fix: harden v2 migration and Flatpak updatesbundle-v2-flatpak-auto-update

6 files changed, 353 insertions, 39 deletions

diff --git a/README.md b/README.md
index 12a8ecc..c0642a4 100644
--- a/README.md
+++ b/README.md
@@ -80,6 +80,7 @@ The plugin:
 - Downloads checksum-pinned x86_64 or Armada/aarch64 lsfg-vk v2 assets from the owner-fork release to `~/.local/lib/`
 - Bundles checksum-pinned lsfg-vk v2 Flatpak extensions for runtimes 23.08, 24.08, and 25.08, and installs the selected local bundle
 - Updates already-installed user Flatpak runtimes and migrates the plugin's legacy app overrides when the native layer is updated
+- Records existing Flatpak runtime commits and rolls back already-updated runtimes when a multi-runtime migration fails, when Flatpak permits it
 - Configures the v2 Vulkan layer in `~/.local/share/vulkan/implicit_layer.d/`
 - Migrates the existing `~/.config/lsfg-vk/conf.toml` from v1 to v2 automatically on layer update, retaining a one-time `.v1.bak` backup
 - Automatically detects your Lossless Scaling DLL installation
diff --git a/py_modules/lsfg_vk/flatpak_service.py b/py_modules/lsfg_vk/flatpak_service.py
index fe965aa..8410496 100644
--- a/py_modules/lsfg_vk/flatpak_service.py
+++ b/py_modules/lsfg_vk/flatpak_service.py
@@ -91,6 +91,52 @@ class FlatpakService(BaseService):
         plugin_dir = Path(__file__).resolve().parent.parent.parent
         return plugin_dir / BIN_DIR / filename
 
+    def _get_extension_id(self, version: str) -> str:
+        """Return the full user-installation ref for a supported runtime branch."""
+        extension_ids = {
+            "23.08": self.extension_id_23_08,
+            "24.08": self.extension_id_24_08,
+            "25.08": self.extension_id_25_08,
+        }
+        try:
+            return extension_ids[version]
+        except KeyError as error:
+            raise ValueError(f"Unsupported Flatpak runtime version: {version}") from error
+
+    def _get_installed_extension_commit(self, version: str) -> str:
+        """Record the deployed commit so a failed multi-runtime migration can roll back."""
+        result = self._run_flatpak_command(
+            ["info", "--user", "--show-commit", self._get_extension_id(version)],
+            capture_output=True,
+            text=True,
+            check=True,
+        )
+        commit = str(result.stdout or "").strip()
+        if not commit:
+            raise OSError(f"Flatpak did not report a deployed commit for {version}")
+        return commit
+
+    def _rollback_extension(self, version: str, commit: str) -> Optional[str]:
+        """Restore one runtime to its previous deployed commit; return an error if it fails."""
+        try:
+            result = self._run_flatpak_command(
+                [
+                    "update",
+                    "--user",
+                    "--assumeyes",
+                    "--noninteractive",
+                    f"--commit={commit}",
+                    self._get_extension_id(version),
+                ],
+                capture_output=True,
+                text=True,
+            )
+        except (OSError, subprocess.SubprocessError) as error:
+            return str(error)
+        if result.returncode != 0:
+            return result.stderr or f"Flatpak rollback failed for {version}"
+        return None
+
     def _get_installed_extension_versions(self) -> list[str]:
         """Return installed lsfg-vk runtime branches from the user Flatpak installation."""
         result = self._run_flatpak_command(
@@ -271,7 +317,7 @@ class FlatpakService(BaseService):
     def _get_flatpak_app_ids(self) -> list[str]:
         """Return application IDs in the user Flatpak installation."""
         result = self._run_flatpak_command(
-            ["list", "--app"],
+            ["list", "--user", "--app"],
             capture_output=True,
             text=True,
             check=True,
@@ -348,6 +394,8 @@ class FlatpakService(BaseService):
         empty_result = {
             "updated_versions": [],
             "failed_versions": [],
+            "rolled_back_versions": [],
+            "rollback_failed_versions": [],
             "migrated_apps": [],
             "failed_apps": [],
         }
@@ -369,6 +417,37 @@ class FlatpakService(BaseService):
                     **empty_result,
                 )
 
