ge-aerospace/shopdb-flask
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

ADR-013 Phase 2: import guard fails closed on non-.py + owns package root

Browse Source 
Third review found the meta_path guard leaked exactly where it delegated to the
stdlib import system:

1. Non-.py submodules (CRITICAL). When a name had no dir and no .py, find_spec
   returned None and the stdlib loaded a planted .so (ExtensionFileLoader) or a
   sourceless .pyc unverified - an attacker deletes a signed .py and drops a
   same-named .so with arbitrary init code, run on a normal request via core's
   `from plugins.<name>.models import ...`. The guard now refuses any name for
   which a non-source importable candidate (EXTENSION_SUFFIXES + BYTECODE_
   SUFFIXES) exists on disk; None is reserved for genuinely-absent modules.

2. Top-level plugins/__init__.py (CRITICAL). It is in no plugin's provenance,
   is attacker-writable, and Python runs it before any guarded submodule. The
   guard now owns `plugins`: it execs an EMPTY package body (search points at
   the plugins dir), so an overwritten plugins/__init__.py never runs.

Also: specs are built with spec_from_file_location so loaded modules get
__file__/__path__ (Flask blueprint root paths need it) while the loader still
execs the verified in-memory bytes - never re-reading the file.

Verified end to end: under PLUGIN_REQUIRE_SIGNED with all 13 bundled plugins
stamped, the app boots and loads every plugin through the guard; a tampered
plugin file is refused at load. 4 new guard tests (planted .so, sourceless
.pyc, absent-module defer, neutralized package root). Prior fixes #3/#4
confirmed still sound by the review. 1065 pass, naming green.
This commit is contained in:
cproudlock
2026-07-18 22:12:20 -04:00
parent dd30ca0c3f
commit c59d2dab56
2 changed files with 119 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

