1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
24 "git.arvados.org/arvados.git/sdk/go/arvados"
25 "git.arvados.org/arvados.git/sdk/go/ctxlog"
26 "github.com/AdRoll/goamz/s3"
31 s3SignAlgorithm = "AWS4-HMAC-SHA256"
32 s3MaxClockSkew = 5 * time.Minute
35 func hmacstring(msg string, key []byte) []byte {
36 h := hmac.New(sha256.New, key)
37 io.WriteString(h, msg)
41 func hashdigest(h hash.Hash, payload string) string {
42 io.WriteString(h, payload)
43 return fmt.Sprintf("%x", h.Sum(nil))
46 // Signing key for given secret key and request attrs.
47 func s3signatureKey(key, datestamp, regionName, serviceName string) []byte {
48 return hmacstring("aws4_request",
49 hmacstring(serviceName,
50 hmacstring(regionName,
51 hmacstring(datestamp, []byte("AWS4"+key)))))
54 // Canonical query string for S3 V4 signature: sorted keys, spaces
55 // escaped as %20 instead of +, keyvalues joined with &.
56 func s3querystring(u *url.URL) string {
57 keys := make([]string, 0, len(u.Query()))
58 values := make(map[string]string, len(u.Query()))
59 for k, vs := range u.Query() {
60 k = strings.Replace(url.QueryEscape(k), "+", "%20", -1)
61 keys = append(keys, k)
62 for _, v := range vs {
63 v = strings.Replace(url.QueryEscape(v), "+", "%20", -1)
67 values[k] += k + "=" + v
71 for i, k := range keys {
74 return strings.Join(keys, "&")
77 func s3stringToSign(alg, scope, signedHeaders string, r *http.Request) (string, error) {
78 timefmt, timestr := "20060102T150405Z", r.Header.Get("X-Amz-Date")
80 timefmt, timestr = time.RFC1123, r.Header.Get("Date")
82 t, err := time.Parse(timefmt, timestr)
84 return "", fmt.Errorf("invalid timestamp %q: %s", timestr, err)
86 if skew := time.Now().Sub(t); skew < -s3MaxClockSkew || skew > s3MaxClockSkew {
87 return "", errors.New("exceeded max clock skew")
90 var canonicalHeaders string
91 for _, h := range strings.Split(signedHeaders, ";") {
93 canonicalHeaders += h + ":" + r.Host + "\n"
95 canonicalHeaders += h + ":" + r.Header.Get(h) + "\n"
99 canonicalRequest := fmt.Sprintf("%s\n%s\n%s\n%s\n%s\n%s", r.Method, r.URL.EscapedPath(), s3querystring(r.URL), canonicalHeaders, signedHeaders, r.Header.Get("X-Amz-Content-Sha256"))
100 ctxlog.FromContext(r.Context()).Debugf("s3stringToSign: canonicalRequest %s", canonicalRequest)
101 return fmt.Sprintf("%s\n%s\n%s\n%s", alg, r.Header.Get("X-Amz-Date"), scope, hashdigest(sha256.New(), canonicalRequest)), nil
104 func s3signature(secretKey, scope, signedHeaders, stringToSign string) (string, error) {
105 // scope is {datestamp}/{region}/{service}/aws4_request
106 drs := strings.Split(scope, "/")
108 return "", fmt.Errorf("invalid scope %q", scope)
110 key := s3signatureKey(secretKey, drs[0], drs[1], drs[2])
111 return hashdigest(hmac.New(sha256.New, key), stringToSign), nil
114 // checks3signature verifies the given S3 V4 signature and returns the
115 // Arvados token that corresponds to the given accessKey. An error is
116 // returned if accessKey is not a valid token UUID or the signature
118 func (h *handler) checks3signature(r *http.Request) (string, error) {
119 var key, scope, signedHeaders, signature string
120 authstring := strings.TrimPrefix(r.Header.Get("Authorization"), s3SignAlgorithm+" ")
121 for _, cmpt := range strings.Split(authstring, ",") {
122 cmpt = strings.TrimSpace(cmpt)
123 split := strings.SplitN(cmpt, "=", 2)
125 case len(split) != 2:
127 case split[0] == "Credential":
128 keyandscope := strings.SplitN(split[1], "/", 2)
129 if len(keyandscope) == 2 {
130 key, scope = keyandscope[0], keyandscope[1]
132 case split[0] == "SignedHeaders":
133 signedHeaders = split[1]
134 case split[0] == "Signature":
139 client := (&arvados.Client{
140 APIHost: h.Config.cluster.Services.Controller.ExternalURL.Host,
141 Insecure: h.Config.cluster.TLS.Insecure,
142 }).WithRequestID(r.Header.Get("X-Request-Id"))
143 var aca arvados.APIClientAuthorization
146 if len(key) == 27 && key[5:12] == "-gj3su-" {
147 // Access key is the UUID of an Arvados token, secret
148 // key is the secret part.
