core-admin/qubes/vm/qubesvm.py

2288 lines
82 KiB
Python
Raw Normal View History

#
2015-01-19 18:03:23 +01:00
# The Qubes OS Project, https://www.qubes-os.org/
#
2015-01-19 18:03:23 +01:00
# Copyright (C) 2010-2015 Joanna Rutkowska <joanna@invisiblethingslab.com>
# Copyright (C) 2013-2015 Marek Marczykowski-Górecki
# <marmarek@invisiblethingslab.com>
# Copyright (C) 2014-2015 Wojtek Porczyk <woju@invisiblethingslab.com>
#
# 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, see <https://www.gnu.org/licenses/>.
#
import asyncio
2016-01-29 17:56:33 +01:00
import base64
import grp
import os
import os.path
import shutil
import string
import subprocess
import uuid
import libvirt # pylint: disable=import-error
import lxml
2014-11-13 14:38:41 +01:00
import qubes
import qubes.config
import qubes.exc
import qubes.storage
2016-06-02 22:02:06 +02:00
import qubes.storage.file
import qubes.utils
2014-11-13 14:38:41 +01:00
import qubes.vm
import qubes.vm.mix.net
qmemman_present = False
try:
2016-06-02 22:02:06 +02:00
import qubes.qmemman.client # pylint: disable=wrong-import-position
2019-10-20 12:21:09 +02:00
qmemman_present = True
except ImportError:
pass
# overhead of per-qube/per-vcpu Xen structures,
# taken from OpenStack nova/virt/xenapi/driver.py
# see https://wiki.openstack.org/wiki/XenServer/Overhead
# add an extra MB because Nova rounds up to MBs
2016-06-02 22:02:06 +02:00
MEM_OVERHEAD_BASE = (3 + 1) * 1024 * 1024
MEM_OVERHEAD_PER_VCPU = 3 * 1024 * 1024 / 2
def _setter_kernel(self, prop, value):
2019-10-20 12:21:09 +02:00
""" Helper for setting the domain kernel and running sanity checks on it.
""" # pylint: disable=unused-argument
if not value:
return ''
value = str(value)
if '/' in value:
2019-10-20 12:21:09 +02:00
raise qubes.exc.QubesPropertyValueError(
self, prop, value,
'Kernel name cannot contain \'/\'')
return value
def _setter_positive_int(self, prop, value):
2019-10-20 12:21:09 +02:00
""" Helper for setting a positive int. Checks that the int is > 0 """
2016-06-02 22:02:06 +02:00
# pylint: disable=unused-argument
value = int(value)
if value <= 0:
raise ValueError('Value must be positive')
return value
2019-10-20 12:21:09 +02:00
def _setter_non_negative_int(self, prop, value):
2019-10-20 12:21:09 +02:00
""" Helper for setting a positive int. Checks that the int is >= 0 """
# pylint: disable=unused-argument
value = int(value)
if value < 0:
raise ValueError('Value must be positive or zero')
return value
def _setter_default_user(self, prop, value):
2019-10-20 12:21:09 +02:00
""" Helper for setting default user """
value = str(value)
2019-10-20 12:21:09 +02:00
# specifically forbid: ':', ' ', """, '"'
allowed_chars = string.ascii_letters + string.digits + '_-+,.'
if not all(c in allowed_chars for c in value):
2019-10-20 12:21:09 +02:00
raise qubes.exc.QubesPropertyValueError(
self, prop, value,
'Username can contain only those characters: ' + allowed_chars)
return value
2019-10-20 12:21:09 +02:00
def _setter_virt_mode(self, prop, value):
value = str(value)
value = value.lower()
2017-10-02 22:23:27 +02:00
if value not in ('hvm', 'pv', 'pvh'):
2019-10-20 12:21:09 +02:00
raise qubes.exc.QubesPropertyValueError(
self, prop, value,
2017-10-02 22:23:27 +02:00
'Invalid virtualization mode, supported values: hvm, pv, pvh')
if value == 'pvh' and list(self.devices['pci'].persistent()):
2019-10-20 12:21:09 +02:00
raise qubes.exc.QubesPropertyValueError(
self, prop, value,
"pvh mode can't be set if pci devices are attached")
return value
2019-10-20 12:21:09 +02:00
def _default_virt_mode(self):
if self.devices['pci'].persistent():
return 'hvm'
try:
return self.template.virt_mode
except AttributeError:
return 'pvh'
2019-10-20 12:21:09 +02:00
def _default_with_template(prop, default):
2019-10-20 12:21:09 +02:00
"""Return a callable for 'default' argument of a property. Use a value
from a template (if any), otherwise *default*
2019-10-20 12:21:09 +02:00
"""
def _func(self):
try:
return getattr(self.template, prop)
except AttributeError:
if callable(default):
return default(self)
return default
return _func
def _default_maxmem(self):
# first check for any reason to _not_ enable qmemman
if not self.is_memory_balancing_possible():
return 0
# Linux specific cap: max memory can't scale beyond 10.79*init_mem
# see https://groups.google.com/forum/#!topic/qubes-devel/VRqkFj1IOtA
if self.features.get('os', None) == 'Linux':
default_maxmem = self.memory * 10
else:
default_maxmem = 4000
# don't use default larger than half of physical ram
default_maxmem = min(default_maxmem,
2019-10-20 12:21:09 +02:00
int(self.app.host.memory_total / 1024 / 2))
return _default_with_template('maxmem', default_maxmem)(self)
2019-10-20 12:21:09 +02:00
def _default_kernelopts(self):
2019-10-20 12:21:09 +02:00
"""
Return default kernel options for the given kernel. If kernel directory
contains 'default-kernelopts-{pci,nopci}.txt' file, use that. Otherwise
use built-in defaults.
For qubes without PCI devices, kernelopts of qube's template are
considered (for template-based qubes).
2019-10-20 12:21:09 +02:00
"""
if not self.kernel:
return ''
if 'kernel' in self.volumes:
kernels_dir = self.storage.kernels_dir
else:
kernels_dir = os.path.join(
qubes.config.system_path['qubes_kernels_base_dir'],
self.kernel)
pci = bool(list(self.devices['pci'].persistent()))
if pci:
path = os.path.join(kernels_dir, 'default-kernelopts-pci.txt')
else:
try:
return self.template.kernelopts
except AttributeError:
pass
path = os.path.join(kernels_dir, 'default-kernelopts-nopci.txt')
if os.path.exists(path):
with open(path) as f_kernelopts:
return f_kernelopts.read().strip()
else:
return (qubes.config.defaults['kernelopts_pcidevs'] if pci else
2019-10-20 12:21:09 +02:00
qubes.config.defaults['kernelopts'])
class QubesVM(qubes.vm.mix.net.NetVMMixin, qubes.vm.BaseVM):
2019-10-20 12:21:09 +02:00
"""Base functionality of Qubes VM shared between all VMs.
The following events are raised on this class or its subclasses:
.. event:: domain-init (subject, event)
Fired at the end of class' constructor.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-init'``)
.. event:: domain-load (subject, event)
Fired after the qube was loaded from :file:`qubes.xml`
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-loaded'``)
2016-10-25 17:27:02 +02:00
.. event:: domain-pre-start \
(subject, event, start_guid, mem_required)
Fired at the beginning of :py:meth:`start` method.
Handler for this event can be asynchronous (a coroutine).
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-pre-start'``)
*other arguments are as in :py:meth:`start`*
.. event:: domain-spawn (subject, event, start_guid)
Fired after creating libvirt domain.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-spawn'``)
Handler for this event can be asynchronous (a coroutine).
*other arguments are as in :py:meth:`start`*
.. event:: domain-start (subject, event, start_guid)
Fired at the end of :py:meth:`start` method.
Handler for this event can be asynchronous (a coroutine).
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-start'``)
*other arguments are as in :py:meth:`start`*
2018-07-11 04:50:37 +02:00
.. event:: domain-start-failed (subject, event, reason)
Fired when :py:meth:`start` method fails.
*reason* argument is a textual error message.
Handler for this event can be asynchronous (a coroutine).
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-start-failed'``)
.. event:: domain-paused (subject, event)
Fired when the domain has been paused.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-paused'``)
.. event:: domain-unpaused (subject, event)
Fired when the domain has been unpaused.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-unpaused'``)
.. event:: domain-stopped (subject, event)
Fired when domain has been stopped.
This event is emitted before ``'domain-shutdown'`` and will trigger
the cleanup in QubesVM. So if you require that the cleanup has
already run use ``'domain-shutdown'``.
Note that you can receive this event as soon as you received
``'domain-pre-start'``. This also can be emitted in case of a
startup failure, before or after ``'domain-start-failed'``.
Handler for this event can be asynchronous (a coroutine).
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-stopped'``)
.. event:: domain-shutdown (subject, event)
Fired when domain has been shut down. It is generated after
``'domain-stopped'``.
Note that you can receive this event as soon as you received
``'domain-pre-start'``. This also can be emitted in case of a
startup failure, before or after ``'domain-start-failed'``.
Handler for this event can be asynchronous (a coroutine).
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-shutdown'``)
.. event:: domain-pre-shutdown (subject, event, force)
Fired at the beginning of :py:meth:`shutdown` method.
Handler for this event can be asynchronous (a coroutine).
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-pre-shutdown'``)
:param force: If the shutdown is to be forceful
.. event:: domain-shutdown-failed (subject, event, reason)
Fired when ``domain-pre-shutdown`` event was sent, but the actual
shutdown operation failed. It can be caused by other
``domain-pre-shutdown`` handler blocking the operation with an
exception, or a shutdown timeout.
Handler for this event can be asynchronous (a coroutine).
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-shutdown-failed'``)
:param reason: Error message
.. event:: domain-cmd-pre-run (subject, event, start_guid)
Fired at the beginning of :py:meth:`run_service` method.
Handler for this event can be asynchronous (a coroutine).
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-cmd-pre-run'``)
:param start_guid: If the gui daemon can be started
.. event:: domain-create-on-disk (subject, event)
Fired at the end of :py:meth:`create_on_disk` method.
Handler for this event can be asynchronous (a coroutine).
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-create-on-disk'``)
.. event:: domain-remove-from-disk (subject, event)
Fired at the beginning of :py:meth:`remove_from_disk` method, before
the qube directory is removed.
Handler for this event can be asynchronous (a coroutine).
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-remove-from-disk'``)
.. event:: domain-clone-files (subject, event, src)
Fired at the end of :py:meth:`clone_disk_files` method.
Handler for this event can be asynchronous (a coroutine).
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-clone-files'``)
:param src: source qube
.. event:: domain-verify-files (subject, event)
Fired at the end of :py:meth:`clone_disk_files` method.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-verify-files'``)
If you think some files are missing or damaged, raise an exception.
.. event:: domain-is-fully-usable (subject, event)
Fired at the end of :py:meth:`clone_disk_files` method.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-is-fully-usable'``)
You may ``yield False`` from the handler if you think the qube is
not fully usable. This will cause the domain to be in "transient"
state in the domain lifecycle.
2017-05-23 15:35:21 +02:00
.. event:: domain-qdb-create (subject, event)
Fired at the end of :py:meth:`create_qdb_entries` method.
:param subject: Event emitter (the qube object)
2017-05-23 15:35:21 +02:00
:param event: Event name (``'domain-qdb-create'``)
This event is a good place to add your custom entries to the qdb.
.. event:: domain-qdb-change:watched-path (subject, event, path)
Fired when watched QubesDB entry is changed. See
:py:meth:`watch_qdb_path`. *watched-path* part of event name is
what path was registered for watching, *path* in event argument
is what actually have changed (which may be different if watching a
directory, i.e. a path with `/` at the end).
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-qdb-change'``)
:param path: changed QubesDB path
.. event:: backup-get-files (subject, event)
Collects additional file to be included in a backup.
:param subject: Event emitter (the qube object)
:param event: Event name (``'backup-get-files'``)
Handlers should yield paths of the files.
.. event:: domain-restore (subject, event)
Domain was just restored from backup, although the storage was not
yet verified and the app object was not yet saved.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-restore'``)
.. event:: domain-feature-pre-set:feature (subject, event, feature,
value [, oldvalue])
A feature will be changed. This event is fired before value is set.
If any handler raises an exception, value will not be set.
*oldvalue* is present only when there was any.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-feature-pre-set:' feature``)
:param feature: feature name
:param value: new value
:param oldvalue: old value, if any
.. event:: domain-feature-set:feature (subject, event, feature, value
[, oldvalue])
A feature was changed. This event is fired before bare
`domain-feature-set` event.
*oldvalue* is present only when there was any.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-feature-set:' feature``)
:param feature: feature name
:param value: new value
:param oldvalue: old value, if any
.. event:: domain-feature-delete:feature (subject, event, feature)
A feature was removed. This event is fired before bare
`domain-feature-delete` event.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-feature-delete:' feature``)
:param feature: feature name
.. event:: domain-feature-pre-delete:feature (subject, event, feature)
A feature will be removed. This event is fired before feature is
removed. If any handler raises an exception,feature will not be
removed.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-feature-pre-delete:' feature``)
:param feature: feature name
.. event:: domain-tag-add:tag (subject, event, tag)
A tag was added.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-tag-add:' tag``)
:param tag: tag name
.. event:: domain-tag-delete:tag (subject, event, tag)
A feature was removed.
:param subject: Event emitter (the qube object)
:param event: Event name (``'domain-tag-delete:' tag``)
:param tag: tag name
.. event:: features-request (subject, event, *, untrusted_features)
The domain is performing a features request.
:param subject: Event emitter (the qube object)
:param event: Event name (``'features-request'``)
:param untrusted_features: :py:class:`dict` containing the feature \
request
The content of the `untrusted_features` variable is, as the name
implies, **UNTRUSTED**. The remind this to programmer, the variable
name has to be exactly as provided.
It is up to the extensions to decide, what to do with request,
ranging from plainly ignoring the request to verbatim copy into
:py:attr:`features` with only minimal sanitisation.
Handler for this event can be asynchronous (a coroutine).
.. event:: firewall-changed (subject, event)
Firewall was changed.
:param subject: Event emitter (the qube object)
:param event: Event name (``'firewall-changed'``)
.. event:: net-domain-connect (subject, event, vm)
Fired after connecting a domiain to this vm.
:param subject: Event emitter (the qube object)
:param event: Event name (``'net-domain-connect'``)
:param vm: The domain that was just connected.
On the `vm` object there was probably ``property-set:netvm`` fired
earlier.
.. event:: template-postinstall (subject, event)
Fired on non-template-based domain (TemplateVM, StandaloneVM) when
it first reports qrexec presence. This happens at the first
domain startup just after its installation and is suitable for
performing various post-installation setup.
Handler for this event can be asynchronous (a coroutine).
2019-10-20 12:21:09 +02:00
"""
#
# per-class properties
#
#: directory in which domains of this class will reside
dir_path_prefix = qubes.config.system_path['qubes_appvms_dir']
#
# properties loaded from XML
#
2019-10-20 12:44:27 +02:00
guivm = qubes.VMProperty('guivm', load_stage=4, allow_none=True,
default=(lambda self: self.app.default_guivm),
doc='VM used for Gui')
2020-02-27 10:31:27 +01:00
audiovm = qubes.VMProperty('audiovm', load_stage=4, allow_none=True,
default=(lambda self: self.app.default_audiovm),
doc='VM used for Audio')
2019-10-20 12:21:09 +02:00
virt_mode = qubes.property(
'virt_mode',
type=str, setter=_setter_virt_mode,
default=_default_virt_mode,
2019-10-20 12:21:09 +02:00
doc="""Virtualisation mode: full virtualisation ("HVM"),
or paravirtualisation ("PV"), or hybrid ("PVH").
TemplateBasedVMs use its template\'s value by default.""")
2016-03-02 12:17:29 +01:00
2019-10-20 12:21:09 +02:00
installed_by_rpm = qubes.property(
'installed_by_rpm',
type=bool, setter=qubes.property.bool,
default=False,
2019-10-20 12:21:09 +02:00
doc="""If this domain's image was installed from package tracked by
package manager.""")
2019-10-20 12:21:09 +02:00
memory = qubes.property(
'memory', type=int,
setter=_setter_positive_int,
2019-10-20 12:21:09 +02:00
default=_default_with_template(
'memory',
lambda self:
qubes.config.defaults[
'hvm_memory' if self.virt_mode == 'hvm' else 'memory']),
doc='Memory currently available for this VM. TemplateBasedVMs use its '
'template\'s value by default.')
2019-10-20 12:21:09 +02:00
maxmem = qubes.property(
'maxmem', type=int,
setter=_setter_non_negative_int,
default=_default_maxmem,
2019-10-20 12:21:09 +02:00
doc="""Maximum amount of memory available for this VM (for the purpose
of the memory balancer). Set to 0 to disable memory balancing for
this qube. TemplateBasedVMs use its template\'s value by default
2019-10-20 12:21:09 +02:00
(unless memory balancing not supported for this qube).""")
2019-10-20 12:21:09 +02:00
stubdom_mem = qubes.property(
'stubdom_mem', type=int,
setter=_setter_positive_int,
default=None,
2018-12-07 02:43:39 +01:00
doc='Memory amount allocated for the stubdom')
2019-10-20 12:21:09 +02:00
vcpus = qubes.property(
'vcpus',
type=int,
setter=_setter_positive_int,
default=_default_with_template('vcpus', 2),
doc='Number of virtual CPUs for a qube. TemplateBasedVMs use its '
'template\'s value by default.')
# CORE2: swallowed uses_default_kernel
2019-10-20 12:21:09 +02:00
kernel = qubes.property(
'kernel', type=str,
setter=_setter_kernel,
default=_default_with_template('kernel',
2019-10-20 12:21:09 +02:00
lambda self: self.app.default_kernel),
doc='Kernel used by this domain. TemplateBasedVMs use its '
'template\'s value by default.')
# CORE2: swallowed uses_default_kernelopts
# pylint: disable=no-member
2019-10-20 12:21:09 +02:00
kernelopts = qubes.property(
'kernelopts', type=str, load_stage=4,
default=_default_kernelopts,
doc='Kernel command line passed to domain. TemplateBasedVMs use its '
'template\'s value by default.')
2019-10-20 12:21:09 +02:00
debug = qubes.property(
'debug', type=bool, default=False,
setter=qubes.property.bool,
doc='Turns on debugging features.')
# XXX what this exactly does?
# XXX shouldn't this go to standalone VM and TemplateVM, and leave here
# only plain property?
2019-10-20 12:21:09 +02:00
default_user = qubes.property(
'default_user', type=str,
2017-04-15 23:48:02 +02:00
# pylint: disable=no-member
2019-10-20 12:21:09 +02:00
default=_default_with_template('default_user',
'user'),
setter=_setter_default_user,
doc='Default user to start applications as. TemplateBasedVMs use its '
'template\'s value by default.')
2019-10-20 12:21:09 +02:00
qrexec_timeout = qubes.property(
'qrexec_timeout', type=int,
default=_default_with_template(
'qrexec_timeout',
lambda self: self.app.default_qrexec_timeout),
setter=_setter_positive_int,
2019-10-20 12:21:09 +02:00
doc="""Time in seconds after which qrexec connection attempt is deemed
failed. Operating system inside VM should be able to boot in this
2019-10-20 12:21:09 +02:00
time.""")
2019-10-20 12:21:09 +02:00
shutdown_timeout = qubes.property(
'shutdown_timeout', type=int,
default=_default_with_template(
'shutdown_timeout',
lambda self: self.app.default_shutdown_timeout),
setter=_setter_positive_int,
2019-10-20 12:21:09 +02:00
doc="""Time in seconds for shutdown of the VM, after which VM may be
forcefully powered off. Operating system inside VM should be
2019-10-20 12:21:09 +02:00
able to fully shutdown in this time.""")
2019-10-20 12:21:09 +02:00
autostart = qubes.property(
'autostart', default=False,
type=bool, setter=qubes.property.bool,
2019-10-20 12:21:09 +02:00
doc="""Setting this to `True` means that VM should be autostarted on
dom0 boot.""")
2019-10-20 12:21:09 +02:00
include_in_backups = qubes.property(
'include_in_backups',
default=True,
type=bool, setter=qubes.property.bool,
doc='If this domain is to be included in default backup.')
2019-10-20 12:21:09 +02:00
backup_timestamp = qubes.property(
'backup_timestamp', default=None,
type=int,
doc='Time of last backup of the qube, in seconds since unix epoch')
2019-10-20 12:21:09 +02:00
default_dispvm = qubes.VMProperty(
'default_dispvm',
load_stage=4,
allow_none=True,
2019-10-20 12:21:09 +02:00
default=(
lambda self: self.app.default_dispvm),
doc='Default VM to be used as Disposable VM for service calls.')
2019-10-20 12:21:09 +02:00
management_dispvm = qubes.VMProperty(
'management_dispvm',
load_stage=4,
allow_none=True,
2019-10-20 12:21:09 +02:00
default=_default_with_template(
'management_dispvm',
(lambda self: self.app.management_dispvm)),
doc='Default DVM template for Disposable VM for managing this VM.')
2019-10-20 12:21:09 +02:00
updateable = qubes.property(
'updateable',
default=(lambda self: not hasattr(self, 'template')),
type=bool,
setter=qubes.property.forbidden,
doc='True if this machine may be updated on its own.')
#
# static, class-wide properties
#
#
# properties not loaded from XML, calculated at run-time
#
def __str__(self):
return self.name
# VMM-related
@qubes.stateless_property
def xid(self):
2019-10-20 12:21:09 +02:00
"""Xen ID.
Or not Xen, but ID.
2019-10-20 12:21:09 +02:00
"""
if self.libvirt_domain is None:
return -1
try:
2020-02-27 10:31:28 +01:00
if self.is_running():
return self.libvirt_domain.ID()
return -1
except libvirt.libvirtError as e:
if e.get_error_code() == libvirt.VIR_ERR_NO_DOMAIN:
return -1
self.log.exception('libvirt error code: {!r}'.format(
e.get_error_code()))
raise
@qubes.stateless_property
def stubdom_xid(self):
if not self.is_running():
return -1
if self.app.vmm.xs is None:
return -1
2019-10-20 12:21:09 +02:00
stubdom_xid_str = self.app.vmm.xs.read(
'', '/local/domain/{}/image/device-model-domid'.format(
self.xid))
if stubdom_xid_str is None or not stubdom_xid_str.isdigit():
return -1
return int(stubdom_xid_str)
2016-06-21 17:37:58 +02:00
@property
def attached_volumes(self):
result = []
xml_desc = self.libvirt_domain.XMLDesc()
xml = lxml.etree.fromstring(xml_desc)
for disk in xml.xpath("//domain/devices/disk"):
if disk.find('backenddomain') is not None:
pool_name = 'p_%s' % disk.find('backenddomain').get('name')
pool = self.app.pools[pool_name]
vid = disk.find('source').get('dev').split('/dev/')[1]
for volume in pool.volumes:
if volume.vid == vid:
result += [volume]
break
2017-03-01 21:50:06 +01:00
return result + list(self.volumes.values())
@property
def libvirt_domain(self):
2019-10-20 12:21:09 +02:00
"""Libvirt domain object from libvirt.
May be :py:obj:`None`, if libvirt knows nothing about this domain.
2019-10-20 12:21:09 +02:00
"""
if self._libvirt_domain is not None:
return self._libvirt_domain
if self.app.vmm.offline_mode:
return None
2015-01-16 15:33:03 +01:00
# XXX _update_libvirt_domain?
try:
self._libvirt_domain = self.app.vmm.libvirt_conn.lookupByUUID(
self.uuid.bytes)
2016-05-21 03:58:08 +02:00
except libvirt.libvirtError as e:
if e.get_error_code() == libvirt.VIR_ERR_NO_DOMAIN:
self._update_libvirt_domain()
else:
raise
return self._libvirt_domain
@property
def block_devices(self):
2019-10-20 12:21:09 +02:00
""" Return all :py:class:`qubes.storage.BlockDevice` for current domain
for serialization in the libvirt XML template as <disk>.
2019-10-20 12:21:09 +02:00
"""
for v in self.volumes.values():
block_dev = v.block_device()
if block_dev is not None:
yield block_dev
@property
def untrusted_qdb(self):
2019-10-20 12:21:09 +02:00
"""QubesDB handle for this domain."""
if self._qdb_connection is None:
if self.is_running():
2016-06-02 22:02:06 +02:00
import qubesdb # pylint: disable=import-error
self._qdb_connection = qubesdb.QubesDB(self.name)
return self._qdb_connection
2016-06-15 19:47:36 +02:00
@property
def dir_path(self):
2019-10-20 12:21:09 +02:00
"""Root directory for files related to this domain"""
2016-06-15 19:47:36 +02:00
return os.path.join(
qubes.config.qubes_base_dir,
2016-06-15 19:47:36 +02:00
self.dir_path_prefix,
self.name)
@property
def icon_path(self):
return os.path.join(self.dir_path, 'icon.png')
@property
def conf_file(self):
return os.path.join(self.dir_path, 'libvirt.xml')
# network-related
#
# constructor
#
2016-06-02 22:02:06 +02:00
def __init__(self, app, xml, volume_config=None, **kwargs):
# migrate renamed properties
if xml is not None:
node_hvm = xml.find('./properties/property[@name=\'hvm\']')
if node_hvm is not None:
if qubes.property.bool(None, None, node_hvm.text):
kwargs['virt_mode'] = 'hvm'
else:
kwargs['virt_mode'] = 'pv'
node_hvm.getparent().remove(node_hvm)
2015-01-15 12:57:44 +01:00
super(QubesVM, self).__init__(app, xml, **kwargs)
2016-06-02 22:02:06 +02:00
if volume_config is None:
volume_config = {}
2016-06-16 12:44:33 +02:00
if hasattr(self, 'volume_config'):
2016-04-01 19:12:44 +02:00
if xml is not None:
for node in xml.xpath('volume-config/volume'):
name = node.get('name')
assert name
2016-06-02 22:02:06 +02:00
for key, value in node.items():
2016-07-12 17:56:49 +02:00
# pylint: disable=no-member
if value == 'True':
value = True
try:
2016-07-12 17:56:49 +02:00
self.volume_config[name][key] = value
except KeyError:
self.volume_config[name] = {key: value}
2016-04-01 19:12:44 +02:00
for name, conf in volume_config.items():
2016-06-02 22:02:06 +02:00
for key, value in conf.items():
2016-07-12 17:56:49 +02:00
# pylint: disable=no-member
try:
self.volume_config[name][key] = value
except KeyError:
self.volume_config[name] = {key: value}
2016-06-16 12:44:33 +02:00
elif volume_config:
raise TypeError(
2016-05-21 03:58:08 +02:00
'volume_config specified, but {} did not expect that.'.format(
2019-10-20 12:21:09 +02:00
self.__class__.__name__))
# Init private attrs
self._libvirt_domain = None
self._qdb_connection = None
# We assume a fully halted VM here. The 'domain-init' handler will
# check if the VM is already running.
self._domain_stopped_event_received = True
self._domain_stopped_event_handled = True
self._domain_stopped_future = None
# Internal lock to ensure ordering between _domain_stopped_coro() and
# start(). This should not be accessed anywhere else.
self._domain_stopped_lock = asyncio.Lock()
if xml is None:
# we are creating new VM and attributes came through kwargs
assert hasattr(self, 'qid')
assert hasattr(self, 'name')
if xml is None:
# new qube, disable updates check if requested for new qubes
# SEE: 1637 when features are done, migrate to plugin
if not self.app.check_updates_vm:
2017-07-17 12:29:25 +02:00
self.features['check-updates'] = False
# will be initialized after loading all the properties
#: operations which shouldn't happen simultaneously with qube startup
# (including another startup of the same qube)
self.startup_lock = asyncio.Lock()
# fire hooks
if xml is None:
self.events_enabled = True
self.fire_event('domain-init')
2017-08-31 20:28:08 +02:00
def close(self):
if self._qdb_connection is not None:
self._qdb_connection.close()
self._qdb_connection = None
if self._libvirt_domain is not None:
self._libvirt_domain = None
2017-08-31 20:28:08 +02:00
super().close()
def __hash__(self):
return self.qid
def __lt__(self, other):
return self.name < other.name
2016-04-01 19:12:44 +02:00
def __xml__(self):
# pylint: disable=no-member
2016-04-01 19:12:44 +02:00
element = super(QubesVM, self).__xml__()
# pylint: enable=no-member
2016-06-02 22:02:06 +02:00
2016-04-01 19:12:44 +02:00
if hasattr(self, 'volumes'):
volume_config_node = lxml.etree.Element('volume-config')
for volume in self.volumes.values():
volume_config_node.append(volume.__xml__())
2016-04-01 19:12:44 +02:00
element.append(volume_config_node)
return element
#
# event handlers
#
2016-03-14 22:16:52 +01:00
@qubes.events.handler('domain-init', 'domain-load')
def on_domain_init_loaded(self, event):
# pylint: disable=unused-argument
if not hasattr(self, 'uuid'):
# pylint: disable=attribute-defined-outside-init
self.uuid = uuid.uuid4()
# Initialize VM image storage class;
# it might be already initialized by a recursive call from a child VM
if self.storage is None:
self.storage = qubes.storage.Storage(self)
if not self.app.vmm.offline_mode and self.is_running():
self.start_qdb_watch()
self._domain_stopped_event_received = False
self._domain_stopped_event_handled = False
@qubes.events.handler('property-set:label')
def on_property_set_label(self, event, name, newvalue, oldvalue=None):
# pylint: disable=unused-argument
if self.icon_path:
try:
os.remove(self.icon_path)
except OSError:
pass
if hasattr(os, "symlink"):
os.symlink(newvalue.icon_path, self.icon_path)
subprocess.call(['sudo', 'xdg-icon-resource', 'forceupdate'])
else:
shutil.copy(newvalue.icon_path, self.icon_path)
@qubes.events.handler('property-pre-set:kernel')
def on_property_pre_set_kernel(self, event, name, newvalue, oldvalue=None):
# pylint: disable=unused-argument
if not newvalue:
return
dirname = os.path.join(
qubes.config.qubes_base_dir,
qubes.config.system_path['qubes_kernels_base_dir'],
newvalue)
if not os.path.exists(dirname):
2019-10-20 12:21:09 +02:00
raise qubes.exc.QubesPropertyValueError(
self, self.property_get_def(name), newvalue,
'Kernel {!r} not installed'.format(
newvalue))
for filename in ('vmlinuz', 'initramfs'):
if not os.path.exists(os.path.join(dirname, filename)):
2019-10-20 12:21:09 +02:00
raise qubes.exc.QubesPropertyValueError(
self, self.property_get_def(name), newvalue,
'Kernel {!r} not properly installed: '
2019-10-20 12:21:09 +02:00
'missing {!r} file'.format(
newvalue, filename))
@qubes.events.handler('property-pre-set:autostart')
def on_property_pre_set_autostart(self, event, name, newvalue,
2019-10-20 12:21:09 +02:00
oldvalue=None):
# pylint: disable=unused-argument
# workaround https://bugzilla.redhat.com/show_bug.cgi?id=1181922
if newvalue:
retcode = subprocess.call(
["sudo", "ln", "-sf",
"/usr/lib/systemd/system/qubes-vm@.service",
2016-09-12 06:09:45 +02:00
"/etc/systemd/system/multi-user.target.wants/qubes-vm@"
"{}.service".format(self.name)])
else:
retcode = subprocess.call(
['sudo', 'systemctl', 'disable',
2019-10-20 12:21:09 +02:00
'qubes-vm@{}.service'.format(self.name)])
if retcode:
raise qubes.exc.QubesException(
'Failed to set autostart for VM in systemd')
@qubes.events.handler('property-pre-del:autostart')
def on_property_pre_del_autostart(self, event, name, oldvalue=None):
# pylint: disable=unused-argument
if oldvalue:
retcode = subprocess.call(
['sudo', 'systemctl', 'disable',
2019-10-20 12:21:09 +02:00
'qubes-vm@{}.service'.format(self.name)])
if retcode:
raise qubes.exc.QubesException(
'Failed to reset autostart for VM in systemd')
@qubes.events.handler('domain-remove-from-disk')
def on_remove_from_disk(self, event, **kwargs):
# pylint: disable=unused-argument
if self.autostart:
subprocess.call(
['sudo', 'systemctl', 'disable',
2019-10-20 12:21:09 +02:00
'qubes-vm@{}.service'.format(self.name)])
@qubes.events.handler('domain-create-on-disk')
def on_create_on_disk(self, event, **kwargs):
# pylint: disable=unused-argument
if self.autostart:
subprocess.call(
['sudo', 'systemctl', 'enable',
2019-10-20 12:21:09 +02:00
'qubes-vm@{}.service'.format(self.name)])
#
# methods for changing domain state
#
vm: call after-shutdown cleanup also from vm.kill and vm.shutdown Cleaning up after domain shutdown (domain-stopped and domain-shutdown events) relies on libvirt events which may be unreliable in some cases (events may be processed with some delay, of if libvirt was restarted in the meantime, may not happen at all). So, instead of ensuring only proper ordering between shutdown cleanup and next startup, also trigger the cleanup when we know for sure domain isn't running: - at vm.kill() - after libvirt confirms domain was destroyed - at vm.shutdown(wait=True) - after successful shutdown - at vm.remove_from_disk() - after ensuring it isn't running but just before actually removing it This fixes various race conditions: - qvm-kill && qvm-remove: remove could happen before shutdown cleanup was done and storage driver would be confused about that - qvm-shutdown --wait && qvm-clone: clone could happen before new content was commited to the original volume, making the copy of previous VM state (and probably more) Previously it wasn't such a big issue on default configuration, because LVM driver was fully synchronous, effectively blocking the whole qubesd for the time the cleanup happened. To avoid code duplication, factor out _ensure_shutdown_handled function calling actual cleanup (and possibly canceling one called with libvirt event). Note that now, "Duplicated stopped event from libvirt received!" warning may happen in normal circumstances, not only because of some bug. It is very important that post-shutdown cleanup happen when domain is not running. To ensure that, take startup_lock and under it 1) ensure its halted and only then 2) execute the cleanup. This isn't necessary when removing it from disk, because its already removed from the collection at that time, which also avoids other calls to it (see also "vm/dispvm: fix DispVM cleanup" commit). Actually, taking the startup_lock in remove_from_disk function would cause a deadlock in DispVM auto cleanup code: - vm.kill (or other trigger for the cleanup) - vm.startup_lock acquire <==== - vm._ensure_shutdown_handled - domain-shutdown event - vm._auto_cleanup (in DispVM class) - vm.remove_from_disk - cannot take vm.startup_lock again
2018-10-21 04:52:27 +02:00
@asyncio.coroutine
def _ensure_shutdown_handled(self):
2019-10-20 12:21:09 +02:00
"""Make sure previous shutdown is fully handled.