104
shopdb/plugins/importguard.py
View File

@@ -20,6 +20,7 @@ is absent and imports behave exactly as before.
import hashlib
import importlib.abc
import importlib.machinery
import importlib.util
import json
import sys
@@ -27,6 +28,14 @@ from pathlib import Path
from . import signing
# Non-source files the stdlib import system would execute for a module name.
# The guard loads ONLY verified .py source, so any of these under a plugin is a
# module the guard cannot vouch for and must refuse (a planted .so or a
# sourceless .pyc standing in for a deleted, signed .py).
_UNVERIFIABLE_SUFFIXES = tuple(
importlib.machinery.EXTENSION_SUFFIXES
+ importlib.machinery.BYTECODE_SUFFIXES)
class PluginVerificationError(ImportError):
"""Raised when a plugins.* module is not covered by a trusted signature."""
@@ -51,6 +60,14 @@ class _VerifiedSourceLoader(importlib.abc.Loader):
def is_package(self, fullname):
return self._is_package
def get_filename(self, fullname):
# gives the loaded module a __file__ so Flask can resolve blueprint
# root paths (spec.origin is set from this).
return self._filepath
def get_source(self, fullname):
return self._source.decode('utf-8', 'replace')
class PluginImportGuard(importlib.abc.MetaPathFinder):
"""Verifies and loads every plugins.<name>.* module from signed bytes."""
@@ -81,25 +98,49 @@ class PluginImportGuard(importlib.abc.MetaPathFinder):
self._filemaps[name] = filemap
return filemap
def _module_file(self, name, fullname):
"""(filepath, is_package, search_locations) for a plugins.<name>.* module,
or (None, ...) when it is not a source module we should load."""
def _refuse_if_unverifiable_present(self, directory, leaf, name):
"""Fail closed if a non-.py importable file (.so / sourceless .pyc) for
`leaf` sits in `directory`. Reserving None for genuinely-absent names is
what stops the stdlib loading a planted extension/bytecode unverified."""
for suffix in _UNVERIFIABLE_SUFFIXES:
if (directory / f'{leaf}{suffix}').exists():
raise PluginVerificationError(
f'plugin {name} module {leaf}{suffix} is not loadable from '
f'trusted source; refusing')
def _source_spec(self, name, fullname, filepath, is_package, search):
plugin_dir = self.plugins_dir / name
tail = fullname.split('.')[2:] # components after plugins.<name>
base = plugin_dir.joinpath(*tail) if tail else plugin_dir
if base.is_dir():
return base / '__init__.py', True, [str(base)]
source = base.with_suffix('.py')
if source.exists():
return source, False, None
return None, False, None
relpath = filepath.relative_to(plugin_dir).as_posix()
source = filepath.read_bytes()
expected = self._filemap(name).get(relpath)
if expected is None or hashlib.sha256(source).hexdigest() != expected:
raise PluginVerificationError(
f'plugin {name} module {relpath} is not covered by a trusted '
f'signature')
loader = _VerifiedSourceLoader(filepath, source, is_package, search)
# spec_from_file_location sets origin -> the module gets __file__/__path__,
# which Flask needs for blueprint root paths; the loader still execs the
# verified in-memory bytes, never re-reading the file.
return importlib.util.spec_from_file_location(
fullname, str(filepath), loader=loader,
submodule_search_locations=search if is_package else None)
def find_spec(self, fullname, path=None, target=None):
# Only our package; the empty top-level `plugins` package loads normally.
if fullname != 'plugins' and not fullname.startswith('plugins.'):
return None
# The top-level `plugins` package is in no plugin's provenance, yet
# Python imports it before any guarded submodule. Own it: execute an
# EMPTY package body so an attacker-overwritten plugins/__init__.py never
# runs, and point package search at the plugins dir.
if fullname == 'plugins':
return None
init_path = self.plugins_dir / '__init__.py'
loader = _VerifiedSourceLoader(
init_path, b'', True, [str(self.plugins_dir)])
return importlib.util.spec_from_file_location(
'plugins', str(init_path), loader=loader,
submodule_search_locations=[str(self.plugins_dir)])
name = fullname.split('.')[1]
plugin_dir = self.plugins_dir / name
@@ -107,27 +148,28 @@ class PluginImportGuard(importlib.abc.MetaPathFinder):
if self.verifier._dev_exempt(plugin_dir):
return None
filemap = self._filemap(name) # verify signature (raises on failure)
filepath, is_package, search = self._module_file(name, fullname)
if filepath is None or not filepath.exists():
# A missing __init__.py (namespace pkg) or non-python target: let the
# normal machinery decide. Any .py it would run is covered above.
self._filemap(name) # verify the plugin signature (raises on failure)
tail = fullname.split('.')[2:] # components after plugins.<name>
base = plugin_dir.joinpath(*tail) if tail else plugin_dir
if base.is_dir():
init = base / '__init__.py'
if init.exists():
return self._source_spec(name, fullname, init, True, [str(base)])
# Regular package with no __init__.py: refuse a planted
# __init__.<ext>; otherwise treat as a namespace package (no body).
self._refuse_if_unverifiable_present(base, '__init__', name)
return None
relpath = filepath.relative_to(plugin_dir).as_posix()
source = filepath.read_bytes()
expected = filemap.get(relpath)
if expected is None or hashlib.sha256(source).hexdigest() != expected:
raise PluginVerificationError(
f'plugin {name} module {relpath} is not covered by a trusted '
f'signature')
source = base.with_suffix('.py')
if source.exists():
return self._source_spec(name, fullname, source, False, None)
loader = _VerifiedSourceLoader(filepath, source, is_package, search)
spec = importlib.util.spec_from_loader(
fullname, loader, is_package=is_package)
if is_package:
spec.submodule_search_locations = search
return spec
# No .py for this name: refuse a planted .so / sourceless .pyc, else defer
# (a genuinely-absent module -> normal ModuleNotFoundError).
self._refuse_if_unverifiable_present(base.parent, base.name, name)
return None
def verified_source(self, name, relpath):
"""Return hash-verified bytes of one plugin file (read once), for callers