def test_eviction(self):
cache = LruCache(2)
cache[1] = 1
cache[2] = 2
self.assertEquals(cache.get(1), 1)
self.assertEquals(cache.get(2), 2)
cache[3] = 3
self.assertEquals(cache.get(1), None)
self.assertEquals(cache.get(2), 2)
self.assertEquals(cache.get(3), 3)
def test_clear(self):
m1 = Mock()
m2 = Mock()
cache = LruCache(5)
cache.set("key1", "value", callbacks=[m1])
cache.set("key2", "value", callbacks=[m2])
self.assertEquals(m1.call_count, 0)
self.assertEquals(m2.call_count, 0)
cache.clear()
self.assertEquals(m1.call_count, 1)
self.assertEquals(m2.call_count, 1)
def __init__(self, hs):
self.hs = hs
self.clock = hs.get_clock()
self.store = hs.get_datastore()
self.state = hs.get_state_handler()
self.TOKEN_NOT_FOUND_HTTP_STATUS = 401
self.token_cache = LruCache(CACHE_SIZE_FACTOR * 10000)
register_cache("cache", "token_cache", self.token_cache)
def test_pop(self):
m = Mock()
cache = LruCache(1)
cache.set("key", "value", callbacks=[m])
self.assertFalse(m.called)
cache.pop("key")
self.assertEquals(m.call_count, 1)
cache.set("key", "value")
self.assertEquals(m.call_count, 1)
cache.pop("key")
self.assertEquals(m.call_count, 1)
def __init__(self, name, max_entries=1000, keylen=1, tree=False):
cache_type = TreeCache if tree else dict
self.cache = LruCache(
max_size=max_entries, keylen=keylen, cache_type=cache_type
)
self.name = name
self.keylen = keylen
self.sequence = 0
self.thread = None
self.metrics = register_cache(name, self.cache)
def test_evict(self):
cache = LruCache(5, size_callback=len)
cache["key1"] = [0]
cache["key2"] = [1, 2]
cache["key3"] = [3]
cache["key4"] = [4]
self.assertEquals(cache["key1"], [0])
self.assertEquals(cache["key2"], [1, 2])
self.assertEquals(cache["key3"], [3])
self.assertEquals(cache["key4"], [4])
self.assertEquals(len(cache), 5)
cache["key5"] = [5, 6]
self.assertEquals(len(cache), 4)
self.assertEquals(cache.get("key1"), None)
self.assertEquals(cache.get("key2"), None)
self.assertEquals(cache["key3"], [3])
self.assertEquals(cache["key4"], [4])
self.assertEquals(cache["key5"], [5, 6])
def __init__(self, name, max_entries=1000):
self.cache = LruCache(max_size=max_entries)
self.name = name
self.sequence = 0
self.thread = None
# caches_by_name[name] = self.cache
class Sentinel(object):
__slots__ = []
self.sentinel = Sentinel()
caches_by_name[name] = self.cache
def __init__(self, name, max_entries=1000):
self.cache = LruCache(max_size=max_entries, size_callback=len)
self.name = name
self.sequence = 0
self.thread = None
# caches_by_name[name] = self.cache
class Sentinel(object):
__slots__ = []
self.sentinel = Sentinel()
self.metrics = register_cache(name, self.cache)
def test_setdefault(self):
cache = LruCache(1)
self.assertEquals(cache.setdefault("key", 1), 1)
self.assertEquals(cache.get("key"), 1)
self.assertEquals(cache.setdefault("key", 2), 1)
self.assertEquals(cache.get("key"), 1)
cache["key"] = 2 # Make sure overriding works.
self.assertEquals(cache.get("key"), 2)
def __init__(self, name, max_entries=1000, keylen=1, tree=False, iterable=False):
cache_type = TreeCache if tree else dict
self._pending_deferred_cache = cache_type()
self.cache = LruCache(
max_size=max_entries, keylen=keylen, cache_type=cache_type,
size_callback=(lambda d: len(d)) if iterable else None,
evicted_callback=self._on_evicted,
)
self.name = name
self.keylen = keylen
self.thread = None
self.metrics = register_cache("cache", name, self.cache)
def test_del_multi(self):
cache = LruCache(4, 2, cache_type=TreeCache)
cache[("animal", "cat")] = "mew"
cache[("animal", "dog")] = "woof"
cache[("vehicles", "car")] = "vroom"
cache[("vehicles", "train")] = "chuff"
self.assertEquals(len(cache), 4)
self.assertEquals(cache.get(("animal", "cat")), "mew")
self.assertEquals(cache.get(("vehicles", "car")), "vroom")
cache.del_multi(("animal",))
self.assertEquals(len(cache), 2)
self.assertEquals(cache.get(("animal", "cat")), None)
self.assertEquals(cache.get(("animal", "dog")), None)
self.assertEquals(cache.get(("vehicles", "car")), "vroom")
self.assertEquals(cache.get(("vehicles", "train")), "chuff")
def test_del_multi(self):
m1 = Mock()
m2 = Mock()
m3 = Mock()
m4 = Mock()
cache = LruCache(4, 2, cache_type=TreeCache)
cache.set(("a", "1"), "value", callbacks=[m1])
cache.set(("a", "2"), "value", callbacks=[m2])
cache.set(("b", "1"), "value", callbacks=[m3])
cache.set(("b", "2"), "value", callbacks=[m4])
self.assertEquals(m1.call_count, 0)
self.assertEquals(m2.call_count, 0)
self.assertEquals(m3.call_count, 0)
self.assertEquals(m4.call_count, 0)
cache.del_multi(("a",))
self.assertEquals(m1.call_count, 1)
self.assertEquals(m2.call_count, 1)
self.assertEquals(m3.call_count, 0)
self.assertEquals(m4.call_count, 0)
def test_pop(self):
cache = LruCache(1)
cache["key"] = 1
self.assertEquals(cache.pop("key"), 1)
self.assertEquals(cache.pop("key"), None)
class DictionaryCache(object):
"""Caches key -> dictionary lookups, supporting caching partial dicts, i.e.
fetching a subset of dictionary keys for a particular key.
"""
def __init__(self, name, max_entries=1000):
self.cache = LruCache(max_size=max_entries, size_callback=len)
self.name = name
self.sequence = 0
self.thread = None
# caches_by_name[name] = self.cache
class Sentinel(object):
__slots__ = []
self.sentinel = Sentinel()
self.metrics = register_cache("dictionary", name, self.cache)
def check_thread(self):
expected_thread = self.thread
if expected_thread is None:
self.thread = threading.current_thread()
else:
if expected_thread is not threading.current_thread():
raise ValueError(
"Cache objects can only be accessed from the main thread")
def get(self, key, dict_keys=None):
"""Fetch an entry out of the cache
Args:
key
dict_key(list): If given a set of keys then return only those keys
that exist in the cache.
Returns:
DictionaryEntry
"""
entry = self.cache.get(key, self.sentinel)
if entry is not self.sentinel:
self.metrics.inc_hits()
if dict_keys is None:
return DictionaryEntry(entry.full, entry.known_absent,
dict(entry.value))
else:
return DictionaryEntry(
entry.full, entry.known_absent,
{k: entry.value[k]
for k in dict_keys if k in entry.value})
self.metrics.inc_misses()
return DictionaryEntry(False, set(), {})
def invalidate(self, key):
self.check_thread()
# Increment the sequence number so that any SELECT statements that
# raced with the INSERT don't update the cache (SYN-369)
self.sequence += 1
self.cache.pop(key, None)
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
def update(self, sequence, key, value, fetched_keys=None):
"""Updates the entry in the cache
Args:
sequence
key (K)
value (dict[X,Y]): The value to update the cache with.
fetched_keys (None|set[X]): All of the dictionary keys which were
fetched from the database.
If None, this is the complete value for key K. Otherwise, it
is used to infer a list of keys which we know don't exist in
the full dict.
"""
self.check_thread()
if self.sequence == sequence:
# Only update the cache if the caches sequence number matches the
# number that the cache had before the SELECT was started (SYN-369)
if fetched_keys is None:
self._insert(key, value, set())
else:
self._update_or_insert(key, value, fetched_keys)
def _update_or_insert(self, key, value, known_absent):
# We pop and reinsert as we need to tell the cache the size may have
# changed
entry = self.cache.pop(key, DictionaryEntry(False, set(), {}))
entry.value.update(value)
entry.known_absent.update(known_absent)
self.cache[key] = entry
def _insert(self, key, value, known_absent):
self.cache[key] = DictionaryEntry(True, known_absent, value)
def test_setdefault(self):
cache = LruCache(1)
self.assertEquals(cache.setdefault("key", 1), 1)
self.assertEquals(cache.get("key"), 1)
self.assertEquals(cache.setdefault("key", 2), 1)
self.assertEquals(cache.get("key"), 1)
class DictionaryCache(object):
"""Caches key -> dictionary lookups, supporting caching partial dicts, i.e.
fetching a subset of dictionary keys for a particular key.
"""
def __init__(self, name, max_entries=1000):
self.cache = LruCache(max_size=max_entries)
self.name = name
self.sequence = 0
self.thread = None
# caches_by_name[name] = self.cache
class Sentinel(object):
__slots__ = []
self.sentinel = Sentinel()
self.metrics = register_cache(name, self.cache)
def check_thread(self):
expected_thread = self.thread
if expected_thread is None:
self.thread = threading.current_thread()
else:
if expected_thread is not threading.current_thread():
raise ValueError(
"Cache objects can only be accessed from the main thread")
def get(self, key, dict_keys=None):
entry = self.cache.get(key, self.sentinel)
if entry is not self.sentinel:
self.metrics.inc_hits()
if dict_keys is None:
return DictionaryEntry(entry.full, dict(entry.value))
else:
return DictionaryEntry(
entry.full,
{k: entry.value[k]
for k in dict_keys if k in entry.value})
self.metrics.inc_misses()
return DictionaryEntry(False, {})
def invalidate(self, key):
self.check_thread()
# Increment the sequence number so that any SELECT statements that
# raced with the INSERT don't update the cache (SYN-369)
self.sequence += 1
self.cache.pop(key, None)
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
def update(self, sequence, key, value, full=False):
self.check_thread()
if self.sequence == sequence:
# Only update the cache if the caches sequence number matches the
# number that the cache had before the SELECT was started (SYN-369)
if full:
self._insert(key, value)
else:
self._update_or_insert(key, value)
def _update_or_insert(self, key, value):
entry = self.cache.setdefault(key, DictionaryEntry(False, {}))
entry.value.update(value)
def _insert(self, key, value):
self.cache[key] = DictionaryEntry(True, value)
class Auth(object):
"""
FIXME: This class contains a mix of functions for authenticating users
of our client-server API and authenticating events added to room graphs.
"""
def __init__(self, hs):
self.hs = hs
self.clock = hs.get_clock()
self.store = hs.get_datastore()
self.state = hs.get_state_handler()
self.token_cache = LruCache(10000)
register_cache("cache", "token_cache", self.token_cache)
self._auth_blocking = AuthBlocking(self.hs)
self._account_validity = hs.config.account_validity
self._track_appservice_user_ips = hs.config.track_appservice_user_ips
self._macaroon_secret_key = hs.config.macaroon_secret_key
@defer.inlineCallbacks
def check_from_context(self,
room_version: str,
event,
context,
do_sig_check=True):
prev_state_ids = yield context.get_prev_state_ids()
auth_events_ids = yield self.compute_auth_events(event,
prev_state_ids,
for_verification=True)
auth_events = yield self.store.get_events(auth_events_ids)
auth_events = {(e.type, e.state_key): e for e in auth_events.values()}
room_version_obj = KNOWN_ROOM_VERSIONS[room_version]
event_auth.check(room_version_obj,
event,
auth_events=auth_events,
do_sig_check=do_sig_check)
@defer.inlineCallbacks
def check_user_in_room(
self,
room_id: str,
user_id: str,
current_state: Optional[StateMap[EventBase]] = None,
allow_departed_users: bool = False,
):
"""Check if the user is in the room, or was at some point.
Args:
room_id: The room to check.
user_id: The user to check.
current_state: Optional map of the current state of the room.
If provided then that map is used to check whether they are a
member of the room. Otherwise the current membership is
loaded from the database.
allow_departed_users: if True, accept users that were previously
members but have now departed.
Raises:
AuthError if the user is/was not in the room.
Returns:
Deferred[Optional[EventBase]]:
Membership event for the user if the user was in the
room. This will be the join event if they are currently joined to
the room. This will be the leave event if they have left the room.
"""
if current_state:
member = current_state.get((EventTypes.Member, user_id), None)
else:
member = yield defer.ensureDeferred(
self.state.get_current_state(room_id=room_id,
event_type=EventTypes.Member,
state_key=user_id))
membership = member.membership if member else None
if membership == Membership.JOIN:
return member
# XXX this looks totally bogus. Why do we not allow users who have been banned,
# or those who were members previously and have been re-invited?
if allow_departed_users and membership == Membership.LEAVE:
forgot = yield self.store.did_forget(user_id, room_id)
if not forgot:
return member
raise AuthError(403, "User %s not in room %s" % (user_id, room_id))
@defer.inlineCallbacks
def check_host_in_room(self, room_id, host):
with Measure(self.clock, "check_host_in_room"):
latest_event_ids = yield self.store.is_host_joined(room_id, host)
return latest_event_ids
def can_federate(self, event, auth_events):
creation_event = auth_events.get((EventTypes.Create, ""))
return creation_event.content.get("m.federate", True) is True
def get_public_keys(self, invite_event):
return event_auth.get_public_keys(invite_event)
@defer.inlineCallbacks
def get_user_by_req(
self,
request: Request,
allow_guest: bool = False,
rights: str = "access",
allow_expired: bool = False,
):
""" Get a registered user's ID.
Args:
request: An HTTP request with an access_token query parameter.
allow_guest: If False, will raise an AuthError if the user making the
request is a guest.
rights: The operation being performed; the access token must allow this
allow_expired: If True, allow the request through even if the account
is expired, or session token lifetime has ended. Note that
/login will deliver access tokens regardless of expiration.
Returns:
defer.Deferred: resolves to a `synapse.types.Requester` object
Raises:
InvalidClientCredentialsError if no user by that token exists or the token
is invalid.
AuthError if access is denied for the user in the access token
"""
try:
ip_addr = self.hs.get_ip_from_request(request)
user_agent = request.requestHeaders.getRawHeaders(
b"User-Agent",
default=[b""])[0].decode("ascii", "surrogateescape")
access_token = self.get_access_token_from_request(request)
user_id, app_service = yield self._get_appservice_user_id(request)
if user_id:
request.authenticated_entity = user_id
opentracing.set_tag("authenticated_entity", user_id)
opentracing.set_tag("appservice_id", app_service.id)
if ip_addr and self._track_appservice_user_ips:
yield self.store.insert_client_ip(
user_id=user_id,
access_token=access_token,
ip=ip_addr,
user_agent=user_agent,
device_id="dummy-device", # stubbed
)
return synapse.types.create_requester(user_id,
app_service=app_service)
user_info = yield self.get_user_by_access_token(
access_token, rights, allow_expired=allow_expired)
user = user_info["user"]
token_id = user_info["token_id"]
is_guest = user_info["is_guest"]
# Deny the request if the user account has expired.
if self._account_validity.enabled and not allow_expired:
user_id = user.to_string()
expiration_ts = yield self.store.get_expiration_ts_for_user(
user_id)
if (expiration_ts is not None
and self.clock.time_msec() >= expiration_ts):
raise AuthError(403,
"User account has expired",
errcode=Codes.EXPIRED_ACCOUNT)
# device_id may not be present if get_user_by_access_token has been
# stubbed out.
device_id = user_info.get("device_id")
if user and access_token and ip_addr:
yield self.store.insert_client_ip(
user_id=user.to_string(),
access_token=access_token,
ip=ip_addr,
user_agent=user_agent,
device_id=device_id,
)
if is_guest and not allow_guest:
raise AuthError(
403,
"Guest access not allowed",
errcode=Codes.GUEST_ACCESS_FORBIDDEN,
)
request.authenticated_entity = user.to_string()
opentracing.set_tag("authenticated_entity", user.to_string())
if device_id:
opentracing.set_tag("device_id", device_id)
return synapse.types.create_requester(user,
token_id,
is_guest,
device_id,
app_service=app_service)
except KeyError:
raise MissingClientTokenError()
@defer.inlineCallbacks
def _get_appservice_user_id(self, request):
app_service = self.store.get_app_service_by_token(
self.get_access_token_from_request(request))
if app_service is None:
return None, None
if app_service.ip_range_whitelist:
ip_address = IPAddress(self.hs.get_ip_from_request(request))
if ip_address not in app_service.ip_range_whitelist:
return None, None
if b"user_id" not in request.args:
return app_service.sender, app_service
user_id = request.args[b"user_id"][0].decode("utf8")
if app_service.sender == user_id:
return app_service.sender, app_service
if not app_service.is_interested_in_user(user_id):
raise AuthError(
403, "Application service cannot masquerade as this user.")
if not (yield self.store.get_user_by_id(user_id)):
raise AuthError(
403, "Application service has not registered this user")
return user_id, app_service
@defer.inlineCallbacks
def get_user_by_access_token(
self,
token: str,
rights: str = "access",
allow_expired: bool = False,
):
""" Validate access token and get user_id from it
Args:
token: The access token to get the user by
rights: The operation being performed; the access token must
allow this
allow_expired: If False, raises an InvalidClientTokenError
if the token is expired
Returns:
Deferred[dict]: dict that includes:
`user` (UserID)
`is_guest` (bool)
`token_id` (int|None): access token id. May be None if guest
`device_id` (str|None): device corresponding to access token
Raises:
InvalidClientTokenError if a user by that token exists, but the token is
expired
InvalidClientCredentialsError if no user by that token exists or the token
is invalid
"""
if rights == "access":
# first look in the database
r = yield self._look_up_user_by_access_token(token)
if r:
valid_until_ms = r["valid_until_ms"]
if (not allow_expired and valid_until_ms is not None
and valid_until_ms < self.clock.time_msec()):
# there was a valid access token, but it has expired.
# soft-logout the user.
raise InvalidClientTokenError(
msg="Access token has expired", soft_logout=True)
return r
# otherwise it needs to be a valid macaroon
try:
user_id, guest = self._parse_and_validate_macaroon(token, rights)
user = UserID.from_string(user_id)
if rights == "access":
if not guest:
# non-guest access tokens must be in the database
logger.warning("Unrecognised access token - not in store.")
raise InvalidClientTokenError()
# Guest access tokens are not stored in the database (there can
# only be one access token per guest, anyway).
#
# In order to prevent guest access tokens being used as regular
# user access tokens (and hence getting around the invalidation
# process), we look up the user id and check that it is indeed
# a guest user.
#
# It would of course be much easier to store guest access
# tokens in the database as well, but that would break existing
# guest tokens.
stored_user = yield self.store.get_user_by_id(user_id)
if not stored_user:
raise InvalidClientTokenError("Unknown user_id %s" %
user_id)
if not stored_user["is_guest"]:
raise InvalidClientTokenError(
"Guest access token used for regular user")
ret = {
"user": user,
"is_guest": True,
"token_id": None,
# all guests get the same device id
"device_id": GUEST_DEVICE_ID,
}
elif rights == "delete_pusher":
# We don't store these tokens in the database
ret = {
"user": user,
"is_guest": False,
"token_id": None,
"device_id": None,
}
else:
raise RuntimeError("Unknown rights setting %s", rights)
return ret
except (
_InvalidMacaroonException,
pymacaroons.exceptions.MacaroonException,
TypeError,
ValueError,
) as e:
logger.warning("Invalid macaroon in auth: %s %s", type(e), e)
raise InvalidClientTokenError("Invalid macaroon passed.")
def _parse_and_validate_macaroon(self, token, rights="access"):
"""Takes a macaroon and tries to parse and validate it. This is cached
if and only if rights == access and there isn't an expiry.
