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.URL.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 // serveS3 handles r and returns true if r is a request from an S3
177 // client, otherwise it returns false.
178 func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
180 if auth := r.Header.Get("Authorization"); strings.HasPrefix(auth, "AWS ") {
181 split := strings.SplitN(auth[4:], ":", 2)
183 http.Error(w, "malformed Authorization header", http.StatusUnauthorized)
187 } else if strings.HasPrefix(auth, s3SignAlgorithm+" ") {
188 t, err := h.checks3signature(r)
190 http.Error(w, "signature verification failed: "+err.Error(), http.StatusForbidden)
198 _, kc, client, release, err := h.getClients(r.Header.Get("X-Request-Id"), token)
200 http.Error(w, "Pool failed: "+h.clientPool.Err().Error(), http.StatusInternalServerError)
205 fs := client.SiteFileSystem(kc)
206 fs.ForwardSlashNameSubstitution(h.Config.cluster.Collections.ForwardSlashNameSubstitution)
208 objectNameGiven := strings.Count(strings.TrimSuffix(r.URL.Path, "/"), "/") > 1
211 case r.Method == http.MethodGet && !objectNameGiven:
212 // Path is "/{uuid}" or "/{uuid}/", has no object name
213 if _, ok := r.URL.Query()["versioning"]; ok {
214 // GetBucketVersioning
215 w.Header().Set("Content-Type", "application/xml")
216 io.WriteString(w, xml.Header)
217 fmt.Fprintln(w, `<VersioningConfiguration xmlns="http://s3.amazonaws.com/doc/2006-03-01/"/>`)
223 case r.Method == http.MethodGet || r.Method == http.MethodHead:
224 fspath := "/by_id" + r.URL.Path
225 fi, err := fs.Stat(fspath)
226 if r.Method == "HEAD" && !objectNameGiven {
228 if err == nil && fi.IsDir() {
229 w.WriteHeader(http.StatusOK)
230 } else if os.IsNotExist(err) {
231 w.WriteHeader(http.StatusNotFound)
233 http.Error(w, err.Error(), http.StatusBadGateway)
237 if err == nil && fi.IsDir() && objectNameGiven && strings.HasSuffix(fspath, "/") && h.Config.cluster.Collections.S3FolderObjects {
238 w.Header().Set("Content-Type", "application/x-directory")
239 w.WriteHeader(http.StatusOK)
242 if os.IsNotExist(err) ||
243 (err != nil && err.Error() == "not a directory") ||
244 (fi != nil && fi.IsDir()) {
245 http.Error(w, "not found", http.StatusNotFound)
248 // shallow copy r, and change URL path
251 http.FileServer(fs).ServeHTTP(w, &r)
253 case r.Method == http.MethodPut:
254 if !objectNameGiven {
255 http.Error(w, "missing object name in PUT request", http.StatusBadRequest)
258 fspath := "by_id" + r.URL.Path
260 if strings.HasSuffix(fspath, "/") {
261 if !h.Config.cluster.Collections.S3FolderObjects {
262 http.Error(w, "invalid object name: trailing slash", http.StatusBadRequest)
265 n, err := r.Body.Read(make([]byte, 1))
266 if err != nil && err != io.EOF {
267 http.Error(w, fmt.Sprintf("error reading request body: %s", err), http.StatusInternalServerError)
270 http.Error(w, "cannot create object with trailing '/' char unless content is empty", http.StatusBadRequest)
272 } else if strings.SplitN(r.Header.Get("Content-Type"), ";", 2)[0] != "application/x-directory" {
273 http.Error(w, "cannot create object with trailing '/' char unless Content-Type is 'application/x-directory'", http.StatusBadRequest)
276 // Given PUT "foo/bar/", we'll use "foo/bar/."
277 // in the "ensure parents exist" block below,
278 // and then we'll be done.
