusterId *string) bool {
227 var filters [][]interface{}
228 err := json.Unmarshal([]byte(params["filters"][0]), &filters)
230 httpserver.Error(w, err.Error(), http.StatusBadRequest)
234 // Split the list of uuids by prefix
235 queryClusters := make(map[string][]string)
237 for _, f1 := range filters {
241 lhs, ok := f1[0].(string)
242 if ok && lhs == "uuid" {
243 op, ok := f1[1].(string)
249 rhs, ok := f1[2].([]interface{})
251 for _, i := range rhs {
254 queryClusters[u[0:5]] = append(queryClusters[u[0:5]], u)
256 expectCount += len(rhs)
258 } else if op == "=" {
259 u, ok := f1[2].(string)
262 queryClusters[u[0:5]] = append(queryClusters[u[0:5]], u)
273 if len(queryClusters) <= 1 {
274 // Query does not search for uuids across multiple
280 if !(len(params["count"]) == 1 && (params["count"][0] == `none` ||
281 params["count"][0] == `"none"`)) {
282 httpserver.Error(w, "Federated multi-object query must have 'count=none'", http.StatusBadRequest)
285 if len(params["limit"]) != 0 || len(params["offset"]) != 0 || len(params["order"]) != 0 {
286 httpserver.Error(w, "Federated multi-object may not provide 'limit', 'offset' or 'order'.", http.StatusBadRequest)
289 if expectCount > h.handler.Cluster.MaxItemsPerResponse {
290 httpserver.Error(w, fmt.Sprintf("Federated multi-object request for %v objects which is more than max page size %v.",
291 expectCount, h.handler.Cluster.MaxItemsPerResponse), http.StatusBadRequest)
294 if len(params["select"]) == 1 {
296 var selects []interface{}
297 err := json.Unmarshal([]byte(params["select"][0]), &selects)
299 httpserver.Error(w, err.Error(), http.StatusBadRequest)
303 for _, r := range selects {
304 if r.(string) == "uuid" {
310 httpserver.Error(w, "Federated multi-object request must include 'uuid' in 'select'", http.StatusBadRequest)
315 // Perform parallel requests to each cluster
317 // use channel as a semaphore to limit the number of parallel
318 // requests at a time
319 sem := make(chan bool, h.handler.Cluster.ParallelRemoteRequests)
321 wg := sync.WaitGroup{}
323 req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
326 var completeResponses []interface{}
329 for k, v := range queryClusters {
335 // blocks until it can put a value into the
336 // channel (which has a max queue capacity)
339 go func(k string, v []string) {
340 rp, kn, err := h.remoteQueryUUIDs(w, req, params, k, v)
343 completeResponses = append(completeResponses, rp...)
346 errors = append(errors, err)
357 for _, e := range errors {
358 strerr = append(strerr, e.Error())
360 httpserver.Errors(w, strerr, http.StatusBadGateway)
364 w.Header().Set("Content-Type", "application/json")
365 w.WriteHeader(http.StatusOK)
366 itemList := make(map[string]interface{})
367 itemList["items"] = completeResponses
368 itemList["kind"] = kind
369 json.NewEncoder(w).Encode(itemList)
374 func (h *genericFederatedRequestHandler) ServeHTTP(w http.ResponseWriter, req *http.Request) {
375 m := h.matcher.FindStringSubmatch(req.URL.Path)
378 if len(m) > 0 && m[2] != "" {
382 // First, parse the query portion of the URL.
383 var params url.Values
385 if params, err = url.ParseQuery(req.URL.RawQuery); err != nil {
386 httpserver.Error(w, err.Error(), http.StatusBadRequest)
390 // Next, if appropriate, merge in parameters from the form POST body.