149 ctx := arvados.ContextWithAuthorization(r.Context(), "Bearer "+h.Config.cluster.SystemRootToken)
150 err = client.RequestAndDecodeContext(ctx, &aca, "GET", "arvados/v1/api_client_authorizations/"+key, nil, nil)
151 secret = aca.APIToken
153 // Access key and secret key are both an entire
154 // Arvados token or OIDC access token.
155 ctx := arvados.ContextWithAuthorization(r.Context(), "Bearer "+key)
156 err = client.RequestAndDecodeContext(ctx, &aca, "GET", "arvados/v1/api_client_authorizations/current", nil, nil)
160 ctxlog.FromContext(r.Context()).WithError(err).WithField("UUID", key).Info("token lookup failed")
161 return "", errors.New("invalid access key")
163 stringToSign, err := s3stringToSign(s3SignAlgorithm, scope, signedHeaders, r)
167 expect, err := s3signature(secret, scope, signedHeaders, stringToSign)
170 } else if expect != signature {
171 return "", fmt.Errorf("signature does not match (scope %q signedHeaders %q stringToSign %q)", scope, signedHeaders, stringToSign)
176 func s3ErrorResponse(w http.ResponseWriter, s3code string, message string, resource string, code int) {
177 w.Header().Set("Content-Type", "application/xml")
178 w.Header().Set("X-Content-Type-Options", "nosniff")
180 var errstruct struct {
186 errstruct.Code = s3code
187 errstruct.Message = message
188 errstruct.Resource = resource
189 errstruct.RequestId = ""
190 enc := xml.NewEncoder(w)
191 fmt.Fprint(w, xml.Header)
192 enc.EncodeElement(errstruct, xml.StartElement{Name: xml.Name{Local: "Error"}})
195 var NoSuchKey = "NoSuchKey"
196 var NoSuchBucket = "NoSuchBucket"
197 var InvalidArgument = "InvalidArgument"
198 var InternalError = "InternalError"
199 var UnauthorizedAccess = "UnauthorizedAccess"
200 var InvalidRequest = "InvalidRequest"
201 var SignatureDoesNotMatch = "SignatureDoesNotMatch"
203 // serveS3 handles r and returns true if r is a request from an S3
204 // client, otherwise it returns false.
205 func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
207 if auth := r.Header.Get("Authorization"); strings.HasPrefix(auth, "AWS ") {
208 split := strings.SplitN(auth[4:], ":", 2)
210 s3ErrorResponse(w, InvalidRequest, "malformed Authorization header", r.URL.Path, http.StatusUnauthorized)
214 } else if strings.HasPrefix(auth, s3SignAlgorithm+" ") {
215 t, err := h.checks3signature(r)
217 s3ErrorResponse(w, SignatureDoesNotMatch, "signature verification failed: "+err.Error(), r.URL.Path, http.StatusForbidden)
225 _, kc, client, release, err := h.getClients(r.Header.Get("X-Request-Id"), token)
227 s3ErrorResponse(w, InternalError, "Pool failed: "+h.clientPool.Err().Error(), r.URL.Path, http.StatusInternalServerError)
232 fs := client.SiteFileSystem(kc)
233 fs.ForwardSlashNameSubstitution(h.Config.cluster.Collections.ForwardSlashNameSubstitution)
235 objectNameGiven := strings.Count(strings.TrimSuffix(r.URL.Path, "/"), "/") > 1
238 case r.Method == http.MethodGet && !objectNameGiven:
239 // Path is "/{uuid}" or "/{uuid}/", has no object name
240 if _, ok := r.URL.Query()["versioning"]; ok {
241 // GetBucketVersioning
242 w.Header().Set("Content-Type", "application/xml")
243 io.WriteString(w, xml.Header)
244 fmt.Fprintln(w, `<VersioningConfiguration xmlns="http://s3.amazonaws.com/doc/2006-03-01/"/>`)
250 case r.Method == http.MethodGet || r.Method == http.MethodHead:
251 fspath := "/by_id" + r.URL.Path
252 fi, err := fs.Stat(fspath)
253 if r.Method == "HEAD" && !objectNameGiven {
255 if err == nil && fi.IsDir() {
256 w.WriteHeader(http.StatusOK)
257 } else if os.IsNotExist(err) {
258 s3ErrorResponse(w, NoSuchBucket, "The specified bucket does not exist.", r.URL.Path, http.StatusNotFound)
