This repository demonstrates a critical bug in tokio-tar and related async Rust tar libraries where PAX extended header size overrides are not applied before calculating the next header position.

Bug Summary

Root Cause: When processing tar files with PAX extended headers that override the file size, the library uses the octal size field (often zero) instead of the PAX override for position calculations.

Impact: This causes the parser to jump into file content and mistake it for tar headers, leading to extraction of wrong files.

Known Affected Libraries:

• async-tar https://github.com/dignifiedquire/async-tar
• tokio-tar https://github.com/vorot93/tokio-tar
• krata-tokio-tar https://github.com/edera-dev/tokio-tar
• astral-tokio-tar https://github.com/astral-sh/tokio-tar

Quick Start

# Requirements: CMake, Rust/Cargo, C++ compiler, system tar command

cmake -S . -B build
cmake --build build --target generate_report

This will:

1. Build all C++ and Rust tools
2. Generate a repro case tar file
3. Run comparisons showing the bug
4. Generate detailed reports in build/output/

Project Structure

├── disclosure/               # Security disclosure documentation tree
│   └── blast_radius/         # Record of projects depending on tokio-tar variants
├── repro_generator/          # C++ tool to generate a repro case tar file
├── tar-bug-detector/         # Rust tool comparing tar libraries  
├── tarwalk/                  # Correct C++ tar parser
│   ├── tarwalk.cpp           # Handles PAX correctly
│   └── tarwalk_bad.cpp       # Reproduces the same bug
├── CMakeLists.txt            # Build system
├── generate_report.cmake     # Report generation
└── README.md                 # This file

The Bug in Detail

Normal TAR Processing

Header -> Content (size from octal field) -> Next Header

PAX Extended TAR Processing (Correct)

PAX Header (size=1024) -> File Header (octal size=0) -> Content (1024 bytes) -> Next Header

PAX Extended TAR Processing (Buggy)

PAX Header (size=1024) -> File Header (octal size=0) -> Content (0 bytes) -> WRONG POSITION
                                                                              ↓
                                                                    Reading content as headers!

Real-World Trigger

Docker save creates tar files with:

• Large layers (>8GB) requiring PAX extensions
• Layer content starting with filesystem tar headers (etc/, usr/)
• When the bug triggers, parsers extract filesystem content instead of image manifests

Reproduction Files Generated

• pax_bug_compact.tar - Minimal reproduction case

Expected Results

Correct libraries (GNU tar, sync tar crate):

normal.txt -> blob.bin -> marker.txt

Buggy libraries (tokio-tar):

normal.txt -> blob.bin -> INNER_FILE -> marker.txt

The appearance of INNER_FILE indicates the bug - the library jumped into blob.bin content and mistook a fake tar header for a real entry.

Technical Details

The fix requires applying PAX overrides before position calculations:

// Read header
let mut file_size = header.size();

// Apply PAX overrides BEFORE calculating next position
if let Some(pax_size) = pending_pax.get("size") {
    file_size = pax_size.parse().unwrap();
}

// Now calculate next header position using effective size
let next_pos = current_pos + 512 + pad_to_512(file_size);

License

This reproduction code is provided for security research and responsible disclosure purposes.

See COPYING for original source code license.