391 if req.Method == "POST" && req.Header.Get("Content-Type") == "application/x-www-form-urlencoded" {
392 if err = loadParamsFromForm(req, params); err != nil {
393 httpserver.Error(w, err.Error(), http.StatusBadRequest)
398 // Check if the parameters have an explicit cluster_id
399 if len(params["cluster_id"]) == 1 {
400 clusterId = params["cluster_id"][0]
403 // Handle the POST-as-GET special case (workaround for large
404 // GET requests that potentially exceed maximum URL length,
405 // like multi-object queries where the filter has 100s of
407 effectiveMethod := req.Method
408 if req.Method == "POST" && len(params["_method"]) == 1 {
409 effectiveMethod = params["_method"][0]
412 if effectiveMethod == "GET" && clusterId == "" && len(params["filters"]) == 1 {
413 if h.handleMultiClusterQuery(w, req, params, &clusterId) {
418 if clusterId == "" || clusterId == h.handler.Cluster.ClusterID {
419 h.next.ServeHTTP(w, req)
421 h.handler.remoteClusterRequest(clusterId, w, req, nil)
425 type rewriteSignaturesClusterId struct {
430 func (rw rewriteSignaturesClusterId) rewriteSignatures(resp *http.Response, requestError error) (newResponse *http.Response, err error) {
431 if requestError != nil {
432 return resp, requestError
435 if resp.StatusCode != 200 {
439 originalBody := resp.Body
440 defer originalBody.Close()
442 var col arvados.Collection
443 err = json.NewDecoder(resp.Body).Decode(&col)
448 // rewriting signatures will make manifest text 5-10% bigger so calculate
449 // capacity accordingly
450 updatedManifest := bytes.NewBuffer(make([]byte, 0, int(float64(len(col.ManifestText))*1.1)))
453 mw := io.MultiWriter(hasher, updatedManifest)
456 scanner := bufio.NewScanner(strings.NewReader(col.ManifestText))
457 scanner.Buffer(make([]byte, 1048576), len(col.ManifestText))
459 line := scanner.Text()
460 tokens := strings.Split(line, " ")
462 return nil, fmt.Errorf("Invalid stream (<3 tokens): %q", line)
465 n, err := mw.Write([]byte(tokens[0]))
467 return nil, fmt.Errorf("Error updating manifest: %v", err)
470 for _, token := range tokens[1:] {
471 n, err = mw.Write([]byte(" "))
473 return nil, fmt.Errorf("Error updating manifest: %v", err)
477 m := keepclient.SignedLocatorRe.FindStringSubmatch(token)
479 // Rewrite the block signature to be a remote signature
480 _, err = fmt.Fprintf(updatedManifest, "%s%s%s+R%s-%s%s", m[1], m[2], m[3], rw.clusterID, m[5][2:], m[8])
482 return nil, fmt.Errorf("Error updating manifest: %v", err)
485 // for hash checking, ignore signatures
486 n, err = fmt.Fprintf(hasher, "%s%s", m[1], m[2])
488 return nil, fmt.Errorf("Error updating manifest: %v", err)
492 n, err = mw.Write([]byte(token))
494 return nil, fmt.Errorf("Error updating manifest: %v", err)
499 n, err = mw.Write([]byte("\n"))
501 return nil, fmt.Errorf("Error updating manifest: %v", err)
506 // Check that expected hash is consistent with
507 // portable_data_hash field of the returned record
508 if rw.expectHash == "" {
509 rw.expectHash = col.PortableDataHash
510 } else if rw.expectHash != col.PortableDataHash {
511 return nil, fmt.Errorf("portable_data_hash %q on returned record did not match expected hash %q ", rw.expectHash, col.PortableDataHash)
514 // Certify that the computed hash of the manifest_text matches our expectation
515 sum := hasher.Sum(nil)
516 computedHash := fmt.Sprintf("%x+%v", sum, sz)
517 if computedHash != rw.expectHash {
518 return nil, fmt.Errorf("Computed manifest_text hash %q did not match expected hash %q", computedHash, rw.expectHash)
521 col.ManifestText = updatedManifest.String()
523 newbody, err := json.Marshal(col)
528 buf := bytes.NewBuffer(newbody)
529 resp.Body = ioutil.NopCloser(buf)
530 resp.ContentLength = int64(buf.Len())
531 resp.Header.Set("Content-Length", fmt.Sprintf("%v", buf.Len()))
536 func filterLocalClusterResponse(resp *http.Response, requestError error) (newResponse *http.Response, err error) {
537 if requestError != nil {
538 return resp, requestError
541 if resp.StatusCode == 404 {
542 // Suppress returning this result, because we want to
543 // search the federation.
