Summary
torchtitan/hf_datasets/multimodal/utils/image.py:36 fetches arbitrary URLs from dataset image paths without SSRF validation, allowing access to internal services/cloud metadata.
Vulnerability
Location: torchtitan/hf_datasets/multimodal/utils/image.py:36
requests.get(image, timeout=10)
The image URL comes from dataset data without validation. No checks for:
- Private/internal IP addresses (10.x, 172.16-31.x, 192.168.x)
- Cloud metadata endpoints (169.254.169.254)
- Loopback addresses
Impact
- SSRF to internal services
- Cloud metadata exfiltration (AWS/GCP/Azure)
- Internal network probing
Recommended Fix
Validate URL host is a public IP, or implement IP-based SSRF protection (block private IPs, metadata endpoints, loopback).
Summary
torchtitan/hf_datasets/multimodal/utils/image.py:36fetches arbitrary URLs from dataset image paths without SSRF validation, allowing access to internal services/cloud metadata.Vulnerability
Location:
torchtitan/hf_datasets/multimodal/utils/image.py:36The
imageURL comes from dataset data without validation. No checks for:Impact
Recommended Fix
Validate URL host is a public IP, or implement IP-based SSRF protection (block private IPs, metadata endpoints, loopback).