2023-11-21 15:29:58 -05:00
|
|
|
#
|
|
|
|
# This file is licensed under the Affero General Public License (AGPL) version 3.
|
|
|
|
#
|
2024-01-23 06:26:48 -05:00
|
|
|
# Copyright 2019-2020 The Matrix.org Foundation C.I.C.
|
|
|
|
# Copyright 2014-2016 OpenMarket Ltd
|
2023-11-21 15:29:58 -05:00
|
|
|
# Copyright (C) 2023 New Vector, Ltd
|
|
|
|
#
|
|
|
|
# This program is free software: you can redistribute it and/or modify
|
|
|
|
# it under the terms of the GNU Affero General Public License as
|
|
|
|
# published by the Free Software Foundation, either version 3 of the
|
|
|
|
# License, or (at your option) any later version.
|
|
|
|
#
|
|
|
|
# See the GNU Affero General Public License for more details:
|
|
|
|
# <https://www.gnu.org/licenses/agpl-3.0.html>.
|
|
|
|
#
|
|
|
|
# Originally licensed under the Apache License, Version 2.0:
|
|
|
|
# <http://www.apache.org/licenses/LICENSE-2.0>.
|
|
|
|
#
|
|
|
|
# [This file includes modifications made by New Vector Limited]
|
|
|
|
#
|
|
|
|
#
|
2018-04-27 06:40:06 -04:00
|
|
|
import logging
|
2020-08-24 13:58:56 -04:00
|
|
|
import random
|
2021-09-28 22:23:16 -04:00
|
|
|
from http import HTTPStatus
|
2021-04-29 07:17:28 -04:00
|
|
|
from typing import TYPE_CHECKING, Any, Dict, List, Mapping, Optional, Tuple
|
2018-04-27 06:40:06 -04:00
|
|
|
|
2020-08-19 07:26:03 -04:00
|
|
|
from canonicaljson import encode_canonical_json
|
2018-07-09 02:09:20 -04:00
|
|
|
|
2019-12-03 14:19:45 -05:00
|
|
|
from twisted.internet.interfaces import IDelayedCall
|
2014-08-27 12:59:36 -04:00
|
|
|
|
2019-07-17 14:08:02 -04:00
|
|
|
from synapse import event_auth
|
2019-12-03 14:19:45 -05:00
|
|
|
from synapse.api.constants import (
|
|
|
|
EventContentFields,
|
|
|
|
EventTypes,
|
2021-09-06 07:17:16 -04:00
|
|
|
GuestAccess,
|
2022-06-01 07:29:51 -04:00
|
|
|
HistoryVisibility,
|
2024-03-19 13:52:53 -04:00
|
|
|
JoinRules,
|
2019-12-03 14:19:45 -05:00
|
|
|
Membership,
|
|
|
|
RelationTypes,
|
|
|
|
UserTypes,
|
|
|
|
)
|
2018-08-15 11:35:22 -04:00
|
|
|
from synapse.api.errors import (
|
|
|
|
AuthError,
|
|
|
|
Codes,
|
|
|
|
ConsentNotGivenError,
|
|
|
|
NotFoundError,
|
2023-02-10 18:31:05 -05:00
|
|
|
PartialStateConflictError,
|
2020-08-24 13:58:56 -04:00
|
|
|
ShadowBanError,
|
2018-08-15 11:35:22 -04:00
|
|
|
SynapseError,
|
2022-07-27 08:44:40 -04:00
|
|
|
UnstableSpecAuthError,
|
2021-09-29 05:57:10 -04:00
|
|
|
UnsupportedRoomVersionError,
|
2018-08-15 11:35:22 -04:00
|
|
|
)
|
2022-06-10 06:01:55 -04:00
|
|
|
from synapse.api.room_versions import KNOWN_ROOM_VERSIONS
|
2018-05-22 03:56:52 -04:00
|
|
|
from synapse.api.urls import ConsentURIBuilder
|
Split `event_auth.check` into two parts (#10940)
Broadly, the existing `event_auth.check` function has two parts:
* a validation section: checks that the event isn't too big, that it has the rught signatures, etc.
This bit is independent of the rest of the state in the room, and so need only be done once
for each event.
* an auth section: ensures that the event is allowed, given the rest of the state in the room.
This gets done multiple times, against various sets of room state, because it forms part of
the state res algorithm.
Currently, this is implemented with `do_sig_check` and `do_size_check` parameters, but I think
that makes everything hard to follow. Instead, we split the function in two and call each part
separately where it is needed.
2021-09-29 13:59:15 -04:00
|
|
|
from synapse.event_auth import validate_event_for_room_version
|
2022-05-16 08:42:45 -04:00
|
|
|
from synapse.events import EventBase, relation_from_event
|
2020-07-22 12:29:15 -04:00
|
|
|
from synapse.events.builder import EventBuilder
|
2023-02-09 16:05:02 -05:00
|
|
|
from synapse.events.snapshot import EventContext, UnpersistedEventContextBase
|
2023-03-06 11:08:39 -05:00
|
|
|
from synapse.events.utils import SerializeEventConfig, maybe_upsert_event_field
|
2014-12-10 12:59:47 -05:00
|
|
|
from synapse.events.validator import EventValidator
|
2021-09-20 08:56:23 -04:00
|
|
|
from synapse.handlers.directory import DirectoryHandler
|
2023-08-16 10:19:54 -04:00
|
|
|
from synapse.handlers.worker_lock import NEW_EVENT_DURING_PURGE_LOCK_NAME
|
2022-08-03 13:19:34 -04:00
|
|
|
from synapse.logging import opentracing
|
2021-05-12 08:17:11 -04:00
|
|
|
from synapse.logging.context import make_deferred_yieldable, run_in_background
|
2019-06-19 06:33:03 -04:00
|
|
|
from synapse.metrics.background_process_metrics import run_as_background_process
|
2018-07-31 08:53:54 -04:00
|
|
|
from synapse.replication.http.send_event import ReplicationSendEventRestServlet
|
2022-09-28 06:11:48 -04:00
|
|
|
from synapse.replication.http.send_events import ReplicationSendEventsRestServlet
|
2020-08-05 16:38:57 -04:00
|
|
|
from synapse.storage.databases.main.events_worker import EventRedactBehaviour
|
2022-05-26 05:48:12 -04:00
|
|
|
from synapse.types import (
|
2022-09-28 06:11:48 -04:00
|
|
|
PersistedEventPosition,
|
2022-05-26 05:48:12 -04:00
|
|
|
Requester,
|
|
|
|
RoomAlias,
|
2022-09-28 05:39:03 -04:00
|
|
|
StateMap,
|
2022-05-26 05:48:12 -04:00
|
|
|
StreamToken,
|
|
|
|
UserID,
|
|
|
|
create_requester,
|
|
|
|
)
|
2022-12-12 11:19:30 -05:00
|
|
|
from synapse.types.state import StateFilter
|
2022-03-01 04:34:30 -05:00
|
|
|
from synapse.util import json_decoder, json_encoder, log_failure, unwrapFirstError
|
|
|
|
from synapse.util.async_helpers import Linearizer, gather_results
|
2021-05-05 11:53:22 -04:00
|
|
|
from synapse.util.caches.expiringcache import ExpiringCache
|
2016-08-23 10:23:39 -04:00
|
|
|
from synapse.util.metrics import measure_func
|
2022-06-01 07:29:51 -04:00
|
|
|
from synapse.visibility import get_effective_room_visibility_from_state
|
2014-12-10 12:59:47 -05:00
|
|
|
|
2020-07-09 05:40:19 -04:00
|
|
|
if TYPE_CHECKING:
|
|
|
|
from synapse.server import HomeServer
|
|
|
|
|
2014-08-27 12:59:36 -04:00
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
|
|
|
2020-09-04 06:54:56 -04:00
|
|
|
class MessageHandler:
|
2021-02-16 17:32:34 -05:00
|
|
|
"""Contains some read only APIs to get state about a room"""
|
2014-08-27 12:59:36 -04:00
|
|
|
|
2021-04-29 07:17:28 -04:00
|
|
|
def __init__(self, hs: "HomeServer"):
|
2018-07-18 10:22:02 -04:00
|
|
|
self.auth = hs.get_auth()
|
|
|
|
self.clock = hs.get_clock()
|
2015-02-09 12:41:29 -05:00
|
|
|
self.state = hs.get_state_handler()
|
2022-02-23 06:04:02 -05:00
|
|
|
self.store = hs.get_datastores().main
|
2022-05-31 08:17:50 -04:00
|
|
|
self._storage_controllers = hs.get_storage_controllers()
|
|
|
|
self._state_storage_controller = self._storage_controllers.state
|
2019-05-09 08:21:57 -04:00
|
|
|
self._event_serializer = hs.get_event_client_serializer()
|
2021-09-29 06:44:15 -04:00
|
|
|
self._ephemeral_events_enabled = hs.config.server.enable_ephemeral_messages
|
2019-12-03 14:19:45 -05:00
|
|
|
|
|
|
|
# The scheduled call to self._expire_event. None if no call is currently
|
|
|
|
# scheduled.
|
2021-07-16 13:22:36 -04:00
|
|
|
self._scheduled_expiry: Optional[IDelayedCall] = None
|
2019-12-03 14:19:45 -05:00
|
|
|
|
2021-09-13 13:07:12 -04:00
|
|
|
if not hs.config.worker.worker_app:
|
2019-12-03 14:19:45 -05:00
|
|
|
run_as_background_process(
|
|
|
|
"_schedule_next_expiry", self._schedule_next_expiry
|
|
|
|
)
|
2018-07-18 10:22:02 -04:00
|
|
|
|
2020-07-22 12:29:15 -04:00
|
|
|
async def get_room_data(
|
2021-02-16 17:32:34 -05:00
|
|
|
self,
|
2022-08-22 09:17:59 -04:00
|
|
|
requester: Requester,
|
2021-02-16 17:32:34 -05:00
|
|
|
room_id: str,
|
|
|
|
event_type: str,
|
|
|
|
state_key: str,
|
2021-04-29 07:17:28 -04:00
|
|
|
) -> Optional[EventBase]:
|
2021-02-16 17:32:34 -05:00
|
|
|
"""Get data from a room.
|
2018-07-18 10:22:02 -04:00
|
|
|
|
|
|
|
Args:
|
2022-08-22 09:17:59 -04:00
|
|
|
requester: The user who did the request.
|
2020-07-22 12:29:15 -04:00
|
|
|
room_id
|
|
|
|
event_type
|
|
|
|
state_key
|
2018-07-18 10:22:02 -04:00
|
|
|
Returns:
|
|
|
|
The path data content.
|
|
|
|
Raises:
|
2020-08-26 10:07:35 -04:00
|
|
|
SynapseError or AuthError if the user is not in the room
|
2018-07-18 10:22:02 -04:00
|
|
|
"""
|
2019-10-31 11:43:24 -04:00
|
|
|
(
|
|
|
|
membership,
|
|
|
|
membership_event_id,
|
2020-07-22 12:29:15 -04:00
|
|
|
) = await self.auth.check_user_in_room_or_world_readable(
|
2022-08-22 09:17:59 -04:00
|
|
|
room_id, requester, allow_departed_users=True
|
2020-02-18 18:14:57 -05:00
|
|
|
)
|
2018-07-18 10:22:02 -04:00
|
|
|
|
|
|
|
if membership == Membership.JOIN:
|
2022-06-06 04:24:12 -04:00
|
|
|
data = await self._storage_controllers.state.get_current_state_event(
|
|
|
|
room_id, event_type, state_key
|
|
|
|
)
|
2018-07-18 10:22:02 -04:00
|
|
|
elif membership == Membership.LEAVE:
|
|
|
|
key = (event_type, state_key)
|
2021-04-29 07:17:28 -04:00
|
|
|
# If the membership is not JOIN, then the event ID should exist.
|
|
|
|
assert (
|
|
|
|
membership_event_id is not None
|
|
|
|
), "check_user_in_room_or_world_readable returned invalid data"
|
2022-05-31 08:17:50 -04:00
|
|
|
room_state = await self._state_storage_controller.get_state_for_events(
|
2018-10-25 12:49:55 -04:00
|
|
|
[membership_event_id], StateFilter.from_types([key])
|
2018-07-18 10:22:02 -04:00
|
|
|
)
|
|
|
|
data = room_state[membership_event_id].get(key)
|
2020-08-26 10:07:35 -04:00
|
|
|
else:
|
|
|
|
# check_user_in_room_or_world_readable, if it doesn't raise an AuthError, should
|
|
|
|
# only ever return a Membership.JOIN/LEAVE object
|
|
|
|
#
|
|
|
|
# Safeguard in case it returned something else
|
|
|
|
logger.error(
|
|
|
|
"Attempted to retrieve data from a room for a user that has never been in it. "
|
|
|
|
"This should not have happened."
|
|
|
|
)
|
2022-07-27 08:44:40 -04:00
|
|
|
raise UnstableSpecAuthError(
|
|
|
|
403,
|
|
|
|
"User not in room",
|
|
|
|
errcode=Codes.NOT_JOINED,
|
|
|
|
)
|
2018-07-18 10:22:02 -04:00
|
|
|
|
2019-07-23 09:00:55 -04:00
|
|
|
return data
|
2018-07-18 10:22:02 -04:00
|
|
|
|
2020-07-22 12:29:15 -04:00
|
|
|
async def get_state_events(
|
2019-06-20 05:32:02 -04:00
|
|
|
self,
|
2022-08-22 09:17:59 -04:00
|
|
|
requester: Requester,
|
2020-07-22 12:29:15 -04:00
|
|
|
room_id: str,
|
2021-04-08 17:38:54 -04:00
|
|
|
state_filter: Optional[StateFilter] = None,
|
2020-07-22 12:29:15 -04:00
|
|
|
at_token: Optional[StreamToken] = None,
|
|
|
|
) -> List[dict]:
|
2018-07-18 10:22:02 -04:00
|
|
|
"""Retrieve all state events for a given room. If the user is
|
|
|
|
joined to the room then return the current state. If the user has
|
2018-08-15 11:35:22 -04:00
|
|
|
left the room return the state events from when they left. If an explicit
|
|
|
|
'at' parameter is passed, return the state events as of that event, if
|
|
|
|
visible.
|
2018-07-18 10:22:02 -04:00
|
|
|
|
|
|
|
Args:
|
2022-08-22 09:17:59 -04:00
|
|
|
requester: The user requesting state events.
|
2020-07-22 12:29:15 -04:00
|
|
|
room_id: The room ID to get all state events from.
|
|
|
|
state_filter: The state filter used to fetch state from the database.
|
|
|
|
at_token: the stream token of the at which we are requesting
|
2018-08-15 11:35:22 -04:00
|
|
|
the stats. If the user is not allowed to view the state as of that
|
|
|
|
stream token, we raise a 403 SynapseError. If None, returns the current
|
|
|
|
state based on the current_state_events table.
|
2018-07-18 10:22:02 -04:00
|
|
|
Returns:
|
|
|
|
A list of dicts representing state events. [{}, {}, {}]
|
2018-08-15 11:35:22 -04:00
|
|
|
Raises:
|
|
|
|
NotFoundError (404) if the at token does not yield an event
|
|
|
|
|
|
|
|
AuthError (403) if the user doesn't have permission to view
|
|
|
|
members of this room.
|
2018-07-18 10:22:02 -04:00
|
|
|
"""
|
2021-04-08 17:38:54 -04:00
|
|
|
state_filter = state_filter or StateFilter.all()
|
2022-08-22 09:17:59 -04:00
|
|
|
user_id = requester.user.to_string()
|
2021-04-08 17:38:54 -04:00
|
|
|
|
2018-08-15 11:35:22 -04:00
|
|
|
if at_token:
|
2022-06-01 07:29:51 -04:00
|
|
|
last_event_id = (
|
2024-06-17 12:27:14 -04:00
|
|
|
await self.store.get_last_event_id_in_room_before_stream_ordering(
|
2022-06-01 07:29:51 -04:00
|
|
|
room_id,
|
|
|
|
end_token=at_token.room_key,
|
|
|
|
)
|
2018-08-15 11:35:22 -04:00
|
|
|
)
|
2018-07-18 10:22:02 -04:00
|
|
|
|
2022-06-01 07:29:51 -04:00
|
|
|
if not last_event_id:
|
2019-06-20 05:32:02 -04:00
|
|
|
raise NotFoundError("Can't find event for token %s" % (at_token,))
|
2021-08-31 05:09:58 -04:00
|
|
|
|
2022-06-01 07:29:51 -04:00
|
|
|
if not await self._user_can_see_state_at_event(
|
|
|
|
user_id, room_id, last_event_id
|
|
|
|
):
|
2018-08-15 11:35:22 -04:00
|
|
|
raise AuthError(
|
|
|
|
403,
|
2019-06-20 05:32:02 -04:00
|
|
|
"User %s not allowed to view events in room %s at token %s"
|
|
|
|
% (user_id, room_id, at_token),
|
2018-08-15 11:35:22 -04:00
|
|
|
)
|
2022-06-01 07:29:51 -04:00
|
|
|
|
|
|
|
room_state_events = (
|
|
|
|
await self._state_storage_controller.get_state_for_events(
|
|
|
|
[last_event_id], state_filter=state_filter
|
|
|
|
)
|
|
|
|
)
|
|
|
|
room_state: Mapping[Any, EventBase] = room_state_events[last_event_id]
|
2018-08-15 11:35:22 -04:00
|
|
|
else:
|
2019-10-31 11:43:24 -04:00
|
|
|
(
|
|
|
|
membership,
|
|
|
|
membership_event_id,
|
2020-07-22 12:29:15 -04:00
|
|
|
) = await self.auth.check_user_in_room_or_world_readable(
|
2022-08-22 09:17:59 -04:00
|
|
|
room_id, requester, allow_departed_users=True
|
2020-02-18 18:14:57 -05:00
|
|
|
)
|
2018-08-15 11:35:22 -04:00
|
|
|
|
|
|
|
if membership == Membership.JOIN:
|
2022-06-01 11:02:53 -04:00
|
|
|
state_ids = await self._state_storage_controller.get_current_state_ids(
|
2019-06-20 05:32:02 -04:00
|
|
|
room_id, state_filter=state_filter
|
2018-08-15 11:35:22 -04:00
|
|
|
)
|
2020-07-22 12:29:15 -04:00
|
|
|
room_state = await self.store.get_events(state_ids.values())
|
2018-08-15 11:35:22 -04:00
|
|
|
elif membership == Membership.LEAVE:
|
2021-04-29 07:17:28 -04:00
|
|
|
# If the membership is not JOIN, then the event ID should exist.
|
|
|
|
assert (
|
|
|
|
membership_event_id is not None
|
|
|
|
), "check_user_in_room_or_world_readable returned invalid data"
|
2022-05-31 08:17:50 -04:00
|
|
|
room_state_events = (
|
|
|
|
await self._state_storage_controller.get_state_for_events(
|
|
|
|
[membership_event_id], state_filter=state_filter
|
|
|
|
)
|
2018-08-15 11:35:22 -04:00
|
|
|
)
|
2021-04-29 07:17:28 -04:00
|
|
|
room_state = room_state_events[membership_event_id]
|
2018-07-18 10:22:02 -04:00
|
|
|
|
2023-10-27 05:04:08 -04:00
|
|
|
events = await self._event_serializer.serialize_events(
|
2023-03-06 11:08:39 -05:00
|
|
|
room_state.values(),
|
|
|
|
self.clock.time_msec(),
|
|
|
|
config=SerializeEventConfig(requester=requester),
|
|
|
|
)
|
2019-07-23 09:00:55 -04:00
|
|
|
return events
|
2018-07-18 10:22:02 -04:00
|
|
|
|
2022-06-01 07:29:51 -04:00
|
|
|
async def _user_can_see_state_at_event(
|
|
|
|
self, user_id: str, room_id: str, event_id: str
|
|
|
|
) -> bool:
|
|
|
|
# check whether the user was in the room, and the history visibility,
|
|
|
|
# at that time.
|
|
|
|
state_map = await self._state_storage_controller.get_state_for_event(
|
|
|
|
event_id,
|
|
|
|
StateFilter.from_types(
|
|
|
|
[
|
|
|
|
(EventTypes.Member, user_id),
|
|
|
|
(EventTypes.RoomHistoryVisibility, ""),
|
|
|
|
]
|
|
|
|
),
|
|
|
|
)
|
|
|
|
|
|
|
|
membership = None
|
|
|
|
membership_event = state_map.get((EventTypes.Member, user_id))
|
|
|
|
if membership_event:
|
|
|
|
membership = membership_event.membership
|
|
|
|
|
|
|
|
# if the user was a member of the room at the time of the event,
|
|
|
|
# they can see it.
|
|
|
|
if membership == Membership.JOIN:
|
|
|
|
return True
|
|
|
|
|
|
|
|
# otherwise, it depends on the history visibility.
|
|
|
|
visibility = get_effective_room_visibility_from_state(state_map)
|
|
|
|
|
|
|
|
if visibility == HistoryVisibility.JOINED:
|
|
|
|
# we weren't a member at the time of the event, so we can't see this event.
|
|
|
|
return False
|
|
|
|
|
|
|
|
# otherwise *invited* is good enough
|
|
|
|
if membership == Membership.INVITE:
|
|
|
|
return True
|
|
|
|
|
|
|
|
if visibility == HistoryVisibility.INVITED:
|
|
|
|
# we weren't invited, so we can't see this event.
|
|
|
|
return False
|
|
|
|
|
|
|
|
if visibility == HistoryVisibility.WORLD_READABLE:
|
|
|
|
return True
|
|
|
|
|
|
|
|
# So it's SHARED, and the user was not a member at the time. The user cannot
|
|
|
|
# see history, unless they have *subsequently* joined the room.
|
|
|
|
#
|
|
|
|
# XXX: if the user has subsequently joined and then left again,
|
|
|
|
# ideally we would share history up to the point they left. But
|
|
|
|
# we don't know when they left. We just treat it as though they
|
|
|
|
# never joined, and restrict access.
|
|
|
|
|
|
|
|
(
|
|
|
|
current_membership,
|
|
|
|
_,
|
|
|
|
) = await self.store.get_local_current_membership_for_user_in_room(
|
|
|
|
user_id, event_id
|
|
|
|
)
|
|
|
|
return current_membership == Membership.JOIN
|
|
|
|
|
2020-07-22 12:29:15 -04:00
|
|
|
async def get_joined_members(self, requester: Requester, room_id: str) -> dict:
|
2018-07-18 10:22:02 -04:00
|
|
|
"""Get all the joined members in the room and their profile information.
|
|
|
|
|
|
|
|
If the user has left the room return the state events from when they left.
|
|
|
|
|
|
|
|
Args:
|
2020-07-22 12:29:15 -04:00
|
|
|
requester: The user requesting state events.
