You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
ansible/library/setup

1682 lines
66 KiB
Plaintext

#!/usr/bin/python
# -*- coding: utf-8 -*-
# (c) 2012, Michael DeHaan <michael.dehaan@gmail.com>
#
#n This file is part of Ansible
#
# Ansible is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# Ansible 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 General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with Ansible. If not, see <http://www.gnu.org/licenses/>.
import os
import array
import fcntl
import fnmatch
import glob
import platform
import re
import socket
import struct
import datetime
import getpass
if not os.path.exists('/sys/devices/virtual/dmi/id/product_name'):
try:
import dmidecode
except ImportError:
import subprocess
DOCUMENTATION = '''
---
module: setup
short_description: Gathers facts about remote hosts
options:
filter:
version_added: "1.1"
description:
- if supplied, only return facts that match this shell-style (fnmatch) wildcard.
required: false
default: '*'
description:
- This module is automatically called by playbooks to gather useful
variables about remote hosts that can be used in playbooks. It can also be
executed directly by C(/usr/bin/ansible) to check what variables are
available to a host. Ansible provides many I(facts) about the system,
automatically.
notes:
- More ansible facts will be added with successive releases. If I(facter) or
I(ohai) are installed, variables from these programs will also be snapshotted
into the JSON file for usage in templating. These variables are prefixed
with C(facter_) and C(ohai_) so it's easy to tell their source. All variables are
bubbled up to the caller. Using the ansible facts and choosing to not
install I(facter) and I(ohai) means you can avoid Ruby-dependencies on your
remote systems. (See also M(facter) and M(ohai).)
- The filter option filters only the first level subkey below ansible_facts.
author: Michael DeHaan
'''
EXAMPLES = """
# Display facts from all hosts and store them indexed by I(hostname) at C(/tmp/facts).
ansible all -m setup --tree /tmp/facts
# Display only facts regarding memory found by ansible on all hosts and output them.
ansible all -m setup -a 'filter=ansible_*_mb'
# Display only facts returned by facter.
ansible all -m setup -a 'filter=facter_*'
# Display only facts returned by facter.
ansible all -m setup -a 'filter=ansible_eth[0-2]'
"""
try:
import selinux
HAVE_SELINUX=True
except ImportError:
HAVE_SELINUX=False
try:
import json
except ImportError:
import simplejson as json
class Facts(object):
"""
This class should only attempt to populate those facts that
are mostly generic to all systems. This includes platform facts,
service facts (eg. ssh keys or selinux), and distribution facts.
Anything that requires extensive code or may have more than one
possible implementation to establish facts for a given topic should
subclass Facts.
"""
_I386RE = re.compile(r'i[3456]86')
# For the most part, we assume that platform.dist() will tell the truth.
# This is the fallback to handle unknowns or exceptions
OSDIST_DICT = { '/etc/redhat-release': 'RedHat',
'/etc/vmware-release': 'VMwareESX',
'/etc/openwrt_release': 'OpenWrt',
'/etc/system-release': 'OtherLinux',
'/etc/alpine-release': 'Alpine' }
SELINUX_MODE_DICT = { 1: 'enforcing', 0: 'permissive', -1: 'disabled' }
# A list of dicts. If there is a platform with more than one
# package manager, put the preferred one last. If there is an
# ansible module, use that as the value for the 'name' key.
PKG_MGRS = [ { 'path' : '/usr/bin/yum', 'name' : 'yum' },
{ 'path' : '/usr/bin/apt-get', 'name' : 'apt' },
{ 'path' : '/usr/bin/zypper', 'name' : 'zypper' },
{ 'path' : '/usr/bin/pacman', 'name' : 'pacman' },
{ 'path' : '/bin/opkg', 'name' : 'opkg' },
{ 'path' : '/opt/local/bin/pkgin', 'name' : 'pkgin' },
{ 'path' : '/opt/local/bin/port', 'name' : 'macports' },
{ 'path' : '/sbin/apk', 'name' : 'apk' },
]
def __init__(self):
self.facts = {}
self.get_platform_facts()
self.get_distribution_facts()
self.get_cmdline()
self.get_public_ssh_host_keys()
self.get_selinux_facts()
self.get_pkg_mgr_facts()
self.get_lsb_facts()
self.get_date_time_facts()
self.get_user_facts()
def populate(self):
return self.facts
# Platform
# patform.system() can be Linux, Darwin, Java, or Windows
def get_platform_facts(self):
self.facts['system'] = platform.system()
self.facts['kernel'] = platform.release()
self.facts['machine'] = platform.machine()
self.facts['python_version'] = platform.python_version()
self.facts['fqdn'] = socket.getfqdn()
self.facts['hostname'] = platform.node().split('.')[0]
self.facts['domain'] = '.'.join(self.facts['fqdn'].split('.')[1:])
if self.facts['machine'] == 'x86_64':
self.facts['architecture'] = self.facts['machine']
elif Facts._I386RE.search(self.facts['machine']):
self.facts['architecture'] = 'i386'
else:
self.facts['architecture'] = self.facts['machine']
if self.facts['system'] == 'Linux':
self.get_distribution_facts()
elif self.facts['system'] == 'AIX':
rc, out, err = module.run_command("/usr/sbin/bootinfo -p")
data = out.split('\n')
self.facts['architecture'] = data[0]
# platform.dist() is deprecated in 2.6
# in 2.6 and newer, you should use platform.linux_distribution()
def get_distribution_facts(self):
# A list with OS Family members
OS_FAMILY = dict(
RedHat = 'RedHat', Fedora = 'RedHat', CentOS = 'RedHat', Scientific = 'RedHat',
SLC = 'RedHat', Ascendos = 'RedHat', CloudLinux = 'RedHat', PSBM = 'RedHat',
OracleLinux = 'RedHat', OVS = 'RedHat', OEL = 'RedHat', Amazon = 'RedHat',
XenServer = 'RedHat', Ubuntu = 'Debian', Debian = 'Debian', SLES = 'Suse',
SLED = 'Suse', OpenSuSE = 'Suse', SuSE = 'Suse', Gentoo = 'Gentoo',
Archlinux = 'Archlinux', Mandriva = 'Mandrake', Mandrake = 'Mandrake',
Solaris = 'Solaris', Nexenta = 'Solaris', OmniOS = 'Solaris', OpenIndiana = 'Solaris',
SmartOS = 'Solaris', AIX = 'AIX', Alpine = 'Alpine', MacOSX = 'Darwin'
)
if self.facts['system'] == 'AIX':
self.facts['distribution'] = 'AIX'
rc, out, err = module.run_command("/usr/bin/oslevel")
data = out.split('.')
self.facts['distribution_version'] = data[0]
self.facts['distribution_release'] = data[1]
elif self.facts['system'] == 'Darwin':
self.facts['distribution'] = 'MacOSX'
rc, out, err = module.run_command("/usr/bin/sw_vers -productVersion")
data = out.split()[-1]
self.facts['distribution_version'] = data
else:
dist = platform.dist()
self.facts['distribution'] = dist[0].capitalize() or 'NA'
self.facts['distribution_version'] = dist[1] or 'NA'
self.facts['distribution_release'] = dist[2] or 'NA'
# Try to handle the exceptions now ...
for (path, name) in Facts.OSDIST_DICT.items():
if os.path.exists(path):
if self.facts['distribution'] == 'Fedora':
pass
elif name == 'RedHat':
data = get_file_content(path)
if 'Red Hat' in data:
self.facts['distribution'] = name
else:
self.facts['distribution'] = data.split()[0]
elif name == 'OtherLinux':
data = get_file_content(path)
if 'Amazon' in data:
self.facts['distribution'] = 'Amazon'
self.facts['distribution_version'] = data.split()[-1]
elif name == 'OpenWrt':
data = get_file_content(path)
if 'OpenWrt' in data:
self.facts['distribution'] = name
version = re.search('DISTRIB_RELEASE="(.*)"', data)
if version:
self.facts['distribution_version'] = version.groups()[0]
release = re.search('DISTRIB_CODENAME="(.*)"', data)
if release:
self.facts['distribution_release'] = release.groups()[0]
elif name == 'Alpine':
data = get_file_content(path)
self.facts['distribution'] = 'Alpine'
self.facts['distribution_version'] = data
else:
self.facts['distribution'] = name
self.facts['os_family'] = self.facts['distribution']
if self.facts['distribution'] in OS_FAMILY:
self.facts['os_family'] = OS_FAMILY[self.facts['distribution']]
def get_cmdline(self):
data = get_file_content('/proc/cmdline')
if data:
self.facts['cmdline'] = {}
for piece in shlex.split(data):
item = piece.split('=', 1)
if len(item) == 1:
self.facts['cmdline'][item[0]] = True
else:
self.facts['cmdline'][item[0]] = item[1]
def get_public_ssh_host_keys(self):
dsa_filename = '/etc/ssh/ssh_host_dsa_key.pub'
rsa_filename = '/etc/ssh/ssh_host_rsa_key.pub'
if self.facts['system'] == 'Darwin':
dsa_filename = '/etc/ssh_host_dsa_key.pub'
rsa_filename = '/etc/ssh_host_rsa_key.pub'
dsa = get_file_content(dsa_filename)
rsa = get_file_content(rsa_filename)
if dsa is None:
dsa = 'NA'
else:
self.facts['ssh_host_key_dsa_public'] = dsa.split()[1]
if rsa is None:
rsa = 'NA'
else:
self.facts['ssh_host_key_rsa_public'] = rsa.split()[1]
def get_pkg_mgr_facts(self):
self.facts['pkg_mgr'] = 'unknown'
for pkg in Facts.PKG_MGRS:
if os.path.exists(pkg['path']):
self.facts['pkg_mgr'] = pkg['name']
if self.facts['system'] == 'OpenBSD':
self.facts['pkg_mgr'] = 'openbsd_pkg'
def get_lsb_facts(self):
lsb_path = module.get_bin_path('lsb_release')
if lsb_path is None:
return self.facts
rc, out, err = module.run_command([lsb_path, "-a"])
if rc == 0:
self.facts['lsb'] = {}
for line in out.split('\n'):
if len(line) < 1:
continue
value = line.split(':', 1)[1].strip()
if 'LSB Version:' in line:
self.facts['lsb']['release'] = value
elif 'Distributor ID:' in line:
self.facts['lsb']['id'] = value
elif 'Description:' in line:
self.facts['lsb']['description'] = value
elif 'Release:' in line:
self.facts['lsb']['release'] = value
elif 'Codename:' in line:
self.facts['lsb']['codename'] = value
if 'lsb' in self.facts and 'release' in self.facts['lsb']:
self.facts['lsb']['major_release'] = self.facts['lsb']['release'].split('.')[0]
def get_selinux_facts(self):
if not HAVE_SELINUX:
self.facts['selinux'] = False
return
self.facts['selinux'] = {}
