Merge branch '16774-keep-web-errors' refs #16774
[arvados.git] / services / keep-web / s3.go
index 49fb2456f5851662bec9573af6e06978d930d741..7fb90789a55b164bd2465bf8d2454cf2d30e4f52 100644 (file)
@@ -16,6 +16,7 @@ import (
        "net/url"
        "os"
        "path/filepath"
+       "regexp"
        "sort"
        "strconv"
        "strings"
@@ -111,6 +112,21 @@ func s3signature(secretKey, scope, signedHeaders, stringToSign string) (string,
        return hashdigest(hmac.New(sha256.New, key), stringToSign), nil
 }
 
+var v2tokenUnderscore = regexp.MustCompile(`^v2_[a-z0-9]{5}-gj3su-[a-z0-9]{15}_`)
+
+func unescapeKey(key string) string {
+       if v2tokenUnderscore.MatchString(key) {
+               // Entire Arvados token, with "/" replaced by "_" to
+               // avoid colliding with the Authorization header
+               // format.
+               return strings.Replace(key, "_", "/", -1)
+       } else if s, err := url.PathUnescape(key); err == nil {
+               return s
+       } else {
+               return key
+       }
+}
+
 // checks3signature verifies the given S3 V4 signature and returns the
 // Arvados token that corresponds to the given accessKey. An error is
 // returned if accessKey is not a valid token UUID or the signature
@@ -152,7 +168,7 @@ func (h *handler) checks3signature(r *http.Request) (string, error) {
        } else {
                // Access key and secret key are both an entire
                // Arvados token or OIDC access token.
-               ctx := arvados.ContextWithAuthorization(r.Context(), "Bearer "+key)
+               ctx := arvados.ContextWithAuthorization(r.Context(), "Bearer "+unescapeKey(key))
                err = client.RequestAndDecodeContext(ctx, &aca, "GET", "arvados/v1/api_client_authorizations/current", nil, nil)
                secret = key
        }
@@ -170,9 +186,36 @@ func (h *handler) checks3signature(r *http.Request) (string, error) {
        } else if expect != signature {
                return "", fmt.Errorf("signature does not match (scope %q signedHeaders %q stringToSign %q)", scope, signedHeaders, stringToSign)
        }
-       return secret, nil
+       return aca.TokenV2(), nil
 }
 
+func s3ErrorResponse(w http.ResponseWriter, s3code string, message string, resource string, code int) {
+       w.Header().Set("Content-Type", "application/xml")
+       w.Header().Set("X-Content-Type-Options", "nosniff")
+       w.WriteHeader(code)
+       var errstruct struct {
+               Code      string
+               Message   string
+               Resource  string
+               RequestId string
+       }
+       errstruct.Code = s3code
+       errstruct.Message = message
+       errstruct.Resource = resource
+       errstruct.RequestId = ""
+       enc := xml.NewEncoder(w)
+       fmt.Fprint(w, xml.Header)
+       enc.EncodeElement(errstruct, xml.StartElement{Name: xml.Name{Local: "Error"}})
+}
+
+var NoSuchKey = "NoSuchKey"
+var NoSuchBucket = "NoSuchBucket"
+var InvalidArgument = "InvalidArgument"
+var InternalError = "InternalError"
+var UnauthorizedAccess = "UnauthorizedAccess"
+var InvalidRequest = "InvalidRequest"
+var SignatureDoesNotMatch = "SignatureDoesNotMatch"
+
 // serveS3 handles r and returns true if r is a request from an S3
 // client, otherwise it returns false.
 func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
@@ -180,14 +223,14 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
        if auth := r.Header.Get("Authorization"); strings.HasPrefix(auth, "AWS ") {
                split := strings.SplitN(auth[4:], ":", 2)
                if len(split) < 2 {
-                       http.Error(w, "malformed Authorization header", http.StatusUnauthorized)
+                       s3ErrorResponse(w, InvalidRequest, "malformed Authorization header", r.URL.Path, http.StatusUnauthorized)
                        return true
                }
-               token = split[0]
+               token = unescapeKey(split[0])
        } else if strings.HasPrefix(auth, s3SignAlgorithm+" ") {
                t, err := h.checks3signature(r)
                if err != nil {
-                       http.Error(w, "signature verification failed: "+err.Error(), http.StatusForbidden)
+                       s3ErrorResponse(w, SignatureDoesNotMatch, "signature verification failed: "+err.Error(), r.URL.Path, http.StatusForbidden)
                        return true
                }
                token = t
@@ -197,7 +240,7 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
 
