1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
22 "git.curoverse.com/arvados.git/sdk/go/arvados"
23 "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
24 "git.curoverse.com/arvados.git/sdk/go/auth"
25 "git.curoverse.com/arvados.git/sdk/go/health"
26 "git.curoverse.com/arvados.git/sdk/go/httpserver"
27 "git.curoverse.com/arvados.git/sdk/go/keepclient"
28 log "github.com/Sirupsen/logrus"
29 "golang.org/x/net/webdav"
34 MetricsAPI http.Handler
35 clientPool *arvadosclient.ClientPool
37 healthHandler http.Handler
38 webdavLS webdav.LockSystem
41 // parseCollectionIDFromDNSName returns a UUID or PDH if s begins with
42 // a UUID or URL-encoded PDH; otherwise "".
43 func parseCollectionIDFromDNSName(s string) string {
45 if i := strings.IndexRune(s, '.'); i >= 0 {
48 // Names like {uuid}--collections.example.com serve the same
49 // purpose as {uuid}.collections.example.com but can reduce
50 // cost/effort of using [additional] wildcard certificates.
51 if i := strings.Index(s, "--"); i >= 0 {
54 if arvadosclient.UUIDMatch(s) {
57 if pdh := strings.Replace(s, "-", "+", 1); arvadosclient.PDHMatch(pdh) {
63 var urlPDHDecoder = strings.NewReplacer(" ", "+", "-", "+")
65 // parseCollectionIDFromURL returns a UUID or PDH if s is a UUID or a
66 // PDH (even if it is a PDH with "+" replaced by " " or "-");
68 func parseCollectionIDFromURL(s string) string {
69 if arvadosclient.UUIDMatch(s) {
72 if pdh := urlPDHDecoder.Replace(s); arvadosclient.PDHMatch(pdh) {
78 func (h *handler) setup() {
79 h.clientPool = arvadosclient.MakeClientPool()
81 keepclient.RefreshServiceDiscoveryOnSIGHUP()
83 h.healthHandler = &health.Handler{
84 Token: h.Config.ManagementToken,
88 // Even though we don't accept LOCK requests, every webdav
89 // handler must have a non-nil LockSystem.
90 h.webdavLS = &noLockSystem{}
93 func (h *handler) serveStatus(w http.ResponseWriter, r *http.Request) {
94 json.NewEncoder(w).Encode(struct{ Version string }{version})
97 // updateOnSuccess wraps httpserver.ResponseWriter. If the handler
98 // sends an HTTP header indicating success, updateOnSuccess first
99 // calls the provided update func. If the update func fails, a 500
100 // response is sent, and the status code and body sent by the handler
101 // are ignored (all response writes return the update error).
102 type updateOnSuccess struct {
103 httpserver.ResponseWriter
109 func (uos *updateOnSuccess) Write(p []byte) (int, error) {
111 uos.WriteHeader(http.StatusOK)
116 return uos.ResponseWriter.Write(p)
119 func (uos *updateOnSuccess) WriteHeader(code int) {
121 uos.sentHeader = true
122 if code >= 200 && code < 400 {
123 if uos.err = uos.update(); uos.err != nil {
124 code := http.StatusInternalServerError
125 if err, ok := uos.err.(*arvados.TransactionError); ok {
126 code = err.StatusCode
128 log.Printf("update() changes response to HTTP %d: %T %q", code, uos.err, uos.err)
129 http.Error(uos.ResponseWriter, uos.err.Error(), code)
134 uos.ResponseWriter.WriteHeader(code)
138 writeMethod = map[string]bool{
146 webdavMethod = map[string]bool{
156 browserMethod = map[string]bool{
161 // top-level dirs to serve with siteFS
162 siteFSDir = map[string]bool{
163 "": true, // root directory
169 // ServeHTTP implements http.Handler.