549 type searchRemoteClusterForPDH struct {
554 sharedContext *context.Context
560 func (s *searchRemoteClusterForPDH) filterRemoteClusterResponse(resp *http.Response, requestError error) (newResponse *http.Response, err error) {
565 // Another request already returned a response
569 if requestError != nil {
570 *s.errors = append(*s.errors, fmt.Sprintf("Request error contacting %q: %v", s.remoteID, requestError))
571 // Record the error and suppress response
575 if resp.StatusCode != 200 {
576 // Suppress returning unsuccessful result. Maybe
577 // another request will find it.
578 // TODO collect and return error responses.
579 *s.errors = append(*s.errors, fmt.Sprintf("Response from %q: %v", s.remoteID, resp.Status))
580 if resp.StatusCode != 404 {
581 // Got a non-404 error response, convert into BadGateway
582 *s.statusCode = http.StatusBadGateway
589 // This reads the response body. We don't want to hold the
590 // lock while doing this because other remote requests could
591 // also have made it to this point, and we don't want a
592 // slow response holding the lock to block a faster response
593 // that is waiting on the lock.
594 newResponse, err = rewriteSignaturesClusterId{s.remoteID, s.pdh}.rewriteSignatures(resp, nil)
599 // Another request already returned a response
604 // Suppress returning unsuccessful result. Maybe
605 // another request will be successful.
606 *s.errors = append(*s.errors, fmt.Sprintf("Error parsing response from %q: %v", s.remoteID, err))
610 // We have a successful response. Suppress/cancel all the
611 // other requests/responses.
612 *s.sentResponse = true
615 return newResponse, nil
618 func (h *collectionFederatedRequestHandler) ServeHTTP(w http.ResponseWriter, req *http.Request) {
619 if req.Method != "GET" {
620 // Only handle GET requests right now
621 h.next.ServeHTTP(w, req)
625 m := collectionByPDHRe.FindStringSubmatch(req.URL.Path)
627 // Not a collection PDH GET request
628 m = collectionRe.FindStringSubmatch(req.URL.Path)
635 if clusterId != "" && clusterId != h.handler.Cluster.ClusterID {
636 // request for remote collection by uuid
637 h.handler.remoteClusterRequest(clusterId, w, req,
638 rewriteSignaturesClusterId{clusterId, ""}.rewriteSignatures)
641 // not a collection UUID request, or it is a request
642 // for a local UUID, either way, continue down the
644 h.next.ServeHTTP(w, req)
648 // Request for collection by PDH. Search the federation.
650 // First, query the local cluster.