On invalid macaroon raises _InvalidMacaroonException
Returns:
(user_id, is_guest)
"""
if rights == "access":
cached = self.token_cache.get(token, None)
if cached:
return cached
try:
macaroon = pymacaroons.Macaroon.deserialize(token)
except Exception: # deserialize can throw more-or-less anything
# doesn't look like a macaroon: treat it as an opaque token which
# must be in the database.
# TODO: it would be nice to get rid of this, but apparently some
# people use access tokens which aren't macaroons
raise _InvalidMacaroonException()
try:
user_id = self.get_user_id_from_macaroon(macaroon)
guest = False
for caveat in macaroon.caveats:
if caveat.caveat_id == "guest = true":
guest = True
self.validate_macaroon(macaroon, rights, user_id=user_id)
except (pymacaroons.exceptions.MacaroonException, TypeError,
ValueError):
raise InvalidClientTokenError("Invalid macaroon passed.")
if rights == "access":
self.token_cache[token] = (user_id, guest)
return user_id, guest
def get_user_id_from_macaroon(self, macaroon):
"""Retrieve the user_id given by the caveats on the macaroon.
Does *not* validate the macaroon.
Args:
macaroon (pymacaroons.Macaroon): The macaroon to validate
Returns:
(str) user id
Raises:
InvalidClientCredentialsError if there is no user_id caveat in the
macaroon
"""
user_prefix = "user_id = "
for caveat in macaroon.caveats:
if caveat.caveat_id.startswith(user_prefix):
return caveat.caveat_id[len(user_prefix):]
raise InvalidClientTokenError("No user caveat in macaroon")
def validate_macaroon(self, macaroon, type_string, user_id):
"""
validate that a Macaroon is understood by and was signed by this server.
Args:
macaroon(pymacaroons.Macaroon): The macaroon to validate
type_string(str): The kind of token required (e.g. "access",
"delete_pusher")
user_id (str): The user_id required
"""
v = pymacaroons.Verifier()
# the verifier runs a test for every caveat on the macaroon, to check
# that it is met for the current request. Each caveat must match at
# least one of the predicates specified by satisfy_exact or
# specify_general.
v.satisfy_exact("gen = 1")
v.satisfy_exact("type = " + type_string)
v.satisfy_exact("user_id = %s" % user_id)
v.satisfy_exact("guest = true")
v.satisfy_general(self._verify_expiry)
# access_tokens include a nonce for uniqueness: any value is acceptable
v.satisfy_general(lambda c: c.startswith("nonce = "))
v.verify(macaroon, self._macaroon_secret_key)
def _verify_expiry(self, caveat):
prefix = "time < "
if not caveat.startswith(prefix):
return False
expiry = int(caveat[len(prefix):])
now = self.hs.get_clock().time_msec()
return now < expiry
@defer.inlineCallbacks
def _look_up_user_by_access_token(self, token):
ret = yield self.store.get_user_by_access_token(token)
if not ret:
return None
# we use ret.get() below because *lots* of unit tests stub out
# get_user_by_access_token in a way where it only returns a couple of
# the fields.
user_info = {
"user": UserID.from_string(ret.get("name")),
"token_id": ret.get("token_id", None),
"is_guest": False,
"device_id": ret.get("device_id"),
"valid_until_ms": ret.get("valid_until_ms"),
}
return user_info
def get_appservice_by_req(self, request):
token = self.get_access_token_from_request(request)
service = self.store.get_app_service_by_token(token)
if not service:
logger.warning("Unrecognised appservice access token.")
raise InvalidClientTokenError()
request.authenticated_entity = service.sender
return defer.succeed(service)
async def is_server_admin(self, user: UserID) -> bool:
""" Check if the given user is a local server admin.
Args:
user: user to check
Returns:
True if the user is an admin
"""
return await self.store.is_server_admin(user)
def compute_auth_events(
self,
event,
current_state_ids: StateMap[str],
for_verification: bool = False,
):
"""Given an event and current state return the list of event IDs used
to auth an event.
If `for_verification` is False then only return auth events that
should be added to the event's `auth_events`.
Returns:
defer.Deferred(list[str]): List of event IDs.
"""
if event.type == EventTypes.Create:
return defer.succeed([])
# Currently we ignore the `for_verification` flag even though there are
# some situations where we can drop particular auth events when adding
# to the event's `auth_events` (e.g. joins pointing to previous joins
# when room is publicly joinable). Dropping event IDs has the
# advantage that the auth chain for the room grows slower, but we use
# the auth chain in state resolution v2 to order events, which means
# care must be taken if dropping events to ensure that it doesn't
# introduce undesirable "state reset" behaviour.
#
# All of which sounds a bit tricky so we don't bother for now.
auth_ids = []
for etype, state_key in event_auth.auth_types_for_event(event):
auth_ev_id = current_state_ids.get((etype, state_key))
if auth_ev_id:
auth_ids.append(auth_ev_id)
return defer.succeed(auth_ids)
async def check_can_change_room_list(self, room_id: str, user: UserID):
"""Determine whether the user is allowed to edit the room's entry in the
published room list.
Args:
room_id
user
"""
is_admin = await self.is_server_admin(user)
if is_admin:
return True
user_id = user.to_string()
await self.check_user_in_room(room_id, user_id)
# We currently require the user is a "moderator" in the room. We do this
# by checking if they would (theoretically) be able to change the
# m.room.canonical_alias events
power_level_event = await self.state.get_current_state(
room_id, EventTypes.PowerLevels, "")
auth_events = {}
if power_level_event:
auth_events[(EventTypes.PowerLevels, "")] = power_level_event
send_level = event_auth.get_send_level(EventTypes.CanonicalAlias, "",
power_level_event)
user_level = event_auth.get_user_power_level(user_id, auth_events)
return user_level >= send_level
@staticmethod
def has_access_token(request: Request):
"""Checks if the request has an access_token.
Returns:
bool: False if no access_token was given, True otherwise.
"""
query_params = request.args.get(b"access_token")
auth_headers = request.requestHeaders.getRawHeaders(b"Authorization")
return bool(query_params) or bool(auth_headers)
@staticmethod
def get_access_token_from_request(request: Request):
"""Extracts the access_token from the request.
Args:
request: The http request.
Returns:
unicode: The access_token
Raises:
MissingClientTokenError: If there isn't a single access_token in the
request
"""
auth_headers = request.requestHeaders.getRawHeaders(b"Authorization")
query_params = request.args.get(b"access_token")
if auth_headers:
# Try the get the access_token from a "Authorization: Bearer"
# header
if query_params is not None:
raise MissingClientTokenError(
"Mixing Authorization headers and access_token query parameters."
)
if len(auth_headers) > 1:
raise MissingClientTokenError(
"Too many Authorization headers.")
parts = auth_headers[0].split(b" ")
if parts[0] == b"Bearer" and len(parts) == 2:
return parts[1].decode("ascii")
else:
raise MissingClientTokenError("Invalid Authorization header.")
else:
# Try to get the access_token from the query params.
if not query_params:
raise MissingClientTokenError()
return query_params[0].decode("ascii")
@defer.inlineCallbacks
def check_user_in_room_or_world_readable(
self,
room_id: str,
user_id: str,
allow_departed_users: bool = False):
"""Checks that the user is or was in the room or the room is world
readable. If it isn't then an exception is raised.
Args:
room_id: room to check
user_id: user to check
allow_departed_users: if True, accept users that were previously
members but have now departed
Returns:
Deferred[tuple[str, str|None]]: Resolves to the current membership of
the user in the room and the membership event ID of the user. If
the user is not in the room and never has been, then
`(Membership.JOIN, None)` is returned.
"""
try:
# check_user_in_room will return the most recent membership
# event for the user if:
# * The user is a non-guest user, and was ever in the room
# * The user is a guest user, and has joined the room
# else it will throw.
member_event = yield self.check_user_in_room(
room_id, user_id, allow_departed_users=allow_departed_users)
return member_event.membership, member_event.event_id
except AuthError:
visibility = yield defer.ensureDeferred(
self.state.get_current_state(room_id,
EventTypes.RoomHistoryVisibility,
""))
if (visibility and visibility.content["history_visibility"]
== "world_readable"):
return Membership.JOIN, None
raise AuthError(
403,
"User %s not in room %s, and room previews are disabled" %
(user_id, room_id),
)
def check_auth_blocking(self, *args, **kwargs):
return self._auth_blocking.check_auth_blocking(*args, **kwargs)
class DictionaryCache(object):
"""Caches key -> dictionary lookups, supporting caching partial dicts, i.e.
fetching a subset of dictionary keys for a particular key.
"""
def __init__(self, name, max_entries=1000):
self.cache = LruCache(max_size=max_entries)
self.name = name
self.sequence = 0
self.thread = None
# caches_by_name[name] = self.cache
class Sentinel(object):
__slots__ = []
self.sentinel = Sentinel()
caches_by_name[name] = self.cache
def check_thread(self):
expected_thread = self.thread
if expected_thread is None:
self.thread = threading.current_thread()
else:
if expected_thread is not threading.current_thread():
raise ValueError(
"Cache objects can only be accessed from the main thread"
)
def get(self, key, dict_keys=None):
entry = self.cache.get(key, self.sentinel)
if entry is not self.sentinel:
cache_counter.inc_hits(self.name)
if dict_keys is None:
return DictionaryEntry(entry.full, dict(entry.value))
else:
return DictionaryEntry(entry.full, {
k: entry.value[k]
for k in dict_keys
if k in entry.value
})
cache_counter.inc_misses(self.name)
return DictionaryEntry(False, {})
def invalidate(self, key):
self.check_thread()
# Increment the sequence number so that any SELECT statements that
# raced with the INSERT don't update the cache (SYN-369)
self.sequence += 1
self.cache.pop(key, None)
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
def update(self, sequence, key, value, full=False):
self.check_thread()
if self.sequence == sequence:
# Only update the cache if the caches sequence number matches the
# number that the cache had before the SELECT was started (SYN-369)
if full:
self._insert(key, value)
else:
self._update_or_insert(key, value)
def _update_or_insert(self, key, value):
entry = self.cache.setdefault(key, DictionaryEntry(False, {}))
entry.value.update(value)
def _insert(self, key, value):
self.cache[key] = DictionaryEntry(True, value)
class EventFederationWorkerStore(EventsWorkerStore, SignatureWorkerStore,
SQLBaseStore):
def __init__(self, database: DatabasePool, db_conn, hs):
super().__init__(database, db_conn, hs)
if hs.config.run_background_tasks:
hs.get_clock().looping_call(self._delete_old_forward_extrem_cache,
60 * 60 * 1000)
# Cache of event ID to list of auth event IDs and their depths.
self._event_auth_cache = LruCache(
500000, "_event_auth_cache",
size_callback=len) # type: LruCache[str, List[Tuple[str, int]]]
async def get_auth_chain(self,
event_ids: Collection[str],
include_given: bool = False) -> List[EventBase]:
"""Get auth events for given event_ids. The events *must* be state events.
Args:
event_ids: state events
include_given: include the given events in result
Returns:
list of events
"""
event_ids = await self.get_auth_chain_ids(event_ids,
include_given=include_given)
return await self.get_events_as_list(event_ids)
async def get_auth_chain_ids(
self,
event_ids: Collection[str],
include_given: bool = False,
) -> List[str]:
"""Get auth events for given event_ids. The events *must* be state events.
Args:
event_ids: state events
include_given: include the given events in result
Returns:
An awaitable which resolve to a list of event_ids
"""
return await self.db_pool.runInteraction(
"get_auth_chain_ids",
self._get_auth_chain_ids_txn,
event_ids,
include_given,
)
def _get_auth_chain_ids_txn(self, txn: LoggingTransaction,
event_ids: Collection[str],
include_given: bool) -> List[str]:
if include_given:
results = set(event_ids)
else:
results = set()
# We pull out the depth simply so that we can populate the
# `_event_auth_cache` cache.
base_sql = """
SELECT a.event_id, auth_id, depth
FROM event_auth AS a
INNER JOIN events AS e ON (e.event_id = a.auth_id)
WHERE
"""
front = set(event_ids)
while front:
new_front = set()
for chunk in batch_iter(front, 100):
# Pull the auth events either from the cache or DB.
to_fetch = [] # Event IDs to fetch from DB # type: List[str]
for event_id in chunk:
res = self._event_auth_cache.get(event_id)
if res is None:
to_fetch.append(event_id)
else:
new_front.update(auth_id for auth_id, depth in res)
if to_fetch:
clause, args = make_in_list_sql_clause(
txn.database_engine, "a.event_id", to_fetch)
txn.execute(base_sql + clause, args)
# Note we need to batch up the results by event ID before
# adding to the cache.
to_cache = {}
for event_id, auth_event_id, auth_event_depth in txn:
to_cache.setdefault(event_id, []).append(
(auth_event_id, auth_event_depth))
new_front.add(auth_event_id)
for event_id, auth_events in to_cache.items():
self._event_auth_cache.set(event_id, auth_events)
new_front -= results
front = new_front
results.update(front)
return list(results)
async def get_auth_chain_difference(
self, room_id: str, state_sets: List[Set[str]]) -> Set[str]:
"""Given sets of state events figure out the auth chain difference (as
per state res v2 algorithm).
This equivalent to fetching the full auth chain for each set of state
and returning the events that don't appear in each and every auth
chain.
Returns:
The set of the difference in auth chains.
"""
# Check if we have indexed the room so we can use the chain cover
# algorithm.
room = await self.get_room(room_id)
if room["has_auth_chain_index"]:
try:
return await self.db_pool.runInteraction(
"get_auth_chain_difference_chains",
self._get_auth_chain_difference_using_cover_index_txn,
room_id,
state_sets,
)
except _NoChainCoverIndex:
# For whatever reason we don't actually have a chain cover index
# for the events in question, so we fall back to the old method.
pass
return await self.db_pool.runInteraction(
"get_auth_chain_difference",
self._get_auth_chain_difference_txn,
state_sets,
)
def _get_auth_chain_difference_using_cover_index_txn(
self, txn: Cursor, room_id: str,
state_sets: List[Set[str]]) -> Set[str]:
"""Calculates the auth chain difference using the chain index.
See docs/auth_chain_difference_algorithm.md for details
"""
# First we look up the chain ID/sequence numbers for all the events, and
# work out the chain/sequence numbers reachable from each state set.
initial_events = set(state_sets[0]).union(*state_sets[1:])
# Map from event_id -> (chain ID, seq no)
chain_info = {} # type: Dict[str, Tuple[int, int]]
# Map from chain ID -> seq no -> event Id
chain_to_event = {} # type: Dict[int, Dict[int, str]]
# All the chains that we've found that are reachable from the state
# sets.
seen_chains = set() # type: Set[int]
sql = """
SELECT event_id, chain_id, sequence_number
FROM event_auth_chains
WHERE %s
"""
for batch in batch_iter(initial_events, 1000):
clause, args = make_in_list_sql_clause(txn.database_engine,
"event_id", batch)
txn.execute(sql % (clause, ), args)
for event_id, chain_id, sequence_number in txn:
chain_info[event_id] = (chain_id, sequence_number)
seen_chains.add(chain_id)
chain_to_event.setdefault(chain_id,
{})[sequence_number] = event_id
# Check that we actually have a chain ID for all the events.
events_missing_chain_info = initial_events.difference(chain_info)
if events_missing_chain_info:
# This can happen due to e.g. downgrade/upgrade of the server. We
# raise an exception and fall back to the previous algorithm.
logger.info(
"Unexpectedly found that events don't have chain IDs in room %s: %s",
room_id,
events_missing_chain_info,
)
raise _NoChainCoverIndex(room_id)
# Corresponds to `state_sets`, except as a map from chain ID to max
# sequence number reachable from the state set.
set_to_chain = [] # type: List[Dict[int, int]]
for state_set in state_sets:
chains = {} # type: Dict[int, int]
set_to_chain.append(chains)
for event_id in state_set:
chain_id, seq_no = chain_info[event_id]
chains[chain_id] = max(seq_no, chains.get(chain_id, 0))
# Now we look up all links for the chains we have, adding chains to
# set_to_chain that are reachable from each set.
sql = """
SELECT
origin_chain_id, origin_sequence_number,
target_chain_id, target_sequence_number
FROM event_auth_chain_links
WHERE %s
"""
# (We need to take a copy of `seen_chains` as we want to mutate it in
# the loop)
for batch in batch_iter(set(seen_chains), 1000):
clause, args = make_in_list_sql_clause(txn.database_engine,
"origin_chain_id", batch)
txn.execute(sql % (clause, ), args)
for (
origin_chain_id,
origin_sequence_number,
target_chain_id,
target_sequence_number,
) in txn:
for chains in set_to_chain:
# chains are only reachable if the origin sequence number of
# the link is less than the max sequence number in the
# origin chain.
if origin_sequence_number <= chains.get(
origin_chain_id, 0):
chains[target_chain_id] = max(
target_sequence_number,
chains.get(target_chain_id, 0),
)
seen_chains.add(target_chain_id)
# Now for each chain we figure out the maximum sequence number reachable
# from *any* state set and the minimum sequence number reachable from
# *all* state sets. Events in that range are in the auth chain
# difference.
result = set()
# Mapping from chain ID to the range of sequence numbers that should be
# pulled from the database.
chain_to_gap = {} # type: Dict[int, Tuple[int, int]]
for chain_id in seen_chains:
min_seq_no = min(
chains.get(chain_id, 0) for chains in set_to_chain)
max_seq_no = max(
chains.get(chain_id, 0) for chains in set_to_chain)
if min_seq_no < max_seq_no:
# We have a non empty gap, try and fill it from the events that
# we have, otherwise add them to the list of gaps to pull out
# from the DB.
for seq_no in range(min_seq_no + 1, max_seq_no + 1):
event_id = chain_to_event.get(chain_id, {}).get(seq_no)
if event_id:
result.add(event_id)
else:
chain_to_gap[chain_id] = (min_seq_no, max_seq_no)
break
if not chain_to_gap:
# If there are no gaps to fetch, we're done!
return result
if isinstance(self.database_engine, PostgresEngine):
# We can use `execute_values` to efficiently fetch the gaps when
# using postgres.
sql = """
SELECT event_id
FROM event_auth_chains AS c, (VALUES ?) AS l(chain_id, min_seq, max_seq)
WHERE
c.chain_id = l.chain_id
AND min_seq < sequence_number AND sequence_number <= max_seq
"""
args = [(chain_id, min_no, max_no)
for chain_id, (min_no, max_no) in chain_to_gap.items()]
rows = txn.execute_values(sql, args)
result.update(r for r, in rows)
else:
# For SQLite we just fall back to doing a noddy for loop.
sql = """
SELECT event_id FROM event_auth_chains
WHERE chain_id = ? AND ? < sequence_number AND sequence_number <= ?
"""
for chain_id, (min_no, max_no) in chain_to_gap.items():
txn.execute(sql, (chain_id, min_no, max_no))
result.update(r for r, in txn)
return result
def _get_auth_chain_difference_txn(self, txn,
state_sets: List[Set[str]]) -> Set[str]:
"""Calculates the auth chain difference using a breadth first search.
This is used when we don't have a cover index for the room.
