1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
25 "git.arvados.org/arvados.git/sdk/go/arvados"
26 "git.arvados.org/arvados.git/sdk/go/ctxlog"
27 "github.com/AdRoll/goamz/s3"
32 s3SignAlgorithm = "AWS4-HMAC-SHA256"
33 s3MaxClockSkew = 5 * time.Minute
36 func hmacstring(msg string, key []byte) []byte {
37 h := hmac.New(sha256.New, key)
38 io.WriteString(h, msg)
42 func hashdigest(h hash.Hash, payload string) string {
43 io.WriteString(h, payload)
44 return fmt.Sprintf("%x", h.Sum(nil))
47 // Signing key for given secret key and request attrs.
48 func s3signatureKey(key, datestamp, regionName, serviceName string) []byte {
49 return hmacstring("aws4_request",
50 hmacstring(serviceName,
51 hmacstring(regionName,
52 hmacstring(datestamp, []byte("AWS4"+key)))))
55 // Canonical query string for S3 V4 signature: sorted keys, spaces
56 // escaped as %20 instead of +, keyvalues joined with &.
57 func s3querystring(u *url.URL) string {
58 keys := make([]string, 0, len(u.Query()))
59 values := make(map[string]string, len(u.Query()))
60 for k, vs := range u.Query() {
61 k = strings.Replace(url.QueryEscape(k), "+", "%20", -1)
62 keys = append(keys, k)
63 for _, v := range vs {
64 v = strings.Replace(url.QueryEscape(v), "+", "%20", -1)
68 values[k] += k + "=" + v
72 for i, k := range keys {
75 return strings.Join(keys, "&")
78 var reMultipleSlashChars = regexp.MustCompile(`//+`)
80 func s3stringToSign(alg, scope, signedHeaders string, r *http.Request) (string, error) {
81 timefmt, timestr := "20060102T150405Z", r.Header.Get("X-Amz-Date")
83 timefmt, timestr = time.RFC1123, r.Header.Get("Date")
85 t, err := time.Parse(timefmt, timestr)
87 return "", fmt.Errorf("invalid timestamp %q: %s", timestr, err)
89 if skew := time.Now().Sub(t); skew < -s3MaxClockSkew || skew > s3MaxClockSkew {
90 return "", errors.New("exceeded max clock skew")
93 var canonicalHeaders string
94 for _, h := range strings.Split(signedHeaders, ";") {
96 canonicalHeaders += h + ":" + r.Host + "\n"
98 canonicalHeaders += h + ":" + r.Header.Get(h) + "\n"
102 normalizedURL := *r.URL
103 normalizedURL.RawPath = ""
104 normalizedURL.Path = reMultipleSlashChars.ReplaceAllString(normalizedURL.Path, "/")
105 canonicalRequest := fmt.Sprintf("%s\n%s\n%s\n%s\n%s\n%s", r.Method, normalizedURL.EscapedPath(), s3querystring(r.URL), canonicalHeaders, signedHeaders, r.Header.Get("X-Amz-Content-Sha256"))
106 ctxlog.FromContext(r.Context()).Debugf("s3stringToSign: canonicalRequest %s", canonicalRequest)
107 return fmt.Sprintf("%s\n%s\n%s\n%s", alg, r.Header.Get("X-Amz-Date"), scope, hashdigest(sha256.New(), canonicalRequest)), nil
110 func s3signature(secretKey, scope, signedHeaders, stringToSign string) (string, error) {
111 // scope is {datestamp}/{region}/{service}/aws4_request
112 drs := strings.Split(scope, "/")
114 return "", fmt.Errorf("invalid scope %q", scope)
116 key := s3signatureKey(secretKey, drs[0], drs[1], drs[2])
117 return hashdigest(hmac.New(sha256.New, key), stringToSign), nil
120 // checks3signature verifies the given S3 V4 signature and returns the
121 // Arvados token that corresponds to the given accessKey. An error is
122 // returned if accessKey is not a valid token UUID or the signature
124 func (h *handler) checks3signature(r *http.Request) (string, error) {
125 var key, scope, signedHeaders, signature string
126 authstring := strings.TrimPrefix(r.Header.Get("Authorization"), s3SignAlgorithm+" ")
127 for _, cmpt := range strings.Split(authstring, ",") {
128 cmpt = strings.TrimSpace(cmpt)
129 split := strings.SplitN(cmpt, "=", 2)
131 case len(split) != 2:
133 case split[0] == "Credential":
134 keyandscope := strings.SplitN(split[1], "/", 2)
135 if len(keyandscope) == 2 {
136 key, scope = keyandscope[0], keyandscope[1]
138 case split[0] == "SignedHeaders":
139 signedHeaders = split[1]
140 case split[0] == "Signature":
145 client := (&arvados.Client{
146 APIHost: h.Config.cluster.Services.Controller.ExternalURL.Host,
147 Insecure: h.Config.cluster.TLS.Insecure,
148 }).WithRequestID(r.Header.Get("X-Request-Id"))
149 var aca arvados.APIClientAuthorization
152 if len(key) == 27 && key[5:12] == "-gj3su-" {
153 // Access key is the UUID of an Arvados token, secret
154 // key is the secret part.