vm: call after-shutdown cleanup also from vm.kill and vm.shutdown Cleaning up after domain shutdown (domain-stopped and domain-shutdown events) relies on libvirt events which may be unreliable in some cases (events may be processed with some delay, of if libvirt was restarted in the meantime, may not happen at all). So, instead of ensuring only proper ordering between shutdown cleanup and next startup, also trigger the cleanup when we know for sure domain isn't running: - at vm.kill() - after libvirt confirms domain was destroyed - at vm.shutdown(wait=True) - after successful shutdown - at vm.remove_from_disk() - after ensuring it isn't running but just before actually removing it This fixes various race conditions: - qvm-kill && qvm-remove: remove could happen before shutdown cleanup was done and storage driver would be confused about that - qvm-shutdown --wait && qvm-clone: clone could happen before new content was commited to the original volume, making the copy of previous VM state (and probably more) Previously it wasn't such a big issue on default configuration, because LVM driver was fully synchronous, effectively blocking the whole qubesd for the time the cleanup happened. To avoid code duplication, factor out _ensure_shutdown_handled function calling actual cleanup (and possibly canceling one called with libvirt event). Note that now, "Duplicated stopped event from libvirt received!" warning may happen in normal circumstances, not only because of some bug. It is very important that post-shutdown cleanup happen when domain is not running. To ensure that, take startup_lock and under it 1) ensure its halted and only then 2) execute the cleanup. This isn't necessary when removing it from disk, because its already removed from the collection at that time, which also avoids other calls to it (see also "vm/dispvm: fix DispVM cleanup" commit). Actually, taking the startup_lock in remove_from_disk function would cause a deadlock in DispVM auto cleanup code: - vm.kill (or other trigger for the cleanup) - vm.startup_lock acquire <==== - vm._ensure_shutdown_handled - domain-shutdown event - vm._auto_cleanup (in DispVM class) - vm.remove_from_disk - cannot take vm.startup_lock again
2018-10-21 04:52:27 +02:00
MUST NOT be called when domain is running.
2019-10-20 12:21:09 +02:00
"""
vm: call after-shutdown cleanup also from vm.kill and vm.shutdown Cleaning up after domain shutdown (domain-stopped and domain-shutdown events) relies on libvirt events which may be unreliable in some cases (events may be processed with some delay, of if libvirt was restarted in the meantime, may not happen at all). So, instead of ensuring only proper ordering between shutdown cleanup and next startup, also trigger the cleanup when we know for sure domain isn't running: - at vm.kill() - after libvirt confirms domain was destroyed - at vm.shutdown(wait=True) - after successful shutdown - at vm.remove_from_disk() - after ensuring it isn't running but just before actually removing it This fixes various race conditions: - qvm-kill && qvm-remove: remove could happen before shutdown cleanup was done and storage driver would be confused about that - qvm-shutdown --wait && qvm-clone: clone could happen before new content was commited to the original volume, making the copy of previous VM state (and probably more) Previously it wasn't such a big issue on default configuration, because LVM driver was fully synchronous, effectively blocking the whole qubesd for the time the cleanup happened. To avoid code duplication, factor out _ensure_shutdown_handled function calling actual cleanup (and possibly canceling one called with libvirt event). Note that now, "Duplicated stopped event from libvirt received!" warning may happen in normal circumstances, not only because of some bug. It is very important that post-shutdown cleanup happen when domain is not running. To ensure that, take startup_lock and under it 1) ensure its halted and only then 2) execute the cleanup. This isn't necessary when removing it from disk, because its already removed from the collection at that time, which also avoids other calls to it (see also "vm/dispvm: fix DispVM cleanup" commit). Actually, taking the startup_lock in remove_from_disk function would cause a deadlock in DispVM auto cleanup code: - vm.kill (or other trigger for the cleanup) - vm.startup_lock acquire <==== - vm._ensure_shutdown_handled - domain-shutdown event - vm._auto_cleanup (in DispVM class) - vm.remove_from_disk - cannot take vm.startup_lock again
2018-10-21 04:52:27 +02:00
with (yield from self._domain_stopped_lock):
# Don't accept any new stopped event's till a new VM has been
# created. If we didn't received any stopped event or it wasn't
# handled yet we will handle this in the next lines.
self._domain_stopped_event_received = True
if self._domain_stopped_future is not None:
# Libvirt stopped event was already received, so cancel the
# future. If it didn't generate the Qubes events yet we
# will do it below.
self._domain_stopped_future.cancel()
self._domain_stopped_future = None
if not self._domain_stopped_event_handled:
# No Qubes domain-stopped events have been generated yet.
# So do this now.
# Set this immediately such that we don't generate events
# twice if an exception gets thrown.
self._domain_stopped_event_handled = True
yield from self.fire_event_async('domain-stopped')
yield from self.fire_event_async('domain-shutdown')
@asyncio.coroutine
def start(self, start_guid=True, notify_function=None,
2019-10-20 12:21:09 +02:00
mem_required=None):
"""Start domain
:param bool start_guid: FIXME
:param collections.Callable notify_function: FIXME
:param int mem_required: FIXME
2019-10-20 12:21:09 +02:00
"""
with (yield from self.startup_lock):
# check if domain wasn't removed in the meantime
if self not in self.app.domains:
raise qubes.exc.QubesVMNotFoundError(self.name)
# Intentionally not used is_running(): eliminate also "Paused",
# "Crashed", "Halting"
if self.get_power_state() != 'Halted':
return self
vm: call after-shutdown cleanup also from vm.kill and vm.shutdown Cleaning up after domain shutdown (domain-stopped and domain-shutdown events) relies on libvirt events which may be unreliable in some cases (events may be processed with some delay, of if libvirt was restarted in the meantime, may not happen at all). So, instead of ensuring only proper ordering between shutdown cleanup and next startup, also trigger the cleanup when we know for sure domain isn't running: - at vm.kill() - after libvirt confirms domain was destroyed - at vm.shutdown(wait=True) - after successful shutdown - at vm.remove_from_disk() - after ensuring it isn't running but just before actually removing it This fixes various race conditions: - qvm-kill && qvm-remove: remove could happen before shutdown cleanup was done and storage driver would be confused about that - qvm-shutdown --wait && qvm-clone: clone could happen before new content was commited to the original volume, making the copy of previous VM state (and probably more) Previously it wasn't such a big issue on default configuration, because LVM driver was fully synchronous, effectively blocking the whole qubesd for the time the cleanup happened. To avoid code duplication, factor out _ensure_shutdown_handled function calling actual cleanup (and possibly canceling one called with libvirt event). Note that now, "Duplicated stopped event from libvirt received!" warning may happen in normal circumstances, not only because of some bug. It is very important that post-shutdown cleanup happen when domain is not running. To ensure that, take startup_lock and under it 1) ensure its halted and only then 2) execute the cleanup. This isn't necessary when removing it from disk, because its already removed from the collection at that time, which also avoids other calls to it (see also "vm/dispvm: fix DispVM cleanup" commit). Actually, taking the startup_lock in remove_from_disk function would cause a deadlock in DispVM auto cleanup code: - vm.kill (or other trigger for the cleanup) - vm.startup_lock acquire <==== - vm._ensure_shutdown_handled - domain-shutdown event - vm._auto_cleanup (in DispVM class) - vm.remove_from_disk - cannot take vm.startup_lock again
2018-10-21 04:52:27 +02:00
yield from self._ensure_shutdown_handled()
self.log.info('Starting {}'.format(self.name))
try:
yield from self.fire_event_async('domain-pre-start',
2019-10-20 12:21:09 +02:00
pre_event=True,
start_guid=start_guid,
mem_required=mem_required)
except Exception as exc:
2019-10-10 19:41:02 +02:00
self.log.error('Start failed: %s', str(exc))
yield from self.fire_event_async('domain-start-failed',
2019-10-20 12:21:09 +02:00
reason=str(exc))
raise
qmemman_client = None
try:
for devclass in self.devices:
for dev in self.devices[devclass].persistent():
if isinstance(dev, qubes.devices.UnknownDevice):
raise qubes.exc.QubesException(
'{} device {} not available'.format(
devclass, dev))
if self.virt_mode == 'pvh' and not self.kernel:
raise qubes.exc.QubesException(
'virt_mode PVH require kernel to be set')
yield from self.storage.verify()
if self.netvm is not None:
# pylint: disable = no-member
if self.netvm.qid != 0:
if not self.netvm.is_running():
2019-10-20 12:21:09 +02:00
yield from self.netvm.start(
start_guid=start_guid,
notify_function=notify_function)
2019-10-20 12:21:09 +02:00
qmemman_client = yield from asyncio.get_event_loop(). \
run_in_executor(None, self.request_memory, mem_required)
yield from self.storage.start()
except Exception as exc:
2019-10-10 19:41:02 +02:00
self.log.error('Start failed: %s', str(exc))
# let anyone receiving domain-pre-start know that startup failed
yield from self.fire_event_async('domain-start-failed',
2019-10-20 12:21:09 +02:00
reason=str(exc))
if qmemman_client:
qmemman_client.close()
raise
try:
self._update_libvirt_domain()
self.libvirt_domain.createWithFlags(
libvirt.VIR_DOMAIN_START_PAUSED)
except libvirt.libvirtError as exc:
# missing IOMMU?
if self.virt_mode == 'hvm' and \
list(self.devices['pci'].persistent()) and \
not self.app.host.is_iommu_supported():
exc = qubes.exc.QubesException(
'Failed to start an HVM qube with PCI devices assigned '
'- hardware does not support IOMMU/VT-d/AMD-Vi')
self.log.error('Start failed: %s', str(exc))
yield from self.fire_event_async('domain-start-failed',
reason=str(exc))
yield from self.storage.stop()
raise exc
except Exception as exc:
2017-11-06 01:19:19 +01:00
self.log.error('Start failed: %s', str(exc))
# let anyone receiving domain-pre-start know that startup failed
yield from self.fire_event_async('domain-start-failed',
2019-10-20 12:21:09 +02:00
reason=str(exc))
yield from self.storage.stop()
raise
finally:
if qmemman_client:
qmemman_client.close()
self._domain_stopped_event_received = False
self._domain_stopped_event_handled = False
try:
yield from self.fire_event_async('domain-spawn',
2019-10-20 12:21:09 +02:00
start_guid=start_guid)
self.log.info('Setting Qubes DB info for the VM')
yield from self.start_qubesdb()
self.create_qdb_entries()
self.start_qdb_watch()
self.log.warning('Activating the {} VM'.format(self.name))
self.libvirt_domain.resume()
yield from self.start_qrexec_daemon()
yield from self.fire_event_async('domain-start',
2019-10-20 12:21:09 +02:00
start_guid=start_guid)
except Exception as exc: # pylint: disable=bare-except
2017-11-06 01:19:19 +01:00
self.log.error('Start failed: %s', str(exc))
# This avoids losing the exception if an exception is
2020-02-15 18:48:32 +01:00
# raised in self._kill_locked(), because the vm is not
# running or paused
try:
yield from self._kill_locked()
except qubes.exc.QubesVMNotStartedError:
pass
# let anyone receiving domain-pre-start know that startup failed
yield from self.fire_event_async('domain-start-failed',
2019-10-20 12:21:09 +02:00
reason=str(exc))
raise
return self
def on_libvirt_domain_stopped(self):
2019-10-20 12:21:09 +02:00
""" Handle VIR_DOMAIN_EVENT_STOPPED events from libvirt.
