CVE-2026-105756

Received Received - Intake

Denial of Service in vLLM via Malformed Cache Salt

Publication date: 2026-10-05

Last updated on: 2026-10-05

Assigner: GitHub, Inc.

Description

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.

CVSS Scores

EPSS Scores

Probability:
Percentile:

Meta Information

Published

2026-10-05

Last Modified

2026-10-05

Generated

2026-10-06

AI Q&A

2026-10-06

EPSS Evaluated

N/A

NVD

CVE-2026-105756

Affected Vendors & Products

Vendor	Product	Version / Range
vllm-project	vllm	< 0.30.0

Helpful Resources

Exploitability

CWE

KEV

CWE ID	Description
CWE-20	The product receives input or data, but it does not validate or incorrectly validates that the input has the properties that are required to process the data safely and correctly.
CWE-248	An exception is thrown from a function, but it is not caught.

Attack-Flow Graph

Executive Summary

This vulnerability affects vLLM, an inference engine for large language models. When using OpenAI-compatible requests with a cache_salt value that contains forbidden characters or exceeds length limits, it triggers an uncaught error during cache lookup. This causes the EngineCore to terminate, disrupting service for all users.

Detection Guidance

Detecting this vulnerability requires checking the vLLM version in use. Run: vllm --version. If the version is below 0.30.0, the system is vulnerable. Additionally, inspect logs for EngineCore terminations during cache lookups.

Impact Analysis

If you use vLLM with LMCache-MP connector, an attacker could send a malicious request causing EngineCore to crash. This would deny service to all concurrent users, leading to downtime and potential loss of access to the language model.

Compliance Impact

This vulnerability does not directly affect compliance with GDPR, HIPAA, or similar standards as it is a service disruption issue rather than a data protection or privacy breach. The denial of service could indirectly impact compliance if it disrupts access to critical systems, but no evidence suggests data exposure or unauthorized access.

Mitigation Strategies

Upgrade vLLM to version 0.30.0 or later immediately. If upgrading is not possible, disable the LMCache-MP connector or restrict cache_salt inputs to prevent forbidden characters and excessive length.

Hi! I’m here to help you understand CVE-2026-105756. Ask me anything about the vulnerability, its impact, or mitigation strategies.

0/70