vLLM is an inference and serving engine for large language models (LLMs). From 0.7.0 to before 0.19.0, the VideoMediaIO.load_base64() method at vllm/multimodal/media/video.py splits video/jpeg data URLs by comma to extract individual JPEG frames, but does not enforce a frame count limit. The num_frames parameter (default: 32), which is enforced by the load_bytes() code path, is completely bypassed in the video/jpeg base64 path. An attacker can send a single API request containing thousands of comma-separated base64-encoded JPEG frames, causing the server to decode all frames into memory and crash with OOM. This vulnerability is fixed in 0.19.0.

Affected: PyPI: vllm

## Summary

The `VideoMediaIO.load_base64()` method at `vllm/multimodal/media/video.py:51-62` splits `video/jpeg` data URLs by comma to extract individual JPEG frames, but does not enforce a frame count limit. The `num_frames` parameter (default: 32), which is enforced by the `load_bytes()` code path at line 47-48, is completely bypassed in the `video/jpeg` base64 path. An attacker can send a single API request containing thousands of comma-separated base64-encoded JPEG frames, causing the server to decode all frames into memory and crash with OOM.

## Details

### Vulnerable code

```python # video.py:51-62 def load_base64(self, media_type: str, data: str) -> tuple[npt.NDArray, dict[str, Any]]: if media_type.lower() == "video/jpeg": load_frame = partial(self.image_io.load_base64, "image/jpeg") return np.stack( [np.asarray(load_frame(frame_data)) for frame_data in data.split(",")] # ^^^^^^^^^^ # Unbounded split — no frame count limit ), {} return self.load_bytes(base64.b64decode(data)) ```

The `load_bytes()` path (line 47-48) properly delegates to a video loader that respects `self.num_frames` (default 32). The `load_base64("video/jpeg", ...)` path bypasses this limit entirely — `data.split(",")` produces an unbounded list and every frame is decoded into a numpy array.

### video/jpeg is part of vLLM's public API

`video/jpeg` is a vLLM-specific MIME typ…

Affected: PyPI: vllm

vLLM is an inference and serving engine for large language models (LLMs). From 0.7.0 to before 0.19.0, the VideoMediaIO.load_base64() method at vllm/multimodal/media/video.py splits video/jpeg data URLs by comma to extract individual JPEG frames, but does not enforce a frame count limit. The num_frames parameter (default: 32), which is enforced by the load_bytes() code path, is completely bypassed in the video/jpeg base64 path. An attacker can send a single API request containing thousands of comma-separated base64-encoded JPEG frames, causing the server to decode all frames into memory and crash with OOM. This vulnerability is fixed in 0.19.0.

giflib: giflib: Denial of Service via buffer overflow in EGifGCBToExtension gnutls: GnuTLS: Denial of Service via DTLS zero-length fragment gnutls: GnuTLS: Denial of Service via heap buffer overflow in DTLS handshake fragment reassembly vLLM: vLLM: Denial of Service due to excessive video frame processing vllm: vLLM: Denial of Service via excessively large 'n' parameter in OpenAI-compatible API vim: arbitrary command execution via modeline sandbox bypass vllm: vLLM: Arbitrary code execution via malicious HuggingFace model gnutls: gnutls: Denial of Service via DTLS packet reordering vulnerability gnutls: gnutls: Authentication Bypass via NUL Character in Username vllm: starlette: vLLM: Critical authentication bypass allows unauthorized API access vllm: vLLM: Denial of Service due to improper floating-point validation starlette: Starlette: request.form() limits silently ignored for application/x-www-form-urlencoded enable DoS

Ein Angreifer kann mehrere Schwachstellen in vllm ausnutzen, um Dateien zu manipulieren, Sicherheitsmaßnahmen zu umgehen, vertrauliche Informationen offenzulegen oder einen Denial-of-Service-Zustand herbeizuführen.