from io import StringIO from pathlib import Path import json import itertools import yaml import jsonschema from typing import Any, Dict, List, Literal, NotRequired, Optional, TypedDict import requests try: from yachalk import chalk yachalk_imported = True except ModuleNotFoundError: yachalk_imported = False dataset_path = Path('dataset') output_path = Path('pages') dataset_info = dataset_path / Path('dataset.json') token = "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" def error(msg: str) -> Exception: print(chalk.red(msg) if yachalk_imported else "Error: {}".format(msg)) return Exception(msg) def warning(msg: str): print(chalk.yellow(msg) if yachalk_imported else "Warning: {}".format(msg)) class License(TypedDict): key: str name: str spdx_id: str url: str node_id: str class Permissions(TypedDict): admin: bool maintain: bool push: bool triage: bool pull: bool class Owner(TypedDict): login: str id: int node_id: str avatar_url: str gravatar_id: str url: str html_url: str followers_url: str following_url: str gists_url: str starred_url: str subscriptions_url: str organizations_url: str repos_url: str events_url: str received_events_url: str type: str site_admin: bool name: NotRequired[str] company: NotRequired[Optional[str]] blog: NotRequired[str] location: NotRequired[Optional[str]] email: NotRequired[Optional[str]] hireable: NotRequired[Optional[bool]] bio: NotRequired[Optional[str]] twitter_username: NotRequired[Optional[str]] public_repos: NotRequired[int] public_gists: NotRequired[int] followers: NotRequired[int] following: NotRequired[int] created_at: NotRequired[str] updated_at: NotRequired[str] class GithubRepositoryInformation(TypedDict): id: int node_id: str name: str full_name: str private: bool owner: Owner html_url: str description: Optional[str] fork: bool url: str forks_url: str keys_url: str collaborators_url: str teams_url: str hooks_url: str issue_events_url: str events_url: str assignees_url: str branches_url: str tags_url: str blobs_url: str git_tags_url: str git_refs_url: str trees_url: str statuses_url: str languages_url: str stargazers_url: str contributors_url: str subscribers_url: str subscription_url: str commits_url: str git_commits_url: str comments_url: str issue_comment_url: str contents_url: str compare_url: str merges_url: str archive_url: str downloads_url: str issues_url: str pulls_url: str milestones_url: str notifications_url: str labels_url: str releases_url: str deployments_url: str created_at: str updated_at: str pushed_at: str git_url: str ssh_url: str clone_url: str svn_url: str homepage: Optional[str] size: int stargazers_count: int watchers_count: int language: str has_issues: bool has_projects: bool has_downloads: bool has_wiki: bool has_pages: bool forks_count: int mirror_url: None archived: bool disabled: bool open_issues_count: int license: Optional[License] allow_forking: bool is_template: bool web_commit_signoff_required: bool topics: List[str] visibility: str forks: int open_issues: int watchers: int default_branch: str permissions: Permissions temp_clone_token: str organization: NotRequired[Owner] network_count: int subscribers_count: int class ModelInformation(TypedDict): title: NotRequired[str] slug: str branch: NotRequired[str] data: GithubRepositoryInformation owner: Owner stars: int forks: int owner_name: str owner_slug: str s: int e: int i: int a: int t: int l: int tech: List[str] Dataset = dict[str, ModelInformation] def open_dataset() -> Dataset: with open(dataset_info, 'r') as f: return json.load(f) def save_dataset(dataset: Dataset): with open(dataset_info, 'w') as f: json.dump(dataset, f, indent=4) def get_json(uri: str): print(uri) resp = requests.get(url=uri, headers={"Authorization": f"Bearer {token}"}) print(resp) if not resp.ok: try: resp_error = resp.json()['message'] except Exception: resp_error = resp.text raise Exception(f"Invalid response: {resp_error}") return resp.json() def get_repo(slug: str): return get_json(f"https://api.github.com/repos/{slug}") def get_user(name: str): return get_json(f"https://api.github.com/users/{name}") def get_file(slug: str, path: str): return get_json(f"https://api.github.com/repos/{slug}/contents/{path}") def plural(amount: int, name: str, plural: str = 's'): return f"{amount} {name}{plural[:amount^1]}" from typing import TypedDict class Artifact(TypedDict): file: str lines: NotRequired[list[int]] repository: NotRequired[str] RuleStatus = Literal["disregarded", "observed", "not applicable", "unknown"] class SecurityRule(TypedDict): status: RuleStatus argument: str | list[str] artifacts: NotRequired[list[Artifact]] rule_schema = yaml.safe_load("""type: object additionalProperties: no required: - status - argument properties: status: type: string enum: - disregarded - observed - not applicable - unknown argument: anyOf: - type: string - type: array items: type: string artifacts: type: array items: additionalProperties: no required: - file type: object properties: file: type: string repository: type: string lines: type: array items: type: integer""") def check_security_rules(security_rules: dict[Any, Any] | None) -> dict[int, SecurityRule]: if security_rules is None: raise Exception("Security rules file is empty!") for n in range(1, 19): try: rule = security_rules.get(n, None) if rule is None: raise jsonschema.ValidationError(f"Rule {n} is not evaluated") jsonschema.validate(rule, rule_schema) rule: SecurityRule if rule["status"] == "unknown": warning(f"Rule {n} is still unknown!") except jsonschema.ValidationError as e: warning("Not checking further rules!") raise Exception("Security rule {n}: {msg} at $.{n}.{path}".format(n=n, msg=e.message, path=e.json_path)) from e return dict(sorted(security_rules.items())) update_dataset = False def get_name(slug: str): return slug[slug.find('/')+1:] def get_tag_slug(tag: str) -> str: return tag.lower().replace(' ', '_') rule_names = { 1: "API Gateway", 2: "Mutual Authentication", 3: "Decoupled Authentication", 4: "Internal Identity Represenation", 5: "Authentication Token Validation", 6: "Login Rate Limiting", 7: "Edge Encryption", 8: "Internal Encryption", 9: "Central Logging Subsystem", 10: "Local Logging Agent", 11: "Log Sanitization", 12: "Log Message Broker", 13: "Circuit Breaker", 14: "Load Balancing", 15: "Service Mesh Usage Limits", 16: "Service Registry Deployment", 17: "Service Registry Validation", 18: "Secret Manager", } def artifact_to_string(info: ModelInformation, artifact: Artifact): file = Path(artifact['file']) filename = file.name file_url = f"https://github.com/{artifact.get('repository', info['slug'])}/blob/{info.get('branch', 'master')}/{artifact['file']}" lines = artifact.get("lines") if lines is None: return f"- {filename}: [File]({file_url})" return f"- {filename}: Line{'s'[:len(lines)^1]}: {', '.join(f'[{line}]({file_url}#L{line})' for line in lines)}" def rule_to_string(info: ModelInformation, id: int, rule: SecurityRule | None): if rule is None: warning(f"Rule {id} is missing!") return "" argument = rule['argument'] argument = argument if isinstance(argument, str) else "".join(f"\n1. {arg}" for arg in argument) text = f"""#### Rule {id}: {rule_names[id]} {{#rule{id:02}}} This rule is {rule['status']}: {argument}""" artifacts = rule.get("artifacts", []) if len(artifacts) > 0: text = text + f""" Artifacts: {chr(10).join(artifact_to_string(info, artifact) for artifact in artifacts)}""" return text def write_security_rules(info: ModelInformation, security_rules: dict[int, SecurityRule]): icons: Dict[RuleStatus | str, str] = { 'disregarded': '', 'observed': '', 'not applicable': '', 'unknown': '', } return f"""## Security Rules {" | ".join(f"R{i}" for