if not selinux.is_selinux_enabled():
self.facts['selinux']['status'] = 'disabled'
else:
self.facts['selinux']['status'] = 'enabled'
try:
self.facts['selinux']['policyvers'] = selinux.security_policyvers()
except OSError, e:
self.facts['selinux']['policyvers'] = 'unknown'
try:
(rc, configmode) = selinux.selinux_getenforcemode()
if rc == 0:
self.facts['selinux']['config_mode'] = Facts.SELINUX_MODE_DICT.get(configmode, 'unknown')
else:
self.facts['selinux']['config_mode'] = 'unknown'
except OSError, e:
self.facts['selinux']['config_mode'] = 'unknown'
try:
mode = selinux.security_getenforce()
self.facts['selinux']['mode'] = Facts.SELINUX_MODE_DICT.get(mode, 'unknown')
except OSError, e:
self.facts['selinux']['mode'] = 'unknown'
try:
(rc, policytype) = selinux.selinux_getpolicytype()
if rc == 0:
self.facts['selinux']['type'] = policytype
else:
self.facts['selinux']['type'] = 'unknown'
except OSError, e:
self.facts['selinux']['type'] = 'unknown'
def get_date_time_facts(self):
self.facts['date_time'] = {}
now = datetime.datetime.now()
self.facts['date_time']['year'] = now.strftime('%Y')
self.facts['date_time']['month'] = now.strftime('%m')
self.facts['date_time']['day'] = now.strftime('%d')
self.facts['date_time']['hour'] = now.strftime('%H')
self.facts['date_time']['minute'] = now.strftime('%M')
self.facts['date_time']['second'] = now.strftime('%S')
self.facts['date_time']['epoch'] = now.strftime('%s')
self.facts['date_time']['date'] = now.strftime('%Y-%m-%d')
self.facts['date_time']['time'] = now.strftime('%H:%M:%S')
self.facts['date_time']['iso8601_micro'] = now.utcnow().strftime("%Y-%m-%dT%H:%M:%S.%fZ")
self.facts['date_time']['iso8601'] = now.utcnow().strftime("%Y-%m-%dT%H:%M:%SZ")
self.facts['date_time']['tz'] = time.strftime("%Z")
# User
def get_user_facts(self):
self.facts['user_id'] = getpass.getuser()
class Hardware(Facts):
"""
This is a generic Hardware subclass of Facts. This should be further
subclassed to implement per platform. If you subclass this, it
should define:
- memfree_mb
- memtotal_mb
- swapfree_mb
- swaptotal_mb
- processor (a list)
- processor_cores
- processor_count
All subclasses MUST define platform.
"""
platform = 'Generic'
def __new__(cls, *arguments, **keyword):
subclass = cls
for sc in Hardware.__subclasses__():
if sc.platform == platform.system():
subclass = sc
return super(cls, subclass).__new__(subclass, *arguments, **keyword)
def __init__(self):
Facts.__init__(self)
def populate(self):
return self.facts
class LinuxHardware(Hardware):
"""
Linux-specific subclass of Hardware. Defines memory and CPU facts:
- memfree_mb
- memtotal_mb
- swapfree_mb
- swaptotal_mb
- processor (a list)
- processor_cores
- processor_count
In addition, it also defines number of DMI facts and device facts.
"""
platform = 'Linux'
MEMORY_FACTS = ['MemTotal', 'SwapTotal', 'MemFree', 'SwapFree']
def __init__(self):
Hardware.__init__(self)
def populate(self):
self.get_cpu_facts()
self.get_memory_facts()
self.get_dmi_facts()
self.get_device_facts()
self.get_mount_facts()
return self.facts
def get_memory_facts(self):
if not os.access("/proc/meminfo", os.R_OK):
return
for line in open("/proc/meminfo").readlines():
data = line.split(":", 1)
key = data[0]
if key in LinuxHardware.MEMORY_FACTS:
val = data[1].strip().split(' ')[0]
self.facts["%s_mb" % key.lower()] = long(val) / 1024
def get_cpu_facts(self):
i = 0
physid = 0
sockets = {}
if not os.access("/proc/cpuinfo", os.R_OK):
return
self.facts['processor'] = []
for line in open("/proc/cpuinfo").readlines():
data = line.split(":", 1)
key = data[0].strip()
# model name is for Intel arch, Processor (mind the uppercase P)
# works for some ARM devices, like the Sheevaplug.
if key == 'model name' or key == 'Processor':
if 'processor' not in self.facts:
self.facts['processor'] = []
self.facts['processor'].append(data[1].strip())
i += 1
elif key == 'physical id':
physid = data[1].strip()
if physid not in sockets:
sockets[physid] = 1
elif key == 'cpu cores':
sockets[physid] = int(data[1].strip())
if len(sockets) > 0:
self.facts['processor_count'] = len(sockets)
self.facts['processor_cores'] = reduce(lambda x, y: x + y, sockets.values())
else:
self.facts['processor_count'] = i
self.facts['processor_cores'] = 'NA'
def get_dmi_facts(self):
def execute(cmd):
p = subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
(out, err) = p.communicate()
if p.returncode or err:
return None
return out.rstrip()
if os.path.exists('/sys/devices/virtual/dmi/id/product_name'):
# Use kernel DMI info, if available
# DMI SPEC -- http://www.dmtf.org/sites/default/files/standards/documents/DSP0134_2.7.0.pdf
FORM_FACTOR = [ "Unknown", "Other", "Unknown", "Desktop",
"Low Profile Desktop", "Pizza Box", "Mini Tower", "Tower",
"Portable", "Laptop", "Notebook", "Hand Held", "Docking Station",
"All In One", "Sub Notebook", "Space-saving", "Lunch Box",
"Main Server Chassis", "Expansion Chassis", "Sub Chassis",
"Bus Expansion Chassis", "Peripheral Chassis", "RAID Chassis",
"Rack Mount Chassis", "Sealed-case PC", "Multi-system",
"CompactPCI", "AdvancedTCA", "Blade" ]
DMI_DICT = dict(
bios_date = '/sys/devices/virtual/dmi/id/bios_date',
bios_version = '/sys/devices/virtual/dmi/id/bios_version',
form_factor = '/sys/devices/virtual/dmi/id/chassis_type',
product_name = '/sys/devices/virtual/dmi/id/product_name',
product_serial = '/sys/devices/virtual/dmi/id/product_serial',
product_uuid = '/sys/devices/virtual/dmi/id/product_uuid',
product_version = '/sys/devices/virtual/dmi/id/product_version',
system_vendor = '/sys/devices/virtual/dmi/id/sys_vendor',
)
for (key,path) in DMI_DICT.items():
data = get_file_content(path)
if data is not None:
if key == 'form_factor':
try:
self.facts['form_factor'] = FORM_FACTOR[int(data)]
except IndexError, e:
self.facts['form_factor'] = 'unknown (%s)' % data
else:
self.facts[key] = data
else:
self.facts[key] = 'NA'
elif 'dmidecode' in sys.modules.keys():
# Use python dmidecode, if available
DMI_DICT = dict(
bios_date = '/dmidecode/BIOSinfo/ReleaseDate',
bios_version = '/dmidecode/BIOSinfo/BIOSrevision',
form_factor = '/dmidecode/ChassisInfo/ChassisType',
product_name = '/dmidecode/SystemInfo/ProductName',
product_serial = '/dmidecode/SystemInfo/SerialNumber',
product_uuid = '/dmidecode/SystemInfo/SystemUUID',
product_version = '/dmidecode/SystemInfo/Version',
system_vendor = '/dmidecode/SystemInfo/Manufacturer',
)
dmixml = dmidecode.dmidecodeXML()
dmixml.SetResultType(dmidecode.DMIXML_DOC)
xmldoc = dmixml.QuerySection('all')
dmixp = xmldoc.xpathNewContext()
for (key,path) in DMI_DICT.items():
try:
data = dmixp.xpathEval(path)
if len(data) > 0:
self.facts[key] = data[0].get_content()
else:
self.facts[key] = 'Error'
except:
self.facts[key] = 'NA'
else:
# Fall back to using dmidecode, if available
self.facts['bios_date'] = execute('dmidecode -s bios-release-date') or 'NA'
self.facts['bios_version'] = execute('dmidecode -s bios-version') or 'NA'
self.facts['form_factor'] = execute('dmidecode -s chassis-type') or 'NA'
self.facts['product_name'] = execute('dmidecode -s system-product-name') or 'NA'
self.facts['product_serial'] = execute('dmidecode -s system-serial-number') or 'NA'
self.facts['product_uuid'] = execute('dmidecode -s system-uuid') or 'NA'
self.facts['product_version'] = execute('dmidecode -s system-version') or 'NA'
self.facts['system_vendor'] = execute('dmidecode -s system-manufacturer') or 'NA'
def get_mount_facts(self):
self.facts['mounts'] = []
mtab = get_file_content('/etc/mtab', '')
for line in mtab.split('\n'):
if line.startswith('/'):
fields = line.rstrip('\n').split()
self.facts['mounts'].append({'mount': fields[1], 'device':fields[0], 'fstype': fields[2], 'options': fields[3]})
def get_device_facts(self):
self.facts['devices'] = {}
lspci = module.get_bin_path('lspci')
if lspci:
rc, pcidata, err = module.run_command(lspci)
else:
pcidata = None
try:
block_devs = os.listdir("/sys/block")
except OSError:
return
for block in block_devs:
virtual = 1
sysfs_no_links = 0
try:
path = os.readlink(os.path.join("/sys/block/", block))
except OSError, e:
if e.errno == errno.EINVAL:
path = block
sysfs_no_links = 1
else:
continue
if "virtual" in path:
continue
sysdir = os.path.join("/sys/block", path)
if sysfs_no_links == 1:
for folder in os.listdir(sysdir):
if "device" in folder:
virtual = 0
break
if virtual:
continue
d = {}
diskname = os.path.basename(sysdir)
for key in ['vendor', 'model']:
d[key] = get_file_content(sysdir + "/device/" + key)
for key,test in [ ('removable','/removable'), \
('support_discard','/queue/discard_granularity'),
]:
d[key] = get_file_content(sysdir + test)
d['partitions'] = {}
for folder in os.listdir(sysdir):
m = re.search("(" + diskname + "\d+)", folder)
if m:
part = {}
partname = m.group(1)
part_sysdir = sysdir + "/" + partname
part['start'] = get_file_content(part_sysdir + "/start",0)
part['sectors'] = get_file_content(part_sysdir + "/size",0)
part['sectorsize'] = get_file_content(part_sysdir + "/queue/hw_sector_size",512)
part['size'] = module.pretty_bytes((float(part['sectors']) * float(part['sectorsize'])))
d['partitions'][partname] = part
d['rotational'] = get_file_content(sysdir + "/queue/rotational")
d['scheduler_mode'] = ""
scheduler = get_file_content(sysdir + "/queue/scheduler")
m = re.match(".*?(\[(.*)\])", scheduler)
if m:
d['scheduler_mode'] = m.group(2)
d['sectors'] = get_file_content(sysdir + "/size")
if not d['sectors']:
d['sectors'] = 0
d['sectorsize'] = get_file_content(sysdir + "/queue/hw_sector_size")
if not d['sectorsize']:
d['sectorsize'] = 512
d['size'] = module.pretty_bytes(float(d['sectors']) * float(d['sectorsize']))
d['host'] = ""
m = re.match(".+/[a-f0-9]+:([a-f0-9]+:[a-f0-9]+\.[a-f0-9]+)/host\d+/", sysdir)
if m and pcidata:
pciid = m.group(1)
did = re.escape(pciid)
m = re.search("^" + did + "\s(.*)$", pcidata, re.MULTILINE)
d['host'] = m.group(1)
d['holders'] = []
for folder in os.listdir(sysdir + "/holders"):
if not folder.startswith("dm-"):
continue
name = get_file_content(sysdir + "/holders/" + folder + "/dm/name")
if name:
d['holders'].append(name)
else:
d['holders'].append(folder)
self.facts['devices'][diskname] = d
class SunOSHardware(Hardware):
"""
In addition to the generic memory and cpu facts, this also sets
swap_reserved_mb and swap_allocated_mb that is available from *swap -s*.
