from __future__ import annotations import asyncio from asyncio import CancelledError from dataclasses import astuple, dataclass import enum from functools import lru_cache, partial import logging import socket import sys import time from typing import TYPE_CHECKING, Any, NoReturn import aiohappyeyeballs from async_interrupt import interrupt from google.protobuf.json_format import MessageToDict import aioesphomeapi.host_resolver as hr from ._frame_helper.base import MAX_NAME_LEN, safe_label_str from ._frame_helper.noise import APINoiseFrameHelper from ._frame_helper.plain_text import APIPlaintextFrameHelper from .api_pb2 import ( # type: ignore[attr-defined] DST_RULE_TYPE_DAY_OF_YEAR as DST_RULE_TYPE_DAY_OF_YEAR_PB, DST_RULE_TYPE_JULIAN_NO_LEAP as DST_RULE_TYPE_JULIAN_NO_LEAP_PB, DST_RULE_TYPE_MONTH_WEEK_DAY as DST_RULE_TYPE_MONTH_WEEK_DAY_PB, DST_RULE_TYPE_NONE as DST_RULE_TYPE_NONE_PB, AuthenticationRequest, AuthenticationResponse, DisconnectRequest, DisconnectResponse, DSTRule as DSTRuleProto, GetTimeRequest, GetTimeResponse, HelloRequest, HelloResponse, ParsedTimezone as ParsedTimezoneProto, PingRequest, PingResponse, ) from .core import ( MESSAGE_TYPE_TO_PROTO, APIConnectionCancelledError, APIConnectionError, BadNameAPIError, ConnectionNotEstablishedAPIError, HandshakeAPIError, InvalidAuthAPIError, PingFailedAPIError, ProtocolAPIError, ReadFailedAPIError, SocketAPIError, SocketClosedAPIError, TimeoutAPIError, UnhandledAPIConnectionError, ) from .model import APIVersion, message_types_to_names from .posix_tz import DSTRule as DSTRuleParsed, DSTRuleType, parse_posix_tz from .timezone import get_timezone from .util import asyncio_timeout if TYPE_CHECKING: from collections.abc import Callable from google.protobuf import message from .zeroconf import ZeroconfManager _LOGGER = logging.getLogger(__name__) MESSAGE_NUMBER_TO_PROTO: tuple[ tuple[Callable[[], message.Message], Callable[[message.Message, bytes], None]], ... ] = tuple((msg, msg.MergeFromString) for msg in MESSAGE_TYPE_TO_PROTO.values()) _MESSAGE_NUMBER_TO_PROTO_LEN = len(MESSAGE_NUMBER_TO_PROTO) PREFERRED_BUFFER_SIZE = 2097152 # Set buffer limit to 2MB MIN_BUFFER_SIZE = 131072 # Minimum buffer size to use DISCONNECT_REQUEST_MESSAGE = DisconnectRequest() DISCONNECT_RESPONSE_MESSAGES = (DisconnectResponse(),) PING_REQUEST_MESSAGES = (PingRequest(),) PING_RESPONSE_MESSAGES = (PingResponse(),) NO_PASSWORD_AUTH_REQUEST = AuthenticationRequest() PROTO_TO_MESSAGE_TYPE: dict[ type[message.Message], tuple[int, Callable[[message.Message], bytes]] ] = {v: (k, v.SerializeToString) for k, v in MESSAGE_TYPE_TO_PROTO.items()} KEEP_ALIVE_TIMEOUT_RATIO = 4.5 # # We use 4.5x the keep-alive time as the timeout for the pong # since the default ping interval is 20s which is about the time # a device takes to reboot and reconnect to the network making # the maximum time it has to respond to a ping at 90s which is # enough time to know that the device has truly disconnected # from the network. # DISCONNECT_CONNECT_TIMEOUT = 5.0 # How long to wait for an existing connection to finish being # setup when requesting a disconnect so we can try to disconnect # gracefully without closing the socket out from under the # the esp device DISCONNECT_RESPONSE_TIMEOUT = 10.0 # Timeout for the noise handshake to complete. 60s is intentionally high # because on ESP8266 with power_save_mode: LIGHT and weak WiFi (-70 dBm+), # TCP segments get dropped and retransmitted with exponential backoff. # The Curve25519 key exchange takes ~300ms on ESP8266, but real-world # handshake times of 28-30s have been observed due to retransmissions. # See https://github.com/esphome/esphome/issues/14999 HANDSHAKE_TIMEOUT = 60.0 CONNECT_REQUEST_TIMEOUT = 30.0 # The connect timeout should be the maximum time we expect the esp to take # to reboot and connect to the network/WiFi. TCP_CONNECT_TIMEOUT = 60.0 WRITE_EXCEPTIONS = (RuntimeError, ConnectionResetError, OSError) # Cap on how many bytes of the offending peer payload we embed in the # "Invalid protobuf message" log entry and ProtocolAPIError text. A peer # can send up to MAX_PLAINTEXT_FRAME_SIZE (65535 bytes) or a 16-bit-length # noise frame of crafted garbage; with repr() escaping that balloons ~4x # in the log file. The truncated prefix is plenty for debugging the # message shape, and the total length is still reported separately. MAX_PROTOBUF_ERROR_DATA_BYTES = 255 _MAX_PROTOBUF_ERROR_DATA_BYTES = MAX_PROTOBUF_ERROR_DATA_BYTES _WIN32 = sys.platform == "win32" _int = int _bytes = bytes _float = float class ConnectionInterruptedError(Exception): """An error that is raised when a connection is interrupted.""" @dataclass class ConnectionParams: addresses: list[str] port: int password: str | None client_info: str keepalive: float zeroconf_manager: ZeroconfManager noise_psk: str | None expected_name: str | None