"""
# Algorithm Description
# ~~~~~~~~~~~~~~~~~~~~~
#
# The idea here is to basically walk the auth graph of each state set in
# tandem, keeping track of which auth events are reachable by each state
# set. If we reach an auth event we've already visited (via a different
# state set) then we mark that auth event and all ancestors as reachable
# by the state set. This requires that we keep track of the auth chains
# in memory.
#
# Doing it in a such a way means that we can stop early if all auth
# events we're currently walking are reachable by all state sets.
#
# *Note*: We can't stop walking an event's auth chain if it is reachable
# by all state sets. This is because other auth chains we're walking
# might be reachable only via the original auth chain. For example,
# given the following auth chain:
#
# A -> C -> D -> E
# / /
# B -´---------´
#
# and state sets {A} and {B} then walking the auth chains of A and B
# would immediately show that C is reachable by both. However, if we
# stopped at C then we'd only reach E via the auth chain of B and so E
# would errornously get included in the returned difference.
#
# The other thing that we do is limit the number of auth chains we walk
# at once, due to practical limits (i.e. we can only query the database
# with a limited set of parameters). We pick the auth chains we walk
# each iteration based on their depth, in the hope that events with a
# lower depth are likely reachable by those with higher depths.
#
# We could use any ordering that we believe would give a rough
# topological ordering, e.g. origin server timestamp. If the ordering
# chosen is not topological then the algorithm still produces the right
# result, but perhaps a bit more inefficiently. This is why it is safe
# to use "depth" here.
initial_events = set(state_sets[0]).union(*state_sets[1:])
# Dict from events in auth chains to which sets *cannot* reach them.
# I.e. if the set is empty then all sets can reach the event.
event_to_missing_sets = {
event_id:
{i
for i, a in enumerate(state_sets) if event_id not in a}
for event_id in initial_events
}
# The sorted list of events whose auth chains we should walk.
search = [] # type: List[Tuple[int, str]]
# We need to get the depth of the initial events for sorting purposes.
sql = """
SELECT depth, event_id FROM events
WHERE %s
"""
# the list can be huge, so let's avoid looking them all up in one massive
# query.
for batch in batch_iter(initial_events, 1000):
clause, args = make_in_list_sql_clause(txn.database_engine,
"event_id", batch)
txn.execute(sql % (clause, ), args)
# I think building a temporary list with fetchall is more efficient than
# just `search.extend(txn)`, but this is unconfirmed
search.extend(txn.fetchall())
# sort by depth
search.sort()
# Map from event to its auth events
event_to_auth_events = {} # type: Dict[str, Set[str]]
base_sql = """
SELECT a.event_id, auth_id, depth
FROM event_auth AS a
INNER JOIN events AS e ON (e.event_id = a.auth_id)
WHERE
"""
while search:
# Check whether all our current walks are reachable by all state
# sets. If so we can bail.
if all(not event_to_missing_sets[eid] for _, eid in search):
break
# Fetch the auth events and their depths of the N last events we're
# currently walking, either from cache or DB.
search, chunk = search[:-100], search[-100:]
found = [] # Results found # type: List[Tuple[str, str, int]]
to_fetch = [] # Event IDs to fetch from DB # type: List[str]
for _, event_id in chunk:
res = self._event_auth_cache.get(event_id)
if res is None:
to_fetch.append(event_id)
else:
found.extend(
(event_id, auth_id, depth) for auth_id, depth in res)
if to_fetch:
clause, args = make_in_list_sql_clause(txn.database_engine,
"a.event_id", to_fetch)
txn.execute(base_sql + clause, args)
# We parse the results and add the to the `found` set and the
# cache (note we need to batch up the results by event ID before
# adding to the cache).
to_cache = {}
for event_id, auth_event_id, auth_event_depth in txn:
to_cache.setdefault(event_id, []).append(
(auth_event_id, auth_event_depth))
found.append((event_id, auth_event_id, auth_event_depth))
for event_id, auth_events in to_cache.items():
self._event_auth_cache.set(event_id, auth_events)
for event_id, auth_event_id, auth_event_depth in found:
event_to_auth_events.setdefault(event_id,
set()).add(auth_event_id)
sets = event_to_missing_sets.get(auth_event_id)
if sets is None:
# First time we're seeing this event, so we add it to the
# queue of things to fetch.
search.append((auth_event_depth, auth_event_id))
# Assume that this event is unreachable from any of the
# state sets until proven otherwise
sets = event_to_missing_sets[auth_event_id] = set(
range(len(state_sets)))
else:
# We've previously seen this event, so look up its auth
# events and recursively mark all ancestors as reachable
# by the current event's state set.
a_ids = event_to_auth_events.get(auth_event_id)
while a_ids:
new_aids = set()
for a_id in a_ids:
event_to_missing_sets[a_id].intersection_update(
event_to_missing_sets[event_id])
b = event_to_auth_events.get(a_id)
if b:
new_aids.update(b)
a_ids = new_aids
# Mark that the auth event is reachable by the approriate sets.
sets.intersection_update(event_to_missing_sets[event_id])
search.sort()
# Return all events where not all sets can reach them.
return {eid for eid, n in event_to_missing_sets.items() if n}
async def get_oldest_events_with_depth_in_room(self, room_id):
return await self.db_pool.runInteraction(
"get_oldest_events_with_depth_in_room",
self.get_oldest_events_with_depth_in_room_txn,
room_id,
)
def get_oldest_events_with_depth_in_room_txn(self, txn, room_id):
sql = ("SELECT b.event_id, MAX(e.depth) FROM events as e"
" INNER JOIN event_edges as g"
" ON g.event_id = e.event_id"
" INNER JOIN event_backward_extremities as b"
" ON g.prev_event_id = b.event_id"
" WHERE b.room_id = ? AND g.is_state is ?"
" GROUP BY b.event_id")
txn.execute(sql, (room_id, False))
return dict(txn)
async def get_max_depth_of(self, event_ids: List[str]) -> int:
"""Returns the max depth of a set of event IDs
Args:
event_ids: The event IDs to calculate the max depth of.
"""
rows = await self.db_pool.simple_select_many_batch(
table="events",
column="event_id",
iterable=event_ids,
retcols=("depth", ),
desc="get_max_depth_of",
)
if not rows:
return 0
else:
return max(row["depth"] for row in rows)
async def get_prev_events_for_room(self, room_id: str) -> List[str]:
"""
Gets a subset of the current forward extremities in the given room.
Limits the result to 10 extremities, so that we can avoid creating
events which refer to hundreds of prev_events.
Args:
room_id: room_id
Returns:
The event ids of the forward extremities.
"""
return await self.db_pool.runInteraction(
"get_prev_events_for_room", self._get_prev_events_for_room_txn,
room_id)
def _get_prev_events_for_room_txn(self, txn, room_id: str):
# we just use the 10 newest events. Older events will become
# prev_events of future events.
sql = """
SELECT e.event_id FROM event_forward_extremities AS f
INNER JOIN events AS e USING (event_id)
WHERE f.room_id = ?
ORDER BY e.depth DESC
LIMIT 10
"""
txn.execute(sql, (room_id, ))
return [row[0] for row in txn]
async def get_rooms_with_many_extremities(
self, min_count: int, limit: int,
room_id_filter: Iterable[str]) -> List[str]:
"""Get the top rooms with at least N extremities.
Args:
min_count: The minimum number of extremities
limit: The maximum number of rooms to return.
room_id_filter: room_ids to exclude from the results
Returns:
At most `limit` room IDs that have at least `min_count` extremities,
sorted by extremity count.
"""
def _get_rooms_with_many_extremities_txn(txn):
where_clause = "1=1"
if room_id_filter:
where_clause = "room_id NOT IN (%s)" % (",".join(
"?" for _ in room_id_filter), )
sql = """
SELECT room_id FROM event_forward_extremities
WHERE %s
GROUP BY room_id
HAVING count(*) > ?
ORDER BY count(*) DESC
LIMIT ?
""" % (where_clause, )
query_args = list(
itertools.chain(room_id_filter, [min_count, limit]))
txn.execute(sql, query_args)
return [room_id for room_id, in txn]
return await self.db_pool.runInteraction(
"get_rooms_with_many_extremities",
_get_rooms_with_many_extremities_txn)
@cached(max_entries=5000, iterable=True)
async def get_latest_event_ids_in_room(self, room_id: str) -> List[str]:
return await self.db_pool.simple_select_onecol(
table="event_forward_extremities",
keyvalues={"room_id": room_id},
retcol="event_id",
desc="get_latest_event_ids_in_room",
)
async def get_min_depth(self, room_id: str) -> int:
"""For the given room, get the minimum depth we have seen for it.
"""
return await self.db_pool.runInteraction(
"get_min_depth", self._get_min_depth_interaction, room_id)
def _get_min_depth_interaction(self, txn, room_id):
min_depth = self.db_pool.simple_select_one_onecol_txn(
txn,
table="room_depth",
keyvalues={"room_id": room_id},
retcol="min_depth",
allow_none=True,
)
return int(min_depth) if min_depth is not None else None
async def get_forward_extremeties_for_room(
self, room_id: str, stream_ordering: int) -> List[str]:
"""For a given room_id and stream_ordering, return the forward
extremeties of the room at that point in "time".
Throws a StoreError if we have since purged the index for
stream_orderings from that point.
Args:
room_id:
stream_ordering:
Returns:
A list of event_ids
"""
# We want to make the cache more effective, so we clamp to the last
# change before the given ordering.
last_change = self._events_stream_cache.get_max_pos_of_last_change(
room_id)
# We don't always have a full stream_to_exterm_id table, e.g. after
# the upgrade that introduced it, so we make sure we never ask for a
# stream_ordering from before a restart
last_change = max(self._stream_order_on_start, last_change)
# provided the last_change is recent enough, we now clamp the requested
# stream_ordering to it.
if last_change > self.stream_ordering_month_ago:
stream_ordering = min(last_change, stream_ordering)
return await self._get_forward_extremeties_for_room(
room_id, stream_ordering)
@cached(max_entries=5000, num_args=2)
async def _get_forward_extremeties_for_room(self, room_id,
stream_ordering):
"""For a given room_id and stream_ordering, return the forward
extremeties of the room at that point in "time".
Throws a StoreError if we have since purged the index for
stream_orderings from that point.
"""
if stream_ordering <= self.stream_ordering_month_ago:
raise StoreError(
400, "stream_ordering too old %s" % (stream_ordering, ))
sql = """
SELECT event_id FROM stream_ordering_to_exterm
INNER JOIN (
SELECT room_id, MAX(stream_ordering) AS stream_ordering
FROM stream_ordering_to_exterm
WHERE stream_ordering <= ? GROUP BY room_id
) AS rms USING (room_id, stream_ordering)
WHERE room_id = ?
"""
def get_forward_extremeties_for_room_txn(txn):
txn.execute(sql, (stream_ordering, room_id))
return [event_id for event_id, in txn]
return await self.db_pool.runInteraction(
"get_forward_extremeties_for_room",
get_forward_extremeties_for_room_txn)
async def get_backfill_events(self, room_id: str, event_list: list,
limit: int):
"""Get a list of Events for a given topic that occurred before (and
including) the events in event_list. Return a list of max size `limit`
Args:
room_id
event_list
limit
"""
event_ids = await self.db_pool.runInteraction(
"get_backfill_events",
self._get_backfill_events,
room_id,
event_list,
limit,
)
events = await self.get_events_as_list(event_ids)
return sorted(events, key=lambda e: -e.depth)
def _get_backfill_events(self, txn, room_id, event_list, limit):
logger.debug("_get_backfill_events: %s, %r, %s", room_id, event_list,
limit)
event_results = set()
# We want to make sure that we do a breadth-first, "depth" ordered
# search.
query = ("SELECT depth, prev_event_id FROM event_edges"
" INNER JOIN events"
" ON prev_event_id = events.event_id"
" WHERE event_edges.event_id = ?"
" AND event_edges.is_state = ?"
" LIMIT ?")
queue = PriorityQueue()
for event_id in event_list:
depth = self.db_pool.simple_select_one_onecol_txn(
txn,
table="events",
keyvalues={
"event_id": event_id,
"room_id": room_id
},
retcol="depth",
allow_none=True,
)
if depth:
queue.put((-depth, event_id))
while not queue.empty() and len(event_results) < limit:
try:
_, event_id = queue.get_nowait()
except Empty:
break
if event_id in event_results:
continue
event_results.add(event_id)
txn.execute(query, (event_id, False, limit - len(event_results)))
for row in txn:
if row[1] not in event_results:
queue.put((-row[0], row[1]))
return event_results
async def get_missing_events(self, room_id, earliest_events, latest_events,
limit):
ids = await self.db_pool.runInteraction(
"get_missing_events",
self._get_missing_events,
room_id,
earliest_events,
latest_events,
limit,
)
return await self.get_events_as_list(ids)
def _get_missing_events(self, txn, room_id, earliest_events, latest_events,
limit):
seen_events = set(earliest_events)
front = set(latest_events) - seen_events
event_results = []
query = ("SELECT prev_event_id FROM event_edges "
"WHERE room_id = ? AND event_id = ? AND is_state = ? "
"LIMIT ?")
while front and len(event_results) < limit:
new_front = set()
for event_id in front:
txn.execute(
query,
(room_id, event_id, False, limit - len(event_results)))
new_results = {t[0] for t in txn} - seen_events
new_front |= new_results
seen_events |= new_results
event_results.extend(new_results)
front = new_front
# we built the list working backwards from latest_events; we now need to
# reverse it so that the events are approximately chronological.
event_results.reverse()
return event_results
async def get_successor_events(self,
event_ids: Iterable[str]) -> List[str]:
"""Fetch all events that have the given events as a prev event
Args:
event_ids: The events to use as the previous events.
"""
rows = await self.db_pool.simple_select_many_batch(
table="event_edges",
column="prev_event_id",
iterable=event_ids,
retcols=("event_id", ),
desc="get_successor_events",
)
return [row["event_id"] for row in rows]
@wrap_as_background_process("delete_old_forward_extrem_cache")
async def _delete_old_forward_extrem_cache(self) -> None:
def _delete_old_forward_extrem_cache_txn(txn):
# Delete entries older than a month, while making sure we don't delete
# the only entries for a room.
sql = """
DELETE FROM stream_ordering_to_exterm
WHERE
room_id IN (
SELECT room_id
FROM stream_ordering_to_exterm
WHERE stream_ordering > ?
) AND stream_ordering < ?
"""
txn.execute(sql, (self.stream_ordering_month_ago,
self.stream_ordering_month_ago))
await self.db_pool.runInteraction(
"_delete_old_forward_extrem_cache",
_delete_old_forward_extrem_cache_txn,
)
def test_eviction(self):
m1 = Mock(name="m1")
m2 = Mock(name="m2")
m3 = Mock(name="m3")
cache = LruCache(2)
cache.set("key1", "value", callbacks=[m1])
cache.set("key2", "value", callbacks=[m2])
self.assertEqual(m1.call_count, 0)
self.assertEqual(m2.call_count, 0)
self.assertEqual(m3.call_count, 0)
cache.set("key3", "value", callbacks=[m3])
self.assertEqual(m1.call_count, 1)
self.assertEqual(m2.call_count, 0)
self.assertEqual(m3.call_count, 0)
cache.set("key3", "value")
self.assertEqual(m1.call_count, 1)
self.assertEqual(m2.call_count, 0)
self.assertEqual(m3.call_count, 0)
cache.get("key2")
self.assertEqual(m1.call_count, 1)
self.assertEqual(m2.call_count, 0)
self.assertEqual(m3.call_count, 0)
cache.set("key1", "value", callbacks=[m1])
self.assertEqual(m1.call_count, 1)
self.assertEqual(m2.call_count, 0)
self.assertEqual(m3.call_count, 1)
class ClientIpStore(ClientIpWorkerStore):
def __init__(self, database: DatabasePool, db_conn, hs):
self.client_ip_last_seen = LruCache(
cache_name="client_ip_last_seen", max_size=50000
)
super().__init__(database, db_conn, hs)
# (user_id, access_token, ip,) -> (user_agent, device_id, last_seen)
self._batch_row_update = {}
self._client_ip_looper = self._clock.looping_call(
self._update_client_ips_batch, 5 * 1000
)
self.hs.get_reactor().addSystemEventTrigger(
"before", "shutdown", self._update_client_ips_batch
)
async def insert_client_ip(
self, user_id, access_token, ip, user_agent, device_id, now=None
):
if not now:
now = int(self._clock.time_msec())
key = (user_id, access_token, ip)
try:
last_seen = self.client_ip_last_seen.get(key)
except KeyError:
last_seen = None
await self.populate_monthly_active_users(user_id)
# Rate-limited inserts
if last_seen is not None and (now - last_seen) < LAST_SEEN_GRANULARITY:
return
self.client_ip_last_seen.set(key, now)
self._batch_row_update[key] = (user_agent, device_id, now)
@wrap_as_background_process("update_client_ips")
async def _update_client_ips_batch(self) -> None:
# If the DB pool has already terminated, don't try updating
if not self.db_pool.is_running():
return
to_update = self._batch_row_update
self._batch_row_update = {}
await self.db_pool.runInteraction(
"_update_client_ips_batch", self._update_client_ips_batch_txn, to_update
)
def _update_client_ips_batch_txn(self, txn, to_update):
if "user_ips" in self.db_pool._unsafe_to_upsert_tables or (
not self.database_engine.can_native_upsert
):
self.database_engine.lock_table(txn, "user_ips")
for entry in to_update.items():
(user_id, access_token, ip), (user_agent, device_id, last_seen) = entry
self.db_pool.simple_upsert_txn(
txn,
table="user_ips",
keyvalues={"user_id": user_id, "access_token": access_token, "ip": ip},
values={
"user_agent": user_agent,
"device_id": device_id,
"last_seen": last_seen,
},
lock=False,
)
# Technically an access token might not be associated with
# a device so we need to check.
if device_id:
# this is always an update rather than an upsert: the row should
# already exist, and if it doesn't, that may be because it has been
# deleted, and we don't want to re-create it.
self.db_pool.simple_update_txn(
txn,
table="devices",
keyvalues={"user_id": user_id, "device_id": device_id},
updatevalues={
"user_agent": user_agent,
"last_seen": last_seen,
"ip": ip,
},
)
async def get_last_client_ip_by_device(
self, user_id: str, device_id: Optional[str]
) -> Dict[Tuple[str, str], dict]:
"""For each device_id listed, give the user_ip it was last seen on
Args:
user_id: The user to fetch devices for.
device_id: If None fetches all devices for the user
Returns:
A dictionary mapping a tuple of (user_id, device_id) to dicts, with
keys giving the column names from the devices table.
"""
ret = await super().get_last_client_ip_by_device(user_id, device_id)
# Update what is retrieved from the database with data which is pending
# insertion, as if it has already been stored in the database.
for key in self._batch_row_update:
uid, _access_token, ip = key
if uid == user_id:
user_agent, did, last_seen = self._batch_row_update[key]
if did is None:
# These updates don't make it to the `devices` table
continue
if not device_id or did == device_id:
ret[(user_id, did)] = {
"user_id": user_id,
"ip": ip,
"user_agent": user_agent,
"device_id": did,
"last_seen": last_seen,
}
return ret
async def get_user_ip_and_agents(
self, user: UserID, since_ts: int = 0
) -> List[Dict[str, Union[str, int]]]:
"""
Fetch IP/User Agent connection since a given timestamp.
"""
user_id = user.to_string()
results = {}
for key in self._batch_row_update:
(
uid,
access_token,
ip,
) = key
if uid == user_id:
user_agent, _, last_seen = self._batch_row_update[key]
if last_seen >= since_ts:
results[(access_token, ip)] = (user_agent, last_seen)
def get_recent(txn):
txn.execute(
"""
SELECT access_token, ip, user_agent, last_seen FROM user_ips
WHERE last_seen >= ? AND user_id = ?
ORDER BY last_seen
DESC
""",
(since_ts, user_id),
)
return txn.fetchall()
rows = await self.db_pool.runInteraction(
desc="get_user_ip_and_agents", func=get_recent
)
results.update(
((access_token, ip), (user_agent, last_seen))
for access_token, ip, user_agent, last_seen in rows
)
return [
{
"access_token": access_token,
"ip": ip,
"user_agent": user_agent,
"last_seen": last_seen,
}
for (access_token, ip), (user_agent, last_seen) in results.items()
]
class Cache(object):
__slots__ = (
"cache",
"max_entries",
"name",
"keylen",
"thread",
"metrics",
"_pending_deferred_cache",
)
def __init__(self, name, max_entries=1000, keylen=1, tree=False, iterable=False):
cache_type = TreeCache if tree else dict
self._pending_deferred_cache = cache_type()
self.cache = LruCache(
max_size=max_entries, keylen=keylen, cache_type=cache_type,
size_callback=(lambda d: len(d)) if iterable else None,
evicted_callback=self._on_evicted,
)
self.name = name
self.keylen = keylen
self.thread = None
self.metrics = register_cache("cache", name, self.cache)
def _on_evicted(self, evicted_count):
self.metrics.inc_evictions(evicted_count)
def check_thread(self):
expected_thread = self.thread
if expected_thread is None:
self.thread = threading.current_thread()
else:
if expected_thread is not threading.current_thread():
raise ValueError(
"Cache objects can only be accessed from the main thread"
)
def get(self, key, default=_CacheSentinel, callback=None, update_metrics=True):
"""Looks the key up in the caches.
