UTC
Medium

vLLM: Mirrored multimodal IPC caches desync after a rejected request — a later request reusing the same media hash trips a receiver assertio

CVE-2026-105753 · GHSA-ph3r-5jfg-f84f · published 2026-10-06

Severity
Medium
CVSS score
6.5
Chance of exploitation (EPSS)
0.43%
Fixed in
0.28.0
  • Exploitation No known exploitation
  • Automatable No
  • Technical impact Partial
  • CISA SSVC
What it is

A remote client submitting multimodal requests can poison a cache identity — render a media item successfully, then have that request rejected — so a later request reusing the same media hash fails on the P1 receiver assertion. This is an availability failure against a shared serving instance. No code execution, memory corruption, or data disclosure is claimed. On this revision the failure is scoped as a request-level preprocessing error (the engine core catches around preprocessaddrequest); public reports show the same assertion cascading into further engine-loop assertions on other revisions. It applies to multimodal models running the default mirrored lru cache.

What to do

Upgrade vllm to 0.28.0 or later. If you start it from an MCP config with npx or uvx, pin that version so an old cached copy is not used.

Two complementary changes: 1. Make the mirrored commit atomic with admission — insert into the P0 sender cache only after the request has passed all admission checks (length, limits) and P1 has acknowledged the item, or roll back the P0 insert on rejection. 2. Defense in depth — convert the P1 receiver assert mmitem is not None into a checked, request-scoped error (fetch-on-miss from P0) so a desync degrades a single request rather than asserting in the engine loop. The core of the rollback half: wrap the post-render length check so a ValueError rejection discards the P0 entries the render just committed, before re-raising.

Affected
PackageRegistryAffectedFixed in
vllmPyPI< 0.28.00.28.0

The impact, patch and workaround text comes from the advisory, shortened (GitHub Advisory Database, CC BY 4.0). Proof-of-concept and reproduction sections and code are left out; the original advisory has them.

Sources: GitHub Advisory Database, reviewed and malware advisories (CC BY 4.0, https://github.com/advisories); advisories published by Cursor (https://github.com/cursor/cursor/security/advisories); advisories published by OpenClaw (https://github.com/openclaw/openclaw/security/advisories); advisories published by vLLM (https://github.com/vllm-project/vllm/security/advisories); advisories published by Open WebUI (https://github.com/open-webui/open-webui/security/advisories); advisories published by n8n (https://github.com/n8n-io/n8n/security/advisories); advisories published by Flowise (https://github.com/FlowiseAI/Flowise/security/advisories); advisories published by Langflow (https://github.com/langflow-ai/langflow/security/advisories); advisories published by the MCP Python SDK project (https://github.com/modelcontextprotocol/python-sdk/security/advisories); advisories published by MLflow (https://github.com/mlflow/mlflow/security/advisories); CISA Known Exploited Vulnerabilities catalog (CC0). Data as of 2026-11-22. ShieldMap describes how each problem works and how to fix it; it does not publish exploit code. Report a correction to [email protected].