This is not a Qubes event handler. Instead we do some sanity checks
and synchronization with start() and then emits Qubes events.
2019-10-20 12:21:09 +02:00
"""
state = self.get_power_state()
if state not in ['Halted', 'Crashed', 'Dying']:
self.log.warning('Stopped event from libvirt received,'
2019-10-20 12:21:09 +02:00
' but domain is in state {}!'.format(state))
# ignore this unexpected event
return
if self._domain_stopped_event_received:
vm: call after-shutdown cleanup also from vm.kill and vm.shutdown Cleaning up after domain shutdown (domain-stopped and domain-shutdown events) relies on libvirt events which may be unreliable in some cases (events may be processed with some delay, of if libvirt was restarted in the meantime, may not happen at all). So, instead of ensuring only proper ordering between shutdown cleanup and next startup, also trigger the cleanup when we know for sure domain isn't running: - at vm.kill() - after libvirt confirms domain was destroyed - at vm.shutdown(wait=True) - after successful shutdown - at vm.remove_from_disk() - after ensuring it isn't running but just before actually removing it This fixes various race conditions: - qvm-kill && qvm-remove: remove could happen before shutdown cleanup was done and storage driver would be confused about that - qvm-shutdown --wait && qvm-clone: clone could happen before new content was commited to the original volume, making the copy of previous VM state (and probably more) Previously it wasn't such a big issue on default configuration, because LVM driver was fully synchronous, effectively blocking the whole qubesd for the time the cleanup happened. To avoid code duplication, factor out _ensure_shutdown_handled function calling actual cleanup (and possibly canceling one called with libvirt event). Note that now, "Duplicated stopped event from libvirt received!" warning may happen in normal circumstances, not only because of some bug. It is very important that post-shutdown cleanup happen when domain is not running. To ensure that, take startup_lock and under it 1) ensure its halted and only then 2) execute the cleanup. This isn't necessary when removing it from disk, because its already removed from the collection at that time, which also avoids other calls to it (see also "vm/dispvm: fix DispVM cleanup" commit). Actually, taking the startup_lock in remove_from_disk function would cause a deadlock in DispVM auto cleanup code: - vm.kill (or other trigger for the cleanup) - vm.startup_lock acquire <==== - vm._ensure_shutdown_handled - domain-shutdown event - vm._auto_cleanup (in DispVM class) - vm.remove_from_disk - cannot take vm.startup_lock again
2018-10-21 04:52:27 +02:00
# ignore this event - already triggered by shutdown(), kill(),
# or subsequent start()
return
self._domain_stopped_event_received = True
self._domain_stopped_future = \
asyncio.ensure_future(self._domain_stopped_coro())
@asyncio.coroutine
def _domain_stopped_coro(self):
with (yield from self._domain_stopped_lock):
assert not self._domain_stopped_event_handled
# Set this immediately such that we don't generate events twice if
# an exception gets thrown.
self._domain_stopped_event_handled = True
while self.get_power_state() == 'Dying':
yield from asyncio.sleep(0.25)
yield from self.fire_event_async('domain-stopped')
yield from self.fire_event_async('domain-shutdown')
@qubes.events.handler('domain-stopped')
@asyncio.coroutine
def on_domain_stopped(self, _event, **_kwargs):
2019-10-20 12:21:09 +02:00
"""Cleanup after domain was stopped"""
try:
yield from self.storage.stop()
except qubes.storage.StoragePoolException:
self.log.exception('Failed to stop storage for domain %s',
2019-10-20 12:21:09 +02:00
self.name)
@asyncio.coroutine
def shutdown(self, force=False, wait=False, timeout=None):
2019-10-20 12:21:09 +02:00
"""Shutdown domain.
:param force: ignored
:param wait: wait for shutdown to complete
:param timeout: shutdown wait timeout (for *wait*=True), defaults to
:py:attr:`shutdown_timeout`
:raises qubes.exc.QubesVMNotStartedError: \
when domain is already shut down.
2019-10-20 12:21:09 +02:00
"""
2016-06-16 21:06:20 +02:00
if self.is_halted():
raise qubes.exc.QubesVMNotStartedError(self)
try:
yield from self.fire_event_async('domain-pre-shutdown',
2019-10-20 12:21:09 +02:00
pre_event=True, force=force)
self.libvirt_domain.shutdown()
if wait:
if timeout is None:
timeout = self.shutdown_timeout
while timeout > 0 and not self.is_halted():
yield from asyncio.sleep(0.25)
timeout -= 0.25
with (yield from self.startup_lock):
if self.is_halted():
# make sure all shutdown tasks are completed
yield from self._ensure_shutdown_handled()
else:
raise qubes.exc.QubesVMShutdownTimeoutError(self)
except Exception as ex:
yield from self.fire_event_async('domain-shutdown-failed',
2019-10-20 12:21:09 +02:00
reason=str(ex))
raise
2016-08-01 15:09:28 +02:00
return self
@asyncio.coroutine
def kill(self):
2019-10-20 12:21:09 +02:00
"""Forcefully shutdown (destroy) domain.
:raises qubes.exc.QubesVMNotStartedError: \
when domain is already shut down.
2019-10-20 12:21:09 +02:00
"""
if not self.is_running() and not self.is_paused():
raise qubes.exc.QubesVMNotStartedError(self)
vm: call after-shutdown cleanup also from vm.kill and vm.shutdown Cleaning up after domain shutdown (domain-stopped and domain-shutdown events) relies on libvirt events which may be unreliable in some cases (events may be processed with some delay, of if libvirt was restarted in the meantime, may not happen at all). So, instead of ensuring only proper ordering between shutdown cleanup and next startup, also trigger the cleanup when we know for sure domain isn't running: - at vm.kill() - after libvirt confirms domain was destroyed - at vm.shutdown(wait=True) - after successful shutdown - at vm.remove_from_disk() - after ensuring it isn't running but just before actually removing it This fixes various race conditions: - qvm-kill && qvm-remove: remove could happen before shutdown cleanup was done and storage driver would be confused about that - qvm-shutdown --wait && qvm-clone: clone could happen before new content was commited to the original volume, making the copy of previous VM state (and probably more) Previously it wasn't such a big issue on default configuration, because LVM driver was fully synchronous, effectively blocking the whole qubesd for the time the cleanup happened. To avoid code duplication, factor out _ensure_shutdown_handled function calling actual cleanup (and possibly canceling one called with libvirt event). Note that now, "Duplicated stopped event from libvirt received!" warning may happen in normal circumstances, not only because of some bug. It is very important that post-shutdown cleanup happen when domain is not running. To ensure that, take startup_lock and under it 1) ensure its halted and only then 2) execute the cleanup. This isn't necessary when removing it from disk, because its already removed from the collection at that time, which also avoids other calls to it (see also "vm/dispvm: fix DispVM cleanup" commit). Actually, taking the startup_lock in remove_from_disk function would cause a deadlock in DispVM auto cleanup code: - vm.kill (or other trigger for the cleanup) - vm.startup_lock acquire <==== - vm._ensure_shutdown_handled - domain-shutdown event - vm._auto_cleanup (in DispVM class) - vm.remove_from_disk - cannot take vm.startup_lock again
2018-10-21 04:52:27 +02:00
with (yield from self.startup_lock):
yield from self._kill_locked()
return self
@asyncio.coroutine
def _kill_locked(self):
2019-10-20 12:21:09 +02:00
"""Forcefully shutdown (destroy) domain.
2019-10-20 12:21:09 +02:00
This function needs to be called with self.startup_lock held."""
try:
self.libvirt_domain.destroy()
except libvirt.libvirtError as e:
if e.get_error_code() == libvirt.VIR_ERR_OPERATION_INVALID:
raise qubes.exc.QubesVMNotStartedError(self)
raise
# make sure all shutdown tasks are completed
yield from self._ensure_shutdown_handled()
@asyncio.coroutine
def suspend(self):
2019-10-20 12:21:09 +02:00
"""Suspend (pause) domain.
:raises qubes.exc.QubesVMNotRunnignError: \
when domain is already shut down.
2019-10-20 12:21:09 +02:00
"""
if not self.is_running() and not self.is_paused():
raise qubes.exc.QubesVMNotRunningError(self)
if list(self.devices['pci'].attached()):
if self.features.check_with_template('qrexec', False):
yield from self.run_service_for_stdio('qubes.SuspendPre',
2019-10-20 12:21:09 +02:00
user='root')
self.libvirt_domain.pMSuspendForDuration(
libvirt.VIR_NODE_SUSPEND_TARGET_MEM, 0, 0)
else:
self.libvirt_domain.suspend()
return self
@asyncio.coroutine
def pause(self):
2019-10-20 12:21:09 +02:00
"""Pause (suspend) domain."""
if not self.is_running():
raise qubes.exc.QubesVMNotRunningError(self)
self.libvirt_domain.suspend()
return self
@asyncio.coroutine
def resume(self):
2019-10-20 12:21:09 +02:00
"""Resume suspended domain.