+            missing_bundles = [
+                {"version": version, "error": f"Bundled Flatpak extension not found for {version}"}
+                for version in installed_versions
+                if not self._get_bundled_extension_path(version).is_file()
+            ]
+            if missing_bundles:
+                return self._error_response(
+                    BaseResponse,
+                    "Flatpak runtime migration was not started because a bundled asset is missing",
+                    skipped=False,
+                    failed_versions=missing_bundles,
+                    **{key: value for key, value in empty_result.items() if key != "failed_versions"},
+                )
+
+            previous_commits = {}
+            commit_failures = []
+            for version in installed_versions:
+                try:
+                    previous_commits[version] = self._get_installed_extension_commit(version)
+                except (subprocess.CalledProcessError, OSError) as error:
+                    stderr = getattr(error, "stderr", None)
+                    commit_failures.append({"version": version, "error": stderr or str(error)})
+            if commit_failures:
+                return self._error_response(
+                    BaseResponse,
+                    "Flatpak runtime migration was not started because existing commits could not be recorded",
+                    skipped=False,
+                    failed_versions=commit_failures,
+                    **{key: value for key, value in empty_result.items() if key != "failed_versions"},
+                )
+
             updated_versions = []
             failed_versions = []
             for version in installed_versions:
@@ -379,12 +458,29 @@ class FlatpakService(BaseService):
                     failed_versions.append({"version": version, "error": result.get("error", "unknown error")})
 
             if failed_versions:
+                rolled_back_versions = []
+                rollback_failed_versions = []
+                for version in reversed(updated_versions):
+                    rollback_error = self._rollback_extension(version, previous_commits[version])
+                    if rollback_error is None:
+                        rolled_back_versions.append(version)
+                    else:
+                        rollback_failed_versions.append({"version": version, "error": rollback_error})
+
+                rollback_message = (
+                    f"rolled back {len(rolled_back_versions)} runtime(s)"
+                    if not rollback_failed_versions
+                    else f"could not roll back {len(rollback_failed_versions)} runtime(s)"
+                )
                 return self._error_response(
                     BaseResponse,
-                    "Flatpak runtime migration failed; legacy app overrides were left unchanged",
+                    f"Flatpak runtime migration failed; {rollback_message}; "
+                    "legacy app overrides were left unchanged",
                     skipped=False,
                     updated_versions=updated_versions,
                     failed_versions=failed_versions,
+                    rolled_back_versions=rolled_back_versions,
+                    rollback_failed_versions=rollback_failed_versions,
                     migrated_apps=[],
                     failed_apps=[],
                 )
@@ -393,6 +489,8 @@ class FlatpakService(BaseService):
             response = {
                 "updated_versions": updated_versions,
                 "failed_versions": [],
+                "rolled_back_versions": [],
+                "rollback_failed_versions": [],
                 "migrated_apps": override_result.get("migrated_apps", []),
                 "failed_apps": override_result.get("failed_apps", []),
             }
@@ -428,12 +526,7 @@ class FlatpakService(BaseService):
             if not self.check_flatpak_available():
                 return self._error_response(BaseResponse, "Flatpak is not available on this system")
 
-            if version == "23.08":
-                extension_id = self.extension_id_23_08
-            elif version == "24.08":
-                extension_id = self.extension_id_24_08
-            else:
-                extension_id = self.extension_id_25_08
+            extension_id = self._get_extension_id(version)
 
             result = self._run_flatpak_command(
                 ["uninstall", "--user", "--noninteractive", extension_id],
diff --git a/py_modules/lsfg_vk/installation.py b/py_modules/lsfg_vk/installation.py
index 7eee848..1ba091a 100644
--- a/py_modules/lsfg_vk/installation.py
+++ b/py_modules/lsfg_vk/installation.py
@@ -45,10 +45,16 @@ class InstallationService(BaseService):
                 return self._error_response(InstallationResponse, error_msg, message="")
 
