CVE-2026-116178vLLM 0.20.0 before 0.30.0 DoS via /inference/v1/generate mm_placeholders

Descripción

vLLM from 0.20.0 before 0.30.0 contains a denial of service vulnerability that allows remote attackers to crash the engine by sending inconsistent multimodal metadata to the disaggregated /inference/v1/generate endpoint. Attackers can replay a valid multimodal item with repeated or overlapping mm_placeholders ranges or mismatched kwargs_data and mm_hashes lists, producing a negative text span in Qwen2.5-VL position construction that terminates the engine.

Datos: CVE.org, NVD, CISA KEV, FIRST EPSS, Ubuntu. Se publican tal como los indican las fuentes, en inglés.

Productos y versiones afectados

Referencias y parches

• GitHub Security Advisory (GHSA-34qw-q2h2-jp3m) vendor-advisoryPatchVendor Advisory
• VulnCheck Advisory: vLLM 0.20.0 before 0.30.0 DoS via /inference/v1/generate mm_placeholders third-party-advisoryThird Party Advisory