chore: 纳入本地 Dify 1.19.0 源码及品牌与免登改动

演示目录只保留可重建的源码。运行密钥、数据卷、依赖和带口令的本地配置不进入仓库。

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
wuzilongandCursor committed 2026-11-18 13:59:25 +08:00
1 parent 60158935f4
commit 6519682885
15831 files changed
+3063428

No files matched your search

@@ -0,0 +1,502 @@
import inspect
import logging
from collections.abc import Callable
from enum import StrEnum, auto
from functools import wraps
from typing import Protocol, cast, overload
from flask import current_app, request
from flask_login import user_logged_in
from flask_restx import Resource
from flask_restx.utils import merge
from pydantic import BaseModel
from sqlalchemy import select
from sqlalchemy.orm import Session
from werkzeug.exceptions import Forbidden, NotFound, ServiceUnavailable, Unauthorized
from configs import dify_config
from controllers.common.access_response import MISSING_BEARER_MESSAGE
from controllers.common.errors import UnauthorizedError
from controllers.common.resource_access_token_errors import resource_access_token_errors
from controllers.service_api.schema import (
USER_FETCH_FROM_ATTR,
USER_FORM_PARAM,
USER_QUERY_PARAM,
USER_REQUIRED_ATTR,
)
from enums import CloudPlan, DeploymentEdition
from extensions.ext_application_services import application_services
from extensions.ext_database import db
from libs.login import current_user
from models import Account, Tenant, TenantAccountJoin, TenantStatus
from models.dataset import Dataset
from models.model import ApiToken, EndUser
from repositories.knowledge import dataset_api_key_bindings
from services.api_token_service import (
ApiTokenCache,
ApiTokenInvalidError,
CachedApiToken,
fetch_token_with_single_flight,
record_token_usage,
)
from services.app_service import AppService
from services.auth.resource_access_token_contracts import DatasetServiceApiGrant, is_resource_access_token
from services.feature_service import FeatureService
from services.knowledge_rate_limit_service import KnowledgeRateLimitExceededError, enforce_knowledge_rate_limit
logger = logging.getLogger(__name__)
class _RestxDocumentedView(Protocol):
"""Callable view object carrying Flask-RESTX documentation metadata."""
__apidoc__: dict[str, object]
class WhereisUserArg(StrEnum):
"""
Enum for whereis_user_arg.
"""
QUERY = auto()
JSON = auto()
FORM = auto()
class FetchUserArg(BaseModel):
fetch_from: WhereisUserArg
required: bool = False
APP_TOKEN_FORBIDDEN_RESPONSE = {
403: "Forbidden - token scope, app, dataset, or workspace access denied",
}
DATASET_TOKEN_AUTH_RESPONSES = {
401: "Unauthorized - invalid API token",
403: "Forbidden - dataset API access or workspace access denied",
}
VECTOR_SPACE_UNAVAILABLE_RESPONSE = {
503: (
"`service_unavailable` : Vector space usage could not be verified. Returned on the Dify Cloud Sandbox "
"plan only; retry the request later."
),
}
def document_app_token_contract(view_func: Callable[..., object], fetch_user_arg: FetchUserArg | None) -> None:
doc: dict[str, object] = {"responses": APP_TOKEN_FORBIDDEN_RESPONSE}
if fetch_user_arg is not None:
setattr(view_func, USER_FETCH_FROM_ATTR, fetch_user_arg.fetch_from.name)
setattr(view_func, USER_REQUIRED_ATTR, fetch_user_arg.required)
match fetch_user_arg.fetch_from:
case WhereisUserArg.QUERY:
doc["params"] = {"user": {**USER_QUERY_PARAM, "required": fetch_user_arg.required}}
case WhereisUserArg.FORM:
doc["params"] = {"user": {**USER_FORM_PARAM, "required": fetch_user_arg.required}}
case WhereisUserArg.JSON:
pass
cast(_RestxDocumentedView, view_func).__apidoc__ = cast(
dict[str, object],
merge(getattr(view_func, "__apidoc__", {}), doc),
)
@overload
def validate_app_token[**P, R](view: Callable[P, R]) -> Callable[P, R]: ...