expected_mac: str | None timezone: str | None = None provide_time: bool = True class ConnectionState(enum.Enum): # The connection is initialized, but connect() wasn't called yet INITIALIZED = 0 # The host has been resolved, but the socket hasn't been opened yet HOST_RESOLVED = 1 # The socket has been opened, but the handshake and login haven't been completed SOCKET_OPENED = 2 # The handshake has been completed, messages can be exchanged HANDSHAKE_COMPLETE = 3 # The connection has been established, authenticated data can be exchanged CONNECTED = 4 CLOSED = 5 CONNECTION_STATE_INITIALIZED = ConnectionState.INITIALIZED CONNECTION_STATE_HOST_RESOLVED = ConnectionState.HOST_RESOLVED CONNECTION_STATE_SOCKET_OPENED = ConnectionState.SOCKET_OPENED CONNECTION_STATE_HANDSHAKE_COMPLETE = ConnectionState.HANDSHAKE_COMPLETE CONNECTION_STATE_CONNECTED = ConnectionState.CONNECTED CONNECTION_STATE_CLOSED = ConnectionState.CLOSED def _make_hello_request(client_info: str) -> HelloRequest: """Make a HelloRequest.""" return HelloRequest( client_info=client_info, api_version_major=1, api_version_minor=14 ) _cached_make_hello_request = lru_cache(maxsize=16)(_make_hello_request) make_hello_request = _cached_make_hello_request _DST_RULE_TYPE_MAP: dict[DSTRuleType, int] = { DSTRuleType.NONE: DST_RULE_TYPE_NONE_PB, DSTRuleType.MONTH_WEEK_DAY: DST_RULE_TYPE_MONTH_WEEK_DAY_PB, DSTRuleType.JULIAN_NO_LEAP: DST_RULE_TYPE_JULIAN_NO_LEAP_PB, DSTRuleType.DAY_OF_YEAR: DST_RULE_TYPE_DAY_OF_YEAR_PB, } def _build_dst_rule_proto(rule: DSTRuleParsed) -> DSTRuleProto: """Build a protobuf DSTRule from a parsed DSTRule.""" return DSTRuleProto( time_seconds=rule.time_seconds, day=rule.day, type=_DST_RULE_TYPE_MAP[rule.type], month=rule.month, week=rule.week, day_of_week=rule.day_of_week, ) @lru_cache(maxsize=8) def _build_parsed_tz_proto(tz_string: str) -> ParsedTimezoneProto | None: """Build a protobuf ParsedTimezone from a POSIX TZ string. Returns None if the string is empty or unparseable. """ if not tz_string: return None try: parsed = parse_posix_tz(tz_string) except ValueError: return None return ParsedTimezoneProto( std_offset_seconds=parsed.std_offset_seconds, dst_offset_seconds=parsed.dst_offset_seconds, dst_start=_build_dst_rule_proto(parsed.dst_start), dst_end=_build_dst_rule_proto(parsed.dst_end), ) def handle_timeout(fut: asyncio.Future[None]) -> None: """Handle a timeout.""" if not fut.done(): fut.set_exception(TimeoutError) _handle_timeout = handle_timeout def handle_complex_message( fut: asyncio.Future[None], responses: list[message.Message], do_append: Callable[[message.Message], bool] | None, do_stop: Callable[[message.Message], bool] | None, resp: message.Message, ) -> None: """Handle a message that is part of a response.""" if not fut.done(): if do_append is None or do_append(resp): responses.append(resp) if do_stop is None or do_stop(resp): fut.set_result(None) _handle_complex_message = handle_complex_message class APIConnection: """Represent _one_ connection to a remote native API device. An instance of this class may only be used once, for every new connection a new instance should be established. Only created from APIClient — do not instantiate directly. """ __slots__ = ( "_addrs_info", "_cached_timezone", "_debug_enabled", "_expected_disconnect", "_fatal_exception", "_finish_connect_future", "_frame_helper", "_handshake_complete", "_keep_alive_interval", "_keep_alive_timeout", "_log_errors", "_loop", "_message_handlers", "_params", "_ping_timer", "_pong_timer", "_read_exception_futures", "_resolve_host_future", "_send_pending_ping", "_socket", "_start_connect_future", "api_version", "connected_address", "connection_state", "is_connected", "log_name", "on_stop", "received_name", ) def __init__( self, params: ConnectionParams, on_stop: Callable[[bool], None] | None, debug_enabled: bool, log_name: str | None, log_errors: bool = True, ) -> None: self._params = params self.on_stop = on_stop self._socket: socket.socket | None = None self._frame_helper: None | APINoiseFrameHelper | APIPlaintextFrameHelper = None self.api_version: APIVersion | None = None self.connection_state = CONNECTION_STATE_INITIALIZED # Message handlers currently subscribed to incoming messages self._message_handlers: dict[Any, set[Callable[[message.Message], None]]] = {} # The friendly name to show for this connection in the logs self.log_name = log_name or ",".join(params.addresses) # futures currently subscribed to exceptions in the read task self._read_exception_futures: set[asyncio.Future[None]] = set() self._ping_timer: asyncio.TimerHandle | None = None self._pong_timer: asyncio.TimerHandle | None = None keepalive = params.keepalive self._keep_alive_interval = keepalive self._keep_alive_timeout = keepalive * KEEP_ALIVE_TIMEOUT_RATIO self._resolve_host_future: asyncio.Future[None] | None = None self._start_connect_future: asyncio.Future[None] | None = None self._finish_connect_future: asyncio.Future[None] | None = None self._fatal_exception: Exception | None = None self._expected_disconnect = False self._send_pending_ping = False self._loop = asyncio.get_running_loop() self.is_connected = False self._handshake_complete = False self._debug_enabled = debug_enabled self.received_name: str = "" self._cached_timezone: str = "" self.connected_address: str | None = None self._addrs_info: list[hr.AddrInfo] = [] self._log_errors = log_errors def set_log_name(self, name: str) -> None: """Set the friendly log name for this connection.""" self.log_name = name if self._frame_helper is not None: self._frame_helper.set_log_name(name) def _cleanup(self) -> None: # noqa: C901 # Cython-hot path; see CLAUDE.md "Cython gotchas" """Clean up all resources that have been allocated. Safe to call multiple times. """ if self.connection_state is CONNECTION_STATE_CLOSED: return was_connected = self.is_connected self._set_connection_state(CONNECTION_STATE_CLOSED) if self._debug_enabled: if self._fatal_exception: _LOGGER.debug( "Cleaning up connection to %s (error: %s)", self.log_name, self._fatal_exception, ) else: _LOGGER.debug("Cleaning up connection to %s", self.log_name) for fut in self._read_exception_futures: if not fut.done(): err = self._fatal_exception or APIConnectionError("Connection closed") new_exc = err if not isinstance(err, APIConnectionError): new_exc = ReadFailedAPIError(str(err) or "Read failed") new_exc.__cause__ = err fut.set_exception(new_exc) self._read_exception_futures.clear() self._set_resolve_host_future() self._set_start_connect_future() self._set_finish_connect_future() if self._frame_helper is not None: self._frame_helper.close() self._frame_helper = None if self._socket is not None: self._socket.close() self._socket = None self._async_cancel_pong_timer() if self._ping_timer is not None: self._ping_timer.cancel() self._ping_timer = None if (on_stop := self.on_stop) is not None and was_connected: self.on_stop = None on_stop(self._expected_disconnect) def set_debug(self, enable: bool) -> None: """Enable or disable debug logging.""" self._debug_enabled = enable async def _connect_socket_connect(self, addrs: list[hr.AddrInfo]) -> None: """Step 2 in connect process: connect the socket.""" if self._debug_enabled: _LOGGER.debug( "%s: Connecting to %s", self.log_name, ", ".join(str(addr.sockaddr) for addr in addrs), ) addr_infos: list[aiohappyeyeballs.AddrInfoType] = [ ( addr.family, addr.type, addr.proto, "", astuple(addr.sockaddr), ) for addr in addrs ] last_exception: Exception | None = None sock: socket.socket | None = None interleave = 1 while addr_infos: try: async with asyncio_timeout(TCP_CONNECT_TIMEOUT): # Devices are likely on the local network so we # only use a 100ms happy eyeballs delay sock = await aiohappyeyeballs.start_connection( addr_infos, happy_eyeballs_delay=0.1, interleave=interleave, loop=self._loop, ) break except (OSError, TimeoutError) as err: last_exception = err aiohappyeyeballs.pop_addr_infos_interleave(addr_infos, interleave) if sock is None: if isinstance(last_exception, TimeoutError): msg = f"Timeout while connecting to {addrs}" raise TimeoutAPIError(msg) from last_exception msg = f"Error connecting to {addrs}: {last_exception}" raise SocketAPIError(msg) from last_exception self._socket = sock sock.setblocking(False) sock.setsockopt(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1) try: sock.setsockopt(socket.IPPROTO_TCP, socket.TCP_QUICKACK, 1) # type: ignore[attr-defined, unused-ignore] except (AttributeError, OSError): # On FreeBSD this may throw OSError _LOGGER.debug( "%s: TCP_QUICKACK not supported", self.log_name, ) self._increase_recv_buffer_size() self.connected_address = sock.getpeername()[0] if self._debug_enabled: _LOGGER.debug( "%s: Opened socket to %s:%s", self.log_name, self.connected_address, self._params.port, ) def _increase_recv_buffer_size(self) -> None: """Increase the recv buffer size.""" if TYPE_CHECKING: assert self._socket is not None new_buffer_size = PREFERRED_BUFFER_SIZE while True: # Try to reduce the pressure on ESPHome device as it measures # ram in bytes and we measure ram in megabytes. try: self._socket.setsockopt( socket.SOL_SOCKET, socket.SO_RCVBUF, new_buffer_size ) except OSError as err: if new_buffer_size <= MIN_BUFFER_SIZE: _LOGGER.warning( "%s: Unable to increase the socket receive buffer size to %s; " "The connection may unstable if the ESPHome device sends " "data at volume (ex. a Bluetooth proxy or camera): %s", self.log_name, new_buffer_size, err, ) return new_buffer_size //= 2 else: return async def _connect_init_frame_helper(self) -> None: """Step 3 in connect process: initialize the frame helper and init read loop.""" fh: APIPlaintextFrameHelper | APINoiseFrameHelper if TYPE_CHECKING: assert self._socket is not None if (noise_psk := self._params.noise_psk) is None: _, fh = await self._loop.create_connection( # type: ignore[type-var] lambda: APIPlaintextFrameHelper( connection=self, client_info=self._params.client_info, log_name=self.log_name, ), sock=self._socket, ) else: _, fh = await self._loop.create_connection( # type: ignore[type-var] lambda: APINoiseFrameHelper( noise_psk=noise_psk, expected_name=self._params.expected_name, expected_mac=self._params.expected_mac, connection=self, client_info=self._params.client_info, log_name=self.log_name, ), sock=self._socket, ) # Set the frame helper right away to ensure # the socket gets closed if we fail to handshake self._frame_helper = fh handshake_handle = self._loop.call_at( self._loop.time() + HANDSHAKE_TIMEOUT, _handle_timeout, self._frame_helper.ready_future, ) try: await self._frame_helper.ready_future except TimeoutError as err: msg = f"Handshake timed out after {HANDSHAKE_TIMEOUT}s" raise TimeoutAPIError(msg) from err except OSError as err: msg = f"Handshake failed: {err}" raise HandshakeAPIError(msg) from err finally: handshake_handle.cancel() self._set_connection_state(CONNECTION_STATE_HANDSHAKE_COMPLETE) async def _connect_hello_login(self, login: bool) -> None: """Step 4 in connect process: send hello and login and get api version.""" messages = [make_hello_request(self._params.client_info)] msg_types = [HelloResponse] if login: messages.append(self._make_auth_request()) # Never wait for AuthenticationResponse - the device will either # send it immediately if authentication fails, or not send it at all # if it doesn't support password authentication responses = await self.send_messages_await_response_complex( tuple(messages), None, lambda resp: ( type(resp) # pylint: disable=unidiomatic-typecheck is HelloResponse ), # Only wait for HelloResponse tuple(msg_types), CONNECT_REQUEST_TIMEOUT, ) resp = responses.pop(0) self._process_hello_resp(resp) def _handle_login_response(self, login_response: AuthenticationResponse) -> None: """Handle a AuthenticationResponse.""" if login_response.invalid_password: self.report_fatal_error(InvalidAuthAPIError("Invalid password!")) def _process_hello_resp(self, resp: HelloResponse) -> None: """Process a HelloResponse.""" if self._debug_enabled: _LOGGER.debug( "%s: Successfully connected ('%s' API=%s.%s)", self.log_name, safe_label_str(resp.server_info, MAX_NAME_LEN), resp.api_version_major, resp.api_version_minor, ) api_version = APIVersion(resp.api_version_major, resp.api_version_minor) if api_version.major > 2: _LOGGER.error( "%s: Incompatible version %s! Closing connection", self.log_name, api_version.major, ) msg = f"Incompatible API version ({api_version})." raise APIConnectionError(msg) self.api_version = api_version expected_name = self._params.expected_name # Compare against the raw decoded value so a peer can't sneak past # expected_name by appending non-printable bytes that the sanitizer # would strip; only the value used for logs and for later self- # storage gets sanitized + length-capped. Same pattern as # _frame_helper/noise.py::_handle_hello. if received_name_raw := resp.name: received_name = safe_label_str(received_name_raw, MAX_NAME_LEN) if expected_name is not None and received_name_raw != expected_name: msg = ( f"Expected '{expected_name}' but server sent " f"a different name: '{received_name}'" ) raise BadNameAPIError( msg, received_name, ) self.received_name = received_name self.set_log_name(self.received_name) def _async_schedule_keep_alive(self, now: _float) -> None: """Start the keep alive task.""" self._send_pending_ping = True self._ping_timer = self._loop.call_at( now + self._keep_alive_interval, self._async_send_keep_alive ) def _async_send_keep_alive(self) -> None: """Send a keep alive message.""" now = self._loop.time() if self._send_pending_ping: self.send_messages(PING_REQUEST_MESSAGES) if self._pong_timer is None: # Do not reset the timer if it's already set # since the only thing we want to reset the timer # is if we receive a pong. self._pong_timer = self._loop.call_at( now + self._keep_alive_timeout, self._async_pong_not_received ) elif self._debug_enabled: # # We haven't reached the ping response (pong) timeout yet # and we haven't seen a response to the last ping # # We send another ping in case the device has # rebooted and dropped the connection without telling # us to force a TCP RST aka connection reset by peer. # _LOGGER.debug( "%s: PingResponse (pong) was not received " "since last keep alive after %s seconds; " "rescheduling keep alive", self.log_name, self._keep_alive_interval, ) self._async_schedule_keep_alive(now) def _async_cancel_pong_timer(self) -> None: """Cancel the pong timer.""" if self._pong_timer is not None: self._pong_timer.cancel() self._pong_timer = None def _async_pong_not_received(self) -> None: """Ping not