             self._ensure_directories()
-            self._extract_and_install_files(archive_path)
-            profile_data = self._create_config_file()
-            self._create_lsfg_launch_script(profile_data)
-            self._remove_legacy_layer_files()
+            with tempfile.TemporaryDirectory(prefix="lsfg-vk-install-") as transaction_dir:
+                snapshots = self._snapshot_installation_files(Path(transaction_dir))
+                try:
+                    self._extract_and_install_files(archive_path)
+                    profile_data = self._create_config_file()
+                    self._create_lsfg_launch_script(profile_data)
+                    self._remove_legacy_layer_files()
+                except Exception:
+                    self._restore_installation_files(snapshots)
+                    raise
 
             message = "lsfg-vk v2 installed successfully"
             if self._config_recovery_backup is not None:
@@ -126,12 +132,54 @@ class InstallationService(BaseService):
         """Point the manifest at ~/.local/lib after Decky installs it."""
         try:
             data = json.loads(source.read_text(encoding="utf-8"))
+            if not isinstance(data, dict) or not isinstance(data.get("layer"), dict):
+                raise ValueError("manifest is missing its layer object")
             data["layer"]["library_path"] = "../../../lib/liblsfg-vk-layer.so"
-            destination.write_text(json.dumps(data, indent=2) + "\n", encoding="utf-8")
-            destination.chmod(0o644)
-        except (json.JSONDecodeError, KeyError, OSError) as error:
+            self._write_file(destination, json.dumps(data, indent=2) + "\n", 0o644)
+        except (json.JSONDecodeError, OSError, TypeError, UnicodeError, ValueError) as error:
             self.log.error("Error fixing JSON file %s: %s", source, error)
-            self._copy_plugin_file(source, destination)
+            raise OSError(f"Invalid Vulkan layer manifest {source}: {error}") from error
+
+    def _installation_paths(self) -> tuple[Path, ...]:
+        """Return every file that can be changed by a native installation."""
+        return (
+            self.lib_file,
+            self.json_file,
+            self.config_file_path,
+            self.lsfg_launch_script_path,
+            self.legacy_lib_file,
+            self.legacy_json_file,
+            self.config_file_path.with_name(f"{self.config_file_path.name}.v1.bak"),
+            self.config_file_path.with_name(f"{self.config_file_path.name}.unrecognized.bak"),
+        )
+
+    def _snapshot_installation_files(self, transaction_dir: Path) -> Dict[Path, Path | None]:
+        """Snapshot installation files so a failed update can be rolled back."""
+        snapshots: Dict[Path, Path | None] = {}
+        for index, path in enumerate(self._installation_paths()):
+            if not path.exists() and not path.is_symlink():
+                snapshots[path] = None
+                continue
+            if not path.is_file():
+                raise OSError(f"Installation target is not a regular file: {path}")
+            backup = transaction_dir / str(index)
+            shutil.copy2(path, backup)
+            snapshots[path] = backup
+        return snapshots
+
+    def _restore_installation_files(self, snapshots: Dict[Path, Path | None]) -> None:
+        """Restore a previous installation after a failed update."""
+        try:
+            for path, backup in snapshots.items():
+                if path.exists() or path.is_symlink():
+                    path.unlink()
+                if backup is not None:
+                    path.parent.mkdir(parents=True, exist_ok=True)
+                    shutil.copy2(backup, path)
+            self.log.warning("Rolled back incomplete lsfg-vk installation")
+        except OSError as error:
+            self.log.error("Could not fully roll back lsfg-vk installation: %s", error)
+            raise
 
     def _create_config_file(self) -> ProfileData:
         """Migrate v1 once, normalize v2, and retain Decky launch workarounds."""
@@ -222,7 +270,17 @@ class InstallationService(BaseService):
     def _merge_config_with_defaults(self, existing: ProfileData, dll_service) -> ProfileData:
         defaults = ConfigurationManager.get_defaults_with_dll_detection(dll_service)
         global_config = dict(existing.get("global_config", {}))
-        global_config.setdefault("dll", defaults["dll"])
+        configured_dll = str(global_config.get("dll", "") or "").strip()
+        detected_dll = str(defaults.get("dll", "") or "").strip()
+        if configured_dll and Path(configured_dll).is_file():
+            global_config["dll"] = configured_dll
+        elif detected_dll and Path(detected_dll).is_file():
+            self.log.warning("Replacing stale configured Lossless.dll path %s with %s", configured_dll, detected_dll)
+            global_config["dll"] = detected_dll
+        else:
+            if configured_dll:
+                self.log.warning("Clearing stale configured Lossless.dll path %s", configured_dll)
+            global_config["dll"] = ""
         global_config.setdefault("allow_fp16", defaults["allow_fp16"])
 