"""
platform = 'SunOS'
def __init__(self):
Hardware.__init__(self)
def populate(self):
self.get_cpu_facts()
self.get_memory_facts()
return self.facts
def get_cpu_facts(self):
rc, out, err = module.run_command("/usr/sbin/psrinfo -v")
self.facts['processor'] = []
for line in out.split('\n'):
if 'processor operates' in line:
if 'processor' not in self.facts:
self.facts['processor'] = []
self.facts['processor'].append(line.strip())
self.facts['processor_cores'] = 'NA'
self.facts['processor_count'] = len(self.facts['processor'])
def get_memory_facts(self):
rc, out, err = module.run_command(["/usr/sbin/prtconf"])
for line in out.split('\n'):
if 'Memory size' in line:
self.facts['memtotal_mb'] = line.split()[2]
rc, out, err = module.run_command("/usr/sbin/swap -s")
allocated = long(out.split()[1][:-1])
reserved = long(out.split()[5][:-1])
used = long(out.split()[8][:-1])
free = long(out.split()[10][:-1])
self.facts['swapfree_mb'] = free / 1024
self.facts['swaptotal_mb'] = (free + used) / 1024
self.facts['swap_allocated_mb'] = allocated / 1024
self.facts['swap_reserved_mb'] = reserved / 1024
class OpenBSDHardware(Hardware):
"""
OpenBSD-specific subclass of Hardware. Defines memory, CPU and device facts:
- memfree_mb
- memtotal_mb
- swapfree_mb
- swaptotal_mb
- processor (a list)
- processor_cores
- processor_count
- processor_speed
- devices
"""
platform = 'OpenBSD'
DMESG_BOOT = '/var/run/dmesg.boot'
def __init__(self):
Hardware.__init__(self)
def populate(self):
self.sysctl = self.get_sysctl()
self.get_memory_facts()
self.get_processor_facts()
self.get_device_facts()
return self.facts
def get_sysctl(self):
rc, out, err = module.run_command(["/sbin/sysctl", "hw"])
if rc != 0:
return dict()
sysctl = dict()
for line in out.splitlines():
(key, value) = line.split('=')
sysctl[key] = value.strip()
return sysctl
def get_memory_facts(self):
# Get free memory. vmstat output looks like:
# procs memory page disks traps cpu
# r b w avm fre flt re pi po fr sr wd0 fd0 int sys cs us sy id
# 0 0 0 47512 28160 51 0 0 0 0 0 1 0 116 89 17 0 1 99
rc, out, err = module.run_command("/usr/bin/vmstat")
if rc == 0:
self.facts['memfree_mb'] = long(out.splitlines()[-1].split()[4]) / 1024
self.facts['memtotal_mb'] = long(self.sysctl['hw.usermem']) / 1024 / 1024
# Get swapctl info. swapctl output looks like:
# total: 69268 1K-blocks allocated, 0 used, 69268 available
# And for older OpenBSD:
# total: 69268k bytes allocated = 0k used, 69268k available
rc, out, err = module.run_command("/sbin/swapctl -sk")
if rc == 0:
data = out.split()
self.facts['swapfree_mb'] = long(data[-2].translate(None, "kmg")) / 1024
self.facts['swaptotal_mb'] = long(data[1].translate(None, "kmg")) / 1024
def get_processor_facts(self):
processor = []
dmesg_boot = get_file_content(OpenBSDHardware.DMESG_BOOT)
if not dmesg_boot:
rc, dmesg_boot, err = module.run_command("/sbin/dmesg")
i = 0
for line in dmesg_boot.splitlines():
if line.split(' ', 1)[0] == 'cpu%i:' % i:
processor.append(line.split(' ', 1)[1])
i = i + 1
processor_count = i
self.facts['processor'] = processor
self.facts['processor_count'] = processor_count
# I found no way to figure out the number of Cores per CPU in OpenBSD
self.facts['processor_cores'] = 'NA'
def get_device_facts(self):
devices = []
devices.extend(self.sysctl['hw.disknames'].split(','))
self.facts['devices'] = devices
class FreeBSDHardware(Hardware):
"""
FreeBSD-specific subclass of Hardware. Defines memory and CPU facts:
- memfree_mb
- memtotal_mb
- swapfree_mb
- swaptotal_mb
- processor (a list)
- processor_cores
- processor_count
- devices
"""
platform = 'FreeBSD'
DMESG_BOOT = '/var/run/dmesg.boot'
def __init__(self):
Hardware.__init__(self)
def populate(self):
self.get_cpu_facts()
self.get_memory_facts()
self.get_device_facts()
self.get_mount_facts()
return self.facts
def get_cpu_facts(self):
self.facts['processor'] = []
rc, out, err = module.run_command("/sbin/sysctl -n hw.ncpu")
self.facts['processor_count'] = out.strip()
dmesg_boot = get_file_content(FreeBSDHardware.DMESG_BOOT)
if not dmesg_boot:
rc, dmesg_boot, err = module.run_command("/sbin/dmesg")
for line in dmesg_boot.split('\n'):
if 'CPU:' in line:
cpu = re.sub(r'CPU:\s+', r"", line)
self.facts['processor'].append(cpu.strip())
if 'Logical CPUs per core' in line:
self.facts['processor_cores'] = line.split()[4]
def get_memory_facts(self):
rc, out, err = module.run_command("/sbin/sysctl vm.stats")
for line in out.split('\n'):
data = line.split()
if 'vm.stats.vm.v_page_size' in line:
pagesize = long(data[1])
if 'vm.stats.vm.v_page_count' in line:
pagecount = long(data[1])
if 'vm.stats.vm.v_free_count' in line:
freecount = long(data[1])