:raises qubes.exc.QubesVMNotSuspendedError: when machine is not paused
:raises qubes.exc.QubesVMError: when machine is suspended
2019-10-20 12:21:09 +02:00
"""
# pylint: disable=not-an-iterable
if self.get_power_state() == "Suspended":
self.libvirt_domain.pMWakeup()
if self.features.check_with_template('qrexec', False):
yield from self.run_service_for_stdio('qubes.SuspendPost',
2019-10-20 12:21:09 +02:00
user='root')
else:
yield from self.unpause()
return self
@asyncio.coroutine
def unpause(self):
2019-10-20 12:21:09 +02:00
"""Resume (unpause) a domain"""
if not self.is_paused():
raise qubes.exc.QubesVMNotPausedError(self)
self.libvirt_domain.resume()
return self
@asyncio.coroutine
def run_service(self, service, source=None, user=None,
2019-10-20 12:21:09 +02:00
filter_esc=False, autostart=False, gui=False, **kwargs):
"""Run service on this VM
:param str service: service name
2017-04-15 20:36:59 +02:00
:param qubes.vm.qubesvm.QubesVM source: source domain as presented to
this VM
:param str user: username to run service as
:param bool filter_esc: filter escape sequences to protect terminal \
emulator
:param bool autostart: if :py:obj:`True`, machine will be started if \
it is not running
:param bool gui: when autostarting, also start gui daemon
:rtype: asyncio.subprocess.Process
.. note::
User ``root`` is redefined to ``SYSTEM`` in the Windows agent code
2019-10-20 12:21:09 +02:00
"""
# UNSUPPORTED from previous incarnation:
# localcmd, wait, passio*, notify_function, `-e` switch
#
# - passio* and friends depend on params to command (like in stdlib)
# - the filter_esc is orthogonal to passio*
# - input: see run_service_for_stdio
# - wait has no purpose since this is asynchronous
# - notify_function is gone
source = 'dom0' if source is None else self.app.domains[source].name
if user is None:
user = self.default_user
if self.is_paused():
# XXX what about autostart?
raise qubes.exc.QubesVMNotRunningError(
self, 'Domain {!r} is paused'.format(self.name))
if not self.is_running():
if not autostart:
raise qubes.exc.QubesVMNotRunningError(self)
yield from self.start(start_guid=gui)
if not self.is_qrexec_running():
raise qubes.exc.QubesVMError(
self, 'Domain {!r}: qrexec not connected'.format(self.name))
yield from self.fire_event_async('domain-cmd-pre-run', pre_event=True,
2019-10-20 12:21:09 +02:00
start_guid=gui)
return (yield from asyncio.create_subprocess_exec(
qubes.config.system_path['qrexec_client_path'],
'-d', str(self.name),
*(('-t', '-T') if filter_esc else ()),
'{}:QUBESRPC {} {}'.format(user, service, source),
**kwargs))
@asyncio.coroutine
def run_service_for_stdio(self, *args, input=None, **kwargs):
2019-10-20 12:21:09 +02:00
"""Run a service, pass an optional input and return (stdout, stderr).
Raises an exception if return code != 0.
*args* and *kwargs* are passed verbatim to :py:meth:`run_service`.
.. warning::
There are some combinations if stdio-related *kwargs*, which are
not filtered for problems originating between the keyboard and the
chair.
2019-10-20 12:21:09 +02:00
""" # pylint: disable=redefined-builtin
kwargs.setdefault('stdin', subprocess.PIPE)
kwargs.setdefault('stdout', subprocess.PIPE)
kwargs.setdefault('stderr', subprocess.PIPE)
if kwargs['stdin'] == subprocess.PIPE and input is None:
# workaround for https://bugs.python.org/issue39744
input = b''
p = yield from self.run_service(*args, **kwargs)
# this one is actually a tuple, but there is no need to unpack it
stdouterr = yield from p.communicate(input=input)
if p.returncode:
raise subprocess.CalledProcessError(p.returncode,
2019-10-20 12:21:09 +02:00
args[0], *stdouterr)
return stdouterr
2017-05-30 15:06:05 +02:00
def run(self, command, user=None, **kwargs):
2019-10-20 12:21:09 +02:00
"""Run a shell command inside the domain using qrexec.
This method is a coroutine.
2019-10-20 12:21:09 +02:00
""" # pylint: disable=redefined-builtin
if user is None:
user = self.default_user
return asyncio.create_subprocess_exec(
qubes.config.system_path['qrexec_client_path'],
'-d', str(self.name),
'{}:{}'.format(user, command),
**kwargs)
@asyncio.coroutine
def run_for_stdio(self, *args, input=None, **kwargs):
"""Run a shell command inside the domain using qrexec.
This method is a coroutine.
2019-10-20 12:21:09 +02:00
""" # pylint: disable=redefined-builtin
kwargs.setdefault('stdin', subprocess.PIPE)
kwargs.setdefault('stdout', subprocess.PIPE)
kwargs.setdefault('stderr', subprocess.PIPE)
if kwargs['stdin'] == subprocess.PIPE and input is None:
# workaround for https://bugs.python.org/issue39744
input = b''
p = yield from self.run(*args, **kwargs)
stdouterr = yield from p.communicate(input=input)
if p.returncode:
raise subprocess.CalledProcessError(p.returncode,
2019-10-20 12:21:09 +02:00
args[0], *stdouterr)
return stdouterr
def is_memory_balancing_possible(self):
2019-10-20 12:21:09 +02:00
"""Check if memory balancing can be enabled.
Reasons to not enable it:
- have PCI devices
- balloon driver not present
We don't have reliable way to detect the second point, but good
heuristic is HVM virt_mode (PV and PVH require OS support and it does
include balloon driver) and lack of qrexec/meminfo-writer service
support (no qubes tools installed).
2019-10-20 12:21:09 +02:00
"""
if list(self.devices['pci'].persistent()):
return False
if self.virt_mode == 'hvm':
# if VM announce any supported service
features_set = set(self.features)
template = getattr(self, 'template', None)
while template is not None:
features_set.update(template.features)
template = getattr(template, 'template', None)
supported_services = any(f.startswith('supported-service.')
2019-10-20 12:21:09 +02:00
for f in features_set)
if (not self.features.check_with_template('qrexec', False) or
2019-10-20 12:21:09 +02:00
(supported_services and
not self.features.check_with_template(
'supported-service.meminfo-writer', False))):
return False
return True
def request_memory(self, mem_required=None):
if not qmemman_present:
return None
if mem_required is None:
if self.virt_mode == 'hvm':
if self.stubdom_mem:
stubdom_mem = self.stubdom_mem
else:
if self.features.check_with_template('linux-stubdom', True):
2019-10-20 12:21:09 +02:00
stubdom_mem = 128 # from libxl_create.c
else:
2019-10-20 12:21:09 +02:00
stubdom_mem = 28 # from libxl_create.c
stubdom_mem += 16 # video ram
else:
stubdom_mem = 0
initial_memory = self.memory
mem_required = int(initial_memory + stubdom_mem) * 1024 * 1024
qmemman_client = qubes.qmemman.client.QMemmanClient()
try:
mem_required_with_overhead = mem_required + MEM_OVERHEAD_BASE \
2019-10-20 12:21:09 +02:00
+ self.vcpus * MEM_OVERHEAD_PER_VCPU
got_memory = qmemman_client.request_memory(
mem_required_with_overhead)
except IOError as e:
raise IOError('Failed to connect to qmemman: {!s}'.format(e))
if not got_memory:
qmemman_client.close()
raise qubes.exc.QubesMemoryError(self)
return qmemman_client
@staticmethod
@asyncio.coroutine
def start_daemon(*command, input=None, **kwargs):
2019-10-20 12:21:09 +02:00
"""Start a daemon for the VM
This function take care to run it as appropriate user.
:param command: command to run (array for
:py:meth:`subprocess.check_call`)
:param kwargs: args for :py:meth:`subprocess.check_call`
:return: None
2019-10-20 12:21:09 +02:00
""" # pylint: disable=redefined-builtin
if os.getuid() == 0:
# try to always have VM daemons running as normal user, otherwise
# some files (like clipboard) may be created as root and cause
# permission problems
qubes_group = grp.getgrnam('qubes')
command = ['runuser', '-u', qubes_group.gr_mem[0], '--'] + \
2019-10-20 12:21:09 +02:00
list(command)
p = yield from asyncio.create_subprocess_exec(*command, **kwargs)
stdout, stderr = yield from p.communicate(input=input)
if p.returncode:
raise subprocess.CalledProcessError(p.returncode, command,
2019-10-20 12:21:09 +02:00
output=stdout, stderr=stderr)
@asyncio.coroutine
def start_qrexec_daemon(self):
2019-10-20 12:21:09 +02:00
"""Start qrexec daemon.
:raises OSError: when starting fails.
2019-10-20 12:21:09 +02:00
"""
self.log.debug('Starting the qrexec daemon')
qrexec_args = [str(self.xid), self.name, self.default_user]
if not self.debug:
qrexec_args.insert(0, "-q")
2016-03-16 18:07:49 +01:00
qrexec_env = os.environ.copy()
if not self.features.check_with_template('qrexec', False):
2016-03-16 18:07:49 +01:00
self.log.debug(
'Starting the qrexec daemon in background, because of features')
qrexec_env['QREXEC_STARTUP_NOWAIT'] = '1'
else:
qrexec_env['QREXEC_STARTUP_TIMEOUT'] = str(self.qrexec_timeout)
try:
yield from self.start_daemon(
qubes.config.system_path['qrexec_daemon_path'], *qrexec_args,
env=qrexec_env, stderr=subprocess.PIPE)
except subprocess.CalledProcessError as err:
if err.returncode == 3:
2019-10-20 12:21:09 +02:00
raise qubes.exc.QubesVMError(
self,
'Cannot connect to qrexec agent for {} seconds, '
'see /var/log/xen/console/guest-{}.log for details'.format(
self.qrexec_timeout, self.name
))
2019-10-20 12:21:09 +02:00
raise qubes.exc.QubesVMError(
self, 'qrexec-daemon startup failed: ' + err.stderr.decode())
@asyncio.coroutine
def start_qubesdb(self):
2019-10-20 12:21:09 +02:00
"""Start QubesDB daemon.
:raises OSError: when starting fails.
2019-10-20 12:21:09 +02:00
"""
# drop old connection to QubesDB, if any
self._qdb_connection = None
self.log.info('Starting Qubes DB')
try:
yield from self.start_daemon(
qubes.config.system_path['qubesdb_daemon_path'],
str(self.xid),
self.name)
except subprocess.CalledProcessError:
raise qubes.exc.QubesException('Cannot execute qubesdb-daemon')
@asyncio.coroutine
2016-07-12 17:58:30 +02:00
def create_on_disk(self, pool=None, pools=None):
2019-10-20 12:21:09 +02:00
"""Create files needed for VM.
