mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-16 09:22:22 +00:00
fix(proxy): walk tuples when checking s3 media and skip plain number lists
This commit is contained in:
parent
91bb894618
commit
5d8f256963
2 changed files with 28 additions and 4 deletions
|
|
@ -8,6 +8,8 @@ from itertools import accumulate, chain, repeat, takewhile
|
|||
from typing import Final
|
||||
from urllib.parse import quote, unquote
|
||||
|
||||
from pydantic import TypeAdapter
|
||||
|
||||
from litellm._uuid import uuid
|
||||
from litellm.litellm_core_utils.core_helpers import as_object_list, as_str_mapping
|
||||
|
||||
|
|
@ -85,6 +87,8 @@ _EMBEDDED_S3_URI: Final = re.compile(r"s3://[^;]*")
|
|||
_MODEL_FILE_ID_MAPPING_KEY: Final = "model_file_id_mapping"
|
||||
_KEYS_THAT_MAY_HOLD_A_FILE_ID: Final = frozenset({"file_id", "file_url", "file_uri", "fileUri"})
|
||||
_KEYS_THAT_MAY_HIDE_AN_S3_URI: Final = _KEYS_THAT_MAY_HOLD_A_FILE_ID | {_MODEL_FILE_ID_MAPPING_KEY}
|
||||
_SCALAR_TYPES: Final = frozenset({int, float, bool, type(None)})
|
||||
_OBJECT_TUPLE: Final = TypeAdapter(tuple[object, ...])
|
||||
|
||||
|
||||
def is_s3_uri(value: str) -> bool:
|
||||
|
|
@ -111,6 +115,12 @@ def _with_embedded_s3_uris(file_id: str) -> Iterator[str]:
|
|||
return chain((file_id,), _s3_uris_embedded_in_file_id(file_id))
|
||||
|
||||
|
||||
def _sequence_items(value: object) -> Sequence[object] | None:
|
||||
if isinstance(value, tuple):
|
||||
return _OBJECT_TUPLE.validate_python(value)
|
||||
return as_object_list(value)
|
||||
|
||||
|
||||
@dataclass(frozen=True, slots=True)
|
||||
class _BodyNode:
|
||||
value: object
|
||||
|
|
@ -128,7 +138,7 @@ def _child_nodes(node: _BodyNode) -> Iterator[_BodyNode]:
|
|||
if fields is not None:
|
||||
inside_media_part: Final = node.inside_media_part or _is_media_part(fields, node.key)
|
||||
return (_BodyNode(child, child_key, inside_media_part) for child_key, child in fields.items())
|
||||
items: Final = as_object_list(node.value)
|
||||
items: Final = _sequence_items(node.value)
|
||||
return (_BodyNode(item, None, node.inside_media_part) for item in items) if items is not None else iter(())
|
||||
|
||||
|
||||
|
|
@ -192,8 +202,8 @@ def _may_name_an_s3_uri(value: object) -> bool:
|
|||
fields: Final = as_str_mapping(value)
|
||||
if fields is not None:
|
||||
return not _KEYS_THAT_MAY_HIDE_AN_S3_URI.isdisjoint(fields) or any(map(_may_name_an_s3_uri, fields.values()))
|
||||
items: Final = as_object_list(value)
|
||||
return items is not None and any(map(_may_name_an_s3_uri, items))
|
||||
items: Final = _sequence_items(value)
|
||||
return items is not None and not _SCALAR_TYPES.issuperset(map(type, items)) and any(map(_may_name_an_s3_uri, items))
|
||||
|
||||
|
||||
def _may_carry_s3_media(request_body: Mapping[str, object]) -> bool:
|
||||
|
|
|
|||
|
|
@ -13472,15 +13472,29 @@ _ADAPTER_BODY_WITH_S3_IMAGE: Final = json.dumps(
|
|||
],
|
||||
}
|
||||
).encode()
|
||||
_ADAPTER_PYTHON_LITERAL_BODY_WITH_S3_IMAGE_IN_TUPLES: Final = repr(
|
||||
{
|
||||
"model": "unknown-model",
|
||||
"messages": (
|
||||
{"role": "user", "content": ({"type": "image_url", "image_url": {"url": "s3://private-bucket/cat.png"}},)},
|
||||
),
|
||||
}
|
||||
).encode()
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
@pytest.mark.parametrize(
|
||||
"body",
|
||||
[_ADAPTER_BODY_WITH_S3_IMAGE, _ADAPTER_PYTHON_LITERAL_BODY_WITH_S3_IMAGE_IN_TUPLES],
|
||||
ids=["json", "python_literal_with_tuples"],
|
||||
)
|
||||
async def test_chat_completion_pass_through_endpoint_refuses_an_s3_image_in_an_unlisted_bucket(
|
||||
monkeypatch: pytest.MonkeyPatch,
|
||||
body: bytes,
|
||||
):
|
||||
proxy_logging: Final = _stub_adapter_proxy(monkeypatch)
|
||||
request: Final = MagicMock(spec=Request)
|
||||
request.body = AsyncMock(return_value=_ADAPTER_BODY_WITH_S3_IMAGE)
|
||||
request.body = AsyncMock(return_value=body)
|
||||
|
||||
with pytest.raises(ProxyException) as raised:
|
||||
await chat_completion_pass_through_endpoint(
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue