Make serve_header.py listen on localhost and limit its CORS header (#5564)

Without a bind address in serve_header.yml, the server listened on all
interfaces, so any machine on the network could fetch the header and
trigger make runs in the working trees. It now listens on localhost
unless configured otherwise; bind: null restores the old behavior.

The header was also sent with Access-Control-Allow-Origin: *, letting
any web page read it. CORS is only needed because Compiler Explorer
downloads #include <https://...> headers in the browser, so the header
now goes only to https://godbolt.org and https://compiler-explorer.com,
configurable with cors_origins.

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

tools/serve_header/README.md

@@ -60,6 +60,10 @@ int main() {
 `serve_header.py` will try to read a configuration file `serve_header.yml` in the top level or project root directory, and will fall back on built-in defaults if the file cannot be read.
 An annotated example configuration can be found in `tools/serve_header/serve_header.yml.example`.
 
+By default, the server listens on `localhost` only, and only web pages from Compiler Explorer (`https://godbolt.org` and `https://compiler-explorer.com`) may read the header.
+Set `bind` to serve other machines as well; anyone who can reach the server can then trigger `make` runs in your working trees.
+Set `cors_origins` to allow other web pages.
+
 ## Serving `json.hpp` from multiple project directory instances or working trees
 
 `serve_header.py` was designed with the goal of supporting multiple project roots or working trees at the same time.

tools/serve_header/serve_header.py

@@ -26,6 +26,10 @@ HEADER = 'json.hpp'
 
 DATETIME_FORMAT = '%Y-%m-%d %H:%M:%S'
 
+# origins whose pages may read the served header from a browser; Compiler
+# Explorer downloads #include <https://...> headers client-side
+DEFAULT_CORS_ORIGINS = ['https://godbolt.org', 'https://compiler-explorer.com']
+
 JSON_VERSION_RE = re.compile(r'\s*#\s*define\s+NLOHMANN_JSON_VERSION_MAJOR\s+')
 
 class ExitHandler(logging.StreamHandler):
@@ -247,6 +251,8 @@ class WorkTrees(FileSystemEventHandler):
         self.observer.join()
 
 class HeaderRequestHandler(SimpleHTTPRequestHandler): # lgtm[py/missing-call-to-init]
+    cors_origins = DEFAULT_CORS_ORIGINS
+
     def __init__(self, request, client_address, server):
         """."""
         self.worktrees = server.worktrees
@@ -310,8 +316,11 @@ class HeaderRequestHandler(SimpleHTTPRequestHandler): # lgtm[py/missing-call-to-
 
         # set content length
         super().send_header('Content-Length', length)
-        # CORS header
-        self.send_header('Access-Control-Allow-Origin', '*')
+        # CORS header; only for the configured origins
+        origin = self.headers.get('Origin')
+        if origin in self.cors_origins:
+            self.send_header('Access-Control-Allow-Origin', origin)
+        self.send_header('Vary', 'Origin')
         # prevent caching
         self.send_header('Cache-Control', 'no-cache, no-store, must-revalidate')
         self.send_header('Pragma', 'no-cache')
@@ -383,8 +392,15 @@ if __name__ == '__main__':
         # find and monitor working trees
         worktrees = WorkTrees(config.get('root', '.'))
 
-        # start web server
-        infos = socket.getaddrinfo(config.get('bind', None), config.get('port', 8443),
+        # origins allowed to read the header from a browser
+        cors_origins = config.get('cors_origins', DEFAULT_CORS_ORIGINS)
+        if isinstance(cors_origins, str):
+            cors_origins = [cors_origins]
+        HeaderRequestHandler.cors_origins = cors_origins
+
+        # start web server; only reachable from this machine unless configured
+        # otherwise (bind: null listens on all interfaces)
+        infos = socket.getaddrinfo(config.get('bind', 'localhost'), config.get('port', 8443),
                                    type=socket.SOCK_STREAM, flags=socket.AI_PASSIVE)
         DualStackServer.address_family = infos[0][0]
         HeaderRequestHandler.protocol_version = 'HTTP/1.0'

tools/serve_header/serve_header.yml.example

@@ -10,6 +10,13 @@
 #  cert_file: localhost.pem
 #  key_file: localhost-key.pem
 
-# address and port for the server to listen on
-# bind: null
+# address and port for the server to listen on; by default, only this machine
+# can connect. Binding to a network address, or to null for all interfaces,
+# lets other machines connect, and every request runs make in a working tree.
+# bind: localhost
 # port: 8443
+
+# origins whose web pages may read the header (CORS)
+# cors_origins:
+#   - https://godbolt.org
+#   - https://compiler-explorer.com