@overload
def validate_app_token[**P, R](
view: None = None, *, fetch_user_arg: FetchUserArg | None = None
) -> Callable[[Callable[P, R]], Callable[P, R]]: ...
def validate_app_token[**P, R](
view: Callable[P, R] | None = None, *, fetch_user_arg: FetchUserArg | None = None
) -> Callable[P, R] | Callable[[Callable[P, R]], Callable[P, R]]:
def decorator(view_func: Callable[P, R]) -> Callable[P, R]:
@wraps(view_func)
def decorated_view(*args: P.args, **kwargs: P.kwargs) -> R:
# TODO: Remove these duplicate app/workspace checks when all legacy
# ORM App/EndUser handlers (including account-only endpoints) migrate
# to scalar admission. Keep resource-token routing and app policy aligned
# with service_api_end_user_admission until then.
auth_token = peek_service_api_bearer_token()
if auth_token and is_resource_access_token(auth_token):
with resource_access_token_errors():
grant = application_services().resource_access_tokens.resolve_app_for_service_api(
token=auth_token,
requested_app_id=request.headers.get("X-Dify-App-ID"),
)
app_model = AppService.get_app_in_workspace(
tenant_id=grant.tenant_id,
app_id=next(iter(grant.app_ids)),
session=db.session(),
)
else:
api_token = validate_and_get_api_token("app")
app_model = AppService.get_app_by_id(api_token.app_id, session=db.session())
if not app_model:
raise Forbidden("The app no longer exists.")
if app_model.status != "normal":
raise Forbidden("The app's status is abnormal.")
if not app_model.enable_api:
raise Forbidden("The app's API service has been disabled.")
tenant = db.session.get(Tenant, app_model.tenant_id)
if tenant is None:
raise ValueError("Tenant does not exist.")
if tenant.status == TenantStatus.ARCHIVE:
raise Forbidden("The workspace's status is archived.")
kwargs["app_model"] = app_model
# If caller needs end-user context, attach EndUser to current_user
if fetch_user_arg:
kwargs["end_user"] = resolve_service_api_end_user(
tenant_id=app_model.tenant_id, app_id=app_model.id, fetch_user_arg=fetch_user_arg
)
else:
# For service API without end-user context, ensure an Account is logged in
# so services relying on current_account_with_tenant() work correctly.
tenant_owner_info = db.session.execute(
select(Tenant, Account)
.join(TenantAccountJoin, Tenant.id == TenantAccountJoin.tenant_id)
.join(Account, TenantAccountJoin.account_id == Account.id)
.where(
Tenant.id == app_model.tenant_id,
TenantAccountJoin.role == "owner",
Tenant.status == TenantStatus.NORMAL,
)
).one_or_none()
if tenant_owner_info:
tenant_model, account = tenant_owner_info
account.set_current_tenant_with_session(tenant_model, session=db.session())
current_app.login_manager._update_request_context_with_user(account) # type: ignore
user_logged_in.send(current_app._get_current_object(), user=current_user) # type: ignore
else:
raise Unauthorized("Tenant owner account not found or tenant is not active.")
return view_func(*args, **kwargs)
document_app_token_contract(decorated_view, fetch_user_arg)
return decorated_view
if view is None:
return decorator
else:
return decorator(view)
def resolve_service_api_end_user(*, tenant_id: str, app_id: str, fetch_user_arg: FetchUserArg) -> EndUser:
"""Share user parsing, provisioning and Flask login identity across API admission paths."""
user_id = None
match fetch_user_arg.fetch_from:
case WhereisUserArg.QUERY:
user_id = request.args.get("user")
case WhereisUserArg.JSON:
user_id = request.get_json().get("user")
case WhereisUserArg.FORM:
user_id = request.form.get("user")
if not user_id and fetch_user_arg.required:
raise ValueError("Arg user must be provided.")
if user_id:
user_id = str(user_id)
end_user = application_services().app_scoped_end_users.commands.get_or_create_end_user(tenant_id, app_id, user_id)
current_app.login_manager._update_request_context_with_user(end_user) # type: ignore
user_logged_in.send(current_app._get_current_object(), user=end_user) # type: ignore
return end_user
def cloud_edition_billing_resource_check[**P, R](
resource: str,
api_token_type: str,
) -> Callable[[Callable[P, R]], Callable[P, R]]:
def interceptor(view: Callable[P, R]):
@wraps(view)
def decorated(*args: P.args, **kwargs: P.kwargs):
api_token = validate_and_get_api_token(api_token_type)
if dify_config.DEPLOYMENT_EDITION != DeploymentEdition.CLOUD:
return view(*args, **kwargs)
tenant_id = api_token.tenant_id
if tenant_id is None:
raise Unauthorized("Tenant id is required for this token.")
if resource == "vector_space":
vector_space = application_services().feature_queries.get_workspace_vector_space(tenant_id)
if vector_space.usage_unknown:
features = FeatureService.get_features(tenant_id, exclude_vector_space=True)
if features.billing.subscription.plan == CloudPlan.SANDBOX:
raise ServiceUnavailable(
"Unable to verify vector space usage right now. Please try again later."
