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
+        # Use the validated immutable ID so retagging after validation cannot
-        # between validation and container creation.
+        # 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})

agent/tests/test_agent.py

@@ -24,6 +24,7 @@ class FakeDocker:
         self.fail_restart = False
         self.lose_reply_after = None
         self.started = 0
+        self.image_id = "sha256:" + "1" * 64
 
     def inspect(self, name):
         return self.containers.get(name)
@@ -34,6 +35,8 @@ class FakeDocker:
             return json.dumps({"ServerVersion": "27.5.1"})
         if args[:2] == ("image", "ls"):
             return "ghcr.io/ayymoss/plutainer:latest\n"
+        if args[:4] == ("image", "inspect", "--format", "{{.Id}}"):
+            return self.image_id
         if args[:2] == ("container", "create"):
             labels = {args[index + 1].split("=", 1)[0]: args[index + 1].split("=", 1)[1] for index, arg in enumerate(args) if arg == "--label"}
             name = args[args.index("--name") + 1]
@@ -989,7 +992,8 @@ class AgentTests(unittest.TestCase):
         payload = self.create()
         creation = next(args for args in self.docker.calls if args[:2] == ("container", "create"))
         self.assertIn("--pull=never", creation)
-        self.assertEqual(creation[-1], payload["image"])
+        self.assertEqual(creation[-1], self.docker.image_id)
+        self.assertEqual(self.agent.spec(INSTANCE, 1)["image"], payload["image"])
         self.assertFalse(any(args[:2] == ("image", "pull") for args in self.docker.calls))
 
     def test_image_disappearing_after_validation_cannot_be_pulled_or_started(self):
@@ -1008,6 +1012,88 @@ class AgentTests(unittest.TestCase):
         self.assertEqual(self.docker.started, 0)
         self.assertFalse(any(args[:2] == ("image", "pull") for args in self.docker.calls))
 
