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 var filters [][]interface{}
159 err := json.Unmarshal([]byte(params["filters"][0]), &filters)
161 httpserver.Error(w, err.Error(), http.StatusBadRequest)
165 // Split the list of uuids by prefix
166 queryClusters := make(map[string][]string)
167 if len(filters) == 1 && len(filters[0]) == 3 {
169 lhs := f1[0].(string)
171 op, ok := f1[1].(string)
177 rhs, ok := f1[2].([]interface{})
179 for _, i := range rhs {
182 queryClusters[u[0:5]] = append(queryClusters[u[0:5]], u)
185 } else if op == "=" {
186 u, ok := f1[2].(string)
189 queryClusters[u[0:5]] = append(queryClusters[u[0:5]], u)
196 if len(queryClusters) <= 1 {
197 // Didn't find ["uuid", "in", ...] filters for multiple clusters
201 if !(len(params["count"]) == 1 && (params["count"][0] == `none` ||
202 params["count"][0] == `"none"`)) {
203 httpserver.Error(w, "Federated multi-object query must have count=none", http.StatusBadRequest)
207 wg := sync.WaitGroup{}
209 // use channel as a semaphore to limit it to 4
210 // parallel requests at a time
211 sem := make(chan bool, 4)
213 req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
215 rc := multiClusterQueryResponseCollector{}
216 for k, v := range queryClusters {
217 // blocks until it can put a value into the
218 // channel (which has a max queue capacity)
221 go func(k string, v []string) {
226 var remoteReq http.Request
227 remoteReq.Header = req.Header
228 remoteReq.Method = "POST"
229 remoteReq.URL = &url.URL{Path: req.URL.Path}
230 remoteParams := make(url.Values)
231 remoteParams["_method"] = []string{"GET"}
232 remoteParams["count"] = []string{"none"}
233 content, err := json.Marshal(v)
236 defer rc.mtx.Unlock()
237 rc.errors = append(rc.errors, err)
240 remoteParams["filters"] = []string{fmt.Sprintf(`[["uuid", "in", %s]]`, content)}
241 enc := remoteParams.Encode()
242 remoteReq.Body = ioutil.NopCloser(bytes.NewBufferString(enc))
244 if k == h.handler.Cluster.ClusterID {
245 h.handler.localClusterRequest(w, &remoteReq,
248 h.handler.remoteClusterRequest(k, w, &remoteReq,
255 if len(rc.errors) > 0 {
258 for _, e := range rc.errors {
259 strerr = append(strerr, e.Error())
261 httpserver.Errors(w, strerr, http.StatusBadGateway)
263 w.Header().Set("Content-Type", "application/json")
264 w.WriteHeader(http.StatusOK)
265 itemList := make(map[string]interface{})
266 itemList["items"] = rc.responses
267 itemList["kind"] = rc.kind
268 json.NewEncoder(w).Encode(itemList)
274 func (h *genericFederatedRequestHandler) ServeHTTP(w http.ResponseWriter, req *http.Request) {
275 m := h.matcher.FindStringSubmatch(req.URL.Path)
278 if len(m) > 0 && m[2] != "" {
282 // First, parse the query portion of the URL.
283 var params url.Values
285 if params, err = url.ParseQuery(req.URL.RawQuery); err != nil {
286 httpserver.Error(w, err.Error(), http.StatusBadRequest)
290 // Next, if appropriate, merge in parameters from the form POST body.