|
|
|
|
room_id: The room ID to get all state events from.
|
2018-07-18 10:22:02 -04:00
|
|
|
Returns:
|
|
|
|
A dict of user_id to profile info
|
|
|
|
"""
|
|
|
|
if not requester.app_service:
|
|
|
|
# We check AS auth after fetching the room membership, as it
|
|
|
|
# requires us to pull out all joined members anyway.
|
2020-07-22 12:29:15 -04:00
|
|
|
membership, _ = await self.auth.check_user_in_room_or_world_readable(
|
2022-08-22 09:17:59 -04:00
|
|
|
room_id, requester, allow_departed_users=True
|
2018-07-18 10:22:02 -04:00
|
|
|
)
|
|
|
|
if membership != Membership.JOIN:
|
2022-08-03 08:26:31 -04:00
|
|
|
raise SynapseError(
|
|
|
|
code=403,
|
|
|
|
errcode=Codes.FORBIDDEN,
|
|
|
|
msg="Getting joined members while not being a current member of the room is forbidden.",
|
2018-07-18 10:22:02 -04:00
|
|
|
)
|
|
|
|
|
2022-08-16 08:16:56 -04:00
|
|
|
users_with_profile = (
|
|
|
|
await self._state_storage_controller.get_users_in_room_with_profiles(
|
|
|
|
room_id
|
|
|
|
)
|
|
|
|
)
|
2018-07-18 10:22:02 -04:00
|
|
|
|
|
|
|
# If this is an AS, double check that they are allowed to see the members.
|
|
|
|
# This can either be because the AS user is in the room or because there
|
|
|
|
# is a user in the room that the AS is "interested in"
|
2022-08-22 09:17:59 -04:00
|
|
|
if (
|
|
|
|
requester.app_service
|
|
|
|
and requester.user.to_string() not in users_with_profile
|
|
|
|
):
|
2018-07-18 10:22:02 -04:00
|
|
|
for uid in users_with_profile:
|
|
|
|
if requester.app_service.is_interested_in_user(uid):
|
|
|
|
break
|
|
|
|
else:
|
|
|
|
# Loop fell through, AS has no interested users in room
|
2022-07-27 08:44:40 -04:00
|
|
|
raise UnstableSpecAuthError(
|
|
|
|
403,
|
|
|
|
"Appservice not in room",
|
|
|
|
errcode=Codes.NOT_JOINED,
|
|
|
|
)
|
2018-07-18 10:22:02 -04:00
|
|
|
|
2019-07-23 09:00:55 -04:00
|
|
|
return {
|
|
|
|
user_id: {
|
|
|
|
"avatar_url": profile.avatar_url,
|
|
|
|
"display_name": profile.display_name,
|
2018-07-18 10:22:02 -04:00
|
|
|
}
|
2020-06-15 07:03:36 -04:00
|
|
|
for user_id, profile in users_with_profile.items()
|
2019-07-23 09:00:55 -04:00
|
|
|
}
|
2018-07-18 10:22:02 -04:00
|
|
|
|
2021-09-20 08:56:23 -04:00
|
|
|
def maybe_schedule_expiry(self, event: EventBase) -> None:
|
2019-12-03 14:19:45 -05:00
|
|
|
"""Schedule the expiry of an event if there's not already one scheduled,
|
|
|
|
or if the one running is for an event that will expire after the provided
|
|
|
|
timestamp.
|
|
|
|
|
|
|
|
This function needs to invalidate the event cache, which is only possible on
|
|
|
|
the master process, and therefore needs to be run on there.
|
|
|
|
|
|
|
|
Args:
|
2020-07-22 12:29:15 -04:00
|
|
|
event: The event to schedule the expiry of.
|
2019-12-03 14:19:45 -05:00
|
|
|
"""
|
|
|
|
|
|
|
|
expiry_ts = event.content.get(EventContentFields.SELF_DESTRUCT_AFTER)
|
2023-08-29 09:41:43 -04:00
|
|
|
if type(expiry_ts) is not int or event.is_state(): # noqa: E721
|
2019-12-03 14:19:45 -05:00
|
|
|
return
|
|
|
|
|
|
|
|
# _schedule_expiry_for_event won't actually schedule anything if there's already
|
|
|
|
# a task scheduled for a timestamp that's sooner than the provided one.
|
|
|
|
self._schedule_expiry_for_event(event.event_id, expiry_ts)
|
|
|
|
|
2021-09-20 08:56:23 -04:00
|
|
|
async def _schedule_next_expiry(self) -> None:
|
2019-12-03 14:19:45 -05:00
|
|
|
"""Retrieve the ID and the expiry timestamp of the next event to be expired,
|
|
|
|
and schedule an expiry task for it.
|
|
|
|
|
|
|
|
If there's no event left to expire, set _expiry_scheduled to None so that a
|
|
|
|
future call to save_expiry_ts can schedule a new expiry task.
|
|
|
|
"""
|
|
|
|
# Try to get the expiry timestamp of the next event to expire.
|
2020-07-22 12:29:15 -04:00
|
|
|
res = await self.store.get_next_event_to_expire()
|
2019-12-03 14:19:45 -05:00
|
|
|
if res:
|
|
|
|
event_id, expiry_ts = res
|
|
|
|
self._schedule_expiry_for_event(event_id, expiry_ts)
|
|
|
|
|
2021-09-20 08:56:23 -04:00
|
|
|
def _schedule_expiry_for_event(self, event_id: str, expiry_ts: int) -> None:
|
2019-12-03 14:19:45 -05:00
|
|
|
"""Schedule an expiry task for the provided event if there's not already one
|
|
|
|
scheduled at a timestamp that's sooner than the provided one.
|
|
|
|
|
|
|
|
Args:
|
2020-07-22 12:29:15 -04:00
|
|
|
event_id: The ID of the event to expire.
|
|
|
|
expiry_ts: The timestamp at which to expire the event.
|
2019-12-03 14:19:45 -05:00
|
|
|
"""
|
|
|
|
if self._scheduled_expiry:
|
|
|
|
# If the provided timestamp refers to a time before the scheduled time of the
|
|
|
|
# next expiry task, cancel that task and reschedule it for this timestamp.
|
|
|
|
next_scheduled_expiry_ts = self._scheduled_expiry.getTime() * 1000
|
|
|
|
if expiry_ts < next_scheduled_expiry_ts:
|
|
|
|
self._scheduled_expiry.cancel()
|
|
|
|
else:
|
|
|
|
return
|
|
|
|
|
|
|
|
# Figure out how many seconds we need to wait before expiring the event.
|
|
|
|
now_ms = self.clock.time_msec()
|
|
|
|
delay = (expiry_ts - now_ms) / 1000
|
|
|
|
|
|
|
|
# callLater doesn't support negative delays, so trim the delay to 0 if we're
|
|
|
|
# in that case.
|
|
|
|
if delay < 0:
|
|
|
|
delay = 0
|
|
|
|
|
|
|
|
logger.info("Scheduling expiry for event %s in %.3fs", event_id, delay)
|
|
|
|
|
|
|
|
self._scheduled_expiry = self.clock.call_later(
|
|
|
|
delay,
|
|
|
|
run_as_background_process,
|
|
|
|
"_expire_event",
|
|
|
|
self._expire_event,
|
|
|
|
event_id,
|
|
|
|
)
|
|
|
|
|
2021-09-20 08:56:23 -04:00
|
|
|
async def _expire_event(self, event_id: str) -> None:
|
2019-12-03 14:19:45 -05:00
|
|
|
"""Retrieve and expire an event that needs to be expired from the database.
|
|
|
|
|
|
|
|
If the event doesn't exist in the database, log it and delete the expiry date
|
|
|
|
from the database (so that we don't try to expire it again).
|
|
|
|
"""
|
|
|
|
assert self._ephemeral_events_enabled
|
|
|
|
|
|
|
|
self._scheduled_expiry = None
|
|
|
|
|
|
|
|
logger.info("Expiring event %s", event_id)
|
|
|
|
|
|
|
|
try:
|
|
|
|
# Expire the event if we know about it. This function also deletes the expiry
|
|
|
|
# date from the database in the same database transaction.
|
2020-07-22 12:29:15 -04:00
|
|
|
await self.store.expire_event(event_id)
|
2019-12-03 14:19:45 -05:00
|
|
|
except Exception as e:
|
|
|
|
logger.error("Could not expire event %s: %r", event_id, e)
|
|
|
|
|
|
|
|
# Schedule the expiry of the next event to expire.
|
2020-07-22 12:29:15 -04:00
|
|
|
await self._schedule_next_expiry()
|
2019-12-03 14:19:45 -05:00
|
|
|
|
2018-07-18 10:22:02 -04:00
|
|
|
|
2019-09-26 06:47:53 -04:00
|
|
|
# The duration (in ms) after which rooms should be removed
|
|
|
|
# `_rooms_to_exclude_from_dummy_event_insertion` (with the effect that we will try
|
|
|
|
# to generate a dummy event for them once more)
|
|
|
|
#
|
|
|
|
_DUMMY_EVENT_ROOM_EXCLUSION_EXPIRY = 7 * 24 * 60 * 60 * 1000
|
|
|
|
|
|
|
|
|
2020-09-04 06:54:56 -04:00
|
|
|
class EventCreationHandler:
|
2020-07-09 05:40:19 -04:00
|
|
|
def __init__(self, hs: "HomeServer"):
|
2018-01-15 11:51:53 -05:00
|
|
|
self.hs = hs
|
2022-06-14 04:51:15 -04:00
|
|
|
self.auth_blocking = hs.get_auth_blocking()
|
2021-07-01 14:25:37 -04:00
|
|
|
self._event_auth_handler = hs.get_event_auth_handler()
|
2022-02-23 06:04:02 -05:00
|
|
|
self.store = hs.get_datastores().main
|
2022-05-31 08:17:50 -04:00
|
|
|
self._storage_controllers = hs.get_storage_controllers()
|
2018-01-15 11:51:53 -05:00
|
|
|
self.state = hs.get_state_handler()
|
|
|
|
self.clock = hs.get_clock()
|
|
|
|
self.validator = EventValidator()
|
|
|
|
self.profile_handler = hs.get_profile_handler()
|
|
|
|
self.event_builder_factory = hs.get_event_builder_factory()
|
|
|
|
self.server_name = hs.hostname
|
|
|
|
self.notifier = hs.get_notifier()
|
2018-02-05 12:22:16 -05:00
|
|
|
self.config = hs.config
|
2021-09-29 06:44:15 -04:00
|
|
|
self.require_membership_for_aliases = (
|
|
|
|
hs.config.server.require_membership_for_aliases
|
|
|
|
)
|
2020-09-14 05:16:41 -04:00
|
|
|
self._events_shard_config = self.config.worker.events_shard_config
|
|
|
|
self._instance_name = hs.get_instance_name()
|
2022-07-19 07:45:17 -04:00
|
|
|
self._notifier = hs.get_notifier()
|
2023-07-31 05:58:03 -04:00
|
|
|
self._worker_lock_handler = hs.get_worker_locks_handler()
|
2018-02-05 12:22:16 -05:00
|
|
|
|
2021-06-09 14:39:51 -04:00
|
|
|
self.room_prejoin_state_types = self.hs.config.api.room_prejoin_state
|
2019-12-09 06:50:34 -05:00
|
|
|
|
2020-05-22 11:11:35 -04:00
|
|
|
self.send_event = ReplicationSendEventRestServlet.make_client(hs)
|
2022-09-28 06:11:48 -04:00
|
|
|
self.send_events = ReplicationSendEventsRestServlet.make_client(hs)
|
2018-01-15 11:51:53 -05:00
|
|
|
|
2021-10-08 07:44:43 -04:00
|
|
|
self.request_ratelimiter = hs.get_request_ratelimiter()
|
2018-01-15 11:51:53 -05:00
|
|
|
|
2023-02-08 14:25:11 -05:00
|
|
|
# We limit concurrent event creation for a room to 1. This prevents state resolution
|
|
|
|
# from occurring when sending bursts of events to a local room
|
|
|
|
self.limiter = Linearizer(max_count=1, name="room_event_creation_limit")
|
2018-01-15 11:51:53 -05:00
|
|
|
|
2022-05-11 07:15:21 -04:00
|
|
|
self._bulk_push_rule_evaluator = hs.get_bulk_push_rule_evaluator()
|
2018-01-15 11:51:53 -05:00
|
|
|
|
2023-04-17 20:57:40 -04:00
|
|
|
self._spam_checker_module_callbacks = hs.get_module_api_callbacks().spam_checker
|
2023-05-04 10:18:22 -04:00
|
|
|
self._third_party_event_rules = (
|
|
|
|
self.hs.get_module_api_callbacks().third_party_event_rules
|
2021-07-16 13:22:36 -04:00
|
|
|
)
|
2018-01-15 11:51:53 -05:00
|
|
|
|
2019-03-19 07:38:59 -04:00
|
|
|
self._block_events_without_consent_error = (
|
2021-09-23 07:13:34 -04:00
|
|
|
self.config.consent.block_events_without_consent_error
|
2019-03-19 07:38:59 -04:00
|
|
|
)
|
|
|
|
|
2020-10-13 08:20:32 -04:00
|
|
|
# we need to construct a ConsentURIBuilder here, as it checks that the necessary
|
|
|
|
# config options, but *only* if we have a configuration for which we are
|
|
|
|
# going to need it.
|
|
|
|
if self._block_events_without_consent_error:
|
|
|
|
self._consent_uri_builder = ConsentURIBuilder(self.config)
|
|
|
|
|
2019-09-26 06:47:53 -04:00
|
|
|
# Rooms which should be excluded from dummy insertion. (For instance,
|
|
|
|
# those without local users who can send events into the room).
|
|
|
|
#
|
|
|
|
# map from room id to time-of-last-attempt.
|
|
|
|
#
|
2021-07-16 13:22:36 -04:00
|
|
|
self._rooms_to_exclude_from_dummy_event_insertion: Dict[str, int] = {}
|
2020-10-13 08:20:32 -04:00
|
|
|
# The number of forward extremeities before a dummy event is sent.
|
2021-09-29 06:44:15 -04:00
|
|
|
self._dummy_events_threshold = hs.config.server.dummy_events_threshold
|
2018-05-22 03:56:52 -04:00
|
|
|
|
2019-06-17 13:04:42 -04:00
|
|
|
if (
|
2021-09-13 13:07:12 -04:00
|
|
|
self.config.worker.run_background_tasks
|
2021-09-29 06:44:15 -04:00
|
|
|
and self.config.server.cleanup_extremities_with_dummy_events
|
2019-06-17 13:04:42 -04:00
|
|
|
):
|
|
|
|
self.clock.looping_call(
|
2019-06-19 06:33:03 -04:00
|
|
|
lambda: run_as_background_process(
|
|
|
|
"send_dummy_events_to_fill_extremities",
|
2019-06-20 05:32:02 -04:00
|
|
|
self._send_dummy_events_to_fill_extremities,
|
2019-06-19 06:33:03 -04:00
|
|
|
),
|
2019-06-17 13:04:42 -04:00
|
|
|
5 * 60 * 1000,
|
|
|
|
)
|
|
|
|
|
2019-12-03 14:19:45 -05:00
|
|
|
self._message_handler = hs.get_message_handler()
|
|
|
|
|
2021-09-29 06:44:15 -04:00
|
|
|
self._ephemeral_events_enabled = hs.config.server.enable_ephemeral_messages
|
2019-12-03 14:19:45 -05:00
|
|
|
|
2021-01-26 08:57:31 -05:00
|
|
|
self._external_cache = hs.get_external_cache()
|
|
|
|
|
2021-05-05 11:53:22 -04:00
|
|
|
# Stores the state groups we've recently added to the joined hosts
|
|
|
|
# external cache. Note that the timeout must be significantly less than
|
|
|
|
# the TTL on the external cache.
|
2021-07-16 13:22:36 -04:00
|
|
|
self._external_cache_joined_hosts_updates: Optional[ExpiringCache] = None
|
2021-05-05 11:53:22 -04:00
|
|
|
if self._external_cache.is_enabled():
|
|
|
|
self._external_cache_joined_hosts_updates = ExpiringCache(
|
|
|
|
"_external_cache_joined_hosts_updates",
|
|
|
|
self.clock,
|
|
|
|
expiry_ms=30 * 60 * 1000,
|
|
|
|
)
|
|
|
|
|
2020-07-22 12:29:15 -04:00
|
|
|
async def create_event(
|
2019-06-20 05:32:02 -04:00
|
|
|
self,
|
2020-07-22 12:29:15 -04:00
|
|
|
requester: Requester,
|
|
|
|
event_dict: dict,
|
|
|
|
txn_id: Optional[str] = None,
|
Fix historical messages backfilling in random order on remote homeservers (MSC2716) (#11114)
Fix https://github.com/matrix-org/synapse/issues/11091
Fix https://github.com/matrix-org/synapse/issues/10764 (side-stepping the issue because we no longer have to deal with `fake_prev_event_id`)
1. Made the `/backfill` response return messages in `(depth, stream_ordering)` order (previously only sorted by `depth`)
- Technically, it shouldn't really matter how `/backfill` returns things but I'm just trying to make the `stream_ordering` a little more consistent from the origin to the remote homeservers in order to get the order of messages from `/messages` consistent ([sorted by `(topological_ordering, stream_ordering)`](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)).
- Even now that we return backfilled messages in order, it still doesn't guarantee the same `stream_ordering` (and more importantly the [`/messages` order](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)) on the other server. For example, if a room has a bunch of history imported and someone visits a permalink to a historical message back in time, their homeserver will skip over the historical messages in between and insert the permalink as the next message in the `stream_order` and totally throw off the sort.
- This will be even more the case when we add the [MSC3030 jump to date API endpoint](https://github.com/matrix-org/matrix-doc/pull/3030) so the static archives can navigate and jump to a certain date.
- We're solving this in the future by switching to [online topological ordering](https://github.com/matrix-org/gomatrixserverlib/issues/187) and [chunking](https://github.com/matrix-org/synapse/issues/3785) which by its nature will apply retroactively to fix any inconsistencies introduced by people permalinking
2. As we're navigating `prev_events` to return in `/backfill`, we order by `depth` first (newest -> oldest) and now also tie-break based on the `stream_ordering` (newest -> oldest). This is technically important because MSC2716 inserts a bunch of historical messages at the same `depth` so it's best to be prescriptive about which ones we should process first. In reality, I think the code already looped over the historical messages as expected because the database is already in order.
3. Making the historical state chain and historical event chain float on their own by having no `prev_events` instead of a fake `prev_event` which caused backfill to get clogged with an unresolvable event. Fixes https://github.com/matrix-org/synapse/issues/11091 and https://github.com/matrix-org/synapse/issues/10764
4. We no longer find connected insertion events by finding a potential `prev_event` connection to the current event we're iterating over. We now solely rely on marker events which when processed, add the insertion event as an extremity and the federating homeserver can ask about it when time calls.
- Related discussion, https://github.com/matrix-org/synapse/pull/11114#discussion_r741514793
Before | After
--- | ---
![](https://user-images.githubusercontent.com/558581/139218681-b465c862-5c49-4702-a59e-466733b0cf45.png) | ![](https://user-images.githubusercontent.com/558581/146453159-a1609e0a-8324-439d-ae44-e4bce43ac6d1.png)
#### Why aren't we sorting topologically when receiving backfill events?
> The main reason we're going to opt to not sort topologically when receiving backfill events is because it's probably best to do whatever is easiest to make it just work. People will probably have opinions once they look at [MSC2716](https://github.com/matrix-org/matrix-doc/pull/2716) which could change whatever implementation anyway.
>
> As mentioned, ideally we would do this but code necessary to make the fake edges but it gets confusing and gives an impression of “just whyyyy” (feels icky). This problem also dissolves with online topological ordering.
>
> -- https://github.com/matrix-org/synapse/pull/11114#discussion_r741517138
See https://github.com/matrix-org/synapse/pull/11114#discussion_r739610091 for the technical difficulties
2022-02-07 16:54:13 -05:00
|
|
|
allow_no_prev_events: bool = False,
|
2020-09-01 08:39:04 -04:00
|
|
|
prev_event_ids: Optional[List[str]] = None,
|
2020-10-13 18:14:35 -04:00
|
|
|
auth_event_ids: Optional[List[str]] = None,
|
2022-03-25 10:21:06 -04:00
|
|
|
state_event_ids: Optional[List[str]] = None,
|
2020-07-22 12:29:15 -04:00
|
|
|
require_consent: bool = True,
|
2021-06-22 05:02:53 -04:00
|
|
|
outlier: bool = False,
|
|
|
|
depth: Optional[int] = None,
|
2022-09-28 05:39:03 -04:00
|
|
|
state_map: Optional[StateMap[str]] = None,
|
|
|
|
for_batch: bool = False,
|
|
|
|
current_state_group: Optional[int] = None,
|
2023-02-24 16:15:29 -05:00
|
|
|
) -> Tuple[EventBase, UnpersistedEventContextBase]:
|
2016-01-15 11:27:26 -05:00
|
|
|
"""
|
2022-09-28 05:39:03 -04:00
|
|
|
Given a dict from a client, create a new event. If bool for_batch is true, will
|
|
|
|
create an event using the prev_event_ids, and will create an event context for
|
|
|
|
the event using the parameters state_map and current_state_group, thus these parameters
|
|
|
|
must be provided in this case if for_batch is True. The subsequently created event
|
|
|
|
and context are suitable for being batched up and bulk persisted to the database
|
|
|
|
with other similarly created events.
|
2014-12-15 12:01:12 -05:00
|
|
|
|
|
|
|
Creates an FrozenEvent object, filling out auth_events, prev_events,
|
|
|
|
etc.
|
|
|
|
|
|
|
|
Args:
|
2017-05-02 06:36:11 -04:00
|
|
|
requester
|
2020-07-22 12:29:15 -04:00
|
|
|
event_dict: An entire event
|
|
|
|
txn_id
|
Fix historical messages backfilling in random order on remote homeservers (MSC2716) (#11114)
Fix https://github.com/matrix-org/synapse/issues/11091
Fix https://github.com/matrix-org/synapse/issues/10764 (side-stepping the issue because we no longer have to deal with `fake_prev_event_id`)
1. Made the `/backfill` response return messages in `(depth, stream_ordering)` order (previously only sorted by `depth`)
- Technically, it shouldn't really matter how `/backfill` returns things but I'm just trying to make the `stream_ordering` a little more consistent from the origin to the remote homeservers in order to get the order of messages from `/messages` consistent ([sorted by `(topological_ordering, stream_ordering)`](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)).
- Even now that we return backfilled messages in order, it still doesn't guarantee the same `stream_ordering` (and more importantly the [`/messages` order](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)) on the other server. For example, if a room has a bunch of history imported and someone visits a permalink to a historical message back in time, their homeserver will skip over the historical messages in between and insert the permalink as the next message in the `stream_order` and totally throw off the sort.
- This will be even more the case when we add the [MSC3030 jump to date API endpoint](https://github.com/matrix-org/matrix-doc/pull/3030) so the static archives can navigate and jump to a certain date.
- We're solving this in the future by switching to [online topological ordering](https://github.com/matrix-org/gomatrixserverlib/issues/187) and [chunking](https://github.com/matrix-org/synapse/issues/3785) which by its nature will apply retroactively to fix any inconsistencies introduced by people permalinking
2. As we're navigating `prev_events` to return in `/backfill`, we order by `depth` first (newest -> oldest) and now also tie-break based on the `stream_ordering` (newest -> oldest). This is technically important because MSC2716 inserts a bunch of historical messages at the same `depth` so it's best to be prescriptive about which ones we should process first. In reality, I think the code already looped over the historical messages as expected because the database is already in order.
3. Making the historical state chain and historical event chain float on their own by having no `prev_events` instead of a fake `prev_event` which caused backfill to get clogged with an unresolvable event. Fixes https://github.com/matrix-org/synapse/issues/11091 and https://github.com/matrix-org/synapse/issues/10764
4. We no longer find connected insertion events by finding a potential `prev_event` connection to the current event we're iterating over. We now solely rely on marker events which when processed, add the insertion event as an extremity and the federating homeserver can ask about it when time calls.
- Related discussion, https://github.com/matrix-org/synapse/pull/11114#discussion_r741514793
Before | After
--- | ---
![](https://user-images.githubusercontent.com/558581/139218681-b465c862-5c49-4702-a59e-466733b0cf45.png) | ![](https://user-images.githubusercontent.com/558581/146453159-a1609e0a-8324-439d-ae44-e4bce43ac6d1.png)
#### Why aren't we sorting topologically when receiving backfill events?
> The main reason we're going to opt to not sort topologically when receiving backfill events is because it's probably best to do whatever is easiest to make it just work. People will probably have opinions once they look at [MSC2716](https://github.com/matrix-org/matrix-doc/pull/2716) which could change whatever implementation anyway.
>
> As mentioned, ideally we would do this but code necessary to make the fake edges but it gets confusing and gives an impression of “just whyyyy” (feels icky). This problem also dissolves with online topological ordering.
>
> -- https://github.com/matrix-org/synapse/pull/11114#discussion_r741517138
See https://github.com/matrix-org/synapse/pull/11114#discussion_r739610091 for the technical difficulties
2022-02-07 16:54:13 -05:00
|
|
|
allow_no_prev_events: Whether to allow this event to be created an empty
|
|
|
|
list of prev_events. Normally this is prohibited just because most
|
|
|
|
events should have a prev_event and we should only use this in special
|
2023-06-16 15:12:24 -04:00
|
|
|
cases (previously useful for MSC2716).
|
2020-01-03 11:19:55 -05:00
|
|
|
prev_event_ids:
|
2018-04-16 13:41:37 -04:00
|
|
|
the forward extremities to use as the prev_events for the
|
2020-01-03 11:19:55 -05:00
|
|
|
new event.
|
2018-04-16 13:41:37 -04:00
|
|
|
|
|
|
|
If None, they will be requested from the database.
|
2020-10-13 18:14:35 -04:00
|
|
|
|
|
|
|
auth_event_ids:
|
|
|
|
The event ids to use as the auth_events for the new event.
|
|
|
|
Should normally be left as None, which will cause them to be calculated
|
|
|
|
based on the room state at the prev_events.
|
|
|
|
|
2021-06-30 07:08:42 -04:00
|
|
|
If non-None, prev_event_ids must also be provided.
|
|
|
|
|
2022-03-25 10:21:06 -04:00
|
|
|
state_event_ids:
|
2023-06-16 15:12:24 -04:00
|
|
|
The full state at a given event. This was previously used particularly
|
|
|
|
by the MSC2716 /batch_send endpoint. This should normally be left as
|
|
|
|
None, which will cause the auth_event_ids to be calculated based on the
|
|
|
|
room state at the prev_events.
|
2022-03-25 10:21:06 -04:00
|
|
|
|
2020-07-22 12:29:15 -04:00
|
|
|
require_consent: Whether to check if the requester has
|
|
|
|
consented to the privacy policy.
|
2021-06-22 05:02:53 -04:00
|
|
|
|
|
|
|
outlier: Indicates whether the event is an `outlier`, i.e. if
|
|
|
|
it's from an arbitrary point and floating in the DAG as
|
|
|
|
opposed to being inline with the current DAG.
|
2022-09-28 05:39:03 -04:00
|
|
|
|
2021-06-22 05:02:53 -04:00
|
|
|
depth: Override the depth used to order the event in the DAG.
|
|
|
|
Should normally be set to None, which will cause the depth to be calculated
|
|
|
|
based on the prev_events.
|
|
|
|
|
2022-09-28 05:39:03 -04:00
|
|
|
state_map: A state map of previously created events, used only when creating events
|
|
|
|
for batch persisting
|
|
|
|
|
|
|
|
for_batch: whether the event is being created for batch persisting to the db
|
|
|
|
|
|
|
|
current_state_group: the current state group, used only for creating events for
|
|
|
|
batch persisting
|
|
|
|
|
2018-08-16 16:25:16 -04:00
|
|
|
Raises:
|
|
|
|
ResourceLimitError if server is blocked to some resource being
|
|
|
|
exceeded
|
2022-09-28 05:39:03 -04:00
|
|
|
|
2016-01-15 11:27:26 -05:00
|
|
|
Returns:
|
2020-07-22 12:29:15 -04:00
|
|
|
Tuple of created event, Context
|
2014-12-15 12:01:12 -05:00
|
|
|
"""
|
2022-06-14 04:51:15 -04:00
|
|
|
await self.auth_blocking.check_auth_blocking(requester=requester)
|
2018-08-16 16:25:16 -04:00
|
|
|
|
2024-06-24 09:21:22 -04:00
|
|
|
if event_dict["type"] == EventTypes.Message:
|
|
|
|
requester_suspended = await self.store.get_user_suspended_status(
|
|
|
|
requester.user.to_string()
|
|
|
|
)
|
|
|
|
if requester_suspended:
|
|
|
|
raise SynapseError(
|
|
|
|
403,
|
|
|
|
"Sending messages while account is suspended is not allowed.",
|
|
|
|
Codes.USER_ACCOUNT_SUSPENDED,
|
|
|
|
)
|
|
|
|
|
2019-01-23 15:21:33 -05:00
|
|
|
if event_dict["type"] == EventTypes.Create and event_dict["state_key"] == "":
|
2021-09-29 05:57:10 -04:00
|
|
|
room_version_id = event_dict["content"]["room_version"]
|
2022-02-21 13:37:04 -05:00
|
|
|
maybe_room_version_obj = KNOWN_ROOM_VERSIONS.get(room_version_id)
|
|
|
|
if not maybe_room_version_obj:
|
2021-09-29 05:57:10 -04:00
|
|
|
# this can happen if support is withdrawn for a room version
|
|
|
|
raise UnsupportedRoomVersionError(room_version_id)
|
2022-02-21 13:37:04 -05:00
|
|
|
room_version_obj = maybe_room_version_obj
|
2019-01-23 15:21:33 -05:00
|
|
|
else:
|
|
|
|
try:
|
2021-09-29 05:57:10 -04:00
|
|
|
room_version_obj = await self.store.get_room_version(
|
2020-01-31 05:06:21 -05:00
|
|
|
event_dict["room_id"]
|
|
|
|
)
|
2019-01-23 15:21:33 -05:00
|
|
|
except NotFoundError:
|
|
|
|
raise AuthError(403, "Unknown room")
|
|
|
|
|
2021-09-29 05:57:10 -04:00
|
|
|
builder = self.event_builder_factory.for_room_version(
|
|
|
|
room_version_obj, event_dict
|
|
|
|
)
|
2014-12-04 10:50:01 -05:00
|
|
|
|
2019-01-28 12:00:14 -05:00
|
|
|
self.validator.validate_builder(builder)
|
2018-04-09 07:07:39 -04:00
|
|
|
|
2020-07-22 12:29:15 -04:00
|
|
|
is_exempt = await self._is_exempt_from_privacy_policy(builder, requester)
|
2019-03-20 13:39:29 -04:00
|
|
|
if require_consent and not is_exempt:
|
2020-07-22 12:29:15 -04:00
|
|
|
await self.assert_accepted_privacy_policy(requester)
|
2018-05-22 03:56:52 -04:00
|
|
|
|
2023-12-01 05:12:00 -05:00
|
|
|
# Save the the device ID and the transaction ID in the event internal metadata.
|
|
|
|
# This is useful to determine if we should echo the transaction_id in events.
|
2023-04-25 04:37:09 -04:00
|
|
|
# See `synapse.events.utils.EventClientSerializer.serialize_event`
|
|
|
|
if requester.device_id is not None:
|
|
|
|
builder.internal_metadata.device_id = requester.device_id
|
|
|
|
|
2018-04-09 07:07:39 -04:00
|
|
|
if txn_id is not None:
|
|
|
|
builder.internal_metadata.txn_id = txn_id
|
|
|
|
|
2021-06-22 05:02:53 -04:00
|
|
|
builder.internal_metadata.outlier = outlier
|
|
|
|
|
2023-02-09 16:05:02 -05:00
|
|
|
event, unpersisted_context = await self.create_new_client_event(
|
2020-10-13 18:14:35 -04:00
|
|
|
builder=builder,
|
|
|
|
requester=requester,
|
Fix historical messages backfilling in random order on remote homeservers (MSC2716) (#11114)
Fix https://github.com/matrix-org/synapse/issues/11091
Fix https://github.com/matrix-org/synapse/issues/10764 (side-stepping the issue because we no longer have to deal with `fake_prev_event_id`)
1. Made the `/backfill` response return messages in `(depth, stream_ordering)` order (previously only sorted by `depth`)
- Technically, it shouldn't really matter how `/backfill` returns things but I'm just trying to make the `stream_ordering` a little more consistent from the origin to the remote homeservers in order to get the order of messages from `/messages` consistent ([sorted by `(topological_ordering, stream_ordering)`](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)).
- Even now that we return backfilled messages in order, it still doesn't guarantee the same `stream_ordering` (and more importantly the [`/messages` order](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)) on the other server. For example, if a room has a bunch of history imported and someone visits a permalink to a historical message back in time, their homeserver will skip over the historical messages in between and insert the permalink as the next message in the `stream_order` and totally throw off the sort.
- This will be even more the case when we add the [MSC3030 jump to date API endpoint](https://github.com/matrix-org/matrix-doc/pull/3030) so the static archives can navigate and jump to a certain date.
- We're solving this in the future by switching to [online topological ordering](https://github.com/matrix-org/gomatrixserverlib/issues/187) and [chunking](https://github.com/matrix-org/synapse/issues/3785) which by its nature will apply retroactively to fix any inconsistencies introduced by people permalinking
2. As we're navigating `prev_events` to return in `/backfill`, we order by `depth` first (newest -> oldest) and now also tie-break based on the `stream_ordering` (newest -> oldest). This is technically important because MSC2716 inserts a bunch of historical messages at the same `depth` so it's best to be prescriptive about which ones we should process first. In reality, I think the code already looped over the historical messages as expected because the database is already in order.
3. Making the historical state chain and historical event chain float on their own by having no `prev_events` instead of a fake `prev_event` which caused backfill to get clogged with an unresolvable event. Fixes https://github.com/matrix-org/synapse/issues/11091 and https://github.com/matrix-org/synapse/issues/10764
4. We no longer find connected insertion events by finding a potential `prev_event` connection to the current event we're iterating over. We now solely rely on marker events which when processed, add the insertion event as an extremity and the federating homeserver can ask about it when time calls.
- Related discussion, https://github.com/matrix-org/synapse/pull/11114#discussion_r741514793
Before | After
--- | ---
![](https://user-images.githubusercontent.com/558581/139218681-b465c862-5c49-4702-a59e-466733b0cf45.png) | ![](https://user-images.githubusercontent.com/558581/146453159-a1609e0a-8324-439d-ae44-e4bce43ac6d1.png)
#### Why aren't we sorting topologically when receiving backfill events?
> The main reason we're going to opt to not sort topologically when receiving backfill events is because it's probably best to do whatever is easiest to make it just work. People will probably have opinions once they look at [MSC2716](https://github.com/matrix-org/matrix-doc/pull/2716) which could change whatever implementation anyway.
>
> As mentioned, ideally we would do this but code necessary to make the fake edges but it gets confusing and gives an impression of “just whyyyy” (feels icky). This problem also dissolves with online topological ordering.
>
> -- https://github.com/matrix-org/synapse/pull/11114#discussion_r741517138
See https://github.com/matrix-org/synapse/pull/11114#discussion_r739610091 for the technical difficulties
2022-02-07 16:54:13 -05:00
|
|
|
allow_no_prev_events=allow_no_prev_events,
|
2020-10-13 18:14:35 -04:00
|
|
|
prev_event_ids=prev_event_ids,
|
|
|
|
auth_event_ids=auth_event_ids,
|
2022-03-25 10:21:06 -04:00
|
|
|
state_event_ids=state_event_ids,
|
2021-06-22 05:02:53 -04:00
|
|
|
depth=depth,
|
2022-09-28 05:39:03 -04:00
|
|
|
state_map=state_map,
|
|
|
|
for_batch=for_batch,
|
|
|
|
current_state_group=current_state_group,
|
2018-04-09 07:07:39 -04:00
|
|
|
)
|
2015-01-28 11:58:23 -05:00
|
|
|
|
2019-05-08 12:01:30 -04:00
|
|
|
# In an ideal world we wouldn't need the second part of this condition. However,
|
|
|
|
# this behaviour isn't spec'd yet, meaning we should be able to deactivate this
|
|
|
|
# behaviour. Another reason is that this code is also evaluated each time a new
|
|
|
|
# m.room.aliases event is created, which includes hitting a /directory route.
|
|
|
|
# Therefore not including this condition here would render the similar one in
|
|
|
|
# synapse.handlers.directory pointless.
|
|
|
|
if builder.type == EventTypes.Aliases and self.require_membership_for_aliases:
|
|
|
|
# Ideally we'd do the membership check in event_auth.check(), which
|
|
|
|
# describes a spec'd algorithm for authenticating events received over
|
|
|
|
# federation as well as those created locally. As of room v3, aliases events
|
|
|
|
# can be created by users that are not in the room, therefore we have to
|
|
|
|
# tolerate them in event_auth.check().
|
2022-09-28 05:39:03 -04:00
|
|
|
if for_batch:
|
|
|
|
assert state_map is not None
|
|
|
|
prev_event_id = state_map.get((EventTypes.Member, event.sender))
|
|
|
|
else:
|
2023-02-24 16:15:29 -05:00
|
|
|
prev_state_ids = await unpersisted_context.get_prev_state_ids(
|
2023-07-20 05:46:37 -04:00
|
|
|
StateFilter.from_types([(EventTypes.Member, event.sender)])
|
2022-09-28 05:39:03 -04:00
|
|
|
)
|
|
|
|
prev_event_id = prev_state_ids.get((EventTypes.Member, event.sender))
|
2019-07-24 08:16:18 -04:00
|
|
|
prev_event = (
|
2020-07-22 12:29:15 -04:00
|
|
|
await self.store.get_event(prev_event_id, allow_none=True)
|
2019-07-24 08:16:18 -04:00
|
|
|
if prev_event_id
|
|
|
|
else None
|
|
|
|
)
|
2019-05-08 12:01:30 -04:00
|
|
|
if not prev_event or prev_event.membership != Membership.JOIN:
|
|
|
|
logger.warning(
|
2019-06-20 05:32:02 -04:00
|
|
|
(
|
|
|
|
"Attempt to send `m.room.aliases` in room %s by user %s but"
|
|
|
|
" membership is %s"
|
|
|
|
),
|
2019-05-08 12:01:30 -04:00
|
|
|
event.room_id,
|
|
|
|
event.sender,
|
|
|
|
prev_event.membership if prev_event else None,
|
|
|
|
)
|
|
|
|
|
|
|
|
raise AuthError(
|
2019-06-20 05:32:02 -04:00
|
|
|
403, "You must be in the room to create an alias for it"
|
2019-05-08 12:01:30 -04:00
|
|
|
)
|
|
|
|
|
2019-11-04 12:09:22 -05:00
|
|
|
self.validator.validate_new(event, self.config)
|
2023-02-24 16:15:29 -05:00
|
|
|
return event, unpersisted_context
|
2016-01-15 11:27:26 -05:00
|
|
|
|
2020-07-22 12:29:15 -04:00
|
|
|
async def _is_exempt_from_privacy_policy(
|
|
|
|
self, builder: EventBuilder, requester: Requester
|
|
|
|
) -> bool:
|
2021-02-16 17:32:34 -05:00
|
|
|
""" "Determine if an event to be sent is exempt from having to consent
|
2018-05-22 03:56:52 -04:00
|
|
|
to the privacy policy
|
|
|
|
|
|
|
|
Args:
|
2020-07-22 12:29:15 -04:00
|
|
|
builder: event being created
|
|
|
|
requester: user requesting this event
|
2018-05-22 03:56:52 -04:00
|
|
|
|
|
|
|
Returns:
|
2020-07-22 12:29:15 -04:00
|
|
|
true if the event can be sent without the user consenting
|
2018-05-22 03:56:52 -04:00
|
|
|
"""
|
|
|
|
# the only thing the user can do is join the server notices room.
|
|
|
|
if builder.type == EventTypes.Member:
|
|
|
|
membership = builder.content.get("membership", None)
|
|
|
|
if membership == Membership.JOIN:
|
2022-09-14 11:53:18 -04:00
|
|
|
return await self.store.is_server_notice_room(builder.room_id)
|
2018-06-25 12:56:10 -04:00
|
|
|
elif membership == Membership.LEAVE:
|
|
|
|
# the user is always allowed to leave (but not kick people)
|
|
|
|
return builder.state_key == requester.user.to_string()
|
2020-07-22 12:29:15 -04:00
|
|
|
return False
|
2018-05-22 03:56:52 -04:00
|
|
|
|
2020-07-22 12:29:15 -04:00
|
|
|
async def assert_accepted_privacy_policy(self, requester: Requester) -> None:
|
2018-05-22 03:56:52 -04:00
|
|
|
"""Check if a user has accepted the privacy policy
|
|
|
|
|
|
|
|
Called when the given user is about to do something that requires
|
|
|
|
privacy consent. We see if the user is exempt and otherwise check that
|
|
|
|
they have given consent. If they have not, a ConsentNotGiven error is
|
|
|
|
raised.
|
|
|
|
|
|
|
|
Args:
|
2020-07-22 12:29:15 -04:00
|
|
|
requester: The user making the request
|
2018-05-22 03:56:52 -04:00
|
|
|
|
|
|
|
Returns:
|
2020-07-22 12:29:15 -04:00
|
|
|
Returns normally if the user has consented or is exempt
|
2018-05-22 03:56:52 -04:00
|
|
|
|
|
|
|
Raises:
|
|
|
|
ConsentNotGivenError: if the user has not given consent yet
|
|
|
|
"""
|
2019-03-19 07:38:59 -04:00
|
|
|
if self._block_events_without_consent_error is None:
|
2018-05-22 03:56:52 -04:00
|
|
|
return
|
|
|
|
|
2019-03-20 13:51:27 -04:00
|
|
|
# exempt AS users from needing consent
|
|
|
|
if requester.app_service is not None:
|
|
|
|
return
|
|
|
|
|
2020-11-17 05:51:25 -05:00
|
|
|
user_id = requester.authenticated_entity
|
|
|
|
if not user_id.startswith("@"):
|
|
|
|
# The authenticated entity might not be a user, e.g. if it's the
|
|
|
|
# server puppetting the user.
|
|
|
|
return
|
|
|
|
|
|
|
|
user = UserID.from_string(user_id)
|
2018-05-22 03:56:52 -04:00
|
|
|
|
|
|
|
# exempt the system notices user
|
|
|
|
if (
|
2021-09-24 07:25:21 -04:00
|
|
|
self.config.servernotices.server_notices_mxid is not None
|
|
|
|
and user_id == self.config.servernotices.server_notices_mxid
|
2018-05-22 03:56:52 -04:00
|
|
|
):
|
|
|
|
return
|
|
|
|
|
2020-07-22 12:29:15 -04:00
|
|
|
u = await self.store.get_user_by_id(user_id)
|
2018-05-22 03:56:52 -04:00
|
|
|
assert u is not None
|
2023-09-14 07:46:30 -04:00
|
|
|
if u.user_type in (UserTypes.SUPPORT, UserTypes.BOT):
|
2019-08-23 05:28:54 -04:00
|
|
|
# support and bot users are not required to consent
|
2019-08-23 04:14:52 -04:00
|
|
|
return
|
2023-09-14 07:46:30 -04:00
|
|
|
if u.appservice_id is not None:
|
2018-05-29 14:54:32 -04:00
|
|
|
# users registered by an appservice are exempt
|
|
|
|
return
|
2023-09-14 07:46:30 -04:00
|
|
|
if u.consent_version == self.config.consent.user_consent_version:
|
2018-05-22 03:56:52 -04:00
|
|
|
return
|
|
|
|
|
2020-11-17 05:51:25 -05:00
|
|
|
consent_uri = self._consent_uri_builder.build_user_consent_uri(user.localpart)
|
2019-06-20 05:32:02 -04:00
|
|
|
msg = self._block_events_without_consent_error % {"consent_uri": consent_uri}
|
|
|
|
raise ConsentNotGivenError(msg=msg, consent_uri=consent_uri)
|
2018-05-22 03:56:52 -04:00
|
|
|
|
2020-07-22 12:29:15 -04:00
|
|
|
async def deduplicate_state_event(
|
|
|
|
self, event: EventBase, context: EventContext
|
2020-08-18 07:53:23 -04:00
|
|
|
) -> Optional[EventBase]:
|
2016-02-16 09:25:23 -05:00
|
|
|
"""
|
|
|
|
Checks whether event is in the latest resolved state in context.
|
|
|
|
|
2020-08-18 07:53:23 -04:00
|
|
|
Args:
|
|
|
|
event: The event to check for duplication.
|
|
|
|
context: The event context.
|
|
|
|
|
|
|
|
Returns:
|
2020-10-23 12:38:40 -04:00
|
|
|
The previous version of the event is returned, if it is found in the
|
2020-08-18 07:53:23 -04:00
|
|
|
event context. Otherwise, None is returned.
|
2016-02-16 09:25:23 -05:00
|
|
|
"""
|
2022-05-10 15:43:13 -04:00
|
|
|
if event.internal_metadata.is_outlier():
|
|
|
|
# This can happen due to out of band memberships
|
|
|
|
return None
|
|
|
|
|
2022-05-20 04:54:12 -04:00
|
|
|
prev_state_ids = await context.get_prev_state_ids(
|
2023-07-20 05:46:37 -04:00
|
|
|
StateFilter.from_types([(event.type, event.state_key)])
|
2022-05-20 04:54:12 -04:00
|
|
|
)
|
2018-07-23 08:00:22 -04:00
|
|
|
prev_event_id = prev_state_ids.get((event.type, event.state_key))
|
2019-07-24 08:16:18 -04:00
|
|
|
if not prev_event_id:
|
2020-08-18 07:53:23 -04:00
|
|
|
return None
|
2020-07-22 12:29:15 -04:00
|
|
|
prev_event = await self.store.get_event(prev_event_id, allow_none=True)
|
2016-08-25 12:32:22 -04:00
|
|
|
if not prev_event:
|
2020-08-18 07:53:23 -04:00
|
|
|
return None
|
2016-08-25 12:32:22 -04:00
|
|
|
|
2016-02-16 09:25:23 -05:00
|
|
|
if prev_event and event.user_id == prev_event.user_id:
|
|
|
|
prev_content = encode_canonical_json(prev_event.content)
|
2016-02-15 13:21:30 -05:00
|
|
|
next_content = encode_canonical_json(event.content)
|
|
|
|
if prev_content == next_content:
|
2019-07-23 09:00:55 -04:00
|
|
|
return prev_event
|
2020-08-18 07:53:23 -04:00
|
|
|
return None
|
2016-02-15 13:21:30 -05:00
|
|
|
|
2023-07-31 08:44:45 -04:00
|
|
|
async def get_event_id_from_transaction(
|
2022-11-03 12:21:31 -04:00
|
|
|
self,
|
|
|
|
requester: Requester,
|
|
|
|
txn_id: str,
|
|
|
|
room_id: str,
|
2023-07-31 08:44:45 -04:00
|
|
|
) -> Optional[str]:
|
|
|
|
"""For the given transaction ID and room ID, check if there is a matching event ID.
|
2022-11-03 12:21:31 -04:00
|
|
|
|
|
|
|
Args:
|
|
|
|
requester: The requester making the request in the context of which we want
|
|
|
|
to fetch the event.
|
|
|
|
txn_id: The transaction ID.
|
|
|
|
room_id: The room ID.
|
|
|
|
|
|
|
|
Returns:
|
2023-07-31 08:44:45 -04:00
|
|
|
An event ID if one could be found, None otherwise.
|
2022-11-03 12:21:31 -04:00
|
|
|
"""
|
2023-07-31 08:44:45 -04:00
|
|
|
existing_event_id = None
|
2023-04-25 04:37:09 -04:00
|
|
|
|
2023-08-04 07:47:18 -04:00
|
|
|
# According to the spec, transactions are scoped to a user's device ID.
|
|
|
|
if requester.device_id:
|
2023-04-25 04:37:09 -04:00
|
|
|
existing_event_id = (
|
|
|
|
await self.store.get_event_id_from_transaction_id_and_device_id(
|
|
|
|
room_id,
|
|
|
|
requester.user.to_string(),
|
|
|
|
requester.device_id,
|
|
|
|
txn_id,
|
|
|
|
)
|
|
|
|
)
|
|
|
|
if existing_event_id:
|
2023-07-31 08:44:45 -04:00
|
|
|
return existing_event_id
|
2023-04-25 04:37:09 -04:00
|
|
|
|
2023-07-31 08:44:45 -04:00
|
|
|
return existing_event_id
|
|
|
|
|
|
|
|
async def get_event_from_transaction(
|
|
|
|
self,
|
|
|
|
requester: Requester,
|
|
|
|
txn_id: str,
|
|
|
|
room_id: str,
|
|
|
|
) -> Optional[EventBase]:
|
|
|
|
"""For the given transaction ID and room ID, check if there is a matching event.
|
|
|
|
If so, fetch it and return it.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
requester: The requester making the request in the context of which we want
|
|
|
|
to fetch the event.
|
|
|
|
txn_id: The transaction ID.
|
|
|
|
room_id: The room ID.
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
An event if one could be found, None otherwise.
|
|
|
|
"""
|
|
|
|
existing_event_id = await self.get_event_id_from_transaction(
|
|
|
|
requester, txn_id, room_id
|
|
|
|
)
|
|
|
|
if existing_event_id:
|
|
|
|
return await self.store.get_event(existing_event_id)
|
2022-11-03 12:21:31 -04:00
|
|
|
return None
|
|
|
|
|
2020-05-01 10:15:36 -04:00
|
|
|
async def create_and_send_nonmember_event(
|
2020-07-22 12:29:15 -04:00
|
|
|
self,
|
|
|
|
requester: Requester,
|
2020-08-12 10:05:50 -04:00
|
|
|
event_dict: dict,
|
Fix historical messages backfilling in random order on remote homeservers (MSC2716) (#11114)
Fix https://github.com/matrix-org/synapse/issues/11091
Fix https://github.com/matrix-org/synapse/issues/10764 (side-stepping the issue because we no longer have to deal with `fake_prev_event_id`)
1. Made the `/backfill` response return messages in `(depth, stream_ordering)` order (previously only sorted by `depth`)
- Technically, it shouldn't really matter how `/backfill` returns things but I'm just trying to make the `stream_ordering` a little more consistent from the origin to the remote homeservers in order to get the order of messages from `/messages` consistent ([sorted by `(topological_ordering, stream_ordering)`](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)).
- Even now that we return backfilled messages in order, it still doesn't guarantee the same `stream_ordering` (and more importantly the [`/messages` order](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)) on the other server. For example, if a room has a bunch of history imported and someone visits a permalink to a historical message back in time, their homeserver will skip over the historical messages in between and insert the permalink as the next message in the `stream_order` and totally throw off the sort.
- This will be even more the case when we add the [MSC3030 jump to date API endpoint](https://github.com/matrix-org/matrix-doc/pull/3030) so the static archives can navigate and jump to a certain date.
- We're solving this in the future by switching to [online topological ordering](https://github.com/matrix-org/gomatrixserverlib/issues/187) and [chunking](https://github.com/matrix-org/synapse/issues/3785) which by its nature will apply retroactively to fix any inconsistencies introduced by people permalinking
2. As we're navigating `prev_events` to return in `/backfill`, we order by `depth` first (newest -> oldest) and now also tie-break based on the `stream_ordering` (newest -> oldest). This is technically important because MSC2716 inserts a bunch of historical messages at the same `depth` so it's best to be prescriptive about which ones we should process first. In reality, I think the code already looped over the historical messages as expected because the database is already in order.
3. Making the historical state chain and historical event chain float on their own by having no `prev_events` instead of a fake `prev_event` which caused backfill to get clogged with an unresolvable event. Fixes https://github.com/matrix-org/synapse/issues/11091 and https://github.com/matrix-org/synapse/issues/10764
4. We no longer find connected insertion events by finding a potential `prev_event` connection to the current event we're iterating over. We now solely rely on marker events which when processed, add the insertion event as an extremity and the federating homeserver can ask about it when time calls.
- Related discussion, https://github.com/matrix-org/synapse/pull/11114#discussion_r741514793
Before | After
--- | ---
![](https://user-images.githubusercontent.com/558581/139218681-b465c862-5c49-4702-a59e-466733b0cf45.png) | ![](https://user-images.githubusercontent.com/558581/146453159-a1609e0a-8324-439d-ae44-e4bce43ac6d1.png)
#### Why aren't we sorting topologically when receiving backfill events?
> The main reason we're going to opt to not sort topologically when receiving backfill events is because it's probably best to do whatever is easiest to make it just work. People will probably have opinions once they look at [MSC2716](https://github.com/matrix-org/matrix-doc/pull/2716) which could change whatever implementation anyway.
>
> As mentioned, ideally we would do this but code necessary to make the fake edges but it gets confusing and gives an impression of “just whyyyy” (feels icky). This problem also dissolves with online topological ordering.
>
> -- https://github.com/matrix-org/synapse/pull/11114#discussion_r741517138
See https://github.com/matrix-org/synapse/pull/11114#discussion_r739610091 for the technical difficulties
2022-02-07 16:54:13 -05:00
|
|
|
allow_no_prev_events: bool = False,
|
2021-06-22 05:02:53 -04:00
|
|
|
prev_event_ids: Optional[List[str]] = None,
|
2022-03-25 10:21:06 -04:00
|
|
|
state_event_ids: Optional[List[str]] = None,
|
2020-07-22 12:29:15 -04:00
|
|
|
ratelimit: bool = True,
|
|
|
|
txn_id: Optional[str] = None,
|
2020-08-24 13:58:56 -04:00
|
|
|
ignore_shadow_ban: bool = False,
|
2021-06-22 05:02:53 -04:00
|
|
|
outlier: bool = False,
|
|
|
|
depth: Optional[int] = None,
|
2020-05-22 09:21:54 -04:00
|
|
|
) -> Tuple[EventBase, int]:
|
2016-01-15 11:27:26 -05:00
|
|
|
"""
|
|
|
|
Creates an event, then sends it.
|
|
|
|
|
2020-10-02 13:10:55 -04:00
|
|
|
See self.create_event and self.handle_new_client_event.
|
2020-08-24 13:58:56 -04:00
|
|
|
|
2020-08-25 10:52:15 -04:00
|
|
|
Args:
|
|
|
|
requester: The requester sending the event.
|
|
|
|
event_dict: An entire event.
|
Fix historical messages backfilling in random order on remote homeservers (MSC2716) (#11114)
Fix https://github.com/matrix-org/synapse/issues/11091
Fix https://github.com/matrix-org/synapse/issues/10764 (side-stepping the issue because we no longer have to deal with `fake_prev_event_id`)
1. Made the `/backfill` response return messages in `(depth, stream_ordering)` order (previously only sorted by `depth`)
- Technically, it shouldn't really matter how `/backfill` returns things but I'm just trying to make the `stream_ordering` a little more consistent from the origin to the remote homeservers in order to get the order of messages from `/messages` consistent ([sorted by `(topological_ordering, stream_ordering)`](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)).
- Even now that we return backfilled messages in order, it still doesn't guarantee the same `stream_ordering` (and more importantly the [`/messages` order](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)) on the other server. For example, if a room has a bunch of history imported and someone visits a permalink to a historical message back in time, their homeserver will skip over the historical messages in between and insert the permalink as the next message in the `stream_order` and totally throw off the sort.
- This will be even more the case when we add the [MSC3030 jump to date API endpoint](https://github.com/matrix-org/matrix-doc/pull/3030) so the static archives can navigate and jump to a certain date.
- We're solving this in the future by switching to [online topological ordering](https://github.com/matrix-org/gomatrixserverlib/issues/187) and [chunking](https://github.com/matrix-org/synapse/issues/3785) which by its nature will apply retroactively to fix any inconsistencies introduced by people permalinking
2. As we're navigating `prev_events` to return in `/backfill`, we order by `depth` first (newest -> oldest) and now also tie-break based on the `stream_ordering` (newest -> oldest). This is technically important because MSC2716 inserts a bunch of historical messages at the same `depth` so it's best to be prescriptive about which ones we should process first. In reality, I think the code already looped over the historical messages as expected because the database is already in order.
3. Making the historical state chain and historical event chain float on their own by having no `prev_events` instead of a fake `prev_event` which caused backfill to get clogged with an unresolvable event. Fixes https://github.com/matrix-org/synapse/issues/11091 and https://github.com/matrix-org/synapse/issues/10764
4. We no longer find connected insertion events by finding a potential `prev_event` connection to the current event we're iterating over. We now solely rely on marker events which when processed, add the insertion event as an extremity and the federating homeserver can ask about it when time calls.
- Related discussion, https://github.com/matrix-org/synapse/pull/11114#discussion_r741514793
Before | After
--- | ---
![](https://user-images.githubusercontent.com/558581/139218681-b465c862-5c49-4702-a59e-466733b0cf45.png) | ![](https://user-images.githubusercontent.com/558581/146453159-a1609e0a-8324-439d-ae44-e4bce43ac6d1.png)
#### Why aren't we sorting topologically when receiving backfill events?
> The main reason we're going to opt to not sort topologically when receiving backfill events is because it's probably best to do whatever is easiest to make it just work. People will probably have opinions once they look at [MSC2716](https://github.com/matrix-org/matrix-doc/pull/2716) which could change whatever implementation anyway.
>
> As mentioned, ideally we would do this but code necessary to make the fake edges but it gets confusing and gives an impression of “just whyyyy” (feels icky). This problem also dissolves with online topological ordering.
>
> -- https://github.com/matrix-org/synapse/pull/11114#discussion_r741517138
See https://github.com/matrix-org/synapse/pull/11114#discussion_r739610091 for the technical difficulties
2022-02-07 16:54:13 -05:00
|
|
|
allow_no_prev_events: Whether to allow this event to be created an empty
|
|
|
|
list of prev_events. Normally this is prohibited just because most
|
|
|
|
events should have a prev_event and we should only use this in special
|
2023-06-16 15:12:24 -04:00
|
|
|
cases (previously useful for MSC2716).
|
2021-06-22 05:02:53 -04:00
|
|
|
prev_event_ids:
|
|
|
|
The event IDs to use as the prev events.
|
|
|
|
Should normally be left as None to automatically request them
|
|
|
|
from the database.
|
2022-03-25 10:21:06 -04:00
|
|
|
state_event_ids:
|
2023-06-16 15:12:24 -04:00
|
|
|
The full state at a given event. This was previously used particularly
|
|
|
|
by the MSC2716 /batch_send endpoint. This should normally be left as
|
|
|
|
None, which will cause the auth_event_ids to be calculated based on the
|
|
|
|
room state at the prev_events.
|
2020-08-25 10:52:15 -04:00
|
|
|
ratelimit: Whether to rate limit this send.
|
|
|
|
txn_id: The transaction ID.
|
|
|
|
ignore_shadow_ban: True if shadow-banned users should be allowed to
|
|
|
|
send this event.
|
2021-06-22 05:02:53 -04:00
|
|
|
outlier: Indicates whether the event is an `outlier`, i.e. if
|
|
|
|
it's from an arbitrary point and floating in the DAG as
|
|
|
|
opposed to being inline with the current DAG.
|
|
|
|
depth: Override the depth used to order the event in the DAG.
|
|
|
|
Should normally be set to None, which will cause the depth to be calculated
|
|
|
|
based on the prev_events.
|
2020-08-25 10:52:15 -04:00
|
|
|
|
2020-10-02 13:10:55 -04:00
|
|
|
Returns:
|
2020-10-13 07:07:56 -04:00
|
|
|
The event, and its stream ordering (if deduplication happened,
|
2020-10-02 13:10:55 -04:00
|
|
|
the previous, duplicate event).
|
|
|
|
|
2020-08-24 13:58:56 -04:00
|
|
|
Raises:
|
|
|
|
ShadowBanError if the requester has been shadow-banned.
|
2016-01-15 11:27:26 -05:00
|
|
|
"""
|
2020-10-02 13:10:55 -04:00
|
|
|
|
|
|
|
if event_dict["type"] == EventTypes.Member:
|
|
|
|
raise SynapseError(
|
|
|
|
500, "Tried to send member event through non-member codepath"
|
|
|
|
)
|
|
|
|
|
2020-08-24 13:58:56 -04:00
|
|
|
if not ignore_shadow_ban and requester.shadow_banned:
|
|
|
|
# We randomly sleep a bit just to annoy the requester.
|
|
|
|
await self.clock.sleep(random.randint(1, 10))
|
|
|
|
raise ShadowBanError()
|
2017-09-19 07:20:11 -04:00
|
|
|
|
2022-06-30 12:22:40 -04:00
|
|
|
if ratelimit:
|
2023-10-27 07:52:40 -04:00
|
|
|
room_id = event_dict["room_id"]
|
|
|
|
try:
|
|
|
|
room_version = await self.store.get_room_version(room_id)
|
|
|
|
except NotFoundError:
|
|
|
|
# The room doesn't exist.
|
|
|
|
raise AuthError(403, f"User {requester.user} not in room {room_id}")
|
|
|
|
|
|
|
|
if room_version.updated_redaction_rules:
|
|
|
|
redacts = event_dict["content"].get("redacts")
|
|
|
|
else:
|
|
|
|
redacts = event_dict.get("redacts")
|
|
|
|
|
|
|
|
is_admin_redaction = await self.is_admin_redaction(
|
|
|
|
event_type=event_dict["type"],
|
|
|
|
sender=event_dict["sender"],
|
|
|
|
redacts=redacts,
|
|
|
|
)
|
|
|
|
await self.request_ratelimiter.ratelimit(
|
|
|
|
requester, is_admin_redaction=is_admin_redaction, update=False
|
|
|
|
)
|
2022-06-30 12:22:40 -04:00
|
|
|
|
2018-04-10 09:00:24 -04:00
|
|
|
# We limit the number of concurrent event sends in a room so that we
|
|
|
|
# don't fork the DAG too much. If we don't limit then we can end up in
|
|
|
|
# a situation where event persistence can't keep up, causing
|
|
|
|
# extremities to pile up, which in turn leads to state resolution
|
|
|
|
# taking longer.
|
2023-03-21 05:13:43 -04:00
|
|
|
room_id = event_dict["room_id"]
|
|
|
|
async with self.limiter.queue(room_id):
|
2022-11-03 12:21:31 -04:00
|
|
|
if txn_id:
|
|
|
|
event = await self.get_event_from_transaction(
|
2023-03-21 05:13:43 -04:00
|
|
|
requester, txn_id, room_id
|
2020-10-13 07:07:56 -04:00
|
|
|
)
|
2022-11-03 12:21:31 -04:00
|
|
|
if event:
|
2020-10-13 07:07:56 -04:00
|
|
|
# we know it was persisted, so must have a stream ordering
|
|
|
|
assert event.internal_metadata.stream_ordering
|
2022-11-03 12:21:31 -04:00
|
|
|
return (
|
|
|
|
event,
|
|
|
|
event.internal_metadata.stream_ordering,
|
|
|
|
)
|
2020-10-13 07:07:56 -04:00
|
|
|
|
2023-07-31 05:58:03 -04:00
|
|
|
async with self._worker_lock_handler.acquire_read_write_lock(
|
2023-08-16 10:19:54 -04:00
|
|
|
NEW_EVENT_DURING_PURGE_LOCK_NAME, room_id, write=False
|
2023-07-31 05:58:03 -04:00
|
|
|
):
|
|
|
|
return await self._create_and_send_nonmember_event_locked(
|
|
|
|
requester=requester,
|
|
|
|
event_dict=event_dict,
|
|
|
|
allow_no_prev_events=allow_no_prev_events,
|
|
|
|
prev_event_ids=prev_event_ids,
|
|
|
|
state_event_ids=state_event_ids,
|
|
|
|
ratelimit=ratelimit,
|
|
|
|
txn_id=txn_id,
|
|
|
|
ignore_shadow_ban=ignore_shadow_ban,
|
|
|
|
outlier=outlier,
|
|
|
|
depth=depth,
|
|
|
|
)
|
|
|
|
|
|
|
|
async def _create_and_send_nonmember_event_locked(
|
|
|
|
self,
|
|
|
|
requester: Requester,
|
|
|
|
event_dict: dict,
|
|
|
|
allow_no_prev_events: bool = False,
|
|
|
|
prev_event_ids: Optional[List[str]] = None,
|
|
|
|
state_event_ids: Optional[List[str]] = None,
|
|
|
|
ratelimit: bool = True,
|
|
|
|
txn_id: Optional[str] = None,
|
|
|
|
ignore_shadow_ban: bool = False,
|
|
|
|
outlier: bool = False,
|
|
|
|
depth: Optional[int] = None,
|
|
|
|
) -> Tuple[EventBase, int]:
|
|
|
|
room_id = event_dict["room_id"]
|
|
|
|
|
2023-03-21 05:13:43 -04:00
|
|
|
# If we don't have any prev event IDs specified then we need to
|
|
|
|
# check that the host is in the room (as otherwise populating the
|
|
|
|
# prev events will fail), at which point we may as well check the
|
|
|
|
# local user is in the room.
|
|
|
|
if not prev_event_ids:
|
|
|
|
user_id = requester.user.to_string()
|
|
|
|
is_user_in_room = await self.store.check_local_user_in_room(
|
|
|
|
user_id, room_id
|
|
|
|
)
|
|
|
|
if not is_user_in_room:
|
|
|
|
raise AuthError(403, f"User {user_id} not in room {room_id}")
|
|
|
|
|
2022-12-15 11:04:23 -05:00
|
|
|
# Try several times, it could fail with PartialStateConflictError
|
|
|
|
# in handle_new_client_event, cf comment in except block.
|
|
|
|
max_retries = 5
|
|
|
|
for i in range(max_retries):
|
|
|
|
try:
|
2023-02-24 16:15:29 -05:00
|
|
|
event, unpersisted_context = await self.create_event(
|
2022-12-15 11:04:23 -05:00
|
|
|
requester,
|
|
|
|
event_dict,
|
|
|
|
txn_id=txn_id,
|
|
|
|
allow_no_prev_events=allow_no_prev_events,
|
|
|
|
prev_event_ids=prev_event_ids,
|
|
|
|
state_event_ids=state_event_ids,
|
|
|
|
outlier=outlier,
|
|
|
|
depth=depth,
|
|
|
|
)
|
2023-02-24 16:15:29 -05:00
|
|
|
context = await unpersisted_context.persist(event)
|
2017-09-19 07:20:11 -04:00
|
|
|
|
2022-12-15 11:04:23 -05:00
|
|
|
assert self.hs.is_mine_id(event.sender), "User must be our own: %s" % (
|
|
|
|
event.sender,
|
|
|
|
)
|
2020-10-05 14:00:50 -04:00
|
|
|
|
2023-04-17 20:57:40 -04:00
|
|
|
spam_check_result = (
|
|
|
|
await self._spam_checker_module_callbacks.check_event_for_spam(
|
|
|
|
event
|
|
|
|
)
|
|
|
|
)
|
|
|
|
if spam_check_result != self._spam_checker_module_callbacks.NOT_SPAM:
|
2022-12-15 11:04:23 -05:00
|
|
|
if isinstance(spam_check_result, tuple):
|
|
|
|
try:
|
|
|
|
[code, dict] = spam_check_result
|
|
|
|
raise SynapseError(
|
|
|
|
403,
|
|
|
|
"This message had been rejected as probable spam",
|
|
|
|
code,
|
|
|
|
dict,
|
|
|
|
)
|
|
|
|
except ValueError:
|
|
|
|
logger.error(
|
|
|
|
"Spam-check module returned invalid error value. Expecting [code, dict], got %s",
|
|
|
|
spam_check_result,
|
|
|
|
)
|
2022-05-31 09:48:22 -04:00
|
|
|
|
2022-12-15 11:04:23 -05:00
|
|
|
raise SynapseError(
|
|
|
|
403,
|
|
|
|
"This message has been rejected as probable spam",
|
|
|
|
Codes.FORBIDDEN,
|
|
|
|
)
|
2022-05-31 06:04:53 -04:00
|
|
|
|
2022-12-15 11:04:23 -05:00
|
|
|
# Backwards compatibility: if the return value is not an error code, it
|
|
|
|
# means the module returned an error message to be included in the
|
|
|
|
# SynapseError (which is now deprecated).
|
|
|
|
raise SynapseError(
|
|
|
|
403,
|
|
|
|
spam_check_result,
|
|
|
|
Codes.FORBIDDEN,
|
|
|
|
)
|
|
|
|
|
|
|
|
ev = await self.handle_new_client_event(
|
|
|
|
requester=requester,
|
|
|
|
events_and_context=[(event, context)],
|
|
|
|
ratelimit=ratelimit,
|
|
|
|
ignore_shadow_ban=ignore_shadow_ban,
|
2022-05-23 13:27:39 -04:00
|
|
|
)
|
2018-04-09 07:07:39 -04:00
|
|
|
|
2022-12-15 11:04:23 -05:00
|
|
|
break
|
|
|
|
except PartialStateConflictError as e:
|
|
|
|
# Persisting couldn't happen because the room got un-partial stated
|
|
|
|
# in the meantime and context needs to be recomputed, so let's do so.
|
|
|
|
if i == max_retries - 1:
|
|
|
|
raise e
|
2020-10-02 13:10:55 -04:00
|
|
|
|
|
|
|
# we know it was persisted, so must have a stream ordering
|
|
|
|
assert ev.internal_metadata.stream_ordering
|
|
|
|
return ev, ev.internal_metadata.stream_ordering
|
2014-12-04 10:50:01 -05:00
|
|
|
|
2018-02-06 11:31:50 -05:00
|
|
|
@measure_func("create_new_client_event")
|
2020-07-22 12:29:15 -04:00
|
|
|
async def create_new_client_event(
|
|
|
|
self,
|
|
|
|
builder: EventBuilder,
|
|
|
|
requester: Optional[Requester] = None,
|
Fix historical messages backfilling in random order on remote homeservers (MSC2716) (#11114)
Fix https://github.com/matrix-org/synapse/issues/11091
Fix https://github.com/matrix-org/synapse/issues/10764 (side-stepping the issue because we no longer have to deal with `fake_prev_event_id`)
1. Made the `/backfill` response return messages in `(depth, stream_ordering)` order (previously only sorted by `depth`)
- Technically, it shouldn't really matter how `/backfill` returns things but I'm just trying to make the `stream_ordering` a little more consistent from the origin to the remote homeservers in order to get the order of messages from `/messages` consistent ([sorted by `(topological_ordering, stream_ordering)`](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)).
- Even now that we return backfilled messages in order, it still doesn't guarantee the same `stream_ordering` (and more importantly the [`/messages` order](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)) on the other server. For example, if a room has a bunch of history imported and someone visits a permalink to a historical message back in time, their homeserver will skip over the historical messages in between and insert the permalink as the next message in the `stream_order` and totally throw off the sort.
- This will be even more the case when we add the [MSC3030 jump to date API endpoint](https://github.com/matrix-org/matrix-doc/pull/3030) so the static archives can navigate and jump to a certain date.
- We're solving this in the future by switching to [online topological ordering](https://github.com/matrix-org/gomatrixserverlib/issues/187) and [chunking](https://github.com/matrix-org/synapse/issues/3785) which by its nature will apply retroactively to fix any inconsistencies introduced by people permalinking
2. As we're navigating `prev_events` to return in `/backfill`, we order by `depth` first (newest -> oldest) and now also tie-break based on the `stream_ordering` (newest -> oldest). This is technically important because MSC2716 inserts a bunch of historical messages at the same `depth` so it's best to be prescriptive about which ones we should process first. In reality, I think the code already looped over the historical messages as expected because the database is already in order.
3. Making the historical state chain and historical event chain float on their own by having no `prev_events` instead of a fake `prev_event` which caused backfill to get clogged with an unresolvable event. Fixes https://github.com/matrix-org/synapse/issues/11091 and https://github.com/matrix-org/synapse/issues/10764
4. We no longer find connected insertion events by finding a potential `prev_event` connection to the current event we're iterating over. We now solely rely on marker events which when processed, add the insertion event as an extremity and the federating homeserver can ask about it when time calls.
- Related discussion, https://github.com/matrix-org/synapse/pull/11114#discussion_r741514793
Before | After
--- | ---
![](https://user-images.githubusercontent.com/558581/139218681-b465c862-5c49-4702-a59e-466733b0cf45.png) | ![](https://user-images.githubusercontent.com/558581/146453159-a1609e0a-8324-439d-ae44-e4bce43ac6d1.png)
#### Why aren't we sorting topologically when receiving backfill events?
> The main reason we're going to opt to not sort topologically when receiving backfill events is because it's probably best to do whatever is easiest to make it just work. People will probably have opinions once they look at [MSC2716](https://github.com/matrix-org/matrix-doc/pull/2716) which could change whatever implementation anyway.
>
> As mentioned, ideally we would do this but code necessary to make the fake edges but it gets confusing and gives an impression of “just whyyyy” (feels icky). This problem also dissolves with online topological ordering.
>
> -- https://github.com/matrix-org/synapse/pull/11114#discussion_r741517138
See https://github.com/matrix-org/synapse/pull/11114#discussion_r739610091 for the technical difficulties
2022-02-07 16:54:13 -05:00
|
|
|
allow_no_prev_events: bool = False,
|
2020-09-01 08:39:04 -04:00
|
|
|
prev_event_ids: Optional[List[str]] = None,
|
2020-10-13 18:14:35 -04:00
|
|
|
auth_event_ids: Optional[List[str]] = None,
|
2022-03-25 10:21:06 -04:00
|
|
|
state_event_ids: Optional[List[str]] = None,
|
2021-06-22 05:02:53 -04:00
|
|
|
depth: Optional[int] = None,
|
2022-09-28 05:39:03 -04:00
|
|
|
state_map: Optional[StateMap[str]] = None,
|
|
|
|
for_batch: bool = False,
|
|
|
|
current_state_group: Optional[int] = None,
|
2023-02-09 16:05:02 -05:00
|
|
|
) -> Tuple[EventBase, UnpersistedEventContextBase]:
|
2022-09-28 05:39:03 -04:00
|
|
|
"""Create a new event for a local client. If bool for_batch is true, will
|
|
|
|
create an event using the prev_event_ids, and will create an event context for
|
|
|
|
the event using the parameters state_map and current_state_group, thus these parameters
|
|
|
|
must be provided in this case if for_batch is True. The subsequently created event
|
|
|
|
and context are suitable for being batched up and bulk persisted to the database
|
2023-02-09 16:05:02 -05:00
|
|
|
with other similarly created events. Note that this returns an UnpersistedEventContext,
|
|
|
|
which must be converted to an EventContext before it can be sent to the DB.
|
2018-04-16 13:41:37 -04:00
|
|
|
|
|
|
|
Args:
|
2020-07-22 12:29:15 -04:00
|
|
|
builder:
|
|
|
|
requester:
|
Fix historical messages backfilling in random order on remote homeservers (MSC2716) (#11114)
Fix https://github.com/matrix-org/synapse/issues/11091
Fix https://github.com/matrix-org/synapse/issues/10764 (side-stepping the issue because we no longer have to deal with `fake_prev_event_id`)
1. Made the `/backfill` response return messages in `(depth, stream_ordering)` order (previously only sorted by `depth`)
- Technically, it shouldn't really matter how `/backfill` returns things but I'm just trying to make the `stream_ordering` a little more consistent from the origin to the remote homeservers in order to get the order of messages from `/messages` consistent ([sorted by `(topological_ordering, stream_ordering)`](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)).
- Even now that we return backfilled messages in order, it still doesn't guarantee the same `stream_ordering` (and more importantly the [`/messages` order](https://github.com/matrix-org/synapse/blob/develop/docs/development/room-dag-concepts.md#depth-and-stream-ordering)) on the other server. For example, if a room has a bunch of history imported and someone visits a permalink to a historical message back in time, their homeserver will skip over the historical messages in between and insert the permalink as the next message in the `stream_order` and totally throw off the sort.
- This will be even more the case when we add the [MSC3030 jump to date API endpoint](https://github.com/matrix-org/matrix-doc/pull/3030) so the static archives can navigate and jump to a certain date.
- We're solving this in the future by switching to [online topological ordering](https://github.com/matrix-org/gomatrixserverlib/issues/187) and [chunking](https://github.com/matrix-org/synapse/issues/3785) which by its nature will apply retroactively to fix any inconsistencies introduced by people permalinking
2. As we're navigating `prev_events` to return in `/backfill`, we order by `depth` first (newest -> oldest) and now also tie-break based on the `stream_ordering` (newest -> oldest). This is technically important because MSC2716 inserts a bunch of historical messages at the same `depth` so it's best to be prescriptive about which ones we should process first. In reality, I think the code already looped over the historical messages as expected because the database is already in order.
3. Making the historical state chain and historical event chain float on their own by having no `prev_events` instead of a fake `prev_event` which caused backfill to get clogged with an unresolvable event. Fixes https://github.com/matrix-org/synapse/issues/11091 and https://github.com/matrix-org/synapse/issues/10764
4. We no longer find connected insertion events by finding a potential `prev_event` connection to the current event we're iterating over. We now solely rely on marker events which when processed, add the insertion event as an extremity and the federating homeserver can ask about it when time calls.
- Related discussion, https://github.com/matrix-org/synapse/pull/11114#discussion_r741514793
Before | After
--- | ---
![](https://user-images.githubusercontent.com/558581/139218681-b465c862-5c49-4702-a59e-466733b0cf45.png) | ![](https://user-images.githubusercontent.com/558581/146453159-a1609e0a-8324-439d-ae44-e4bce43ac6d1.png)
#### Why aren't we sorting topologically when receiving backfill events?
> The main reason we're going to opt to not sort topologically when receiving backfill events is because it's probably best to do whatever is easiest to make it just work. People will probably have opinions once they look at [MSC2716](https://github.com/matrix-org/matrix-doc/pull/2716) which could change whatever implementation anyway.
>
> As mentioned, ideally we would do this but code necessary to make the fake edges but it gets confusing and gives an impression of “just whyyyy” (feels icky). This problem also dissolves with online topological ordering.
>
> -- https://github.com/matrix-org/synapse/pull/11114#discussion_r741517138
See https://github.com/matrix-org/synapse/pull/11114#discussion_r739610091 for the technical difficulties
2022-02-07 16:54:13 -05:00
|
|
|
allow_no_prev_events: Whether to allow this event to be created an empty
|
|
|
|
list of prev_events. Normally this is prohibited just because most
|
|
|
|
events should have a prev_event and we should only use this in special
|
2023-06-16 15:12:24 -04:00
|
|
|
cases (previously useful for MSC2716).
|
2020-01-03 11:16:09 -05:00
|
|
|
prev_event_ids:
|
2018-04-16 13:41:37 -04:00
|
|
|
the forward extremities to use as the prev_events for the
|
2020-01-03 11:16:09 -05:00
|
|
|
new event.
|
2018-04-16 13:41:37 -04:00
|
|
|
|
|
|
|
If None, they will be requested from the database.
|
|
|
|
|
2020-10-13 18:14:35 -04:00
|
|
|
auth_event_ids:
|
|
|
|
The event ids to use as the auth_events for the new event.
|
|
|
|
Should normally be left as None, which will cause them to be calculated
|
|
|
|
based on the room state at the prev_events.
|
|
|
|
|
2022-03-25 10:21:06 -04:00
|
|
|
state_event_ids:
|
2023-06-16 15:12:24 -04:00
|
|
|
The full state at a given event. This was previously used particularly
|
|
|
|
by the MSC2716 /batch_send endpoint. This should normally be left as
|
|
|
|
None, which will cause the auth_event_ids to be calculated based on the
|
|
|
|
room state at the prev_events.
|
2022-03-25 10:21:06 -04:00
|
|
|
|
2021-06-22 05:02:53 -04:00
|
|
|
depth: Override the depth used to order the event in the DAG.
|
|
|
|
Should normally be set to None, which will cause the depth to be calculated
|
|
|
|
based on the prev_events.
|
|
|
|
|
2022-09-28 05:39:03 -04:00
|
|
|
state_map: A state map of previously created events, used only when creating events
|
|
|
|
for batch persisting
|
|
|
|
|
|
|
|
for_batch: whether the event is being created for batch persisting to the db
|
|
|
|
|
|
|
|
current_state_group: the current state group, used only for creating events for
|
|
|
|
batch persisting
|
|
|
|
|
2018-04-16 13:41:37 -04:00
|
|
|
Returns:
|
2023-02-09 16:05:02 -05:00
|
|
|
Tuple of created event, UnpersistedEventContext
|
2018-04-16 13:41:37 -04:00
|
|
|
"""
|
2022-03-25 10:21:06 -04:00
|
|
|
# Strip down the state_event_ids to only what we need to auth the event.
|
2021-10-13 18:44:00 -04:00
|
|
|
# For example, we don't need extra m.room.member that don't match event.sender
|
2022-03-25 10:21:06 -04:00
|
|
|
if state_event_ids is not None:
|
|
|
|
# Do a quick check to make sure that prev_event_ids is present to
|
|
|
|
# make the type-checking around `builder.build` happy.
|
2021-12-11 00:08:51 -05:00
|
|
|
# prev_event_ids could be an empty array though.
|
2021-10-13 18:44:00 -04:00
|
|
|
assert prev_event_ids is not None
|
|
|
|
|
|
|
|
temp_event = await builder.build(
|
|
|
|
prev_event_ids=prev_event_ids,
|
2022-03-25 10:21:06 -04:00
|
|
|
auth_event_ids=state_event_ids,
|
2021-10-13 18:44:00 -04:00
|
|
|
depth=depth,
|
|
|
|
)
|
2022-03-25 10:21:06 -04:00
|
|
|
state_events = await self.store.get_events_as_list(state_event_ids)
|
2021-10-13 18:44:00 -04:00
|
|
|
# Create a StateMap[str]
|
2022-11-28 22:18:12 -05:00
|
|
|
current_state_ids = {
|
|
|
|
(e.type, e.state_key): e.event_id for e in state_events
|
|
|
|
}
|
2022-03-25 10:21:06 -04:00
|
|
|
# Actually strip down and only use the necessary auth events
|
2021-10-13 18:44:00 -04:00
|
|
|
auth_event_ids = self._event_auth_handler.compute_auth_events(
|
|
|
|
event=temp_event,
|
2022-11-28 22:18:12 -05:00
|
|
|
current_state_ids=current_state_ids,
|
2021-10-13 18:44:00 -04:00
|
|
|
for_verification=False,
|
|
|
|
)
|
|
|
|
|
2020-01-03 11:16:09 -05:00
|
|
|
if prev_event_ids is not None:
|
2024-09-02 07:39:04 -04:00
|
|
|
assert len(prev_event_ids) <= 10, (
|
|
|
|
"Attempting to create an event with %i prev_events"
|
|
|
|
% (len(prev_event_ids),)
|
2016-05-11 04:09:20 -04:00
|
|
|
)
|
2018-04-16 13:41:37 -04:00
|
|
|
else:
|
2020-07-22 12:29:15 -04:00
|
|
|
prev_event_ids = await self.store.get_prev_events_for_room(builder.room_id)
|
2016-05-11 04:09:20 -04:00
|
|
|
|
2021-12-11 00:08:51 -05:00
|
|
|
# Do a quick sanity check here, rather than waiting until we've created the
|
2020-08-10 07:29:47 -04:00
|
|
|
# event and then try to auth it (which fails with a somewhat confusing "No
|
|
|
|
# create event in auth events")
|
2021-12-11 00:08:51 -05:00
|
|
|
if allow_no_prev_events:
|
|
|
|
# We allow events with no `prev_events` but it better have some `auth_events`
|
|
|
|
assert (
|
|
|
|
builder.type == EventTypes.Create
|
|
|
|
# Allow an event to have empty list of prev_event_ids
|
|
|
|
# only if it has auth_event_ids.
|
|
|
|
or auth_event_ids
|
|
|
|
), "Attempting to create a non-m.room.create event with no prev_events or auth_event_ids"
|
|
|
|
else:
|
|
|
|
# we now ought to have some prev_events (unless it's a create event).
|
|
|
|
assert (
|
|
|
|
builder.type == EventTypes.Create or prev_event_ids
|
|
|
|
), "Attempting to create a non-m.room.create event with no prev_events"
|
2020-08-10 07:29:47 -04:00
|
|
|
|
2022-09-28 05:39:03 -04:00
|
|
|
if for_batch:
|
|
|
|
assert prev_event_ids is not None
|
|
|
|
assert state_map is not None
|
|
|
|
auth_ids = self._event_auth_handler.compute_auth_events(builder, state_map)
|
|
|
|
event = await builder.build(
|
|
|
|
prev_event_ids=prev_event_ids, auth_event_ids=auth_ids, depth=depth
|
|
|
|
)
|
2023-02-09 16:05:02 -05:00
|
|
|
|
|
|
|
context: UnpersistedEventContextBase = (
|
|
|
|
await self.state.calculate_context_info(
|
|
|
|
event,
|
|
|
|
state_ids_before_event=state_map,
|
|
|
|
partial_state=False,
|
|
|
|
state_group_before_event=current_state_group,
|
|
|
|
)
|
2022-09-28 05:39:03 -04:00
|
|
|
)
|
2023-02-09 16:05:02 -05:00
|
|
|
|
2022-09-28 05:39:03 -04:00
|
|
|
else:
|
|
|
|
event = await builder.build(
|
|
|
|
prev_event_ids=prev_event_ids,
|
|
|
|
auth_event_ids=auth_event_ids,
|
|
|
|
depth=depth,
|
|
|
|
)
|
2021-06-22 05:02:53 -04:00
|
|
|
|
2022-09-28 05:39:03 -04:00
|
|
|
# Pass on the outlier property from the builder to the event
|
|
|
|
# after it is created
|
|
|
|
if builder.internal_metadata.outlier:
|
|
|
|
event.internal_metadata.outlier = True
|
|
|
|
context = EventContext.for_outlier(self._storage_controllers)
|
|
|
|
else:
|
2023-02-09 16:05:02 -05:00
|
|
|
context = await self.state.calculate_context_info(event)
|
2021-06-22 05:02:53 -04:00
|
|
|
|
2017-05-02 06:36:11 -04:00
|
|
|
if requester:
|
|
|
|
context.app_service = requester.app_service
|
2016-05-11 04:09:20 -04:00
|
|
|
|
2023-05-04 10:18:22 -04:00
|
|
|
res, new_content = await self._third_party_event_rules.check_event_allowed(
|
2020-10-13 10:44:54 -04:00
|
|
|
event, context
|
|
|
|
)
|
2021-07-20 06:39:46 -04:00
|
|
|
if res is False:
|
2020-10-13 10:44:54 -04:00
|
|
|
logger.info(
|
2021-02-16 17:32:34 -05:00
|
|
|
"Event %s forbidden by third-party rules",
|
|
|
|
event,
|
2020-10-13 10:44:54 -04:00
|
|
|
)
|
|
|
|
raise SynapseError(
|
|
|
|
403, "This event is not allowed in this context", Codes.FORBIDDEN
|
|
|
|
)
|
2021-07-20 06:39:46 -04:00
|
|
|
elif new_content is not None:
|
2020-10-13 13:53:56 -04:00
|
|
|
# the third-party rules want to replace the event. We'll need to build a new
|
|
|
|
# event.
|
|
|
|
event, context = await self._rebuild_event_after_third_party_rules(
|
2021-07-20 06:39:46 -04:00
|
|
|
new_content, event
|
2020-10-13 13:53:56 -04:00
|
|
|
)
|
2020-10-13 10:44:54 -04:00
|
|
|
|
2019-11-04 12:09:22 -05:00
|
|
|
self.validator.validate_new(event, self.config)
|
2021-11-18 08:43:09 -05:00
|
|
|
await self._validate_event_relation(event)
|
2024-03-19 13:52:53 -04:00
|
|
|
|
|
|
|
if event.type == EventTypes.CallInvite:
|
|
|
|
room_id = event.room_id
|
|
|
|
room_info = await self.store.get_room_with_stats(room_id)
|
|
|
|
assert room_info is not None
|
|
|
|
|
|
|
|
if room_info.join_rules == JoinRules.PUBLIC:
|
|
|
|
raise SynapseError(
|
|
|
|
403,
|
|
|
|
"Call invites are not allowed in public rooms.",
|
|
|
|
Codes.FORBIDDEN,
|
|
|
|
)
|
2021-11-18 08:43:09 -05:00
|
|
|
logger.debug("Created event %s", event.event_id)
|
|
|
|
|
|
|
|
return event, context
|
|
|
|
|
|
|
|
async def _validate_event_relation(self, event: EventBase) -> None:
|
|
|
|
"""
|
|
|
|
Ensure the relation data on a new event is not bogus.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
event: The event being created.
|
|
|
|
|
|
|
|
Raises:
|
|
|
|
SynapseError if the event is invalid.
|
|
|
|
"""
|
|
|
|
|
2022-05-16 08:42:45 -04:00
|
|
|
relation = relation_from_event(event)
|
2021-11-18 08:43:09 -05:00
|
|
|
if not relation:
|
|
|
|
return
|
|
|
|
|
2022-05-16 08:42:45 -04:00
|
|
|
parent_event = await self.store.get_event(relation.parent_id, allow_none=True)
|
2021-11-18 08:43:09 -05:00
|
|
|
if parent_event:
|
|
|
|
# And in the same room.
|
|
|
|
if parent_event.room_id != event.room_id:
|
|
|
|
raise SynapseError(400, "Relations must be in the same room")
|
|
|
|
|
|
|
|
else:
|
|
|
|
# There must be some reason that the client knows the event exists,
|
|
|
|
# see if there are existing relations. If so, assume everything is fine.
|
2022-05-16 08:42:45 -04:00
|
|
|
if not await self.store.event_is_target_of_relation(relation.parent_id):
|
2021-11-18 08:43:09 -05:00
|
|
|
# Otherwise, the client can't know about the parent event!
|
|
|
|
raise SynapseError(400, "Can't send relation to unknown event")
|
2016-05-11 04:09:20 -04:00
|
|
|
|
2019-05-21 11:51:45 -04:00
|
|
|
# If this event is an annotation then we check that that the sender
|
|
|
|
# can't annotate the same way twice (e.g. stops users from liking an
|
|
|
|
# event multiple times).
|
2022-05-16 08:42:45 -04:00
|
|
|
if relation.rel_type == RelationTypes.ANNOTATION:
|
|
|
|
aggregation_key = relation.aggregation_key
|
|
|
|
|
|
|
|
if aggregation_key is None:
|
|
|
|
raise SynapseError(400, "Missing aggregation key")
|
2019-05-20 12:39:05 -04:00
|
|
|
|
2022-03-03 05:52:35 -05:00
|
|
|
if len(aggregation_key) > 500:
|
|
|
|
raise SynapseError(400, "Aggregation key is too long")
|
|
|
|
|
2020-07-22 12:29:15 -04:00
|
|
|
already_exists = await self.store.has_user_annotated_event(
|
2022-05-16 08:42:45 -04:00
|
|
|
relation.parent_id, event.type, aggregation_key, event.sender
|
2019-05-20 12:39:05 -04:00
|
|
|
)
|
|
|
|
if already_exists:
|
2023-02-15 06:47:57 -05:00
|
|
|
raise SynapseError(
|
|
|
|
400,
|
|
|
|
"Can't send same reaction twice",
|
|
|
|
errcode=Codes.DUPLICATE_ANNOTATION,
|
|
|
|
)
|
2019-05-20 12:39:05 -04:00
|
|
|
|
2021-11-18 08:43:09 -05:00
|
|
|
# Don't attempt to start a thread if the parent event is a relation.
|
2022-05-16 08:42:45 -04:00
|
|
|
elif relation.rel_type == RelationTypes.THREAD:
|
|
|
|
if await self.store.event_includes_relation(relation.parent_id):
|
2021-11-18 08:43:09 -05:00
|
|
|
raise SynapseError(
|
|
|
|
400, "Cannot start threads from an event with a relation"
|
|
|
|
)
|
2016-05-11 04:09:20 -04:00
|
|
|
|
2016-08-23 10:23:39 -04:00
|
|
|
@measure_func("handle_new_client_event")
|
2020-05-01 10:15:36 -04:00
|
|
|
async def handle_new_client_event(
|
2020-07-22 12:29:15 -04:00
|
|
|
self,
|
|
|
|
requester: Requester,
|
2022-09-28 06:11:48 -04:00
|
|
|
events_and_context: List[Tuple[EventBase, EventContext]],
|
2020-07-22 12:29:15 -04:00
|
|
|
ratelimit: bool = True,
|
2021-04-08 17:38:54 -04:00
|
|
|
extra_users: Optional[List[UserID]] = None,
|
2020-10-02 13:03:21 -04:00
|
|
|
ignore_shadow_ban: bool = False,
|
2020-10-02 11:45:41 -04:00
|
|
|
) -> EventBase:
|
2022-09-28 06:11:48 -04:00
|
|
|
"""Processes new events. Please note that if batch persisting events, an error in
|
|
|
|
handling any one of these events will result in all of the events being dropped.
|
2020-10-02 11:45:41 -04:00
|
|
|
|
|
|
|
This includes deduplicating, checking auth, persisting,
|
2018-02-15 11:30:10 -05:00
|
|
|
notifying users, sending to remote servers, etc.
|
|
|
|
|
|
|
|
If called from a worker will hit out to the master process for final
|
|
|
|
processing.
|
|
|
|
|
|
|
|
Args:
|
2020-07-22 12:29:15 -04:00
|
|
|
requester
|
2022-09-28 06:11:48 -04:00
|
|
|
events_and_context: A list of one or more tuples of event, context to be persisted
|
2020-07-22 12:29:15 -04:00
|
|
|
ratelimit
|
|
|
|
extra_users: Any extra users to notify about event
|
2020-05-22 09:21:54 -04:00
|
|
|
|
2020-10-02 13:03:21 -04:00
|
|
|
ignore_shadow_ban: True if shadow-banned users should be allowed to
|
|
|
|
send this event.
|
|
|
|
|
2020-05-22 09:21:54 -04:00
|
|
|
Return:
|
2020-10-02 11:45:41 -04:00
|
|
|
If the event was deduplicated, the previous, duplicate, event. Otherwise,
|
|
|
|
`event`.
|
2020-10-02 13:03:21 -04:00
|
|
|
|
|
|
|
Raises:
|
|
|
|
ShadowBanError if the requester has been shadow-banned.
|
2022-12-15 11:04:23 -05:00
|
|
|
PartialStateConflictError if attempting to persist a partial state event in
|
Handle race between persisting an event and un-partial stating a room (#13100)
Whenever we want to persist an event, we first compute an event context,
which includes the state at the event and a flag indicating whether the
state is partial. After a lot of processing, we finally try to store the
event in the database, which can fail for partial state events when the
containing room has been un-partial stated in the meantime.
We detect the race as a foreign key constraint failure in the data store
layer and turn it into a special `PartialStateConflictError` exception,
which makes its way up to the method in which we computed the event
context.
To make things difficult, the exception needs to cross a replication
request: `/fed_send_events` for events coming over federation and
`/send_event` for events from clients. We transport the
`PartialStateConflictError` as a `409 Conflict` over replication and
turn `409`s back into `PartialStateConflictError`s on the worker making
the request.
All client events go through
`EventCreationHandler.handle_new_client_event`, which is called in
*a lot* of places. Instead of trying to update all the code which
creates client events, we turn the `PartialStateConflictError` into a
`429 Too Many Requests` in
`EventCreationHandler.handle_new_client_event` and hope that clients
take it as a hint to retry their request.
On the federation event side, there are 7 places which compute event
contexts. 4 of them use outlier event contexts:
`FederationEventHandler._auth_and_persist_outliers_inner`,
`FederationHandler.do_knock`, `FederationHandler.on_invite_request` and
`FederationHandler.do_remotely_reject_invite`. These events won't have
the partial state flag, so we do not need to do anything for then.
The remaining 3 paths which create events are
`FederationEventHandler.process_remote_join`,
`FederationEventHandler.on_send_membership_event` and
`FederationEventHandler._process_received_pdu`.
We can't experience the race in `process_remote_join`, unless we're
handling an additional join into a partial state room, which currently
blocks, so we make no attempt to handle it correctly.
`on_send_membership_event` is only called by
`FederationServer._on_send_membership_event`, so we catch the
`PartialStateConflictError` there and retry just once.
`_process_received_pdu` is called by `on_receive_pdu` for incoming
events and `_process_pulled_event` for backfill. The latter should never
try to persist partial state events, so we ignore it. We catch the
`PartialStateConflictError` in `on_receive_pdu` and retry just once.
Refering to the graph of code paths in
https://github.com/matrix-org/synapse/issues/12988#issuecomment-1156857648
may make the above make more sense.
Signed-off-by: Sean Quah <seanq@matrix.org>
2022-07-05 11:12:52 -04:00
|
|
|
a room that has been un-partial stated.
|
2018-02-15 11:30:10 -05:00
|
|
|
"""
|
2021-04-08 17:38:54 -04:00
|
|
|
extra_users = extra_users or []
|
2018-02-15 11:30:10 -05:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
for event, context in events_and_context:
|
|
|
|
# we don't apply shadow-banning to membership events here. Invites are blocked
|
|
|
|
# higher up the stack, and we allow shadow-banned users to send join and leave
|
|
|
|
# events as normal.
|
|
|
|
if (
|
|
|
|
event.type != EventTypes.Member
|
|
|
|
and not ignore_shadow_ban
|
|
|
|
and requester.shadow_banned
|
|
|
|
):
|
|
|
|
# We randomly sleep a bit just to annoy the requester.
|
|
|
|
await self.clock.sleep(random.randint(1, 10))
|
|
|
|
raise ShadowBanError()
|
2020-10-02 13:03:21 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
if event.is_state():
|
|
|
|
prev_event = await self.deduplicate_state_event(event, context)
|
|
|
|
if prev_event is not None:
|
|
|
|
logger.info(
|
|
|
|
"Not bothering to persist state event %s duplicated by %s",
|
|
|
|
event.event_id,
|
|
|
|
prev_event.event_id,
|
|
|
|
)
|
|
|
|
return prev_event
|
2020-10-02 11:45:41 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
if event.internal_metadata.is_out_of_band_membership():
|
|
|
|
# the only sort of out-of-band-membership events we expect to see here are
|
|
|
|
# invite rejections and rescinded knocks that we have generated ourselves.
|
|
|
|
assert event.type == EventTypes.Member
|
|
|
|
assert event.content["membership"] == Membership.LEAVE
|
|
|
|
else:
|
|
|
|
try:
|
|
|
|
validate_event_for_room_version(event)
|
2022-10-18 12:58:47 -04:00
|
|
|
# If we are persisting a batch of events the event(s) needed to auth the
|
|
|
|
# current event may be part of the batch and will not be in the DB yet
|
|
|
|
event_id_to_event = {e.event_id: e for e, _ in events_and_context}
|
|
|
|
batched_auth_events = {}
|
|
|
|
for event_id in event.auth_event_ids():
|
|
|
|
auth_event = event_id_to_event.get(event_id)
|
|
|
|
if auth_event:
|
|
|
|
batched_auth_events[event_id] = auth_event
|
2022-09-28 06:11:48 -04:00
|
|
|
await self._event_auth_handler.check_auth_rules_from_context(
|
2022-10-18 12:58:47 -04:00
|
|
|
event, batched_auth_events
|
2022-09-28 06:11:48 -04:00
|
|
|
)
|
|
|
|
except AuthError as err:
|
|
|
|
logger.warning("Denying new event %r because %s", event, err)
|
|
|
|
raise err
|
2018-03-01 05:18:33 -05:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
# Ensure that we can round trip before trying to persist in db
|
|
|
|
try:
|
|
|
|
dump = json_encoder.encode(event.content)
|
|
|
|
json_decoder.decode(dump)
|
|
|
|
except Exception:
|
|
|
|
logger.exception("Failed to encode content: %r", event.content)
|
|
|
|
raise
|
2018-03-01 05:18:33 -05:00
|
|
|
|
2021-05-12 08:17:11 -04:00
|
|
|
# We now persist the event (and update the cache in parallel, since we
|
|
|
|
# don't want to block on it).
|
2023-08-24 10:53:07 -04:00
|
|
|
#
|
|
|
|
# Note: mypy gets confused if we inline dl and check with twisted#11770.
|
|
|
|
# Some kind of bug in mypy's deduction?
|
|
|
|
deferreds = (
|
|
|
|
run_in_background(
|
|
|
|
self._persist_events,
|
|
|
|
requester=requester,
|
|
|
|
events_and_context=events_and_context,
|
|
|
|
ratelimit=ratelimit,
|
|
|
|
extra_users=extra_users,
|
|
|
|
),
|
|
|
|
run_in_background(
|
|
|
|
self.cache_joined_hosts_for_events, events_and_context
|
|
|
|
).addErrback(log_failure, "cache_joined_hosts_for_event failed"),
|
|
|
|
)
|
2022-12-15 11:04:23 -05:00
|
|
|
result, _ = await make_deferred_yieldable(
|
2023-08-24 10:53:07 -04:00
|
|
|
gather_results(deferreds, consumeErrors=True)
|
2022-12-15 11:04:23 -05:00
|
|
|
).addErrback(unwrapFirstError)
|
2021-05-12 08:17:11 -04:00
|
|
|
|
2021-12-14 12:35:28 -05:00
|
|
|
return result
|
2021-05-12 08:17:11 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
async def _persist_events(
|
2021-05-12 08:17:11 -04:00
|
|
|
self,
|
|
|
|
requester: Requester,
|
2022-09-28 06:11:48 -04:00
|
|
|
events_and_context: List[Tuple[EventBase, EventContext]],
|
2021-05-12 08:17:11 -04:00
|
|
|
ratelimit: bool = True,
|
|
|
|
extra_users: Optional[List[UserID]] = None,
|
|
|
|
) -> EventBase:
|
2022-09-28 06:11:48 -04:00
|
|
|
"""Actually persists new events. Should only be called by
|
2021-05-12 08:17:11 -04:00
|
|
|
`handle_new_client_event`, and see its docstring for documentation of
|
2022-09-28 06:11:48 -04:00
|
|
|
the arguments. Please note that if batch persisting events, an error in
|
|
|
|
handling any one of these events will result in all of the events being dropped.
|
Handle race between persisting an event and un-partial stating a room (#13100)
Whenever we want to persist an event, we first compute an event context,
which includes the state at the event and a flag indicating whether the
state is partial. After a lot of processing, we finally try to store the
event in the database, which can fail for partial state events when the
containing room has been un-partial stated in the meantime.
We detect the race as a foreign key constraint failure in the data store
layer and turn it into a special `PartialStateConflictError` exception,
which makes its way up to the method in which we computed the event
context.
To make things difficult, the exception needs to cross a replication
request: `/fed_send_events` for events coming over federation and
`/send_event` for events from clients. We transport the
`PartialStateConflictError` as a `409 Conflict` over replication and
turn `409`s back into `PartialStateConflictError`s on the worker making
the request.
All client events go through
`EventCreationHandler.handle_new_client_event`, which is called in
*a lot* of places. Instead of trying to update all the code which
creates client events, we turn the `PartialStateConflictError` into a
`429 Too Many Requests` in
`EventCreationHandler.handle_new_client_event` and hope that clients
take it as a hint to retry their request.
On the federation event side, there are 7 places which compute event
contexts. 4 of them use outlier event contexts:
`FederationEventHandler._auth_and_persist_outliers_inner`,
`FederationHandler.do_knock`, `FederationHandler.on_invite_request` and
`FederationHandler.do_remotely_reject_invite`. These events won't have
the partial state flag, so we do not need to do anything for then.
The remaining 3 paths which create events are
`FederationEventHandler.process_remote_join`,
`FederationEventHandler.on_send_membership_event` and
`FederationEventHandler._process_received_pdu`.
We can't experience the race in `process_remote_join`, unless we're
handling an additional join into a partial state room, which currently
blocks, so we make no attempt to handle it correctly.
`on_send_membership_event` is only called by
`FederationServer._on_send_membership_event`, so we catch the
`PartialStateConflictError` there and retry just once.
`_process_received_pdu` is called by `on_receive_pdu` for incoming
events and `_process_pulled_event` for backfill. The latter should never
try to persist partial state events, so we ignore it. We catch the
`PartialStateConflictError` in `on_receive_pdu` and retry just once.
Refering to the graph of code paths in
https://github.com/matrix-org/synapse/issues/12988#issuecomment-1156857648
may make the above make more sense.
Signed-off-by: Sean Quah <seanq@matrix.org>
2022-07-05 11:12:52 -04:00
|
|
|
|
|
|
|
PartialStateConflictError: if attempting to persist a partial state event in
|
|
|
|
a room that has been un-partial stated.
|
2021-05-12 08:17:11 -04:00
|
|
|
"""
|
2018-02-15 11:30:10 -05:00
|
|
|
|
2022-10-21 13:46:22 -04:00
|
|
|
await self._bulk_push_rule_evaluator.action_for_events_by_user(
|
|
|
|
events_and_context
|
|
|
|
)
|
2021-01-26 08:57:31 -05:00
|
|
|
|
2018-02-15 11:30:10 -05:00
|
|
|
try:
|
|
|
|
# If we're a worker we need to hit out to the master.
|
2022-09-28 06:11:48 -04:00
|
|
|
first_event, _ = events_and_context[0]
|
|
|
|
writer_instance = self._events_shard_config.get_instance(
|
|
|
|
first_event.room_id
|
|
|
|
)
|
2020-09-14 05:16:41 -04:00
|
|
|
if writer_instance != self._instance_name:
|
2023-10-27 07:52:40 -04:00
|
|
|
# Ratelimit before sending to the other event persister, to
|
|
|
|
# ensure that we correctly have ratelimits on both the event
|
|
|
|
# creators and event persisters.
|
|
|
|
if ratelimit:
|
|
|
|
for event, _ in events_and_context:
|
|
|
|
is_admin_redaction = await self.is_admin_redaction(
|
|
|
|
event.type, event.sender, event.redacts
|
|
|
|
)
|
|
|
|
await self.request_ratelimiter.ratelimit(
|
|
|
|
requester, is_admin_redaction=is_admin_redaction
|
|
|
|
)
|
|
|
|
|
Handle race between persisting an event and un-partial stating a room (#13100)
Whenever we want to persist an event, we first compute an event context,
which includes the state at the event and a flag indicating whether the
state is partial. After a lot of processing, we finally try to store the
event in the database, which can fail for partial state events when the
containing room has been un-partial stated in the meantime.
We detect the race as a foreign key constraint failure in the data store
layer and turn it into a special `PartialStateConflictError` exception,
which makes its way up to the method in which we computed the event
context.
To make things difficult, the exception needs to cross a replication
request: `/fed_send_events` for events coming over federation and
`/send_event` for events from clients. We transport the
`PartialStateConflictError` as a `409 Conflict` over replication and
turn `409`s back into `PartialStateConflictError`s on the worker making
the request.
All client events go through
`EventCreationHandler.handle_new_client_event`, which is called in
*a lot* of places. Instead of trying to update all the code which
creates client events, we turn the `PartialStateConflictError` into a
`429 Too Many Requests` in
`EventCreationHandler.handle_new_client_event` and hope that clients
take it as a hint to retry their request.
On the federation event side, there are 7 places which compute event
contexts. 4 of them use outlier event contexts:
`FederationEventHandler._auth_and_persist_outliers_inner`,
`FederationHandler.do_knock`, `FederationHandler.on_invite_request` and
`FederationHandler.do_remotely_reject_invite`. These events won't have
the partial state flag, so we do not need to do anything for then.
The remaining 3 paths which create events are
`FederationEventHandler.process_remote_join`,
`FederationEventHandler.on_send_membership_event` and
`FederationEventHandler._process_received_pdu`.
We can't experience the race in `process_remote_join`, unless we're
handling an additional join into a partial state room, which currently
blocks, so we make no attempt to handle it correctly.
`on_send_membership_event` is only called by
`FederationServer._on_send_membership_event`, so we catch the
`PartialStateConflictError` there and retry just once.
`_process_received_pdu` is called by `on_receive_pdu` for incoming
events and `_process_pulled_event` for backfill. The latter should never
try to persist partial state events, so we ignore it. We catch the
`PartialStateConflictError` in `on_receive_pdu` and retry just once.
Refering to the graph of code paths in
https://github.com/matrix-org/synapse/issues/12988#issuecomment-1156857648
may make the above make more sense.
Signed-off-by: Sean Quah <seanq@matrix.org>
2022-07-05 11:12:52 -04:00
|
|
|
try:
|
2022-09-28 06:11:48 -04:00
|
|
|
result = await self.send_events(
|
Handle race between persisting an event and un-partial stating a room (#13100)
Whenever we want to persist an event, we first compute an event context,
which includes the state at the event and a flag indicating whether the
state is partial. After a lot of processing, we finally try to store the
event in the database, which can fail for partial state events when the
containing room has been un-partial stated in the meantime.
We detect the race as a foreign key constraint failure in the data store
layer and turn it into a special `PartialStateConflictError` exception,
which makes its way up to the method in which we computed the event
context.
To make things difficult, the exception needs to cross a replication
request: `/fed_send_events` for events coming over federation and
`/send_event` for events from clients. We transport the
`PartialStateConflictError` as a `409 Conflict` over replication and
turn `409`s back into `PartialStateConflictError`s on the worker making
the request.
All client events go through
`EventCreationHandler.handle_new_client_event`, which is called in
*a lot* of places. Instead of trying to update all the code which
creates client events, we turn the `PartialStateConflictError` into a
`429 Too Many Requests` in
`EventCreationHandler.handle_new_client_event` and hope that clients
take it as a hint to retry their request.
On the federation event side, there are 7 places which compute event
contexts. 4 of them use outlier event contexts:
`FederationEventHandler._auth_and_persist_outliers_inner`,
`FederationHandler.do_knock`, `FederationHandler.on_invite_request` and
`FederationHandler.do_remotely_reject_invite`. These events won't have
the partial state flag, so we do not need to do anything for then.
The remaining 3 paths which create events are
`FederationEventHandler.process_remote_join`,
`FederationEventHandler.on_send_membership_event` and
`FederationEventHandler._process_received_pdu`.
We can't experience the race in `process_remote_join`, unless we're
handling an additional join into a partial state room, which currently
blocks, so we make no attempt to handle it correctly.
`on_send_membership_event` is only called by
`FederationServer._on_send_membership_event`, so we catch the
`PartialStateConflictError` there and retry just once.
`_process_received_pdu` is called by `on_receive_pdu` for incoming
events and `_process_pulled_event` for backfill. The latter should never
try to persist partial state events, so we ignore it. We catch the
`PartialStateConflictError` in `on_receive_pdu` and retry just once.
Refering to the graph of code paths in
https://github.com/matrix-org/synapse/issues/12988#issuecomment-1156857648
may make the above make more sense.
Signed-off-by: Sean Quah <seanq@matrix.org>
2022-07-05 11:12:52 -04:00
|
|
|
instance_name=writer_instance,
|
2022-09-28 06:11:48 -04:00
|
|
|
events_and_context=events_and_context,
|
Handle race between persisting an event and un-partial stating a room (#13100)
Whenever we want to persist an event, we first compute an event context,
which includes the state at the event and a flag indicating whether the
state is partial. After a lot of processing, we finally try to store the
event in the database, which can fail for partial state events when the
containing room has been un-partial stated in the meantime.
We detect the race as a foreign key constraint failure in the data store
layer and turn it into a special `PartialStateConflictError` exception,
which makes its way up to the method in which we computed the event
context.
To make things difficult, the exception needs to cross a replication
request: `/fed_send_events` for events coming over federation and
`/send_event` for events from clients. We transport the
`PartialStateConflictError` as a `409 Conflict` over replication and
turn `409`s back into `PartialStateConflictError`s on the worker making
the request.
All client events go through
`EventCreationHandler.handle_new_client_event`, which is called in
*a lot* of places. Instead of trying to update all the code which
creates client events, we turn the `PartialStateConflictError` into a
`429 Too Many Requests` in
`EventCreationHandler.handle_new_client_event` and hope that clients
take it as a hint to retry their request.
On the federation event side, there are 7 places which compute event
contexts. 4 of them use outlier event contexts:
`FederationEventHandler._auth_and_persist_outliers_inner`,
`FederationHandler.do_knock`, `FederationHandler.on_invite_request` and
`FederationHandler.do_remotely_reject_invite`. These events won't have
the partial state flag, so we do not need to do anything for then.
The remaining 3 paths which create events are
`FederationEventHandler.process_remote_join`,
`FederationEventHandler.on_send_membership_event` and
`FederationEventHandler._process_received_pdu`.
We can't experience the race in `process_remote_join`, unless we're
handling an additional join into a partial state room, which currently
blocks, so we make no attempt to handle it correctly.
`on_send_membership_event` is only called by
`FederationServer._on_send_membership_event`, so we catch the
`PartialStateConflictError` there and retry just once.
`_process_received_pdu` is called by `on_receive_pdu` for incoming
events and `_process_pulled_event` for backfill. The latter should never
try to persist partial state events, so we ignore it. We catch the
`PartialStateConflictError` in `on_receive_pdu` and retry just once.
Refering to the graph of code paths in
https://github.com/matrix-org/synapse/issues/12988#issuecomment-1156857648
may make the above make more sense.
Signed-off-by: Sean Quah <seanq@matrix.org>
2022-07-05 11:12:52 -04:00
|
|
|
store=self.store,
|
|
|
|
requester=requester,
|
|
|
|
ratelimit=ratelimit,
|
|
|
|
extra_users=extra_users,
|
|
|
|
)
|
|
|
|
except SynapseError as e:
|
|
|
|
if e.code == HTTPStatus.CONFLICT:
|
|
|
|
raise PartialStateConflictError()
|
|
|
|
raise
|
2020-05-22 09:21:54 -04:00
|
|
|
stream_id = result["stream_id"]
|
2020-10-13 07:07:56 -04:00
|
|
|
event_id = result["event_id"]
|
2022-09-28 06:11:48 -04:00
|
|
|
|
|
|
|
# If we batch persisted events we return the last persisted event, otherwise
|
|
|
|
# we return the one event that was persisted
|
|
|
|
event, _ = events_and_context[-1]
|
|
|
|
|
2020-10-13 07:07:56 -04:00
|
|
|
if event_id != event.event_id:
|
|
|
|
# If we get a different event back then it means that its
|
|
|
|
# been de-duplicated, so we replace the given event with the
|
|
|
|
# one already persisted.
|
|
|
|
event = await self.store.get_event(event_id)
|
|
|
|
else:
|
|
|
|
# If we newly persisted the event then we need to update its
|
|
|
|
# stream_ordering entry manually (as it was persisted on
|
|
|
|
# another worker).
|
|
|
|
event.internal_metadata.stream_ordering = stream_id
|
2024-06-13 12:32:50 -04:00
|
|
|
event.internal_metadata.instance_name = writer_instance
|
2023-10-27 07:52:40 -04:00
|
|
|
|
2020-10-02 11:45:41 -04:00
|
|
|
return event
|
2018-02-15 11:30:10 -05:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
event = await self.persist_and_notify_client_events(
|
|
|
|
requester,
|
|
|
|
events_and_context,
|
|
|
|
ratelimit=ratelimit,
|
|
|
|
extra_users=extra_users,
|
2018-02-05 12:22:16 -05:00
|
|
|
)
|
2018-04-27 06:40:06 -04:00
|
|
|
|
2020-10-02 11:45:41 -04:00
|
|
|
return event
|
2020-07-27 08:35:56 -04:00
|
|
|
except Exception:
|
2022-09-28 06:11:48 -04:00
|
|
|
for event, _ in events_and_context:
|
|
|
|
# Ensure that we actually remove the entries in the push actions
|
|
|
|
# staging area, if we calculated them.
|
|
|
|
await self.store.remove_push_actions_from_staging(event.event_id)
|
2020-07-27 08:35:56 -04:00
|
|
|
raise
|
2018-02-15 11:30:10 -05:00
|
|
|
|
2022-10-12 14:01:00 -04:00
|
|
|
async def cache_joined_hosts_for_events(
|
|
|
|
self, events_and_context: List[Tuple[EventBase, EventContext]]
|
2021-05-05 11:53:22 -04:00
|
|
|
) -> None:
|
2022-10-12 14:01:00 -04:00
|
|
|
"""Precalculate the joined hosts at each of the given events, when using Redis, so that
|
2021-01-26 08:57:31 -05:00
|
|
|
external federation senders don't have to recalculate it themselves.
|
|
|
|
"""
|
|
|
|
|
2023-01-11 08:21:53 -05:00
|
|
|
if not self._external_cache.is_enabled():
|
|
|
|
return
|
2021-05-05 11:53:22 -04:00
|
|
|
|
2023-01-11 08:21:53 -05:00
|
|
|
# If external cache is enabled we should always have this.
|
|
|
|
assert self._external_cache_joined_hosts_updates is not None
|
|
|
|
|
|
|
|
for event, event_context in events_and_context:
|
|
|
|
if event_context.partial_state:
|
|
|
|
# To populate the cache for a partial-state event, we either have to
|
|
|
|
# block until full state, which the code below does, or change the
|
|
|
|
# meaning of cache values to be the list of hosts to which we plan to
|
|
|
|
# send events and calculate that instead.
|
|
|
|
#
|
|
|
|
# The federation senders don't use the external cache when sending
|
|
|
|
# events in partial-state rooms anyway, so let's not bother populating
|
|
|
|
# the cache.
|
|
|
|
continue
|
2021-01-26 08:57:31 -05:00
|
|
|
|
2022-10-12 14:01:00 -04:00
|
|
|
# We actually store two mappings, event ID -> prev state group,
|
|
|
|
# state group -> joined hosts, which is much more space efficient
|
|
|
|
# than event ID -> joined hosts.
|
|
|
|
#
|
|
|
|
# Note: We have to cache event ID -> prev state group, as we don't
|
|
|
|
# store that in the DB.
|
|
|
|
#
|
|
|
|
# Note: We set the state group -> joined hosts cache if it hasn't been
|
|
|
|
# set for a while, so that the expiry time is reset.
|
2021-01-26 08:57:31 -05:00
|
|
|
|
2022-10-12 14:01:00 -04:00
|
|
|
state_entry = await self.state.resolve_state_groups_for_events(
|
|
|
|
event.room_id, event_ids=event.prev_event_ids()
|
2021-05-07 09:54:09 -04:00
|
|
|
)
|
|
|
|
|
2022-10-12 14:01:00 -04:00
|
|
|
if state_entry.state_group:
|
|
|
|
await self._external_cache.set(
|
|
|
|
"event_to_prev_state_group",
|
|
|
|
event.event_id,
|
|
|
|
state_entry.state_group,
|
|
|
|
expiry_ms=60 * 60 * 1000,
|
|
|
|
)
|
2021-05-05 11:53:22 -04:00
|
|
|
|
2022-10-12 14:01:00 -04:00
|
|
|
if state_entry.state_group in self._external_cache_joined_hosts_updates:
|
|
|
|
return
|
|
|
|
|
|
|
|
with opentracing.start_active_span("get_joined_hosts"):
|
2023-07-20 10:51:28 -04:00
|
|
|
joined_hosts = (
|
|
|
|
await self._storage_controllers.state.get_joined_hosts(
|
|
|
|
event.room_id, state_entry
|
|
|
|
)
|
2022-10-12 14:01:00 -04:00
|
|
|
)
|
2021-01-26 08:57:31 -05:00
|
|
|
|
2022-10-12 14:01:00 -04:00
|
|
|
# Note that the expiry times must be larger than the expiry time in
|
|
|
|
# _external_cache_joined_hosts_updates.
|
|
|
|
await self._external_cache.set(
|
|
|
|
"get_joined_hosts",
|
|
|
|
str(state_entry.state_group),
|
|
|
|
list(joined_hosts),
|
|
|
|
expiry_ms=60 * 60 * 1000,
|
|
|
|
)
|
2021-01-26 08:57:31 -05:00
|
|
|
|
2024-03-13 12:46:44 -04:00
|
|
|
self._external_cache_joined_hosts_updates[state_entry.state_group] = (
|
|
|
|
None
|
|
|
|
)
|
2021-05-05 11:53:22 -04:00
|
|
|
|
2020-07-22 12:29:15 -04:00
|
|
|
async def _validate_canonical_alias(
|
2021-09-20 08:56:23 -04:00
|
|
|
self,
|
|
|
|
directory_handler: DirectoryHandler,
|
|
|
|
room_alias_str: str,
|
|
|
|
expected_room_id: str,
|
2020-07-22 12:29:15 -04:00
|
|
|
) -> None:
|
2020-03-23 15:21:54 -04:00
|
|
|
"""
|
|
|
|
Ensure that the given room alias points to the expected room ID.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
directory_handler: The directory handler object.
|
|
|
|
room_alias_str: The room alias to check.
|
|
|
|
expected_room_id: The room ID that the alias should point to.
|
|
|
|
"""
|
|
|
|
room_alias = RoomAlias.from_string(room_alias_str)
|
|
|
|
try:
|
2020-07-22 12:29:15 -04:00
|
|
|
mapping = await directory_handler.get_association(room_alias)
|
2020-03-23 15:21:54 -04:00
|
|
|
except SynapseError as e:
|
|
|
|
# Turn M_NOT_FOUND errors into M_BAD_ALIAS errors.
|
|
|
|
if e.errcode == Codes.NOT_FOUND:
|
|
|
|
raise SynapseError(
|
|
|
|
400,
|
|
|
|
"Room alias %s does not point to the room" % (room_alias_str,),
|
|
|
|
Codes.BAD_ALIAS,
|
|
|
|
)
|
|
|
|
raise
|
|
|
|
|
|
|
|
if mapping["room_id"] != expected_room_id:
|
|
|
|
raise SynapseError(
|
|
|
|
400,
|
|
|
|
"Room alias %s does not point to the room" % (room_alias_str,),
|
|
|
|
Codes.BAD_ALIAS,
|
|
|
|
)
|
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
async def persist_and_notify_client_events(
|
2020-07-22 12:29:15 -04:00
|
|
|
self,
|
|
|
|
requester: Requester,
|
2022-09-28 06:11:48 -04:00
|
|
|
events_and_context: List[Tuple[EventBase, EventContext]],
|
2020-07-22 12:29:15 -04:00
|
|
|
ratelimit: bool = True,
|
2021-04-08 17:38:54 -04:00
|
|
|
extra_users: Optional[List[UserID]] = None,
|
2020-10-13 07:07:56 -04:00
|
|
|
) -> EventBase:
|
2022-09-28 06:11:48 -04:00
|
|
|
"""Called when we have fully built the events, have already
|
|
|
|
calculated the push actions for the events, and checked auth.
|
2018-03-01 05:05:27 -05:00
|
|
|
|
2020-05-22 11:11:35 -04:00
|
|
|
This should only be run on the instance in charge of persisting events.
|
2020-10-13 07:07:56 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
Please note that if batch persisting events, an error in
|
|
|
|
handling any one of these events will result in all of the events being dropped.
|
|
|
|
|
2020-10-13 07:07:56 -04:00
|
|
|
Returns:
|
2022-09-28 06:11:48 -04:00
|
|
|
The persisted event, if one event is passed in, or the last event in the
|
|
|
|
list in the case of batch persisting. If only one event was persisted, the
|
|
|
|
returned event may be different than the given event if it was de-duplicated
|
|
|
|
(e.g. because we had already persisted an event with the same transaction ID.)
|
Handle race between persisting an event and un-partial stating a room (#13100)
Whenever we want to persist an event, we first compute an event context,
which includes the state at the event and a flag indicating whether the
state is partial. After a lot of processing, we finally try to store the
event in the database, which can fail for partial state events when the
containing room has been un-partial stated in the meantime.
We detect the race as a foreign key constraint failure in the data store
layer and turn it into a special `PartialStateConflictError` exception,
which makes its way up to the method in which we computed the event
context.
To make things difficult, the exception needs to cross a replication
request: `/fed_send_events` for events coming over federation and
`/send_event` for events from clients. We transport the
`PartialStateConflictError` as a `409 Conflict` over replication and
turn `409`s back into `PartialStateConflictError`s on the worker making
the request.
All client events go through
`EventCreationHandler.handle_new_client_event`, which is called in
*a lot* of places. Instead of trying to update all the code which
creates client events, we turn the `PartialStateConflictError` into a
`429 Too Many Requests` in
`EventCreationHandler.handle_new_client_event` and hope that clients
take it as a hint to retry their request.
On the federation event side, there are 7 places which compute event
contexts. 4 of them use outlier event contexts:
`FederationEventHandler._auth_and_persist_outliers_inner`,
`FederationHandler.do_knock`, `FederationHandler.on_invite_request` and
`FederationHandler.do_remotely_reject_invite`. These events won't have
the partial state flag, so we do not need to do anything for then.
The remaining 3 paths which create events are
`FederationEventHandler.process_remote_join`,
`FederationEventHandler.on_send_membership_event` and
`FederationEventHandler._process_received_pdu`.
We can't experience the race in `process_remote_join`, unless we're
handling an additional join into a partial state room, which currently
blocks, so we make no attempt to handle it correctly.
`on_send_membership_event` is only called by
`FederationServer._on_send_membership_event`, so we catch the
`PartialStateConflictError` there and retry just once.
`_process_received_pdu` is called by `on_receive_pdu` for incoming
events and `_process_pulled_event` for backfill. The latter should never
try to persist partial state events, so we ignore it. We catch the
`PartialStateConflictError` in `on_receive_pdu` and retry just once.
Refering to the graph of code paths in
https://github.com/matrix-org/synapse/issues/12988#issuecomment-1156857648
may make the above make more sense.
Signed-off-by: Sean Quah <seanq@matrix.org>
2022-07-05 11:12:52 -04:00
|
|
|
|
|
|
|
Raises:
|
|
|
|
PartialStateConflictError: if attempting to persist a partial state event in
|
|
|
|
a room that has been un-partial stated.
|
2018-02-15 11:30:10 -05:00
|
|
|
"""
|
2021-04-08 17:38:54 -04:00
|
|
|
extra_users = extra_users or []
|
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
for event, context in events_and_context:
|
|
|
|
assert self._events_shard_config.should_handle(
|
|
|
|
self._instance_name, event.room_id
|
|
|
|
)
|
2018-02-05 12:22:16 -05:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
if ratelimit:
|
|
|
|
# We check if this is a room admin redacting an event so that we
|
|
|
|
# can apply different ratelimiting. We do this by simply checking
|
|
|
|
# it's not a self-redaction (to avoid having to look up whether the
|
|
|
|
# user is actually admin or not).
|
2023-10-27 07:52:40 -04:00
|
|
|
is_admin_redaction = await self.is_admin_redaction(
|
|
|
|
event.type, event.sender, event.redacts
|
|
|
|
)
|
2022-09-28 06:11:48 -04:00
|
|
|
|
|
|
|
await self.request_ratelimiter.ratelimit(
|
|
|
|
requester, is_admin_redaction=is_admin_redaction
|
2019-09-11 06:16:17 -04:00
|
|
|
)
|
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
# run checks/actions on event based on type
|
|
|
|
if event.type == EventTypes.Member and event.membership == Membership.JOIN:
|
|
|
|
(
|
|
|
|
current_membership,
|
|
|
|
_,
|
|
|
|
) = await self.store.get_local_current_membership_for_user_in_room(
|
|
|
|
event.state_key, event.room_id
|
2019-09-11 06:16:17 -04:00
|
|
|
)
|
2022-09-28 06:11:48 -04:00
|
|
|
if current_membership != Membership.JOIN:
|
|
|
|
self._notifier.notify_user_joined_room(
|
|
|
|
event.event_id, event.room_id
|
|
|
|
)
|
2019-09-11 06:16:17 -04:00
|
|
|
|
2023-09-26 11:57:50 -04:00
|
|
|
if event.type == EventTypes.ServerACL:
|
|
|
|
self._storage_controllers.state.get_server_acl_for_room.invalidate(
|
|
|
|
(event.room_id,)
|
|
|
|
)
|
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
await self._maybe_kick_guest_users(event, context)
|
2016-05-11 04:09:20 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
if event.type == EventTypes.CanonicalAlias:
|
|
|
|
# Validate a newly added alias or newly added alt_aliases.
|
2022-07-19 07:45:17 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
original_alias = None
|
|
|
|
original_alt_aliases: object = []
|
2016-05-11 04:09:20 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
original_event_id = event.unsigned.get("replaces_state")
|
|
|
|
if original_event_id:
|
|
|
|
original_alias_event = await self.store.get_event(original_event_id)
|
2020-03-03 07:12:45 -05:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
if original_alias_event:
|
|
|
|
original_alias = original_alias_event.content.get("alias", None)
|
|
|
|
original_alt_aliases = original_alias_event.content.get(
|
|
|
|
"alt_aliases", []
|
|
|
|
)
|
2020-03-03 07:12:45 -05:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
# Check the alias is currently valid (if it has changed).
|
|
|
|
room_alias_str = event.content.get("alias", None)
|
|
|
|
directory_handler = self.hs.get_directory_handler()
|
|
|
|
if room_alias_str and room_alias_str != original_alias:
|
|
|
|
await self._validate_canonical_alias(
|
|
|
|
directory_handler, room_alias_str, event.room_id
|
|
|
|
)
|
2020-03-03 07:12:45 -05:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
# Check that alt_aliases is the proper form.
|
|
|
|
alt_aliases = event.content.get("alt_aliases", [])
|
|
|
|
if not isinstance(alt_aliases, (list, tuple)):
|
|
|
|
raise SynapseError(
|
|
|
|
400,
|
|
|
|
"The alt_aliases property must be a list.",
|
|
|
|
Codes.INVALID_PARAM,
|
|
|
|
)
|
2016-05-11 04:09:20 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
# If the old version of alt_aliases is of an unknown form,
|
|
|
|
# completely replace it.
|
|
|
|
if not isinstance(original_alt_aliases, (list, tuple)):
|
|
|
|
# TODO: check that the original_alt_aliases' entries are all strings
|
|
|
|
original_alt_aliases = []
|
|
|
|
|
|
|
|
# Check that each alias is currently valid.
|
|
|
|
new_alt_aliases = set(alt_aliases) - set(original_alt_aliases)
|
|
|
|
if new_alt_aliases:
|
|
|
|
for alias_str in new_alt_aliases:
|
|
|
|
await self._validate_canonical_alias(
|
|
|
|
directory_handler, alias_str, event.room_id
|
|
|
|
)
|
2020-03-03 07:12:45 -05:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
federation_handler = self.hs.get_federation_handler()
|
2020-03-03 07:12:45 -05:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
if event.type == EventTypes.Member:
|
|
|
|
if event.content["membership"] == Membership.INVITE:
|
2022-12-12 19:54:46 -05:00
|
|
|
maybe_upsert_event_field(
|
|
|
|
event,
|
|
|
|
event.unsigned,
|
|
|
|
"invite_room_state",
|
|
|
|
await self.store.get_stripped_room_state_from_event_context(
|
|
|
|
context,
|
|
|
|
self.room_prejoin_state_types,
|
|
|
|
membership_user_id=event.sender,
|
|
|
|
),
|
2020-03-23 15:21:54 -04:00
|
|
|
)
|
2020-03-03 07:12:45 -05:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
invitee = UserID.from_string(event.state_key)
|
|
|
|
if not self.hs.is_mine(invitee):
|
|
|
|
# TODO: Can we add signature from remote server in a nicer
|
|
|
|
# way? If we have been invited by a remote server, we need
|
|
|
|
# to get them to sign the event.
|
2016-05-11 04:09:20 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
returned_invite = await federation_handler.send_invite(
|
|
|
|
invitee.domain, event
|
|
|
|
)
|
|
|
|
event.unsigned.pop("room_state", None)
|
2016-05-11 04:09:20 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
# TODO: Make sure the signatures actually are correct.
|
|
|
|
event.signatures.update(returned_invite.signatures)
|
2016-05-11 04:09:20 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
if event.content["membership"] == Membership.KNOCK:
|
2022-12-12 19:54:46 -05:00
|
|
|
maybe_upsert_event_field(
|
|
|
|
event,
|
|
|
|
event.unsigned,
|
|
|
|
"knock_room_state",
|
|
|
|
await self.store.get_stripped_room_state_from_event_context(
|
|
|
|
context,
|
|
|
|
self.room_prejoin_state_types,
|
|
|
|
),
|
2016-05-11 04:09:20 -04:00
|
|
|
)
|
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
if event.type == EventTypes.Redaction:
|
|
|
|
assert event.redacts is not None
|
2016-05-11 04:09:20 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
original_event = await self.store.get_event(
|
|
|
|
event.redacts,
|
|
|
|
redact_behaviour=EventRedactBehaviour.as_is,
|
|
|
|
get_prev_content=False,
|
|
|
|
allow_rejected=False,
|
|
|
|
allow_none=True,
|
2021-06-09 14:39:51 -04:00
|
|
|
)
|
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
room_version = await self.store.get_room_version_id(event.room_id)
|
|
|
|
room_version_obj = KNOWN_ROOM_VERSIONS[room_version]
|
2021-11-02 09:55:52 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
# we can make some additional checks now if we have the original event.
|
|
|
|
if original_event:
|
|
|
|
if original_event.type == EventTypes.Create:
|
|
|
|
raise AuthError(403, "Redacting create events is not permitted")
|
2019-07-17 14:08:02 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
if original_event.room_id != event.room_id:
|
|
|
|
raise SynapseError(
|
|
|
|
400, "Cannot redact event from a different room"
|
|
|
|
)
|
Allow room creator to send MSC2716 related events in existing room versions (#10566)
* Allow room creator to send MSC2716 related events in existing room versions
Discussed at https://github.com/matrix-org/matrix-doc/pull/2716/#discussion_r682474869
Restoring `get_create_event_for_room_txn` from,
https://github.com/matrix-org/synapse/pull/10245/commits/44bb3f0cf5cb365ef9281554daceeecfb17cc94d
* Add changelog
* Stop people from trying to redact MSC2716 events in unsupported room versions
* Populate rooms.creator column for easy lookup
> From some [out of band discussion](https://matrix.to/#/!UytJQHLQYfvYWsGrGY:jki.re/$p2fKESoFst038x6pOOmsY0C49S2gLKMr0jhNMz_JJz0?via=jki.re&via=matrix.org), my plan is to use `rooms.creator`. But currently, we don't fill in `creator` for remote rooms when a user is invited to a room for example. So we need to add some code to fill in `creator` wherever we add to the `rooms` table. And also add a background update to fill in the rows missing `creator` (we can use the same logic that `get_create_event_for_room_txn` is doing by looking in the state events to get the `creator`).
>
> https://github.com/matrix-org/synapse/pull/10566#issuecomment-901616642
* Remove and switch away from get_create_event_for_room_txn
* Fix no create event being found because no state events persisted yet
* Fix and add tests for rooms creator bg update
* Populate rooms.creator field for easy lookup
Part of https://github.com/matrix-org/synapse/pull/10566
- Fill in creator whenever we insert into the rooms table
- Add background update to backfill any missing creator values
* Add changelog
* Fix usage
* Remove extra delta already included in #10697
* Don't worry about setting creator for invite
* Only iterate over rows missing the creator
See https://github.com/matrix-org/synapse/pull/10697#discussion_r695940898
* Use constant to fetch room creator field
See https://github.com/matrix-org/synapse/pull/10697#discussion_r696803029
* More protection from other random types
See https://github.com/matrix-org/synapse/pull/10697#discussion_r696806853
* Move new background update to end of list
See https://github.com/matrix-org/synapse/pull/10697#discussion_r696814181
* Fix query casing
* Fix ambiguity iterating over cursor instead of list
Fix `psycopg2.ProgrammingError: no results to fetch` error
when tests run with Postgres.
```
SYNAPSE_POSTGRES=1 SYNAPSE_TEST_LOG_LEVEL=INFO python -m twisted.trial tests.storage.databases.main.test_room
```
---
We use `txn.fetchall` because it will return the results as a
list or an empty list when there are no results.
Docs:
> `cursor` objects are iterable, so, instead of calling explicitly fetchone() in a loop, the object itself can be used:
>
> https://www.psycopg.org/docs/cursor.html#cursor-iterable
And I'm guessing iterating over a raw cursor does something weird when there are no results.
---
Test CI failure: https://github.com/matrix-org/synapse/pull/10697/checks?check_run_id=3468916530
```
tests.test_visibility.FilterEventsForServerTestCase.test_large_room
===============================================================================
[FAIL]
Traceback (most recent call last):
File "/home/runner/work/synapse/synapse/tests/storage/databases/main/test_room.py", line 85, in test_background_populate_rooms_creator_column
self.get_success(
File "/home/runner/work/synapse/synapse/tests/unittest.py", line 500, in get_success
return self.successResultOf(d)
File "/home/runner/work/synapse/synapse/.tox/py/lib/python3.9/site-packages/twisted/trial/_synctest.py", line 700, in successResultOf
self.fail(
twisted.trial.unittest.FailTest: Success result expected on <Deferred at 0x7f4022f3eb50 current result: None>, found failure result instead:
Traceback (most recent call last):
File "/home/runner/work/synapse/synapse/.tox/py/lib/python3.9/site-packages/twisted/internet/defer.py", line 701, in errback
self._startRunCallbacks(fail)
File "/home/runner/work/synapse/synapse/.tox/py/lib/python3.9/site-packages/twisted/internet/defer.py", line 764, in _startRunCallbacks
self._runCallbacks()
File "/home/runner/work/synapse/synapse/.tox/py/lib/python3.9/site-packages/twisted/internet/defer.py", line 858, in _runCallbacks
current.result = callback( # type: ignore[misc]
File "/home/runner/work/synapse/synapse/.tox/py/lib/python3.9/site-packages/twisted/internet/defer.py", line 1751, in gotResult
current_context.run(_inlineCallbacks, r, gen, status)
--- <exception caught here> ---
File "/home/runner/work/synapse/synapse/.tox/py/lib/python3.9/site-packages/twisted/internet/defer.py", line 1657, in _inlineCallbacks
result = current_context.run(
File "/home/runner/work/synapse/synapse/.tox/py/lib/python3.9/site-packages/twisted/python/failure.py", line 500, in throwExceptionIntoGenerator
return g.throw(self.type, self.value, self.tb)
File "/home/runner/work/synapse/synapse/synapse/storage/background_updates.py", line 224, in do_next_background_update
await self._do_background_update(desired_duration_ms)
File "/home/runner/work/synapse/synapse/synapse/storage/background_updates.py", line 261, in _do_background_update
items_updated = await update_handler(progress, batch_size)
File "/home/runner/work/synapse/synapse/synapse/storage/databases/main/room.py", line 1399, in _background_populate_rooms_creator_column
end = await self.db_pool.runInteraction(
File "/home/runner/work/synapse/synapse/synapse/storage/database.py", line 686, in runInteraction
result = await self.runWithConnection(
File "/home/runner/work/synapse/synapse/synapse/storage/database.py", line 791, in runWithConnection
return await make_deferred_yieldable(
File "/home/runner/work/synapse/synapse/.tox/py/lib/python3.9/site-packages/twisted/internet/defer.py", line 858, in _runCallbacks
current.result = callback( # type: ignore[misc]
File "/home/runner/work/synapse/synapse/tests/server.py", line 425, in <lambda>
d.addCallback(lambda x: function(*args, **kwargs))
File "/home/runner/work/synapse/synapse/.tox/py/lib/python3.9/site-packages/twisted/enterprise/adbapi.py", line 293, in _runWithConnection
compat.reraise(excValue, excTraceback)
File "/home/runner/work/synapse/synapse/.tox/py/lib/python3.9/site-packages/twisted/python/deprecate.py", line 298, in deprecatedFunction
return function(*args, **kwargs)
File "/home/runner/work/synapse/synapse/.tox/py/lib/python3.9/site-packages/twisted/python/compat.py", line 404, in reraise
raise exception.with_traceback(traceback)
File "/home/runner/work/synapse/synapse/.tox/py/lib/python3.9/site-packages/twisted/enterprise/adbapi.py", line 284, in _runWithConnection
result = func(conn, *args, **kw)
File "/home/runner/work/synapse/synapse/synapse/storage/database.py", line 786, in inner_func
return func(db_conn, *args, **kwargs)
File "/home/runner/work/synapse/synapse/synapse/storage/database.py", line 554, in new_transaction
r = func(cursor, *args, **kwargs)
File "/home/runner/work/synapse/synapse/synapse/storage/databases/main/room.py", line 1375, in _background_populate_rooms_creator_column_txn
for room_id, event_json in txn:
psycopg2.ProgrammingError: no results to fetch
```
* Move code not under the MSC2716 room version underneath an experimental config option
See https://github.com/matrix-org/synapse/pull/10566#issuecomment-906437909
* Add ordering to rooms creator background update
See https://github.com/matrix-org/synapse/pull/10697#discussion_r696815277
* Add comment to better document constant
See https://github.com/matrix-org/synapse/pull/10697#discussion_r699674458
* Use constant field
2021-09-04 01:58:49 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
if original_event.type == EventTypes.ServerACL:
|
|
|
|
raise AuthError(
|
|
|
|
403, "Redacting server ACL events is not permitted"
|
|
|
|
)
|
2022-05-20 04:54:12 -04:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
event_types = event_auth.auth_types_for_event(event.room_version, event)
|
|
|
|
prev_state_ids = await context.get_prev_state_ids(
|
|
|
|
StateFilter.from_types(event_types)
|
2021-09-28 22:23:16 -04:00
|
|
|
)
|
2022-09-28 06:11:48 -04:00
|
|
|
|
|
|
|
auth_events_ids = self._event_auth_handler.compute_auth_events(
|
|
|
|
event, prev_state_ids, for_verification=True
|
|
|
|
)
|
|
|
|
auth_events_map = await self.store.get_events(auth_events_ids)
|
|
|
|
auth_events = {
|
|
|
|
(e.type, e.state_key): e for e in auth_events_map.values()
|
|
|
|
}
|
|
|
|
|
|
|
|
if event_auth.check_redaction(
|
|
|
|
room_version_obj, event, auth_events=auth_events
|
|
|
|
):
|
|
|
|
# this user doesn't have 'redact' rights, so we need to do some more
|
|
|
|
# checks on the original event. Let's start by checking the original
|
|
|
|
# event exists.
|
|
|
|
if not original_event:
|
|
|
|
raise NotFoundError(
|
|
|
|
"Could not find event %s" % (event.redacts,)
|
2021-11-02 09:55:52 -04:00
|
|
|
)
|
2022-09-28 06:11:48 -04:00
|
|
|
|
|
|
|
if event.user_id != original_event.user_id:
|
|
|
|
raise AuthError(
|
|
|
|
403, "You don't have permission to redact events"
|
|
|
|
)
|
|
|
|
|
|
|
|
# all the checks are done.
|
|
|
|
event.internal_metadata.recheck_redaction = False
|
|
|
|
|
|
|
|
if event.type == EventTypes.Create:
|
|
|
|
prev_state_ids = await context.get_prev_state_ids()
|
|
|
|
if prev_state_ids:
|
|
|
|
raise AuthError(403, "Changing the room create event is forbidden")
|
|
|
|
|
|
|
|
assert self._storage_controllers.persistence is not None
|
2020-10-13 07:07:56 -04:00
|
|
|
(
|
2022-09-28 06:11:48 -04:00
|
|
|
persisted_events,
|
2020-10-13 07:07:56 -04:00
|
|
|
max_stream_token,
|
2022-09-28 06:11:48 -04:00
|
|
|
) = await self._storage_controllers.persistence.persist_events(
|
2023-06-16 15:12:24 -04:00
|
|
|
events_and_context,
|
2021-06-22 05:02:53 -04:00
|
|
|
)
|
2016-05-11 04:09:20 -04:00
|
|
|
|
2022-10-05 13:12:48 -04:00
|
|
|
events_and_pos = []
|
2022-09-28 06:11:48 -04:00
|
|
|
for event in persisted_events:
|
|
|
|
if self._ephemeral_events_enabled:
|
|
|
|
# If there's an expiry timestamp on the event, schedule its expiry.
|
|
|
|
self._message_handler.maybe_schedule_expiry(event)
|
2019-12-03 14:19:45 -05:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
stream_ordering = event.internal_metadata.stream_ordering
|
|
|
|
assert stream_ordering is not None
|
|
|
|
pos = PersistedEventPosition(self._instance_name, stream_ordering)
|
2022-10-05 13:12:48 -04:00
|
|
|
events_and_pos.append((event, pos))
|
2018-02-06 12:27:00 -05:00
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
if event.type == EventTypes.Message:
|
|
|
|
# We don't want to block sending messages on any presence code. This
|
|
|
|
# matters as sometimes presence code can take a while.
|
2023-01-31 06:31:52 -05:00
|
|
|
run_as_background_process(
|
2023-08-28 13:08:49 -04:00
|
|
|
"bump_presence_active_time",
|
|
|
|
self._bump_active_time,
|
|
|
|
requester.user,
|
|
|
|
requester.device_id,
|
2023-01-31 06:31:52 -05:00
|
|
|
)
|
2018-04-27 06:07:40 -04:00
|
|
|
|
2022-10-05 13:12:48 -04:00
|
|
|
async def _notify() -> None:
|
|
|
|
try:
|
|
|
|
await self.notifier.on_new_room_events(
|
|
|
|
events_and_pos, max_stream_token, extra_users=extra_users
|
|
|
|
)
|
|
|
|
except Exception:
|
|
|
|
logger.exception("Error notifying about new room events")
|
|
|
|
|
|
|
|
run_in_background(_notify)
|
|
|
|
|
2022-09-28 06:11:48 -04:00
|
|
|
return persisted_events[-1]
|
2020-05-22 09:21:54 -04:00
|
|
|
|
2023-10-27 07:52:40 -04:00
|
|
|
async def is_admin_redaction(
|
|
|
|
self, event_type: str, sender: str, redacts: Optional[str]
|
|
|
|
) -> bool:
|
|
|
|
"""Return whether the event is a redaction made by an admin, and thus
|
|
|
|
should use a different ratelimiter.
|
|
|
|
"""
|
|
|
|
if event_type != EventTypes.Redaction:
|
|
|
|
return False
|
|
|
|
|
|
|
|
assert redacts is not None
|
|
|
|
|
|
|
|
original_event = await self.store.get_event(
|
|
|
|
redacts,
|
|
|
|
redact_behaviour=EventRedactBehaviour.as_is,
|
|
|
|
get_prev_content=False,
|
|
|
|
allow_rejected=False,
|
|
|
|
allow_none=True,
|
|
|
|
)
|
|
|
|
|
|
|
|
return bool(original_event and sender != original_event.sender)
|
|
|
|
|
2021-09-06 07:17:16 -04:00
|
|
|
async def _maybe_kick_guest_users(
|
|
|
|
self, event: EventBase, context: EventContext
|
|
|
|
) -> None:
|
|
|
|
if event.type != EventTypes.GuestAccess:
|
|
|
|
return
|
|
|
|
|
|
|
|
guest_access = event.content.get(EventContentFields.GUEST_ACCESS)
|
|
|
|
if guest_access == GuestAccess.CAN_JOIN:
|
|
|
|
return
|
|
|
|
|
|
|
|
current_state_ids = await context.get_current_state_ids()
|
|
|
|
|
|
|
|
# since this is a client-generated event, it cannot be an outlier and we must
|
|
|
|
# therefore have the state ids.
|
|
|
|
assert current_state_ids is not None
|
|
|
|
current_state_dict = await self.store.get_events(
|
|
|
|
list(current_state_ids.values())
|
|
|
|
)
|
|
|
|
current_state = list(current_state_dict.values())
|
|
|
|
logger.info("maybe_kick_guest_users %r", current_state)
|
|
|
|
await self.hs.get_room_member_handler().kick_guest_users(current_state)
|
|
|
|
|
2023-08-28 13:08:49 -04:00
|
|
|
async def _bump_active_time(self, user: UserID, device_id: Optional[str]) -> None:
|
2018-04-27 06:07:40 -04:00
|
|
|
try:
|
|
|
|
presence = self.hs.get_presence_handler()
|
2023-08-28 13:08:49 -04:00
|
|
|
await presence.bump_presence_active_time(user, device_id)
|
2018-04-27 06:07:40 -04:00
|
|
|
except Exception:
|
|
|
|
logger.exception("Error bumping presence active time")
|
2019-06-17 13:04:42 -04:00
|
|
|
|
2021-09-20 08:56:23 -04:00
|
|
|
async def _send_dummy_events_to_fill_extremities(self) -> None:
|
2019-06-17 13:04:42 -04:00
|
|
|
"""Background task to send dummy events into rooms that have a large
|
|
|
|
number of extremities
|
|
|
|
"""
|
2019-09-26 06:47:53 -04:00
|
|
|
self._expire_rooms_to_exclude_from_dummy_event_insertion()
|
2020-05-01 10:15:36 -04:00
|
|
|
room_ids = await self.store.get_rooms_with_many_extremities(
|
2020-05-07 05:35:23 -04:00
|
|
|
min_count=self._dummy_events_threshold,
|
2019-09-26 06:47:53 -04:00
|
|
|
limit=5,
|
|
|
|
room_id_filter=self._rooms_to_exclude_from_dummy_event_insertion.keys(),
|
2019-06-17 13:04:42 -04:00
|
|
|
)
|
|
|
|
|
|
|
|
for room_id in room_ids:
|
2023-07-31 05:58:03 -04:00
|
|
|
async with self._worker_lock_handler.acquire_read_write_lock(
|
2023-08-16 10:19:54 -04:00
|
|
|
NEW_EVENT_DURING_PURGE_LOCK_NAME, room_id, write=False
|
2023-07-31 05:58:03 -04:00
|
|
|
):
|
|
|
|
dummy_event_sent = await self._send_dummy_event_for_room(room_id)
|
2019-06-17 13:04:42 -04:00
|
|
|
|
2019-09-26 06:47:53 -04:00
|
|
|
if not dummy_event_sent:
|
|
|
|
# Did not find a valid user in the room, so remove from future attempts
|
|
|
|
# Exclusion is time limited, so the room will be rechecked in the future
|
|
|
|
# dependent on _DUMMY_EVENT_ROOM_EXCLUSION_EXPIRY
|
|
|
|
logger.info(
|
|
|
|
"Failed to send dummy event into room %s. Will exclude it from "
|
|
|
|
"future attempts until cache expires" % (room_id,)
|
|
|
|
)
|
|
|
|
now = self.clock.time_msec()
|
|
|
|
self._rooms_to_exclude_from_dummy_event_insertion[room_id] = now
|
|
|
|
|
2020-09-23 13:18:43 -04:00
|
|
|
async def _send_dummy_event_for_room(self, room_id: str) -> bool:
|
|
|
|
"""Attempt to send a dummy event for the given room.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
room_id: room to try to send an event from
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
True if a dummy event was successfully sent. False if no user was able
|
|
|
|
to send an event.
|
|
|
|
"""
|
|
|
|
|
|
|
|
# For each room we need to find a joined member we can use to send
|
|
|
|
# the dummy event with.
|
2022-07-18 09:19:11 -04:00
|
|
|
members = await self.store.get_local_users_in_room(room_id)
|
2020-09-23 13:18:43 -04:00
|
|
|
for user_id in members:
|
2020-11-17 05:51:25 -05:00
|
|
|
requester = create_requester(user_id, authenticated_entity=self.server_name)
|
2020-09-23 13:18:43 -04:00
|
|
|
try:
|
2022-12-15 11:04:23 -05:00
|
|
|
# Try several times, it could fail with PartialStateConflictError
|
|
|
|
# in handle_new_client_event, cf comment in except block.
|
|
|
|
max_retries = 5
|
|
|
|
for i in range(max_retries):
|
|
|
|
try:
|
2023-02-24 16:15:29 -05:00
|
|
|
event, unpersisted_context = await self.create_event(
|
2022-12-15 11:04:23 -05:00
|
|
|
requester,
|
|
|
|
{
|
|
|
|
"type": EventTypes.Dummy,
|
|
|
|
"content": {},
|
|
|
|
"room_id": room_id,
|
|
|
|
"sender": user_id,
|
|
|
|
},
|
|
|
|
)
|
2023-02-24 16:15:29 -05:00
|
|
|
context = await unpersisted_context.persist(event)
|
2020-09-23 13:18:43 -04:00
|
|
|
|
2022-12-15 11:04:23 -05:00
|
|
|
event.internal_metadata.proactively_send = False
|
2020-09-23 13:18:43 -04:00
|
|
|
|
2022-12-15 11:04:23 -05:00
|
|
|
# Since this is a dummy-event it is OK if it is sent by a
|
|
|
|
# shadow-banned user.
|
|
|
|
await self.handle_new_client_event(
|
|
|
|
requester,
|
|
|
|
events_and_context=[(event, context)],
|
|
|
|
ratelimit=False,
|
|
|
|
ignore_shadow_ban=True,
|
|
|
|
)
|
|
|
|
|
|
|
|
break
|
|
|
|
except PartialStateConflictError as e:
|
|
|
|
# Persisting couldn't happen because the room got un-partial stated
|
|
|
|
# in the meantime and context needs to be recomputed, so let's do so.
|
|
|
|
if i == max_retries - 1:
|
|
|
|
raise e
|
2020-09-23 13:18:43 -04:00
|
|
|
return True
|
|
|
|
except AuthError:
|
|
|
|
logger.info(
|
|
|
|
"Failed to send dummy event into room %s for user %s due to "
|
|
|
|
"lack of power. Will try another user" % (room_id, user_id)
|
|
|
|
)
|
|
|
|
return False
|
|
|
|
|
2021-09-20 08:56:23 -04:00
|
|
|
def _expire_rooms_to_exclude_from_dummy_event_insertion(self) -> None:
|
2019-09-26 06:47:53 -04:00
|
|
|
expire_before = self.clock.time_msec() - _DUMMY_EVENT_ROOM_EXCLUSION_EXPIRY
|
|
|
|
to_expire = set()
|
|
|
|
for room_id, time in self._rooms_to_exclude_from_dummy_event_insertion.items():
|
|
|
|
if time < expire_before:
|
|
|
|
to_expire.add(room_id)
|
|
|
|
for room_id in to_expire:
|
|
|
|
logger.debug(
|
|
|
|
"Expiring room id %s from dummy event insertion exclusion cache",
|
|
|
|
room_id,
|
2019-06-17 13:04:42 -04:00
|
|
|
)
|
2019-09-26 06:47:53 -04:00
|
|
|
del self._rooms_to_exclude_from_dummy_event_insertion[room_id]
|
2020-10-13 13:53:56 -04:00
|
|
|
|
|
|
|
async def _rebuild_event_after_third_party_rules(
|
|
|
|
self, third_party_result: dict, original_event: EventBase
|
2023-02-09 16:05:02 -05:00
|
|
|
) -> Tuple[EventBase, UnpersistedEventContextBase]:
|
2020-10-13 13:53:56 -04:00
|
|
|
# the third_party_event_rules want to replace the event.
|
2023-02-09 16:05:02 -05:00
|
|
|
# we do some basic checks, and then return the replacement event.
|
2020-10-13 13:53:56 -04:00
|
|
|
|
|
|
|
# Construct a new EventBuilder and validate it, which helps with the
|
|
|
|
# rest of these checks.
|
|
|
|
try:
|
|
|
|
builder = self.event_builder_factory.for_room_version(
|
|
|
|
original_event.room_version, third_party_result
|
|
|
|
)
|
|
|
|
self.validator.validate_builder(builder)
|
|
|
|
except SynapseError as e:
|
|
|
|
raise Exception(
|
|
|
|
"Third party rules module created an invalid event: " + e.msg,
|
|
|
|
)
|
|
|
|
|
|
|
|
immutable_fields = [
|
|
|
|
# changing the room is going to break things: we've already checked that the
|
|
|
|
# room exists, and are holding a concurrency limiter token for that room.
|
|
|
|
# Also, we might need to use a different room version.
|
|
|
|
"room_id",
|
|
|
|
# changing the type or state key might work, but we'd need to check that the
|
|
|
|
# calling functions aren't making assumptions about them.
|
|
|
|
"type",
|
|
|
|
"state_key",
|
|
|
|
]
|
|
|
|
|
|
|
|
for k in immutable_fields:
|
|
|
|
if getattr(builder, k, None) != original_event.get(k):
|
|
|
|
raise Exception(
|
|
|
|
"Third party rules module created an invalid event: "
|
|
|
|
"cannot change field " + k
|
|
|
|
)
|
|
|
|
|
|
|
|
# check that the new sender belongs to this HS
|
|
|
|
if not self.hs.is_mine_id(builder.sender):
|
|
|
|
raise Exception(
|
|
|
|
"Third party rules module created an invalid event: "
|
|
|
|
"invalid sender " + builder.sender
|
|
|
|
)
|
|
|
|
|
|
|
|
# copy over the original internal metadata
|
|
|
|
for k, v in original_event.internal_metadata.get_dict().items():
|
|
|
|
setattr(builder.internal_metadata, k, v)
|
|
|
|
|
2021-07-08 07:00:05 -04:00
|
|
|
# modules can send new state events, so we re-calculate the auth events just in
|
|
|
|
# case.
|
|
|
|
prev_event_ids = await self.store.get_prev_events_for_room(builder.room_id)
|
|
|
|
|
2020-10-16 08:39:46 -04:00
|
|
|
event = await builder.build(
|
2021-07-08 07:00:05 -04:00
|
|
|
prev_event_ids=prev_event_ids,
|
|
|
|
auth_event_ids=None,
|
2020-10-16 08:39:46 -04:00
|
|
|
)
|
2020-10-13 13:53:56 -04:00
|
|
|
|
|
|
|
# we rebuild the event context, to be on the safe side. If nothing else,
|
|
|
|
# delta_ids might need an update.
|
2023-02-09 16:05:02 -05:00
|
|
|
context = await self.state.calculate_context_info(event)
|
|
|
|
|
2020-10-13 13:53:56 -04:00
|
|
|
return event, context
|