651 if h.handler.localClusterRequest(w, req, filterLocalClusterResponse) {
655 sharedContext, cancelFunc := context.WithCancel(req.Context())
657 req = req.WithContext(sharedContext)
659 // Create a goroutine for each cluster in the
660 // RemoteClusters map. The first valid result gets
661 // returned to the client. When that happens, all
662 // other outstanding requests are cancelled or
664 sentResponse := false
666 wg := sync.WaitGroup{}
668 var errorCode int = 404
670 // use channel as a semaphore to limit the number of parallel
671 // requests at a time
672 sem := make(chan bool, h.handler.Cluster.ParallelRemoteRequests)
674 for remoteID := range h.handler.Cluster.RemoteClusters {
675 // blocks until it can put a value into the
676 // channel (which has a max queue capacity)
681 search := &searchRemoteClusterForPDH{m[1], remoteID, &mtx, &sentResponse,
682 &sharedContext, cancelFunc, &errors, &errorCode}
685 h.handler.remoteClusterRequest(search.remoteID, w, req, search.filterRemoteClusterResponse)
696 // No successful responses, so return the error
697 httpserver.Errors(w, errors, errorCode)
700 func (h *Handler) setupProxyRemoteCluster(next http.Handler) http.Handler {
701 mux := http.NewServeMux()
702 mux.Handle("/arvados/v1/workflows", &genericFederatedRequestHandler{next, h, wfRe})
703 mux.Handle("/arvados/v1/workflows/", &genericFederatedRequestHandler{next, h, wfRe})
704 mux.Handle("/arvados/v1/containers", &genericFederatedRequestHandler{next, h, containersRe})
705 mux.Handle("/arvados/v1/containers/", &genericFederatedRequestHandler{next, h, containersRe})
706 mux.Handle("/arvados/v1/container_requests", &genericFederatedRequestHandler{next, h, containerRequestsRe})
707 mux.Handle("/arvados/v1/container_requests/", &genericFederatedRequestHandler{next, h, containerRequestsRe})
708 mux.Handle("/arvados/v1/collections", next)
709 mux.Handle("/arvados/v1/collections/", &collectionFederatedRequestHandler{next, h})
710 mux.Handle("/", next)
712 return http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
713 parts := strings.Split(req.Header.Get("Authorization"), "/")
714 alreadySalted := (len(parts) == 3 && parts[0] == "Bearer v2" && len(parts[2]) == 40)
717 strings.Index(req.Header.Get("Via"), "arvados-controller") != -1 {
718 // The token is already salted, or this is a
719 // request from another instance of
720 // arvados-controller. In either case, we
721 // don't want to proxy this query, so just
722 // continue down the instance handler stack.
723 next.ServeHTTP(w, req)
727 mux.ServeHTTP(w, req)
733 type CurrentUser struct {
734 Authorization arvados.APIClientAuthorization
738 func (h *Handler) validateAPItoken(req *http.Request, user *CurrentUser) error {
743 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)
746 // Extract the auth token supplied in req, and replace it with a
747 // salted token for the remote cluster.
748 func (h *Handler) saltAuthToken(req *http.Request, remote string) error {
749 creds := auth.NewCredentials()
750 creds.LoadTokensFromHTTPRequest(req)
751 if len(creds.Tokens) == 0 && req.Header.Get("Content-Type") == "application/x-www-form-encoded" {
752 // Override ParseForm's 10MiB limit by ensuring
753 // req.Body is a *http.maxBytesReader.
754 req.Body = http.MaxBytesReader(nil, req.Body, 1<<28) // 256MiB. TODO: use MaxRequestSize from discovery doc or config.
755 if err := creds.LoadTokensFromHTTPRequestBody(req); err != nil {
758 // Replace req.Body with a buffer that re-encodes the
759 // form without api_token, in case we end up
760 // forwarding the request.
761 if req.PostForm != nil {
762 req.PostForm.Del("api_token")
764 req.Body = ioutil.NopCloser(bytes.NewBufferString(req.PostForm.Encode()))
766 if len(creds.Tokens) == 0 {
769 token, err := auth.SaltToken(creds.Tokens[0], remote)
770 if err == auth.ErrObsoleteToken {
771 // If the token exists in our own database, salt it
772 // for the remote. Otherwise, assume it was issued by
773 // the remote, and pass it through unmodified.
774 currentUser := CurrentUser{Authorization: arvados.APIClientAuthorization{APIToken: creds.Tokens[0]}}
775 err = h.validateAPItoken(req, ¤tUser)
776 if err == sql.ErrNoRows {
777 // Not ours; pass through unmodified.
778 token = currentUser.Authorization.APIToken
779 } else if err != nil {
782 // Found; make V2 version and salt it.
783 token, err = auth.SaltToken(currentUser.Authorization.TokenV2(), remote)
788 } else if err != nil {
791 req.Header.Set("Authorization", "Bearer "+token)
793 // Remove api_token=... from the the query string, in case we
794 // end up forwarding the request.
795 if values, err := url.ParseQuery(req.URL.RawQuery); err != nil {
797 } else if _, ok := values["api_token"]; ok {
798 delete(values, "api_token")
799 req.URL.RawQuery = values.Encode()