291 if req.Method == "POST" && req.Header.Get("Content-Type") == "application/x-www-form-urlencoded" {
292 if err = loadParamsFromForm(req, params); err != nil {
293 httpserver.Error(w, err.Error(), http.StatusBadRequest)
298 // Check if the parameters have an explicit cluster_id
299 if len(params["cluster_id"]) == 1 {
300 clusterId = params["cluster_id"][0]
303 // Handle the POST-as-GET special case (workaround for large
304 // GET requests that potentially exceed maximum URL length,
305 // like multi-object queries where the filter has 100s of
307 effectiveMethod := req.Method
308 if req.Method == "POST" && len(params["_method"]) == 1 {
309 effectiveMethod = params["_method"][0]
312 if effectiveMethod == "GET" && clusterId == "" && len(params["filters"]) == 1 {
313 if h.handleMultiClusterQuery(w, req, params, &clusterId) {
318 if clusterId == "" || clusterId == h.handler.Cluster.ClusterID {
319 h.next.ServeHTTP(w, req)
321 h.handler.remoteClusterRequest(clusterId, w, req, nil)
325 type rewriteSignaturesClusterId struct {
330 func (rw rewriteSignaturesClusterId) rewriteSignatures(resp *http.Response, requestError error) (newResponse *http.Response, err error) {
331 if requestError != nil {
332 return resp, requestError
335 if resp.StatusCode != 200 {
339 originalBody := resp.Body
340 defer originalBody.Close()
342 var col arvados.Collection
343 err = json.NewDecoder(resp.Body).Decode(&col)
348 // rewriting signatures will make manifest text 5-10% bigger so calculate
349 // capacity accordingly
350 updatedManifest := bytes.NewBuffer(make([]byte, 0, int(float64(len(col.ManifestText))*1.1)))
353 mw := io.MultiWriter(hasher, updatedManifest)
356 scanner := bufio.NewScanner(strings.NewReader(col.ManifestText))
357 scanner.Buffer(make([]byte, 1048576), len(col.ManifestText))
359 line := scanner.Text()
360 tokens := strings.Split(line, " ")
362 return nil, fmt.Errorf("Invalid stream (<3 tokens): %q", line)
365 n, err := mw.Write([]byte(tokens[0]))
367 return nil, fmt.Errorf("Error updating manifest: %v", err)
370 for _, token := range tokens[1:] {
371 n, err = mw.Write([]byte(" "))
373 return nil, fmt.Errorf("Error updating manifest: %v", err)
377 m := keepclient.SignedLocatorRe.FindStringSubmatch(token)
379 // Rewrite the block signature to be a remote signature
380 _, err = fmt.Fprintf(updatedManifest, "%s%s%s+R%s-%s%s", m[1], m[2], m[3], rw.clusterID, m[5][2:], m[8])
382 return nil, fmt.Errorf("Error updating manifest: %v", err)
385 // for hash checking, ignore signatures
386 n, err = fmt.Fprintf(hasher, "%s%s", m[1], m[2])
388 return nil, fmt.Errorf("Error updating manifest: %v", err)
392 n, err = mw.Write([]byte(token))
394 return nil, fmt.Errorf("Error updating manifest: %v", err)
399 n, err = mw.Write([]byte("\n"))
401 return nil, fmt.Errorf("Error updating manifest: %v", err)
406 // Check that expected hash is consistent with
407 // portable_data_hash field of the returned record
408 if rw.expectHash == "" {
409 rw.expectHash = col.PortableDataHash
410 } else if rw.expectHash != col.PortableDataHash {
411 return nil, fmt.Errorf("portable_data_hash %q on returned record did not match expected hash %q ", rw.expectHash, col.PortableDataHash)
414 // Certify that the computed hash of the manifest_text matches our expectation
415 sum := hasher.Sum(nil)
416 computedHash := fmt.Sprintf("%x+%v", sum, sz)
417 if computedHash != rw.expectHash {
418 return nil, fmt.Errorf("Computed manifest_text hash %q did not match expected hash %q", computedHash, rw.expectHash)
421 col.ManifestText = updatedManifest.String()
423 newbody, err := json.Marshal(col)
428 buf := bytes.NewBuffer(newbody)
429 resp.Body = ioutil.NopCloser(buf)
430 resp.ContentLength = int64(buf.Len())
431 resp.Header.Set("Content-Length", fmt.Sprintf("%v", buf.Len()))
436 func filterLocalClusterResponse(resp *http.Response, requestError error) (newResponse *http.Response, err error) {
437 if requestError != nil {
438 return resp, requestError
441 if resp.StatusCode == 404 {
442 // Suppress returning this result, because we want to
443 // search the federation.