"""
self.log.info('Creating directory: {0}'.format(self.dir_path))
os.makedirs(self.dir_path, mode=0o775, exist_ok=True)
2016-07-12 17:58:30 +02:00
if pool or pools:
2016-07-12 18:24:43 +02:00
# pylint: disable=attribute-defined-outside-init
2016-07-12 17:58:30 +02:00
self.volume_config = _patch_volume_config(self.volume_config, pool,
pools)
self.storage = qubes.storage.Storage(self)
try:
yield from self.storage.create()
except:
try:
yield from self.storage.remove()
os.rmdir(self.dir_path)
except: # pylint: disable=bare-except
self.log.exception('failed to cleanup {} after failed VM '
'creation'.format(self.dir_path))
raise
if os.path.exists(self.icon_path):
os.unlink(self.icon_path)
self.log.info('Creating icon symlink: {} -> {}'.format(
self.icon_path, self.label.icon_path))
if hasattr(os, "symlink"):
os.symlink(self.label.icon_path, self.icon_path)
else:
shutil.copy(self.label.icon_path, self.icon_path)
# fire hooks
yield from self.fire_event_async('domain-create-on-disk')
@asyncio.coroutine
def remove_from_disk(self):
2019-10-20 12:21:09 +02:00
"""Remove domain remnants from disk."""
if not self.is_halted():
raise qubes.exc.QubesVMNotHaltedError(
2019-06-28 12:29:32 +02:00
"Can't remove VM {!s}, because it's in state {!r}.".format(
self, self.get_power_state()))
vm: call after-shutdown cleanup also from vm.kill and vm.shutdown Cleaning up after domain shutdown (domain-stopped and domain-shutdown events) relies on libvirt events which may be unreliable in some cases (events may be processed with some delay, of if libvirt was restarted in the meantime, may not happen at all). So, instead of ensuring only proper ordering between shutdown cleanup and next startup, also trigger the cleanup when we know for sure domain isn't running: - at vm.kill() - after libvirt confirms domain was destroyed - at vm.shutdown(wait=True) - after successful shutdown - at vm.remove_from_disk() - after ensuring it isn't running but just before actually removing it This fixes various race conditions: - qvm-kill && qvm-remove: remove could happen before shutdown cleanup was done and storage driver would be confused about that - qvm-shutdown --wait && qvm-clone: clone could happen before new content was commited to the original volume, making the copy of previous VM state (and probably more) Previously it wasn't such a big issue on default configuration, because LVM driver was fully synchronous, effectively blocking the whole qubesd for the time the cleanup happened. To avoid code duplication, factor out _ensure_shutdown_handled function calling actual cleanup (and possibly canceling one called with libvirt event). Note that now, "Duplicated stopped event from libvirt received!" warning may happen in normal circumstances, not only because of some bug. It is very important that post-shutdown cleanup happen when domain is not running. To ensure that, take startup_lock and under it 1) ensure its halted and only then 2) execute the cleanup. This isn't necessary when removing it from disk, because its already removed from the collection at that time, which also avoids other calls to it (see also "vm/dispvm: fix DispVM cleanup" commit). Actually, taking the startup_lock in remove_from_disk function would cause a deadlock in DispVM auto cleanup code: - vm.kill (or other trigger for the cleanup) - vm.startup_lock acquire <==== - vm._ensure_shutdown_handled - domain-shutdown event - vm._auto_cleanup (in DispVM class) - vm.remove_from_disk - cannot take vm.startup_lock again
2018-10-21 04:52:27 +02:00
# make sure shutdown is handled before removing anything, but only if
# handling is pending; if not, we may be called from within
# domain-shutdown event (DispVM._auto_cleanup), which would deadlock
if not self._domain_stopped_event_handled:
yield from self._ensure_shutdown_handled()
yield from self.fire_event_async('domain-remove-from-disk')
try:
yield from self.storage.remove()
finally:
try:
# TODO: make it async?
shutil.rmtree(self.dir_path)
except FileNotFoundError:
pass
@asyncio.coroutine
2016-07-12 18:24:43 +02:00
def clone_disk_files(self, src, pool=None, pools=None, ):
2019-10-20 12:21:09 +02:00
"""Clone files from other vm.
:param qubes.vm.qubesvm.QubesVM src: source VM
2019-10-20 12:21:09 +02:00
"""
# If the current vm name is not a part of `self.app.domains.keys()`,
# then the current vm is in creation process. Calling
# `self.is_halted()` at this point, would instantiate libvirt, we want
# avoid that.
if self.name in self.app.domains.keys() and not self.is_halted():
raise qubes.exc.QubesVMNotHaltedError(
self, 'Cannot clone a running domain {!r}'.format(self.name))
msg = "Destination {!s} already exists".format(self.dir_path)
assert not os.path.exists(self.dir_path), msg
self.log.info('Creating directory: {0}'.format(self.dir_path))
os.makedirs(self.dir_path, mode=0o775, exist_ok=True)
2016-07-12 18:24:43 +02:00
if pool or pools:
2016-06-16 13:17:24 +02:00
# pylint: disable=attribute-defined-outside-init
2016-07-12 18:24:43 +02:00
self.volume_config = _patch_volume_config(self.volume_config, pool,
pools)
self.storage = qubes.storage.Storage(self)
yield from self.storage.clone(src)
2016-07-12 17:58:30 +02:00
self.storage.verify()
assert self.volumes != {}
if src.icon_path is not None \
and os.path.exists(src.icon_path) \
and self.icon_path is not None:
if os.path.islink(src.icon_path):
icon_path = os.readlink(src.icon_path)
self.log.info(
'Creating icon symlink {} -> {}'.format(
self.icon_path, icon_path))
os.symlink(icon_path, self.icon_path)
else:
self.log.info(
'Copying icon {} -> {}'.format(
src.icon_path, self.icon_path))
shutil.copy(src.icon_path, self.icon_path)
# fire hooks
yield from self.fire_event_async('domain-clone-files', src=src)
#
# methods for querying domain state
#
# state of the machine
def get_power_state(self):
2019-10-20 12:21:09 +02:00
"""Return power state description string.
Return value may be one of those:
=============== ========================================================
return value meaning
=============== ========================================================
``'Halted'`` Machine is not active.
``'Transient'`` Machine is running, but does not have :program:`guid`
or :program:`qrexec` available.
``'Running'`` Machine is ready and running.
``'Paused'`` Machine is paused.
``'Suspended'`` Machine is S3-suspended.
``'Halting'`` Machine is in process of shutting down.
``'Dying'`` Machine is still in process of shutting down.
``'Crashed'`` Machine crashed and is unusable, probably because of
bug in dom0.
``'NA'`` Machine is in unknown state (most likely libvirt domain
is undefined).
=============== ========================================================
FIXME: graph below may be incomplete and wrong. Click on method name to
see its documentation.
.. graphviz::
digraph {
node [fontname="sans-serif"];
edge [fontname="mono"];
Halted;
NA;
Dying;
Crashed;
Transient;
Halting;
Running;
Paused [color=gray75 fontcolor=gray75];
Suspended;
NA -> Halted;
Halted -> NA [constraint=false];
Halted -> Transient
[xlabel="start()" URL="#qubes.vm.qubesvm.QubesVM.start"];
Transient -> Running;
Running -> Halting
[xlabel="shutdown()"
URL="#qubes.vm.qubesvm.QubesVM.shutdown"
constraint=false];
Halting -> Dying -> Halted [constraint=false];
/* cosmetic, invisible edges to put rank constraint */
Dying -> Halting [style="invis"];
Halting -> Transient [style="invis"];
Running -> Halted
2020-02-15 18:48:32 +01:00
[label="kill()"
URL="#qubes.vm.qubesvm.QubesVM.kill"
constraint=false];
Running -> Crashed [constraint=false];
Crashed -> Halted [constraint=false];
Running -> Paused
[label="pause()" URL="#qubes.vm.qubesvm.QubesVM.pause"
color=gray75 fontcolor=gray75];
Running -> Suspended
[label="suspend()" URL="#qubes.vm.qubesvm.QubesVM.suspend"
color=gray50 fontcolor=gray50];
Paused -> Running
[label="unpause()" URL="#qubes.vm.qubesvm.QubesVM.unpause"
color=gray75 fontcolor=gray75];
Suspended -> Running
[label="resume()" URL="#qubes.vm.qubesvm.QubesVM.resume"
color=gray50 fontcolor=gray50];
Running -> Suspended
[label="suspend()" URL="#qubes.vm.qubesvm.QubesVM.suspend"];
Suspended -> Running
[label="resume()" URL="#qubes.vm.qubesvm.QubesVM.resume"];
{ rank=source; Halted NA };
{ rank=same; Transient Halting };
{ rank=same; Crashed Dying };
{ rank=sink; Paused Suspended };
}
.. seealso::
http://wiki.libvirt.org/page/VM_lifecycle
Description of VM life cycle from the point of view of libvirt.
https://libvirt.org/html/libvirt-libvirt-domain.html#virDomainState
Libvirt's enum describing precise state of a domain.
2019-10-20 12:21:09 +02:00
""" # pylint: disable=too-many-return-statements
# don't try to define libvirt domain, if it isn't there, VM surely
# isn't running
# reason for this "if": allow vm.is_running() in PCI (or other
# device) extension while constructing libvirt XML
if self.app.vmm.offline_mode:
return 'Halted'
if self._libvirt_domain is None:
try:
self._libvirt_domain = self.app.vmm.libvirt_conn.lookupByUUID(
self.uuid.bytes)
except libvirt.libvirtError as e:
if e.get_error_code() == libvirt.VIR_ERR_NO_DOMAIN:
return 'Halted'
raise
libvirt_domain = self.libvirt_domain
if libvirt_domain is None:
return 'Halted'
try:
if libvirt_domain.isActive():
2016-01-29 17:56:33 +01:00
# pylint: disable=line-too-long
if libvirt_domain.state()[0] == libvirt.VIR_DOMAIN_PAUSED:
return "Paused"
if libvirt_domain.state()[0] == libvirt.VIR_DOMAIN_CRASHED:
return "Crashed"
if libvirt_domain.state()[0] == libvirt.VIR_DOMAIN_SHUTDOWN:
return "Halting"
if libvirt_domain.state()[0] == libvirt.VIR_DOMAIN_SHUTOFF:
return "Dying"
2019-10-20 12:21:09 +02:00
if libvirt_domain.state()[
0] == libvirt.VIR_DOMAIN_PMSUSPENDED: # nopep8
return "Suspended"
if not self.is_fully_usable():
return "Transient"
return "Running"
2017-04-15 20:04:38 +02:00
return 'Halted'
except libvirt.libvirtError as e:
if e.get_error_code() == libvirt.VIR_ERR_NO_DOMAIN:
return 'Halted'
raise
assert False
2016-06-16 21:06:20 +02:00
def is_halted(self):
2019-10-20 12:21:09 +02:00
""" Check whether this domain's state is 'Halted'
2016-06-16 21:06:20 +02:00
:returns: :py:obj:`True` if this domain is halted, \
:py:obj:`False` otherwise.
:rtype: bool
2019-10-20 12:21:09 +02:00
"""
2016-06-16 21:06:20 +02:00
return self.get_power_state() == 'Halted'
def is_running(self):
2019-10-20 12:21:09 +02:00
"""Check whether this domain is running.
:returns: :py:obj:`True` if this domain is started, \
:py:obj:`False` otherwise.
:rtype: bool
2019-10-20 12:21:09 +02:00
"""
if self.app.vmm.offline_mode:
return False
# don't try to define libvirt domain, if it isn't there, VM surely
# isn't running
# reason for this "if": allow vm.is_running() in PCI (or other
# device) extension while constructing libvirt XML
if self._libvirt_domain is None:
try:
self._libvirt_domain = self.app.vmm.libvirt_conn.lookupByUUID(
self.uuid.bytes)
except libvirt.libvirtError as e:
if e.get_error_code() == libvirt.VIR_ERR_NO_DOMAIN:
return False
raise
2018-02-21 04:30:47 +01:00
return bool(self.libvirt_domain.isActive())
def is_paused(self):
2019-10-20 12:21:09 +02:00
"""Check whether this domain is paused.
:returns: :py:obj:`True` if this domain is paused, \
:py:obj:`False` otherwise.
:rtype: bool
2019-10-20 12:21:09 +02:00
"""
return self.libvirt_domain \
2019-10-20 12:21:09 +02:00
and self.libvirt_domain.state()[0] == libvirt.VIR_DOMAIN_PAUSED
def is_qrexec_running(self):
2019-10-20 12:21:09 +02:00
"""Check whether qrexec for this domain is available.