Args:
key(tuple)
default: What is returned if key is not in the caches. If not
specified then function throws KeyError instead
callback(fn): Gets called when the entry in the cache is invalidated
update_metrics (bool): whether to update the cache hit rate metrics
Returns:
Either a Deferred or the raw result
"""
callbacks = [callback] if callback else []
val = self._pending_deferred_cache.get(key, _CacheSentinel)
if val is not _CacheSentinel:
val.callbacks.update(callbacks)
if update_metrics:
self.metrics.inc_hits()
return val.deferred
val = self.cache.get(key, _CacheSentinel, callbacks=callbacks)
if val is not _CacheSentinel:
self.metrics.inc_hits()
return val
if update_metrics:
self.metrics.inc_misses()
if default is _CacheSentinel:
raise KeyError()
else:
return default
def set(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.check_thread()
entry = CacheEntry(
deferred=value,
callbacks=callbacks,
)
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry:
existing_entry.invalidate()
self._pending_deferred_cache[key] = entry
def shuffle(result):
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry is entry:
self.cache.set(key, result, entry.callbacks)
else:
# oops, the _pending_deferred_cache has been updated since
# we started our query, so we are out of date.
#
# Better put back whatever we took out. (We do it this way
# round, rather than peeking into the _pending_deferred_cache
# and then removing on a match, to make the common case faster)
if existing_entry is not None:
self._pending_deferred_cache[key] = existing_entry
# we're not going to put this entry into the cache, so need
# to make sure that the invalidation callbacks are called.
# That was probably done when _pending_deferred_cache was
# updated, but it's possible that `set` was called without
# `invalidate` being previously called, in which case it may
# not have been. Either way, let's double-check now.
entry.invalidate()
return result
entry.deferred.addCallback(shuffle)
def prefill(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.cache.set(key, value, callbacks=callbacks)
def invalidate(self, key):
self.check_thread()
self.cache.pop(key, None)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, which will (a) stop it being returned
# for future queries and (b) stop it being persisted as a proper entry
# in self.cache.
entry = self._pending_deferred_cache.pop(key, None)
# run the invalidation callbacks now, rather than waiting for the
# deferred to resolve.
if entry:
entry.invalidate()
def invalidate_many(self, key):
self.check_thread()
if not isinstance(key, tuple):
raise TypeError(
"The cache key must be a tuple not %r" % (type(key),)
)
self.cache.del_multi(key)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, as above
entry_dict = self._pending_deferred_cache.pop(key, None)
if entry_dict is not None:
for entry in iterate_tree_cache_entry(entry_dict):
entry.invalidate()
def invalidate_all(self):
self.check_thread()
self.cache.clear()
for entry in itervalues(self._pending_deferred_cache):
entry.invalidate()
self._pending_deferred_cache.clear()
r = regex_cache.get((display_name, False, True), None)
if not r:
r1 = re.escape(display_name)
r1 = _re_word_boundary(r1)
r = re.compile(r1, flags=re.IGNORECASE)
regex_cache[(display_name, False, True)] = r
return bool(r.search(body))
def _get_value(self, dotted_key: str) -> Optional[str]:
return self._value_cache.get(dotted_key, None)
# Caches (string, is_glob, word_boundary) -> regex for push. See _glob_matches
regex_cache = LruCache(
50000,
"regex_push_cache") # type: LruCache[Tuple[str, bool, bool], Pattern]
def _glob_matches(glob: str, value: str, word_boundary: bool = False) -> bool:
"""Tests if value matches glob.
Args:
glob
value: String to test against glob.
word_boundary: Whether to match against word boundaries or entire
string. Defaults to False.
"""
try:
r = regex_cache.get((glob, True, word_boundary), None)
class Cache(object):
__slots__ = (
"cache",
"name",
"keylen",
"thread",
"metrics",
"_pending_deferred_cache",
)
def __init__(
self,
name: str,
max_entries: int = 1000,
keylen: int = 1,
tree: bool = False,
iterable: bool = False,
apply_cache_factor_from_config: bool = True,
):
"""
Args:
name: The name of the cache
max_entries: Maximum amount of entries that the cache will hold
keylen: The length of the tuple used as the cache key
tree: Use a TreeCache instead of a dict as the underlying cache type
iterable: If True, count each item in the cached object as an entry,
rather than each cached object
apply_cache_factor_from_config: Whether cache factors specified in the
config file affect `max_entries`
Returns:
Cache
"""
cache_type = TreeCache if tree else dict
self._pending_deferred_cache = cache_type()
self.cache = LruCache(
max_size=max_entries,
keylen=keylen,
cache_type=cache_type,
size_callback=(lambda d: len(d)) if iterable else None,
evicted_callback=self._on_evicted,
apply_cache_factor_from_config=apply_cache_factor_from_config,
)
self.name = name
self.keylen = keylen
self.thread = None
self.metrics = register_cache(
"cache",
name,
self.cache,
collect_callback=self._metrics_collection_callback,
)
@property
def max_entries(self):
return self.cache.max_size
def _on_evicted(self, evicted_count):
self.metrics.inc_evictions(evicted_count)
def _metrics_collection_callback(self):
cache_pending_metric.labels(self.name).set(len(self._pending_deferred_cache))
def check_thread(self):
expected_thread = self.thread
if expected_thread is None:
self.thread = threading.current_thread()
else:
if expected_thread is not threading.current_thread():
raise ValueError(
"Cache objects can only be accessed from the main thread"
)
def get(self, key, default=_CacheSentinel, callback=None, update_metrics=True):
"""Looks the key up in the caches.
Args:
key(tuple)
default: What is returned if key is not in the caches. If not
specified then function throws KeyError instead
callback(fn): Gets called when the entry in the cache is invalidated
update_metrics (bool): whether to update the cache hit rate metrics
Returns:
Either an ObservableDeferred or the raw result
"""
callbacks = [callback] if callback else []
val = self._pending_deferred_cache.get(key, _CacheSentinel)
if val is not _CacheSentinel:
val.callbacks.update(callbacks)
if update_metrics:
self.metrics.inc_hits()
return val.deferred
val = self.cache.get(key, _CacheSentinel, callbacks=callbacks)
if val is not _CacheSentinel:
self.metrics.inc_hits()
return val
if update_metrics:
self.metrics.inc_misses()
if default is _CacheSentinel:
raise KeyError()
else:
return default
def set(self, key, value, callback=None):
if not isinstance(value, defer.Deferred):
raise TypeError("not a Deferred")
callbacks = [callback] if callback else []
self.check_thread()
observable = ObservableDeferred(value, consumeErrors=True)
observer = observable.observe()
entry = CacheEntry(deferred=observable, callbacks=callbacks)
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry:
existing_entry.invalidate()
self._pending_deferred_cache[key] = entry
def compare_and_pop():
"""Check if our entry is still the one in _pending_deferred_cache, and
if so, pop it.
Returns true if the entries matched.
"""
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry is entry:
return True
# oops, the _pending_deferred_cache has been updated since
# we started our query, so we are out of date.
#
# Better put back whatever we took out. (We do it this way
# round, rather than peeking into the _pending_deferred_cache
# and then removing on a match, to make the common case faster)
if existing_entry is not None:
self._pending_deferred_cache[key] = existing_entry
return False
def cb(result):
if compare_and_pop():
self.cache.set(key, result, entry.callbacks)
else:
# we're not going to put this entry into the cache, so need
# to make sure that the invalidation callbacks are called.
# That was probably done when _pending_deferred_cache was
# updated, but it's possible that `set` was called without
# `invalidate` being previously called, in which case it may
# not have been. Either way, let's double-check now.
entry.invalidate()
def eb(_fail):
compare_and_pop()
entry.invalidate()
# once the deferred completes, we can move the entry from the
# _pending_deferred_cache to the real cache.
#
observer.addCallbacks(cb, eb)
return observable
def prefill(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.cache.set(key, value, callbacks=callbacks)
def invalidate(self, key):
self.check_thread()
self.cache.pop(key, None)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, which will (a) stop it being returned
# for future queries and (b) stop it being persisted as a proper entry
# in self.cache.
entry = self._pending_deferred_cache.pop(key, None)
# run the invalidation callbacks now, rather than waiting for the
# deferred to resolve.
if entry:
entry.invalidate()
def invalidate_many(self, key):
self.check_thread()
if not isinstance(key, tuple):
raise TypeError("The cache key must be a tuple not %r" % (type(key),))
self.cache.del_multi(key)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, as above
entry_dict = self._pending_deferred_cache.pop(key, None)
if entry_dict is not None:
for entry in iterate_tree_cache_entry(entry_dict):
entry.invalidate()
def invalidate_all(self):
self.check_thread()
self.cache.clear()
for entry in self._pending_deferred_cache.values():
entry.invalidate()
self._pending_deferred_cache.clear()
class Auth(object):
"""
FIXME: This class contains a mix of functions for authenticating users
of our client-server API and authenticating events added to room graphs.
"""
def __init__(self, hs):
self.hs = hs
self.clock = hs.get_clock()
self.store = hs.get_datastore()
self.state = hs.get_state_handler()
self.TOKEN_NOT_FOUND_HTTP_STATUS = 401
self.token_cache = LruCache(CACHE_SIZE_FACTOR * 10000)
register_cache("cache", "token_cache", self.token_cache)
@defer.inlineCallbacks
def check_from_context(self,
room_version,
event,
context,
do_sig_check=True):
prev_state_ids = yield context.get_prev_state_ids(self.store)
auth_events_ids = yield self.compute_auth_events(
event,
prev_state_ids,
for_verification=True,
)
auth_events = yield self.store.get_events(auth_events_ids)
auth_events = {(e.type, e.state_key): e
for e in itervalues(auth_events)}
self.check(
room_version,
event,
auth_events=auth_events,
do_sig_check=do_sig_check,
)
def check(self, room_version, event, auth_events, do_sig_check=True):
""" Checks if this event is correctly authed.
Args:
room_version (str): version of the room
event: the event being checked.
auth_events (dict: event-key -> event): the existing room state.
Returns:
True if the auth checks pass.
"""
with Measure(self.clock, "auth.check"):
event_auth.check(room_version,
event,
auth_events,
do_sig_check=do_sig_check)
@defer.inlineCallbacks
def check_joined_room(self, room_id, user_id, current_state=None):
"""Check if the user is currently joined in the room
Args:
room_id(str): The room to check.
user_id(str): The user to check.
current_state(dict): Optional map of the current state of the room.
If provided then that map is used to check whether they are a
member of the room. Otherwise the current membership is
loaded from the database.
Raises:
AuthError if the user is not in the room.
Returns:
A deferred membership event for the user if the user is in
the room.
"""
if current_state:
member = current_state.get((EventTypes.Member, user_id), None)
else:
member = yield self.state.get_current_state(
room_id=room_id,
event_type=EventTypes.Member,
state_key=user_id)
self._check_joined_room(member, user_id, room_id)
defer.returnValue(member)
@defer.inlineCallbacks
def check_user_was_in_room(self, room_id, user_id):
"""Check if the user was in the room at some point.
Args:
room_id(str): The room to check.
user_id(str): The user to check.
Raises:
AuthError if the user was never in the room.
Returns:
A deferred membership event for the user if the user was in the
room. This will be the join event if they are currently joined to
the room. This will be the leave event if they have left the room.
"""
member = yield self.state.get_current_state(
room_id=room_id, event_type=EventTypes.Member, state_key=user_id)
membership = member.membership if member else None
if membership not in (Membership.JOIN, Membership.LEAVE):
raise AuthError(403, "User %s not in room %s" % (user_id, room_id))
if membership == Membership.LEAVE:
forgot = yield self.store.did_forget(user_id, room_id)
if forgot:
raise AuthError(403,
"User %s not in room %s" % (user_id, room_id))
defer.returnValue(member)
@defer.inlineCallbacks
def check_host_in_room(self, room_id, host):
with Measure(self.clock, "check_host_in_room"):
latest_event_ids = yield self.store.is_host_joined(room_id, host)
defer.returnValue(latest_event_ids)
def _check_joined_room(self, member, user_id, room_id):
if not member or member.membership != Membership.JOIN:
raise AuthError(
403, "User %s not in room %s (%s)" %
(user_id, room_id, repr(member)))
def can_federate(self, event, auth_events):
creation_event = auth_events.get((EventTypes.Create, ""))
return creation_event.content.get("m.federate", True) is True
def get_public_keys(self, invite_event):
return event_auth.get_public_keys(invite_event)
@defer.inlineCallbacks
def get_user_by_req(self, request, allow_guest=False, rights="access"):
""" Get a registered user's ID.
Args:
request - An HTTP request with an access_token query parameter.
Returns:
defer.Deferred: resolves to a ``synapse.types.Requester`` object
Raises:
AuthError if no user by that token exists or the token is invalid.
"""
# Can optionally look elsewhere in the request (e.g. headers)
try:
ip_addr = self.hs.get_ip_from_request(request)
user_agent = request.requestHeaders.getRawHeaders(
b"User-Agent",
default=[b""])[0].decode('ascii', 'surrogateescape')
access_token = self.get_access_token_from_request(
request, self.TOKEN_NOT_FOUND_HTTP_STATUS)
user_id, app_service = yield self._get_appservice_user_id(request)
if user_id:
request.authenticated_entity = user_id
if ip_addr and self.hs.config.track_appservice_user_ips:
yield self.store.insert_client_ip(
user_id=user_id,
access_token=access_token,
ip=ip_addr,
user_agent=user_agent,
device_id="dummy-device", # stubbed
)
defer.returnValue(
synapse.types.create_requester(user_id,
app_service=app_service))
user_info = yield self.get_user_by_access_token(
access_token, rights)
user = user_info["user"]
token_id = user_info["token_id"]
is_guest = user_info["is_guest"]
# device_id may not be present if get_user_by_access_token has been
# stubbed out.
device_id = user_info.get("device_id")
if user and access_token and ip_addr:
yield self.store.insert_client_ip(
user_id=user.to_string(),
access_token=access_token,
ip=ip_addr,
user_agent=user_agent,
device_id=device_id,
)
if is_guest and not allow_guest:
raise AuthError(403,
"Guest access not allowed",
errcode=Codes.GUEST_ACCESS_FORBIDDEN)
request.authenticated_entity = user.to_string()
defer.returnValue(
synapse.types.create_requester(user,
token_id,
is_guest,
device_id,
app_service=app_service))
except KeyError:
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Missing access token.",
errcode=Codes.MISSING_TOKEN)
@defer.inlineCallbacks
def _get_appservice_user_id(self, request):
app_service = self.store.get_app_service_by_token(
self.get_access_token_from_request(
request, self.TOKEN_NOT_FOUND_HTTP_STATUS))
if app_service is None:
defer.returnValue((None, None))
if app_service.ip_range_whitelist:
ip_address = IPAddress(self.hs.get_ip_from_request(request))
if ip_address not in app_service.ip_range_whitelist:
defer.returnValue((None, None))
if b"user_id" not in request.args:
defer.returnValue((app_service.sender, app_service))
user_id = request.args[b"user_id"][0].decode('utf8')
if app_service.sender == user_id:
defer.returnValue((app_service.sender, app_service))
if not app_service.is_interested_in_user(user_id):
raise AuthError(
403, "Application service cannot masquerade as this user.")
if not (yield self.store.get_user_by_id(user_id)):
raise AuthError(
403, "Application service has not registered this user")
defer.returnValue((user_id, app_service))
@defer.inlineCallbacks
def get_user_by_access_token(self, token, rights="access"):
""" Validate access token and get user_id from it
Args:
token (str): The access token to get the user by.
rights (str): The operation being performed; the access token must
allow this.
Returns:
Deferred[dict]: dict that includes:
`user` (UserID)
`is_guest` (bool)
`token_id` (int|None): access token id. May be None if guest
`device_id` (str|None): device corresponding to access token
Raises:
AuthError if no user by that token exists or the token is invalid.
"""
if rights == "access":
# first look in the database
r = yield self._look_up_user_by_access_token(token)
if r:
defer.returnValue(r)
# otherwise it needs to be a valid macaroon
try:
user_id, guest = self._parse_and_validate_macaroon(token, rights)
user = UserID.from_string(user_id)
if rights == "access":
if not guest:
# non-guest access tokens must be in the database
logger.warning("Unrecognised access token - not in store.")
raise AuthError(
self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Unrecognised access token.",
errcode=Codes.UNKNOWN_TOKEN,
)
# Guest access tokens are not stored in the database (there can
# only be one access token per guest, anyway).
#
# In order to prevent guest access tokens being used as regular
# user access tokens (and hence getting around the invalidation
# process), we look up the user id and check that it is indeed
# a guest user.
#
# It would of course be much easier to store guest access
# tokens in the database as well, but that would break existing
# guest tokens.
stored_user = yield self.store.get_user_by_id(user_id)
if not stored_user:
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Unknown user_id %s" % user_id,
errcode=Codes.UNKNOWN_TOKEN)
if not stored_user["is_guest"]:
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Guest access token used for regular user",
errcode=Codes.UNKNOWN_TOKEN)
ret = {
"user": user,
"is_guest": True,
"token_id": None,
# all guests get the same device id
"device_id": GUEST_DEVICE_ID,
}
elif rights == "delete_pusher":
# We don't store these tokens in the database
ret = {
"user": user,
"is_guest": False,
"token_id": None,
"device_id": None,
}
else:
raise RuntimeError("Unknown rights setting %s", rights)
defer.returnValue(ret)
except (
_InvalidMacaroonException,
pymacaroons.exceptions.MacaroonException,
TypeError,
ValueError,
) as e:
logger.warning("Invalid macaroon in auth: %s %s", type(e), e)
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Invalid macaroon passed.",
errcode=Codes.UNKNOWN_TOKEN)
def _parse_and_validate_macaroon(self, token, rights="access"):
"""Takes a macaroon and tries to parse and validate it. This is cached
if and only if rights == access and there isn't an expiry.
On invalid macaroon raises _InvalidMacaroonException
Returns:
(user_id, is_guest)
"""
if rights == "access":
cached = self.token_cache.get(token, None)
if cached:
return cached
try:
macaroon = pymacaroons.Macaroon.deserialize(token)
except Exception: # deserialize can throw more-or-less anything
# doesn't look like a macaroon: treat it as an opaque token which
# must be in the database.
# TODO: it would be nice to get rid of this, but apparently some
# people use access tokens which aren't macaroons
raise _InvalidMacaroonException()
try:
user_id = self.get_user_id_from_macaroon(macaroon)
has_expiry = False
guest = False
for caveat in macaroon.caveats:
if caveat.caveat_id.startswith("time "):
has_expiry = True
elif caveat.caveat_id == "guest = true":
guest = True
self.validate_macaroon(
macaroon,
rights,
self.hs.config.expire_access_token,
user_id=user_id,
)
except (pymacaroons.exceptions.MacaroonException, TypeError,
ValueError):
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Invalid macaroon passed.",
errcode=Codes.UNKNOWN_TOKEN)
if not has_expiry and rights == "access":
self.token_cache[token] = (user_id, guest)
return user_id, guest
def get_user_id_from_macaroon(self, macaroon):
"""Retrieve the user_id given by the caveats on the macaroon.
Does *not* validate the macaroon.
Args:
macaroon (pymacaroons.Macaroon): The macaroon to validate
Returns:
(str) user id
Raises:
AuthError if there is no user_id caveat in the macaroon
"""
user_prefix = "user_id = "
for caveat in macaroon.caveats:
if caveat.caveat_id.startswith(user_prefix):
return caveat.caveat_id[len(user_prefix):]
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"No user caveat in macaroon",
errcode=Codes.UNKNOWN_TOKEN)
def validate_macaroon(self, macaroon, type_string, verify_expiry, user_id):
"""
validate that a Macaroon is understood by and was signed by this server.
Args:
macaroon(pymacaroons.Macaroon): The macaroon to validate
type_string(str): The kind of token required (e.g. "access",
"delete_pusher")
verify_expiry(bool): Whether to verify whether the macaroon has expired.
user_id (str): The user_id required
"""
v = pymacaroons.Verifier()
# the verifier runs a test for every caveat on the macaroon, to check
# that it is met for the current request. Each caveat must match at
# least one of the predicates specified by satisfy_exact or
# specify_general.
v.satisfy_exact("gen = 1")
v.satisfy_exact("type = " + type_string)
v.satisfy_exact("user_id = %s" % user_id)
v.satisfy_exact("guest = true")
# verify_expiry should really always be True, but there exist access
# tokens in the wild which expire when they should not, so we can't
# enforce expiry yet (so we have to allow any caveat starting with
# 'time < ' in access tokens).
#
# On the other hand, short-term login tokens (as used by CAS login, for
# example) have an expiry time which we do want to enforce.
if verify_expiry:
v.satisfy_general(self._verify_expiry)
else:
v.satisfy_general(lambda c: c.startswith("time < "))
# access_tokens include a nonce for uniqueness: any value is acceptable
v.satisfy_general(lambda c: c.startswith("nonce = "))
v.verify(macaroon, self.hs.config.macaroon_secret_key)
def _verify_expiry(self, caveat):
prefix = "time < "
if not caveat.startswith(prefix):
return False
expiry = int(caveat[len(prefix):])
now = self.hs.get_clock().time_msec()
return now < expiry
@defer.inlineCallbacks
def _look_up_user_by_access_token(self, token):
ret = yield self.store.get_user_by_access_token(token)
if not ret:
defer.returnValue(None)
# we use ret.get() below because *lots* of unit tests stub out
# get_user_by_access_token in a way where it only returns a couple of
# the fields.
user_info = {
"user": UserID.from_string(ret.get("name")),
"token_id": ret.get("token_id", None),
"is_guest": False,
"device_id": ret.get("device_id"),
}
defer.returnValue(user_info)
def get_appservice_by_req(self, request):
try:
token = self.get_access_token_from_request(
request, self.TOKEN_NOT_FOUND_HTTP_STATUS)
service = self.store.get_app_service_by_token(token)
if not service:
logger.warn("Unrecognised appservice access token.")
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Unrecognised access token.",
errcode=Codes.UNKNOWN_TOKEN)
request.authenticated_entity = service.sender
return defer.succeed(service)
except KeyError:
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Missing access token.")
def is_server_admin(self, user):
""" Check if the given user is a local server admin.
Args:
user (str): mxid of user to check
Returns:
bool: True if the user is an admin
"""
return self.store.is_server_admin(user)
@defer.inlineCallbacks
def compute_auth_events(self,
event,
current_state_ids,
for_verification=False):
if event.type == EventTypes.Create:
defer.returnValue([])
auth_ids = []
key = (
EventTypes.PowerLevels,
"",
)
power_level_event_id = current_state_ids.get(key)
if power_level_event_id:
auth_ids.append(power_level_event_id)
key = (
EventTypes.JoinRules,
"",
)
join_rule_event_id = current_state_ids.get(key)
key = (
EventTypes.Member,
event.sender,
)
member_event_id = current_state_ids.get(key)
key = (
EventTypes.Create,
"",
)
create_event_id = current_state_ids.get(key)
if create_event_id:
auth_ids.append(create_event_id)
if join_rule_event_id:
join_rule_event = yield self.store.get_event(join_rule_event_id)
join_rule = join_rule_event.content.get("join_rule")
is_public = join_rule == JoinRules.PUBLIC if join_rule else False
else:
is_public = False
if event.type == EventTypes.Member:
e_type = event.content["membership"]
if e_type in [Membership.JOIN, Membership.INVITE]:
if join_rule_event_id:
auth_ids.append(join_rule_event_id)
if e_type == Membership.JOIN:
if member_event_id and not is_public:
auth_ids.append(member_event_id)
else:
if member_event_id:
auth_ids.append(member_event_id)
if for_verification:
key = (
EventTypes.Member,
event.state_key,
)
existing_event_id = current_state_ids.get(key)
if existing_event_id:
auth_ids.append(existing_event_id)
if e_type == Membership.INVITE:
if "third_party_invite" in event.content:
key = (
EventTypes.ThirdPartyInvite,
event.content["third_party_invite"]["signed"]["token"])
third_party_invite_id = current_state_ids.get(key)
if third_party_invite_id:
auth_ids.append(third_party_invite_id)
elif member_event_id:
member_event = yield self.store.get_event(member_event_id)
if member_event.content["membership"] == Membership.JOIN:
auth_ids.append(member_event.event_id)
defer.returnValue(auth_ids)
def check_redaction(self, room_version, event, auth_events):
"""Check whether the event sender is allowed to redact the target event.
Returns:
True if the the sender is allowed to redact the target event if the
target event was created by them.
False if the sender is allowed to redact the target event with no
further checks.
Raises:
AuthError if the event sender is definitely not allowed to redact
the target event.
"""
return event_auth.check_redaction(room_version, event, auth_events)
@defer.inlineCallbacks
def check_can_change_room_list(self, room_id, user):
"""Check if the user is allowed to edit the room's entry in the
published room list.
Args:
room_id (str)
user (UserID)
"""
is_admin = yield self.is_server_admin(user)
if is_admin:
defer.returnValue(True)
user_id = user.to_string()
yield self.check_joined_room(room_id, user_id)
# We currently require the user is a "moderator" in the room. We do this
# by checking if they would (theoretically) be able to change the
# m.room.aliases events
power_level_event = yield self.state.get_current_state(
room_id, EventTypes.PowerLevels, "")
auth_events = {}
if power_level_event:
auth_events[(EventTypes.PowerLevels, "")] = power_level_event
send_level = event_auth.get_send_level(
EventTypes.Aliases,
"",
power_level_event,
)
user_level = event_auth.get_user_power_level(user_id, auth_events)
if user_level < send_level:
raise AuthError(
403,
"This server requires you to be a moderator in the room to"
" edit its room list entry")
@staticmethod
def has_access_token(request):
"""Checks if the request has an access_token.
Returns:
bool: False if no access_token was given, True otherwise.
"""
query_params = request.args.get(b"access_token")
auth_headers = request.requestHeaders.getRawHeaders(b"Authorization")
return bool(query_params) or bool(auth_headers)
@staticmethod
def get_access_token_from_request(request,
token_not_found_http_status=401):
"""Extracts the access_token from the request.
Args:
request: The http request.
token_not_found_http_status(int): The HTTP status code to set in the
AuthError if the token isn't found. This is used in some of the
legacy APIs to change the status code to 403 from the default of
401 since some of the old clients depended on auth errors returning
403.
Returns:
unicode: The access_token
Raises:
AuthError: If there isn't an access_token in the request.
"""
auth_headers = request.requestHeaders.getRawHeaders(b"Authorization")
query_params = request.args.get(b"access_token")
if auth_headers:
# Try the get the access_token from a "Authorization: Bearer"
# header
if query_params is not None:
raise AuthError(
token_not_found_http_status,
"Mixing Authorization headers and access_token query parameters.",
errcode=Codes.MISSING_TOKEN,
)
if len(auth_headers) > 1:
raise AuthError(
token_not_found_http_status,
"Too many Authorization headers.",
errcode=Codes.MISSING_TOKEN,
)
parts = auth_headers[0].split(b" ")
if parts[0] == b"Bearer" and len(parts) == 2:
return parts[1].decode('ascii')
else:
raise AuthError(
token_not_found_http_status,
"Invalid Authorization header.",
errcode=Codes.MISSING_TOKEN,
)
else:
# Try to get the access_token from the query params.
if not query_params:
raise AuthError(token_not_found_http_status,
"Missing access token.",
errcode=Codes.MISSING_TOKEN)
return query_params[0].decode('ascii')
@defer.inlineCallbacks
def check_in_room_or_world_readable(self, room_id, user_id):
"""Checks that the user is or was in the room or the room is world
readable. If it isn't then an exception is raised.
Returns:
Deferred[tuple[str, str|None]]: Resolves to the current membership of
the user in the room and the membership event ID of the user. If
the user is not in the room and never has been, then
`(Membership.JOIN, None)` is returned.
"""
try:
# check_user_was_in_room will return the most recent membership
# event for the user if:
# * The user is a non-guest user, and was ever in the room
# * The user is a guest user, and has joined the room
# else it will throw.
member_event = yield self.check_user_was_in_room(room_id, user_id)
defer.returnValue((member_event.membership, member_event.event_id))
except AuthError:
visibility = yield self.state.get_current_state(
room_id, EventTypes.RoomHistoryVisibility, "")
if (visibility and visibility.content["history_visibility"]
== "world_readable"):
defer.returnValue((Membership.JOIN, None))
return
raise AuthError(403,
"Guest access not allowed",
errcode=Codes.GUEST_ACCESS_FORBIDDEN)
@defer.inlineCallbacks
def check_auth_blocking(self, user_id=None, threepid=None):
"""Checks if the user should be rejected for some external reason,
such as monthly active user limiting or global disable flag
Args:
user_id(str|None): If present, checks for presence against existing
MAU cohort
threepid(dict|None): If present, checks for presence against configured
reserved threepid. Used in cases where the user is trying register
with a MAU blocked server, normally they would be rejected but their
threepid is on the reserved list. user_id and
threepid should never be set at the same time.
"""
# Never fail an auth check for the server notices users or support user
# This can be a problem where event creation is prohibited due to blocking
if user_id is not None:
if user_id == self.hs.config.server_notices_mxid:
return
if (yield self.store.is_support_user(user_id)):
return
if self.hs.config.hs_disabled:
raise ResourceLimitError(
403,
self.hs.config.hs_disabled_message,
errcode=Codes.RESOURCE_LIMIT_EXCEEDED,
admin_contact=self.hs.config.admin_contact,
limit_type=self.hs.config.hs_disabled_limit_type)
if self.hs.config.limit_usage_by_mau is True:
assert not (user_id and threepid)
# If the user is already part of the MAU cohort or a trial user
if user_id:
timestamp = yield self.store.user_last_seen_monthly_active(
user_id)
if timestamp:
return
is_trial = yield self.store.is_trial_user(user_id)
if is_trial:
return
elif threepid:
# If the user does not exist yet, but is signing up with a
# reserved threepid then pass auth check
if is_threepid_reserved(
self.hs.config.mau_limits_reserved_threepids,
threepid):
return
# Else if there is no room in the MAU bucket, bail
current_mau = yield self.store.get_monthly_active_count()
if current_mau >= self.hs.config.max_mau_value:
raise ResourceLimitError(
403,
"Monthly Active User Limit Exceeded",
admin_contact=self.hs.config.admin_contact,
errcode=Codes.RESOURCE_LIMIT_EXCEEDED,
limit_type="monthly_active_user")
def test_clear(self):
cache = LruCache(1)
cache["key"] = 1
cache.clear()
self.assertEquals(len(cache), 0)
def test_get_set(self):
cache = LruCache(1)
cache["key"] = "value"
self.assertEquals(cache.get("key"), "value")
self.assertEquals(cache["key"], "value")
# Similar to _glob_matches, but do not treat display_name as a glob.
r = regex_cache.get((display_name, False, True), None)
if not r:
r = re.escape(display_name)
r = _re_word_boundary(r)
r = re.compile(r, flags=re.IGNORECASE)
regex_cache[(display_name, False, True)] = r
return r.search(body)
def _get_value(self, dotted_key: str) -> str:
return self._value_cache.get(dotted_key, None)
# Caches (string, is_glob, word_boundary) -> regex for push. See _glob_matches
regex_cache = LruCache(50000)
register_cache("cache", "regex_push_cache", regex_cache)
def _glob_matches(glob: str, value: str, word_boundary: bool = False) -> bool:
"""Tests if value matches glob.
Args:
glob
value: String to test against glob.
word_boundary: Whether to match against word boundaries or entire
string. Defaults to False.
"""
try:
r = regex_cache.get((glob, True, word_boundary), None)
class Auth(object):
"""
FIXME: This class contains a mix of functions for authenticating users
of our client-server API and authenticating events added to room graphs.
"""
def __init__(self, hs):
self.hs = hs
self.clock = hs.get_clock()
self.store = hs.get_datastore()
self.state = hs.get_state_handler()
self.TOKEN_NOT_FOUND_HTTP_STATUS = 401
self.token_cache = LruCache(CACHE_SIZE_FACTOR * 10000)
register_cache("cache", "token_cache", self.token_cache)
self._account_validity = hs.config.account_validity
@defer.inlineCallbacks
def check_from_context(self, room_version, event, context, do_sig_check=True):
prev_state_ids = yield context.get_prev_state_ids(self.store)
auth_events_ids = yield self.compute_auth_events(
event, prev_state_ids, for_verification=True,
)
auth_events = yield self.store.get_events(auth_events_ids)
auth_events = {
(e.type, e.state_key): e for e in itervalues(auth_events)
}
self.check(
room_version, event,
auth_events=auth_events, do_sig_check=do_sig_check,
)
def check(self, room_version, event, auth_events, do_sig_check=True):
""" Checks if this event is correctly authed.
Args:
room_version (str): version of the room
event: the event being checked.
auth_events (dict: event-key -> event): the existing room state.
Returns:
True if the auth checks pass.
"""
with Measure(self.clock, "auth.check"):
event_auth.check(
room_version, event, auth_events, do_sig_check=do_sig_check
)
@defer.inlineCallbacks
def check_joined_room(self, room_id, user_id, current_state=None):
"""Check if the user is currently joined in the room
Args:
room_id(str): The room to check.
user_id(str): The user to check.
current_state(dict): Optional map of the current state of the room.
If provided then that map is used to check whether they are a
member of the room. Otherwise the current membership is
loaded from the database.
Raises:
AuthError if the user is not in the room.
Returns:
A deferred membership event for the user if the user is in
the room.
"""
if current_state:
member = current_state.get(
(EventTypes.Member, user_id),
None
)
else:
member = yield self.state.get_current_state(
room_id=room_id,
event_type=EventTypes.Member,
state_key=user_id
)
self._check_joined_room(member, user_id, room_id)
defer.returnValue(member)
@defer.inlineCallbacks
def check_user_was_in_room(self, room_id, user_id):
"""Check if the user was in the room at some point.
Args:
room_id(str): The room to check.
user_id(str): The user to check.
Raises:
AuthError if the user was never in the room.
Returns:
A deferred membership event for the user if the user was in the
room. This will be the join event if they are currently joined to
the room. This will be the leave event if they have left the room.
"""
member = yield self.state.get_current_state(
room_id=room_id,
event_type=EventTypes.Member,
state_key=user_id
)
membership = member.membership if member else None
if membership not in (Membership.JOIN, Membership.LEAVE):
raise AuthError(403, "User %s not in room %s" % (
user_id, room_id
))
if membership == Membership.LEAVE:
forgot = yield self.store.did_forget(user_id, room_id)
if forgot:
raise AuthError(403, "User %s not in room %s" % (
user_id, room_id
))
defer.returnValue(member)
@defer.inlineCallbacks
def check_host_in_room(self, room_id, host):
with Measure(self.clock, "check_host_in_room"):
latest_event_ids = yield self.store.is_host_joined(room_id, host)
defer.returnValue(latest_event_ids)
def _check_joined_room(self, member, user_id, room_id):
if not member or member.membership != Membership.JOIN:
raise AuthError(403, "User %s not in room %s (%s)" % (
user_id, room_id, repr(member)
))
def can_federate(self, event, auth_events):
creation_event = auth_events.get((EventTypes.Create, ""))
return creation_event.content.get("m.federate", True) is True
def get_public_keys(self, invite_event):
return event_auth.get_public_keys(invite_event)
@defer.inlineCallbacks
def get_user_by_req(self, request, allow_guest=False, rights="access"):
""" Get a registered user's ID.
Args:
request - An HTTP request with an access_token query parameter.
Returns:
defer.Deferred: resolves to a ``synapse.types.Requester`` object
Raises:
AuthError if no user by that token exists or the token is invalid.
"""
# Can optionally look elsewhere in the request (e.g. headers)
try:
ip_addr = self.hs.get_ip_from_request(request)
user_agent = request.requestHeaders.getRawHeaders(
b"User-Agent",
default=[b""]
)[0].decode('ascii', 'surrogateescape')
access_token = self.get_access_token_from_request(
request, self.TOKEN_NOT_FOUND_HTTP_STATUS
)
user_id, app_service = yield self._get_appservice_user_id(request)
if user_id:
request.authenticated_entity = user_id
if ip_addr and self.hs.config.track_appservice_user_ips:
yield self.store.insert_client_ip(
user_id=user_id,
access_token=access_token,
ip=ip_addr,
user_agent=user_agent,
device_id="dummy-device", # stubbed
)
defer.returnValue(
synapse.types.create_requester(user_id, app_service=app_service)
)
user_info = yield self.get_user_by_access_token(access_token, rights)
user = user_info["user"]
token_id = user_info["token_id"]
is_guest = user_info["is_guest"]
# Deny the request if the user account has expired.
if self._account_validity.enabled:
user_id = user.to_string()
expiration_ts = yield self.store.get_expiration_ts_for_user(user_id)
if expiration_ts is not None and self.clock.time_msec() >= expiration_ts:
raise AuthError(
403,
"User account has expired",
errcode=Codes.EXPIRED_ACCOUNT,
)
# device_id may not be present if get_user_by_access_token has been
# stubbed out.
device_id = user_info.get("device_id")
if user and access_token and ip_addr:
yield self.store.insert_client_ip(
user_id=user.to_string(),
access_token=access_token,
ip=ip_addr,
user_agent=user_agent,
device_id=device_id,
)
if is_guest and not allow_guest:
raise AuthError(
403, "Guest access not allowed", errcode=Codes.GUEST_ACCESS_FORBIDDEN
)
request.authenticated_entity = user.to_string()
defer.returnValue(synapse.types.create_requester(
user, token_id, is_guest, device_id, app_service=app_service)
)
except KeyError:
raise AuthError(
self.TOKEN_NOT_FOUND_HTTP_STATUS, "Missing access token.",
errcode=Codes.MISSING_TOKEN
)
@defer.inlineCallbacks
def _get_appservice_user_id(self, request):
app_service = self.store.get_app_service_by_token(
self.get_access_token_from_request(
request, self.TOKEN_NOT_FOUND_HTTP_STATUS
)
)
if app_service is None:
defer.returnValue((None, None))
if app_service.ip_range_whitelist:
ip_address = IPAddress(self.hs.get_ip_from_request(request))
if ip_address not in app_service.ip_range_whitelist:
defer.returnValue((None, None))
if b"user_id" not in request.args:
defer.returnValue((app_service.sender, app_service))
user_id = request.args[b"user_id"][0].decode('utf8')
if app_service.sender == user_id:
defer.returnValue((app_service.sender, app_service))
if not app_service.is_interested_in_user(user_id):
raise AuthError(
403,
"Application service cannot masquerade as this user."
)
if not (yield self.store.get_user_by_id(user_id)):
raise AuthError(
403,
"Application service has not registered this user"
)
defer.returnValue((user_id, app_service))
@defer.inlineCallbacks
def get_user_by_access_token(self, token, rights="access"):
""" Validate access token and get user_id from it
Args:
token (str): The access token to get the user by.
rights (str): The operation being performed; the access token must
allow this.
Returns:
Deferred[dict]: dict that includes:
`user` (UserID)
`is_guest` (bool)
`token_id` (int|None): access token id. May be None if guest
`device_id` (str|None): device corresponding to access token
Raises:
AuthError if no user by that token exists or the token is invalid.