)
if 0 < vector_space.limit <= vector_space.size:
raise Forbidden("The capacity of the vector space has reached the limit of your subscription.")
return view(*args, **kwargs)
features = FeatureService.get_features(tenant_id, exclude_vector_space=True)
members = features.members
apps = features.apps
documents_upload_quota = features.documents_upload_quota
if resource == "members" and 0 < members.limit <= members.size:
raise Forbidden("The number of members has reached the limit of your subscription.")
elif resource == "apps" and 0 < apps.limit <= apps.size:
raise Forbidden("The number of apps has reached the limit of your subscription.")
elif resource == "documents" and 0 < documents_upload_quota.limit <= documents_upload_quota.size:
raise Forbidden("The number of documents has reached the limit of your subscription.")
return view(*args, **kwargs)
if resource == "vector_space":
cast(_RestxDocumentedView, decorated).__apidoc__ = cast(
dict[str, object],
merge(decorated.__dict__.get("__apidoc__", {}), {"responses": VECTOR_SPACE_UNAVAILABLE_RESPONSE}),
)
return decorated
return interceptor
def cloud_edition_billing_knowledge_limit_check[**P, R](
resource: str,
api_token_type: str,
) -> Callable[[Callable[P, R]], Callable[P, R]]:
def interceptor(view: Callable[P, R]):
@wraps(view)
def decorated(*args: P.args, **kwargs: P.kwargs):
api_token = validate_and_get_api_token(api_token_type)
if dify_config.DEPLOYMENT_EDITION != DeploymentEdition.CLOUD or resource != "add_segment":
return view(*args, **kwargs)
tenant_id = api_token.tenant_id
if tenant_id is None:
raise Unauthorized("Tenant id is required for this token.")
features = FeatureService.get_features(tenant_id, exclude_vector_space=True)
if features.billing.subscription.plan == CloudPlan.SANDBOX:
raise Forbidden(
"To unlock this feature and elevate your Dify experience, please upgrade to a paid plan."
)
return view(*args, **kwargs)
return decorated
return interceptor
def check_knowledge_rate_limit(api_token: ApiToken | CachedApiToken) -> None:
"""Enforce the cloud knowledge base request rate limit for a dataset API key's workspace.
Shared by the ``cloud_edition_billing_rate_limit_check`` decorator and the KnowledgeFS
service routes, which authenticate inside their own profile helper rather than through
a decorator. No-op unless the workspace has a knowledge rate limit (cloud billing).
"""
if api_token.tenant_id is None:
raise Unauthorized("Tenant id is required for this token.")
try:
enforce_knowledge_rate_limit(api_token.tenant_id)
except KnowledgeRateLimitExceededError as error:
raise Forbidden(str(error)) from error
def cloud_edition_billing_rate_limit_check[**P, R](
resource: str,
api_token_type: str,
) -> Callable[[Callable[P, R]], Callable[P, R]]:
def interceptor(view: Callable[P, R]):
@wraps(view)
def decorated(*args: P.args, **kwargs: P.kwargs):
api_token = validate_and_get_api_token(api_token_type)
if api_token.tenant_id is None:
raise Forbidden("API token is not associated with a workspace")
if resource == "knowledge":
check_knowledge_rate_limit(api_token)
return view(*args, **kwargs)
return decorated
return interceptor
def validate_dataset_token[R](view: Callable[..., R]) -> Callable[..., R]:
view_parameters = inspect.signature(view).parameters
positional_parameters = [
parameter
for parameter in view_parameters.values()
if parameter.kind in (inspect.Parameter.POSITIONAL_ONLY, inspect.Parameter.POSITIONAL_OR_KEYWORD)
]
expects_bound_instance = bool(positional_parameters and positional_parameters[0].name in {"self", "cls"})
accepts_dataset_grant = "dataset_grant" in view_parameters
@wraps(view)
def decorated(*args: object, **kwargs: object) -> R:
# Flask may pass URL path parameters positionally, so inspect both kwargs and args.
dataset_id = kwargs.get("dataset_id")
if not dataset_id and args:
potential_id = args[0]
try:
str_id = str(potential_id)
if len(str_id) == 36 and str_id.count("-") == 4:
dataset_id = str_id
except Exception:
logger.exception("Failed to parse dataset_id from positional args")
auth_token = peek_service_api_bearer_token()
if auth_token and is_resource_access_token(auth_token):
with resource_access_token_errors():
dataset_grant = application_services().resource_access_tokens.resolve_datasets_for_service_api(
token=auth_token,
dataset_id=str(dataset_id) if dataset_id else None,
)
tenant_id = dataset_grant.tenant_id
else:
api_token = validate_and_get_api_token("dataset")
tenant_id = api_token.tenant_id
# Per-knowledge-base scoping is expressed by DatasetApiTokenBinding rows:
# no rows -> the key can reach every dataset in its tenant (default / back-compat)
# N rows -> the key is limited to exactly those resources
# Only legacy-dataset bindings count here: a key bound solely to KnowledgeFS spaces
# is restricted but has no reachable legacy dataset, so it is rejected. A bound key
# may only call endpoints carrying one of its dataset ids; endpoints without a
# dataset id (e.g. list/create datasets) are rejected. The scope is queried per
# request (not cached) so changes take effect immediately.
# db.session is Flask-SQLAlchemy's scoped_session proxy; cast so the plain-Session
# typed helper accepts it (runtime proxies every Session method through unchanged).
if tenant_id is None:
raise Unauthorized("Tenant id is required for this token.")
