Directory Traversal in Python tarfile extract and extractall

This repository is a self-contained Docker lab demonstrating exploitation of CVE-2007-4559 (Python TarSlip) rather than a generic exploit toolkit. The core vulnerable component is vulnerable-api/app.py, a Flask application exposing POST /upload, GET /admin, and GET /health. The exploit path is straightforward: the API saves an uploaded tarball to /tmp/upload.tar.gz, creates /shared/uploads/<uuid>, and calls tarfile.extractall() on attacker-controlled content. In vulnerable mode, no path validation is performed, so a tar member named ../../../etc/passwd escapes the extraction directory and overwrites /etc/passwd. The app’s /admin endpoint intentionally authenticates by reading plaintext credentials from /etc/passwd, so the overwrite plants admin:hacked and converts the file write into an authentication bypass. Repository structure supports a full demo chain. attacker/craft_malicious.py generates two archives: innocent.tar.gz for baseline behavior and tarslip_passwd.tar.gz containing the traversal entry and malicious passwd contents. attacker/demo.py automates the attack workflow: upload archive, probe /admin with HTTP Basic auth, and optionally verify that the fixed configuration blocks the same payload. docker-compose.vulnerable.yml launches the vulnerable API, nginx file server, and attacker container on an isolated bridge network; docker-compose.fixed.yml launches the same app with USE_SAFE_EXTRACTION=true so extractall(..., filter='data') is used. run_demo.sh orchestrates the full demonstration interactively. Main exploit capabilities: crafting a malicious tar.gz archive, exploiting unsafe archive extraction over a web upload endpoint, arbitrary file overwrite via path traversal, and post-overwrite access to a protected admin endpoint. There is no reverse shell or code execution payload in the repository; the demonstrated impact is file overwrite leading to auth bypass. The exploit is operational because it includes working payload generation and an end-to-end driver, but payloads are hardcoded for the lab environment.

This repository contains a proof-of-concept (PoC) exploit for CVE-2007-4559, a long-standing path traversal vulnerability in Python's tarfile module. The main file, 'CVE-2007-4559-PoC.py', demonstrates how a malicious tar archive can be crafted to include a file with a path that escapes the intended extraction directory (e.g., '../../../../../../../..//tmp/passwd.etc'). When a vulnerable extraction is performed (using tarfile.extractall() without path sanitization), the file is written outside the target directory, confirming the vulnerability. The script creates a test file, adds it to a tar archive with a traversal path, extracts the archive, and checks if the file was written to the target location. The README.md provides background on the vulnerability, affected systems (notably Debian 11/12 and Ubuntu 24), and the ongoing risk due to lack of patching in some distributions. The exploit is a local PoC and does not require network access. The repository is structured simply, with one Python exploit script and a markdown readme.

This repository demonstrates a proof-of-concept exploit for CVE-2007-4559, a directory traversal vulnerability in tar file extraction. The main application is a Flask-based web app (src/app.py) that allows users to upload files, including .tar archives, which are extracted into a user-specific directory. The exploit leverages the fact that tar files can contain entries with directory traversal sequences (e.g., '../../flag'), allowing an attacker to write files outside the intended directory. The provided test/poc.py script uploads a malicious tar archive (test/pay.tar) containing a reference to '../../flag', then accesses the extracted file via the /api/access/get_flag endpoint. The repository includes Dockerfile and deployment instructions, a sample 'flag' file, and both the vulnerable application and exploit scripts. The main attack vector is network-based, targeting the /api/upload endpoint. The exploit is a working proof-of-concept and not weaponized, as it requires manual crafting of the tar file and direct interaction with the endpoints.