anonymousland-synapse/synapse/handlers/federation.py

2788 lines
105 KiB
Python
Raw Normal View History

2014-08-12 10:10:52 -04:00
# -*- coding: utf-8 -*-
2016-01-06 23:26:29 -05:00
# Copyright 2014-2016 OpenMarket Ltd
# Copyright 2017-2018 New Vector Ltd
# Copyright 2019 The Matrix.org Foundation C.I.C.
2014-08-12 10:10:52 -04:00
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
2014-08-12 10:10:52 -04:00
"""Contains handlers for federation events."""
import itertools
import logging
import six
from six import iteritems, itervalues
from six.moves import http_client, zip
2018-07-09 02:09:20 -04:00
from signedjson.key import decode_verify_key_bytes
from signedjson.sign import verify_signed_json
from unpaddedbase64 import decode_base64
2014-08-12 10:10:52 -04:00
2018-07-09 02:09:20 -04:00
from twisted.internet import defer
2014-08-12 10:10:52 -04:00
from synapse import event_auth
from synapse.api.constants import EventTypes, Membership, RejectedReason
from synapse.api.errors import (
2018-07-09 02:09:20 -04:00
AuthError,
CodeMessageException,
Codes,
FederationDeniedError,
2018-07-09 02:09:20 -04:00
FederationError,
RequestSendFailed,
2018-07-09 02:09:20 -04:00
StoreError,
SynapseError,
)
from synapse.api.room_versions import KNOWN_ROOM_VERSIONS, RoomVersions
2019-01-23 15:05:44 -05:00
from synapse.crypto.event_signing import compute_event_signature
2019-02-12 05:31:21 -05:00
from synapse.event_auth import auth_types_for_event
from synapse.events.snapshot import EventContext
2018-07-09 02:09:20 -04:00
from synapse.events.validator import EventValidator
from synapse.logging.context import (
make_deferred_yieldable,
nested_logging_context,
preserve_fn,
run_in_background,
)
from synapse.logging.utils import log_function
from synapse.replication.http.federation import (
2018-08-09 05:29:48 -04:00
ReplicationCleanRoomRestServlet,
ReplicationFederationSendEventsRestServlet,
)
from synapse.replication.http.membership import ReplicationUserJoinedLeftRoomRestServlet
2018-10-24 04:44:22 -04:00
from synapse.state import StateResolutionStore, resolve_events_with_store
from synapse.types import UserID, get_domain_from_id
from synapse.util import unwrapFirstError
from synapse.util.async_helpers import Linearizer
2018-07-09 02:09:20 -04:00
from synapse.util.distributor import user_joined_room
from synapse.util.retryutils import NotRetryingDestination
from synapse.visibility import filter_events_for_server
2018-07-09 02:09:20 -04:00
from ._base import BaseHandler
2014-08-12 10:10:52 -04:00
logger = logging.getLogger(__name__)
def shortstr(iterable, maxitems=5):
"""If iterable has maxitems or fewer, return the stringification of a list
containing those items.
Otherwise, return the stringification of a a list with the first maxitems items,
followed by "...".
Args:
iterable (Iterable): iterable to truncate
maxitems (int): number of items to return before truncating
Returns:
unicode
"""
items = list(itertools.islice(iterable, maxitems + 1))
if len(items) <= maxitems:
return str(items)
2019-06-20 05:32:02 -04:00
return "[" + ", ".join(repr(r) for r in items[:maxitems]) + ", ...]"
2014-08-12 10:10:52 -04:00
class FederationHandler(BaseHandler):
"""Handles events that originated from federation.
Responsible for:
a) handling received Pdus before handing them on as Events to the rest
of the home server (including auth and state conflict resoultion)
b) converting events that were produced by local clients that may need
to be sent to remote home servers.
c) doing the necessary dances to invite remote users and join remote
rooms.
"""
2014-08-12 10:10:52 -04:00
def __init__(self, hs):
super(FederationHandler, self).__init__(hs)
self.hs = hs
2019-01-30 05:53:17 -05:00
self.store = hs.get_datastore()
2019-10-23 07:02:36 -04:00
self.storage = hs.get_storage()
2019-10-23 12:25:54 -04:00
self.state_store = self.storage.state
self.federation_client = hs.get_federation_client()
self.state_handler = hs.get_state_handler()
self.server_name = hs.hostname
2014-11-14 11:45:39 -05:00
self.keyring = hs.get_keyring()
2017-05-18 13:17:40 -04:00
self.action_generator = hs.get_action_generator()
self.is_mine_id = hs.is_mine_id
self.pusher_pool = hs.get_pusherpool()
self.spam_checker = hs.get_spam_checker()
2018-01-15 11:52:07 -05:00
self.event_creation_handler = hs.get_event_creation_handler()
self._server_notices_mxid = hs.config.server_notices_mxid
self.config = hs.config
self.http_client = hs.get_simple_http_client()
2019-06-20 05:32:02 -04:00
self._send_events_to_master = ReplicationFederationSendEventsRestServlet.make_client(
hs
)
2019-06-20 05:32:02 -04:00
self._notify_user_membership_change = ReplicationUserJoinedLeftRoomRestServlet.make_client(
hs
)
2019-06-20 05:32:02 -04:00
self._clean_room_for_join_client = ReplicationCleanRoomRestServlet.make_client(
hs
2018-08-09 05:29:48 -04:00
)
# When joining a room we need to queue any events for that room up
self.room_queues = {}
self._room_pdu_linearizer = Linearizer("fed_room_pdu")
self.third_party_event_rules = hs.get_third_party_event_rules()
@defer.inlineCallbacks
2019-06-20 05:32:02 -04:00
def on_receive_pdu(self, origin, pdu, sent_to_us_directly=False):
""" Process a PDU received via a federation /send/ transaction, or
via backfill of missing prev_events
Args:
origin (str): server which initiated the /send/ transaction. Will
be used to fetch missing events or state.
pdu (FrozenEvent): received PDU
sent_to_us_directly (bool): True if this event was pushed to us; False if
we pulled it as the result of a missing prev_event.
Returns (Deferred): completes with None
"""
room_id = pdu.room_id
event_id = pdu.event_id
2019-06-20 05:32:02 -04:00
logger.info("[%s %s] handling received PDU: %s", room_id, event_id, pdu)
# We reprocess pdus when we have seen them only as outliers
existing = yield self.store.get_event(
2019-06-20 05:32:02 -04:00
event_id, allow_none=True, allow_rejected=True
)
# FIXME: Currently we fetch an event again when we already have it
# if it has been marked as an outlier.
2019-06-20 05:32:02 -04:00
already_seen = existing and (
not existing.internal_metadata.is_outlier()
or pdu.internal_metadata.is_outlier()
)
if already_seen:
logger.debug("[%s %s]: Already seen pdu", room_id, event_id)
return
# do some initial sanity-checking of the event. In particular, make
# sure it doesn't have hundreds of prev_events or auth_events, which
2018-04-17 18:41:12 -04:00
# could cause a huge state resolution or cascade of event fetches.
try:
self._sanity_check_event(pdu)
except SynapseError as err:
logger.warning(
2019-06-20 05:32:02 -04:00
"[%s %s] Received event failed sanity checks", room_id, event_id
)
2019-06-20 05:32:02 -04:00
raise FederationError("ERROR", err.code, err.msg, affected=pdu.event_id)
# If we are currently in the process of joining this room, then we
# queue up events for later processing.
if room_id in self.room_queues:
logger.info(
"[%s %s] Queuing PDU from %s for now: join in progress",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
origin,
)
self.room_queues[room_id].append((pdu, origin))
return
# If we're not in the room just ditch the event entirely. This is
# probably an old server that has come back and thinks we're still in
# the room (or we've been rejoined to the room by a state reset).
#
# Note that if we were never in the room then we would have already
# dropped the event, since we wouldn't know the room version.
2019-06-20 05:32:02 -04:00
is_in_room = yield self.auth.check_host_in_room(room_id, self.server_name)
if not is_in_room:
logger.info(
"[%s %s] Ignoring PDU from %s as we're not in the room",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
origin,
)
return None
state = None
auth_chain = []
# Get missing pdus if necessary.
if not pdu.internal_metadata.is_outlier():
# We only backfill backwards to the min depth.
2019-06-20 05:32:02 -04:00
min_depth = yield self.get_min_depth_for_context(pdu.room_id)
2019-06-20 05:32:02 -04:00
logger.debug("[%s %s] min_depth: %d", room_id, event_id, min_depth)
prevs = set(pdu.prev_event_ids())
seen = yield self.store.have_seen_events(prevs)
if min_depth and pdu.depth < min_depth:
# This is so that we don't notify the user about this
# message, to work around the fact that some events will
# reference really really old events we really don't want to
# send to the clients.
pdu.internal_metadata.outlier = True
elif min_depth and pdu.depth > min_depth:
missing_prevs = prevs - seen
if sent_to_us_directly and missing_prevs:
# If we're missing stuff, ensure we only fetch stuff one
# at a time.
logger.info(
"[%s %s] Acquiring room lock to fetch %d missing prev_events: %s",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
len(missing_prevs),
shortstr(missing_prevs),
)
with (yield self._room_pdu_linearizer.queue(pdu.room_id)):
logger.info(
"[%s %s] Acquired room lock to fetch %d missing prev_events",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
len(missing_prevs),
)
yield self._get_missing_events_for_pdu(
origin, pdu, prevs, min_depth
)
# Update the set of things we've seen after trying to
# fetch the missing stuff
seen = yield self.store.have_seen_events(prevs)
if not prevs - seen:
logger.info(
"[%s %s] Found all missing prev_events",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
)
elif missing_prevs:
logger.info(
"[%s %s] Not recursively fetching %d missing prev_events: %s",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
len(missing_prevs),
shortstr(missing_prevs),
)
if prevs - seen:
# We've still not been able to get all of the prev_events for this event.
#
# In this case, we need to fall back to asking another server in the
# federation for the state at this event. That's ok provided we then
# resolve the state against other bits of the DAG before using it (which
# will ensure that you can't just take over a room by sending an event,
# withholding its prev_events, and declaring yourself to be an admin in
# the subsequent state request).
#
# Now, if we're pulling this event as a missing prev_event, then clearly
# this event is not going to become the only forward-extremity and we are
# guaranteed to resolve its state against our existing forward
# extremities, so that should be fine.
#
# On the other hand, if this event was pushed to us, it is possible for
# it to become the only forward-extremity in the room, and we would then
# trust its state to be the state for the whole room. This is very bad.
# Further, if the event was pushed to us, there is no excuse for us not to
# have all the prev_events. We therefore reject any such events.
#
# XXX this really feels like it could/should be merged with the above,
# but there is an interaction with min_depth that I'm not really
# following.
if sent_to_us_directly:
logger.warning(
"[%s %s] Rejecting: failed to fetch %d prev events: %s",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
len(prevs - seen),
shortstr(prevs - seen),
)
raise FederationError(
"ERROR",
403,
(
"Your server isn't divulging details about prev_events "
"referenced in this event."
),
affected=pdu.event_id,
)
# Calculate the state after each of the previous events, and
# resolve them to find the correct state at the current event.
auth_chains = set()
2019-06-20 05:32:02 -04:00
event_map = {event_id: pdu}
try:
2018-06-27 06:36:03 -04:00
# Get the state of the events we know about
2019-10-23 12:25:54 -04:00
ours = yield self.state_store.get_state_groups_ids(room_id, seen)
# state_maps is a list of mappings from (type, state_key) to event_id
state_maps = list(
ours.values()
) # type: list[dict[tuple[str, str], str]]
# we don't need this any more, let's delete it.
del ours
2018-06-27 06:36:03 -04:00
# Ask the remote server for the states we don't
# know about
for p in prevs - seen:
logger.info(
"[%s %s] Requesting state at missing prev_event %s",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
p,
)
room_version = yield self.store.get_room_version(room_id)
with nested_logging_context(p):
2018-09-26 14:17:36 -04:00
# note that if any of the missing prevs share missing state or
# auth events, the requests to fetch those events are deduped
# by the get_pdu_cache in federation_client.
(
remote_state,
got_auth_chain,
) = yield self.federation_client.get_state_for_room(
origin, room_id, p
2018-06-27 06:40:27 -04:00
)
# we want the state *after* p; get_state_for_room returns the
# state *before* p.
remote_event = yield self.federation_client.get_pdu(
2019-06-20 05:32:02 -04:00
[origin], p, room_version, outlier=True
)
if remote_event is None:
raise Exception(
2019-06-20 05:32:02 -04:00
"Unable to get missing prev_event %s" % (p,)
)
if remote_event.is_state():
remote_state.append(remote_event)
# XXX hrm I'm not convinced that duplicate events will compare
# for equality, so I'm not sure this does what the author
# hoped.
auth_chains.update(got_auth_chain)
remote_state_map = {
(x.type, x.state_key): x.event_id for x in remote_state
}
state_maps.append(remote_state_map)
for x in remote_state:
event_map[x.event_id] = x
2018-10-24 04:44:22 -04:00
state_map = yield resolve_events_with_store(
2019-06-20 05:32:02 -04:00
room_version,
state_maps,
event_map,
state_res_store=StateResolutionStore(self.store),
2018-06-27 06:36:03 -04:00
)
2018-10-24 04:47:49 -04:00
# We need to give _process_received_pdu the actual state events
# rather than event ids, so generate that now.
2018-10-24 04:47:49 -04:00
# First though we need to fetch all the events that are in
# state_map, so we can build up the state below.
evs = yield self.store.get_events(
list(state_map.values()),
get_prev_content=False,
check_redacted=False,
2018-06-27 06:36:03 -04:00
)
event_map.update(evs)
2019-06-20 05:32:02 -04:00
state = [event_map[e] for e in six.itervalues(state_map)]
2018-06-27 06:36:03 -04:00
auth_chain = list(auth_chains)
except Exception:
logger.warning(
"[%s %s] Error attempting to resolve state at missing "
"prev_events",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
exc_info=True,
)
raise FederationError(
"ERROR",
403,
"We can't get valid state history.",
affected=event_id,
)
yield self._process_received_pdu(
2019-06-20 05:32:02 -04:00
origin, pdu, state=state, auth_chain=auth_chain
)
@defer.inlineCallbacks
def _get_missing_events_for_pdu(self, origin, pdu, prevs, min_depth):
"""
Args:
origin (str): Origin of the pdu. Will be called to get the missing events
pdu: received pdu
2017-04-28 07:46:53 -04:00
prevs (set(str)): List of event ids which we are missing
min_depth (int): Minimum depth of events to return.
"""
room_id = pdu.room_id
event_id = pdu.event_id
seen = yield self.store.have_seen_events(prevs)
if not prevs - seen:
2017-04-28 07:46:53 -04:00
return
latest = yield self.store.get_latest_event_ids_in_room(room_id)
# We add the prev events that we have seen to the latest
# list to ensure the remote server doesn't give them to us
latest = set(latest)
latest |= seen
logger.info(
"[%s %s]: Requesting missing events between %s and %s",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
shortstr(latest),
event_id,
)
# XXX: we set timeout to 10s to help workaround
# https://github.com/matrix-org/synapse/issues/1733.
# The reason is to avoid holding the linearizer lock
# whilst processing inbound /send transactions, causing
# FDs to stack up and block other inbound transactions
# which empirically can currently take up to 30 minutes.
#
# N.B. this explicitly disables retry attempts.
#
# N.B. this also increases our chances of falling back to
# fetching fresh state for the room if the missing event
# can't be found, which slightly reduces our security.
# it may also increase our DAG extremity count for the room,
# causing additional state resolution? See #1760.
# However, fetching state doesn't hold the linearizer lock
# apparently.
#
# see https://github.com/matrix-org/synapse/pull/1744
#
# ----
#
# Update richvdh 2018/09/18: There are a number of problems with timing this
# request out agressively on the client side:
#
# - it plays badly with the server-side rate-limiter, which starts tarpitting you
# if you send too many requests at once, so you end up with the server carefully
# working through the backlog of your requests, which you have already timed
# out.
#
# - for this request in particular, we now (as of
# https://github.com/matrix-org/synapse/pull/3456) reject any PDUs where the
# server can't produce a plausible-looking set of prev_events - so we becone
# much more likely to reject the event.
#
# - contrary to what it says above, we do *not* fall back to fetching fresh state
# for the room if get_missing_events times out. Rather, we give up processing
# the PDU whose prevs we are missing, which then makes it much more likely that
# we'll end up back here for the *next* PDU in the list, which exacerbates the
# problem.
#
# - the agressive 10s timeout was introduced to deal with incoming federation
# requests taking 8 hours to process. It's not entirely clear why that was going
# on; certainly there were other issues causing traffic storms which are now
# resolved, and I think in any case we may be more sensible about our locking
# now. We're *certainly* more sensible about our logging.
#
# All that said: Let's try increasing the timout to 60s and see what happens.
try:
missing_events = yield self.federation_client.get_missing_events(
origin,
room_id,
earliest_events_ids=list(latest),
latest_events=[pdu],
limit=10,
min_depth=min_depth,
timeout=60000,
)
except RequestSendFailed as e:
# We failed to get the missing events, but since we need to handle
# the case of `get_missing_events` not returning the necessary
# events anyway, it is safe to simply log the error and continue.
logger.warning(
"[%s %s]: Failed to get prev_events: %s", room_id, event_id, e
)
return
logger.info(
"[%s %s]: Got %d prev_events: %s",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
len(missing_events),
shortstr(missing_events),
)
# We want to sort these by depth so we process them and
# tell clients about them in order.
missing_events.sort(key=lambda x: x.depth)
for ev in missing_events:
logger.info(
"[%s %s] Handling received prev_event %s",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
ev.event_id,
)
with nested_logging_context(ev.event_id):
try:
2019-06-20 05:32:02 -04:00
yield self.on_receive_pdu(origin, ev, sent_to_us_directly=False)
except FederationError as e:
if e.code == 403:
logger.warning(
"[%s %s] Received prev_event %s failed history check.",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
ev.event_id,
)
else:
raise
@defer.inlineCallbacks
def _process_received_pdu(self, origin, event, state, auth_chain):
""" Called when we have a new pdu. We need to do auth checks and put it
through the StateHandler.
"""
room_id = event.room_id
event_id = event.event_id
2019-06-20 05:32:02 -04:00
logger.debug("[%s %s] Processing event: %s", room_id, event_id, event)
2018-11-09 09:58:09 -05:00
event_ids = set()
if state:
event_ids |= {e.event_id for e in state}
if auth_chain:
event_ids |= {e.event_id for e in auth_chain}
seen_ids = yield self.store.have_seen_events(event_ids)
if state and auth_chain is not None:
# If we have any state or auth_chain given to us by the replication
# layer, then we should handle them (if we haven't before.)
event_infos = []
for e in itertools.chain(auth_chain, state):
if e.event_id in seen_ids:
continue
e.internal_metadata.outlier = True
auth_ids = e.auth_event_ids()
auth = {
2019-06-20 05:32:02 -04:00
(e.type, e.state_key): e
for e in auth_chain
2018-11-09 09:58:09 -05:00
if e.event_id in auth_ids or e.type == EventTypes.Create
}
2019-06-20 05:32:02 -04:00
event_infos.append({"event": e, "auth_events": auth})
2018-11-09 09:58:09 -05:00
seen_ids.add(e.event_id)
logger.info(
2018-11-09 09:58:09 -05:00
"[%s %s] persisting newly-received auth/state events %s",
2019-06-20 05:32:02 -04:00
room_id,
event_id,
[e["event"].event_id for e in event_infos],
)
2018-11-09 09:58:09 -05:00
yield self._handle_new_events(origin, event_infos)
2018-11-09 09:58:09 -05:00
try:
2019-06-20 05:32:02 -04:00
context = yield self._handle_new_event(origin, event, state=state)
2018-11-09 09:58:09 -05:00
except AuthError as e:
2019-06-20 05:32:02 -04:00
raise FederationError("ERROR", e.code, e.msg, affected=event.event_id)
2014-12-16 10:24:03 -05:00
room = yield self.store.get_room(room_id)
2014-08-12 10:10:52 -04:00
if not room:
try:
yield self.store.store_room(
2019-06-20 05:32:02 -04:00
room_id=room_id, room_creator_user_id="", is_public=False
)
except StoreError:
logger.exception("Failed to store room.")
2014-08-19 09:20:03 -04:00
2014-12-16 06:29:05 -05:00
if event.type == EventTypes.Member:
if event.membership == Membership.JOIN:
# Only fire user_joined_room if the user has acutally
# joined the room. Don't bother if the user is just
# changing their profile info.
newly_joined = True
2018-07-23 08:00:22 -04:00
prev_state_ids = yield context.get_prev_state_ids(self.store)
2019-06-20 05:32:02 -04:00
prev_state_id = prev_state_ids.get((event.type, event.state_key))
if prev_state_id:
prev_state = yield self.store.get_event(
2019-06-20 05:32:02 -04:00
prev_state_id, allow_none=True
)
if prev_state and prev_state.membership == Membership.JOIN:
newly_joined = False
if newly_joined:
user = UserID.from_string(event.state_key)
yield self.user_joined_room(user, room_id)
2014-08-19 09:20:03 -04:00
@log_function
@defer.inlineCallbacks
def backfill(self, dest, room_id, limit, extremities):
""" Trigger a backfill request to `dest` for the given `room_id`
2016-04-12 07:04:19 -04:00
2018-04-17 18:41:36 -04:00
This will attempt to get more events from the remote. If the other side
has no new events to offer, this will return an empty list.
As the events are received, we check their signatures, and also do some
sanity-checking on them. If any of the backfilled events are invalid,
this method throws a SynapseError.
TODO: make this more useful to distinguish failures of the remote
server from invalid events (there is probably no point in trying to
re-fetch invalid events from every other HS in the room.)
"""
2016-04-05 07:56:29 -04:00
if dest == self.server_name:
raise SynapseError(400, "Can't backfill from self.")
room_version = yield self.store.get_room_version(room_id)
events = yield self.federation_client.backfill(
2019-06-20 05:32:02 -04:00
dest, room_id, limit=limit, extremities=extremities
)
2018-04-20 06:41:03 -04:00
# ideally we'd sanity check the events here for excess prev_events etc,
# but it's hard to reject events at this point without completely
# breaking backfill in the same way that it is currently broken by
# events whose signature we cannot verify (#3121).
#
# So for now we accept the events anyway. #3124 tracks this.
#
# for ev in events:
# self._sanity_check_event(ev)
2018-04-17 18:41:12 -04:00
2016-04-12 06:54:41 -04:00
# Don't bother processing events we already have.
seen_events = yield self.store.have_events_in_timeline(
set(e.event_id for e in events)
)
events = [e for e in events if e.event_id not in seen_events]
if not events:
return []
event_map = {e.event_id: e for e in events}
event_ids = set(e.event_id for e in events)
2019-06-20 05:32:02 -04:00
edges = [ev.event_id for ev in events if set(ev.prev_event_ids()) - event_ids]
2019-06-20 05:32:02 -04:00
logger.info("backfill: Got %d events with %d edges", len(events), len(edges))
2015-06-02 05:28:14 -04:00
# For each edge get the current state.
auth_events = {}
state_events = {}
events_to_state = {}
for e_id in edges:
state, auth = yield self.federation_client.get_state_for_room(
2019-06-20 05:32:02 -04:00
destination=dest, room_id=room_id, event_id=e_id
)
auth_events.update({a.event_id: a for a in auth})
auth_events.update({s.event_id: s for s in state})
state_events.update({s.event_id: s for s in state})
events_to_state[e_id] = state
required_auth = set(
2016-08-05 07:59:04 -04:00
a_id
2019-06-20 05:32:02 -04:00
for event in events
+ list(state_events.values())
+ list(auth_events.values())
for a_id in event.auth_event_ids()
)
2019-06-20 05:32:02 -04:00
auth_events.update(
{e_id: event_map[e_id] for e_id in required_auth if e_id in event_map}
)
missing_auth = required_auth - set(auth_events)
2016-08-05 07:59:04 -04:00
failed_to_fetch = set()
# Try and fetch any missing auth events from both DB and remote servers.
# We repeatedly do this until we stop finding new auth events.
while missing_auth - failed_to_fetch:
logger.info("Missing auth for backfill: %r", missing_auth)
2016-08-05 07:59:04 -04:00
ret_events = yield self.store.get_events(missing_auth - failed_to_fetch)
auth_events.update(ret_events)
required_auth.update(
a_id for event in ret_events.values() for a_id in event.auth_event_ids()
2016-08-05 07:59:04 -04:00
)
missing_auth = required_auth - set(auth_events)
if missing_auth - failed_to_fetch:
logger.info(
"Fetching missing auth for backfill: %r",
2019-06-20 05:32:02 -04:00
missing_auth - failed_to_fetch,
2016-08-05 07:59:04 -04:00
)
results = yield make_deferred_yieldable(
2019-06-20 05:32:02 -04:00
defer.gatherResults(
[
run_in_background(
2019-06-20 05:32:02 -04:00
self.federation_client.get_pdu,
[dest],
event_id,
room_version=room_version,
outlier=True,
timeout=10000,
)
for event_id in missing_auth - failed_to_fetch
],
consumeErrors=True,
)
).addErrback(unwrapFirstError)
2016-08-24 05:31:05 -04:00
auth_events.update({a.event_id: a for a in results if a})
2016-08-05 07:59:04 -04:00
required_auth.update(
2016-08-25 05:30:09 -04:00
a_id
2019-06-20 05:32:02 -04:00
for event in results
if event
for a_id in event.auth_event_ids()
2016-08-05 07:59:04 -04:00
)
missing_auth = required_auth - set(auth_events)
failed_to_fetch = missing_auth - set(auth_events)
seen_events = yield self.store.have_seen_events(
2016-08-05 07:59:04 -04:00
set(auth_events.keys()) | set(state_events.keys())
)
2019-02-25 05:02:12 -05:00
# We now have a chunk of events plus associated state and auth chain to
# persist. We do the persistence in two steps:
# 1. Auth events and state get persisted as outliers, plus the
# backward extremities get persisted (as non-outliers).
# 2. The rest of the events in the chunk get persisted one by one, as
# each one depends on the previous event for its state.
#
# The important thing is that events in the chunk get persisted as
# non-outliers, including when those events are also in the state or
# auth chain. Caution must therefore be taken to ensure that they are
# not accidentally marked as outliers.
2019-02-25 09:45:02 -05:00
# Step 1a: persist auth events that *don't* appear in the chunk
2015-06-25 12:18:19 -04:00
ev_infos = []
for a in auth_events.values():
# We only want to persist auth events as outliers that we haven't
# seen and aren't about to persist as part of the backfilled chunk.
if a.event_id in seen_events or a.event_id in event_map:
2015-06-25 12:18:19 -04:00
continue
a.internal_metadata.outlier = True
2019-06-20 05:32:02 -04:00
ev_infos.append(
{
"event": a,
"auth_events": {
(
auth_events[a_id].type,
auth_events[a_id].state_key,
): auth_events[a_id]
for a_id in a.auth_event_ids()
if a_id in auth_events
},
2015-06-25 12:18:19 -04:00
}
2019-06-20 05:32:02 -04:00
)
2015-06-25 12:18:19 -04:00
2019-02-25 09:45:02 -05:00
# Step 1b: persist the events in the chunk we fetched state for (i.e.
# the backwards extremities) as non-outliers.
2015-06-25 12:18:19 -04:00
for e_id in events_to_state:
2019-02-25 05:02:12 -05:00
# For paranoia we ensure that these events are marked as
# non-outliers
ev = event_map[e_id]
2019-06-20 05:32:02 -04:00
assert not ev.internal_metadata.is_outlier()
ev_infos.append(
{
"event": ev,
"state": events_to_state[e_id],
"auth_events": {
(
auth_events[a_id].type,
auth_events[a_id].state_key,
): auth_events[a_id]
for a_id in ev.auth_event_ids()
if a_id in auth_events
},
2015-06-25 12:18:19 -04:00
}
2019-06-20 05:32:02 -04:00
)
2019-06-20 05:32:02 -04:00
yield self._handle_new_events(dest, ev_infos, backfilled=True)
2019-02-25 09:45:02 -05:00
# Step 2: Persist the rest of the events in the chunk one by one
events.sort(key=lambda e: e.depth)
for event in events:
if event in events_to_state:
continue
2019-02-25 05:02:12 -05:00
# For paranoia we ensure that these events are marked as
# non-outliers
2019-06-20 05:32:02 -04:00
assert not event.internal_metadata.is_outlier()
2019-02-25 05:02:12 -05:00
2016-04-12 07:48:30 -04:00
# We store these one at a time since each event depends on the
# previous to work out the state.
# TODO: We can probably do something more clever here.
2019-06-20 05:32:02 -04:00
yield self._handle_new_event(dest, event, backfilled=True)
return events
@defer.inlineCallbacks
def maybe_backfill(self, room_id, current_depth):
"""Checks the database to see if we should backfill before paginating,
and if so do.
"""
2019-06-20 05:32:02 -04:00
extremities = yield self.store.get_oldest_events_with_depth_in_room(room_id)
if not extremities:
logger.debug("Not backfilling as no extremeties found.")
return
# We only want to paginate if we can actually see the events we'll get,
# as otherwise we'll just spend a lot of resources to get redacted
# events.
#
2019-03-05 04:16:35 -05:00
# We do this by filtering all the backwards extremities and seeing if
# any remain. Given we don't have the extremity events themselves, we
# need to actually check the events that reference them.
2019-02-27 08:06:10 -05:00
#
# *Note*: the spec wants us to keep backfilling until we reach the start
# of the room in case we are allowed to see some of the history. However
# in practice that causes more issues than its worth, as a) its
# relatively rare for there to be any visible history and b) even when
# there is its often sufficiently long ago that clients would stop
# attempting to paginate before backfill reached the visible history.
#
2019-03-05 04:16:35 -05:00
# TODO: If we do do a backfill then we should filter the backwards
# extremities to only include those that point to visible portions of
# history.
#
# TODO: Correctly handle the case where we are allowed to see the
2019-03-05 04:16:35 -05:00
# forward event but not the backward extremity, e.g. in the case of
# initial join of the server where we are allowed to see the join
# event but not anything before it. This would require looking at the
# state *before* the event, ignoring the special casing certain event
# types have.
2019-06-20 05:32:02 -04:00
forward_events = yield self.store.get_successor_events(list(extremities))
extremities_events = yield self.store.get_events(
2019-06-20 05:32:02 -04:00
forward_events, check_redacted=False, get_prev_content=False
)
# We set `check_history_visibility_only` as we might otherwise get false
# positives from users having been erased.
filtered_extremities = yield filter_events_for_server(
2019-10-23 12:25:54 -04:00
self.storage,
2019-06-20 05:32:02 -04:00
self.server_name,
list(extremities_events.values()),
redact=False,
check_history_visibility_only=True,
)
if not filtered_extremities:
return False
# Check if we reached a point where we should start backfilling.
2019-06-20 05:32:02 -04:00
sorted_extremeties_tuple = sorted(extremities.items(), key=lambda e: -int(e[1]))
max_depth = sorted_extremeties_tuple[0][1]
# We don't want to specify too many extremities as it causes the backfill
# request URI to be too long.
extremities = dict(sorted_extremeties_tuple[:5])
if current_depth > max_depth:
logger.debug(
2019-06-20 05:32:02 -04:00
"Not backfilling as we don't need to. %d < %d", max_depth, current_depth
)
return
# Now we need to decide which hosts to hit first.
# First we try hosts that are already in the room
# TODO: HEURISTIC ALERT.
curr_state = yield self.state_handler.get_current_state(room_id)
def get_domains_from_state(state):
"""Get joined domains from state
Args:
state (dict[tuple, FrozenEvent]): State map from type/state
key to event.
Returns:
list[tuple[str, int]]: Returns a list of servers with the
lowest depth of their joins. Sorted by lowest depth first.
"""
joined_users = [
(state_key, int(event.depth))
for (e_type, state_key), event in iteritems(state)
2019-06-20 05:32:02 -04:00
if e_type == EventTypes.Member and event.membership == Membership.JOIN
]
joined_domains = {}
for u, d in joined_users:
try:
dom = get_domain_from_id(u)
old_d = joined_domains.get(dom)
if old_d:
joined_domains[dom] = min(d, old_d)
else:
joined_domains[dom] = d
except Exception:
pass
return sorted(joined_domains.items(), key=lambda d: d[1])
curr_domains = get_domains_from_state(curr_state)
likely_domains = [
2019-06-20 05:32:02 -04:00
domain for domain, depth in curr_domains if domain != self.server_name
]
@defer.inlineCallbacks
def try_backfill(domains):
# TODO: Should we try multiple of these at a time?
for dom in domains:
try:
yield self.backfill(
2019-06-20 05:32:02 -04:00
dom, room_id, limit=100, extremities=extremities
)
2016-04-12 07:04:19 -04:00
# If this succeeded then we probably already have the
# appropriate stuff.
2016-04-12 07:48:30 -04:00
# TODO: We can probably do something more intelligent here.
return True
except SynapseError as e:
2019-06-20 05:32:02 -04:00
logger.info("Failed to backfill from %s because %s", dom, e)
continue
except CodeMessageException as e:
if 400 <= e.code < 500:
raise
2019-06-20 05:32:02 -04:00
logger.info("Failed to backfill from %s because %s", dom, e)
continue
except NotRetryingDestination as e:
2018-09-06 10:22:23 -04:00
logger.info(str(e))
continue
except RequestSendFailed as e:
logger.info("Falied to get backfill from %s because %s", dom, e)
continue
except FederationDeniedError as e:
logger.info(e)
continue
except Exception as e:
2019-06-20 05:32:02 -04:00
logger.exception("Failed to backfill from %s because %s", dom, e)
continue
return False
success = yield try_backfill(likely_domains)
if success:
return True
# Huh, well *those* domains didn't work out. Lets try some domains
# from the time.
tried_domains = set(likely_domains)
tried_domains.add(self.server_name)
event_ids = list(extremities.keys())
logger.debug("calling resolve_state_groups in _maybe_backfill")
resolve = preserve_fn(self.state_handler.resolve_state_groups_for_events)
states = yield make_deferred_yieldable(
2019-06-20 05:32:02 -04:00
defer.gatherResults(
[resolve(room_id, [e]) for e in event_ids], consumeErrors=True
)
)
2018-05-22 14:00:48 -04:00
# dict[str, dict[tuple, str]], a map from event_id to state map of
# event_ids.
2017-01-13 08:21:04 -05:00
states = dict(zip(event_ids, [s.state for s in states]))
state_map = yield self.store.get_events(
[e_id for ids in itervalues(states) for e_id in itervalues(ids)],
2019-06-20 05:32:02 -04:00
get_prev_content=False,
)
states = {
key: {
k: state_map[e_id]
for k, e_id in iteritems(state_dict)
if e_id in state_map
2019-06-20 05:32:02 -04:00
}
for key, state_dict in iteritems(states)
}
for e_id, _ in sorted_extremeties_tuple:
2015-05-12 09:02:01 -04:00
likely_domains = get_domains_from_state(states[e_id])
2019-06-20 05:32:02 -04:00
success = yield try_backfill(
[dom for dom, _ in likely_domains if dom not in tried_domains]
)
if success:
return True
tried_domains.update(dom for dom, _ in likely_domains)
return False
2018-04-17 18:41:12 -04:00
def _sanity_check_event(self, ev):
"""
Do some early sanity checks of a received event
In particular, checks it doesn't have an excessive number of
prev_events or auth_events, which could cause a huge state resolution
or cascade of event fetches.
Args:
ev (synapse.events.EventBase): event to be checked
Returns: None
Raises:
SynapseError if the event does not pass muster
"""
if len(ev.prev_event_ids()) > 20:
logger.warning(
2019-06-20 05:32:02 -04:00
"Rejecting event %s which has %i prev_events",
ev.event_id,
len(ev.prev_event_ids()),
2018-04-17 18:41:12 -04:00
)
2019-06-20 05:32:02 -04:00
raise SynapseError(http_client.BAD_REQUEST, "Too many prev_events")
2018-04-17 18:41:12 -04:00
if len(ev.auth_event_ids()) > 10:
logger.warning(
2019-06-20 05:32:02 -04:00
"Rejecting event %s which has %i auth_events",
ev.event_id,
len(ev.auth_event_ids()),
2018-04-17 18:41:12 -04:00
)
2019-06-20 05:32:02 -04:00
raise SynapseError(http_client.BAD_REQUEST, "Too many auth_events")
2018-04-17 18:41:12 -04:00
@defer.inlineCallbacks
def send_invite(self, target_host, event):
""" Sends the invite to the remote server for signing.
Invites must be signed by the invitee's server before distribution.
"""
pdu = yield self.federation_client.send_invite(
destination=target_host,
room_id=event.room_id,
event_id=event.event_id,
2019-06-20 05:32:02 -04:00
pdu=event,
)
return pdu
2014-11-07 10:35:53 -05:00
@defer.inlineCallbacks
def on_event_auth(self, event_id):
2017-05-24 09:22:41 -04:00
event = yield self.store.get_event(event_id)
auth = yield self.store.get_auth_chain(
2019-06-20 05:32:02 -04:00
[auth_id for auth_id in event.auth_event_ids()], include_given=True
2017-05-24 09:22:41 -04:00
)
return [e for e in auth]
2014-11-07 10:35:53 -05:00
@log_function
@defer.inlineCallbacks
def do_invite_join(self, target_hosts, room_id, joinee, content):
""" Attempts to join the `joinee` to the room `room_id` via the
servers contained in `target_hosts`.
This first triggers a /make_join/ request that returns a partial
event that we can fill out and sign. This is then sent to the
remote server via /send_join/ which responds with the state at that
event and the auth_chains.
We suspend processing of any received events from this room until we
have finished processing the join.
Args:
target_hosts (Iterable[str]): List of servers to attempt to join the room with.
room_id (str): The ID of the room to join.
joinee (str): The User ID of the joining user.
content (dict): The event content to use for the join event.
"""
logger.debug("Joining %s to %s", joinee, room_id)
origin, event, event_format_version = yield self._make_and_verify_event(
target_hosts,
room_id,
joinee,
"join",
content,
2019-06-20 05:32:02 -04:00
params={"ver": KNOWN_ROOM_VERSIONS},
)
# This shouldn't happen, because the RoomMemberHandler has a
# linearizer lock which only allows one operation per user per room
# at a time - so this is just paranoia.
2019-06-20 05:32:02 -04:00
assert room_id not in self.room_queues
self.room_queues[room_id] = []
yield self._clean_room_for_join(room_id)
handled_events = set()
try:
# Try the host we successfully got a response to /make_join/
# request first.
try:
target_hosts.remove(origin)
target_hosts.insert(0, origin)
except ValueError:
pass
ret = yield self.federation_client.send_join(
2019-06-20 05:32:02 -04:00
target_hosts, event, event_format_version
)
origin = ret["origin"]
state = ret["state"]
auth_chain = ret["auth_chain"]
auth_chain.sort(key=lambda e: e.depth)
handled_events.update([s.event_id for s in state])
handled_events.update([a.event_id for a in auth_chain])
handled_events.add(event.event_id)
logger.debug("do_invite_join auth_chain: %s", auth_chain)
logger.debug("do_invite_join state: %s", state)
logger.debug("do_invite_join event: %s", event)
try:
yield self.store.store_room(
2019-06-20 05:32:02 -04:00
room_id=room_id, room_creator_user_id="", is_public=False
)
except Exception:
# FIXME
pass
2019-06-20 05:32:02 -04:00
yield self._persist_auth_tree(origin, auth_chain, state, event)
# Check whether this room is the result of an upgrade of a room we already know
# about. If so, migrate over user information
predecessor = yield self.store.get_room_predecessor(room_id)
if not predecessor:
return
old_room_id = predecessor["room_id"]
logger.debug(
"Found predecessor for %s during remote join: %s", room_id, old_room_id
)
# We retrieve the room member handler here as to not cause a cyclic dependency
member_handler = self.hs.get_room_member_handler()
yield member_handler.transfer_room_state_on_room_upgrade(
old_room_id, room_id
)
logger.debug("Finished joining %s to %s", joinee, room_id)
finally:
room_queue = self.room_queues[room_id]
del self.room_queues[room_id]
# we don't need to wait for the queued events to be processed -
# it's just a best-effort thing at this point. We do want to do
# them roughly in order, though, otherwise we'll end up making
# lots of requests for missing prev_events which we do actually
# have. Hence we fire off the deferred, but don't wait for it.
run_in_background(self._handle_queued_pdus, room_queue)
return True
@defer.inlineCallbacks
def _handle_queued_pdus(self, room_queue):
"""Process PDUs which got queued up while we were busy send_joining.
Args:
room_queue (list[FrozenEvent, str]): list of PDUs to be processed
and the servers that sent them
"""
for p, origin in room_queue:
try:
2019-06-20 05:32:02 -04:00
logger.info(
"Processing queued PDU %s which was received "
"while we were joining %s",
p.event_id,
p.room_id,
)
with nested_logging_context(p.event_id):
yield self.on_receive_pdu(origin, p, sent_to_us_directly=True)
except Exception as e:
logger.warning(
2019-06-20 05:32:02 -04:00
"Error handling queued PDU %s from %s: %s", p.event_id, origin, e
)
@defer.inlineCallbacks
@log_function
def on_make_join_request(self, origin, room_id, user_id):
""" We've received a /make_join/ request, so we create a partial
join event for the room and return that. We do *not* persist or
process it until the other server has signed it and sent it back.
Args:
origin (str): The (verified) server name of the requesting server.
room_id (str): Room to create join event in
user_id (str): The user to create the join for
Returns:
Deferred[FrozenEvent]
"""
if get_domain_from_id(user_id) != origin:
logger.info(
"Got /make_join request for user %r from different origin %s, ignoring",
user_id,
origin,
)
raise SynapseError(403, "User not from origin", Codes.FORBIDDEN)
event_content = {"membership": Membership.JOIN}
room_version = yield self.store.get_room_version(room_id)
builder = self.event_builder_factory.new(
room_version,
{
"type": EventTypes.Member,
"content": event_content,
"room_id": room_id,
"sender": user_id,
"state_key": user_id,
2019-06-20 05:32:02 -04:00
},
)
2016-04-13 06:11:46 -04:00
try:
event, context = yield self.event_creation_handler.create_new_client_event(
2019-06-20 05:32:02 -04:00
builder=builder
2016-04-13 06:11:46 -04:00
)
except AuthError as e:
logger.warning("Failed to create join to %s because %s", room_id, e)
2016-04-13 06:11:46 -04:00
raise e
event_allowed = yield self.third_party_event_rules.check_event_allowed(
2019-06-20 05:32:02 -04:00
event, context
)
if not event_allowed:
logger.info("Creation of join %s forbidden by third-party rules", event)
raise SynapseError(
2019-06-20 05:32:02 -04:00
403, "This event is not allowed in this context", Codes.FORBIDDEN
)
2016-07-15 04:29:54 -04:00
# The remote hasn't signed it yet, obviously. We'll do the full checks
# when we get the event back in `on_send_join_request`
yield self.auth.check_from_context(
2019-06-20 05:32:02 -04:00
room_version, event, context, do_sig_check=False
)
return event
@defer.inlineCallbacks
@log_function
def on_send_join_request(self, origin, pdu):
""" We have received a join event for a room. Fully process it and
respond with the current state and auth chains.
"""
2014-11-14 16:25:02 -05:00
event = pdu
logger.debug(
"on_send_join_request from %s: Got event: %s, signatures: %s",
origin,
event.event_id,
event.signatures,
)
if get_domain_from_id(event.sender) != origin:
logger.info(
"Got /send_join request for user %r from different origin %s",
event.sender,
origin,
)
raise SynapseError(403, "User not from origin", Codes.FORBIDDEN)
2014-12-08 04:08:26 -05:00
event.internal_metadata.outlier = False
# Send this event on behalf of the origin server.
#
# The reasons we have the destination server rather than the origin
# server send it are slightly mysterious: the origin server should have
# all the neccessary state once it gets the response to the send_join,
# so it could send the event itself if it wanted to. It may be that
# doing it this way reduces failure modes, or avoids certain attacks
# where a new server selectively tells a subset of the federation that
# it has joined.
#
# The fact is that, as of the current writing, Synapse doesn't send out
# the join event over federation after joining, and changing it now
# would introduce the danger of backwards-compatibility problems.
event.internal_metadata.send_on_behalf_of = origin
2019-06-20 05:32:02 -04:00
context = yield self._handle_new_event(origin, event)
event_allowed = yield self.third_party_event_rules.check_event_allowed(
2019-06-20 05:32:02 -04:00
event, context
)
if not event_allowed:
logger.info("Sending of join %s forbidden by third-party rules", event)
raise SynapseError(
2019-06-20 05:32:02 -04:00
403, "This event is not allowed in this context", Codes.FORBIDDEN
)
logger.debug(
"on_send_join_request: After _handle_new_event: %s, sigs: %s",
event.event_id,
event.signatures,
)
2014-12-16 06:29:05 -05:00
if event.type == EventTypes.Member:
if event.content["membership"] == Membership.JOIN:
user = UserID.from_string(event.state_key)
yield self.user_joined_room(user, event.room_id)
2018-07-23 08:00:22 -04:00
prev_state_ids = yield context.get_prev_state_ids(self.store)
state_ids = list(prev_state_ids.values())
2017-05-24 09:22:41 -04:00
auth_chain = yield self.store.get_auth_chain(state_ids)
2018-07-23 08:00:22 -04:00
state = yield self.store.get_events(list(prev_state_ids.values()))
return {"state": list(state.values()), "auth_chain": auth_chain}
@defer.inlineCallbacks
def on_invite_request(self, origin, pdu):
""" We've got an invite event. Process and persist it. Sign it.
Respond with the now signed event.
"""
2014-11-14 16:25:02 -05:00
event = pdu
if event.state_key is None:
raise SynapseError(400, "The invite event did not have a state key")
2017-06-19 07:36:28 -04:00
is_blocked = yield self.store.is_room_blocked(event.room_id)
if is_blocked:
raise SynapseError(403, "This room has been blocked on this server")
if self.hs.config.block_non_admin_invites:
raise SynapseError(403, "This server does not accept room invites")
if not self.spam_checker.user_may_invite(
2019-06-20 05:32:02 -04:00
event.sender, event.state_key, event.room_id
):
2017-10-03 09:04:10 -04:00
raise SynapseError(
403, "This user is not permitted to send invites to this server/user"
2017-10-03 09:04:10 -04:00
)
membership = event.content.get("membership")
if event.type != EventTypes.Member or membership != Membership.INVITE:
raise SynapseError(400, "The event was not an m.room.member invite event")
sender_domain = get_domain_from_id(event.sender)
if sender_domain != origin:
2019-06-20 05:32:02 -04:00
raise SynapseError(
400, "The invite event was not from the server sending it"
)
if not self.is_mine_id(event.state_key):
raise SynapseError(400, "The invite event must be for this server")
# block any attempts to invite the server notices mxid
if event.state_key == self._server_notices_mxid:
2019-06-20 05:32:02 -04:00
raise SynapseError(http_client.FORBIDDEN, "Cannot invite this user")
event.internal_metadata.outlier = True
event.internal_metadata.out_of_band_membership = True
event.signatures.update(
compute_event_signature(
2019-06-20 05:32:02 -04:00
event.get_pdu_json(), self.hs.hostname, self.hs.config.signing_key[0]
)
)
context = yield self.state_handler.compute_event_context(event)
yield self.persist_events_and_notify([(event, context)])
return event
@defer.inlineCallbacks
def do_remotely_reject_invite(self, target_hosts, room_id, user_id):
origin, event, event_format_version = yield self._make_and_verify_event(
2019-06-20 05:32:02 -04:00
target_hosts, room_id, user_id, "leave"
)
2017-06-09 08:05:05 -04:00
# Mark as outlier as we don't have any state for this event; we're not
# even in the room.
event.internal_metadata.outlier = True
event.internal_metadata.out_of_band_membership = True
# Try the host that we succesfully called /make_leave/ on first for
# the /send_leave/ request.
try:
target_hosts.remove(origin)
target_hosts.insert(0, origin)
except ValueError:
pass
2019-06-20 05:32:02 -04:00
yield self.federation_client.send_leave(target_hosts, event)
context = yield self.state_handler.compute_event_context(event)
yield self.persist_events_and_notify([(event, context)])
return event
@defer.inlineCallbacks
2019-06-20 05:32:02 -04:00
def _make_and_verify_event(
self, target_hosts, room_id, user_id, membership, content={}, params=None
):
origin, event, format_ver = yield self.federation_client.make_membership_event(
2019-06-20 05:32:02 -04:00
target_hosts, room_id, user_id, membership, content, params=params
)
logger.debug("Got response to make_%s: %s", membership, event)
# We should assert some things.
# FIXME: Do this in a nicer way
2019-06-20 05:32:02 -04:00
assert event.type == EventTypes.Member
assert event.user_id == user_id
assert event.state_key == user_id
assert event.room_id == room_id
return origin, event, format_ver
@defer.inlineCallbacks
@log_function
def on_make_leave_request(self, origin, room_id, user_id):
""" We've received a /make_leave/ request, so we create a partial
leave event for the room and return that. We do *not* persist or
process it until the other server has signed it and sent it back.
Args:
origin (str): The (verified) server name of the requesting server.
room_id (str): Room to create leave event in
user_id (str): The user to create the leave for
Returns:
Deferred[FrozenEvent]
"""
if get_domain_from_id(user_id) != origin:
logger.info(
"Got /make_leave request for user %r from different origin %s, ignoring",
user_id,
origin,
)
raise SynapseError(403, "User not from origin", Codes.FORBIDDEN)
room_version = yield self.store.get_room_version(room_id)
builder = self.event_builder_factory.new(
room_version,
{
"type": EventTypes.Member,
"content": {"membership": Membership.LEAVE},
"room_id": room_id,
"sender": user_id,
"state_key": user_id,
2019-06-20 05:32:02 -04:00
},
)
event, context = yield self.event_creation_handler.create_new_client_event(
2019-06-20 05:32:02 -04:00
builder=builder
)
event_allowed = yield self.third_party_event_rules.check_event_allowed(
2019-06-20 05:32:02 -04:00
event, context
)
if not event_allowed:
logger.warning("Creation of leave %s forbidden by third-party rules", event)
raise SynapseError(
2019-06-20 05:32:02 -04:00
403, "This event is not allowed in this context", Codes.FORBIDDEN
)
2016-04-13 06:11:46 -04:00
try:
2016-07-15 04:29:54 -04:00
# The remote hasn't signed it yet, obviously. We'll do the full checks
# when we get the event back in `on_send_leave_request`
yield self.auth.check_from_context(
2019-06-20 05:32:02 -04:00
room_version, event, context, do_sig_check=False
)
2016-04-13 06:11:46 -04:00
except AuthError as e:
logger.warning("Failed to create new leave %r because %s", event, e)
2016-04-13 06:11:46 -04:00
raise e
return event
@defer.inlineCallbacks
@log_function
def on_send_leave_request(self, origin, pdu):
""" We have received a leave event for a room. Fully process it."""
event = pdu
logger.debug(
"on_send_leave_request: Got event: %s, signatures: %s",
event.event_id,
event.signatures,
)
if get_domain_from_id(event.sender) != origin:
logger.info(
"Got /send_leave request for user %r from different origin %s",
event.sender,
origin,
)
raise SynapseError(403, "User not from origin", Codes.FORBIDDEN)
event.internal_metadata.outlier = False
2019-06-20 05:32:02 -04:00
context = yield self._handle_new_event(origin, event)
event_allowed = yield self.third_party_event_rules.check_event_allowed(
2019-06-20 05:32:02 -04:00
event, context
)
if not event_allowed:
logger.info("Sending of leave %s forbidden by third-party rules", event)
raise SynapseError(
2019-06-20 05:32:02 -04:00
403, "This event is not allowed in this context", Codes.FORBIDDEN
)
logger.debug(
"on_send_leave_request: After _handle_new_event: %s, sigs: %s",
event.event_id,
event.signatures,
)
return None
@defer.inlineCallbacks
2016-07-21 05:30:12 -04:00
def get_state_for_pdu(self, room_id, event_id):
2016-09-02 09:19:22 -04:00
"""Returns the state at the event. i.e. not including said event.
"""
event = yield self.store.get_event(
2019-06-20 05:32:02 -04:00
event_id, allow_none=False, check_room_id=room_id
)
2019-10-23 12:25:54 -04:00
state_groups = yield self.state_store.get_state_groups(room_id, [event_id])
if state_groups:
_, state = list(iteritems(state_groups)).pop()
2019-06-20 05:32:02 -04:00
results = {(e.type, e.state_key): e for e in state}
if event.is_state():
# Get previous state
if "replaces_state" in event.unsigned:
prev_id = event.unsigned["replaces_state"]
if prev_id != event.event_id:
prev_event = yield self.store.get_event(prev_id)
results[(event.type, event.state_key)] = prev_event
else:
del results[(event.type, event.state_key)]
res = list(results.values())
return res
else:
return []
@defer.inlineCallbacks
def get_state_ids_for_pdu(self, room_id, event_id):
2016-09-02 09:19:22 -04:00
"""Returns the state at the event. i.e. not including said event.
"""
event = yield self.store.get_event(
2019-06-20 05:32:02 -04:00
event_id, allow_none=False, check_room_id=room_id
)
2019-10-23 12:25:54 -04:00
state_groups = yield self.state_store.get_state_groups_ids(room_id, [event_id])
if state_groups:
2018-09-06 10:22:23 -04:00
_, state = list(state_groups.items()).pop()
results = state
if event.is_state():
# Get previous state
if "replaces_state" in event.unsigned:
prev_id = event.unsigned["replaces_state"]
if prev_id != event.event_id:
results[(event.type, event.state_key)] = prev_id
else:
results.pop((event.type, event.state_key), None)
return list(results.values())
else:
return []
@defer.inlineCallbacks
@log_function
def on_backfill_request(self, origin, room_id, pdu_list, limit):
in_room = yield self.auth.check_host_in_room(room_id, origin)
if not in_room:
raise AuthError(403, "Host not in room.")
2019-06-20 05:32:02 -04:00
events = yield self.store.get_backfill_events(room_id, pdu_list, limit)
2019-10-23 12:25:54 -04:00
events = yield filter_events_for_server(self.storage, origin, events)
return events
2014-10-31 06:47:34 -04:00
@defer.inlineCallbacks
@log_function
def get_persisted_pdu(self, origin, event_id):
"""Get an event from the database for the given server.
Args:
origin [str]: hostname of server which is requesting the event; we
will check that the server is allowed to see it.
event_id [str]: id of the event being requested
2014-10-31 06:47:34 -04:00
Returns:
Deferred[EventBase|None]: None if we know nothing about the event;
otherwise the (possibly-redacted) event.
Raises:
AuthError if the server is not currently in the room
2014-10-31 06:47:34 -04:00
"""
event = yield self.store.get_event(
2019-06-20 05:32:02 -04:00
event_id, allow_none=True, allow_rejected=True
2014-10-31 06:47:34 -04:00
)
if event:
2019-06-20 05:32:02 -04:00
in_room = yield self.auth.check_host_in_room(event.room_id, origin)
if not in_room:
raise AuthError(403, "Host not in room.")
2016-08-10 08:22:20 -04:00
2019-10-23 12:25:54 -04:00
events = yield filter_events_for_server(self.storage, origin, [event])
event = events[0]
return event
2014-10-31 06:47:34 -04:00
else:
return None
2014-10-31 06:47:34 -04:00
def get_min_depth_for_context(self, context):
return self.store.get_min_depth(context)
@defer.inlineCallbacks
2019-06-20 05:32:02 -04:00
def _handle_new_event(
self, origin, event, state=None, auth_events=None, backfilled=False
):
2015-06-25 12:18:19 -04:00
context = yield self._prep_event(
2019-06-20 05:32:02 -04:00
origin, event, state=state, auth_events=auth_events, backfilled=backfilled
2015-06-25 12:18:19 -04:00
)
2018-10-01 13:48:51 -04:00
# reraise does not allow inlineCallbacks to preserve the stacktrace, so we
# hack around with a try/finally instead.
success = False
try:
if not event.internal_metadata.is_outlier() and not backfilled:
yield self.action_generator.handle_push_actions_for_event(
event, context
)
yield self.persist_events_and_notify(
2019-06-20 05:32:02 -04:00
[(event, context)], backfilled=backfilled
)
2018-10-01 13:48:51 -04:00
success = True
finally:
if not success:
run_in_background(
2019-06-20 05:32:02 -04:00
self.store.remove_push_actions_from_staging, event.event_id
2018-10-01 13:48:51 -04:00
)
2015-06-25 12:18:19 -04:00
return context
2015-06-25 12:18:19 -04:00
@defer.inlineCallbacks
def _handle_new_events(self, origin, event_infos, backfilled=False):
2016-04-12 07:48:30 -04:00
"""Creates the appropriate contexts and persists events. The events
should not depend on one another, e.g. this should be used to persist
a bunch of outliers, but not a chunk of individual events that depend
on each other for state calculations.
2018-08-01 08:39:14 -04:00
Notifies about the events where appropriate.
2016-04-12 07:48:30 -04:00
"""
@defer.inlineCallbacks
def prep(ev_info):
event = ev_info["event"]
with nested_logging_context(suffix=event.event_id):
res = yield self._prep_event(
2015-06-25 12:18:19 -04:00
origin,
event,
2015-06-25 12:18:19 -04:00
state=ev_info.get("state"),
auth_events=ev_info.get("auth_events"),
2019-02-12 05:31:21 -05:00
backfilled=backfilled,
2015-06-25 12:18:19 -04:00
)
return res
contexts = yield make_deferred_yieldable(
2019-06-20 05:32:02 -04:00
defer.gatherResults(
[run_in_background(prep, ev_info) for ev_info in event_infos],
2019-06-20 05:32:02 -04:00
consumeErrors=True,
)
)
2015-06-25 12:18:19 -04:00
yield self.persist_events_and_notify(
2015-06-25 12:18:19 -04:00
[
(ev_info["event"], context)
for ev_info, context in zip(event_infos, contexts)
2015-06-25 12:18:19 -04:00
],
backfilled=backfilled,
)
@defer.inlineCallbacks
def _persist_auth_tree(self, origin, auth_events, state, event):
"""Checks the auth chain is valid (and passes auth checks) for the
state and event. Then persists the auth chain and state atomically.
2018-08-01 08:39:14 -04:00
Persists the event separately. Notifies about the persisted events
where appropriate.
Will attempt to fetch missing auth events.
2016-07-29 05:45:05 -04:00
Args:
origin (str): Where the events came from
auth_events (list)
state (list)
event (Event)
Returns:
2018-08-01 08:39:14 -04:00
Deferred
"""
events_to_context = {}
for e in itertools.chain(auth_events, state):
e.internal_metadata.outlier = True
ctx = yield self.state_handler.compute_event_context(e)
events_to_context[e.event_id] = ctx
event_map = {
2019-06-20 05:32:02 -04:00
e.event_id: e for e in itertools.chain(auth_events, state, [event])
}
create_event = None
for e in auth_events:
if (e.type, e.state_key) == (EventTypes.Create, ""):
create_event = e
break
2019-01-24 13:31:23 -05:00
if create_event is None:
# If the state doesn't have a create event then the room is
# invalid, and it would fail auth checks anyway.
raise SynapseError(400, "No create event in state")
room_version = create_event.content.get(
2019-06-20 05:32:02 -04:00
"room_version", RoomVersions.V1.identifier
)
missing_auth_events = set()
for e in itertools.chain(auth_events, state, [event]):
for e_id in e.auth_event_ids():
if e_id not in event_map:
missing_auth_events.add(e_id)
for e_id in missing_auth_events:
m_ev = yield self.federation_client.get_pdu(
2019-06-20 05:32:02 -04:00
[origin], e_id, room_version=room_version, outlier=True, timeout=10000
)
if m_ev and m_ev.event_id == e_id:
event_map[e_id] = m_ev
else:
logger.info("Failed to find auth event %r", e_id)
for e in itertools.chain(auth_events, state, [event]):
auth_for_e = {
(event_map[e_id].type, event_map[e_id].state_key): event_map[e_id]
for e_id in e.auth_event_ids()
if e_id in event_map
}
if create_event:
auth_for_e[(EventTypes.Create, "")] = create_event
try:
event_auth.check(room_version, e, auth_events=auth_for_e)
except SynapseError as err:
# we may get SynapseErrors here as well as AuthErrors. For
# instance, there are a couple of (ancient) events in some
# rooms whose senders do not have the correct sigil; these
# cause SynapseErrors in auth.check. We don't want to give up
# the attempt to federate altogether in such cases.
logger.warning("Rejecting %s because %s", e.event_id, err.msg)
if e == event:
raise
events_to_context[e.event_id].rejected = RejectedReason.AUTH_ERROR
yield self.persist_events_and_notify(
[
(e, events_to_context[e.event_id])
for e in itertools.chain(auth_events, state)
2019-06-20 05:32:02 -04:00
]
)
new_event_context = yield self.state_handler.compute_event_context(
event, old_state=state
)
2019-06-20 05:32:02 -04:00
yield self.persist_events_and_notify([(event, new_event_context)])
2015-06-25 12:18:19 -04:00
@defer.inlineCallbacks
2019-02-12 05:31:21 -05:00
def _prep_event(self, origin, event, state, auth_events, backfilled):
"""
Args:
origin:
event:
state:
auth_events:
2019-02-12 05:31:21 -05:00
backfilled (bool)
2015-06-03 11:30:01 -04:00
Returns:
Deferred, which resolves to synapse.events.snapshot.EventContext
"""
2019-06-20 05:32:02 -04:00
context = yield self.state_handler.compute_event_context(event, old_state=state)
if not auth_events:
2018-07-23 08:00:22 -04:00
prev_state_ids = yield context.get_prev_state_ids(self.store)
auth_events_ids = yield self.auth.compute_auth_events(
2019-06-20 05:32:02 -04:00
event, prev_state_ids, for_verification=True
)
auth_events = yield self.store.get_events(auth_events_ids)
2019-06-20 05:32:02 -04:00
auth_events = {(e.type, e.state_key): e for e in auth_events.values()}
# This is a hack to fix some old rooms where the initial join event
# didn't reference the create event in its auth events.
if event.type == EventTypes.Member and not event.auth_event_ids():
if len(event.prev_event_ids()) == 1 and event.depth < 5:
c = yield self.store.get_event(
2019-06-20 05:32:02 -04:00
event.prev_event_ids()[0], allow_none=True
)
if c and c.type == EventTypes.Create:
auth_events[(c.type, c.state_key)] = c
context = yield self.do_auth(origin, event, context, auth_events=auth_events)
2015-01-28 11:16:53 -05:00
2019-03-08 06:26:33 -05:00
if not context.rejected:
yield self._check_for_soft_fail(event, state, backfilled)
if event.type == EventTypes.GuestAccess and not context.rejected:
yield self.maybe_kick_guest_users(event)
return context
2019-03-08 06:26:33 -05:00
@defer.inlineCallbacks
def _check_for_soft_fail(self, event, state, backfilled):
"""Checks if we should soft fail the event, if so marks the event as
such.
Args:
event (FrozenEvent)
state (dict|None): The state at the event if we don't have all the
event's prev events
backfilled (bool): Whether the event is from backfill
Returns:
Deferred
"""
2019-02-12 05:31:21 -05:00
# For new (non-backfilled and non-outlier) events we check if the event
# passes auth based on the current state. If it doesn't then we
# "soft-fail" the event.
do_soft_fail_check = not backfilled and not event.internal_metadata.is_outlier()
if do_soft_fail_check:
2019-06-20 05:32:02 -04:00
extrem_ids = yield self.store.get_latest_event_ids_in_room(event.room_id)
2019-02-12 05:31:21 -05:00
extrem_ids = set(extrem_ids)
prev_event_ids = set(event.prev_event_ids())
if extrem_ids == prev_event_ids:
# If they're the same then the current state is the same as the
# state at the event, so no point rechecking auth for soft fail.
do_soft_fail_check = False
if do_soft_fail_check:
room_version = yield self.store.get_room_version(event.room_id)
# Calculate the "current state".
if state is not None:
# If we're explicitly given the state then we won't have all the
# prev events, and so we have a gap in the graph. In this case
# we want to be a little careful as we might have been down for
# a while and have an incorrect view of the current state,
# however we still want to do checks as gaps are easy to
# maliciously manufacture.
#
# So we use a "current state" that is actually a state
# resolution across the current forward extremities and the
# given state at the event. This should correctly handle cases
# like bans, especially with state res v2.
2019-10-23 12:25:54 -04:00
state_sets = yield self.state_store.get_state_groups(
2019-06-20 05:32:02 -04:00
event.room_id, extrem_ids
2019-02-12 05:31:21 -05:00
)
state_sets = list(state_sets.values())
state_sets.append(state)
current_state_ids = yield self.state_handler.resolve_events(
2019-06-20 05:32:02 -04:00
room_version, state_sets, event
2019-02-12 05:31:21 -05:00
)
current_state_ids = {
k: e.event_id for k, e in iteritems(current_state_ids)
}
else:
current_state_ids = yield self.state_handler.get_current_state_ids(
2019-06-20 05:32:02 -04:00
event.room_id, latest_event_ids=extrem_ids
2019-02-12 05:31:21 -05:00
)
logger.debug(
"Doing soft-fail check for %s: state %s",
2019-06-20 05:32:02 -04:00
event.event_id,
current_state_ids,
)
2019-02-12 05:31:21 -05:00
# Now check if event pass auth against said current state
auth_types = auth_types_for_event(event)
current_state_ids = [
2019-06-20 05:32:02 -04:00
e for k, e in iteritems(current_state_ids) if k in auth_types
2019-02-12 05:31:21 -05:00
]
current_auth_events = yield self.store.get_events(current_state_ids)
current_auth_events = {
(e.type, e.state_key): e for e in current_auth_events.values()
}
try:
event_auth.check(room_version, event, auth_events=current_auth_events)
2019-02-12 05:31:21 -05:00
except AuthError as e:
logger.warning("Soft-failing %r because %s", event, e)
2019-02-12 05:31:21 -05:00
event.internal_metadata.soft_failed = True
@defer.inlineCallbacks
2019-06-20 05:32:02 -04:00
def on_query_auth(
self, origin, event_id, room_id, remote_auth_chain, rejects, missing
):
in_room = yield self.auth.check_host_in_room(room_id, origin)
if not in_room:
raise AuthError(403, "Host not in room.")
event = yield self.store.get_event(
event_id, allow_none=False, check_room_id=room_id
)
# Just go through and process each event in `remote_auth_chain`. We
# don't want to fall into the trap of `missing` being wrong.
for e in remote_auth_chain:
try:
yield self._handle_new_event(origin, e)
except AuthError:
pass
# Now get the current auth_chain for the event.
2017-05-24 09:22:41 -04:00
local_auth_chain = yield self.store.get_auth_chain(
2019-06-20 05:32:02 -04:00
[auth_id for auth_id in event.auth_event_ids()], include_given=True
2017-05-24 09:22:41 -04:00
)
# TODO: Check if we would now reject event_id. If so we need to tell
# everyone.
2019-06-20 05:32:02 -04:00
ret = yield self.construct_auth_difference(local_auth_chain, remote_auth_chain)
2015-02-03 08:23:58 -05:00
logger.debug("on_query_auth returning: %s", ret)
return ret
2015-01-28 11:16:53 -05:00
@defer.inlineCallbacks
2019-06-20 05:32:02 -04:00
def on_get_missing_events(
self, origin, room_id, earliest_events, latest_events, limit
):
in_room = yield self.auth.check_host_in_room(room_id, origin)
if not in_room:
raise AuthError(403, "Host not in room.")
limit = min(limit, 20)
missing_events = yield self.store.get_missing_events(
room_id=room_id,
earliest_events=earliest_events,
latest_events=latest_events,
limit=limit,
)
missing_events = yield filter_events_for_server(
2019-10-23 12:25:54 -04:00
self.storage, origin, missing_events
)
return missing_events
2015-01-28 11:16:53 -05:00
@defer.inlineCallbacks
@log_function
def do_auth(self, origin, event, context, auth_events):
"""
Args:
origin (str):
event (synapse.events.EventBase):
context (synapse.events.snapshot.EventContext):
auth_events (dict[(str, str)->synapse.events.EventBase]):
Map from (event_type, state_key) to event
What we expect the event's auth_events to be, based on the event's
position in the dag. I think? maybe??
Also NB that this function adds entries to it.
Returns:
defer.Deferred[EventContext]: updated context object
"""
room_version = yield self.store.get_room_version(event.room_id)
try:
context = yield self._update_auth_events_and_context_for_auth(
origin, event, context, auth_events
)
except Exception:
# We don't really mind if the above fails, so lets not fail
# processing if it does. However, it really shouldn't fail so
# let's still log as an exception since we'll still want to fix
# any bugs.
logger.exception(
"Failed to double check auth events for %s with remote. "
"Ignoring failure and continuing processing of event.",
event.event_id,
)
try:
event_auth.check(room_version, event, auth_events=auth_events)
except AuthError as e:
logger.warning("Failed auth resolution for %r because %s", event, e)
context.rejected = RejectedReason.AUTH_ERROR
return context
@defer.inlineCallbacks
def _update_auth_events_and_context_for_auth(
self, origin, event, context, auth_events
):
"""Helper for do_auth. See there for docs.
Checks whether a given event has the expected auth events. If it
doesn't then we talk to the remote server to compare state to see if
we can come to a consensus (e.g. if one server missed some valid
state).
This attempts to resovle any potential divergence of state between
servers, but is not essential and so failures should not block further
processing of the event.
Args:
origin (str):
event (synapse.events.EventBase):
context (synapse.events.snapshot.EventContext):
auth_events (dict[(str, str)->synapse.events.EventBase]):
Returns:
defer.Deferred[EventContext]: updated context
"""
event_auth_events = set(event.auth_event_ids())
2015-06-25 12:18:19 -04:00
if event.is_state():
event_key = (event.type, event.state_key)
else:
event_key = None
# if the event's auth_events refers to events which are not in our
# calculated auth_events, we need to fetch those events from somewhere.
#
# we start by fetching them from the store, and then try calling /event_auth/.
missing_auth = event_auth_events.difference(
e.event_id for e in auth_events.values()
)
if missing_auth:
# TODO: can we use store.have_seen_events here instead?
2019-06-20 05:32:02 -04:00
have_events = yield self.store.get_seen_events_with_rejections(missing_auth)
logger.debug("Got events %s from store", have_events)
missing_auth.difference_update(have_events.keys())
2015-06-25 12:18:19 -04:00
else:
have_events = {}
2019-06-20 05:32:02 -04:00
have_events.update({e.event_id: "" for e in auth_events.values()})
2015-06-25 12:18:19 -04:00
2015-01-28 11:16:53 -05:00
if missing_auth:
# If we don't have all the auth events, we need to get them.
2019-06-20 05:32:02 -04:00
logger.info("auth_events contains unknown events: %s", missing_auth)
try:
try:
remote_auth_chain = yield self.federation_client.get_event_auth(
origin, event.room_id, event.event_id
)
except RequestSendFailed as e:
# The other side isn't around or doesn't implement the
# endpoint, so lets just bail out.
logger.info("Failed to get event auth from remote: %s", e)
return context
seen_remotes = yield self.store.have_seen_events(
[e.event_id for e in remote_auth_chain]
)
for e in remote_auth_chain:
if e.event_id in seen_remotes:
continue
if e.event_id == event.event_id:
continue
try:
auth_ids = e.auth_event_ids()
auth = {
2019-06-20 05:32:02 -04:00
(e.type, e.state_key): e
for e in remote_auth_chain
if e.event_id in auth_ids or e.type == EventTypes.Create
}
e.internal_metadata.outlier = True
logger.debug(
2019-06-20 05:32:02 -04:00
"do_auth %s missing_auth: %s", event.event_id, e.event_id
)
2019-06-20 05:32:02 -04:00
yield self._handle_new_event(origin, e, auth_events=auth)
if e.event_id in event_auth_events:
auth_events[(e.type, e.state_key)] = e
except AuthError:
pass
have_events = yield self.store.get_seen_events_with_rejections(
event.auth_event_ids()
)
except Exception:
# FIXME:
logger.exception("Failed to get auth chain")
if event.internal_metadata.is_outlier():
logger.info("Skipping auth_event fetch for outlier")
return context
# FIXME: Assumes we have and stored all the state for all the
# prev_events
different_auth = event_auth_events.difference(
e.event_id for e in auth_events.values()
)
if not different_auth:
return context
logger.info(
"auth_events refers to events which are not in our calculated auth "
"chain: %s",
different_auth,
)
room_version = yield self.store.get_room_version(event.room_id)
different_events = yield make_deferred_yieldable(
2019-06-20 05:32:02 -04:00
defer.gatherResults(
[
run_in_background(
2019-06-20 05:32:02 -04:00
self.store.get_event, d, allow_none=True, allow_rejected=False
)
for d in different_auth
if d in have_events and not have_events[d]
],
consumeErrors=True,
)
).addErrback(unwrapFirstError)
if different_events:
local_view = dict(auth_events)
remote_view = dict(auth_events)
2019-06-20 05:32:02 -04:00
remote_view.update(
{(d.type, d.state_key): d for d in different_events if d}
)
new_state = yield self.state_handler.resolve_events(
room_version,
[list(local_view.values()), list(remote_view.values())],
2019-06-20 05:32:02 -04:00
event,
)
logger.info(
"After state res: updating auth_events with new state %s",
{
2019-06-20 05:32:02 -04:00
(d.type, d.state_key): d.event_id
for d in new_state.values()
if auth_events.get((d.type, d.state_key)) != d
},
)
auth_events.update(new_state)
context = yield self._update_context_for_auth_events(
2019-06-20 05:32:02 -04:00
event, context, auth_events, event_key
)
return context
@defer.inlineCallbacks
2019-06-20 05:32:02 -04:00
def _update_context_for_auth_events(self, event, context, auth_events, event_key):
"""Update the state_ids in an event context after auth event resolution,
storing the changes as a new state group.
Args:
event (Event): The event we're handling the context for
context (synapse.events.snapshot.EventContext): initial event context
auth_events (dict[(str, str)->str]): Events to update in the event
context.
event_key ((str, str)): (type, state_key) for the current event.
this will not be included in the current_state in the context.
Returns:
Deferred[EventContext]: new event context
"""
state_updates = {
2019-06-20 05:32:02 -04:00
k: a.event_id for k, a in iteritems(auth_events) if k != event_key
}
2018-07-23 08:02:09 -04:00
current_state_ids = yield context.get_current_state_ids(self.store)
current_state_ids = dict(current_state_ids)
current_state_ids.update(state_updates)
prev_state_ids = yield context.get_prev_state_ids(self.store)
prev_state_ids = dict(prev_state_ids)
2019-06-20 05:32:02 -04:00
prev_state_ids.update({k: a.event_id for k, a in iteritems(auth_events)})
2018-07-23 08:02:09 -04:00
# create a new state group as a delta from the existing one.
prev_group = context.state_group
2019-10-23 12:25:54 -04:00
state_group = yield self.state_store.store_state_group(
event.event_id,
event.room_id,
prev_group=prev_group,
delta_ids=state_updates,
2018-07-23 08:02:09 -04:00
current_state_ids=current_state_ids,
)
return EventContext.with_state(
2018-07-23 08:02:09 -04:00
state_group=state_group,
current_state_ids=current_state_ids,
prev_state_ids=prev_state_ids,
prev_group=prev_group,
delta_ids=state_updates,
)
2015-01-28 11:16:53 -05:00
@defer.inlineCallbacks
def construct_auth_difference(self, local_auth, remote_auth):
""" Given a local and remote auth chain, find the differences. This
assumes that we have already processed all events in remote_auth
Params:
local_auth (list)
remote_auth (list)
Returns:
dict
"""
logger.debug("construct_auth_difference Start!")
2015-01-28 11:16:53 -05:00
# TODO: Make sure we are OK with local_auth or remote_auth having more
# auth events in them than strictly necessary.
def sort_fun(ev):
return ev.depth, ev.event_id
logger.debug("construct_auth_difference after sort_fun!")
2015-01-28 11:16:53 -05:00
# We find the differences by starting at the "bottom" of each list
# and iterating up on both lists. The lists are ordered by depth and
# then event_id, we iterate up both lists until we find the event ids
# don't match. Then we look at depth/event_id to see which side is
# missing that event, and iterate only up that list. Repeat.
remote_list = list(remote_auth)
remote_list.sort(key=sort_fun)
local_list = list(local_auth)
local_list.sort(key=sort_fun)
local_iter = iter(local_list)
remote_iter = iter(remote_list)
logger.debug("construct_auth_difference before get_next!")
2015-01-28 11:16:53 -05:00
def get_next(it, opt=None):
try:
return next(it)
except Exception:
return opt
current_local = get_next(local_iter)
current_remote = get_next(remote_iter)
logger.debug("construct_auth_difference before while")
2015-01-28 11:16:53 -05:00
missing_remotes = []
missing_locals = []
while current_local or current_remote:
2015-01-28 11:16:53 -05:00
if current_remote is None:
missing_locals.append(current_local)
current_local = get_next(local_iter)
continue
if current_local is None:
missing_remotes.append(current_remote)
current_remote = get_next(remote_iter)
continue
if current_local.event_id == current_remote.event_id:
current_local = get_next(local_iter)
current_remote = get_next(remote_iter)
continue
if current_local.depth < current_remote.depth:
missing_locals.append(current_local)
current_local = get_next(local_iter)
continue
if current_local.depth > current_remote.depth:
missing_remotes.append(current_remote)
current_remote = get_next(remote_iter)
continue
# They have the same depth, so we fall back to the event_id order
if current_local.event_id < current_remote.event_id:
missing_locals.append(current_local)
current_local = get_next(local_iter)
if current_local.event_id > current_remote.event_id:
missing_remotes.append(current_remote)
current_remote = get_next(remote_iter)
continue
logger.debug("construct_auth_difference after while")
2015-01-28 11:16:53 -05:00
# missing locals should be sent to the server
# We should find why we are missing remotes, as they will have been
# rejected.
# Remove events from missing_remotes if they are referencing a missing
# remote. We only care about the "root" rejected ones.
missing_remote_ids = [e.event_id for e in missing_remotes]
base_remote_rejected = list(missing_remotes)
for e in missing_remotes:
for e_id in e.auth_event_ids():
2015-01-28 11:16:53 -05:00
if e_id in missing_remote_ids:
try:
base_remote_rejected.remove(e)
except ValueError:
pass
2015-01-28 11:16:53 -05:00
reason_map = {}
for e in base_remote_rejected:
reason = yield self.store.get_rejection_reason(e.event_id)
if reason is None:
# TODO: e is not in the current state, so we should
# construct some proof of that.
continue
2015-01-28 11:16:53 -05:00
reason_map[e.event_id] = reason
logger.debug("construct_auth_difference returning")
return {
"auth_chain": local_auth,
"rejects": {
e.event_id: {"reason": reason_map[e.event_id], "proof": None}
for e in base_remote_rejected
},
"missing": [e.event_id for e in missing_locals],
}
@defer.inlineCallbacks
@log_function
def exchange_third_party_invite(
2019-06-20 05:32:02 -04:00
self, sender_user_id, target_user_id, room_id, signed
):
2019-06-20 05:32:02 -04:00
third_party_invite = {"signed": signed}
event_dict = {
"type": EventTypes.Member,
"content": {
"membership": Membership.INVITE,
"third_party_invite": third_party_invite,
},
"room_id": room_id,
"sender": sender_user_id,
"state_key": target_user_id,
}
if (yield self.auth.check_host_in_room(room_id, self.hs.hostname)):
room_version = yield self.store.get_room_version(room_id)
builder = self.event_builder_factory.new(room_version, event_dict)
EventValidator().validate_builder(builder)
event, context = yield self.event_creation_handler.create_new_client_event(
builder=builder
)
event_allowed = yield self.third_party_event_rules.check_event_allowed(
2019-06-20 05:32:02 -04:00
event, context
)
if not event_allowed:
logger.info(
"Creation of threepid invite %s forbidden by third-party rules",
event,
)
raise SynapseError(
2019-06-20 05:32:02 -04:00
403, "This event is not allowed in this context", Codes.FORBIDDEN
)
event, context = yield self.add_display_name_to_third_party_invite(
room_version, event_dict, event, context
)
EventValidator().validate_new(event)
# We need to tell the transaction queue to send this out, even
# though the sender isn't a local user.
event.internal_metadata.send_on_behalf_of = self.hs.hostname
2016-04-13 06:11:46 -04:00
try:
yield self.auth.check_from_context(room_version, event, context)
2016-04-13 06:11:46 -04:00
except AuthError as e:
logger.warning("Denying new third party invite %r because %s", event, e)
2016-04-13 06:11:46 -04:00
raise e
yield self._check_signature(event, context)
# We retrieve the room member handler here as to not cause a cyclic dependency
2018-03-01 05:54:37 -05:00
member_handler = self.hs.get_room_member_handler()
yield member_handler.send_membership_event(None, event, context)
else:
destinations = set(x.split(":", 1)[-1] for x in (sender_user_id, room_id))
yield self.federation_client.forward_third_party_invite(
2019-06-20 05:32:02 -04:00
destinations, room_id, event_dict
)
@defer.inlineCallbacks
@log_function
def on_exchange_third_party_invite_request(self, room_id, event_dict):
"""Handle an exchange_third_party_invite request from a remote server
The remote server will call this when it wants to turn a 3pid invite
into a normal m.room.member invite.
Args:
room_id (str): The ID of the room.
event_dict (dict[str, Any]): Dictionary containing the event body.
Returns:
Deferred: resolves (to None)
"""
room_version = yield self.store.get_room_version(room_id)
# NB: event_dict has a particular specced format we might need to fudge
# if we change event formats too much.
builder = self.event_builder_factory.new(room_version, event_dict)
event, context = yield self.event_creation_handler.create_new_client_event(
2019-06-20 05:32:02 -04:00
builder=builder
)
event_allowed = yield self.third_party_event_rules.check_event_allowed(
2019-06-20 05:32:02 -04:00
event, context
)
if not event_allowed:
logger.warning(
2019-06-20 05:32:02 -04:00
"Exchange of threepid invite %s forbidden by third-party rules", event
)
raise SynapseError(
2019-06-20 05:32:02 -04:00
403, "This event is not allowed in this context", Codes.FORBIDDEN
)
event, context = yield self.add_display_name_to_third_party_invite(
room_version, event_dict, event, context
)
2016-04-13 06:11:46 -04:00
try:
2019-10-02 09:44:58 -04:00
yield self.auth.check_from_context(room_version, event, context)
2016-04-13 06:11:46 -04:00
except AuthError as e:
logger.warning("Denying third party invite %r because %s", event, e)
2016-04-13 06:11:46 -04:00
raise e
yield self._check_signature(event, context)
# We need to tell the transaction queue to send this out, even
# though the sender isn't a local user.
event.internal_metadata.send_on_behalf_of = get_domain_from_id(event.sender)
# We retrieve the room member handler here as to not cause a cyclic dependency
2018-03-01 05:54:37 -05:00
member_handler = self.hs.get_room_member_handler()
yield member_handler.send_membership_event(None, event, context)
@defer.inlineCallbacks
2019-06-20 05:32:02 -04:00
def add_display_name_to_third_party_invite(
self, room_version, event_dict, event, context
):
key = (
EventTypes.ThirdPartyInvite,
2019-06-20 05:32:02 -04:00
event.content["third_party_invite"]["signed"]["token"],
)
original_invite = None
2018-07-23 08:00:22 -04:00
prev_state_ids = yield context.get_prev_state_ids(self.store)
original_invite_id = prev_state_ids.get(key)
if original_invite_id:
original_invite = yield self.store.get_event(
original_invite_id, allow_none=True
)
if original_invite:
# If the m.room.third_party_invite event's content is empty, it means the
2019-10-04 06:16:19 -04:00
# invite has been revoked. In this case, we don't have to raise an error here
# because the auth check will fail on the invite (because it's not able to
# fetch public keys from the m.room.third_party_invite event's content, which
2019-10-04 06:21:24 -04:00
# is empty).
2019-10-04 06:16:19 -04:00
display_name = original_invite.content.get("display_name")
2019-10-04 06:18:28 -04:00
event_dict["content"]["third_party_invite"]["display_name"] = display_name
else:
logger.info(
2019-06-20 05:32:02 -04:00
"Could not find invite event for third_party_invite: %r", event_dict
)
2016-09-22 06:59:46 -04:00
# We don't discard here as this is not the appropriate place to do
# auth checks. If we need the invite and don't have it then the
# auth check code will explode appropriately.
builder = self.event_builder_factory.new(room_version, event_dict)
EventValidator().validate_builder(builder)
event, context = yield self.event_creation_handler.create_new_client_event(
2019-06-20 05:32:02 -04:00
builder=builder
2018-01-15 11:52:07 -05:00
)
EventValidator().validate_new(event)
return (event, context)
@defer.inlineCallbacks
def _check_signature(self, event, context):
"""
Checks that the signature in the event is consistent with its invite.
Args:
event (Event): The m.room.member event to check
context (EventContext):
Raises:
AuthError: if signature didn't match any keys, or key has been
revoked,
SynapseError: if a transient error meant a key couldn't be checked
for revocation.
"""
signed = event.content["third_party_invite"]["signed"]
token = signed["token"]
2018-07-23 08:00:22 -04:00
prev_state_ids = yield context.get_prev_state_ids(self.store)
2019-06-20 05:32:02 -04:00
invite_event_id = prev_state_ids.get((EventTypes.ThirdPartyInvite, token))
invite_event = None
if invite_event_id:
invite_event = yield self.store.get_event(invite_event_id, allow_none=True)
if not invite_event:
raise AuthError(403, "Could not find invite")
logger.debug("Checking auth on event %r", event.content)
last_exception = None
# for each public key in the 3pid invite event
for public_key_object in self.hs.get_auth().get_public_keys(invite_event):
try:
# for each sig on the third_party_invite block of the actual invite
for server, signature_block in signed["signatures"].items():
for key_name, encoded_signature in signature_block.items():
if not key_name.startswith("ed25519:"):
continue
logger.debug(
"Attempting to verify sig with key %s from %r "
"against pubkey %r",
2019-06-20 05:32:02 -04:00
key_name,
server,
public_key_object,
)
try:
public_key = public_key_object["public_key"]
verify_key = decode_verify_key_bytes(
2019-06-20 05:32:02 -04:00
key_name, decode_base64(public_key)
)
verify_signed_json(signed, server, verify_key)
logger.debug(
"Successfully verified sig with key %s from %r "
"against pubkey %r",
2019-06-20 05:32:02 -04:00
key_name,
server,
public_key_object,
)
except Exception:
logger.info(
"Failed to verify sig with key %s from %r "
"against pubkey %r",
2019-06-20 05:32:02 -04:00
key_name,
server,
public_key_object,
)
raise
try:
if "key_validity_url" in public_key_object:
yield self._check_key_revocation(
2019-06-20 05:32:02 -04:00
public_key, public_key_object["key_validity_url"]
)
except Exception:
logger.info(
"Failed to query key_validity_url %s",
2019-06-20 05:32:02 -04:00
public_key_object["key_validity_url"],
)
raise
return
except Exception as e:
last_exception = e
raise last_exception
@defer.inlineCallbacks
def _check_key_revocation(self, public_key, url):
"""
Checks whether public_key has been revoked.
Args:
public_key (str): base-64 encoded public key.
url (str): Key revocation URL.
Raises:
AuthError: if they key has been revoked.
SynapseError: if a transient error meant a key couldn't be checked
for revocation.
"""
try:
2019-06-20 05:32:02 -04:00
response = yield self.http_client.get_json(url, {"public_key": public_key})
except Exception:
2019-06-20 05:32:02 -04:00
raise SynapseError(502, "Third party certificate could not be checked")
if "valid" not in response or not response["valid"]:
raise AuthError(403, "Third party certificate was invalid")
@defer.inlineCallbacks
def persist_events_and_notify(self, event_and_contexts, backfilled=False):
"""Persists events and tells the notifier/pushers about them, if
necessary.
Args:
event_and_contexts(list[tuple[FrozenEvent, EventContext]])
backfilled (bool): Whether these events are a result of
backfilling or not
Returns:
Deferred
"""
if self.config.worker_app:
yield self._send_events_to_master(
store=self.store,
event_and_contexts=event_and_contexts,
2019-06-20 05:32:02 -04:00
backfilled=backfilled,
)
else:
2019-10-23 07:02:36 -04:00
max_stream_id = yield self.storage.persistence.persist_events(
2019-06-20 05:32:02 -04:00
event_and_contexts, backfilled=backfilled
)
if not backfilled: # Never notify for backfilled events
for event, _ in event_and_contexts:
yield self._notify_persisted_event(event, max_stream_id)
def _notify_persisted_event(self, event, max_stream_id):
"""Checks to see if notifier/pushers should be notified about the
event or not.
Args:
event (FrozenEvent)
max_stream_id (int): The max_stream_id returned by persist_events
"""
extra_users = []
if event.type == EventTypes.Member:
target_user_id = event.state_key
# We notify for memberships if its an invite for one of our
# users
if event.internal_metadata.is_outlier():
if event.membership != Membership.INVITE:
if not self.is_mine_id(target_user_id):
return
target_user = UserID.from_string(target_user_id)
extra_users.append(target_user)
elif event.internal_metadata.is_outlier():
return
event_stream_id = event.internal_metadata.stream_ordering
self.notifier.on_new_room_event(
2019-06-20 05:32:02 -04:00
event, event_stream_id, max_stream_id, extra_users=extra_users
)
2019-06-20 05:32:02 -04:00
return self.pusher_pool.on_new_notifications(event_stream_id, max_stream_id)
def _clean_room_for_join(self, room_id):
2018-08-09 05:29:48 -04:00
"""Called to clean up any data in DB for a given room, ready for the
server to join the room.
Args:
room_id (str)
"""
if self.config.worker_app:
return self._clean_room_for_join_client(room_id)
else:
return self.store.clean_room_for_join(room_id)
def user_joined_room(self, user, room_id):
"""Called when a new user has joined the room
"""
if self.config.worker_app:
return self._notify_user_membership_change(
2019-06-20 05:32:02 -04:00
room_id=room_id, user_id=user.to_string(), change="joined"
)
else:
return user_joined_room(self.distributor, user, room_id)
@defer.inlineCallbacks
def get_room_complexity(self, remote_room_hosts, room_id):
"""
Fetch the complexity of a remote room over federation.
Args:
remote_room_hosts (list[str]): The remote servers to ask.
room_id (str): The room ID to ask about.
Returns:
Deferred[dict] or Deferred[None]: Dict contains the complexity
metric versions, while None means we could not fetch the complexity.
"""
for host in remote_room_hosts:
res = yield self.federation_client.get_room_complexity(host, room_id)
# We got a result, return it.
if res:
defer.returnValue(res)
# We fell off the bottom, couldn't get the complexity from anyone. Oh
# well.
defer.returnValue(None)