155 ctx := arvados.ContextWithAuthorization(r.Context(), "Bearer "+h.Config.cluster.SystemRootToken)
156 err = client.RequestAndDecodeContext(ctx, &aca, "GET", "arvados/v1/api_client_authorizations/"+key, nil, nil)
157 secret = aca.APIToken
159 // Access key and secret key are both an entire
160 // Arvados token or OIDC access token.
161 ctx := arvados.ContextWithAuthorization(r.Context(), "Bearer "+key)
162 err = client.RequestAndDecodeContext(ctx, &aca, "GET", "arvados/v1/api_client_authorizations/current", nil, nil)
166 ctxlog.FromContext(r.Context()).WithError(err).WithField("UUID", key).Info("token lookup failed")
167 return "", errors.New("invalid access key")
169 stringToSign, err := s3stringToSign(s3SignAlgorithm, scope, signedHeaders, r)
173 expect, err := s3signature(secret, scope, signedHeaders, stringToSign)
176 } else if expect != signature {
177 return "", fmt.Errorf("signature does not match (scope %q signedHeaders %q stringToSign %q)", scope, signedHeaders, stringToSign)
182 func s3ErrorResponse(w http.ResponseWriter, s3code string, message string, resource string, code int) {
183 w.Header().Set("Content-Type", "application/xml")
184 w.Header().Set("X-Content-Type-Options", "nosniff")
186 var errstruct struct {
192 errstruct.Code = s3code
193 errstruct.Message = message
194 errstruct.Resource = resource
195 errstruct.RequestId = ""
196 enc := xml.NewEncoder(w)
197 fmt.Fprint(w, xml.Header)
198 enc.EncodeElement(errstruct, xml.StartElement{Name: xml.Name{Local: "Error"}})
201 var NoSuchKey = "NoSuchKey"
202 var NoSuchBucket = "NoSuchBucket"
203 var InvalidArgument = "InvalidArgument"
204 var InternalError = "InternalError"
205 var UnauthorizedAccess = "UnauthorizedAccess"
206 var InvalidRequest = "InvalidRequest"
207 var SignatureDoesNotMatch = "SignatureDoesNotMatch"
209 // serveS3 handles r and returns true if r is a request from an S3
210 // client, otherwise it returns false.
211 func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
213 if auth := r.Header.Get("Authorization"); strings.HasPrefix(auth, "AWS ") {
214 split := strings.SplitN(auth[4:], ":", 2)
216 s3ErrorResponse(w, InvalidRequest, "malformed Authorization header", r.URL.Path, http.StatusUnauthorized)
220 } else if strings.HasPrefix(auth, s3SignAlgorithm+" ") {
221 t, err := h.checks3signature(r)
223 s3ErrorResponse(w, SignatureDoesNotMatch, "signature verification failed: "+err.Error(), r.URL.Path, http.StatusForbidden)
232 var fs arvados.CustomFileSystem
233 if r.Method == http.MethodGet || r.Method == http.MethodHead {
234 // Use a single session (cached FileSystem) across
235 // multiple read requests.