:returns: :py:obj:`True` if qrexec is running, \
:py:obj:`False` otherwise.
:rtype: bool
2019-10-20 12:21:09 +02:00
"""
if self.xid < 0: # pylint: disable=comparison-with-callable
return False
return os.path.exists('/var/run/qubes/qrexec.%s' % self.name)
def is_fully_usable(self):
return all(self.fire_event('domain-is-fully-usable'))
@qubes.events.handler('domain-is-fully-usable')
def on_domain_is_fully_usable(self, event):
2019-10-20 12:21:09 +02:00
"""Check whether domain is running and sane.
Currently this checks for running qrexec.
2019-10-20 12:21:09 +02:00
""" # pylint: disable=unused-argument
if self.features.check_with_template('qrexec', False):
# Running gui-daemon implies also VM running
yield self.is_qrexec_running()
else:
yield True
# memory and disk
def get_mem(self):
2019-10-20 12:21:09 +02:00
"""Get current memory usage from VM.
:returns: Memory usage [FIXME unit].
:rtype: FIXME
2019-10-20 12:21:09 +02:00
"""
if self.libvirt_domain is None:
return 0
try:
if not self.libvirt_domain.isActive():
return 0
return self.libvirt_domain.info()[1]
except libvirt.libvirtError as e:
if e.get_error_code() in (
# qube no longer exists
libvirt.VIR_ERR_NO_DOMAIN,
# libxl_domain_info failed (race condition from isActive)
libvirt.VIR_ERR_INTERNAL_ERROR):
return 0
self.log.exception(
'libvirt error code: {!r}'.format(e.get_error_code()))
raise
def get_mem_static_max(self):
2019-10-20 12:21:09 +02:00
"""Get maximum memory available to VM.
:returns: Memory limit [FIXME unit].
:rtype: FIXME
2019-10-20 12:21:09 +02:00
"""
if self.libvirt_domain is None:
return 0
try:
return self.libvirt_domain.maxMemory()
except libvirt.libvirtError as e:
if e.get_error_code() in (
# qube no longer exists
libvirt.VIR_ERR_NO_DOMAIN,
# libxl_domain_info failed (race condition from isActive)
libvirt.VIR_ERR_INTERNAL_ERROR):
return 0
self.log.exception(
'libvirt error code: {!r}'.format(e.get_error_code()))
raise
def get_cputime(self):
2019-10-20 12:21:09 +02:00
"""Get total CPU time burned by this domain since start.
:returns: CPU time usage [FIXME unit].
:rtype: FIXME
2019-10-20 12:21:09 +02:00
"""
if self.libvirt_domain is None:
return 0
if self.libvirt_domain is None:
return 0
if not self.libvirt_domain.isActive():
return 0
try:
if not self.libvirt_domain.isActive():
return 0
2019-10-20 12:21:09 +02:00
# this does not work, because libvirt
2019-10-20 16:40:40 +02:00
# return self.libvirt_domain.getCPUStats(
# libvirt.VIR_NODE_CPU_STATS_ALL_CPUS, 0)[0]['cpu_time']/10**9
return self.libvirt_domain.info()[4]
except libvirt.libvirtError as e:
if e.get_error_code() in (
# qube no longer exists
libvirt.VIR_ERR_NO_DOMAIN,
# libxl_domain_info failed (race condition from isActive)
libvirt.VIR_ERR_INTERNAL_ERROR):
return 0
self.log.exception(
'libvirt error code: {!r}'.format(e.get_error_code()))
raise
# miscellanous
@qubes.stateless_property
def start_time(self):
2019-10-20 12:21:09 +02:00
"""Tell when machine was started.
:rtype: float or None
2019-10-20 12:21:09 +02:00
"""
if not self.is_running():
return None
# TODO shouldn't this be qubesdb?
start_time = self.app.vmm.xs.read('',
2019-10-20 12:21:09 +02:00
'/vm/{}/start_time'.format(self.uuid))
if start_time != '':
return float(start_time)
2017-04-15 20:04:38 +02:00
return None
@property
def kernelopts_common(self):
2019-10-20 12:21:09 +02:00
"""Kernel options which should be used in addition to *kernelopts*
property.
This is specific to kernel (and initrd if any)
2019-10-20 12:21:09 +02:00
"""
if not self.kernel:
return ''
kernels_dir = self.storage.kernels_dir
kernelopts_path = os.path.join(kernels_dir,
2019-10-20 12:21:09 +02:00
'default-kernelopts-common.txt')
if os.path.exists(kernelopts_path):
with open(kernelopts_path) as f_kernelopts:
return f_kernelopts.read().rstrip('\n\r')
else:
return qubes.config.defaults['kernelopts_common']
#
# helper methods
#
def relative_path(self, path):
2019-10-20 12:21:09 +02:00
"""Return path relative to py:attr:`dir_path`.
:param str path: Path in question.
:returns: Relative path.
2019-10-20 12:21:09 +02:00
"""
return os.path.relpath(path, self.dir_path)
def create_qdb_entries(self):
2019-10-20 12:21:09 +02:00
"""Create entries in Qubes DB.
"""
# pylint: disable=no-member
self.untrusted_qdb.write('/name', self.name)
self.untrusted_qdb.write('/type', self.__class__.__name__)
self.untrusted_qdb.write('/default-user', self.default_user)
self.untrusted_qdb.write('/qubes-vm-updateable', str(self.updateable))
self.untrusted_qdb.write('/qubes-vm-persistence',
2019-10-20 12:21:09 +02:00
'full' if self.updateable else 'rw-only')
self.untrusted_qdb.write('/qubes-debug-mode', str(int(self.debug)))
try:
self.untrusted_qdb.write('/qubes-base-template', self.template.name)
except AttributeError:
self.untrusted_qdb.write('/qubes-base-template', '')
self.untrusted_qdb.write('/qubes-random-seed',
2019-10-20 12:21:09 +02:00
base64.b64encode(qubes.utils.urandom(64)))
if self.provides_network:
# '/qubes-netvm-network' value is only checked for being non empty
self.untrusted_qdb.write('/qubes-netvm-network', str(self.gateway))
self.untrusted_qdb.write('/qubes-netvm-gateway', str(self.gateway))
2019-09-25 01:18:09 +02:00
if self.gateway6: # pylint: disable=using-constant-test
self.untrusted_qdb.write('/qubes-netvm-gateway6',
2019-10-20 12:21:09 +02:00
str(self.gateway6))
self.untrusted_qdb.write('/qubes-netvm-netmask', str(self.netmask))
for i, addr in zip(('primary', 'secondary'), self.dns):
self.untrusted_qdb.write('/qubes-netvm-{}-dns'.format(i), addr)
if self.netvm is not None:
2019-05-16 17:23:47 +02:00
self.untrusted_qdb.write('/qubes-mac', str(self.mac))
self.untrusted_qdb.write('/qubes-ip', str(self.visible_ip))
self.untrusted_qdb.write('/qubes-netmask',
2019-10-20 12:21:09 +02:00
str(self.visible_netmask))
self.untrusted_qdb.write('/qubes-gateway',
2019-10-20 12:21:09 +02:00
str(self.visible_gateway))
for i, addr in zip(('primary', 'secondary'), self.dns):
self.untrusted_qdb.write('/qubes-{}-dns'.format(i), str(addr))
2019-09-25 01:18:09 +02:00
if self.visible_ip6: # pylint: disable=using-constant-test
self.untrusted_qdb.write('/qubes-ip6', str(self.visible_ip6))
2019-09-25 01:18:09 +02:00
if self.visible_gateway6: # pylint: disable=using-constant-test
self.untrusted_qdb.write('/qubes-gateway6',
2019-10-20 12:21:09 +02:00
str(self.visible_gateway6))
tzname = qubes.utils.get_timezone()
if tzname:
self.untrusted_qdb.write('/qubes-timezone', tzname)
self.untrusted_qdb.write('/qubes-block-devices', '')
self.untrusted_qdb.write('/qubes-usb-devices', '')
# TODO: Currently the whole qmemman is quite Xen-specific, so stay with
# xenstore for it until decided otherwise
if qmemman_present:
self.app.vmm.xs.set_permissions('',
2019-10-20 12:21:09 +02:00
'/local/domain/{}/memory'.format(
self.xid),
[{'dom': self.xid}])
self.fire_event('domain-qdb-create')
# TODO async; update this in constructor
def _update_libvirt_domain(self):
2019-10-20 12:21:09 +02:00
"""Re-initialise :py:attr:`libvirt_domain`."""
domain_config = self.create_config_file()
try:
self._libvirt_domain = self.app.vmm.libvirt_conn.defineXML(
domain_config)
except libvirt.libvirtError as e:
if e.get_error_code() == libvirt.VIR_ERR_OS_TYPE \
and e.get_str2() == 'hvm':
2019-10-20 12:21:09 +02:00
raise qubes.exc.QubesVMError(
self,
'HVM qubes are not supported on this machine. '
'Check BIOS settings for VT-x/AMD-V extensions.')
raise
#
# workshop -- those are to be reworked later
#
def get_prefmem(self):
# TODO: qmemman is still xen specific
2019-10-20 12:21:09 +02:00
untrusted_meminfo_key = self.app.vmm.xs.read(
'', '/local/domain/{}/memory/meminfo'.format(self.xid))
if untrusted_meminfo_key is None or untrusted_meminfo_key == '':
return 0
domain = qubes.qmemman.DomainState(self.xid)
qubes.qmemman.algo.refresh_meminfo_for_domain(
domain, untrusted_meminfo_key)
if domain.mem_used is None:
# apparently invalid xenstore content
return 0
domain.memory_maximum = self.get_mem_static_max() * 1024
return qubes.qmemman.algo.prefmem(domain) / 1024
2016-07-12 18:24:43 +02:00
def _clean_volume_config(config):
common_attributes = ['name', 'pool', 'size',
'revisions_to_keep', 'rw', 'snap_on_start',
'save_on_stop', 'source']
return {k: v for k, v in config.items() if k in common_attributes}
2016-07-12 18:24:43 +02:00
def _patch_pool_config(config, pool=None, pools=None):
assert pool is not None or pools is not None
is_snapshot = config['snap_on_start']
is_rw = config['rw']
2016-07-12 18:24:43 +02:00
name = config['name']
if pool and not is_snapshot and is_rw:
2016-07-12 18:24:43 +02:00
config['pool'] = str(pool)
elif pool:
2016-07-12 18:24:43 +02:00
pass
elif pools and name in pools.keys():
if not is_snapshot:
2016-07-12 18:24:43 +02:00
config['pool'] = str(pools[name])
else:
msg = "Snapshot volume {0!s} must be in the same pool as its " \
"origin ({0!s} volume of template)," \
"cannot move to pool {1!s} " \
2016-07-12 18:24:43 +02:00
.format(name, pools[name])
raise qubes.exc.QubesException(msg)
return config
2019-10-20 12:21:09 +02:00
2016-07-12 18:24:43 +02:00
def _patch_volume_config(volume_config, pool=None, pools=None):
assert not (pool and pools), \
'You can not pass pool & pools parameter at same time'
assert pool or pools
result = {}
for name, config in volume_config.items():
# copy only the subset of volume_config key/values
dst_config = _clean_volume_config(config)
if pool is not None or pools is not None:
dst_config = _patch_pool_config(dst_config, pool, pools)
result[name] = dst_config
return result