+    def test_retargeted_prepared_image_requires_a_new_admin_pull_before_creation(self):
+        image = "registry.example/custom-runtime:stable"
+        payload = {"game": "t5", "libraryId": self.library_id, "image": image, "port": 28960,
+                   "rconPassword": "rcon-secret", "env": {"PLUTAINER_KEY": "server-secret"}}
+        self.assertEqual(self.agent.execute(self.operation("runtime.pull", {"image": image}, "prepare"))["state"], "completed")
+        prepared_id = self.docker.image_id
+        self.docker.image_id = "sha256:" + "2" * 64
+        original = self.docker.call
+
+        def call(*args, **kwargs):
+            if args[:2] == ("image", "ls"):
+                # The old image still exists, so cached-ID inventory alone
+                # cannot prove that this mutable tag is safe to launch.
+                return (f"registry.example/custom-runtime|stable|<none>|{self.docker.image_id}\n"
+                        f"<none>|<none>|<none>|{prepared_id}\n")
+            return original(*args, **kwargs)
+
+        with patch.object(self.docker, "call", side_effect=call):
+            self.assertIn(image, self.agent.inventory()["preparedRuntimeImages"])
+            result = self.agent.execute(self.operation("instance.create", payload, "retargeted"))
+        self.assertEqual(result["state"], "failed")
+        self.assertIn("Prepared runtime image has changed", result["error"])
+        self.assertFalse(any(args[:2] == ("container", "create") for args in self.docker.calls))
+        self.assertEqual(self.docker.started, 0)
+        self.assertFalse((self.agent.instance_dir(INSTANCE, 1) / "environment").exists())
+
+        self.assertEqual(self.agent.execute(self.operation("runtime.pull", {"image": image}, "prepare-again"))["state"], "completed")
+        self.assertEqual(self.agent.execute(self.operation("instance.create", payload, "retry"))["state"], "completed")
+        creation = next(args for args in self.docker.calls if args[:2] == ("container", "create"))
+        self.assertEqual(creation[-1], self.docker.image_id)
+
+    def test_retag_after_validation_cannot_replace_the_image_used_for_creation(self):
+        image = "ghcr.io/ayymoss/plutainer:latest"
+        self.agent.dispatch(self.operation("runtime.pull", {"image": image}))
+        prepared_id = self.docker.image_id
+        original = self.agent.write_initial_files
+
+        def retag(*args):
+            self.docker.image_id = "sha256:" + "2" * 64
+            return original(*args)
+
+        with patch.object(self.agent, "write_initial_files", side_effect=retag):
+            payload = self.create()
+        creation = next(args for args in self.docker.calls if args[:2] == ("container", "create"))
+        self.assertIn("--pull=never", creation)
+        self.assertEqual(creation[-1], prepared_id)
+        self.assertNotEqual(creation[-1], self.docker.image_id)
+        self.assertEqual(self.agent.spec(INSTANCE, 1), payload)
+        # Retagging must not invalidate an already created immutable instance.
+        calls = list(self.docker.calls)
+        self.assertTrue(self.agent.create(INSTANCE, 1, payload)["exists"])
+        self.assertEqual(self.docker.calls, calls)
+
+    def test_rebuild_rejects_retargeted_image_before_stopping_existing_container(self):
+        image = "ghcr.io/ayymoss/plutainer:latest"
+        self.agent.dispatch(self.operation("runtime.pull", {"image": image}))
+        payload = self.create()
+        self.docker.image_id = "sha256:" + "2" * 64
+        self.docker.calls.clear()
+        result = self.agent.execute(self.operation("instance.rebuild", {"nextGeneration": 2, "spec": payload}, "rebuild"))
+        self.assertEqual(result["state"], "failed")
+        self.assertIn("Prepared runtime image has changed", result["error"])
+        self.assertFalse(any(args[:2] in (("container", "stop"), ("container", "rm"), ("container", "create")) for args in self.docker.calls))
+        self.assertTrue(self.docker.inspect(self.agent.container_name(INSTANCE, 1))["State"]["Running"])
+
+    def test_launch_rejects_invalid_image_ids_and_preparation_records(self):
+        payload = {"game": "t5", "libraryId": self.library_id, "image": "prepared:runtime", "port": 28960}
+        for image_id in ("", "prepared:runtime", "sha256:short"):
+            with self.subTest(image_id=image_id):
+                self.docker.image_id = image_id
+                with self.assertRaisesRegex(AgentError, "valid runtime image ID"):
+                    self.agent.create(INSTANCE, 1, payload)
+        self.docker.image_id = "sha256:" + "1" * 64
+        aliases = self.agent.root / "agent/runtime-images.json"
+        for raw in ("invalid json", "[]", "null"):
+            with self.subTest(records=raw):
+                aliases.write_text(raw)
+                with self.assertRaisesRegex(AgentError, "Runtime preparation records"):
+                    self.agent.create(INSTANCE, 1, payload)
+        self.assertFalse(any(args[:2] == ("container", "create") for args in self.docker.calls))
+        self.assertEqual(self.docker.started, 0)
+
     def test_manifest_is_advertised_and_reconciles_as_a_read_operation(self):
         self.create()
         operation = self.operation("files.manifest", {"roots": ["configs", "scripts"], "ignoreNames": [], "paths": ["configs/server.cfg"]})

docs/guides/native-hosts.md

@@ -253,6 +253,12 @@ still run the configured default, but must update before accepting custom runtim
 Queued launches and migrations never trigger custom image pulls, and container
 creation cannot implicitly pull a missing image.
 
+Agent 1.8.7 also checks a runtime's current image ID against its recorded preparation
+before creating or rebuilding a container. If another program retags a prepared
+image, an admin must prepare it again from the host page before launching it.
+Containers are created using the validated immutable image ID, so a tag change
+after validation cannot substitute another image. Existing containers keep running.
+
 ### Resource samples and learned allowances
 
 Only provisioning reserves a host and port. Agent **1.4.6** fixes container

static/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
+        # Use the validated immutable ID so retagging after validation cannot
-        # between validation and container creation.
+        # 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})

static/agent/manifest.json

@@ -1,8 +1,8 @@
 {
-  "version": "1.8.6",
+  "version": "1.8.7",
   "protocolVersion": 1,
   "platform": "debian-amd64",
-  "sha256": "79702dc8cf80d2521153bfda9241f7240285014fe746fc921f0a7793dd81212b",
+  "sha256": "e773c907141c97eab2055b354306973b248a67109a51337f1f1d76d80aaa8701",
   "updaterProtocolVersion": 1,
   "updaterSha256": "066bb896726600ece9ce72697ce523c9a707cc9a621505931ba0892dec59d3c8"
 }