Reject oversized media before fully downloading it (#51896)

Signed-off-by: Clinton Thomas <1033162+KernelClint@users.noreply.github.com>
Co-authored-by: Lucas Bourtoule <35483370+dhalf@users.noreply.github.com>
This commit is contained in:
Clinton Thomas
2026-08-24 01:15:09 +00:00
committed by GitHub
co-authored by Lucas Bourtoule
parent 585bb07c70
commit 2ec6f0d71e
13 changed files with 718 additions and 24 deletions
+26 -5
View File
@@ -10,7 +10,7 @@ from argparse import Namespace
from collections.abc import Awaitable, Callable
from http import HTTPStatus
from io import BytesIO, StringIO
from typing import Any, TypeAlias
from typing import Any, NoReturn, TypeAlias
from urllib.parse import urlparse
import aiohttp
@@ -28,7 +28,7 @@ from urllib3.util import parse_url
import vllm.envs as envs
from vllm.config import config
from vllm.connections import global_http_connection
from vllm.connections import HTTPResponseSizeExceededError, global_http_connection
from vllm.engine.arg_utils import AsyncEngineArgs
from vllm.engine.protocol import EngineClient
from vllm.entrypoints.launchers.api_server.entry import init_app_state
@@ -68,11 +68,24 @@ from vllm.logger import init_logger
from vllm.reasoning import ReasoningParserManager
from vllm.utils import random_uuid
from vllm.utils.argparse_utils import FlexibleArgumentParser
from vllm.utils.mem_constants import MiB_bytes
from vllm.version import __version__ as VLLM_VERSION
logger = init_logger(__name__)
def _get_audio_download_limit_bytes() -> int:
return int(envs.VLLM_MAX_AUDIO_CLIP_FILESIZE_MB * MiB_bytes)
def _raise_audio_filesize_limit(size_bytes: int) -> NoReturn:
raise VLLMValidationError(
"Maximum file size exceeded",
parameter="audio_filesize_mb",
value=size_bytes / MiB_bytes,
)
class BatchTranscriptionRequest(TranscriptionRequest):
"""
Batch transcription request that uses file_url instead of file.
@@ -474,6 +487,9 @@ async def download_bytes_from_url(
if "," in url:
header, data = url.split(",", 1)
if "base64" in header:
decoded_size_bytes = len(data.rstrip("=")) * 3 // 4
if decoded_size_bytes > _get_audio_download_limit_bytes():
_raise_audio_filesize_limit(decoded_size_bytes)
return base64.b64decode(data)
else:
raise VLLMValidationError(
@@ -502,9 +518,14 @@ async def download_bytes_from_url(
# between urllib3 and aiohttp (e.g. backslash-@ attacks).
url = url_spec.url
return await global_http_connection.async_get_bytes(
url, allow_redirects=envs.VLLM_MEDIA_URL_ALLOW_REDIRECTS
)
try:
return await global_http_connection.async_get_bytes(
url,
allow_redirects=envs.VLLM_MEDIA_URL_ALLOW_REDIRECTS,
max_bytes=_get_audio_download_limit_bytes(),
)
except HTTPResponseSizeExceededError as exc:
_raise_audio_filesize_limit(exc.received_bytes)
else:
raise VLLMValidationError(
@@ -124,7 +124,6 @@ class SpeechToTextBaseServing(GenerateBaseServing):
self.enable_force_include_usage = enable_force_include_usage
self.max_audio_filesize_mb = envs.VLLM_MAX_AUDIO_CLIP_FILESIZE_MB
self.max_audio_decode_duration_s: int = envs.VLLM_MAX_AUDIO_DECODE_DURATION_S
self.max_audio_decode_bytes: int = envs.VLLM_MAX_AUDIO_DECODE_BYTES
if self.model_cls.supports_segment_timestamp:
@@ -274,13 +273,6 @@ class SpeechToTextBaseServing(GenerateBaseServing):
else None
)
if len(audio_data) / 1024**2 > self.max_audio_filesize_mb:
raise VLLMValidationError(
"Maximum file size exceeded",
parameter="audio_filesize_mb",
value=len(audio_data) / 1024**2,
)
# Run cpu intensive preprocess step in a separate thread pool executor.
chunks, duration = await self._decode_and_chunk_speech_async(audio_data)