# -*- coding: utf-8 -*- # Copyright (C) 2010, 2011, 2012, 2013 Sebastian Wiesner # This library is free software; you can redistribute it and/or modify it # under the terms of the GNU Lesser General Public License as published by the # Free Software Foundation; either version 2.1 of the License, or (at your # option) any later version. # This library is distributed in the hope that it will be useful, but WITHOUT # ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or # FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License # for more details. # You should have received a copy of the GNU Lesser General Public License # along with this library; if not, write to the Free Software Foundation, # Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA """ pyudev.monitor ============== Monitor implementation. .. moduleauthor:: Sebastian Wiesner """ from __future__ import (print_function, division, unicode_literals, absolute_import) import os import errno from threading import Thread from functools import partial from .device import Device from ._util import eintr_retry_call from ._util import ensure_byte_string from ._os import pipe from ._os import poll class Monitor(object): """ A synchronous device event monitor. A :class:`Monitor` objects connects to the udev daemon and listens for changes to the device list. A monitor is created by connecting to the kernel daemon through netlink (see :meth:`from_netlink`): >>> from pyudev import Context, Monitor >>> context = Context() >>> monitor = Monitor.from_netlink(context) Once the monitor is created, you can add a filter using :meth:`filter_by()` or :meth:`filter_by_tag()` to drop incoming events in subsystems, which are not of interest to the application: >>> monitor.filter_by('input') When the monitor is eventually set up, you can either poll for events synchronously: >>> device = monitor.poll(timeout=3) >>> if device: ... print('{0.action}: {0}'.format(device)) ... Or you can monitor events asynchronously with :class:`MonitorObserver`. To integrate into various event processing frameworks, the monitor provides a :func:`selectable ` file description by :meth:`fileno()`. However, do *not* read or write directly on this file descriptor. Instances of this class can directly be given as ``udev_monitor *`` to functions wrapped through :mod:`ctypes`. .. versionchanged:: 0.16 Remove :meth:`from_socket()` which is deprecated, and even removed in recent udev versions. """ def __init__(self, context, monitor_p): self.context = context self._as_parameter_ = monitor_p self._libudev = context._libudev self._started = False def __del__(self): self._libudev.udev_monitor_unref(self) @classmethod def from_netlink(cls, context, source='udev'): """ Create a monitor by connecting to the kernel daemon through netlink. ``context`` is the :class:`Context` to use. ``source`` is a string, describing the event source. Two sources are available: ``'udev'`` (the default) Events emitted after udev as registered and configured the device. This is the absolutely recommended source for applications. ``'kernel'`` Events emitted directly after the kernel has seen the device. The device has not yet been configured by udev and might not be usable at all. **Never** use this, unless you know what you are doing. Return a new :class:`Monitor` object, which is connected to the given source. Raise :exc:`~exceptions.ValueError`, if an invalid source has been specified. Raise :exc:`~exceptions.EnvironmentError`, if the creation of the monitor failed. """ if source not in ('kernel', 'udev'): raise ValueError('Invalid source: {0!r}. Must be one of "udev" ' 'or "kernel"'.format(source)) monitor = context._libudev.udev_monitor_new_from_netlink( context, ensure_byte_string(source)) if not monitor: raise EnvironmentError('Could not create udev monitor') return cls(context, monitor) @property def started(self): """ ``True``, if this monitor was started, ``False`` otherwise. Readonly. .. seealso:: :meth:`start()` .. versionadded:: 0.16 """ return self._started def fileno(self): # pylint: disable=anomalous-backslash-in-string """ Return the file description associated with this monitor as integer. This is really a real file descriptor ;), which can be watched and :func:`select.select`\ ed. """ return self._libudev.udev_monitor_get_fd(self) def filter_by(self, subsystem, device_type=None): """ Filter incoming events. ``subsystem`` is a byte or unicode string with the name of a subsystem (e.g. ``'input'``). Only events originating from the given subsystem pass the filter and are handed to the caller. If given, ``device_type`` is a byte or unicode string specifying the device type. Only devices with the given device type are propagated to the caller. If ``device_type`` is not given, no additional filter for a specific device type is installed. These filters are executed inside the kernel, and client processes will usually not be woken up for device, that do not match these filters. .. versionchanged:: 0.15 This method can also be after :meth:`start()` now. """ subsystem = ensure_byte_string(subsystem) if device_type is not None: device_type = ensure_byte_string(device_type) self._libudev.udev_monitor_filter_add_match_subsystem_devtype( self, subsystem, device_type) self._libudev.udev_monitor_filter_update(self) def filter_by_tag(self, tag): """ Filter incoming events by the given ``tag``. ``tag`` is a byte or unicode string with the name of a tag. Only events for devices which have this tag attached pass the filter and are handed to the caller. Like with :meth:`filter_by` this filter is also executed inside the kernel, so that client processes are usually not woken up for devices without the given ``tag``. .. udevversion:: 154 .. versionadded:: 0.9 .. versionchanged:: 0.15 This method can also be after :meth:`start()` now. """ self._libudev.udev_monitor_filter_add_match_tag( self, ensure_byte_string(tag)) self._libudev.udev_monitor_filter_update(self) def remove_filter(self): """ Remove any filters installed with :meth:`filter_by()` or :meth:`filter_by_tag()` from this monitor. .. warning:: Up to udev 181 (and possibly even later versions) the underlying ``udev_monitor_filter_remove()`` seems to be broken. If used with affected versions this method always raises :exc:`~exceptions.ValueError`. Raise :exc:`~exceptions.EnvironmentError` if removal of installed filters failed. .. versionadded:: 0.15 """ self._libudev.udev_monitor_filter_remove(self) self._libudev.udev_monitor_filter_update(self) def enable_receiving(self): """ Switch the monitor into listing mode. Connect to the event source and receive incoming events. Only after calling this method, the monitor listens for incoming events. .. note:: This method is implicitly called by :meth:`__iter__`. You don't need to call it explicitly, if you are iterating over the monitor. .. deprecated:: 0.16 Will be removed in 1.0. Use :meth:`start()` instead. """ import warnings warnings.warn('Will be removed in 1.0. Use Monitor.start() instead.', DeprecationWarning) self.start() def start(self): """ Start this monitor. The monitor will not receive events until this method is called. This method does nothing if called on an already started :class:`Monitor`. .. note:: Typically you don't need to call this method. It is implicitly called by :meth:`poll()` and :meth:`__iter__()`. .. seealso:: :attr:`started` .. versionchanged:: 0.16 This method does nothing if the :class:`Monitor` was already started. """ if not self._started: self._libudev.udev_monitor_enable_receiving(self) # Force monitor FD into non-blocking mode pipe.set_fd_status_flag(self, os.O_NONBLOCK) self._started = True def set_receive_buffer_size(self, size): """ Set the receive buffer ``size``. ``size`` is the requested buffer size in bytes, as integer. .. note:: The CAP_NET_ADMIN capability must be contained in the effective capability set of the caller for this method to succeed. Otherwise :exc:`~exceptions.EnvironmentError` will be raised, with ``errno`` set to :data:`~errno.EPERM`. Unprivileged processes typically lack this capability. You can check the capabilities of the current process with the python-prctl_ module: >>> import prctl >>> prctl.cap_effective.net_admin Raise :exc:`~exceptions.EnvironmentError`, if the buffer size could not bet set. .. versionadded:: 0.13 .. _python-prctl: http://packages.python.org/python-prctl """ self._libudev.udev_monitor_set_receive_buffer_size(self, size) def _receive_device(self): """Receive a single device from the monitor. Return the received :class:`Device`, or ``None`` if no device could be received. """ while True: try: device_p = self._libudev.udev_monitor_receive_device(self) return Device(self.context, device_p) if device_p else None except EnvironmentError as error: if error.errno in (errno.EAGAIN, errno.EWOULDBLOCK): # No data available return None elif error.errno == errno.EINTR: # Try again if our system call was interrupted continue else: raise def poll(self, timeout=None): """ Poll for a device event. You can use this method together with :func:`iter()` to synchronously monitor events in the current thread:: for device in iter(monitor.poll, None): print('{0.action} on {0.device_path}'.format(device)) Since this method will never return ``None`` if no ``timeout`` is specified, this is effectively an endless loop. With :func:`functools.partial()` you can also create a loop that only waits for a specified time:: for device in iter(partial(monitor.poll, 3), None): print('{0.action} on {0.device_path}'.format(device)) This loop will only wait three seconds for a new device event. If no device event occurred after three seconds, the loop will exit. ``timeout`` is a floating point number that specifies a time-out in seconds. If omitted or ``None``, this method blocks until a device event is available. If ``0``, this method just polls and will never block. .. note:: This method implicitly calls :meth:`start()`. Return the received :class:`Device`, or ``None`` if a timeout occurred. Raise :exc:`~exceptions.EnvironmentError` if event retrieval failed. .. seealso:: :attr:`Device.action` The action that created this event. :attr:`Device.sequence_number` The sequence number of this event. .. versionadded:: 0.16 """ if timeout is not None and timeout > 0: # .poll() takes timeout in milliseconds timeout = int(timeout * 1000) self.start() if eintr_retry_call(poll.Poll.for_events((self, 'r')).poll, timeout): return self._receive_device() else: return None def receive_device(self): """ Receive a single device from the monitor. .. warning:: You *must* call :meth:`start()` before calling this method. The caller must make sure, that there are events available in the event queue. The call blocks, until a device is available. If a device was available, return ``(action, device)``. ``device`` is the :class:`Device` object describing the device. ``action`` is a string describing the action. Usual actions are: ``'add'`` A device has been added (e.g. a USB device was plugged in) ``'remove'`` A device has been removed (e.g. a USB device was unplugged) ``'change'`` Something about the device changed (e.g. a device property) ``'online'`` The device is online now ``'offline'`` The device is offline now Raise :exc:`~exceptions.EnvironmentError`, if no