Verify prepared runtime IDs at launch (Refs #93)

agent/ec_agent.py

@@ -35,7 +35,7 @@ import urllib.parse
 import urllib.request
 import uuid
 
-VERSION = "1.8.6"
+VERSION = "1.8.7"
 LABEL = "io.electric-cherry"
 MAX_FILE = 32 * 1024 * 1024
 MANIFEST_MAX_ENTRIES = 10000
@@ -1643,7 +1643,20 @@ class Agent:
             raise AgentError("Invalid runtime environment")
         # Runtime download is a separate preparation operation. A deployment
         # uses the already prepared image even during a registry outage.
-        self.docker.call("image", "inspect", image)
+        image_id = self.docker.call("image", "inspect", "--format", "{{.Id}}", image).strip()
+        if not re.fullmatch(r"sha256:[a-f0-9]{64}", image_id):
+            raise AgentError("Docker did not report a valid runtime image ID")
+        try:
+            raw = managed_read(self.root / "agent", "runtime-images.json")
+            prepared = json.loads(raw) if raw is not None else {}
+        except (OSError, ValueError):
+            raise AgentError("Runtime preparation records are unavailable") from None
+        if not isinstance(prepared, dict):
+            raise AgentError("Runtime preparation records are invalid")
+        # The controller also authorizes its configured default without a pull
+        # record. Whenever preparation was recorded, enforce that exact image.
+        if image in prepared and prepared[image] != image_id:
+            raise AgentError("Prepared runtime image has changed. An admin must prepare it again before launching.")
         files = payload.get("files", [])
         if not isinstance(files, list) or len(files) > 10000:
             raise Rejected("Invalid initial file list")
@@ -1656,7 +1669,7 @@ class Agent:
             total += len(self.decode_file(item))
             if total > 64 * 1024 * 1024:
                 raise Rejected("Initial files exceed the 64 MiB operation limit")
-        return game, library, image, port, bind, env
+        return game, library, image, image_id, port, bind, env
 
     @staticmethod
     def plutonium_manifest(data):
@@ -1890,7 +1903,7 @@ class Agent:
                 raise AgentError("Instance already exists with a different specification")
             return existing
         self.operation_phase = "validating-create-spec"
-        game, library, image, port, bind, env = self.validate_spec(payload)
+        game, library, image, image_id, port, bind, env = self.validate_spec(payload)
         app = directory / "app"
         self.operation_phase = "preparing-create-directory"
         app.mkdir(parents=True, exist_ok=True)
@@ -1905,15 +1918,15 @@ class Agent:
         atomic_bytes(env_path, "".join(f"{key}={value}\n" for key, value in env.items()).encode())
         name = self.container_name(instance_id, generation)
         spec_sha = digest(json.dumps(payload, sort_keys=True, separators=(",", ":")).encode())
-        # Do not let Docker fetch a replacement if the prepared image disappears
-        # between validation and container creation.
+        # Use the validated immutable ID so retagging after validation cannot
+        # substitute another image. A deleted image must not trigger a pull.
         args = ["container", "create", "--pull=never", "--name", name, "--restart", "unless-stopped", "--label", LABEL + ".managed=true", "--label", LABEL + ".instance=" + instance_id, "--label", LABEL + ".generation=" + str(generation), "--label", LABEL + ".operation=" + (self.executing_id or "direct"), "--label", LABEL + ".spec-sha256=" + spec_sha, "--mount", f"type=bind,src={app},dst=/home/plutainer/app", "--mount", f"type=bind,src={library / 'content'},dst=/home/plutainer/gamefiles,readonly", "--env-file", str(env_path)]
         if self.config.get("hostId"):
             args.extend(["--label", LABEL + ".host=" + identifier(self.config["hostId"])])
         bind_arg = "[" + bind + "]" if ":" in bind else bind
         for protocol in ("udp", "tcp"):
             args.extend(["--publish", f"{bind_arg}:{port}:{port}/{protocol}"])
-        args.append(image)
+        args.append(image_id)
         self.check_creation_cancellation()
         self.operation_phase = "creating-container"
         self.checkpoint({"phase": "creating", "generation": generation, "specSha256": spec_sha})