mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-30 01:52:18 +00:00
85e61375d9
Auth-time validation of caller-supplied clientside credentials (api_base /
api_key / base_url) checks the opt-in for the model the caller DECLARED
only. Server-side router fallbacks re-dispatch the same kwargs to a
DIFFERENT deployment: if model-a opts into configurable api_base and the
config lists model-b as model-a's fallback, a caller can send
{"model": "model-a", "api_base": "http://attacker/"} (passes auth for
model-a), make their host answer 500, and the fallback re-dispatch would
forward the caller's api_base to model-b's dispatch — overriding model-b's
api_base while keeping model-b's own api_key, exfiltrating that provider
key to the caller-chosen host.
Three coordinated pieces:
- litellm_pre_call_utils: stamp the authoritative clientside-credential
opt-in scope into request metadata ("proxy_wide" when
general_settings.allow_client_side_credentials is true, else
"per_model"), and strip caller-forged values for that metadata key
(_UNTRUSTED_METADATA_CONTROL_FIELDS) so the scope cannot be spoofed.
- router: at dispatch time, drop clientside credential keys the deployment
being dispatched did not opt in to
(strip_clientside_credentials_without_deployment_opt_in) before the
clientside-credential branch can forward them.
- clientside_credential_handler: the scope constants, the per-deployment
opt-in check mirroring proxy auth's _is_param_allowed (kept local to
avoid a proxy -> router_utils circular import), and the strip helper.
Deliberately narrow scoping so the SDK router's per-call credential
feature is unchanged: no scope stamp (plain SDK call) -> unchanged;
proxy_wide -> unchanged; per_model -> each clientside credential key must
be opted in by the deployment being dispatched, else it is stripped and
the deployment's own config is used.
Regression: model-a(opt-in) -> 500 -> fallback model-b previously served
the fallback with model-b's dispatch pointed at the attacker URL and
leaked "Authorization: Bearer sk-backend-B-credential" to it; with this
change the fallback uses model-b's configured upstream, only the opted-in
model-a ever honors the caller's api_base, and the direct control
(model-b + api_base, no fallback) still 401s.
|
||
|---|---|---|
| .. | ||
| agent_tests | ||
| audio_tests | ||
| base_sdk_tests | ||
| basic_proxy_startup_tests | ||
| batches_tests | ||
| benchmarks | ||
| code_coverage_tests | ||
| documentation_tests | ||
| e2e | ||
| enterprise | ||
| guardrails_tests | ||
| image_gen_tests | ||
| integration | ||
| litellm-proxy-extras | ||
| litellm_utils_tests | ||
| llm_responses_api_testing | ||
| llm_translation | ||
| load_tests | ||
| local_testing | ||
| logging_callback_tests | ||
| mcp_tests | ||
| multi_instance_e2e_tests | ||
| ocr_tests | ||
| openai_endpoints_tests | ||
| otel_tests | ||
| pass_through_tests | ||
| pass_through_unit_tests | ||
| proxy_admin_ui_tests | ||
| proxy_behavior | ||
| proxy_e2e_anthropic_messages_tests | ||
| proxy_migration_tests | ||
| proxy_security_tests | ||
| proxy_unit_tests | ||
| router_unit_tests | ||
| rust-python-harness | ||
| search_tests | ||
| spend_tracking_tests | ||
| store_model_in_db_tests | ||
| test_litellm | ||
| unified_google_tests | ||
| vector_store_tests | ||
| windows_tests | ||
| __init__.py | ||
| _fake_openai_endpoint_server.py | ||
| _flush_vcr_cache.py | ||
| _live_test_helpers.py | ||
| _openai_record_replay_proxy.py | ||
| _vcr_conftest_common.py | ||
| _vcr_redis_persister.py | ||
| _wait_helpers.py | ||
| _ws_vcr.py | ||
| eval_swe_bench.py | ||
| fake_openai_endpoint.py | ||
| gettysburg.wav | ||
| large_text.py | ||
| openai_batch_completions.jsonl | ||
| pyrightconfig.json | ||
| README.MD | ||
| test_anthropic_compaction_usage.py | ||
| test_budget_management.py | ||
| test_callbacks_on_proxy.py | ||
| test_debug_warning.py | ||
| test_default_encoding_non_root.py | ||
| test_end_users.py | ||
| test_fallbacks.py | ||
| test_gpt5_azure_temperature_support.py | ||
| test_health.py | ||
| test_keys.py | ||
| test_litellm_proxy_responses_config.py | ||
| test_logging.conf | ||
| test_models.py | ||
| test_new_vector_store_endpoints.py | ||
| test_openai_endpoints.py | ||
| test_organizations.py | ||
| test_otel_thread_leak.py | ||
| test_presidio_latency.py | ||
| test_proxy_server_non_root.py | ||
| test_ratelimit.py | ||
| test_resource_cleanup.py | ||
| test_rust_python_harness.py | ||
| test_service_logger_otel.py | ||
| test_spend_logs.py | ||
| test_team.py | ||
| test_team_logging.py | ||
| test_team_members.py | ||
| test_users.py | ||
In total litellm runs 1000+ tests
[02/20/2025] Update:
To make it easier to contribute and map what behavior is tested,
we've started mapping the litellm directory in tests/test_litellm
This folder can only run mock tests.