        _, kc, client, release, err := h.getClients(r.Header.Get("X-Request-Id"), token)
        if err != nil {
-               http.Error(w, "Pool failed: "+h.clientPool.Err().Error(), http.StatusInternalServerError)
+               s3ErrorResponse(w, InternalError, "Pool failed: "+h.clientPool.Err().Error(), r.URL.Path, http.StatusInternalServerError)
                return true
        }
        defer release()
@@ -205,7 +248,15 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
        fs := client.SiteFileSystem(kc)
        fs.ForwardSlashNameSubstitution(h.Config.cluster.Collections.ForwardSlashNameSubstitution)
 
-       objectNameGiven := strings.Count(strings.TrimSuffix(r.URL.Path, "/"), "/") > 1
+       var objectNameGiven bool
+       fspath := "/by_id"
+       if id := parseCollectionIDFromDNSName(r.Host); id != "" {
+               fspath += "/" + id
+               objectNameGiven = strings.Count(strings.TrimSuffix(r.URL.Path, "/"), "/") > 0
+       } else {
+               objectNameGiven = strings.Count(strings.TrimSuffix(r.URL.Path, "/"), "/") > 1
+       }
+       fspath += r.URL.Path
 
        switch {
        case r.Method == http.MethodGet && !objectNameGiven:
@@ -221,16 +272,15 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
                }
                return true
        case r.Method == http.MethodGet || r.Method == http.MethodHead:
-               fspath := "/by_id" + r.URL.Path
                fi, err := fs.Stat(fspath)
                if r.Method == "HEAD" && !objectNameGiven {
                        // HeadBucket
                        if err == nil && fi.IsDir() {
                                w.WriteHeader(http.StatusOK)
                        } else if os.IsNotExist(err) {
-                               w.WriteHeader(http.StatusNotFound)
+                               s3ErrorResponse(w, NoSuchBucket, "The specified bucket does not exist.", r.URL.Path, http.StatusNotFound)
                        } else {
-                               http.Error(w, err.Error(), http.StatusBadGateway)
+                               s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusBadGateway)
                        }
                        return true
                }
@@ -242,7 +292,7 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
                if os.IsNotExist(err) ||
                        (err != nil && err.Error() == "not a directory") ||
                        (fi != nil && fi.IsDir()) {
-                       http.Error(w, "not found", http.StatusNotFound)
+                       s3ErrorResponse(w, NoSuchKey, "The specified key does not exist.", r.URL.Path, http.StatusNotFound)
                        return true
                }
                // shallow copy r, and change URL path
@@ -252,25 +302,24 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
                return true
        case r.Method == http.MethodPut:
                if !objectNameGiven {
-                       http.Error(w, "missing object name in PUT request", http.StatusBadRequest)
+                       s3ErrorResponse(w, InvalidArgument, "Missing object name in PUT request.", r.URL.Path, http.StatusBadRequest)
                        return true
                }
-               fspath := "by_id" + r.URL.Path
                var objectIsDir bool
                if strings.HasSuffix(fspath, "/") {
                        if !h.Config.cluster.Collections.S3FolderObjects {
-                               http.Error(w, "invalid object name: trailing slash", http.StatusBadRequest)
+                               s3ErrorResponse(w, InvalidArgument, "invalid object name: trailing slash", r.URL.Path, http.StatusBadRequest)
                                return true
                        }
                        n, err := r.Body.Read(make([]byte, 1))
                        if err != nil && err != io.EOF {
-                               http.Error(w, fmt.Sprintf("error reading request body: %s", err), http.StatusInternalServerError)
+                               s3ErrorResponse(w, InternalError, fmt.Sprintf("error reading request body: %s", err), r.URL.Path, http.StatusInternalServerError)
                                return true
                        } else if n > 0 {
-                               http.Error(w, "cannot create object with trailing '/' char unless content is empty", http.StatusBadRequest)
+                               s3ErrorResponse(w, InvalidArgument, "cannot create object with trailing '/' char unless content is empty", r.URL.Path, http.StatusBadRequest)
                                return true
                        } else if strings.SplitN(r.Header.Get("Content-Type"), ";", 2)[0] != "application/x-directory" {
-                               http.Error(w, "cannot create object with trailing '/' char unless Content-Type is 'application/x-directory'", http.StatusBadRequest)
+                               s3ErrorResponse(w, InvalidArgument, "cannot create object with trailing '/' char unless Content-Type is 'application/x-directory'", r.URL.Path, http.StatusBadRequest)
                                return true
                        }
                        // Given PUT "foo/bar/", we'll use "foo/bar/."
@@ -282,12 +331,12 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
                fi, err := fs.Stat(fspath)
                if err != nil && err.Error() == "not a directory" {
                        // requested foo/bar, but foo is a file
-                       http.Error(w, "object name conflicts with existing object", http.StatusBadRequest)
+                       s3ErrorResponse(w, InvalidArgument, "object name conflicts with existing object", r.URL.Path, http.StatusBadRequest)
                        return true
                }
                if strings.HasSuffix(r.URL.Path, "/") && err == nil && !fi.IsDir() {
                        // requested foo/bar/, but foo/bar is a file