         profiles: Dict[str, Any] = {}
diff --git a/src/api/lsfgApi.ts b/src/api/lsfgApi.ts
index f112d39..07f4b72 100644
--- a/src/api/lsfgApi.ts
+++ b/src/api/lsfgApi.ts
@@ -17,6 +17,8 @@ export interface FlatpakMigrationResult {
   skipped?: boolean;
   updated_versions: string[];
   failed_versions: Array<{ version: string; error: string }>;
+  rolled_back_versions: string[];
+  rollback_failed_versions: Array<{ version: string; error: string }>;
   migrated_apps: string[];
   failed_apps: Array<{ app_id: string; error: string }>;
 }
diff --git a/src/components/InstallationButton.tsx b/src/components/InstallationButton.tsx
index 40ba955..de7d454 100644
--- a/src/components/InstallationButton.tsx
+++ b/src/components/InstallationButton.tsx
@@ -16,15 +16,24 @@ export function InstallationButton({
   onInstall,
   onUninstall
 }: InstallationButtonProps) {
-  const renderButtonContent = () => {
+  const renderInstallButtonContent = () => {
     if (isInstalling) {
       return (
         <div style={{ display: "flex", alignItems: "center", gap: "8px" }}>
-          <div>Installing...</div>
+          <div>{isInstalled ? "Updating..." : "Installing..."}</div>
         </div>
       );
     }
 
+    return (
+      <div style={{ display: "flex", alignItems: "center", gap: "8px" }}>
+        <FaDownload />
+        <div>{isInstalled ? "Update LSFG-VK" : "Install / Update LSFG-VK"}</div>
+      </div>
+    );
+  };
+
+  const renderUninstallButtonContent = () => {
     if (isUninstalling) {
       return (
         <div style={{ display: "flex", alignItems: "center", gap: "8px" }}>
@@ -33,32 +42,37 @@ export function InstallationButton({
       );
     }
 
-    if (isInstalled) {
-      return (
-        <div style={{ display: "flex", alignItems: "center", gap: "8px" }}>
-          <FaTrash />
-          <div>Uninstall LSFG-VK</div>
-        </div>
-      );
-    }
-
     return (
       <div style={{ display: "flex", alignItems: "center", gap: "8px" }}>
-        <FaDownload />
-        <div>Install / Update LSFG-VK</div>
+        <FaTrash />
+        <div>Uninstall LSFG-VK</div>
       </div>
     );
   };
 
   return (
-    <PanelSectionRow>
-      <ButtonItem
-        layout="below"
-        onClick={isInstalled ? onUninstall : onInstall}
-        disabled={isInstalling || isUninstalling}
-      >
-        {renderButtonContent()}
-      </ButtonItem>
-    </PanelSectionRow>
+    <>
+      <PanelSectionRow>
+        <ButtonItem
+          layout="below"
+          onClick={onInstall}
+          disabled={isInstalling || isUninstalling}
+        >
+          {renderInstallButtonContent()}
+        </ButtonItem>
+      </PanelSectionRow>
+
+      {isInstalled && (
+        <PanelSectionRow>
+          <ButtonItem
+            layout="below"
+            onClick={onUninstall}
+            disabled={isInstalling || isUninstalling}
+          >
+            {renderUninstallButtonContent()}
+          </ButtonItem>
+        </PanelSectionRow>
+      )}
+    </>
   );
 }
diff --git a/tests/test_lsfg_v2_migration.py b/tests/test_lsfg_v2_migration.py
index ee1fcba..8b2e2a3 100644
--- a/tests/test_lsfg_v2_migration.py
+++ b/tests/test_lsfg_v2_migration.py
@@ -192,6 +192,95 @@ class InstallerInfrastructureTests(unittest.TestCase):
             self.assertIn("version = 2", config_path.read_text(encoding="utf-8"))
             self.assertEqual(service._config_recovery_backup, config_dir / "conf.toml.unrecognized.bak")
 