i in range(1, 19))} {" | ".join("--" for _ in range(1, 19))} {" | ".join(f'{icons[security_rules.get(i, {"status": "unknown"})["status"]]}' for i in range(1, 19))} ### Authentication / Authorization {(chr(10)*2).join(rule_to_string(info, i, security_rules.get(i)) for i in range(1, 7))} ### Encryption {(chr(10)*2).join(rule_to_string(info, i, security_rules.get(i)) for i in range(7, 9))} ### Logging {(chr(10)*2).join(rule_to_string(info, i, security_rules.get(i)) for i in range(9, 13))} ### Availability {(chr(10)*2).join(rule_to_string(info, i, security_rules.get(i)) for i in range(13, 16))} ### Service Registry {(chr(10)*2).join(rule_to_string(info, i, security_rules.get(i)) for i in range(16, 18))} ### Secret Management {(chr(10)*2).join(rule_to_string(info, i, security_rules.get(i)) for i in range(18, 19))}""" def write_file_if_changed(file: Path, content: str, encoding: str = "utf-8"): old_content = None if file.exists(): with file.open('r', encoding=encoding) as f: old_content = f.read() if old_content is None or old_content != content: print(f"Writing changed file: {file}") with file.open('w', encoding=encoding) as f: f.write(content) def write_model_readmes(dataset: Dataset): for model_id, info in dataset.items(): dir = output_path / 'dataset' readme = dir / f'{model_id}.md' slug = info['slug'] data = info.get('data') if not data: data = get_repo(slug) info['data'] = data owner_url = data.get('owner', {}).get('url') if not owner_url: raise Exception(f'No owner in repo {slug}!') owner = info.get('owner') if not owner: owner = get_json(owner_url) info['owner'] = owner owner_name = owner.get('name') if not owner_name: raise Exception(f'No owner name in repo {slug}!') stars = data['stargazers_count'] forks = data['forks'] owner_slug = owner['login'] info['stars'] = stars info['forks'] = forks info['owner_name'] = owner_name info['owner_slug'] = owner_slug model_path = dataset_path / model_id security_rules_file = model_path / 'security_rules.yaml' model_file = model_path / f"{model_id}.py" with model_file.open("r") as f: model = f.read() security_rules = None try: with security_rules_file.open('r') as f: security_rules = check_security_rules(yaml.safe_load(f)) except FileNotFoundError: warning("Security rules file not found at {}".format(security_rules_file)) except Exception as e: warning("Security rules file at {} is invalid: {}".format(security_rules_file, e)) dir.mkdir(exist_ok=True) write_file_if_changed(readme, f"""--- title: {slug} keywords: model TODO tags: [{', '.join(get_tag_slug(tech) for tech in info['tech'])}] sidebar: datasetdoc_sidebar permalink: {model_id}.html toc: false --- ## Repository Information Repository: [GitHub](https://github.com/{slug}) Owner: [{owner_name}](https://github.com/{owner_slug}) The repository has {plural(stars, 'star')} and was forked {plural(forks, 'time')}. The codebase consists of {plural(info['l'], 'line')} of code and makes use of the following technologies: {chr(10).join(f'{tech}' for tech in info['tech'])} ## Data Flow Diagram ### DFD Model {{% include note.html content="Download the [model file](../../dataset/{model_id}/{model_id}.py)" %}} The images below were generated by executing the model file. The DFD is represented as a CodeableModels file. ```python {model} ``` ### Statistics The Application consists of a total of {plural(info['t'], 'element')}: Element | Count -- | -- Services | {info['s']} External Entities | {info['e']} Information Flows | {info['i']} Annotations | {info['a']} Total Items | {info['t']} ### Diagram Formats: - [PlantUML Model](../../dataset/{model_id}/{model_id}/{model_id}.txt) - [SVG Vector Image](../../dataset/{model_id}/{model_id}/{model_id}.svg) - [PNG Raster Image](../../dataset/{model_id}/{model_id}/{model_id}.png) ![Data Flow Diagram](../../dataset/{model_id}/{model_id}/{model_id}.svg) {"" if