-                       http.Error(w, "object name conflicts with existing object", http.StatusBadRequest)
+                       s3ErrorResponse(w, InvalidArgument, "object name conflicts with existing object", r.URL.Path, http.StatusBadRequest)
                        return true
                }
                // create missing parent/intermediate directories, if any
@@ -296,7 +345,7 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
                                dir := fspath[:i]
                                if strings.HasSuffix(dir, "/") {
                                        err = errors.New("invalid object name (consecutive '/' chars)")
-                                       http.Error(w, err.Error(), http.StatusBadRequest)
+                                       s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
                                        return true
                                }
                                err = fs.Mkdir(dir, 0755)
@@ -304,11 +353,11 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
                                        // Cannot create a directory
                                        // here.
                                        err = fmt.Errorf("mkdir %q failed: %w", dir, err)
-                                       http.Error(w, err.Error(), http.StatusBadRequest)
+                                       s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
                                        return true
                                } else if err != nil && !os.IsExist(err) {
                                        err = fmt.Errorf("mkdir %q failed: %w", dir, err)
-                                       http.Error(w, err.Error(), http.StatusInternalServerError)
+                                       s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
                                        return true
                                }
                        }
@@ -320,37 +369,36 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
                        }
                        if err != nil {
                                err = fmt.Errorf("open %q failed: %w", r.URL.Path, err)
-                               http.Error(w, err.Error(), http.StatusBadRequest)
+                               s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
                                return true
                        }
                        defer f.Close()
                        _, err = io.Copy(f, r.Body)
                        if err != nil {
                                err = fmt.Errorf("write to %q failed: %w", r.URL.Path, err)
-                               http.Error(w, err.Error(), http.StatusBadGateway)
+                               s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusBadGateway)
                                return true
                        }
                        err = f.Close()
                        if err != nil {
                                err = fmt.Errorf("write to %q failed: close: %w", r.URL.Path, err)
-                               http.Error(w, err.Error(), http.StatusBadGateway)
+                               s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusBadGateway)
                                return true
                        }
                }
                err = fs.Sync()
                if err != nil {
                        err = fmt.Errorf("sync failed: %w", err)
-                       http.Error(w, err.Error(), http.StatusInternalServerError)
+                       s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
                        return true
                }
                w.WriteHeader(http.StatusOK)
                return true
        case r.Method == http.MethodDelete:
                if !objectNameGiven || r.URL.Path == "/" {
-                       http.Error(w, "missing object name in DELETE request", http.StatusBadRequest)
+                       s3ErrorResponse(w, InvalidArgument, "missing object name in DELETE request", r.URL.Path, http.StatusBadRequest)
                        return true
                }
-               fspath := "by_id" + r.URL.Path
                if strings.HasSuffix(fspath, "/") {
                        fspath = strings.TrimSuffix(fspath, "/")
                        fi, err := fs.Stat(fspath)
@@ -358,7 +406,7 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
                                w.WriteHeader(http.StatusNoContent)
                                return true
                        } else if err != nil {
-                               http.Error(w, err.Error(), http.StatusInternalServerError)
+                               s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
                                return true
                        } else if !fi.IsDir() {
                                // if "foo" exists and is a file, then
@@ -382,19 +430,20 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
                }
                if err != nil {
                        err = fmt.Errorf("rm failed: %w", err)
-                       http.Error(w, err.Error(), http.StatusBadRequest)
+                       s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
                        return true
                }
                err = fs.Sync()
                if err != nil {
                        err = fmt.Errorf("sync failed: %w", err)
-                       http.Error(w, err.Error(), http.StatusInternalServerError)
+                       s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
                        return true
                }
                w.WriteHeader(http.StatusNoContent)
                return true
        default:
-               http.Error(w, "method not allowed", http.StatusMethodNotAllowed)
+               s3ErrorResponse(w, InvalidRequest, "method not allowed", r.URL.Path, http.StatusMethodNotAllowed)
+
                return true
        }
 }