self.facts['memtotal_mb'] = pagesize * pagecount / 1024 / 1024
self.facts['memfree_mb'] = pagesize * freecount / 1024 / 1024
# Get swapinfo. swapinfo output looks like:
# Device 1M-blocks Used Avail Capacity
# /dev/ada0p3 314368 0 314368 0%
#
rc, out, err = module.run_command("/usr/sbin/swapinfo -m")
lines = out.split('\n')
if len(lines[-1]) == 0:
lines.pop()
data = lines[-1].split()
self.facts['swaptotal_mb'] = data[1]
self.facts['swapfree_mb'] = data[3]
def get_mount_facts(self):
self.facts['mounts'] = []
fstab = get_file_content('/etc/fstab')
if fstab:
for line in fstab.split('\n'):
if line.startswith('#') or line.strip() == '':
continue
fields = re.sub(r'\s+',' ',line.rstrip('\n')).split()
self.facts['mounts'].append({'mount': fields[1] , 'device': fields[0], 'fstype' : fields[2], 'options': fields[3]})
def get_device_facts(self):
sysdir = '/dev'
self.facts['devices'] = {}
drives = re.compile('(ada?\d+|da\d+|a?cd\d+)') #TODO: rc, disks, err = module.run_command("/sbin/sysctl kern.disks")
slices = re.compile('(ada?\d+s\d+\w*|da\d+s\d+\w*)')
if os.path.isdir(sysdir):
dirlist = sorted(os.listdir(sysdir))
for device in dirlist:
d = drives.match(device)
if d:
self.facts['devices'][d.group(1)] = []
s = slices.match(device)
if s:
self.facts['devices'][d.group(1)].append(s.group(1))
class AIX(Hardware):
"""
AIX-specific subclass of Hardware. Defines memory and CPU facts:
- memfree_mb
- memtotal_mb
- swapfree_mb
- swaptotal_mb
- processor (a list)
- processor_cores
- processor_count
"""
platform = 'AIX'
def __init__(self):
Hardware.__init__(self)
def populate(self):
self.get_cpu_facts()
self.get_memory_facts()
self.get_dmi_facts()
return self.facts
def get_cpu_facts(self):
self.facts['processor'] = []
rc, out, err = module.run_command("/usr/sbin/lsattr -El proc0 -a type")
data = out.split(' ')
self.facts['processor'] = data[1]
rc, out, err = module.run_command("/usr/sbin/lsattr -El proc0 -a smt_threads")
data = out.split(' ')
self.facts['processor_cores'] = int(data[1])
rc, out, err = module.run_command("/usr/sbin/lsdev -Cc processor")
i = 0
for line in out.split('\n'):
data = line.split(':')
if 'Available' in line:
i += 1
self.facts['processor_count'] = int(i)
def get_memory_facts(self):
pagesize = 4096
rc, out, err = module.run_command("/usr/bin/vmstat -v")
for line in out.split('\n'):
data = line.split()
if 'memory pages' in line:
pagecount = long(data[0])
if 'free pages' in line:
freecount = long(data[0])
self.facts['memtotal_mb'] = pagesize * pagecount / 1024 / 1024
self.facts['memfree_mb'] = pagesize * freecount / 1024 / 1024
# Get swapinfo. swapinfo output looks like:
# Device 1M-blocks Used Avail Capacity
# /dev/ada0p3 314368 0 314368 0%
#
rc, out, err = module.run_command("/usr/sbin/lsps -s")
lines = out.split('\n')
data = lines[1].split()
swaptotal_mb = long(data[0].rstrip('MB'))
percused = int(data[1].rstrip('%'))
self.facts['swaptotal_mb'] = swaptotal_mb
self.facts['swapfree_mb'] = long(swaptotal_mb * ( 100 - percused ) / 100)
def get_dmi_facts(self):
rc, out, err = module.run_command("/usr/sbin/lsattr -El sys0 -a fwversion")
data = out.split()
self.facts['firmware_version'] = data[1].strip('IBM,')
class Darwin(Hardware):
"""
Darwin-specific subclass of Hardware. Defines memory and CPU facts:
- processor
- processor_cores
- memtotal_mb
- memfree_mb
- model
- osversion
- osrevision
"""
platform = 'Darwin'
def __init__(self):
Hardware.__init__(self)
def populate(self):
self.sysctl = self.get_sysctl()
self.get_mac_facts()
self.get_cpu_facts()
self.get_memory_facts()
return self.facts
def get_sysctl(self):
rc, out, err = module.run_command(["/usr/sbin/sysctl", "hw", "machdep", "kern"])
if rc != 0:
return dict()
sysctl = dict()
for line in out.splitlines():
if line.rstrip("\n"):
(key, value) = re.split(' = |: ', line, maxsplit=1)
sysctl[key] = value.strip()
return sysctl
def get_mac_facts(self):
self.facts['model'] = self.sysctl['hw.model']
self.facts['osversion'] = self.sysctl['kern.osversion']
self.facts['osrevision'] = self.sysctl['kern.osrevision']
def get_cpu_facts(self):
self.facts['processor'] = self.sysctl['machdep.cpu.brand_string']
self.facts['processor_cores'] = self.sysctl['machdep.cpu.core_count']
def get_memory_facts(self):
self.facts['memtotal_mb'] = long(self.sysctl['hw.memsize']) / 1024 / 1024
self.facts['memfree_mb'] = long(self.sysctl['hw.usermem']) / 1024 / 1024
class Network(Facts):
"""
This is a generic Network subclass of Facts. This should be further
subclassed to implement per platform. If you subclass this,
you must define:
- interfaces (a list of interface names)
- interface_<name> dictionary of ipv4, ipv6, and mac address information.