"""
if rights == "access":
# first look in the database
r = yield self._look_up_user_by_access_token(token)
if r:
defer.returnValue(r)
# otherwise it needs to be a valid macaroon
try:
user_id, guest = self._parse_and_validate_macaroon(token, rights)
user = UserID.from_string(user_id)
if rights == "access":
if not guest:
# non-guest access tokens must be in the database
logger.warning("Unrecognised access token - not in store.")
raise AuthError(
self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Unrecognised access token.",
errcode=Codes.UNKNOWN_TOKEN,
)
# Guest access tokens are not stored in the database (there can
# only be one access token per guest, anyway).
#
# In order to prevent guest access tokens being used as regular
# user access tokens (and hence getting around the invalidation
# process), we look up the user id and check that it is indeed
# a guest user.
#
# It would of course be much easier to store guest access
# tokens in the database as well, but that would break existing
# guest tokens.
stored_user = yield self.store.get_user_by_id(user_id)
if not stored_user:
raise AuthError(
self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Unknown user_id %s" % user_id,
errcode=Codes.UNKNOWN_TOKEN
)
if not stored_user["is_guest"]:
raise AuthError(
self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Guest access token used for regular user",
errcode=Codes.UNKNOWN_TOKEN
)
ret = {
"user": user,
"is_guest": True,
"token_id": None,
# all guests get the same device id
"device_id": GUEST_DEVICE_ID,
}
elif rights == "delete_pusher":
# We don't store these tokens in the database
ret = {
"user": user,
"is_guest": False,
"token_id": None,
"device_id": None,
}
else:
raise RuntimeError("Unknown rights setting %s", rights)
defer.returnValue(ret)
except (
_InvalidMacaroonException,
pymacaroons.exceptions.MacaroonException,
TypeError,
ValueError,
) as e:
logger.warning("Invalid macaroon in auth: %s %s", type(e), e)
raise AuthError(
self.TOKEN_NOT_FOUND_HTTP_STATUS, "Invalid macaroon passed.",
errcode=Codes.UNKNOWN_TOKEN
)
def _parse_and_validate_macaroon(self, token, rights="access"):
"""Takes a macaroon and tries to parse and validate it. This is cached
if and only if rights == access and there isn't an expiry.
On invalid macaroon raises _InvalidMacaroonException
Returns:
(user_id, is_guest)
"""
if rights == "access":
cached = self.token_cache.get(token, None)
if cached:
return cached
try:
macaroon = pymacaroons.Macaroon.deserialize(token)
except Exception: # deserialize can throw more-or-less anything
# doesn't look like a macaroon: treat it as an opaque token which
# must be in the database.
# TODO: it would be nice to get rid of this, but apparently some
# people use access tokens which aren't macaroons
raise _InvalidMacaroonException()
try:
user_id = self.get_user_id_from_macaroon(macaroon)
has_expiry = False
guest = False
for caveat in macaroon.caveats:
if caveat.caveat_id.startswith("time "):
has_expiry = True
elif caveat.caveat_id == "guest = true":
guest = True
self.validate_macaroon(
macaroon, rights, self.hs.config.expire_access_token,
user_id=user_id,
)
except (pymacaroons.exceptions.MacaroonException, TypeError, ValueError):
raise AuthError(
self.TOKEN_NOT_FOUND_HTTP_STATUS, "Invalid macaroon passed.",
errcode=Codes.UNKNOWN_TOKEN
)
if not has_expiry and rights == "access":
self.token_cache[token] = (user_id, guest)
return user_id, guest
def get_user_id_from_macaroon(self, macaroon):
"""Retrieve the user_id given by the caveats on the macaroon.
Does *not* validate the macaroon.
Args:
macaroon (pymacaroons.Macaroon): The macaroon to validate
Returns:
(str) user id
Raises:
AuthError if there is no user_id caveat in the macaroon
"""
user_prefix = "user_id = "
for caveat in macaroon.caveats:
if caveat.caveat_id.startswith(user_prefix):
return caveat.caveat_id[len(user_prefix):]
raise AuthError(
self.TOKEN_NOT_FOUND_HTTP_STATUS, "No user caveat in macaroon",
errcode=Codes.UNKNOWN_TOKEN
)
def validate_macaroon(self, macaroon, type_string, verify_expiry, user_id):
"""
validate that a Macaroon is understood by and was signed by this server.
Args:
macaroon(pymacaroons.Macaroon): The macaroon to validate
type_string(str): The kind of token required (e.g. "access",
"delete_pusher")
verify_expiry(bool): Whether to verify whether the macaroon has expired.
user_id (str): The user_id required
"""
v = pymacaroons.Verifier()
# the verifier runs a test for every caveat on the macaroon, to check
# that it is met for the current request. Each caveat must match at
# least one of the predicates specified by satisfy_exact or
# specify_general.
v.satisfy_exact("gen = 1")
v.satisfy_exact("type = " + type_string)
v.satisfy_exact("user_id = %s" % user_id)
v.satisfy_exact("guest = true")
# verify_expiry should really always be True, but there exist access
# tokens in the wild which expire when they should not, so we can't
# enforce expiry yet (so we have to allow any caveat starting with
# 'time < ' in access tokens).
#
# On the other hand, short-term login tokens (as used by CAS login, for
# example) have an expiry time which we do want to enforce.
if verify_expiry:
v.satisfy_general(self._verify_expiry)
else:
v.satisfy_general(lambda c: c.startswith("time < "))
# access_tokens include a nonce for uniqueness: any value is acceptable
v.satisfy_general(lambda c: c.startswith("nonce = "))
v.verify(macaroon, self.hs.config.macaroon_secret_key)
def _verify_expiry(self, caveat):
prefix = "time < "
if not caveat.startswith(prefix):
return False
expiry = int(caveat[len(prefix):])
now = self.hs.get_clock().time_msec()
return now < expiry
@defer.inlineCallbacks
def _look_up_user_by_access_token(self, token):
ret = yield self.store.get_user_by_access_token(token)
if not ret:
defer.returnValue(None)
# we use ret.get() below because *lots* of unit tests stub out
# get_user_by_access_token in a way where it only returns a couple of
# the fields.
user_info = {
"user": UserID.from_string(ret.get("name")),
"token_id": ret.get("token_id", None),
"is_guest": False,
"device_id": ret.get("device_id"),
}
defer.returnValue(user_info)
def get_appservice_by_req(self, request):
try:
token = self.get_access_token_from_request(
request, self.TOKEN_NOT_FOUND_HTTP_STATUS
)
service = self.store.get_app_service_by_token(token)
if not service:
logger.warn("Unrecognised appservice access token.")
raise AuthError(
self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Unrecognised access token.",
errcode=Codes.UNKNOWN_TOKEN
)
request.authenticated_entity = service.sender
return defer.succeed(service)
except KeyError:
raise AuthError(
self.TOKEN_NOT_FOUND_HTTP_STATUS, "Missing access token."
)
def is_server_admin(self, user):
""" Check if the given user is a local server admin.
Args:
user (UserID): user to check
Returns:
bool: True if the user is an admin
"""
return self.store.is_server_admin(user)
@defer.inlineCallbacks
def compute_auth_events(self, event, current_state_ids, for_verification=False):
if event.type == EventTypes.Create:
defer.returnValue([])
auth_ids = []
key = (EventTypes.PowerLevels, "", )
power_level_event_id = current_state_ids.get(key)
if power_level_event_id:
auth_ids.append(power_level_event_id)
key = (EventTypes.JoinRules, "", )
join_rule_event_id = current_state_ids.get(key)
key = (EventTypes.Member, event.sender, )
member_event_id = current_state_ids.get(key)
key = (EventTypes.Create, "", )
create_event_id = current_state_ids.get(key)
if create_event_id:
auth_ids.append(create_event_id)
if join_rule_event_id:
join_rule_event = yield self.store.get_event(join_rule_event_id)
join_rule = join_rule_event.content.get("join_rule")
is_public = join_rule == JoinRules.PUBLIC if join_rule else False
else:
is_public = False
if event.type == EventTypes.Member:
e_type = event.content["membership"]
if e_type in [Membership.JOIN, Membership.INVITE]:
if join_rule_event_id:
auth_ids.append(join_rule_event_id)
if e_type == Membership.JOIN:
if member_event_id and not is_public:
auth_ids.append(member_event_id)
else:
if member_event_id:
auth_ids.append(member_event_id)
if for_verification:
key = (EventTypes.Member, event.state_key, )
existing_event_id = current_state_ids.get(key)
if existing_event_id:
auth_ids.append(existing_event_id)
if e_type == Membership.INVITE:
if "third_party_invite" in event.content:
key = (
EventTypes.ThirdPartyInvite,
event.content["third_party_invite"]["signed"]["token"]
)
third_party_invite_id = current_state_ids.get(key)
if third_party_invite_id:
auth_ids.append(third_party_invite_id)
elif member_event_id:
member_event = yield self.store.get_event(member_event_id)
if member_event.content["membership"] == Membership.JOIN:
auth_ids.append(member_event.event_id)
defer.returnValue(auth_ids)
def check_redaction(self, room_version, event, auth_events):
"""Check whether the event sender is allowed to redact the target event.
Returns:
True if the the sender is allowed to redact the target event if the
target event was created by them.
False if the sender is allowed to redact the target event with no
further checks.
Raises:
AuthError if the event sender is definitely not allowed to redact
the target event.
"""
return event_auth.check_redaction(room_version, event, auth_events)
@defer.inlineCallbacks
def check_can_change_room_list(self, room_id, user):
"""Check if the user is allowed to edit the room's entry in the
published room list.
Args:
room_id (str)
user (UserID)
"""
is_admin = yield self.is_server_admin(user)
if is_admin:
defer.returnValue(True)
user_id = user.to_string()
yield self.check_joined_room(room_id, user_id)
# We currently require the user is a "moderator" in the room. We do this
# by checking if they would (theoretically) be able to change the
# m.room.aliases events
power_level_event = yield self.state.get_current_state(
room_id, EventTypes.PowerLevels, ""
)
auth_events = {}
if power_level_event:
auth_events[(EventTypes.PowerLevels, "")] = power_level_event
send_level = event_auth.get_send_level(
EventTypes.Aliases, "", power_level_event,
)
user_level = event_auth.get_user_power_level(user_id, auth_events)
if user_level < send_level:
raise AuthError(
403,
"This server requires you to be a moderator in the room to"
" edit its room list entry"
)
@staticmethod
def has_access_token(request):
"""Checks if the request has an access_token.
Returns:
bool: False if no access_token was given, True otherwise.
"""
query_params = request.args.get(b"access_token")
auth_headers = request.requestHeaders.getRawHeaders(b"Authorization")
return bool(query_params) or bool(auth_headers)
@staticmethod
def get_access_token_from_request(request, token_not_found_http_status=401):
"""Extracts the access_token from the request.
Args:
request: The http request.
token_not_found_http_status(int): The HTTP status code to set in the
AuthError if the token isn't found. This is used in some of the
legacy APIs to change the status code to 403 from the default of
401 since some of the old clients depended on auth errors returning
403.
Returns:
unicode: The access_token
Raises:
AuthError: If there isn't an access_token in the request.
"""
auth_headers = request.requestHeaders.getRawHeaders(b"Authorization")
query_params = request.args.get(b"access_token")
if auth_headers:
# Try the get the access_token from a "Authorization: Bearer"
# header
if query_params is not None:
raise AuthError(
token_not_found_http_status,
"Mixing Authorization headers and access_token query parameters.",
errcode=Codes.MISSING_TOKEN,
)
if len(auth_headers) > 1:
raise AuthError(
token_not_found_http_status,
"Too many Authorization headers.",
errcode=Codes.MISSING_TOKEN,
)
parts = auth_headers[0].split(b" ")
if parts[0] == b"Bearer" and len(parts) == 2:
return parts[1].decode('ascii')
else:
raise AuthError(
token_not_found_http_status,
"Invalid Authorization header.",
errcode=Codes.MISSING_TOKEN,
)
else:
# Try to get the access_token from the query params.
if not query_params:
raise AuthError(
token_not_found_http_status,
"Missing access token.",
errcode=Codes.MISSING_TOKEN
)
return query_params[0].decode('ascii')
@defer.inlineCallbacks
def check_in_room_or_world_readable(self, room_id, user_id):
"""Checks that the user is or was in the room or the room is world
readable. If it isn't then an exception is raised.
Returns:
Deferred[tuple[str, str|None]]: Resolves to the current membership of
the user in the room and the membership event ID of the user. If
the user is not in the room and never has been, then
`(Membership.JOIN, None)` is returned.
"""
try:
# check_user_was_in_room will return the most recent membership
# event for the user if:
# * The user is a non-guest user, and was ever in the room
# * The user is a guest user, and has joined the room
# else it will throw.
member_event = yield self.check_user_was_in_room(room_id, user_id)
defer.returnValue((member_event.membership, member_event.event_id))
except AuthError:
visibility = yield self.state.get_current_state(
room_id, EventTypes.RoomHistoryVisibility, ""
)
if (
visibility and
visibility.content["history_visibility"] == "world_readable"
):
defer.returnValue((Membership.JOIN, None))
return
raise AuthError(
403, "Guest access not allowed", errcode=Codes.GUEST_ACCESS_FORBIDDEN
)
@defer.inlineCallbacks
def check_auth_blocking(self, user_id=None, threepid=None):
"""Checks if the user should be rejected for some external reason,
such as monthly active user limiting or global disable flag
Args:
user_id(str|None): If present, checks for presence against existing
MAU cohort
threepid(dict|None): If present, checks for presence against configured
reserved threepid. Used in cases where the user is trying register
with a MAU blocked server, normally they would be rejected but their
threepid is on the reserved list. user_id and
threepid should never be set at the same time.
"""
# Never fail an auth check for the server notices users or support user
# This can be a problem where event creation is prohibited due to blocking
if user_id is not None:
if user_id == self.hs.config.server_notices_mxid:
return
if (yield self.store.is_support_user(user_id)):
return
if self.hs.config.hs_disabled:
raise ResourceLimitError(
403, self.hs.config.hs_disabled_message,
errcode=Codes.RESOURCE_LIMIT_EXCEEDED,
admin_contact=self.hs.config.admin_contact,
limit_type=self.hs.config.hs_disabled_limit_type
)
if self.hs.config.limit_usage_by_mau is True:
assert not (user_id and threepid)
# If the user is already part of the MAU cohort or a trial user
if user_id:
timestamp = yield self.store.user_last_seen_monthly_active(user_id)
if timestamp:
return
is_trial = yield self.store.is_trial_user(user_id)
if is_trial:
return
elif threepid:
# If the user does not exist yet, but is signing up with a
# reserved threepid then pass auth check
if is_threepid_reserved(
self.hs.config.mau_limits_reserved_threepids, threepid
):
return
# Else if there is no room in the MAU bucket, bail
current_mau = yield self.store.get_monthly_active_count()
if current_mau >= self.hs.config.max_mau_value:
raise ResourceLimitError(
403, "Monthly Active User Limit Exceeded",
admin_contact=self.hs.config.admin_contact,
errcode=Codes.RESOURCE_LIMIT_EXCEEDED,
limit_type="monthly_active_user"
)
class Auth(object):
"""
FIXME: This class contains a mix of functions for authenticating users
of our client-server API and authenticating events added to room graphs.
"""
def __init__(self, hs):
self.hs = hs
self.clock = hs.get_clock()
self.store = hs.get_datastore()
self.state = hs.get_state_handler()
self.TOKEN_NOT_FOUND_HTTP_STATUS = 401
self.token_cache = LruCache(CACHE_SIZE_FACTOR * 10000)
register_cache("token_cache", self.token_cache)
@defer.inlineCallbacks
def check_from_context(self, event, context, do_sig_check=True):
auth_events_ids = yield self.compute_auth_events(
event,
context.prev_state_ids,
for_verification=True,
)
auth_events = yield self.store.get_events(auth_events_ids)
auth_events = {(e.type, e.state_key): e for e in auth_events.values()}
self.check(event, auth_events=auth_events, do_sig_check=do_sig_check)
def check(self, event, auth_events, do_sig_check=True):
""" Checks if this event is correctly authed.
Args:
event: the event being checked.
auth_events (dict: event-key -> event): the existing room state.
Returns:
True if the auth checks pass.
"""
with Measure(self.clock, "auth.check"):
event_auth.check(event, auth_events, do_sig_check=do_sig_check)
@defer.inlineCallbacks
def check_joined_room(self, room_id, user_id, current_state=None):
"""Check if the user is currently joined in the room
Args:
room_id(str): The room to check.
user_id(str): The user to check.
current_state(dict): Optional map of the current state of the room.
If provided then that map is used to check whether they are a
member of the room. Otherwise the current membership is
loaded from the database.
Raises:
AuthError if the user is not in the room.
Returns:
A deferred membership event for the user if the user is in
the room.
"""
if current_state:
member = current_state.get((EventTypes.Member, user_id), None)
else:
member = yield self.state.get_current_state(
room_id=room_id,
event_type=EventTypes.Member,
state_key=user_id)
self._check_joined_room(member, user_id, room_id)
defer.returnValue(member)
@defer.inlineCallbacks
def check_user_was_in_room(self, room_id, user_id):
"""Check if the user was in the room at some point.
Args:
room_id(str): The room to check.
user_id(str): The user to check.
Raises:
AuthError if the user was never in the room.
Returns:
A deferred membership event for the user if the user was in the
room. This will be the join event if they are currently joined to
the room. This will be the leave event if they have left the room.
"""
member = yield self.state.get_current_state(
room_id=room_id, event_type=EventTypes.Member, state_key=user_id)
membership = member.membership if member else None
if membership not in (Membership.JOIN, Membership.LEAVE):
raise AuthError(403, "User %s not in room %s" % (user_id, room_id))
if membership == Membership.LEAVE:
forgot = yield self.store.did_forget(user_id, room_id)
if forgot:
raise AuthError(403,
"User %s not in room %s" % (user_id, room_id))
defer.returnValue(member)
@defer.inlineCallbacks
def check_host_in_room(self, room_id, host):
with Measure(self.clock, "check_host_in_room"):
latest_event_ids = yield self.store.is_host_joined(room_id, host)
defer.returnValue(latest_event_ids)
def _check_joined_room(self, member, user_id, room_id):
if not member or member.membership != Membership.JOIN:
raise AuthError(
403, "User %s not in room %s (%s)" %
(user_id, room_id, repr(member)))
def can_federate(self, event, auth_events):
creation_event = auth_events.get((EventTypes.Create, ""))
return creation_event.content.get("m.federate", True) is True
def get_public_keys(self, invite_event):
return event_auth.get_public_keys(invite_event)
@defer.inlineCallbacks
def get_user_by_req(self, request, allow_guest=False, rights="access"):
""" Get a registered user's ID.
Args:
request - An HTTP request with an access_token query parameter.
Returns:
defer.Deferred: resolves to a ``synapse.types.Requester`` object
Raises:
AuthError if no user by that token exists or the token is invalid.
"""
# Can optionally look elsewhere in the request (e.g. headers)
try:
user_id, app_service = yield self._get_appservice_user_id(request)
if user_id:
request.authenticated_entity = user_id
defer.returnValue(
synapse.types.create_requester(user_id,
app_service=app_service))
access_token = get_access_token_from_request(
request, self.TOKEN_NOT_FOUND_HTTP_STATUS)
user_info = yield self.get_user_by_access_token(
access_token, rights)
user = user_info["user"]
token_id = user_info["token_id"]
is_guest = user_info["is_guest"]
# device_id may not be present if get_user_by_access_token has been
# stubbed out.
device_id = user_info.get("device_id")
ip_addr = self.hs.get_ip_from_request(request)
user_agent = request.requestHeaders.getRawHeaders("User-Agent",
default=[""])[0]
if user and access_token and ip_addr:
self.store.insert_client_ip(
user_id=user.to_string(),
access_token=access_token,
ip=ip_addr,
user_agent=user_agent,
device_id=device_id,
)
if is_guest and not allow_guest:
raise AuthError(403,
"Guest access not allowed",
errcode=Codes.GUEST_ACCESS_FORBIDDEN)
request.authenticated_entity = user.to_string()
defer.returnValue(
synapse.types.create_requester(user,
token_id,
is_guest,
device_id,
app_service=app_service))
except KeyError:
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Missing access token.",
errcode=Codes.MISSING_TOKEN)
@defer.inlineCallbacks
def _get_appservice_user_id(self, request):
app_service = self.store.get_app_service_by_token(
get_access_token_from_request(request,
self.TOKEN_NOT_FOUND_HTTP_STATUS))
if app_service is None:
defer.returnValue((None, None))
if "user_id" not in request.args:
defer.returnValue((app_service.sender, app_service))
user_id = request.args["user_id"][0]
if app_service.sender == user_id:
defer.returnValue((app_service.sender, app_service))
if not app_service.is_interested_in_user(user_id):
raise AuthError(
403, "Application service cannot masquerade as this user.")
if not (yield self.store.get_user_by_id(user_id)):
raise AuthError(
403, "Application service has not registered this user")
defer.returnValue((user_id, app_service))
@defer.inlineCallbacks
def get_user_by_access_token(self, token, rights="access"):
""" Validate access token and get user_id from it
Args:
token (str): The access token to get the user by.
rights (str): The operation being performed; the access token must
allow this.