449 type searchRemoteClusterForPDH struct {
454 sharedContext *context.Context
460 func (s *searchRemoteClusterForPDH) filterRemoteClusterResponse(resp *http.Response, requestError error) (newResponse *http.Response, err error) {
465 // Another request already returned a response
469 if requestError != nil {
470 *s.errors = append(*s.errors, fmt.Sprintf("Request error contacting %q: %v", s.remoteID, requestError))
471 // Record the error and suppress response
475 if resp.StatusCode != 200 {
476 // Suppress returning unsuccessful result. Maybe
477 // another request will find it.
478 // TODO collect and return error responses.
479 *s.errors = append(*s.errors, fmt.Sprintf("Response from %q: %v", s.remoteID, resp.Status))
480 if resp.StatusCode != 404 {
481 // Got a non-404 error response, convert into BadGateway
482 *s.statusCode = http.StatusBadGateway
489 // This reads the response body. We don't want to hold the
490 // lock while doing this because other remote requests could
491 // also have made it to this point, and we don't want a
492 // slow response holding the lock to block a faster response
493 // that is waiting on the lock.
494 newResponse, err = rewriteSignaturesClusterId{s.remoteID, s.pdh}.rewriteSignatures(resp, nil)
499 // Another request already returned a response
504 // Suppress returning unsuccessful result. Maybe
505 // another request will be successful.
506 *s.errors = append(*s.errors, fmt.Sprintf("Error parsing response from %q: %v", s.remoteID, err))
510 // We have a successful response. Suppress/cancel all the
511 // other requests/responses.
512 *s.sentResponse = true
515 return newResponse, nil
518 func (h *collectionFederatedRequestHandler) ServeHTTP(w http.ResponseWriter, req *http.Request) {
519 if req.Method != "GET" {
520 // Only handle GET requests right now
521 h.next.ServeHTTP(w, req)
525 m := collectionByPDHRe.FindStringSubmatch(req.URL.Path)
527 // Not a collection PDH GET request
528 m = collectionRe.FindStringSubmatch(req.URL.Path)
535 if clusterId != "" && clusterId != h.handler.Cluster.ClusterID {
536 // request for remote collection by uuid
537 h.handler.remoteClusterRequest(clusterId, w, req,
538 rewriteSignaturesClusterId{clusterId, ""}.rewriteSignatures)
541 // not a collection UUID request, or it is a request
542 // for a local UUID, either way, continue down the
544 h.next.ServeHTTP(w, req)
548 // Request for collection by PDH. Search the federation.
550 // First, query the local cluster.
551 if h.handler.localClusterRequest(w, req, filterLocalClusterResponse) {
555 sharedContext, cancelFunc := context.WithCancel(req.Context())
557 req = req.WithContext(sharedContext)
559 // Create a goroutine for each cluster in the
560 // RemoteClusters map. The first valid result gets
561 // returned to the client. When that happens, all
562 // other outstanding requests are cancelled or
564 sentResponse := false
566 wg := sync.WaitGroup{}
568 var errorCode int = 404
570 // use channel as a semaphore to limit it to 4
571 // parallel requests at a time
572 sem := make(chan bool, 4)
574 for remoteID := range h.handler.Cluster.RemoteClusters {
575 // blocks until it can put a value into the
576 // channel (which has a max queue capacity)
581 search := &searchRemoteClusterForPDH{m[1], remoteID, &mtx, &sentResponse,
582 &sharedContext, cancelFunc, &errors, &errorCode}
585 h.handler.remoteClusterRequest(search.remoteID, w, req, search.filterRemoteClusterResponse)
596 // No successful responses, so return the error
597 httpserver.Errors(w, errors, errorCode)
600 func (h *Handler) setupProxyRemoteCluster(next http.Handler) http.Handler {
601 mux := http.NewServeMux()
602 mux.Handle("/arvados/v1/workflows", &genericFederatedRequestHandler{next, h, wfRe})
603 mux.Handle("/arvados/v1/workflows/", &genericFederatedRequestHandler{next, h, wfRe})
604 mux.Handle("/arvados/v1/containers", &genericFederatedRequestHandler{next, h, containersRe})
605 mux.Handle("/arvados/v1/containers/", &genericFederatedRequestHandler{next, h, containersRe})
606 mux.Handle("/arvados/v1/container_requests", &genericFederatedRequestHandler{next, h, containerRequestsRe})
607 mux.Handle("/arvados/v1/container_requests/", &genericFederatedRequestHandler{next, h, containerRequestsRe})
608 mux.Handle("/arvados/v1/collections", next)
609 mux.Handle("/arvados/v1/collections/", &collectionFederatedRequestHandler{next, h})
610 mux.Handle("/", next)
612 return http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
613 parts := strings.Split(req.Header.Get("Authorization"), "/")
614 alreadySalted := (len(parts) == 3 && parts[0] == "Bearer v2" && len(parts[2]) == 40)
617 strings.Index(req.Header.Get("Via"), "arvados-controller") != -1 {
618 // The token is already salted, or this is a
619 // request from another instance of
620 // arvados-controller. In either case, we
621 // don't want to proxy this query, so just
622 // continue down the instance handler stack.
623 next.ServeHTTP(w, req)
627 mux.ServeHTTP(w, req)
633 type CurrentUser struct {
634 Authorization arvados.APIClientAuthorization
638 func (h *Handler) validateAPItoken(req *http.Request, user *CurrentUser) error {
643 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)
646 // Extract the auth token supplied in req, and replace it with a
647 // salted token for the remote cluster.
648 func (h *Handler) saltAuthToken(req *http.Request, remote string) error {
649 creds := auth.NewCredentials()
650 creds.LoadTokensFromHTTPRequest(req)
651 if len(creds.Tokens) == 0 && req.Header.Get("Content-Type") == "application/x-www-form-encoded" {
652 // Override ParseForm's 10MiB limit by ensuring
653 // req.Body is a *http.maxBytesReader.
654 req.Body = http.MaxBytesReader(nil, req.Body, 1<<28) // 256MiB. TODO: use MaxRequestSize from discovery doc or config.
655 if err := creds.LoadTokensFromHTTPRequestBody(req); err != nil {
658 // Replace req.Body with a buffer that re-encodes the
659 // form without api_token, in case we end up
660 // forwarding the request.
661 if req.PostForm != nil {
662 req.PostForm.Del("api_token")
664 req.Body = ioutil.NopCloser(bytes.NewBufferString(req.PostForm.Encode()))
666 if len(creds.Tokens) == 0 {
669 token, err := auth.SaltToken(creds.Tokens[0], remote)
670 if err == auth.ErrObsoleteToken {
671 // If the token exists in our own database, salt it
672 // for the remote. Otherwise, assume it was issued by
673 // the remote, and pass it through unmodified.
674 currentUser := CurrentUser{Authorization: arvados.APIClientAuthorization{APIToken: creds.Tokens[0]}}
675 err = h.validateAPItoken(req, ¤tUser)
676 if err == sql.ErrNoRows {
677 // Not ours; pass through unmodified.
678 token = currentUser.Authorization.APIToken
679 } else if err != nil {
682 // Found; make V2 version and salt it.
683 token, err = auth.SaltToken(currentUser.Authorization.TokenV2(), remote)
688 } else if err != nil {
691 req.Header.Set("Authorization", "Bearer "+token)
693 // Remove api_token=... from the the query string, in case we
694 // end up forwarding the request.
695 if values, err := url.ParseQuery(req.URL.RawQuery); err != nil {
697 } else if _, ok := values["api_token"]; ok {
698 delete(values, "api_token")
699 req.URL.RawQuery = values.Encode()