Bug 2560972 (CVE-2026-115916)

Summary:	CVE-2026-115916 vllm: vLLM: Denial of Service via out-of-bounds prompt token ID
Product:	[Other] Security Response	Reporter:	OSIDB Bzimport <bzimport>
Component:	vulnerability	Assignee:	Product Security <prodsec-ir-bot>
Status:	NEW ---	QA Contact:
Severity:	medium	Docs Contact:
Priority:	medium
Version:	unspecified	CC:	alinfoot, bbrownin, dtrifiro, mwoodson, weaton
Target Milestone:	---	Keywords:	Security
Target Release:	---
Hardware:	All
OS:	Linux
Whiteboard:
Fixed In Version:		Doc Type:	---
Doc Text:	A flaw was found in vLLM. This vulnerability allows a remote attacker to cause a Denial of Service (DoS) by submitting a request containing an out-of-bounds pre-tokenized prompt token identifier. Due to an off-by-one error in input validation, an invalid token identifier equal to the vocabulary size is accepted, which triggers an assertion failure during embedding processing and terminates the inference engine.	Story Points:	---
Clone Of:		Environment:
Last Closed:		Type:	---
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:

Description OSIDB Bzimport 2026-10-26 17:26:22 UTC

vLLM is an inference and serving engine for large language models. Prior to 0.32.0, vllm/tokenizers/hf.py computes max_token_id with an off-by-one upper bound, allowing a pre-tokenized prompt ID equal to the vocabulary size to pass validation. For a tensor-parallel size of one and an exact multiple-of-64 vocabulary, VocabParallelEmbedding.forward does not mask that value before F.embedding uses it, causing a CUDA device-side assertion and terminating the engine. This issue is fixed in version 0.32.0.