Returns:
Deferred[dict]: dict that includes:
`user` (UserID)
`is_guest` (bool)
`token_id` (int|None): access token id. May be None if guest
`device_id` (str|None): device corresponding to access token
Raises:
AuthError if no user by that token exists or the token is invalid.
"""
try:
user_id, guest = self._parse_and_validate_macaroon(token, rights)
except _InvalidMacaroonException:
# doesn't look like a macaroon: treat it as an opaque token which
# must be in the database.
# TODO: it would be nice to get rid of this, but apparently some
# people use access tokens which aren't macaroons
r = yield self._look_up_user_by_access_token(token)
defer.returnValue(r)
try:
user = UserID.from_string(user_id)
if guest:
# Guest access tokens are not stored in the database (there can
# only be one access token per guest, anyway).
#
# In order to prevent guest access tokens being used as regular
# user access tokens (and hence getting around the invalidation
# process), we look up the user id and check that it is indeed
# a guest user.
#
# It would of course be much easier to store guest access
# tokens in the database as well, but that would break existing
# guest tokens.
stored_user = yield self.store.get_user_by_id(user_id)
if not stored_user:
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Unknown user_id %s" % user_id,
errcode=Codes.UNKNOWN_TOKEN)
if not stored_user["is_guest"]:
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Guest access token used for regular user",
errcode=Codes.UNKNOWN_TOKEN)
ret = {
"user": user,
"is_guest": True,
"token_id": None,
# all guests get the same device id
"device_id": GUEST_DEVICE_ID,
}
elif rights == "delete_pusher":
# We don't store these tokens in the database
ret = {
"user": user,
"is_guest": False,
"token_id": None,
"device_id": None,
}
else:
# This codepath exists for several reasons:
# * so that we can actually return a token ID, which is used
# in some parts of the schema (where we probably ought to
# use device IDs instead)
# * the only way we currently have to invalidate an
# access_token is by removing it from the database, so we
# have to check here that it is still in the db
# * some attributes (notably device_id) aren't stored in the
# macaroon. They probably should be.
# TODO: build the dictionary from the macaroon once the
# above are fixed
ret = yield self._look_up_user_by_access_token(token)
if ret["user"] != user:
logger.error("Macaroon user (%s) != DB user (%s)", user,
ret["user"])
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"User mismatch in macaroon",
errcode=Codes.UNKNOWN_TOKEN)
defer.returnValue(ret)
except (pymacaroons.exceptions.MacaroonException, TypeError,
ValueError):
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Invalid macaroon passed.",
errcode=Codes.UNKNOWN_TOKEN)
def _parse_and_validate_macaroon(self, token, rights="access"):
"""Takes a macaroon and tries to parse and validate it. This is cached
if and only if rights == access and there isn't an expiry.
On invalid macaroon raises _InvalidMacaroonException
Returns:
(user_id, is_guest)
"""
if rights == "access":
cached = self.token_cache.get(token, None)
if cached:
return cached
try:
macaroon = pymacaroons.Macaroon.deserialize(token)
except Exception: # deserialize can throw more-or-less anything
# doesn't look like a macaroon: treat it as an opaque token which
# must be in the database.
# TODO: it would be nice to get rid of this, but apparently some
# people use access tokens which aren't macaroons
raise _InvalidMacaroonException()
try:
user_id = self.get_user_id_from_macaroon(macaroon)
has_expiry = False
guest = False
for caveat in macaroon.caveats:
if caveat.caveat_id.startswith("time "):
has_expiry = True
elif caveat.caveat_id == "guest = true":
guest = True
self.validate_macaroon(
macaroon,
rights,
self.hs.config.expire_access_token,
user_id=user_id,
)
except (pymacaroons.exceptions.MacaroonException, TypeError,
ValueError):
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Invalid macaroon passed.",
errcode=Codes.UNKNOWN_TOKEN)
if not has_expiry and rights == "access":
self.token_cache[token] = (user_id, guest)
return user_id, guest
def get_user_id_from_macaroon(self, macaroon):
"""Retrieve the user_id given by the caveats on the macaroon.
Does *not* validate the macaroon.
Args:
macaroon (pymacaroons.Macaroon): The macaroon to validate
Returns:
(str) user id
Raises:
AuthError if there is no user_id caveat in the macaroon
"""
user_prefix = "user_id = "
for caveat in macaroon.caveats:
if caveat.caveat_id.startswith(user_prefix):
return caveat.caveat_id[len(user_prefix):]
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"No user caveat in macaroon",
errcode=Codes.UNKNOWN_TOKEN)
def validate_macaroon(self, macaroon, type_string, verify_expiry, user_id):
"""
validate that a Macaroon is understood by and was signed by this server.
Args:
macaroon(pymacaroons.Macaroon): The macaroon to validate
type_string(str): The kind of token required (e.g. "access",
"delete_pusher")
verify_expiry(bool): Whether to verify whether the macaroon has expired.
user_id (str): The user_id required
"""
v = pymacaroons.Verifier()
# the verifier runs a test for every caveat on the macaroon, to check
# that it is met for the current request. Each caveat must match at
# least one of the predicates specified by satisfy_exact or
# specify_general.
v.satisfy_exact("gen = 1")
v.satisfy_exact("type = " + type_string)
v.satisfy_exact("user_id = %s" % user_id)
v.satisfy_exact("guest = true")
# verify_expiry should really always be True, but there exist access
# tokens in the wild which expire when they should not, so we can't
# enforce expiry yet (so we have to allow any caveat starting with
# 'time < ' in access tokens).
#
# On the other hand, short-term login tokens (as used by CAS login, for
# example) have an expiry time which we do want to enforce.
if verify_expiry:
v.satisfy_general(self._verify_expiry)
else:
v.satisfy_general(lambda c: c.startswith("time < "))
# access_tokens include a nonce for uniqueness: any value is acceptable
v.satisfy_general(lambda c: c.startswith("nonce = "))
v.verify(macaroon, self.hs.config.macaroon_secret_key)
def _verify_expiry(self, caveat):
prefix = "time < "
if not caveat.startswith(prefix):
return False
expiry = int(caveat[len(prefix):])
now = self.hs.get_clock().time_msec()
return now < expiry
@defer.inlineCallbacks
def _look_up_user_by_access_token(self, token):
ret = yield self.store.get_user_by_access_token(token)
if not ret:
logger.warn("Unrecognised access token - not in store: %s" %
(token, ))
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Unrecognised access token.",
errcode=Codes.UNKNOWN_TOKEN)
# we use ret.get() below because *lots* of unit tests stub out
# get_user_by_access_token in a way where it only returns a couple of
# the fields.
user_info = {
"user": UserID.from_string(ret.get("name")),
"token_id": ret.get("token_id", None),
"is_guest": False,
"device_id": ret.get("device_id"),
}
defer.returnValue(user_info)
def get_appservice_by_req(self, request):
try:
token = get_access_token_from_request(
request, self.TOKEN_NOT_FOUND_HTTP_STATUS)
service = self.store.get_app_service_by_token(token)
if not service:
logger.warn("Unrecognised appservice access token: %s" %
(token, ))
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Unrecognised access token.",
errcode=Codes.UNKNOWN_TOKEN)
request.authenticated_entity = service.sender
return defer.succeed(service)
except KeyError:
raise AuthError(self.TOKEN_NOT_FOUND_HTTP_STATUS,
"Missing access token.")
def is_server_admin(self, user):
""" Check if the given user is a local server admin.
Args:
user (str): mxid of user to check
Returns:
bool: True if the user is an admin
"""
return self.store.is_server_admin(user)
@defer.inlineCallbacks
def add_auth_events(self, builder, context):
auth_ids = yield self.compute_auth_events(builder,
context.prev_state_ids)
auth_events_entries = yield self.store.add_event_hashes(auth_ids)
builder.auth_events = auth_events_entries
@defer.inlineCallbacks
def compute_auth_events(self,
event,
current_state_ids,
for_verification=False):
if event.type == EventTypes.Create:
defer.returnValue([])
auth_ids = []
key = (
EventTypes.PowerLevels,
"",
)
power_level_event_id = current_state_ids.get(key)
if power_level_event_id:
auth_ids.append(power_level_event_id)
key = (
EventTypes.JoinRules,
"",
)
join_rule_event_id = current_state_ids.get(key)
key = (
EventTypes.Member,
event.user_id,
)
member_event_id = current_state_ids.get(key)
key = (
EventTypes.Create,
"",
)
create_event_id = current_state_ids.get(key)
if create_event_id:
auth_ids.append(create_event_id)
if join_rule_event_id:
join_rule_event = yield self.store.get_event(join_rule_event_id)
join_rule = join_rule_event.content.get("join_rule")
is_public = join_rule == JoinRules.PUBLIC if join_rule else False
else:
is_public = False
if event.type == EventTypes.Member:
e_type = event.content["membership"]
if e_type in [Membership.JOIN, Membership.INVITE]:
if join_rule_event_id:
auth_ids.append(join_rule_event_id)
if e_type == Membership.JOIN:
if member_event_id and not is_public:
auth_ids.append(member_event_id)
else:
if member_event_id:
auth_ids.append(member_event_id)
if for_verification:
key = (
EventTypes.Member,
event.state_key,
)
existing_event_id = current_state_ids.get(key)
if existing_event_id:
auth_ids.append(existing_event_id)
if e_type == Membership.INVITE:
if "third_party_invite" in event.content:
key = (
EventTypes.ThirdPartyInvite,
event.content["third_party_invite"]["signed"]["token"])
third_party_invite_id = current_state_ids.get(key)
if third_party_invite_id:
auth_ids.append(third_party_invite_id)
elif member_event_id:
member_event = yield self.store.get_event(member_event_id)
if member_event.content["membership"] == Membership.JOIN:
auth_ids.append(member_event.event_id)
defer.returnValue(auth_ids)
def check_redaction(self, event, auth_events):
"""Check whether the event sender is allowed to redact the target event.
Returns:
True if the the sender is allowed to redact the target event if the
target event was created by them.
False if the sender is allowed to redact the target event with no
further checks.
Raises:
AuthError if the event sender is definitely not allowed to redact
the target event.
"""
return event_auth.check_redaction(event, auth_events)
@defer.inlineCallbacks
def check_can_change_room_list(self, room_id, user):
"""Check if the user is allowed to edit the room's entry in the
published room list.
Args:
room_id (str)
user (UserID)
"""
is_admin = yield self.is_server_admin(user)
if is_admin:
defer.returnValue(True)
user_id = user.to_string()
yield self.check_joined_room(room_id, user_id)
# We currently require the user is a "moderator" in the room. We do this
# by checking if they would (theoretically) be able to change the
# m.room.aliases events
power_level_event = yield self.state.get_current_state(
room_id, EventTypes.PowerLevels, "")
auth_events = {}
if power_level_event:
auth_events[(EventTypes.PowerLevels, "")] = power_level_event
send_level = event_auth.get_send_level(EventTypes.Aliases, "",
auth_events)
user_level = event_auth.get_user_power_level(user_id, auth_events)
if user_level < send_level:
raise AuthError(
403,
"This server requires you to be a moderator in the room to"
" edit its room list entry")
def test_eviction(self):
m1 = Mock(name="m1")
m2 = Mock(name="m2")
m3 = Mock(name="m3")
cache = LruCache(2)
cache.set("key1", "value", callbacks=[m1])
cache.set("key2", "value", callbacks=[m2])
self.assertEquals(m1.call_count, 0)
self.assertEquals(m2.call_count, 0)
self.assertEquals(m3.call_count, 0)
cache.set("key3", "value", callbacks=[m3])
self.assertEquals(m1.call_count, 1)
self.assertEquals(m2.call_count, 0)
self.assertEquals(m3.call_count, 0)
cache.set("key3", "value")
self.assertEquals(m1.call_count, 1)
self.assertEquals(m2.call_count, 0)
self.assertEquals(m3.call_count, 0)
cache.get("key2")
self.assertEquals(m1.call_count, 1)
self.assertEquals(m2.call_count, 0)
self.assertEquals(m3.call_count, 0)
cache.set("key1", "value", callbacks=[m1])
self.assertEquals(m1.call_count, 1)
self.assertEquals(m2.call_count, 0)
self.assertEquals(m3.call_count, 1)
def _contains_display_name(self, display_name):
if not display_name:
return False
body = self._event["content"].get("body", None)
if not body:
return False
return _glob_matches(display_name, body, word_boundary=True)
def _get_value(self, dotted_key):
return self._value_cache.get(dotted_key, None)
# Caches (glob, word_boundary) -> regex for push. See _glob_matches
regex_cache = LruCache(50000 * CACHE_SIZE_FACTOR)
register_cache("regex_push_cache", regex_cache)
def _glob_matches(glob, value, word_boundary=False):
"""Tests if value matches glob.
Args:
glob (string)
value (string): String to test against glob.
word_boundary (bool): Whether to match against word boundaries or entire
string. Defaults to False.
Returns:
bool
"""
class DeferredCache(Generic[KT, VT]):
"""Wraps an LruCache, adding support for Deferred results.
It expects that each entry added with set() will be a Deferred; likewise get()
will return a Deferred.
"""
__slots__ = (
"cache",
"thread",
"_pending_deferred_cache",
)
def __init__(
self,
name: str,
max_entries: int = 1000,
keylen: int = 1,
tree: bool = False,
iterable: bool = False,
apply_cache_factor_from_config: bool = True,
):
"""
Args:
name: The name of the cache
max_entries: Maximum amount of entries that the cache will hold
keylen: The length of the tuple used as the cache key. Ignored unless
`tree` is True.
tree: Use a TreeCache instead of a dict as the underlying cache type
iterable: If True, count each item in the cached object as an entry,
rather than each cached object
apply_cache_factor_from_config: Whether cache factors specified in the
config file affect `max_entries`
"""
cache_type = TreeCache if tree else dict
# _pending_deferred_cache maps from the key value to a `CacheEntry` object.
self._pending_deferred_cache = (
cache_type()) # type: MutableMapping[KT, CacheEntry]
def metrics_cb():
cache_pending_metric.labels(name).set(
len(self._pending_deferred_cache))
# cache is used for completed results and maps to the result itself, rather than
# a Deferred.
self.cache = LruCache(
max_size=max_entries,
keylen=keylen,
cache_name=name,
cache_type=cache_type,
size_callback=(lambda d: len(d)) if iterable else None,
metrics_collection_callback=metrics_cb,
apply_cache_factor_from_config=apply_cache_factor_from_config,
) # type: LruCache[KT, VT]
self.thread = None # type: Optional[threading.Thread]
@property
def max_entries(self):
return self.cache.max_size
def check_thread(self):
expected_thread = self.thread
if expected_thread is None:
self.thread = threading.current_thread()
else:
if expected_thread is not threading.current_thread():
raise ValueError(
"Cache objects can only be accessed from the main thread")
def get(
self,
key: KT,
callback: Optional[Callable[[], None]] = None,
update_metrics: bool = True,
) -> defer.Deferred:
"""Looks the key up in the caches.
For symmetry with set(), this method does *not* follow the synapse logcontext
rules: the logcontext will not be cleared on return, and the Deferred will run
its callbacks in the sentinel context. In other words: wrap the result with
make_deferred_yieldable() before `await`ing it.
Args:
key:
callback: Gets called when the entry in the cache is invalidated
update_metrics (bool): whether to update the cache hit rate metrics
Returns:
A Deferred which completes with the result. Note that this may later fail
if there is an ongoing set() operation which later completes with a failure.
Raises:
KeyError if the key is not found in the cache
"""
callbacks = [callback] if callback else []
val = self._pending_deferred_cache.get(key, _Sentinel.sentinel)
if val is not _Sentinel.sentinel:
val.callbacks.update(callbacks)
if update_metrics:
m = self.cache.metrics
assert m # we always have a name, so should always have metrics
m.inc_hits()
return val.deferred.observe()
val2 = self.cache.get(key,
_Sentinel.sentinel,
callbacks=callbacks,
update_metrics=update_metrics)
if val2 is _Sentinel.sentinel:
raise KeyError()
else:
return defer.succeed(val2)
def get_immediate(self,
key: KT,
default: T,
update_metrics: bool = True) -> Union[VT, T]:
"""If we have a *completed* cached value, return it."""
return self.cache.get(key, default, update_metrics=update_metrics)
def set(
self,
key: KT,
value: defer.Deferred,
callback: Optional[Callable[[], None]] = None,
) -> defer.Deferred:
"""Adds a new entry to the cache (or updates an existing one).
The given `value` *must* be a Deferred.
First any existing entry for the same key is invalidated. Then a new entry
is added to the cache for the given key.
Until the `value` completes, calls to `get()` for the key will also result in an
incomplete Deferred, which will ultimately complete with the same result as
`value`.
If `value` completes successfully, subsequent calls to `get()` will then return
a completed deferred with the same result. If it *fails*, the cache is
invalidated and subequent calls to `get()` will raise a KeyError.
If another call to `set()` happens before `value` completes, then (a) any
invalidation callbacks registered in the interim will be called, (b) any
`get()`s in the interim will continue to complete with the result from the
*original* `value`, (c) any future calls to `get()` will complete with the
result from the *new* `value`.
It is expected that `value` does *not* follow the synapse logcontext rules - ie,
if it is incomplete, it runs its callbacks in the sentinel context.
Args:
key: Key to be set
value: a deferred which will complete with a result to add to the cache
callback: An optional callback to be called when the entry is invalidated
"""
if not isinstance(value, defer.Deferred):
raise TypeError("not a Deferred")
callbacks = [callback] if callback else []
self.check_thread()
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry:
existing_entry.invalidate()
# XXX: why don't we invalidate the entry in `self.cache` yet?
# we can save a whole load of effort if the deferred is ready.
if value.called:
result = value.result
if not isinstance(result, failure.Failure):
self.cache.set(key, result, callbacks)
return value
# otherwise, we'll add an entry to the _pending_deferred_cache for now,
# and add callbacks to add it to the cache properly later.
observable = ObservableDeferred(value, consumeErrors=True)
observer = observable.observe()
entry = CacheEntry(deferred=observable, callbacks=callbacks)
self._pending_deferred_cache[key] = entry
def compare_and_pop():
"""Check if our entry is still the one in _pending_deferred_cache, and
if so, pop it.
Returns true if the entries matched.