236 fs, err = h.Config.Cache.GetSession(token)
238 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
242 // Create a FileSystem for this request, to avoid
243 // exposing incomplete write operations to concurrent
245 _, kc, client, release, err := h.getClients(r.Header.Get("X-Request-Id"), token)
247 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
251 fs = client.SiteFileSystem(kc)
252 fs.ForwardSlashNameSubstitution(h.Config.cluster.Collections.ForwardSlashNameSubstitution)
255 var objectNameGiven bool
256 var bucketName string
258 if id := parseCollectionIDFromDNSName(r.Host); id != "" {
261 objectNameGiven = strings.Count(strings.TrimSuffix(r.URL.Path, "/"), "/") > 0
263 bucketName = strings.SplitN(strings.TrimPrefix(r.URL.Path, "/"), "/", 2)[0]
264 objectNameGiven = strings.Count(strings.TrimSuffix(r.URL.Path, "/"), "/") > 1
266 fspath += reMultipleSlashChars.ReplaceAllString(r.URL.Path, "/")
269 case r.Method == http.MethodGet && !objectNameGiven:
270 // Path is "/{uuid}" or "/{uuid}/", has no object name
271 if _, ok := r.URL.Query()["versioning"]; ok {
272 // GetBucketVersioning
273 w.Header().Set("Content-Type", "application/xml")
274 io.WriteString(w, xml.Header)
275 fmt.Fprintln(w, `<VersioningConfiguration xmlns="http://s3.amazonaws.com/doc/2006-03-01/"/>`)
276 } else if _, ok = r.URL.Query()["location"]; ok {
278 w.Header().Set("Content-Type", "application/xml")
279 io.WriteString(w, xml.Header)
280 fmt.Fprintln(w, `<LocationConstraint xmlns="http://s3.amazonaws.com/doc/2006-03-01/">`+h.Config.cluster.ClusterID+`</LocationConstraint>`)
283 h.s3list(bucketName, w, r, fs)
286 case r.Method == http.MethodGet || r.Method == http.MethodHead:
287 fi, err := fs.Stat(fspath)
288 if r.Method == "HEAD" && !objectNameGiven {
290 if err == nil && fi.IsDir() {
291 w.WriteHeader(http.StatusOK)
292 } else if os.IsNotExist(err) {
293 s3ErrorResponse(w, NoSuchBucket, "The specified bucket does not exist.", r.URL.Path, http.StatusNotFound)
295 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusBadGateway)
299 if err == nil && fi.IsDir() && objectNameGiven && strings.HasSuffix(fspath, "/") && h.Config.cluster.Collections.S3FolderObjects {
300 w.Header().Set("Content-Type", "application/x-directory")
301 w.WriteHeader(http.StatusOK)
304 if os.IsNotExist(err) ||
305 (err != nil && err.Error() == "not a directory") ||
306 (fi != nil && fi.IsDir()) {
307 s3ErrorResponse(w, NoSuchKey, "The specified key does not exist.", r.URL.Path, http.StatusNotFound)
310 // shallow copy r, and change URL path
313 http.FileServer(fs).ServeHTTP(w, &r)
315 case r.Method == http.MethodPut:
316 if !objectNameGiven {
317 s3ErrorResponse(w, InvalidArgument, "Missing object name in PUT request.", r.URL.Path, http.StatusBadRequest)
321 if strings.HasSuffix(fspath, "/") {
322 if !h.Config.cluster.Collections.S3FolderObjects {
323 s3ErrorResponse(w, InvalidArgument, "invalid object name: trailing slash", r.URL.Path, http.StatusBadRequest)
326 n, err := r.Body.Read(make([]byte, 1))
327 if err != nil && err != io.EOF {
328 s3ErrorResponse(w, InternalError, fmt.Sprintf("error reading request body: %s", err), r.URL.Path, http.StatusInternalServerError)
331 s3ErrorResponse(w, InvalidArgument, "cannot create object with trailing '/' char unless content is empty", r.URL.Path, http.StatusBadRequest)
333 } else if strings.SplitN(r.Header.Get("Content-Type"), ";", 2)[0] != "application/x-directory" {
334 s3ErrorResponse(w, InvalidArgument, "cannot create object with trailing '/' char unless Content-Type is 'application/x-directory'", r.URL.Path, http.StatusBadRequest)
337 // Given PUT "foo/bar/", we'll use "foo/bar/."
338 // in the "ensure parents exist" block below,
339 // and then we'll be done.