260 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusBadGateway)
264 if err == nil && fi.IsDir() && objectNameGiven && strings.HasSuffix(fspath, "/") && h.Config.cluster.Collections.S3FolderObjects {
265 w.Header().Set("Content-Type", "application/x-directory")
266 w.WriteHeader(http.StatusOK)
269 if os.IsNotExist(err) ||
270 (err != nil && err.Error() == "not a directory") ||
271 (fi != nil && fi.IsDir()) {
272 s3ErrorResponse(w, NoSuchKey, "The specified key does not exist.", r.URL.Path, http.StatusNotFound)
275 // shallow copy r, and change URL path
278 http.FileServer(fs).ServeHTTP(w, &r)
280 case r.Method == http.MethodPut:
281 if !objectNameGiven {
282 s3ErrorResponse(w, InvalidArgument, "Missing object name in PUT request.", r.URL.Path, http.StatusBadRequest)
285 fspath := "by_id" + r.URL.Path
287 if strings.HasSuffix(fspath, "/") {
288 if !h.Config.cluster.Collections.S3FolderObjects {
289 s3ErrorResponse(w, InvalidArgument, "invalid object name: trailing slash", r.URL.Path, http.StatusBadRequest)
292 n, err := r.Body.Read(make([]byte, 1))
293 if err != nil && err != io.EOF {
294 s3ErrorResponse(w, InternalError, fmt.Sprintf("error reading request body: %s", err), r.URL.Path, http.StatusInternalServerError)
297 s3ErrorResponse(w, InvalidArgument, "cannot create object with trailing '/' char unless content is empty", r.URL.Path, http.StatusBadRequest)
299 } else if strings.SplitN(r.Header.Get("Content-Type"), ";", 2)[0] != "application/x-directory" {
300 s3ErrorResponse(w, InvalidArgument, "cannot create object with trailing '/' char unless Content-Type is 'application/x-directory'", r.URL.Path, http.StatusBadRequest)
303 // Given PUT "foo/bar/", we'll use "foo/bar/."
304 // in the "ensure parents exist" block below,
305 // and then we'll be done.
309 fi, err := fs.Stat(fspath)
310 if err != nil && err.Error() == "not a directory" {
311 // requested foo/bar, but foo is a file
312 s3ErrorResponse(w, InvalidArgument, "object name conflicts with existing object", r.URL.Path, http.StatusBadRequest)
315 if strings.HasSuffix(r.URL.Path, "/") && err == nil && !fi.IsDir() {
316 // requested foo/bar/, but foo/bar is a file
317 s3ErrorResponse(w, InvalidArgument, "object name conflicts with existing object", r.URL.Path, http.StatusBadRequest)
320 // create missing parent/intermediate directories, if any
321 for i, c := range fspath {
322 if i > 0 && c == '/' {
324 if strings.HasSuffix(dir, "/") {
325 err = errors.New("invalid object name (consecutive '/' chars)")
326 s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
329 err = fs.Mkdir(dir, 0755)
330 if err == arvados.ErrInvalidArgument {
331 // Cannot create a directory
333 err = fmt.Errorf("mkdir %q failed: %w", dir, err)
334 s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
336 } else if err != nil && !os.IsExist(err) {
337 err = fmt.Errorf("mkdir %q failed: %w", dir, err)
338 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
344 f, err := fs.OpenFile(fspath, os.O_WRONLY|os.O_TRUNC|os.O_CREATE, 0644)
345 if os.IsNotExist(err) {
346 f, err = fs.OpenFile(fspath, os.O_WRONLY|os.O_TRUNC|os.O_CREATE, 0644)
349 err = fmt.Errorf("open %q failed: %w", r.URL.Path, err)
350 s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
354 _, err = io.Copy(f, r.Body)
356 err = fmt.Errorf("write to %q failed: %w", r.URL.Path, err)
357 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusBadGateway)
362 err = fmt.Errorf("write to %q failed: close: %w", r.URL.Path, err)
363 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusBadGateway)
369 err = fmt.Errorf("sync failed: %w", err)
370 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
373 w.WriteHeader(http.StatusOK)