170 func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
171 h.setupOnce.Do(h.setup)
174 var statusText string
176 remoteAddr := r.RemoteAddr
177 if xff := r.Header.Get("X-Forwarded-For"); xff != "" {
178 remoteAddr = xff + "," + remoteAddr
180 if xfp := r.Header.Get("X-Forwarded-Proto"); xfp != "" && xfp != "http" {
184 w := httpserver.WrapResponseWriter(wOrig)
187 statusCode = w.WroteStatus()
188 } else if w.WroteStatus() == 0 {
189 w.WriteHeader(statusCode)
190 } else if w.WroteStatus() != statusCode {
191 log.WithField("RequestID", r.Header.Get("X-Request-Id")).Warn(
192 fmt.Sprintf("Our status changed from %d to %d after we sent headers", w.WroteStatus(), statusCode))
194 if statusText == "" {
195 statusText = http.StatusText(statusCode)
199 if strings.HasPrefix(r.URL.Path, "/_health/") && r.Method == "GET" {
200 h.healthHandler.ServeHTTP(w, r)
204 if method := r.Header.Get("Access-Control-Request-Method"); method != "" && r.Method == "OPTIONS" {
205 if !browserMethod[method] && !webdavMethod[method] {
206 statusCode = http.StatusMethodNotAllowed
209 w.Header().Set("Access-Control-Allow-Headers", "Authorization, Content-Type, Range")
210 w.Header().Set("Access-Control-Allow-Methods", "COPY, DELETE, GET, MKCOL, MOVE, OPTIONS, POST, PROPFIND, PUT, RMCOL")
211 w.Header().Set("Access-Control-Allow-Origin", "*")
212 w.Header().Set("Access-Control-Max-Age", "86400")
213 statusCode = http.StatusOK
217 if !browserMethod[r.Method] && !webdavMethod[r.Method] {
218 statusCode, statusText = http.StatusMethodNotAllowed, r.Method
222 if r.Header.Get("Origin") != "" {
223 // Allow simple cross-origin requests without user
224 // credentials ("user credentials" as defined by CORS,
225 // i.e., cookies, HTTP authentication, and client-side
226 // SSL certificates. See
227 // http://www.w3.org/TR/cors/#user-credentials).
228 w.Header().Set("Access-Control-Allow-Origin", "*")
229 w.Header().Set("Access-Control-Expose-Headers", "Content-Range")
232 pathParts := strings.Split(r.URL.Path[1:], "/")
235 var collectionID string
237 var reqTokens []string
241 credentialsOK := h.Config.TrustAllContent
243 if r.Host != "" && r.Host == h.Config.AttachmentOnlyHost {
246 } else if r.FormValue("disposition") == "attachment" {
250 if collectionID = parseCollectionIDFromDNSName(r.Host); collectionID != "" {
251 // http://ID.collections.example/PATH...
253 } else if r.URL.Path == "/status.json" {
256 } else if strings.HasPrefix(r.URL.Path, "/metrics") {
257 h.MetricsAPI.ServeHTTP(w, r)
259 } else if siteFSDir[pathParts[0]] {
261 } else if len(pathParts) >= 1 && strings.HasPrefix(pathParts[0], "c=") {
263 collectionID = parseCollectionIDFromURL(pathParts[0][2:])
265 } else if len(pathParts) >= 2 && pathParts[0] == "collections" {
266 if len(pathParts) >= 4 && pathParts[1] == "download" {
267 // /collections/download/ID/TOKEN/PATH...
268 collectionID = parseCollectionIDFromURL(pathParts[2])
269 tokens = []string{pathParts[3]}
273 // /collections/ID/PATH...
274 collectionID = parseCollectionIDFromURL(pathParts[1])
275 tokens = h.Config.AnonymousTokens
280 if collectionID == "" && !useSiteFS {
281 statusCode = http.StatusNotFound
286 if cc := r.Header.Get("Cache-Control"); strings.Contains(cc, "no-cache") || strings.Contains(cc, "must-revalidate") {
290 formToken := r.FormValue("api_token")
291 if formToken != "" && r.Header.Get("Origin") != "" && attachment && r.URL.Query().Get("api_token") == "" {
292 // The client provided an explicit token in the POST
293 // body. The Origin header indicates this *might* be
294 // an AJAX request, in which case redirect-with-cookie
295 // won't work: we should just serve the content in the
296 // POST response. This is safe because:
298 // * We're supplying an attachment, not inline
299 // content, so we don't need to convert the POST to
300 // a GET and avoid the "really resubmit form?"
303 // * The token isn't embedded in the URL, so we don't
304 // need to worry about bookmarks and copy/paste.
305 tokens = append(tokens, formToken)
306 } else if formToken != "" && browserMethod[r.Method] {
307 // The client provided an explicit token in the query
308 // string, or a form in POST body. We must put the
309 // token in an HttpOnly cookie, and redirect to the
310 // same URL with the query param redacted and method =
312 h.seeOtherWithCookie(w, r, "", credentialsOK)
318 tokens = auth.NewCredentialsFromHTTPRequest(r).Tokens
320 h.serveSiteFS(w, r, tokens, credentialsOK, attachment)
324 targetPath := pathParts[stripParts:]
325 if tokens == nil && len(targetPath) > 0 && strings.HasPrefix(targetPath[0], "t=") {
326 // http://ID.example/t=TOKEN/PATH...