343 fi, err := fs.Stat(fspath)
344 if err != nil && err.Error() == "not a directory" {
345 // requested foo/bar, but foo is a file
346 s3ErrorResponse(w, InvalidArgument, "object name conflicts with existing object", r.URL.Path, http.StatusBadRequest)
349 if strings.HasSuffix(r.URL.Path, "/") && err == nil && !fi.IsDir() {
350 // requested foo/bar/, but foo/bar is a file
351 s3ErrorResponse(w, InvalidArgument, "object name conflicts with existing object", r.URL.Path, http.StatusBadRequest)
354 // create missing parent/intermediate directories, if any
355 for i, c := range fspath {
356 if i > 0 && c == '/' {
358 if strings.HasSuffix(dir, "/") {
359 err = errors.New("invalid object name (consecutive '/' chars)")
360 s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
363 err = fs.Mkdir(dir, 0755)
364 if err == arvados.ErrInvalidArgument {
365 // Cannot create a directory
367 err = fmt.Errorf("mkdir %q failed: %w", dir, err)
368 s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
370 } else if err != nil && !os.IsExist(err) {
371 err = fmt.Errorf("mkdir %q failed: %w", dir, err)
372 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
378 f, err := fs.OpenFile(fspath, os.O_WRONLY|os.O_TRUNC|os.O_CREATE, 0644)
379 if os.IsNotExist(err) {
380 f, err = fs.OpenFile(fspath, os.O_WRONLY|os.O_TRUNC|os.O_CREATE, 0644)
383 err = fmt.Errorf("open %q failed: %w", r.URL.Path, err)
384 s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
388 _, err = io.Copy(f, r.Body)
390 err = fmt.Errorf("write to %q failed: %w", r.URL.Path, err)
391 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusBadGateway)
396 err = fmt.Errorf("write to %q failed: close: %w", r.URL.Path, err)
397 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusBadGateway)
403 err = fmt.Errorf("sync failed: %w", err)
404 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
407 // Ensure a subsequent read operation will see the changes.
408 h.Config.Cache.ResetSession(token)
409 w.WriteHeader(http.StatusOK)
411 case r.Method == http.MethodDelete:
412 if !objectNameGiven || r.URL.Path == "/" {
413 s3ErrorResponse(w, InvalidArgument, "missing object name in DELETE request", r.URL.Path, http.StatusBadRequest)
416 if strings.HasSuffix(fspath, "/") {
417 fspath = strings.TrimSuffix(fspath, "/")
418 fi, err := fs.Stat(fspath)
419 if os.IsNotExist(err) {
420 w.WriteHeader(http.StatusNoContent)
422 } else if err != nil {
423 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
425 } else if !fi.IsDir() {
426 // if "foo" exists and is a file, then
427 // "foo/" doesn't exist, so we say
428 // delete was successful.
429 w.WriteHeader(http.StatusNoContent)
432 } else if fi, err := fs.Stat(fspath); err == nil && fi.IsDir() {
433 // if "foo" is a dir, it is visible via S3
434 // only as "foo/", not "foo" -- so we leave
435 // the dir alone and return 204 to indicate
436 // that "foo" does not exist.
437 w.WriteHeader(http.StatusNoContent)
440 err = fs.Remove(fspath)
441 if os.IsNotExist(err) {
442 w.WriteHeader(http.StatusNoContent)
446 err = fmt.Errorf("rm failed: %w", err)
447 s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
452 err = fmt.Errorf("sync failed: %w", err)
453 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
456 // Ensure a subsequent read operation will see the changes.
457 h.Config.Cache.ResetSession(token)
458 w.WriteHeader(http.StatusNoContent)
461 s3ErrorResponse(w, InvalidRequest, "method not allowed", r.URL.Path, http.StatusMethodNotAllowed)
466 // Call fn on the given path (directory) and its contents, in
467 // lexicographic order.
469 // If isRoot==true and path is not a directory, return nil.
471 // If fn returns filepath.SkipDir when called on a directory, don't
472 // descend into that directory.
473 func walkFS(fs arvados.CustomFileSystem, path string, isRoot bool, fn func(path string, fi os.FileInfo) error) error {
475 fi, err := fs.Stat(path)
476 if os.IsNotExist(err) || (err == nil && !fi.IsDir()) {
478 } else if err != nil {
482 if err == filepath.SkipDir {
484 } else if err != nil {
488 f, err := fs.Open(path)
489 if os.IsNotExist(err) && isRoot {
491 } else if err != nil {
492 return fmt.Errorf("open %q: %w", path, err)
498 fis, err := f.Readdir(-1)
502 sort.Slice(fis, func(i, j int) bool { return fis[i].Name() < fis[j].Name() })
503 for _, fi := range fis {
504 err = fn(path+"/"+fi.Name(), fi)
505 if err == filepath.SkipDir {
507 } else if err != nil {
511 err = walkFS(fs, path+"/"+fi.Name(), false, fn)
520 var errDone = errors.New("done")
522 func (h *handler) s3list(bucket string, w http.ResponseWriter, r *http.Request, fs arvados.CustomFileSystem) {
529 params.delimiter = r.FormValue("delimiter")
530 params.marker = r.FormValue("marker")
531 if mk, _ := strconv.ParseInt(r.FormValue("max-keys"), 10, 64); mk > 0 && mk < s3MaxKeys {
532 params.maxKeys = int(mk)
534 params.maxKeys = s3MaxKeys
536 params.prefix = r.FormValue("prefix")
538 bucketdir := "by_id/" + bucket
539 // walkpath is the directory (relative to bucketdir) we need
540 // to walk: the innermost directory that is guaranteed to
541 // contain all paths that have the requested prefix. Examples:
542 // prefix "foo/bar" => walkpath "foo"
543 // prefix "foo/bar/" => walkpath "foo/bar"
544 // prefix "foo" => walkpath ""
545 // prefix "" => walkpath ""
546 walkpath := params.prefix
547 if cut := strings.LastIndex(walkpath, "/"); cut >= 0 {
548 walkpath = walkpath[:cut]
553 type commonPrefix struct {
556 type listResp struct {
557 XMLName string `xml:"http://s3.amazonaws.com/doc/2006-03-01/ ListBucketResult"`
559 // s3.ListResp marshals an empty tag when
560 // CommonPrefixes is nil, which confuses some clients.