375 case r.Method == http.MethodDelete:
376 if !objectNameGiven || r.URL.Path == "/" {
377 s3ErrorResponse(w, InvalidArgument, "missing object name in DELETE request", r.URL.Path, http.StatusBadRequest)
380 fspath := "by_id" + r.URL.Path
381 if strings.HasSuffix(fspath, "/") {
382 fspath = strings.TrimSuffix(fspath, "/")
383 fi, err := fs.Stat(fspath)
384 if os.IsNotExist(err) {
385 w.WriteHeader(http.StatusNoContent)
387 } else if err != nil {
388 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
390 } else if !fi.IsDir() {
391 // if "foo" exists and is a file, then
392 // "foo/" doesn't exist, so we say
393 // delete was successful.
394 w.WriteHeader(http.StatusNoContent)
397 } else if fi, err := fs.Stat(fspath); err == nil && fi.IsDir() {
398 // if "foo" is a dir, it is visible via S3
399 // only as "foo/", not "foo" -- so we leave
400 // the dir alone and return 204 to indicate
401 // that "foo" does not exist.
402 w.WriteHeader(http.StatusNoContent)
405 err = fs.Remove(fspath)
406 if os.IsNotExist(err) {
407 w.WriteHeader(http.StatusNoContent)
411 err = fmt.Errorf("rm failed: %w", err)
412 s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
417 err = fmt.Errorf("sync failed: %w", err)
418 s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
421 w.WriteHeader(http.StatusNoContent)
424 s3ErrorResponse(w, InvalidRequest, "method not allowed", r.URL.Path, http.StatusMethodNotAllowed)
430 // Call fn on the given path (directory) and its contents, in
431 // lexicographic order.
433 // If isRoot==true and path is not a directory, return nil.
435 // If fn returns filepath.SkipDir when called on a directory, don't
436 // descend into that directory.
437 func walkFS(fs arvados.CustomFileSystem, path string, isRoot bool, fn func(path string, fi os.FileInfo) error) error {
439 fi, err := fs.Stat(path)
440 if os.IsNotExist(err) || (err == nil && !fi.IsDir()) {
442 } else if err != nil {
446 if err == filepath.SkipDir {
448 } else if err != nil {
452 f, err := fs.Open(path)
453 if os.IsNotExist(err) && isRoot {
455 } else if err != nil {
456 return fmt.Errorf("open %q: %w", path, err)
462 fis, err := f.Readdir(-1)
466 sort.Slice(fis, func(i, j int) bool { return fis[i].Name() < fis[j].Name() })
467 for _, fi := range fis {
468 err = fn(path+"/"+fi.Name(), fi)
469 if err == filepath.SkipDir {
471 } else if err != nil {
475 err = walkFS(fs, path+"/"+fi.Name(), false, fn)
484 var errDone = errors.New("done")
486 func (h *handler) s3list(w http.ResponseWriter, r *http.Request, fs arvados.CustomFileSystem) {
494 params.bucket = strings.SplitN(r.URL.Path[1:], "/", 2)[0]
495 params.delimiter = r.FormValue("delimiter")
496 params.marker = r.FormValue("marker")
497 if mk, _ := strconv.ParseInt(r.FormValue("max-keys"), 10, 64); mk > 0 && mk < s3MaxKeys {
498 params.maxKeys = int(mk)
500 params.maxKeys = s3MaxKeys
502 params.prefix = r.FormValue("prefix")
504 bucketdir := "by_id/" + params.bucket
505 // walkpath is the directory (relative to bucketdir) we need
506 // to walk: the innermost directory that is guaranteed to
507 // contain all paths that have the requested prefix. Examples:
508 // prefix "foo/bar" => walkpath "foo"
509 // prefix "foo/bar/" => walkpath "foo/bar"
510 // prefix "foo" => walkpath ""
511 // prefix "" => walkpath ""
512 walkpath := params.prefix
513 if cut := strings.LastIndex(walkpath, "/"); cut >= 0 {
514 walkpath = walkpath[:cut]
519 type commonPrefix struct {
522 type listResp struct {
523 XMLName string `xml:"http://s3.amazonaws.com/doc/2006-03-01/ ListBucketResult"`
525 // s3.ListResp marshals an empty tag when
526 // CommonPrefixes is nil, which confuses some clients.
