fix(pipelines): prevent arbitrary file read and path traversal in /files endpoint

[AdvaitVarhade]

Summary of Changes

This pull request resolves a path traversal and arbitrary file read vulnerability in the pipelines REST API /files endpoint (slm/pipelines/rest_api/controller/file_upload.py).

Problem

In the previous implementation of download_file, os.path.join(FILE_PARSE_PATH, file_name) was used directly without path sanitization or boundary verification. Supplying absolute paths (e.g., /etc/passwd or C:\Windows\win.ini) or directory traversal sequences (e.g., ../../file) caused os.path.join to discard FILE_PARSE_PATH or traverse outside the intended directory.

Solution

1. Input Sanitization: Strip directory components using os.path.basename(file_name).
2. Boundary Validation: Check that abs_target starts with abs_base + os.sep (where abs_base = os.path.abspath(FILE_PARSE_PATH)).
3. Invalid Request Handling: If file_name contains path traversal attempts or resolves outside FILE_PARSE_PATH, raise HTTPException(status_code=400, detail="Invalid file name").
4. Regular File Verification: Ensure only existing regular files (os.path.isfile) are returned via FileResponse.
5. Unit Testing: Added unit tests in slm/pipelines/tests/test_file_upload.py covering valid file downloads, default filename behavior, absolute path traversal attempts, relative directory traversal, and non-existent files.

We welcome feedback from the maintainers on this implementation!

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

AdvaitVarhade seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}