561 // Fix by using this nested struct instead.
562 CommonPrefixes []commonPrefix
563 // Similarly, we need omitempty here, because an empty
564 // tag confuses some clients (e.g.,
565 // github.com/aws/aws-sdk-net never terminates its
567 NextMarker string `xml:"NextMarker,omitempty"`
568 // ListObjectsV2 has a KeyCount response field.
572 ListResp: s3.ListResp{
574 Prefix: params.prefix,
575 Delimiter: params.delimiter,
576 Marker: params.marker,
577 MaxKeys: params.maxKeys,
580 commonPrefixes := map[string]bool{}
581 err := walkFS(fs, strings.TrimSuffix(bucketdir+"/"+walkpath, "/"), true, func(path string, fi os.FileInfo) error {
582 if path == bucketdir {
585 path = path[len(bucketdir)+1:]
586 filesize := fi.Size()
591 if len(path) <= len(params.prefix) {
592 if path > params.prefix[:len(path)] {
593 // with prefix "foobar", walking "fooz" means we're done
596 if path < params.prefix[:len(path)] {
597 // with prefix "foobar", walking "foobag" is pointless
598 return filepath.SkipDir
600 if fi.IsDir() && !strings.HasPrefix(params.prefix+"/", path) {
601 // with prefix "foo/bar", walking "fo"
602 // is pointless (but walking "foo" or
603 // "foo/bar" is necessary)
604 return filepath.SkipDir
606 if len(path) < len(params.prefix) {
607 // can't skip anything, and this entry
608 // isn't in the results, so just
613 if path[:len(params.prefix)] > params.prefix {
614 // with prefix "foobar", nothing we
615 // see after "foozzz" is relevant
619 if path < params.marker || path < params.prefix {
622 if fi.IsDir() && !h.Config.cluster.Collections.S3FolderObjects {
623 // Note we don't add anything to
624 // commonPrefixes here even if delimiter is
625 // "/". We descend into the directory, and
626 // return a commonPrefix only if we end up
627 // finding a regular file inside it.
630 if params.delimiter != "" {
631 idx := strings.Index(path[len(params.prefix):], params.delimiter)
633 // with prefix "foobar" and delimiter
634 // "z", when we hit "foobar/baz", we
635 // add "/baz" to commonPrefixes and
637 commonPrefixes[path[:len(params.prefix)+idx+1]] = true
638 return filepath.SkipDir
641 if len(resp.Contents)+len(commonPrefixes) >= params.maxKeys {
642 resp.IsTruncated = true
643 if params.delimiter != "" {
644 resp.NextMarker = path
648 resp.Contents = append(resp.Contents, s3.Key{
650 LastModified: fi.ModTime().UTC().Format("2006-01-02T15:04:05.999") + "Z",
655 if err != nil && err != errDone {
656 http.Error(w, err.Error(), http.StatusInternalServerError)
659 if params.delimiter != "" {
660 resp.CommonPrefixes = make([]commonPrefix, 0, len(commonPrefixes))
661 for prefix := range commonPrefixes {
662 resp.CommonPrefixes = append(resp.CommonPrefixes, commonPrefix{prefix})
664 sort.Slice(resp.CommonPrefixes, func(i, j int) bool { return resp.CommonPrefixes[i].Prefix < resp.CommonPrefixes[j].Prefix })
666 resp.KeyCount = len(resp.Contents)
667 w.Header().Set("Content-Type", "application/xml")
668 io.WriteString(w, xml.Header)
669 if err := xml.NewEncoder(w).Encode(resp); err != nil {
670 ctxlog.FromContext(r.Context()).WithError(err).Error("error writing xml response")