scope = dataset_api_key_bindings.get_active_key_scope(
cast(Session, db.session), api_token_id=api_token.id, tenant_id=tenant_id
)
if scope is None:
raise Unauthorized("API key is no longer valid.")
if not scope.allows_dataset(str(dataset_id) if dataset_id else None):
raise Forbidden("The API key is not authorized to access this knowledge base.")
dataset_grant = DatasetServiceApiGrant(
token_id=str(api_token.id),
tenant_id=tenant_id,
dataset_ids=None,
)
if dataset_id:
dataset_id = str(dataset_id)
dataset = db.session.scalar(
select(Dataset)
.where(
Dataset.id == dataset_id,
Dataset.tenant_id == tenant_id,
)
.limit(1)
)
if not dataset:
raise NotFound("Dataset not found.")
if not dataset.enable_api:
raise Forbidden("Dataset api access is not enabled.")
tenant_account_join = db.session.execute(
select(Tenant, TenantAccountJoin).where(
Tenant.id == tenant_id,
TenantAccountJoin.tenant_id == Tenant.id,
TenantAccountJoin.role.in_(["owner"]),
Tenant.status == TenantStatus.NORMAL,
)
).one_or_none() # TODO: only owner information is required, so only one is returned.
if tenant_account_join:
tenant, ta = tenant_account_join
account = db.session.get(Account, ta.account_id)
# Login admin
if account:
account.set_current_tenant_with_session(tenant, session=db.session())
current_app.login_manager._update_request_context_with_user(account) # type: ignore
user_logged_in.send(current_app._get_current_object(), user=current_user) # type: ignore
else:
raise Unauthorized("Tenant owner account does not exist.")
else:
raise Unauthorized("Tenant does not exist.")
if accepts_dataset_grant:
kwargs["dataset_grant"] = dataset_grant
if expects_bound_instance:
if not args:
raise TypeError("validate_dataset_token expected a bound resource instance.")
return view(args[0], tenant_id, *args[1:], **kwargs)
return view(tenant_id, *args, **kwargs)
return decorated
def validate_and_get_api_token(scope: str | None = None) -> ApiToken | CachedApiToken:
"""
Validate and get API token with Redis caching.
This function uses a two-tier approach:
1. First checks Redis cache for the token
2. If not cached, queries database and caches the result
The last_used_at field is updated asynchronously via Celery task
to avoid blocking the request.
"""
auth_token = extract_service_api_bearer_token()
# Try to get token from cache first
# Returns a CachedApiToken (plain Python object), not a SQLAlchemy model
cached_token = ApiTokenCache.get(auth_token, scope)
if cached_token is not None:
logger.debug("Token validation served from cache for scope: %s", scope)
# Record usage in Redis for later batch update (no Celery task per request)
record_token_usage(auth_token, scope)
return cached_token
# Cache miss - use Redis lock for single-flight mode
# This ensures only one request queries DB for the same token concurrently
try:
return fetch_token_with_single_flight(auth_token, scope)
except ApiTokenInvalidError as error:
raise UnauthorizedError(str(error)) from error
def extract_service_api_bearer_token() -> str:
auth_token = peek_service_api_bearer_token()
if auth_token is None:
raise UnauthorizedError(MISSING_BEARER_MESSAGE)
return auth_token
def peek_service_api_bearer_token() -> str | None:
# Match the Gateway's HTTP SP/HTAB parser for both API keys and resource
# access tokens. Unicode whitespace must not bypass policy identity lookup.
auth_header = request.headers.get("Authorization") or ""
auth_parts = [part for part in auth_header.replace("\t", " ").split(" ") if part]
if len(auth_parts) != 2 or auth_parts[0].lower() != "bearer":
return None
return auth_parts[1]
class DatasetApiResource(Resource):
__apidoc__ = {"responses": DATASET_TOKEN_AUTH_RESPONSES}
method_decorators = [validate_dataset_token]
def get_dataset(self, dataset_id: str, tenant_id: str) -> Dataset:
dataset = db.session.scalar(
select(Dataset).where(Dataset.id == dataset_id, Dataset.tenant_id == tenant_id).limit(1)
)
if not dataset:
raise NotFound("Dataset not found.")
return dataset