"""
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry is entry:
return True
# oops, the _pending_deferred_cache has been updated since
# we started our query, so we are out of date.
#
# Better put back whatever we took out. (We do it this way
# round, rather than peeking into the _pending_deferred_cache
# and then removing on a match, to make the common case faster)
if existing_entry is not None:
self._pending_deferred_cache[key] = existing_entry
return False
def cb(result):
if compare_and_pop():
self.cache.set(key, result, entry.callbacks)
else:
# we're not going to put this entry into the cache, so need
# to make sure that the invalidation callbacks are called.
# That was probably done when _pending_deferred_cache was
# updated, but it's possible that `set` was called without
# `invalidate` being previously called, in which case it may
# not have been. Either way, let's double-check now.
entry.invalidate()
def eb(_fail):
compare_and_pop()
entry.invalidate()
# once the deferred completes, we can move the entry from the
# _pending_deferred_cache to the real cache.
#
observer.addCallbacks(cb, eb)
# we return a new Deferred which will be called before any subsequent observers.
return observable.observe()
def prefill(self, key: KT, value: VT, callback: Callable[[], None] = None):
callbacks = [callback] if callback else []
self.cache.set(key, value, callbacks=callbacks)
def invalidate(self, key):
self.check_thread()
self.cache.pop(key, None)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, which will (a) stop it being returned
# for future queries and (b) stop it being persisted as a proper entry
# in self.cache.
entry = self._pending_deferred_cache.pop(key, None)
# run the invalidation callbacks now, rather than waiting for the
# deferred to resolve.
if entry:
entry.invalidate()
def invalidate_many(self, key: KT):
self.check_thread()
if not isinstance(key, tuple):
raise TypeError("The cache key must be a tuple not %r" %
(type(key), ))
key = cast(KT, key)
self.cache.del_multi(key)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, as above
entry_dict = self._pending_deferred_cache.pop(key, None)
if entry_dict is not None:
for entry in iterate_tree_cache_entry(entry_dict):
entry.invalidate()
def invalidate_all(self):
self.check_thread()
self.cache.clear()
for entry in self._pending_deferred_cache.values():
entry.invalidate()
self._pending_deferred_cache.clear()
class Cache(object):
__slots__ = (
"cache",
"max_entries",
"name",
"keylen",
"thread",
"metrics",
"_pending_deferred_cache",
)
def __init__(self, name, max_entries=1000, keylen=1, tree=False, iterable=False):
cache_type = TreeCache if tree else dict
self._pending_deferred_cache = cache_type()
self.cache = LruCache(
max_size=max_entries, keylen=keylen, cache_type=cache_type,
size_callback=(lambda d: len(d)) if iterable else None,
evicted_callback=self._on_evicted,
)
self.name = name
self.keylen = keylen
self.thread = None
self.metrics = register_cache("cache", name, self.cache)
def _on_evicted(self, evicted_count):
self.metrics.inc_evictions(evicted_count)
def check_thread(self):
expected_thread = self.thread
if expected_thread is None:
self.thread = threading.current_thread()
else:
if expected_thread is not threading.current_thread():
raise ValueError(
"Cache objects can only be accessed from the main thread"
)
def get(self, key, default=_CacheSentinel, callback=None, update_metrics=True):
"""Looks the key up in the caches.
Args:
key(tuple)
default: What is returned if key is not in the caches. If not
specified then function throws KeyError instead
callback(fn): Gets called when the entry in the cache is invalidated
update_metrics (bool): whether to update the cache hit rate metrics
Returns:
Either a Deferred or the raw result
"""
callbacks = [callback] if callback else []
val = self._pending_deferred_cache.get(key, _CacheSentinel)
if val is not _CacheSentinel:
val.callbacks.update(callbacks)
if update_metrics:
self.metrics.inc_hits()
return val.deferred
val = self.cache.get(key, _CacheSentinel, callbacks=callbacks)
if val is not _CacheSentinel:
self.metrics.inc_hits()
return val
if update_metrics:
self.metrics.inc_misses()
if default is _CacheSentinel:
raise KeyError()
else:
return default
def set(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.check_thread()
entry = CacheEntry(
deferred=value,
callbacks=callbacks,
)
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry:
existing_entry.invalidate()
self._pending_deferred_cache[key] = entry
def shuffle(result):
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry is entry:
self.cache.set(key, result, entry.callbacks)
else:
# oops, the _pending_deferred_cache has been updated since
# we started our query, so we are out of date.
#
# Better put back whatever we took out. (We do it this way
# round, rather than peeking into the _pending_deferred_cache
# and then removing on a match, to make the common case faster)
if existing_entry is not None:
self._pending_deferred_cache[key] = existing_entry
# we're not going to put this entry into the cache, so need
# to make sure that the invalidation callbacks are called.
# That was probably done when _pending_deferred_cache was
# updated, but it's possible that `set` was called without
# `invalidate` being previously called, in which case it may
# not have been. Either way, let's double-check now.
entry.invalidate()
return result
entry.deferred.addCallback(shuffle)
def prefill(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.cache.set(key, value, callbacks=callbacks)
def invalidate(self, key):
self.check_thread()
self.cache.pop(key, None)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, which will (a) stop it being returned
# for future queries and (b) stop it being persisted as a proper entry
# in self.cache.
entry = self._pending_deferred_cache.pop(key, None)
# run the invalidation callbacks now, rather than waiting for the
# deferred to resolve.
if entry:
entry.invalidate()
def invalidate_many(self, key):
self.check_thread()
if not isinstance(key, tuple):
raise TypeError(
"The cache key must be a tuple not %r" % (type(key),)
)
self.cache.del_multi(key)
# if we have a pending lookup for this key, remove it from the
# _pending_deferred_cache, as above
entry_dict = self._pending_deferred_cache.pop(key, None)
if entry_dict is not None:
for entry in iterate_tree_cache_entry(entry_dict):
entry.invalidate()
def invalidate_all(self):
self.check_thread()
self.cache.clear()
for entry in itervalues(self._pending_deferred_cache):
entry.invalidate()
self._pending_deferred_cache.clear()
def test_get_set(self):
cache = LruCache(1)
cache["key"] = "value"
self.assertEquals(cache.get("key"), "value")
self.assertEquals(cache["key"], "value")
class DictionaryCache(Generic[KT, DKT]):
"""Caches key -> dictionary lookups, supporting caching partial dicts, i.e.
fetching a subset of dictionary keys for a particular key.
"""
def __init__(self, name: str, max_entries: int = 1000):
self.cache = LruCache(
max_size=max_entries, cache_name=name, size_callback=len
) # type: LruCache[KT, DictionaryEntry]
self.name = name
self.sequence = 0
self.thread = None # type: Optional[threading.Thread]
def check_thread(self) -> None:
expected_thread = self.thread
if expected_thread is None:
self.thread = threading.current_thread()
else:
if expected_thread is not threading.current_thread():
raise ValueError(
"Cache objects can only be accessed from the main thread"
)
def get(
self, key: KT, dict_keys: Optional[Iterable[DKT]] = None
) -> DictionaryEntry:
"""Fetch an entry out of the cache
Args:
key
dict_key: If given a set of keys then return only those keys
that exist in the cache.
Returns:
DictionaryEntry
"""
entry = self.cache.get(key, _Sentinel.sentinel)
if entry is not _Sentinel.sentinel:
if dict_keys is None:
return DictionaryEntry(
entry.full, entry.known_absent, dict(entry.value)
)
else:
return DictionaryEntry(
entry.full,
entry.known_absent,
{k: entry.value[k] for k in dict_keys if k in entry.value},
)
return DictionaryEntry(False, set(), {})
def invalidate(self, key: KT) -> None:
self.check_thread()
# Increment the sequence number so that any SELECT statements that
# raced with the INSERT don't update the cache (SYN-369)
self.sequence += 1
self.cache.pop(key, None)
def invalidate_all(self) -> None:
self.check_thread()
self.sequence += 1
self.cache.clear()
def update(
self,
sequence: int,
key: KT,
value: Dict[DKT, Any],
fetched_keys: Optional[Set[DKT]] = None,
) -> None:
"""Updates the entry in the cache
Args:
sequence
key
value: The value to update the cache with.
fetched_keys: All of the dictionary keys which were
fetched from the database.
If None, this is the complete value for key K. Otherwise, it
is used to infer a list of keys which we know don't exist in
the full dict.
"""
self.check_thread()
if self.sequence == sequence:
# Only update the cache if the caches sequence number matches the
# number that the cache had before the SELECT was started (SYN-369)
if fetched_keys is None:
self._insert(key, value, set())
else:
self._update_or_insert(key, value, fetched_keys)
def _update_or_insert(
self, key: KT, value: Dict[DKT, Any], known_absent: Set[DKT]
) -> None:
# We pop and reinsert as we need to tell the cache the size may have
# changed
entry = self.cache.pop(key, DictionaryEntry(False, set(), {}))
entry.value.update(value)
entry.known_absent.update(known_absent)
self.cache[key] = entry
def _insert(self, key: KT, value: Dict[DKT, Any], known_absent: Set[DKT]) -> None:
self.cache[key] = DictionaryEntry(True, known_absent, value)
def test_setdefault(self):
cache = LruCache(1)
self.assertEquals(cache.setdefault("key", 1), 1)
self.assertEquals(cache.get("key"), 1)
self.assertEquals(cache.setdefault("key", 2), 1)
self.assertEquals(cache.get("key"), 1)
class Cache(object):
__slots__ = (
"cache",
"max_entries",
"name",
"keylen",
"sequence",
"thread",
"metrics",
)
def __init__(self, name, max_entries=1000, keylen=1, tree=False):
cache_type = TreeCache if tree else dict
self.cache = LruCache(
max_size=max_entries, keylen=keylen, cache_type=cache_type
)
self.name = name
self.keylen = keylen
self.sequence = 0
self.thread = None
self.metrics = register_cache(name, self.cache)
def check_thread(self):
expected_thread = self.thread
if expected_thread is None:
self.thread = threading.current_thread()
else:
if expected_thread is not threading.current_thread():
raise ValueError(
"Cache objects can only be accessed from the main thread"
)
def get(self, key, default=_CacheSentinel, callback=None):
val = self.cache.get(key, _CacheSentinel, callback=callback)
if val is not _CacheSentinel:
self.metrics.inc_hits()
return val
self.metrics.inc_misses()
if default is _CacheSentinel:
raise KeyError()
else:
return default
def update(self, sequence, key, value, callback=None):
self.check_thread()
if self.sequence == sequence:
# Only update the cache if the caches sequence number matches the
# number that the cache had before the SELECT was started (SYN-369)
self.prefill(key, value, callback=callback)
def prefill(self, key, value, callback=None):
self.cache.set(key, value, callback=callback)
def invalidate(self, key):
self.check_thread()
if not isinstance(key, tuple):
raise TypeError(
"The cache key must be a tuple not %r" % (type(key),)
)
# Increment the sequence number so that any SELECT statements that
# raced with the INSERT don't update the cache (SYN-369)
self.sequence += 1
self.cache.pop(key, None)
def invalidate_many(self, key):
self.check_thread()
if not isinstance(key, tuple):
raise TypeError(
"The cache key must be a tuple not %r" % (type(key),)
)
self.sequence += 1
self.cache.del_multi(key)
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
def __init__(self, database: DatabasePool, db_conn, hs: "HomeServer"):
super().__init__(database, db_conn, hs)
self.client_ip_last_seen: LruCache[tuple, int] = LruCache(
cache_name="client_ip_last_seen", max_size=50000)
def test_special_size(self):
cache = LruCache(10, "mycache")
self.assertEqual(cache.max_size, 100)
class DictionaryCache(object):
"""Caches key -> dictionary lookups, supporting caching partial dicts, i.e.
fetching a subset of dictionary keys for a particular key.
"""
def __init__(self, name, max_entries=1000):
self.cache = LruCache(max_size=max_entries, size_callback=len)
self.name = name
self.sequence = 0
self.thread = None
# caches_by_name[name] = self.cache
class Sentinel(object):
__slots__ = []
self.sentinel = Sentinel()
self.metrics = register_cache(name, self.cache)
def check_thread(self):
expected_thread = self.thread
if expected_thread is None:
self.thread = threading.current_thread()
else:
if expected_thread is not threading.current_thread():
raise ValueError(
"Cache objects can only be accessed from the main thread"
)
def get(self, key, dict_keys=None):
"""Fetch an entry out of the cache
Args:
key
dict_key(list): If given a set of keys then return only those keys
that exist in the cache.
Returns:
DictionaryEntry
"""
entry = self.cache.get(key, self.sentinel)
if entry is not self.sentinel:
self.metrics.inc_hits()
if dict_keys is None:
return DictionaryEntry(entry.full, entry.known_absent, dict(entry.value))
else:
return DictionaryEntry(entry.full, entry.known_absent, {
k: entry.value[k]
for k in dict_keys
if k in entry.value
})
self.metrics.inc_misses()
return DictionaryEntry(False, set(), {})
def invalidate(self, key):
self.check_thread()
# Increment the sequence number so that any SELECT statements that
# raced with the INSERT don't update the cache (SYN-369)
self.sequence += 1
self.cache.pop(key, None)
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
def update(self, sequence, key, value, full=False, known_absent=None):
"""Updates the entry in the cache
Args:
sequence
key
value (dict): The value to update the cache with.
full (bool): Whether the given value is the full dict, or just a
partial subset there of. If not full then any existing entries
for the key will be updated.
known_absent (set): Set of keys that we know don't exist in the full
dict.
"""
self.check_thread()
if self.sequence == sequence:
# Only update the cache if the caches sequence number matches the
# number that the cache had before the SELECT was started (SYN-369)
if known_absent is None:
known_absent = set()
if full:
self._insert(key, value, known_absent)
else:
self._update_or_insert(key, value, known_absent)
def _update_or_insert(self, key, value, known_absent):
# We pop and reinsert as we need to tell the cache the size may have
# changed
entry = self.cache.pop(key, DictionaryEntry(False, set(), {}))
entry.value.update(value)
entry.known_absent.update(known_absent)
self.cache[key] = entry
def _insert(self, key, value, known_absent):
self.cache[key] = DictionaryEntry(True, known_absent, value)
def test_get(self):
m = Mock()
cache = LruCache(1)
cache.set("key", "value")
self.assertFalse(m.called)
cache.get("key", callbacks=[m])
self.assertFalse(m.called)
cache.get("key", "value")
self.assertFalse(m.called)
cache.set("key", "value2")
self.assertEqual(m.call_count, 1)
cache.set("key", "value")
self.assertEqual(m.call_count, 1)
class DictionaryCache(object):
"""Caches key -> dictionary lookups, supporting caching partial dicts, i.e.
fetching a subset of dictionary keys for a particular key.
"""
def __init__(self, name, max_entries=1000):
self.cache = LruCache(max_size=max_entries, size_callback=len)
self.name = name
self.sequence = 0
self.thread = None
# caches_by_name[name] = self.cache
class Sentinel(object):
__slots__ = []
self.sentinel = Sentinel()
self.metrics = register_cache(name, self.cache)
def check_thread(self):
expected_thread = self.thread
if expected_thread is None:
self.thread = threading.current_thread()
else:
if expected_thread is not threading.current_thread():
raise ValueError(
"Cache objects can only be accessed from the main thread")
def get(self, key, dict_keys=None):
"""Fetch an entry out of the cache
Args:
key
dict_key(list): If given a set of keys then return only those keys
that exist in the cache.
Returns:
DictionaryEntry
"""
entry = self.cache.get(key, self.sentinel)
if entry is not self.sentinel:
self.metrics.inc_hits()
if dict_keys is None:
return DictionaryEntry(entry.full, entry.known_absent,
dict(entry.value))
else:
return DictionaryEntry(
entry.full, entry.known_absent,
{k: entry.value[k]
for k in dict_keys if k in entry.value})
self.metrics.inc_misses()
return DictionaryEntry(False, set(), {})
def invalidate(self, key):
self.check_thread()
# Increment the sequence number so that any SELECT statements that
# raced with the INSERT don't update the cache (SYN-369)
self.sequence += 1
self.cache.pop(key, None)
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
def update(self, sequence, key, value, full=False, known_absent=None):
"""Updates the entry in the cache
Args:
sequence
key
value (dict): The value to update the cache with.
full (bool): Whether the given value is the full dict, or just a
partial subset there of. If not full then any existing entries
for the key will be updated.
known_absent (set): Set of keys that we know don't exist in the full
dict.
"""
self.check_thread()
if self.sequence == sequence:
# Only update the cache if the caches sequence number matches the
# number that the cache had before the SELECT was started (SYN-369)
if known_absent is None:
known_absent = set()
if full:
self._insert(key, value, known_absent)
else:
self._update_or_insert(key, value, known_absent)
def _update_or_insert(self, key, value, known_absent):
entry = self.cache.setdefault(key, DictionaryEntry(False, set(), {}))
entry.value.update(value)
entry.known_absent.update(known_absent)
def _insert(self, key, value, known_absent):
self.cache[key] = DictionaryEntry(True, known_absent, value)
def test_pop(self):
cache = LruCache(1)
cache["key"] = 1
self.assertEquals(cache.pop("key"), 1)
self.assertEquals(cache.pop("key"), None)
class Cache(object):
__slots__ = (
"cache",
"max_entries",
"name",
"keylen",
"sequence",
"thread",
"metrics",
"_pending_deferred_cache",
)
def __init__(self, name, max_entries=1000, keylen=1, tree=False, iterable=False):
cache_type = TreeCache if tree else dict
self._pending_deferred_cache = cache_type()
self.cache = LruCache(
max_size=max_entries, keylen=keylen, cache_type=cache_type,
size_callback=(lambda d: len(d)) if iterable else None,
evicted_callback=self._on_evicted,
)
self.name = name
self.keylen = keylen
self.sequence = 0
self.thread = None
self.metrics = register_cache(name, self.cache)
def _on_evicted(self, evicted_count):
self.metrics.inc_evictions(evicted_count)
def check_thread(self):
expected_thread = self.thread
if expected_thread is None:
self.thread = threading.current_thread()
else:
if expected_thread is not threading.current_thread():
raise ValueError(
"Cache objects can only be accessed from the main thread"
)
def get(self, key, default=_CacheSentinel, callback=None, update_metrics=True):
"""Looks the key up in the caches.
Args:
key(tuple)
default: What is returned if key is not in the caches. If not
specified then function throws KeyError instead
callback(fn): Gets called when the entry in the cache is invalidated
update_metrics (bool): whether to update the cache hit rate metrics
Returns:
Either a Deferred or the raw result
"""
callbacks = [callback] if callback else []
val = self._pending_deferred_cache.get(key, _CacheSentinel)
if val is not _CacheSentinel:
if val.sequence == self.sequence:
val.callbacks.update(callbacks)
if update_metrics:
self.metrics.inc_hits()
return val.deferred
val = self.cache.get(key, _CacheSentinel, callbacks=callbacks)
if val is not _CacheSentinel:
self.metrics.inc_hits()
return val
if update_metrics:
self.metrics.inc_misses()
if default is _CacheSentinel:
raise KeyError()
else:
return default
def set(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.check_thread()
entry = CacheEntry(
deferred=value,
sequence=self.sequence,
callbacks=callbacks,
)
entry.callbacks.update(callbacks)
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry:
existing_entry.invalidate()
self._pending_deferred_cache[key] = entry
def shuffle(result):
if self.sequence == entry.sequence:
existing_entry = self._pending_deferred_cache.pop(key, None)
if existing_entry is entry:
self.cache.set(key, result, entry.callbacks)
else:
entry.invalidate()
else:
entry.invalidate()
return result
entry.deferred.addCallback(shuffle)
def prefill(self, key, value, callback=None):
callbacks = [callback] if callback else []
self.cache.set(key, value, callbacks=callbacks)
def invalidate(self, key):
self.check_thread()
# Increment the sequence number so that any SELECT statements that
# raced with the INSERT don't update the cache (SYN-369)
self.sequence += 1
entry = self._pending_deferred_cache.pop(key, None)
if entry:
entry.invalidate()
self.cache.pop(key, None)
def invalidate_many(self, key):
self.check_thread()
if not isinstance(key, tuple):
raise TypeError(
"The cache key must be a tuple not %r" % (type(key),)
)
self.sequence += 1
self.cache.del_multi(key)
entry_dict = self._pending_deferred_cache.pop(key, None)
if entry_dict is not None:
for entry in iterate_tree_cache_entry(entry_dict):
entry.invalidate()
def invalidate_all(self):
self.check_thread()
self.sequence += 1
self.cache.clear()
def test_clear(self):
cache = LruCache(1)
cache["key"] = 1
cache.clear()
self.assertEquals(len(cache), 0)
r = regex_cache.get((display_name, False, True), None)
if not r:
r1 = re.escape(display_name)
r1 = re_word_boundary(r1)
r = re.compile(r1, flags=re.IGNORECASE)
regex_cache[(display_name, False, True)] = r
return bool(r.search(body))
def _get_value(self, dotted_key: str) -> Optional[str]:
return self._value_cache.get(dotted_key, None)
# Caches (string, is_glob, word_boundary) -> regex for push. See _glob_matches
regex_cache: LruCache[Tuple[str, bool, bool],
Pattern] = LruCache(50000, "regex_push_cache")
def _glob_matches(glob: str, value: str, word_boundary: bool = False) -> bool:
"""Tests if value matches glob.
Args:
glob
value: String to test against glob.
word_boundary: Whether to match against word boundaries or entire
string. Defaults to False.
"""
try:
r = regex_cache.get((glob, True, word_boundary), None)
if not r:
