Look for bad get_current_user scripts (#1370)

* initial script to find get_current_user calls

* handle pre and post scripts as well

* refactor

* omit absolute root dir path when printing

* avoid printing NS

* consolidate lxml parsing in one place

---------

Co-authored-by: burnettk <burnettk@users.noreply.github.com>
This commit is contained in:
Kevin Burnett 2024-04-10 16:44:13 +00:00 committed by GitHub
parent 2bec24aa63
commit f2c982a96c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 81 additions and 10 deletions

View File

@ -112,7 +112,7 @@ pre-commit:
$(IN_ARENA) poetry run pre-commit run --verbose --all-files
ruff:
$(IN_ARENA) poetry run ruff --fix spiffworkflow-backend
$(IN_ARENA) poetry run ruff check --fix spiffworkflow-backend
run-pyl: fe-lint-fix ruff pre-commit be-mypy be-tests-par
@true

View File

@ -24,7 +24,7 @@ function get_python_dirs() {
function run_autofixers() {
python_dirs="$(get_python_dirs) bin"
# shellcheck disable=2086
poetry run ruff --fix $python_dirs || echo ''
poetry run ruff check --fix $python_dirs || echo ''
}
function run_pre_commmit() {

View File

@ -0,0 +1,70 @@
import glob
import os
from typing import NoReturn
from lxml import etree
from spiffworkflow_backend import create_app
from spiffworkflow_backend.services.file_system_service import FileSystemService
def find_script_tasks_with_get_current_user(bpmn_file_path: str, root_path: str) -> None:
from spiffworkflow_backend.services.spec_file_service import SpecFileService
try:
with open(bpmn_file_path, "rb") as bpmn_file:
binary_data = bpmn_file.read()
tree = SpecFileService.get_etree_from_xml_bytes(binary_data)
check_script_and_prescript_elements(tree, bpmn_file_path, root_path)
except etree.XMLSyntaxError:
print(f"Error parsing XML in file {bpmn_file_path}. Please check for syntax issues.")
return
def check_script_and_prescript_elements(tree, bpmn_file_path: str, root_path: str) -> None:
# Define the namespace map to search for elements
nsmap = {
"bpmn": "http://www.omg.org/spec/BPMN/20100524/MODEL",
"spiffworkflow": "http://spiffworkflow.org/bpmn/schema/1.0/core",
}
# Find all script tasks and preScript elements
script_tasks = tree.xpath("//bpmn:scriptTask", namespaces=nsmap)
pre_scripts = tree.xpath("//spiffworkflow:preScript", namespaces=nsmap)
# Check script tasks for get_current_user() calls
for task in script_tasks:
script = task.find("bpmn:script", namespaces=nsmap)
if script is not None and script.text is not None and "get_current_user()" in script.text:
print(f'Found get_current_user() in script task {task.get("id")} of file {bpmn_file_path}')
# Check preScript elements for get_current_user() calls
check_scripts_for_get_current_user(pre_scripts, bpmn_file_path, "preScript", root_path)
post_scripts = tree.xpath("//spiffworkflow:postScript", namespaces=nsmap)
check_scripts_for_get_current_user(post_scripts, bpmn_file_path, "postScript", root_path)
def check_scripts_for_get_current_user(scripts, bpmn_file_path: str, script_type: str, root_path: str) -> None:
for script in scripts:
if script is not None and script.text is not None and "get_current_user()" in script.text:
# Get the parent of the parent to find the actual BPMN element
parent = script.getparent().getparent()
relative_path = os.path.relpath(bpmn_file_path, root_path)
tag_without_namespace = etree.QName(parent.tag).localname
if tag_without_namespace not in ["manualTask", "userTask"]:
print(
f"Found get_current_user() in {script_type} of {tag_without_namespace} "
f'with id {parent.get("id")} in file {relative_path}'
)
def main() -> NoReturn:
app = create_app()
with app.app_context():
hot_dir = FileSystemService.root_path()
# Search for BPMN files and check for get_current_user() calls in script tasks
bpmn_files = glob.glob(os.path.expanduser(f"{hot_dir}/**/*.bpmn"), recursive=True)
for bpmn_file in bpmn_files:
find_script_tasks_with_get_current_user(bpmn_file, hot_dir)
if __name__ == "__main__":
main()

View File

@ -29,6 +29,7 @@ from spiffworkflow_backend.scripts.script import Script
from spiffworkflow_backend.services.custom_parser import MyCustomParser
from spiffworkflow_backend.services.jinja_service import JinjaHelpers
from spiffworkflow_backend.services.process_instance_processor import CustomScriptEngineEnvironment
from spiffworkflow_backend.services.spec_file_service import SpecFileService
DEFAULT_NSMAP = {
"bpmn": "http://www.omg.org/spec/BPMN/20100524/MODEL",
@ -283,11 +284,9 @@ class ProcessModelTestRunnerMostlyPureSpiffDelegate(ProcessModelTestRunnerDelega
parser.add_dmn_files_by_glob(dmn_file_glob)
def _get_etree_from_bpmn_file(self, bpmn_file: str) -> etree._Element:
data = None
with open(bpmn_file, "rb") as f_handle:
data = f_handle.read()
etree_xml_parser = etree.XMLParser(resolve_entities=False)
return etree.fromstring(data, parser=etree_xml_parser) # noqa: S320
return SpecFileService.get_etree_from_xml_bytes(data)
def _find_related_bpmn_files(self, bpmn_file: str) -> list[str]:
related_bpmn_files = []

View File

@ -56,15 +56,17 @@ class SpecFileService(FileSystemService):
@classmethod
def get_references_for_file(cls, file: File, process_model_info: ProcessModelInfo) -> list[Reference]:
full_file_path = SpecFileService.full_file_path(process_model_info, file.name)
file_contents: bytes = b""
with open(full_file_path) as f:
file_contents = f.read().encode()
full_file_path = cls.full_file_path(process_model_info, file.name)
with open(full_file_path, "rb") as f:
file_contents = f.read()
return cls.get_references_for_file_contents(process_model_info, file.name, file_contents)
# This is designed to isolate xml parsing, which is a security issue, and make it as safe as possible.
# S320 indicates that xml parsing with lxml is unsafe. To mitigate this, we add options to the parser
# to make it as safe as we can. No exploits have been demonstrated with this parser, but we will try to stay alert.
@classmethod
def get_etree_from_xml_bytes(cls, binary_data: bytes) -> etree.Element:
etree_xml_parser = etree.XMLParser(resolve_entities=False)
etree_xml_parser = etree.XMLParser(resolve_entities=False, remove_comments=True, no_network=True)
return etree.fromstring(binary_data, parser=etree_xml_parser) # noqa: S320
@classmethod