received.""" if self._debug_enabled: _LOGGER.debug( "%s: Ping response not received after %s seconds", self.log_name, self._keep_alive_timeout, ) self.report_fatal_error( PingFailedAPIError( f"Ping response not received after {self._keep_alive_timeout} seconds" ) ) async def start_resolve_host(self) -> None: """Start the host resolution process. This part of the process resolves the hostnames to IP addresses and prepares the connection for the next step. """ if self.connection_state is not CONNECTION_STATE_INITIALIZED: msg = "Connection can only be used once, connection is not in init state" raise RuntimeError(msg) self._resolve_host_future = self._loop.create_future() try: async with interrupt( self._resolve_host_future, ConnectionInterruptedError, None ): self._addrs_info = await hr.async_resolve_host( self._params.addresses, self._params.port, self._params.zeroconf_manager, ) except (Exception, CancelledError) as ex: # noqa: BLE001 # Clean up the connection; re-raise the original CancelledError # if the task is actually being cancelled so TaskGroup / # asyncio.timeout semantics are preserved. self._raise_fatal_connection_exception("resolving", ex) finally: self._set_resolve_host_future() self._set_connection_state(CONNECTION_STATE_HOST_RESOLVED) def _set_resolve_host_future(self) -> None: if ( self._resolve_host_future is not None and not self._resolve_host_future.done() ): self._resolve_host_future.set_result(None) self._resolve_host_future = None async def start_connection(self) -> None: """Start the connection process. This part of the process establishes the socket connection but does not initialize the frame helper or send the hello message. """ if self.connection_state is not CONNECTION_STATE_HOST_RESOLVED: msg = "Connection must be in HOST_RESOLVED state to start connection" raise RuntimeError(msg) self._start_connect_future = self._loop.create_future() try: async with interrupt( self._start_connect_future, ConnectionInterruptedError, None ): await self._connect_socket_connect(self._addrs_info) except (Exception, CancelledError) as ex: # noqa: BLE001 # Clean up the connection; re-raise the original CancelledError # if the task is actually being cancelled so TaskGroup / # asyncio.timeout semantics are preserved. self._raise_fatal_connection_exception("starting", ex) finally: self._set_start_connect_future() self._set_connection_state(CONNECTION_STATE_SOCKET_OPENED) def _set_start_connect_future(self) -> None: if ( self._start_connect_future is not None and not self._start_connect_future.done() ): self._start_connect_future.set_result(None) self._start_connect_future = None def _raise_fatal_connection_exception( self, action: str, ex: BaseException ) -> NoReturn: """Report a fatal connection exception and raise the appropriate error. If the awaiting task is actually being cancelled by a parent (for example, a ``TaskGroup`` or ``asyncio.timeout``), the original ``CancelledError`` is re-raised so the task's cancellation state is preserved. Otherwise the exception is wrapped as an ``APIConnectionError`` subclass as before. """ fatal_exc = self._wrap_fatal_connection_exception(action, ex) self.report_fatal_error(fatal_exc) if isinstance(ex, CancelledError) and ( (current_task := asyncio.current_task()) is not None and current_task.cancelling() > 0 ): raise ex raise fatal_exc def _wrap_fatal_connection_exception( self, action: str, ex: BaseException ) -> APIConnectionError: """Ensure a fatal exception is wrapped as as an APIConnectionError.""" if isinstance(ex, APIConnectionError): return ex cause: BaseException | None = None if isinstance(ex, (ConnectionInterruptedError, CancelledError)): err_str = f"{action.title()} connection cancelled" if self._fatal_exception: err_str += f" due to fatal exception: {self._fatal_exception}" cause = self._fatal_exception else: err_str = str(ex) or type(ex).__name__ cause = ex if isinstance(self._fatal_exception, APIConnectionError): klass = type(self._fatal_exception) elif isinstance(ex, CancelledError): klass = APIConnectionCancelledError elif isinstance(ex, OSError): klass = SocketAPIError else: klass = UnhandledAPIConnectionError new_exc = klass(f"Error while {action} connection: {err_str}") new_exc.