Fix path traversal in tar/zip extraction

[AAtomical]

Summary

_uncompress_file_tar() and _uncompress_file_zip() in paddlespeech/cli/download.py extract archive members using tarfile.extract() / zipfile.extract() with no path validation. A crafted archive containing ../ entries can write files outside the intended extraction directory.

This is especially relevant because dataset download URLs use plaintext HTTP (http://www.openslr.org/, http://openslr.elda.org/), making MITM injection of malicious archives feasible.

Changes

• Validate all member names before extraction in both _uncompress_file_tar() and _uncompress_file_zip()
• Reject members with absolute paths or .. path components

Test

import io, tarfile, tempfile
from pathlib import Path
from paddlespeech.cli.download import _uncompress_file_tar

tar_path = Path(tempfile.mkdtemp()) / "test.tar.gz"
with tarfile.open(str(tar_path), "w:gz") as tf:
    info = tarfile.TarInfo(name="../../../tmp/proof.txt")
    data = b"traversal\n"
    info.size = len(data)
    tf.addfile(info, io.BytesIO(data))

try:
    _uncompress_file_tar(str(tar_path))
    print("FAIL: no exception raised")
except ValueError as e:
    print(f"PASS: {e}")

Before fix: file written to /tmp/proof.txt
After fix: ValueError: Path traversal detected in tar member: ../../../tmp/proof.txt

[paddle-bot]

Thanks for your contribution!

[AAtomical]

Fix #4175