327 // /c=ID/t=TOKEN/PATH...
329 // This form must only be used to pass scoped tokens
330 // that give permission for a single collection. See
331 // FormValue case above.
332 tokens = []string{targetPath[0][2:]}
334 targetPath = targetPath[1:]
340 reqTokens = auth.NewCredentialsFromHTTPRequest(r).Tokens
342 tokens = append(reqTokens, h.Config.AnonymousTokens...)
345 if len(targetPath) > 0 && targetPath[0] == "_" {
346 // If a collection has a directory called "t=foo" or
347 // "_", it can be served at
348 // //collections.example/_/t=foo/ or
349 // //collections.example/_/_/ respectively:
350 // //collections.example/t=foo/ won't work because
351 // t=foo will be interpreted as a token "foo".
352 targetPath = targetPath[1:]
356 arv := h.clientPool.Get()
358 statusCode, statusText = http.StatusInternalServerError, "Pool failed: "+h.clientPool.Err().Error()
361 defer h.clientPool.Put(arv)
363 var collection *arvados.Collection
364 tokenResult := make(map[string]int)
365 for _, arv.ApiToken = range tokens {
367 collection, err = h.Config.Cache.Get(arv, collectionID, forceReload)
372 if srvErr, ok := err.(arvadosclient.APIServerError); ok {
373 switch srvErr.HttpStatusCode {
375 // Token broken or insufficient to
376 // retrieve collection
377 tokenResult[arv.ApiToken] = srvErr.HttpStatusCode
381 // Something more serious is wrong
382 statusCode, statusText = http.StatusInternalServerError, err.Error()
385 if collection == nil {
386 if pathToken || !credentialsOK {
387 // Either the URL is a "secret sharing link"
388 // that didn't work out (and asking the client
389 // for additional credentials would just be
390 // confusing), or we don't even accept
391 // credentials at this path.
392 statusCode = http.StatusNotFound
395 for _, t := range reqTokens {
396 if tokenResult[t] == 404 {
397 // The client provided valid token(s), but the
398 // collection was not found.
399 statusCode = http.StatusNotFound
403 // The client's token was invalid (e.g., expired), or
404 // the client didn't even provide one. Propagate the
405 // 401 to encourage the client to use a [different]
408 // TODO(TC): This response would be confusing to
409 // someone trying (anonymously) to download public
410 // data that has been deleted. Allow a referrer to
411 // provide this context somehow?
412 w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"")
413 statusCode = http.StatusUnauthorized
417 kc, err := keepclient.MakeKeepClient(arv)
419 statusCode, statusText = http.StatusInternalServerError, err.Error()
422 kc.RequestID = r.Header.Get("X-Request-Id")
425 if len(targetPath) > 0 {
426 basename = targetPath[len(targetPath)-1]
428 applyContentDispositionHdr(w, r, basename, attachment)
430 client := (&arvados.Client{
431 APIHost: arv.ApiServer,
432 AuthToken: arv.ApiToken,
433 Insecure: arv.ApiInsecure,
434 }).WithRequestID(r.Header.Get("X-Request-Id"))
436 fs, err := collection.FileSystem(client, kc)
438 statusCode, statusText = http.StatusInternalServerError, err.Error()
442 writefs, writeOK := fs.(arvados.CollectionFileSystem)
443 targetIsPDH := arvadosclient.PDHMatch(collectionID)
444 if (targetIsPDH || !writeOK) && writeMethod[r.Method] {
445 statusCode, statusText = http.StatusMethodNotAllowed, errReadOnly.Error()
449 if webdavMethod[r.Method] {
450 if writeMethod[r.Method] {
451 // Save the collection only if/when all
452 // webdav->filesystem operations succeed --
453 // and send a 500 error if the modified
454 // collection can't be saved.