__cause__ = cause or ex return new_exc async def _do_finish_connect(self, login: bool) -> None: """Finish the connection process.""" # Cache timezone before registering handlers to ensure it's available # when GetTimeRequest is received # Use provided timezone from params (converted from IANA to POSIX if needed), # or fall back to local timezone detection self._cached_timezone = await get_timezone(self._params.timezone) # Register internal handlers before # connecting the helper so we can ensure # we handle any messages that are received immediately self._register_internal_message_handlers() await self._connect_init_frame_helper() await self._connect_hello_login(login) self._async_schedule_keep_alive(self._loop.time()) async def finish_connection(self, *, login: bool) -> None: """Finish the connection process. This part of the process initializes the frame helper and sends the hello message than starts the keep alive process. """ if self.connection_state is not CONNECTION_STATE_SOCKET_OPENED: msg = "Connection must be in SOCKET_OPENED state to finish connection" raise RuntimeError(msg) self._finish_connect_future = self._loop.create_future() try: async with interrupt( self._finish_connect_future, ConnectionInterruptedError, None ): await self._do_finish_connect(login) except (Exception, CancelledError) as ex: # noqa: BLE001 # Clean up the connection; re-raise the original CancelledError # if the task is actually being cancelled so TaskGroup / # asyncio.timeout semantics are preserved. self._raise_fatal_connection_exception("finishing", ex) finally: self._set_finish_connect_future() self._set_connection_state(CONNECTION_STATE_CONNECTED) def _set_finish_connect_future(self) -> None: if ( self._finish_connect_future is not None and not self._finish_connect_future.done() ): self._finish_connect_future.set_result(None) self._finish_connect_future = None def _set_connection_state(self, state: ConnectionState) -> None: """Set the connection state and log the change.""" self.connection_state = state self.is_connected = state is CONNECTION_STATE_CONNECTED self._handshake_complete = ( state is CONNECTION_STATE_HANDSHAKE_COMPLETE or state is CONNECTION_STATE_CONNECTED ) def _make_auth_request(self) -> AuthenticationRequest: """Make a AuthenticationRequest.""" if self._params.password is not None: return AuthenticationRequest(password=self._params.password) return NO_PASSWORD_AUTH_REQUEST def send_message(self, msg: message.Message) -> None: """Send a message to the remote.""" self.send_messages((msg,)) def send_messages(self, msgs: tuple[message.Message, ...]) -> None: """Send a protobuf message to the remote.""" if not self._handshake_complete: error_msg = f"Connection isn't established yet ({self.connection_state})" raise ConnectionNotEstablishedAPIError(error_msg) packets: list[tuple[int, bytes]] = [ (msg_type[0], msg_type[1](msg)) for msg in msgs if (msg_type := PROTO_TO_MESSAGE_TYPE[type(msg)]) ] if self._debug_enabled: for msg in msgs: _LOGGER.debug( "%s: Sending %s: %s", self.log_name, type(msg).__name__, # calling __str__ on the message may crash on # Windows systems due to a bug in the protobuf library # so we call MessageToDict instead MessageToDict(msg) if _WIN32 else msg, ) # Check frame_helper is not None to prevent segfault in Cython code if self._frame_helper is None: raise self._fatal_exception or ConnectionNotEstablishedAPIError( f"Connection is closed ({self.connection_state})" ) try: self._frame_helper.write_packets(packets, self._debug_enabled) except WRITE_EXCEPTIONS as err: # If writing packet fails, we don't know what state the frames # are in anymore and we have to close the connection _LOGGER.info("%s: Error writing packets: %s", self.log_name, err) wrapped_err = SocketClosedAPIError( f"{self.log_name}: Error writing packets: {err}" ) wrapped_err.__cause__ = err self.report_fatal_error(wrapped_err) raise wrapped_err from err def _add_message_callback_without_remove( self, on_message: Callable[[Any], None], msg_types: tuple[type[Any], ...] ) -> None: """Add a message callback without returning a remove callable.""" for msg_type in msg_types: if (handlers := self._message_handlers.get(msg_type)) is None: self._message_handlers[msg_type] = {on_message} else: handlers.add(on_message) def add_message_callback( self, on_message: Callable[[Any], None], msg_types: tuple[type[Any], ...] ) -> Callable[[], None]: """Add a message callback.""" self._add_message_callback_without_remove(on_message, msg_types) return partial(self._remove_message_callback, on_message, msg_types) def _remove_message_callback( self, on_message: Callable[[Any], None], msg_types: tuple[type[Any], ...] ) -> None: """Remove a message callback.""" for msg_type in msg_types: handlers = self._message_handlers[msg_type] handlers.discard(on_message) def send_message_callback_response( self, send_msg: message.Message, on_message: Callable[[Any], None], msg_types: tuple[type[Any], ...], ) -> Callable[[], None]: """Send a message to the remote and register the given message handler.""" self.send_messages((send_msg,)) # Since we do not return control to the event loop (no awaits) # between sending the message and registering the handler # we can be sure that we will not miss any messages even though # we register the handler after sending the message return self.add_message_callback(on_message, msg_types) async def send_messages_await_response_complex( # pylint: disable=too-many-locals self, messages: tuple[message.Message, ...], do_append: Callable[[message.Message], bool] | None, do_stop: Callable[[message.Message], bool] | None, msg_types: tuple[type[Any], ...], timeout: _float, ) -> list[message.Message]: """Send a message to the remote and build up a list response. :param send_msg: The message (request) to send. :param do_append: Predicate to check if a received message is part of the response. :param do_stop: Predicate to check if a received message is the stop response. :param timeout: The maximum amount of time to wait for the stop response. :raises TimeoutAPIError: if a timeout occurred """ # Send the message right away to reduce latency. # This is safe because we are not awaiting between # sending the message and registering the handler self.send_messages(messages) # Unsafe to await between sending the message and registering the handler fut: asyncio.Future[None] = self._loop.create_future() responses: list[message.Message] = [] on_message = partial( _handle_complex_message, fut, responses, do_append, do_stop ) self._add_message_callback_without_remove(on_message, msg_types) self._read_exception_futures.add(fut) # Now safe to await since we have registered the handler # We must not await without a finally or # the message could fail to be removed if the # the await is cancelled timeout_handle = self._loop.call_at( self._loop.time() + timeout, _handle_timeout, fut ) timeout_expired = False try: await fut except TimeoutError as err: timeout_expired = True response_names = message_types_to_names(msg_types) msg = f"Timeout waiting for {response_names} after {timeout}s" raise TimeoutAPIError(msg) from err finally: if not timeout_expired: timeout_handle.cancel() self._remove_message_callback(on_message, msg_types) self._read_exception_futures.discard(fut) return responses async def send_message_await_response( self, send_msg: message.Message, response_type: Any, timeout: _float = 10.0 ) -> Any: [response] = await self.send_messages_await_response_complex( (send_msg,), None, # we will only get responses of `response_type` None, # we will only get responses of `response_type` (response_type,), timeout, ) return response def report_fatal_error(self, err: Exception) -> None: """Report a fatal error that occurred during an operation. This should only be called for errors that mean the connection can no longer be used. The connection will be closed, all exception handlers notified. This method does not log the error, the call site should do so. """ if self._fatal_exception is None: if self._expected_disconnect is False and self._log_errors: # Only log the first error log_level = ( logging.DEBUG if isinstance(err, APIConnectionCancelledError) else logging.WARNING ) _LOGGER.log( log_level, "%s: Connection error occurred: %s", self.log_name, err or type(err), exc_info=not str(err) or isinstance( err, UnhandledAPIConnectionError ), # Log the full stack on empty error string ) # Only set the first error since otherwise the original # error will be lost (ie RequiresEncryptionAPIError) and than # SocketClosedAPIError will be raised instead self._set_fatal_exception_if_unset(err) self._cleanup() def _set_fatal_exception_if_unset(self, err: Exception) -> None: """Set the fatal exception if it hasn't been set yet.""" if self._fatal_exception is None: self._fatal_exception = err def process_packet(self, msg_type_proto: _int, data: _bytes) -> None: # noqa: C901 # Cython-hot path; see CLAUDE.md "Cython gotchas" """Process an incoming packet.""" # This method is HOT and extremely performance critical # since its called for every incoming packet. Take # extra care when modifying this method. # Bounds-check before indexing: msg_type_proto is attacker- # controlled, and `MESSAGE_NUMBER_TO_PROTO[msg_type_proto - 1]` # would otherwise wrap (0 -> -1) into the last registered class # or raise an IndexError on every out-of-range value, both of # which are avoidable on a hot path. if msg_type_proto < 1 or msg_type_proto > _MESSAGE_NUMBER_TO_PROTO_LEN: if self._debug_enabled: _LOGGER.debug( "%s: Skipping unknown message type %s", self.log_name, msg_type_proto, ) return # MESSAGE_NUMBER_TO_PROTO is 0-indexed but the message type is 1-indexed klass, merge = MESSAGE_NUMBER_TO_PROTO[msg_type_proto - 1] msg = klass() try: merge(msg, data) except Exception as e: data_len = len(data) truncated = data[:_MAX_PROTOBUF_ERROR_DATA_BYTES] suffix = ( "" if data_len <= _MAX_PROTOBUF_ERROR_DATA_BYTES else f"...(+{data_len - _MAX_PROTOBUF_ERROR_DATA_BYTES} bytes)" ) _LOGGER.exception( "%s: Invalid protobuf message: type=%s len=%d data=%r%s", self.log_name, klass.__name__, data_len, truncated, suffix, ) self.report_fatal_error( ProtocolAPIError( f"Invalid protobuf message: type={klass.__name__} " f"len={data_len} data={truncated!r}{suffix}: {e}" ) ) raise if self._debug_enabled: _LOGGER.debug( "%s: Got message of type %s (len=%d): %s", self.log_name, type(msg).