security_rules is None else write_security_rules(info, security_rules)} """) def write_root_readme(dataset: Dataset): overview_dir = output_path / 'overview' index_file = Path('index.md') write_file_if_changed(index_file, f"""--- title: code2DFD Documentation keywords: code2DFD introduction tags: [overview] sidebar: datasetdoc_sidebar permalink: index.html summary: Dataset of dataflow diagrams of microservice applications. toc: false --- ## DaFD {{% include image.html file="TUHH_logo-wortmarke_en_rgb.svg" alt="TUHH Logo" max-width="350" %}} {{% include image.html file="company_logo_big.png" alt="SoftSec Institute Logo" max-width="350" %}} This is DaFD, a dataset containing Dataflow Diagrams (DFDs) of microservices written in Java. The models correspond to actual implementation code of open-source applications found on GitHub. The DFDs are presented in multiple formats and contain full traceability of all model items to code, indicating the evidence for their implementation. Additionally to the models themselves, we present a mapping to a list of 17 architectural security best-practices, i.e. a table indicating whether each rules is followed or not. For those that are not followed, we created model variants that do follow the rule. These variants were crafted purely on the model-level and the added items do not correspond to code anymore. All artifacts were created manually by researchers of the Institute of Software Security at Hamburg University of Technology. ## Table of Contents - [Overview](index.html) - [Dataflow Diagrams](dfds.html) - [Use-Cases](usecases.html) - [Models](models.html) """) models_file = overview_dir / 'models.md' write_file_if_changed(models_file, f"""--- title: Models keywords: dataset models tags: [overview] sidebar: datasetdoc_sidebar permalink: models.html summary: Dataset of dataflow diagrams of microservice applications. datatable: true --- The following table presents the models in this dataset. It shows some properties about their popularity and size of the models. Column `Source` links directly to the corresponding repository on GitHub. If you click on the name of an entry, you will be referred to the model and all artifacts. Please select a model in column `Name`
Name | Source | LoC | Stars | Forks | DFD Items | Technologies -- | -- | -- | -- | -- | -- | -- {chr(10).join(f"[{info['slug']}]({model_id}.html) | [GitHub](https://github.com/{info['slug']}) | {info['l']} | {info['stars']} | {info['forks']} | {info['t']} | {len(info['tech'])}" for model_id, info in dataset.items())}
""") def write_tag_readme(dataset: Dataset): tag_dir = output_path / 'tags' known_tech = set(tech for model in dataset.values() for tech in model['tech']) tags_data_path = Path('_data') tags_data_file = tags_data_path / 'tags.yml' if tags_data_file.exists(): tags_data_path.mkdir(exist_ok=True, parents=True) with tags_data_file.open('r') as f: tags: dict[Any, Any] = yaml.safe_load(f) else: tags = {} tags['allowed-tags'] = list(sorted(set(itertools.chain(tags.get('allowed-tags', []), (get_tag_slug(tech) for tech in known_tech))))) with StringIO() as f: yaml.dump(tags, f) tags_content = f.getvalue() write_file_if_changed(tags_data_file, tags_content) for tech in known_tech: slug = get_tag_slug(tech) info_file = tag_dir / f'tag_{slug}.md' tag_dir.mkdir(exist_ok=True, parents=True) write_file_if_changed(info_file, f"""--- title: "{tech}" tagName: {slug} search: exclude permalink: tag_{slug}.html sidebar: datasetdoc_sidebar hide_sidebar: true folder: tags --- {{% include taglogic.html %}} {{% include links.html %}} """) def main(): global known_tags dataset = open_dataset() write_tag_readme(dataset) write_root_readme(dataset) write_model_readmes(dataset) if update_dataset: save_dataset(dataset) yaml.dump if __name__ == '__main__': main()