vLLM is a library for LLM inference and serving. vllm/model_executor/weight_utils.py implements hf_model_weights_iterator to load the model checkpoint, which is downloaded from huggingface. It uses the torch.load function and the weights_only parameter defaults to False. When torch.load loads malicious pickle data, it will execute arbitrary code during unpickling.
This vulnerability is fixed in v0.7.0.
Attack Parameters
Attack Vector
Network
Can be exploited remotely
Attack Complexity
Low
Easy to exploit
Privileges Required
None
No privileges needed
User Interaction
Required
User action required
Impact Assessment
Confidentiality
High
Complete data leak
Integrity
High
Complete data modification
Availability
High
Complete denial of service
CVSS Vector v3.1
Weakness Type (CWE)
Vulnerable Products 1
| Configuration | From (including) | Up to (excluding) |
|---|---|---|
|
Vllm Vllm
cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*
|
— |
0.7.0
|
References 4
https://github.com/vllm-project/vllm/commit/d3d6bb13fb62da3234addf6574922a4ec05…
security-advisories@github.com
https://github.com/vllm-project/vllm/pull/12366
security-advisories@github.com
https://github.com/vllm-project/vllm/security/advisories/GHSA-rh4j-5rhw-hr54
security-advisories@github.com
https://pytorch.org/docs/stable/generated/torch.load.html
security-advisories@github.com