+    def test_stale_dll_path_is_replaced_when_detection_finds_a_valid_file(self):
+        from lsfg_vk.installation import InstallationService
+
+        with tempfile.TemporaryDirectory() as directory:
+            home = Path(directory)
+            config_dir = home / ".config" / "lsfg-vk"
+            config_dir.mkdir(parents=True)
+            config_path = config_dir / "conf.toml"
+            detected_dll = home / "Lossless.dll"
+            detected_dll.write_bytes(b"dll")
+            profile_data = {
+                "current_profile": "decky-lsfg-vk",
+                "profiles": {"decky-lsfg-vk": ConfigurationManager.get_defaults()},
+                "global_config": {"dll": str(home / "missing" / "Lossless.dll"), "allow_fp16": True},
+            }
+            config_path.write_text(
+                ConfigurationManager.generate_toml_content_multi_profile(profile_data),
+                encoding="utf-8",
+            )
+
+            service = InstallationService.__new__(InstallationService)
+            service.log = mock.Mock()
+            service.user_home = home
+            service.config_dir = config_dir
+            service.config_file_path = config_path
+            service.lsfg_script_path = home / "lsfg"
+            service.lsfg_launch_script_path = home / "lsfg"
+
+            with mock.patch("lsfg_vk.dll_detection.DllDetectionService") as detection_service:
+                detection_service.return_value.check_lossless_scaling_dll.return_value = {
+                    "detected": True,
+                    "path": str(detected_dll),
+                }
+                migrated = service._create_config_file()
+
+            self.assertEqual(migrated["global_config"]["dll"], str(detected_dll))
+            self.assertIn(str(detected_dll), config_path.read_text(encoding="utf-8"))
+
+    def test_invalid_layer_manifest_fails_without_copying_unmodified_json(self):
+        from lsfg_vk.installation import InstallationService
+
+        with tempfile.TemporaryDirectory() as directory:
+            source = Path(directory) / "manifest.json"
+            destination = Path(directory) / "installed.json"
+            source.write_text('{"not_layer": true}', encoding="utf-8")
+
+            service = InstallationService.__new__(InstallationService)
+            service.log = mock.Mock()
+
+            with self.assertRaises(OSError):
+                service._copy_and_fix_json_file(source, destination)
+            self.assertFalse(destination.exists())
+
+    def test_installation_snapshot_restores_previous_files(self):
+        from lsfg_vk.installation import InstallationService
+
+        with tempfile.TemporaryDirectory() as directory:
+            home = Path(directory)
+            service = InstallationService.__new__(InstallationService)
+            service.log = mock.Mock()
+            service.local_lib_dir = home / "lib"
+            service.local_share_dir = home / "share"
+            service.config_file_path = home / "config" / "conf.toml"
+            service.lsfg_launch_script_path = home / "lsfg"
+            service.legacy_lib_file = service.local_lib_dir / "liblsfg-vk.so"
+            service.legacy_json_file = service.local_share_dir / "legacy.json"
+            service.lib_file = service.local_lib_dir / "liblsfg-vk-layer.so"
+            service.json_file = service.local_share_dir / "manifest.json"
+            service.config_file_path.parent.mkdir(parents=True)
+            service.local_lib_dir.mkdir(parents=True)
+            service.local_share_dir.mkdir(parents=True)
+            service.lib_file.write_bytes(b"old layer")
+            service.json_file.write_text("old manifest", encoding="utf-8")
+            service.config_file_path.write_text("old config", encoding="utf-8")
+            service.lsfg_launch_script_path.write_text("old script", encoding="utf-8")
+
+            with tempfile.TemporaryDirectory() as transaction_dir:
+                snapshots = service._snapshot_installation_files(Path(transaction_dir))
+                service.lib_file.write_bytes(b"new layer")
+                service.json_file.write_text("new manifest", encoding="utf-8")
+                service.config_file_path.unlink()
+                service.lsfg_launch_script_path.write_text("new script", encoding="utf-8")
+                service._restore_installation_files(snapshots)
+
+            self.assertEqual(service.lib_file.read_bytes(), b"old layer")
+            self.assertEqual(service.json_file.read_text(encoding="utf-8"), "old manifest")
+            self.assertEqual(service.config_file_path.read_text(encoding="utf-8"), "old config")
+            self.assertEqual(service.lsfg_launch_script_path.read_text(encoding="utf-8"), "old script")
+
     def test_writes_replace_the_config_atomically(self):
         service = BaseService.__new__(BaseService)
         service.log = mock.Mock()
@@ -246,6 +335,10 @@ class FlatpakMigrationTests(unittest.TestCase):
         service.log = mock.Mock()
         service.check_flatpak_available = mock.Mock(return_value=True)
         service._get_installed_extension_versions = mock.Mock(return_value=["23.08", "25.08"])
+        bundle = mock.Mock()
+        bundle.is_file.return_value = True
+        service._get_bundled_extension_path = mock.Mock(return_value=bundle)
+        service._get_installed_extension_commit = mock.Mock(side_effect=lambda version: f"old-{version}")
         service.install_extension = mock.Mock(
             side_effect=[{"success": True}, {"success": True}]
         )
@@ -298,6 +391,10 @@ class FlatpakMigrationTests(unittest.TestCase):
         service.log = mock.Mock()
         service.check_flatpak_available = mock.Mock(return_value=True)
         service._get_installed_extension_versions = mock.Mock(return_value=["24.08"])
+        bundle = mock.Mock()
+        bundle.is_file.return_value = True
+        service._get_bundled_extension_path = mock.Mock(return_value=bundle)
+        service._get_installed_extension_commit = mock.Mock(return_value="old-24.08")
         service.install_extension = mock.Mock(
             return_value={"success": False, "error": "bundle unavailable"}
         )
@@ -309,6 +406,55 @@ class FlatpakMigrationTests(unittest.TestCase):
         self.assertEqual(result["failed_versions"][0]["version"], "24.08")
         service._migrate_legacy_app_overrides.assert_not_called()
 
