NXD xblob: don't let net/http sniff content type

This has security implications

https://gitlab.com/gitlab-org/gitlab-foss/-/issues/9079

internal/git/xblob.go

@@ -152,21 +152,44 @@ func emitBlob(w http.ResponseWriter, repopath string, refpath string) {
 	w.Header().Set("Content-Transfer-Encoding", "binary")
 	w.Header().Set("Content-Length", fmt.Sprintf("%d", size))
 	w.Header().Set("X-Content-Type-Options", "nosniff")
-	// net/http sniffs stream and automatically detects and sets
-	// Content-Type header. We do not have to do it ourselves.
-
 	w.Header().Set("Cache-Control", "private") // Rails sets this for IE compatibility
 	w.Header().Set("ETag", fmt.Sprintf(`"%s"`, sha1))
 
-	w.WriteHeader(http.StatusOK) // Don't bother with HTTP 500 from this point on, just return
-	// XXX better use queryStdout instead of queryReader, but we could be
-	// holding some tail bytes in queryReader after chat phase
-	_, err = io.CopyN(w, queryReader, size)
+	// read a little bit buffer to sniff content type, then we'll use io.CopyN to copy
+	// the rest
+	var buffsize int64 = 1024
+	if buffsize > size {
+		buffsize = size
+	}
+	buffer := make([]byte, buffsize)
+	_, err = io.ReadFull(queryReader, buffer)
 	if err != nil {
-		helper.LogError(fmt.Errorf("io.CopyN: %v", err))
+		helper.Fail500(w, fmt.Errorf("io.ReadFull: %v", err))
 		return
 	}
 
+	// Serve html, javascript, etc as text/plain, using utf-8 charset because it's more likely to be correct
+	// than the default latin1. Keep the original content type only for images, the /raw/ of an image is
+	// accessed from the /blob/ view.
+	// See https://gitlab.com/gitlab-org/gitlab-foss/-/blob/77c7b561/app/helpers/blob_helper.rb#L138-144
+	contentType := http.DetectContentType(buffer)
+	if strings.HasPrefix(contentType, "text/") {
+		w.Header().Set("Content-Type", "text/plain; charset=utf-8")
+	} else if strings.HasPrefix(contentType, "image/") {
+		w.Header().Set("Content-Type", contentType)
+	} else {
+		w.Header().Set("Content-Type", "application/octet-stream")
+	}
+	w.WriteHeader(http.StatusOK) // Don't bother with HTTP 500 from this point on, just return
+	w.Write(buffer)
+	if size > buffsize {
+		_, err = io.CopyN(w, queryReader, size-buffsize)
+		if err != nil {
+			helper.LogError(fmt.Errorf("io.CopyN: %v", err))
+			return
+		}
+	}
+
 	// close git stdin explicitly, so it can exit cleanly
 	err = queryStdin.Close()
 	if err != nil {

main_test.go

@@ -814,6 +814,18 @@ func TestBlobDownload(t *testing.T) {
 	dl.ExpectSha1("/5f923865/files/ruby/popen.rb", "68990cc20fa74383358797a27967fa2b45d7d8f6")
 	dl.ExpectSha1("/874797c3/files/ruby/popen.rb", "4c266708f2bfd7ca3fed3f7ec74253f92ff3fe73")
 	dl.ExpectCode("/master/non-existing-file", 404)
+
+	dl.ExpectHeader("/5f923865/README", "content-type", "text/plain; charset=utf-8")
+	dl.ExpectHeader("/5f923865/files/html/500.html", "content-type", "text/plain; charset=utf-8")
+	dl.ExpectHeader("/5f923865/files/js/application.js", "content-type", "text/plain; charset=utf-8")
+	dl.ExpectHeader("/5f923865/files/images/logo-white.png", "content-type", "image/png")
+	dl.ExpectHeader("/5f923865/files/images/6049019_460s.jpg", "content-type", "image/jpeg")
+	dl.ExpectHeader("/5f923865/files/images/wm.svg", "content-type", "text/plain; charset=utf-8")
+	dl.ExpectHeader("/5f923865/Gemfile.zip", "content-type", "application/octet-stream")
+	dl.ExpectHeader("/5f923865/files/images/wm.svg", "content-type", "text/plain; charset=utf-8")
+	dl.ExpectHeader("/5f923865/VERSION", "content-type", "text/plain; charset=utf-8")
+	// empty file
+	dl.ExpectHeader("/5f923865/files/flat/path/correct/content.txt", "content-type", "text/plain; charset=utf-8")
 }
 
 func TestDeniedBlobDownload(t *testing.T) {