mirror of
https://github.com/MobSF/Mobile-Security-Framework-MobSF.git
synced 2026-10-11 22:50:09 +00:00
Security Fixes [GHSA-8j49-mmcx-4mp5] Arbitrary File Read via Path Traversal in ZIP/APK Icon Extraction The android:icon attribute from an APK manifest was interpolated into file paths without validation, allowing a crafted APK to read arbitrary files from the server. Fixed by adding is_path_traversal() and is_safe_path() guards in find_icon_path_zip. Dependency bump included. [GHSA-3p54-567p-2wpr] CSRF Checks Not Enforced After Django Middleware Migration The migration from the deprecated MIDDLEWARE_CLASSES to MIDDLEWARE omitted CsrfViewMiddleware, SecurityMiddleware, and XFrameOptionsMiddleware, leaving CSRF, HSTS, and clickjacking protections silently disabled. All three have been restored to the active MIDDLEWARE tuple. The now-dead MIDDLEWARE_CLASSES block has been removed to prevent future confusion. [GHSA-x768-8642-mmq9] Zip Bomb Denial of Service via Per-File Size Limit Bypass The per-file size check in ZIP extraction logged a warning on oversized members but was missing a continue, allowing files exceeding ZIP_MAX_UNCOMPRESSED_FILE_SIZE to be extracted anyway as long as the aggregate limit was not reached. Fixed by adding the missing continue. [GHSA-95px-34x5-p37h] SSRF Port Restriction Bypass in assetlinks_check valid_host() was called with only the bare hostname, allowing android:port to be appended afterward without going through port validation, bypassing the HTTP/HTTPS-only restriction. Fixed with a two-layer approach: primary port allowlist check before URL assembly in get_browsable_activities, and a defence-in-depth re-check in _check_url. Hardening The hand-rolled safe_extract used os.path.abspath (symlink-blind) instead of os.path.realpath, creating a TOCTOU window where a symlink could redirect extraction outside the destination directory. Replaced with Python 3.12's tarfile.extractall(filter='data') (PEP 706), which rejects symlinks, hardlinks, absolute paths, and traversal members per-member before extraction. A robust realpath-based fallback is included for older Python versions. Both download_app_data call sites in the iOS dynamic analyser now wrap app_container and tarfile with shlex.quote when building the SSH tar command, as defensive coding hygiene. Refactoring is_path_traversal, is_safe_path, clean_filename, cmd_injection_check, is_pipe_or_link, and is_attack_pattern were scattered across utils.py and shared.py. All have been moved to mobsf/MobSF/security.py as the single authoritative location for security primitives. Developer Tooling AGENTS.md / CLAUDE.md - AI agent guidelines Added a coding-standards document for AI coding agents (Cursor, Claude, Codex) covering MobSF-specific secure-by-default patterns: path traversal guards, archive extraction safety, Django middleware active-tuple hygiene, split-validation anti-patterns (SSRF port bypass), guard completeness (continue/return/raise after every security check), and Django-specific security features (form validators, decorators, middleware). CLAUDE.md is a symlink to AGENTS.md.
Symbolic link
9 B
Symbolic link
9 B
AGENTS.md