__name__, len(data), # calling __str__ on the message may crash on # Windows systems due to a bug in the protobuf library # so we call MessageToDict instead MessageToDict(msg) if _WIN32 else msg, ) if self._pong_timer is not None: # Any valid message from the remote cancels the pong timer # as we know the connection is still alive self._async_cancel_pong_timer() if self._send_pending_ping: # Any valid message from the remove cancels the pending ping # since we know the connection is still alive self._send_pending_ping = False if (handlers := self._message_handlers.get(type(msg))) is None: return if len(handlers) > 1: # Handlers are allowed to remove themselves # so we need to copy the set to avoid a # runtime error if the set is modified during # iteration. This can only if there is more # than one handler registered for the message # type. handlers_copy = handlers.copy() for handler in handlers_copy: # Isolate user-callback exceptions so a buggy # handler does not propagate through asyncio's # data_received path and tear the whole session # down. See issue #1755. try: handler(msg) except Exception: _LOGGER.exception( "%s: Unexpected error in message handler %r for %s", self.log_name, handler, type(msg).__name__, ) return # Most common case, only one handler: # no need to copy the set. We still # use a loop here even though there is # only one handler because Cython will # poorly optimize next(iter(handlers)) for handler in handlers: try: handler(msg) except Exception: _LOGGER.exception( "%s: Unexpected error in message handler %r for %s", self.log_name, handler, type(msg).__name__, ) break def _register_internal_message_handlers(self) -> None: """Register internal message handlers.""" self._add_message_callback_without_remove( self._handle_disconnect_request_internal, (DisconnectRequest,) ) self._add_message_callback_without_remove( self._handle_ping_request_internal, (PingRequest,) ) if self._params.provide_time: self._add_message_callback_without_remove( self._handle_get_time_request_internal, (GetTimeRequest,) ) self._add_message_callback_without_remove( self._handle_login_response, (AuthenticationResponse,) ) def _handle_disconnect_request_internal( # pylint: disable=unused-argument self, _msg: DisconnectRequest ) -> None: """Handle a DisconnectRequest.""" # Set _expected_disconnect to True before sending # the response if for some reason sending the response # fails we will still mark the disconnect as expected self._expected_disconnect = True self.send_messages(DISCONNECT_RESPONSE_MESSAGES) self._cleanup() def _handle_ping_request_internal( # pylint: disable=unused-argument self, _msg: PingRequest ) -> None: """Handle a PingRequest.""" self.send_messages(PING_RESPONSE_MESSAGES) def _handle_get_time_request_internal( # pylint: disable=unused-argument self, _msg: GetTimeRequest ) -> None: """Handle a GetTimeRequest.""" resp = GetTimeResponse() resp.epoch_seconds = int(time.time()) resp.timezone = self._cached_timezone parsed_tz = _build_parsed_tz_proto(self._cached_timezone) if parsed_tz is not None: resp.parsed_timezone.CopyFrom(parsed_tz) self.send_messages((resp,)) async def disconnect(self) -> None: """Disconnect from the API.""" if self._finish_connect_future is not None: # Try to wait for the handshake to finish so we can send # a disconnect request. If it doesn't finish in time # we will just close the socket. _, pending = await asyncio.wait( [self._finish_connect_future], timeout=DISCONNECT_CONNECT_TIMEOUT ) if pending: self._set_fatal_exception_if_unset( TimeoutAPIError( "Timed out waiting to finish connect before disconnecting" ) ) if self._debug_enabled: _LOGGER.debug( "%s: Connect task didn't finish before disconnect", self.log_name, ) self._expected_disconnect = True if self._handshake_complete and self._frame_helper is not None: # We still want to send a disconnect request even # if the hello phase isn't finished to ensure we # the esp will clean up the connection as soon # as possible. # But skip if frame_helper is None (connection already cleaned up) try: await self.send_message_await_response( DISCONNECT_REQUEST_MESSAGE, DisconnectResponse, timeout=DISCONNECT_RESPONSE_TIMEOUT, ) except APIConnectionError: _LOGGER.exception("%s: disconnect request failed", self.log_name) self._cleanup() def force_disconnect(self) -> None: """Disconnect from the API immediately, without graceful shutdown.""" self._expected_disconnect = True if self._handshake_complete and self._frame_helper is not None: # Still try to tell the esp to disconnect gracefully # but don't wait for it to finish # But skip if frame_helper is None (connection already cleaned up) try: self.send_messages((DISCONNECT_REQUEST_MESSAGE,)) except APIConnectionError: _LOGGER.exception( "%s: Failed to send (forced) disconnect request", self.log_name, ) self._cleanup()