455 w = &updateOnSuccess{
457 update: func() error {
458 return h.Config.Cache.Update(client, *collection, writefs)
462 Prefix: "/" + strings.Join(pathParts[:stripParts], "/"),
463 FileSystem: &webdavFS{
465 writing: writeMethod[r.Method],
466 alwaysReadEOF: r.Method == "PROPFIND",
468 LockSystem: h.webdavLS,
469 Logger: func(_ *http.Request, err error) {
471 log.Printf("error from webdav handler: %q", err)
479 openPath := "/" + strings.Join(targetPath, "/")
480 if f, err := fs.Open(openPath); os.IsNotExist(err) {
481 // Requested non-existent path
482 statusCode = http.StatusNotFound
483 } else if err != nil {
484 // Some other (unexpected) error
485 statusCode, statusText = http.StatusInternalServerError, err.Error()
486 } else if stat, err := f.Stat(); err != nil {
487 // Can't get Size/IsDir (shouldn't happen with a collectionFS!)
488 statusCode, statusText = http.StatusInternalServerError, err.Error()
489 } else if stat.IsDir() && !strings.HasSuffix(r.URL.Path, "/") {
490 // If client requests ".../dirname", redirect to
491 // ".../dirname/". This way, relative links in the
492 // listing for "dirname" can always be "fnm", never
494 h.seeOtherWithCookie(w, r, r.URL.Path+"/", credentialsOK)
495 } else if stat.IsDir() {
496 h.serveDirectory(w, r, collection.Name, fs, openPath, true)
498 http.ServeContent(w, r, basename, stat.ModTime(), f)
499 if r.Header.Get("Range") == "" && int64(w.WroteBodyBytes()) != stat.Size() {
500 // If we wrote fewer bytes than expected, it's
501 // too late to change the real response code
502 // or send an error message to the client, but
503 // at least we can try to put some useful
504 // debugging info in the logs.
505 n, err := f.Read(make([]byte, 1024))
506 statusCode, statusText = http.StatusInternalServerError, fmt.Sprintf("f.Size()==%d but only wrote %d bytes; read(1024) returns %d, %s", stat.Size(), w.WroteBodyBytes(), n, err)
512 func (h *handler) serveSiteFS(w http.ResponseWriter, r *http.Request, tokens []string, credentialsOK, attachment bool) {
513 if len(tokens) == 0 {
514 w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"")
515 http.Error(w, http.StatusText(http.StatusUnauthorized), http.StatusUnauthorized)
518 if writeMethod[r.Method] {
519 http.Error(w, errReadOnly.Error(), http.StatusMethodNotAllowed)
522 arv := h.clientPool.Get()
524 http.Error(w, "Pool failed: "+h.clientPool.Err().Error(), http.StatusInternalServerError)
527 defer h.clientPool.Put(arv)
528 arv.ApiToken = tokens[0]
530 kc, err := keepclient.MakeKeepClient(arv)
532 http.Error(w, err.Error(), http.StatusInternalServerError)
535 kc.RequestID = r.Header.Get("X-Request-Id")
536 client := (&arvados.Client{
537 APIHost: arv.ApiServer,
538 AuthToken: arv.ApiToken,
539 Insecure: arv.ApiInsecure,
540 }).WithRequestID(r.Header.Get("X-Request-Id"))
541 fs := client.SiteFileSystem(kc)
542 f, err := fs.Open(r.URL.Path)
543 if os.IsNotExist(err) {
544 http.Error(w, err.Error(), http.StatusNotFound)
546 } else if err != nil {
547 http.Error(w, err.Error(), http.StatusInternalServerError)
551 if fi, err := f.Stat(); err == nil && fi.IsDir() && r.Method == "GET" {
552 if !strings.HasSuffix(r.URL.Path, "/") {
553 h.seeOtherWithCookie(w, r, r.URL.Path+"/", credentialsOK)
555 h.serveDirectory(w, r, fi.Name(), fs, r.URL.Path, false)
559 if r.Method == "GET" {
560 _, basename := filepath.Split(r.URL.Path)
561 applyContentDispositionHdr(w, r, basename, attachment)
563 wh := webdav.Handler{
565 FileSystem: &webdavFS{
567 writing: writeMethod[r.Method],
568 alwaysReadEOF: r.Method == "PROPFIND",
570 LockSystem: h.webdavLS,
571 Logger: func(_ *http.Request, err error) {
573 log.Printf("error from webdav handler: %q", err)
580 var dirListingTemplate = `<!DOCTYPE HTML>
582 <META name="robots" content="NOINDEX">
583 <TITLE>{{ .CollectionName }}</TITLE>
584 <STYLE type="text/css">
589 background-color: #D9EDF7;
590 border-radius: .25em;
601 font-family: monospace;
608 <H1>{{ .CollectionName }}</H1>
610 <P>This collection of data files is being shared with you through
611 Arvados. You can download individual files listed below. To download
612 the entire directory tree with wget, try:</P>
614 <PRE>$ wget --mirror --no-parent --no-host --cut-dirs={{ .StripParts }} https://{{ .Request.Host }}{{ .Request.URL.Path }}</PRE>
616 <H2>File Listing</H2>
622 <LI>{{" " | printf "%15s " | nbsp}}<A href="{{print "./" .Name}}/">{{.Name}}/</A></LI>
624 <LI>{{.Size | printf "%15d " | nbsp}}<A href="{{print "./" .Name}}">{{.Name}}</A></LI>
629 <P>(No files; this collection is empty.)</P>
636 Arvados is a free and open source software bioinformatics platform.
637 To learn more, visit arvados.org.
638 Arvados is not responsible for the files listed on this page.
645 type fileListEnt struct {
651 func (h *handler) serveDirectory(w http.ResponseWriter, r *http.Request, collectionName string, fs http.FileSystem, base string, recurse bool) {
652 var files []fileListEnt
653 var walk func(string) error
654 if !strings.HasSuffix(base, "/") {
657 walk = func(path string) error {
658 dirname := base + path
660 dirname = strings.TrimSuffix(dirname, "/")
662 d, err := fs.Open(dirname)
666 ents, err := d.Readdir(-1)
670 for _, ent := range ents {
671 if recurse && ent.IsDir() {
672 err = walk(path + ent.Name() + "/")
677 files = append(files, fileListEnt{
678 Name: path + ent.Name(),
686 if err := walk(""); err != nil {
687 http.Error(w, err.Error(), http.StatusInternalServerError)
691 funcs := template.FuncMap{
692 "nbsp": func(s string) template.HTML {
693 return template.HTML(strings.Replace(s, " ", " ", -1))
696 tmpl, err := template.New("dir").Funcs(funcs).Parse(dirListingTemplate)
698 http.Error(w, err.Error(), http.StatusInternalServerError)
701 sort.Slice(files, func(i, j int) bool {
702 return files[i].Name < files[j].Name
704 w.WriteHeader(http.StatusOK)
705 tmpl.Execute(w, map[string]interface{}{
706 "CollectionName": collectionName,
709 "StripParts": strings.Count(strings.TrimRight(r.URL.Path, "/"), "/"),
713 func applyContentDispositionHdr(w http.ResponseWriter, r *http.Request, filename string, isAttachment bool) {
714 disposition := "inline"
716 disposition = "attachment"
718 if strings.ContainsRune(r.RequestURI, '?') {
719 // Help the UA realize that the filename is just
720 // "filename.txt", not
721 // "filename.txt?disposition=attachment".
723 // TODO(TC): Follow advice at RFC 6266 appendix D
724 disposition += "; filename=" + strconv.QuoteToASCII(filename)
726 if disposition != "inline" {
727 w.Header().Set("Content-Disposition", disposition)
731 func (h *handler) seeOtherWithCookie(w http.ResponseWriter, r *http.Request, location string, credentialsOK bool) {
732 if formToken := r.FormValue("api_token"); formToken != "" {
734 // It is not safe to copy the provided token
735 // into a cookie unless the current vhost
736 // (origin) serves only a single collection or
737 // we are in TrustAllContent mode.
738 w.WriteHeader(http.StatusBadRequest)
742 // The HttpOnly flag is necessary to prevent
743 // JavaScript code (included in, or loaded by, a page
744 // in the collection being served) from employing the
745 // user's token beyond reading other files in the same
746 // domain, i.e., same collection.
748 // The 303 redirect is necessary in the case of a GET
749 // request to avoid exposing the token in the Location
750 // bar, and in the case of a POST request to avoid
751 // raising warnings when the user refreshes the
753 http.SetCookie(w, &http.Cookie{
754 Name: "arvados_api_token",
755 Value: auth.EncodeTokenCookie([]byte(formToken)),
761 // Propagate query parameters (except api_token) from
762 // the original request.
763 redirQuery := r.URL.Query()
764 redirQuery.Del("api_token")
768 newu, err := u.Parse(location)
770 w.WriteHeader(http.StatusInternalServerError)
776 Scheme: r.URL.Scheme,
779 RawQuery: redirQuery.Encode(),
782 w.Header().Add("Location", redir)
783 w.WriteHeader(http.StatusSeeOther)
784 io.WriteString(w, `<A href="`)
785 io.WriteString(w, html.EscapeString(redir))
786 io.WriteString(w, `">Continue</A>`)