All subclasses MUST define platform.
"""
platform = 'Generic'
IPV6_SCOPE = { '0' : 'global',
'10' : 'host',
'20' : 'link',
'40' : 'admin',
'50' : 'site',
'80' : 'organization' }
def __new__(cls, *arguments, **keyword):
subclass = cls
for sc in Network.__subclasses__():
if sc.platform == platform.system():
subclass = sc
return super(cls, subclass).__new__(subclass, *arguments, **keyword)
def __init__(self):
Facts.__init__(self)
def populate(self):
return self.facts
class LinuxNetwork(Network):
"""
This is a Linux-specific subclass of Network. It defines
- interfaces (a list of interface names)
- interface_<name> dictionary of ipv4, ipv6, and mac address information.
- all_ipv4_addresses and all_ipv6_addresses: lists of all configured addresses.
- ipv4_address and ipv6_address: the first non-local address for each family.
"""
platform = 'Linux'
def __init__(self):
Network.__init__(self)
def populate(self):
ip_path = module.get_bin_path('ip')
if ip_path is None:
return self.facts
default_ipv4, default_ipv6 = self.get_default_interfaces(ip_path)
interfaces, ips = self.get_interfaces_info(ip_path, default_ipv4, default_ipv6)
self.facts['interfaces'] = interfaces.keys()
for iface in interfaces:
self.facts[iface] = interfaces[iface]
self.facts['default_ipv4'] = default_ipv4
self.facts['default_ipv6'] = default_ipv6
self.facts['all_ipv4_addresses'] = ips['all_ipv4_addresses']
self.facts['all_ipv6_addresses'] = ips['all_ipv6_addresses']
return self.facts
def get_default_interfaces(self, ip_path):
# Use the commands:
# ip -4 route get 8.8.8.8 -> Google public DNS
# ip -6 route get 2404:6800:400a:800::1012 -> ipv6.google.com
# to find out the default outgoing interface, address, and gateway
command = dict(
v4 = [ip_path, '-4', 'route', 'get', '8.8.8.8'],
v6 = [ip_path, '-6', 'route', 'get', '2404:6800:400a:800::1012']
)
interface = dict(v4 = {}, v6 = {})
for v in 'v4', 'v6':
12 years ago
if v == 'v6' and not socket.has_ipv6:
continue
rc, out, err = module.run_command(command[v])
if not out:
# v6 routing may result in
# RTNETLINK answers: Invalid argument
continue
words = out.split('\n')[0].split()
# A valid output starts with the queried address on the first line
if len(words) > 0 and words[0] == command[v][-1]:
for i in range(len(words) - 1):
if words[i] == 'dev':
interface[v]['interface'] = words[i+1]
elif words[i] == 'src':
interface[v]['address'] = words[i+1]
elif words[i] == 'via' and words[i+1] != command[v][-1]:
interface[v]['gateway'] = words[i+1]
return interface['v4'], interface['v6']
def get_interfaces_info(self, ip_path, default_ipv4, default_ipv6):
interfaces = {}
ips = dict(
all_ipv4_addresses = [],
all_ipv6_addresses = [],
)
for path in glob.glob('/sys/class/net/*'):
if not os.path.isdir(path): continue
device = os.path.basename(path)
interfaces[device] = { 'device': device }
if os.path.exists(os.path.join(path, 'address')):
macaddress = open(os.path.join(path, 'address')).read().strip()
if macaddress and macaddress != '00:00:00:00:00:00':
interfaces[device]['macaddress'] = macaddress
if os.path.exists(os.path.join(path, 'mtu')):
interfaces[device]['mtu'] = int(open(os.path.join(path, 'mtu')).read().strip())
if os.path.exists(os.path.join(path, 'operstate')):
interfaces[device]['active'] = open(os.path.join(path, 'operstate')).read().strip() != 'down'
# if os.path.exists(os.path.join(path, 'carrier')):
# interfaces[device]['link'] = open(os.path.join(path, 'carrier')).read().strip() == '1'
if os.path.exists(os.path.join(path, 'device','driver', 'module')):
interfaces[device]['module'] = os.path.basename(os.path.realpath(os.path.join(path, 'device', 'driver', 'module')))
if os.path.exists(os.path.join(path, 'type')):
type = open(os.path.join(path, 'type')).read().strip()
if type == '1':
interfaces[device]['type'] = 'ether'
elif type == '512':
interfaces[device]['type'] = 'ppp'
elif type == '772':
interfaces[device]['type'] = 'loopback'
if os.path.exists(os.path.join(path, 'bridge')):
interfaces[device]['type'] = 'bridge'
interfaces[device]['interfaces'] = [ os.path.basename(b) for b in glob.glob(os.path.join(path, 'brif', '*')) ]
if os.path.exists(os.path.join(path, 'bridge', 'bridge_id')):
interfaces[device]['id'] = open(os.path.join(path, 'bridge', 'bridge_id')).read().strip()
if os.path.exists(os.path.join(path, 'bridge', 'stp_state')):
interfaces[device]['stp'] = open(os.path.join(path, 'bridge', 'stp_state')).read().strip() == '1'
if os.path.exists(os.path.join(path, 'bonding')):
interfaces[device]['type'] = 'bonding'
interfaces[device]['slaves'] = open(os.path.join(path, 'bonding', 'slaves')).read().split()
interfaces[device]['mode'] = open(os.path.join(path, 'bonding', 'mode')).read().split()[0]
