Allocation of Resources Without Limits or Throttling Affecting vllm package, versions [,0.29.0)
Threat Intelligence
Snyk has a proof-of-concept or detailed explanation of how to exploit this vulnerability.
The probability is the direct output of the EPSS model, and conveys an overall sense of the threat of exploitation in the wild. The percentile measures the EPSS probability relative to all known EPSS scores. Note: This data is updated daily, relying on the latest available EPSS model version. Check out the EPSS documentation for more details.
Do your applications use this vulnerable package?
In a few clicks we can analyze your entire application and see what components are vulnerable in your application, and suggest you quick fixes.
Test your applicationsSnyk Learn
Learn about Allocation of Resources Without Limits or Throttling vulnerabilities in an interactive lesson.
Start learning- Snyk IDSNYK-PYTHON-VLLM-20158369
- published27 Sept 2026
- disclosed26 Sept 2026
- creditUnknown
Introduced: 26 Sep 2026
NewCVE-2026-100649 (opens in a new tab)How to fix?
Upgrade vllm to version 0.29.0 or higher.
Overview
vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs
Affected versions of this package are vulnerable to Allocation of Resources Without Limits or Throttling via PyNvVideoCodecVideoBackendMixin, where sampler subclass shadowing causes independent _active_decoder_slots counter increments. Because _active_decoder_slots is a class-level integer and the augmented assignment cls._active_decoder_slots += 1 writes into the concrete subclass's __dict__ rather than the mixin, each subclass (e.g. VideoBackend, Qwen3VLVideoBackend, Qwen2VLVideoBackend) maintains its own counter while sharing the same decoder pool. An unauthenticated attacker can issue video requests that select different registered sampler subclasses, causing each subclass to independently pass the active < max guard and allocate up to hw_decoders slots, multiplying the total decoder allocations by the number of reachable subclasses and exhausting GPU memory beyond the startup reservation.
Note: This is only exploitable when the server is started with the PyNvVideoCodec backend explicitly configured (e.g. {"video":{"backend":"pynvvideocodec","hw_decoders":2}}).