527 // Fix by using this nested struct instead.
528 CommonPrefixes []commonPrefix
529 // Similarly, we need omitempty here, because an empty
530 // tag confuses some clients (e.g.,
531 // github.com/aws/aws-sdk-net never terminates its
533 NextMarker string `xml:"NextMarker,omitempty"`
534 // ListObjectsV2 has a KeyCount response field.
538 ListResp: s3.ListResp{
539 Name: strings.SplitN(r.URL.Path[1:], "/", 2)[0],
540 Prefix: params.prefix,
541 Delimiter: params.delimiter,
542 Marker: params.marker,
543 MaxKeys: params.maxKeys,
546 commonPrefixes := map[string]bool{}
547 err := walkFS(fs, strings.TrimSuffix(bucketdir+"/"+walkpath, "/"), true, func(path string, fi os.FileInfo) error {
548 if path == bucketdir {
551 path = path[len(bucketdir)+1:]
552 filesize := fi.Size()
557 if len(path) <= len(params.prefix) {
558 if path > params.prefix[:len(path)] {
559 // with prefix "foobar", walking "fooz" means we're done
562 if path < params.prefix[:len(path)] {
563 // with prefix "foobar", walking "foobag" is pointless
564 return filepath.SkipDir
566 if fi.IsDir() && !strings.HasPrefix(params.prefix+"/", path) {
567 // with prefix "foo/bar", walking "fo"
568 // is pointless (but walking "foo" or
569 // "foo/bar" is necessary)
570 return filepath.SkipDir
572 if len(path) < len(params.prefix) {
573 // can't skip anything, and this entry
574 // isn't in the results, so just
579 if path[:len(params.prefix)] > params.prefix {
580 // with prefix "foobar", nothing we
581 // see after "foozzz" is relevant
585 if path < params.marker || path < params.prefix {
588 if fi.IsDir() && !h.Config.cluster.Collections.S3FolderObjects {
589 // Note we don't add anything to
590 // commonPrefixes here even if delimiter is
591 // "/". We descend into the directory, and
592 // return a commonPrefix only if we end up
593 // finding a regular file inside it.
596 if params.delimiter != "" {
597 idx := strings.Index(path[len(params.prefix):], params.delimiter)
599 // with prefix "foobar" and delimiter
600 // "z", when we hit "foobar/baz", we
601 // add "/baz" to commonPrefixes and
603 commonPrefixes[path[:len(params.prefix)+idx+1]] = true
604 return filepath.SkipDir
607 if len(resp.Contents)+len(commonPrefixes) >= params.maxKeys {
608 resp.IsTruncated = true
609 if params.delimiter != "" {
610 resp.NextMarker = path
614 resp.Contents = append(resp.Contents, s3.Key{
616 LastModified: fi.ModTime().UTC().Format("2006-01-02T15:04:05.999") + "Z",
621 if err != nil && err != errDone {
622 http.Error(w, err.Error(), http.StatusInternalServerError)
625 if params.delimiter != "" {
626 resp.CommonPrefixes = make([]commonPrefix, 0, len(commonPrefixes))
627 for prefix := range commonPrefixes {
628 resp.CommonPrefixes = append(resp.CommonPrefixes, commonPrefix{prefix})
630 sort.Slice(resp.CommonPrefixes, func(i, j int) bool { return resp.CommonPrefixes[i].Prefix < resp.CommonPrefixes[j].Prefix })
632 resp.KeyCount = len(resp.Contents)
633 w.Header().Set("Content-Type", "application/xml")
634 io.WriteString(w, xml.Header)
635 if err := xml.NewEncoder(w).Encode(resp); err != nil {
636 ctxlog.FromContext(r.Context()).WithError(err).Error("error writing xml response")