Database/AI/ML frameworks & serving
vLLM: unvalidated cache_salt raises an uncaught ValueError and terminates EngineCore
Impact
The OpenAI-compatible request models accepted any non-empty cache_salt without applying the character and length restrictions that LMCache-MP's IPCCacheServerKey consumer requires. On a deployment using the LMCache-MP connector, a salt containing a forbidden character or exceeding the permitted length raises an uncaught ValueError during scheduler cache lookup and terminates EngineCore, denying service to every concurrent user of that replica. This is a one-field, one-request kill of a shared GPU serving process - cheap to trigger repeatedly, and each recovery costs a weight reload. Only deployments wiring up the LMCache-MP KV connector are affected; no data disclosure is claimed.
Who can reach it
Any authenticated API client that can set cache_salt on a request to the OpenAI-compatible endpoint (CVSS AV:N/PR:L), on a deployment configured with the LMCache-MP connector.
What to do
Upgrade vLLM to 0.30.0 and restart the serving replicas. If upgrading has to wait, strip or validate cache_salt at the gateway in front of vLLM, or stop using the LMCache-MP connector - both are config changes plus a replica restart rather than node maintenance.
References
Related entries
- vLLM: structured-output request failures escape request scope and terminate the shared engineCVE-2026-105757 · vLLM structured-output path (grammar compilation, ngram_gpu speculative decoding, Rust frontend validation)Medium
- MLflow: unauthenticated scorer/invoke deserialization lets a caller drive outbound requests to internal hostsCVE-2026-115886 · MLflow scorer endpoints (Scorer.model_validate_json deserialization)Medium
- vLLM: unbounded chat-template rendering lets one request wedge or exhaust the serving processCVE-2026-115911 · vLLM OpenAI-compatible server (request-supplied Jinja chat template rendering)Medium
- vLLM: Phi-4 audio encoder materializes quadratic attention matrices, letting one long clip OOM the engineCVE-2026-115913 · vLLM Phi-4 multimodal audio encoder (phi4mm_audio.py eager attention)Medium
- vLLM: an all-zero image on the DeepSeek-OCR path trips a fatal assertion and shuts the engine downCVE-2026-115914 · vLLM DeepSeek-OCR image path (_parse_and_validate_image_input / sanity_check_mm_encoder_outputs)Medium
- vLLM: off-by-one prompt-token-ID bound lets a request crash the inference engineCVE-2026-115916 · vLLM tokenizer input validation (vllm/tokenizers/hf.py max_token_id)Medium
This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.