282 fi, err := fs.Stat(fspath)
283 if err != nil && err.Error() == "not a directory" {
284 // requested foo/bar, but foo is a file
285 http.Error(w, "object name conflicts with existing object", http.StatusBadRequest)
288 if strings.HasSuffix(r.URL.Path, "/") && err == nil && !fi.IsDir() {
289 // requested foo/bar/, but foo/bar is a file
290 http.Error(w, "object name conflicts with existing object", http.StatusBadRequest)
293 // create missing parent/intermediate directories, if any
294 for i, c := range fspath {
295 if i > 0 && c == '/' {
297 if strings.HasSuffix(dir, "/") {
298 err = errors.New("invalid object name (consecutive '/' chars)")
299 http.Error(w, err.Error(), http.StatusBadRequest)
302 err = fs.Mkdir(dir, 0755)
303 if err == arvados.ErrInvalidArgument {
304 // Cannot create a directory
306 err = fmt.Errorf("mkdir %q failed: %w", dir, err)
307 http.Error(w, err.Error(), http.StatusBadRequest)
309 } else if err != nil && !os.IsExist(err) {
310 err = fmt.Errorf("mkdir %q failed: %w", dir, err)
311 http.Error(w, err.Error(), http.StatusInternalServerError)
317 f, err := fs.OpenFile(fspath, os.O_WRONLY|os.O_TRUNC|os.O_CREATE, 0644)
318 if os.IsNotExist(err) {
319 f, err = fs.OpenFile(fspath, os.O_WRONLY|os.O_TRUNC|os.O_CREATE, 0644)
322 err = fmt.Errorf("open %q failed: %w", r.URL.Path, err)
323 http.Error(w, err.Error(), http.StatusBadRequest)
327 _, err = io.Copy(f, r.Body)
329 err = fmt.Errorf("write to %q failed: %w", r.URL.Path, err)
330 http.Error(w, err.Error(), http.StatusBadGateway)
335 err = fmt.Errorf("write to %q failed: close: %w", r.URL.Path, err)
336 http.Error(w, err.Error(), http.StatusBadGateway)
342 err = fmt.Errorf("sync failed: %w", err)
343 http.Error(w, err.Error(), http.StatusInternalServerError)
346 w.WriteHeader(http.StatusOK)
348 case r.Method == http.MethodDelete:
349 if !objectNameGiven || r.URL.Path == "/" {
350 http.Error(w, "missing object name in DELETE request", http.StatusBadRequest)
353 fspath := "by_id" + r.URL.Path
354 if strings.HasSuffix(fspath, "/") {
355 fspath = strings.TrimSuffix(fspath, "/")
356 fi, err := fs.Stat(fspath)
357 if os.IsNotExist(err) {
358 w.WriteHeader(http.StatusNoContent)
360 } else if err != nil {
361 http.Error(w, err.Error(), http.StatusInternalServerError)
363 } else if !fi.IsDir() {
364 // if "foo" exists and is a file, then
365 // "foo/" doesn't exist, so we say
366 // delete was successful.
367 w.WriteHeader(http.StatusNoContent)
370 } else if fi, err := fs.Stat(fspath); err == nil && fi.IsDir() {
371 // if "foo" is a dir, it is visible via S3
372 // only as "foo/", not "foo" -- so we leave
373 // the dir alone and return 204 to indicate
374 // that "foo" does not exist.
375 w.WriteHeader(http.StatusNoContent)
378 err = fs.Remove(fspath)
379 if os.IsNotExist(err) {
380 w.WriteHeader(http.StatusNoContent)
384 err = fmt.Errorf("rm failed: %w", err)
385 http.Error(w, err.Error(), http.StatusBadRequest)
390 err = fmt.Errorf("sync failed: %w", err)
391 http.Error(w, err.Error(), http.StatusInternalServerError)
394 w.WriteHeader(http.StatusNoContent)
397 http.Error(w, "method not allowed", http.StatusMethodNotAllowed)
402 // Call fn on the given path (directory) and its contents, in
403 // lexicographic order.
405 // If isRoot==true and path is not a directory, return nil.
407 // If fn returns filepath.SkipDir when called on a directory, don't
408 // descend into that directory.
409 func walkFS(fs arvados.CustomFileSystem, path string, isRoot bool, fn func(path string, fi os.FileInfo) error) error {
411 fi, err := fs.Stat(path)
412 if os.IsNotExist(err) || (err == nil && !fi.IsDir()) {
414 } else if err != nil {
418 if err == filepath.SkipDir {
420 } else if err != nil {
424 f, err := fs.Open(path)
425 if os.IsNotExist(err) && isRoot {
427 } else if err != nil {
428 return fmt.Errorf("open %q: %w", path, err)
434 fis, err := f.Readdir(-1)
438 sort.Slice(fis, func(i, j int) bool { return fis[i].Name() < fis[j].Name() })
439 for _, fi := range fis {
440 err = fn(path+"/"+fi.Name(), fi)
441 if err == filepath.SkipDir {
443 } else if err != nil {
447 err = walkFS(fs, path+"/"+fi.Name(), false, fn)
456 var errDone = errors.New("done")
458 func (h *handler) s3list(w http.ResponseWriter, r *http.Request, fs arvados.CustomFileSystem) {
466 params.bucket = strings.SplitN(r.URL.Path[1:], "/", 2)[0]
467 params.delimiter = r.FormValue("delimiter")
468 params.marker = r.FormValue("marker")
469 if mk, _ := strconv.ParseInt(r.FormValue("max-keys"), 10, 64); mk > 0 && mk < s3MaxKeys {
470 params.maxKeys = int(mk)
472 params.maxKeys = s3MaxKeys
474 params.prefix = r.FormValue("prefix")
476 bucketdir := "by_id/" + params.bucket
477 // walkpath is the directory (relative to bucketdir) we need
478 // to walk: the innermost directory that is guaranteed to
479 // contain all paths that have the requested prefix. Examples:
480 // prefix "foo/bar" => walkpath "foo"
481 // prefix "foo/bar/" => walkpath "foo/bar"
482 // prefix "foo" => walkpath ""
483 // prefix "" => walkpath ""
484 walkpath := params.prefix
485 if cut := strings.LastIndex(walkpath, "/"); cut >= 0 {
486 walkpath = walkpath[:cut]
491 type commonPrefix struct {
494 type listResp struct {
495 XMLName string `xml:"http://s3.amazonaws.com/doc/2006-03-01/ ListBucketResult"`
497 // s3.ListResp marshals an empty tag when
498 // CommonPrefixes is nil, which confuses some clients.