device could be read. .. deprecated:: 0.16 Will be removed in 1.0. Use :meth:`Monitor.poll()` instead. """ import warnings warnings.warn('Will be removed in 1.0. Use Monitor.poll() instead.', DeprecationWarning) device = self.poll() return device.action, device def __iter__(self): """ Wait for incoming events and receive them upon arrival. This methods implicitly calls :meth:`start()`, and starts polling the :meth:`fileno` of this monitor. If a event comes in, it receives the corresponding device and yields it to the caller. The returned iterator is endless, and continues receiving devices without ever stopping. Yields ``(action, device)`` (see :meth:`receive_device` for a description). .. deprecated:: 0.16 Will be removed in 1.0. Use an explicit loop over :meth:`poll()` instead, or monitor asynchronously with :class:`MonitorObserver`. """ import warnings warnings.warn('Will be removed in 1.0. Use an explicit loop over ' '"poll()" instead, or monitor asynchronously with ' '"MonitorObserver".', DeprecationWarning) self.start() while True: device = self.poll() if device is not None: yield device.action, device class MonitorObserver(Thread): """ An asynchronous observer for device events. This class subclasses :class:`~threading.Thread` class to asynchronously observe a :class:`Monitor` in a background thread: >>> from pyudev import Context, Monitor, MonitorObserver >>> context = Context() >>> monitor = Monitor.from_netlink(context) >>> monitor.filter_by(subsystem='input') >>> def print_device_event(device): ... print('background event {0.action}: {0.device_path}'.format(device)) >>> observer = MonitorObserver(monitor, callback=print_device_event, name='monitor-observer') >>> observer.daemon True >>> observer.start() In the above example, input device events will be printed in background, until :meth:`stop()` is called on ``observer``. .. note:: Instances of this class are always created as daemon thread. If you do not want to use daemon threads for monitoring, you need explicitly set :attr:`~threading.Thread.daemon` to ``False`` before invoking :meth:`~threading.Thread.start()`. .. seealso:: :attr:`Device.action` The action that created this event. :attr:`Device.sequence_number` The sequence number of this event. .. versionadded:: 0.14 .. versionchanged:: 0.15 :meth:`Monitor.start()` is implicitly called when the thread is started. """ def __init__(self, monitor, event_handler=None, callback=None, *args, **kwargs): """ Create a new observer for the given ``monitor``. ``monitor`` is the :class:`Monitor` to observe. ``callback`` is the callable to invoke on events, with the signature ``callback(device)`` where ``device`` is the :class:`Device` that caused the event. .. warning:: ``callback`` is invoked in the observer thread, hence the observer is blocked while callback executes. ``args`` and ``kwargs`` are passed unchanged to the constructor of :class:`~threading.Thread`. .. deprecated:: 0.16 The ``event_handler`` argument will be removed in 1.0. Use the ``callback`` argument instead. .. versionchanged:: 0.16 Add ``callback`` argument. """ if callback is None and event_handler is None: raise ValueError('callback missing') elif callback is not None and event_handler is not None: raise ValueError('Use either callback or event handler') Thread.__init__(self, *args, **kwargs) self.monitor = monitor # observer threads should not keep the interpreter alive self.daemon = True self._stop_event = None if event_handler is not None: import warnings warnings.warn('"event_handler" argument will be removed in 1.0. ' 'Use Monitor.poll() instead.', DeprecationWarning) callback = lambda d: event_handler(d.action, d) self._callback = callback def start(self): """Start the observer thread.""" if not self.is_alive(): self._stop_event = pipe.Pipe.open() Thread.start(self) def run(self): self.monitor.start() notifier = poll.Poll.for_events( (self.monitor, 'r'), (self._stop_event.source, 'r')) while True: for file_descriptor, event in eintr_retry_call(notifier.poll): if file_descriptor == self._stop_event.source.fileno(): # in case of a stop event, close our pipe side, and # return from the thread self._stop_event.source.close() return elif file_descriptor == self.monitor.fileno() and event == 'r': read_device = partial(eintr_retry_call, self.monitor.poll, timeout=0) for device in iter(read_device, None): self._callback(device) else: raise EnvironmentError('Observed monitor hung up') def send_stop(self): """ Send a stop signal to the background thread. The background thread will eventually exit, but it may still be running when this method returns. This method is essentially the asynchronous equivalent to :meth:`stop()`. .. note:: The underlying :attr:`monitor` is *not* stopped. """ if self._stop_event is None: return with self._stop_event.sink: # emit a stop event to the thread eintr_retry_call(self._stop_event.sink.write, b'\x01') self._stop_event.sink.flush() def stop(self): """ Synchronously stop the background thread. .. note:: This method can safely be called from the observer thread. In this case it is equivalent to :meth:`send_stop()`. Send a stop signal to the backgroud (see :meth:`send_stop`), and waits for the background thread to exit (see :meth:`~threading.Thread.join`) if the current thread is *not* the observer thread. After this method returns in a thread *that is not the observer thread*, the ``callback`` is guaranteed to not be invoked again anymore. .. note:: The underlying :attr:`monitor` is *not* stopped. .. versionchanged:: 0.16 This method can be called from the observer thread. """ self.send_stop() try: self.join() except RuntimeError: pass