Add regexp import to fix merge mistake
[arvados.git] / services / keep-web / s3.go
index f3edf9894e243c1e99dc2cfd1c669a250e342843..66f89db9e3f282fb9373f9686b5099b4f37e2dba 100644 (file)
@@ -16,6 +16,7 @@ import (
        "net/url"
        "os"
        "path/filepath"
+       "regexp"
        "sort"
        "strconv"
        "strings"
@@ -74,6 +75,8 @@ func s3querystring(u *url.URL) string {
        return strings.Join(keys, "&")
 }
 
+var reMultipleSlashChars = regexp.MustCompile(`//+`)
+
 func s3stringToSign(alg, scope, signedHeaders string, r *http.Request) (string, error) {
        timefmt, timestr := "20060102T150405Z", r.Header.Get("X-Amz-Date")
        if timestr == "" {
@@ -96,7 +99,10 @@ func s3stringToSign(alg, scope, signedHeaders string, r *http.Request) (string,
                }
        }
 
-       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"))
+       normalizedURL := *r.URL
+       normalizedURL.RawPath = ""
+       normalizedURL.Path = reMultipleSlashChars.ReplaceAllString(normalizedURL.Path, "/")
+       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"))
        ctxlog.FromContext(r.Context()).Debugf("s3stringToSign: canonicalRequest %s", canonicalRequest)
        return fmt.Sprintf("%s\n%s\n%s\n%s", alg, r.Header.Get("X-Amz-Date"), scope, hashdigest(sha256.New(), canonicalRequest)), nil
 }
@@ -173,6 +179,33 @@ func (h *handler) checks3signature(r *http.Request) (string, error) {
        return secret, 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 +213,14 @@ 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]
        } 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 +230,7 @@ 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 +238,18 @@ 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
+       var bucketName string
+       fspath := "/by_id"
+       if id := parseCollectionIDFromDNSName(r.Host); id != "" {
+               fspath += "/" + id
+               bucketName = id
+               objectNameGiven = strings.Count(strings.TrimSuffix(r.URL.Path, "/"), "/") > 0
+       } else {
+               bucketName = strings.SplitN(strings.TrimPrefix(r.URL.Path, "/"), "/", 2)[0]
+               objectNameGiven = strings.Count(strings.TrimSuffix(r.URL.Path, "/"), "/") > 1
+       }
+       fspath += reMultipleSlashChars.ReplaceAllString(r.URL.Path, "/")
 
        switch {
        case r.Method == http.MethodGet && !objectNameGiven:
@@ -217,20 +261,19 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
                        fmt.Fprintln(w, `<VersioningConfiguration xmlns="http://s3.amazonaws.com/doc/2006-03-01/"/>`)
                } else {
                        // ListObjects
-                       h.s3list(w, r, fs)
+                       h.s3list(bucketName, w, r, fs)
                }
                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 +285,7 @@ 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 +295,24 @@ 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 +324,12 @@ 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 +338,7 @@ 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 +346,11 @@ 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 +362,36 @@ 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 +399,7 @@ 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 +423,20 @@ 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
        }
 }
@@ -455,15 +497,13 @@ func walkFS(fs arvados.CustomFileSystem, path string, isRoot bool, fn func(path
 
 var errDone = errors.New("done")
 
-func (h *handler) s3list(w http.ResponseWriter, r *http.Request, fs arvados.CustomFileSystem) {
+func (h *handler) s3list(bucket string, w http.ResponseWriter, r *http.Request, fs arvados.CustomFileSystem) {
        var params struct {
-               bucket    string
                delimiter string
                marker    string
                maxKeys   int
                prefix    string
        }
-       params.bucket = strings.SplitN(r.URL.Path[1:], "/", 2)[0]
        params.delimiter = r.FormValue("delimiter")
        params.marker = r.FormValue("marker")
        if mk, _ := strconv.ParseInt(r.FormValue("max-keys"), 10, 64); mk > 0 && mk < s3MaxKeys {
@@ -473,7 +513,7 @@ func (h *handler) s3list(w http.ResponseWriter, r *http.Request, fs arvados.Cust
        }
        params.prefix = r.FormValue("prefix")
 
-       bucketdir := "by_id/" + params.bucket
+       bucketdir := "by_id/" + bucket
        // walkpath is the directory (relative to bucketdir) we need
        // to walk: the innermost directory that is guaranteed to
        // contain all paths that have the requested prefix. Examples:
@@ -503,10 +543,12 @@ func (h *handler) s3list(w http.ResponseWriter, r *http.Request, fs arvados.Cust
                // github.com/aws/aws-sdk-net never terminates its
                // paging loop).
                NextMarker string `xml:"NextMarker,omitempty"`
+               // ListObjectsV2 has a KeyCount response field.
+               KeyCount int
        }
        resp := listResp{
                ListResp: s3.ListResp{
-                       Name:      strings.SplitN(r.URL.Path[1:], "/", 2)[0],
+                       Name:      bucket,
                        Prefix:    params.prefix,
                        Delimiter: params.delimiter,
                        Marker:    params.marker,
@@ -599,6 +641,7 @@ func (h *handler) s3list(w http.ResponseWriter, r *http.Request, fs arvados.Cust
                }
                sort.Slice(resp.CommonPrefixes, func(i, j int) bool { return resp.CommonPrefixes[i].Prefix < resp.CommonPrefixes[j].Prefix })
        }
+       resp.KeyCount = len(resp.Contents)
        w.Header().Set("Content-Type", "application/xml")
        io.WriteString(w, xml.Header)
        if err := xml.NewEncoder(w).Encode(resp); err != nil {