CVE-2026-109227: SGLang through 0.5.21 Memory Exhaustion DoS via Qwen-VL Video Preprocessing
SGLang through 0.5.21 contains an unbounded memory allocation vulnerability in preprocessvideo in qwenvl.py that decodes frames at full source resolution before applying maxpixels limits. Remote attackers can submit high-resolution videos to /v1/chat/completions on Qwen-VL deployments, and concurrent requests exhaust worker memory causing denial of service.
Affected Software
Event History
Frequently Asked Questions
Which deployments are exposed to this issue?
Deployments using Qwen-VL through the affected SGLang versions are exposed if their /v1/chat/completions endpoint accepts video inputs. The issue is in Qwen-VL video preprocessing.
What does an attacker need to exploit it?
An attacker only needs network access to submit high-resolution videos to /v1/chat/completions. No privileges or user interaction are required.
Does limiting max_pixels prevent exploitation?
No. Frames are decoded at their full source resolution before max_pixels limits are applied, so the limit does not prevent the vulnerable memory allocation.
What is the operational impact?
Concurrent malicious video requests can exhaust worker memory and cause a denial of service. The available information does not indicate confidentiality or integrity impact.