499 // Fix by using this nested struct instead.
500 CommonPrefixes []commonPrefix
501 // Similarly, we need omitempty here, because an empty
502 // tag confuses some clients (e.g.,
503 // github.com/aws/aws-sdk-net never terminates its
505 NextMarker string `xml:"NextMarker,omitempty"`
508 ListResp: s3.ListResp{
509 Name: strings.SplitN(r.URL.Path[1:], "/", 2)[0],
510 Prefix: params.prefix,
511 Delimiter: params.delimiter,
512 Marker: params.marker,
513 MaxKeys: params.maxKeys,
516 commonPrefixes := map[string]bool{}
517 err := walkFS(fs, strings.TrimSuffix(bucketdir+"/"+walkpath, "/"), true, func(path string, fi os.FileInfo) error {
518 if path == bucketdir {
521 path = path[len(bucketdir)+1:]
522 filesize := fi.Size()
527 if len(path) <= len(params.prefix) {
528 if path > params.prefix[:len(path)] {
529 // with prefix "foobar", walking "fooz" means we're done
532 if path < params.prefix[:len(path)] {
533 // with prefix "foobar", walking "foobag" is pointless
534 return filepath.SkipDir
536 if fi.IsDir() && !strings.HasPrefix(params.prefix+"/", path) {
537 // with prefix "foo/bar", walking "fo"
538 // is pointless (but walking "foo" or
539 // "foo/bar" is necessary)
540 return filepath.SkipDir
542 if len(path) < len(params.prefix) {
543 // can't skip anything, and this entry
544 // isn't in the results, so just
549 if path[:len(params.prefix)] > params.prefix {
550 // with prefix "foobar", nothing we
551 // see after "foozzz" is relevant
555 if path < params.marker || path < params.prefix {
558 if fi.IsDir() && !h.Config.cluster.Collections.S3FolderObjects {
559 // Note we don't add anything to
560 // commonPrefixes here even if delimiter is
561 // "/". We descend into the directory, and
562 // return a commonPrefix only if we end up
563 // finding a regular file inside it.
566 if params.delimiter != "" {
567 idx := strings.Index(path[len(params.prefix):], params.delimiter)
569 // with prefix "foobar" and delimiter
570 // "z", when we hit "foobar/baz", we
571 // add "/baz" to commonPrefixes and
573 commonPrefixes[path[:len(params.prefix)+idx+1]] = true
574 return filepath.SkipDir
577 if len(resp.Contents)+len(commonPrefixes) >= params.maxKeys {
578 resp.IsTruncated = true
579 if params.delimiter != "" {
580 resp.NextMarker = path
584 resp.Contents = append(resp.Contents, s3.Key{
586 LastModified: fi.ModTime().UTC().Format("2006-01-02T15:04:05.999") + "Z",
591 if err != nil && err != errDone {
592 http.Error(w, err.Error(), http.StatusInternalServerError)
595 if params.delimiter != "" {
596 resp.CommonPrefixes = make([]commonPrefix, 0, len(commonPrefixes))
597 for prefix := range commonPrefixes {
598 resp.CommonPrefixes = append(resp.CommonPrefixes, commonPrefix{prefix})
600 sort.Slice(resp.CommonPrefixes, func(i, j int) bool { return resp.CommonPrefixes[i].Prefix < resp.CommonPrefixes[j].Prefix })
602 w.Header().Set("Content-Type", "application/xml")
603 io.WriteString(w, xml.Header)
604 if err := xml.NewEncoder(w).Encode(resp); err != nil {
605 ctxlog.FromContext(r.Context()).WithError(err).Error("error writing xml response")