CVE-2026-105756: vLLM: Loose `cache_salt` validation lets a single request kill EngineCore on LMCache-MP deployments — uncaught downstream `ValueError` denial of service
vllm-project · vllm
No fix listed · Affected: vllm-project · Exploitation: not established
At a glance
Higher than 26% of scored CVEs
What it isNVD
vLLM is an inference and serving engine for large language models.
Prior to 0.30.0, OpenAI-compatible request models accept a non-empty cache_salt value without enforcing the character and length restrictions required by the IPCCacheServerKey consumer in LMCache-MP. On deployments using the LMCache-MP connector, a salt that contains a forbidden character or exceeds the permitted length can raise an uncaught ValueError during scheduler cache lookup, causing EngineCore to terminate and denying service to all concurrent users.
This issue is fixed in version 0.30.0.
ExploitationNot established
Not in CISA KEV as of catalogue 2026.11.20 (20 Nov)
- Not establishedNothing in the sources we check
Not seen so far: in CISA KEV · exploited (reported) · scanning observed · proof of concept public.
Weakness and public checks
- Public check
- No public nuclei template
- Weakness
- Uncaught Exception · Input Validation and Logic
Other CVEs in vllm
Every published CVE for vllm-project vllm, by how far exploitation has gone. This one is among them.
Coverage1 article · 1 outlet
First in our coverage: The Hacker News, 7 Oct.
- LMCache flaw lets hackers run code on LLM cache servers remotelyThe Hacker News · first in our coverage · mentions it in passing
New articles also arrive in the RSS feed.