+    def test_failed_runtime_update_rolls_back_previously_updated_runtimes(self):
+        from lsfg_vk.flatpak_service import FlatpakService
+
+        service = FlatpakService.__new__(FlatpakService)
+        service.log = mock.Mock()
+        service.check_flatpak_available = mock.Mock(return_value=True)
+        service._get_installed_extension_versions = mock.Mock(return_value=["23.08", "24.08"])
+        bundle = mock.Mock()
+        bundle.is_file.return_value = True
+        service._get_bundled_extension_path = mock.Mock(return_value=bundle)
+        service._get_installed_extension_commit = mock.Mock(
+            side_effect=lambda version: f"old-{version}"
+        )
+        service.install_extension = mock.Mock(
+            side_effect=[
+                {"success": True},
+                {"success": False, "error": "transaction failed"},
+            ]
+        )
+        service._rollback_extension = mock.Mock(return_value=None)
+        service._migrate_legacy_app_overrides = mock.Mock()
+
+        result = service.update_installed_extensions()
+
+        self.assertFalse(result["success"])
+        self.assertEqual(result["rolled_back_versions"], ["23.08"])
+        service._rollback_extension.assert_called_once_with("23.08", "old-23.08")
+        service._migrate_legacy_app_overrides.assert_not_called()
+
+    def test_flatpak_app_scan_is_scoped_to_user_installation(self):
+        from lsfg_vk.flatpak_service import FlatpakService
+
+        service = FlatpakService.__new__(FlatpakService)
+        service._run_flatpak_command = mock.Mock(
+            return_value=types.SimpleNamespace(
+                returncode=0,
+                stdout="Game\torg.example.Game\t1.0\tx86_64\n",
+                stderr="",
+            )
+        )
+
+        self.assertEqual(service._get_flatpak_app_ids(), ["org.example.Game"])
+        service._run_flatpak_command.assert_called_once_with(
+            ["list", "--user", "--app"],
+            capture_output=True,
+            text=True,
+            check=True,
+        )
+
     def test_native_install_runs_flatpak_migration_after_success(self):
         from lsfg_vk.plugin import Plugin