interfaces[device]['miimon'] = open(os.path.join(path, 'bonding', 'miimon')).read().split()[0]
interfaces[device]['lacp_rate'] = open(os.path.join(path, 'bonding', 'lacp_rate')).read().split()[0]
primary = open(os.path.join(path, 'bonding', 'primary')).read()
if primary:
interfaces[device]['primary'] = primary
interfaces[device]['all_slaves_active'] = open(os.path.join(path, 'bonding', 'all_slaves_active')).read() == '1'
output = subprocess.Popen(['/sbin/ip', 'addr', 'show', device], stdout=subprocess.PIPE, stderr=subprocess.PIPE).communicate()[0]
for line in output.split('\n'):
if not line:
continue
words = line.split()
if words[0] == 'inet':
if '/' in words[1]:
address, netmask_length = words[1].split('/')
else:
# pointopoint interfaces do not have a prefix
address = words[1]
netmask_length = "32"
address_bin = struct.unpack('!L', socket.inet_aton(address))[0]
netmask_bin = (1<<32) - (1<<32>>int(netmask_length))
netmask = socket.inet_ntoa(struct.pack('!L', netmask_bin))
network = socket.inet_ntoa(struct.pack('!L', address_bin & netmask_bin))
iface = words[-1]
if iface != device:
interfaces[iface] = {}
interfaces[iface].update(interfaces[device])
interfaces[iface]['ipv4'] = {'address': address,
'netmask': netmask,
'network': network}
# If this is the default address, update default_ipv4
if 'address' in default_ipv4 and default_ipv4['address'] == address:
default_ipv4['netmask'] = netmask
default_ipv4['network'] = network
default_ipv4['macaddress'] = macaddress
default_ipv4['mtu'] = interfaces[device]['mtu']
default_ipv4['type'] = interfaces[device]['type']
default_ipv4['alias'] = words[-1]
if not address.startswith('127.'):
ips['all_ipv4_addresses'].append(address)
elif words[0] == 'inet6':
address, prefix = words[1].split('/')
scope = words[3]
if 'ipv6' not in interfaces[device]:
interfaces[device]['ipv6'] = []
interfaces[device]['ipv6'].append({
'address' : address,
'prefix' : prefix,
'scope' : scope
})
# If this is the default address, update default_ipv6
if 'address' in default_ipv6 and default_ipv6['address'] == address:
default_ipv6['prefix'] = prefix
default_ipv6['scope'] = scope
default_ipv6['macaddress'] = macaddress
default_ipv6['mtu'] = interfaces[device]['mtu']
default_ipv6['type'] = interfaces[device]['type']
if not address == '::1':
ips['all_ipv6_addresses'].append(address)
return interfaces, ips
class GenericBsdIfconfigNetwork(Network):
"""
This is a generic BSD subclass of Network using the ifconfig command.
It defines
- interfaces (a list of interface names)
- interface_<name> dictionary of ipv4, ipv6, and mac address information.
- all_ipv4_addresses and all_ipv6_addresses: lists of all configured addresses.
It currently does not define
- default_ipv4 and default_ipv6
- type, mtu and network on interfaces
"""
platform = 'Generic_BSD_Ifconfig'
def __init__(self):
Network.__init__(self)
def populate(self):
ifconfig_path = module.get_bin_path('ifconfig')
if ifconfig_path is None:
return self.facts
route_path = module.get_bin_path('route')
if route_path is None:
return self.facts
default_ipv4, default_ipv6 = self.get_default_interfaces(route_path)
interfaces, ips = self.get_interfaces_info(ifconfig_path)
self.merge_default_interface(default_ipv4, interfaces, 'ipv4')
self.merge_default_interface(default_ipv6, interfaces, 'ipv6')
self.facts['interfaces'] = interfaces.keys()
for iface in interfaces:
self.facts[iface] = interfaces[iface]
self.facts['default_ipv4'] = default_ipv4
self.facts['default_ipv6'] = default_ipv6
self.facts['all_ipv4_addresses'] = ips['all_ipv4_addresses']
self.facts['all_ipv6_addresses'] = ips['all_ipv6_addresses']
return self.facts
def get_default_interfaces(self, route_path):
# Use the commands:
# route -n get 8.8.8.8 -> Google public DNS
# route -n get 2404:6800:400a:800::1012 -> ipv6.google.com
# to find out the default outgoing interface, address, and gateway
command = dict(
v4 = [route_path, '-n', 'get', '8.8.8.8'],
v6 = [route_path, '-n', 'get', '2404:6800:400a:800::1012']
)
interface = dict(v4 = {}, v6 = {})
for v in 'v4', 'v6':
if v == 'v6' and not socket.has_ipv6:
continue
rc, out, err = module.run_command(command[v])
if not out:
# v6 routing may result in
# RTNETLINK answers: Invalid argument
continue
lines = out.split('\n')
for line in lines:
words = line.split()
# look for first word starting interface
if len(words) > 0 and words[0] == 'interface:':
interface[v]['interface'] = words[1]
return interface['v4'], interface['v6']
def get_interfaces_info(self, ifconfig_path):
interfaces = {}
current_if = {}
ips = dict(
all_ipv4_addresses = [],
all_ipv6_addresses = [],
)
rc, out, err = module.run_command([ifconfig_path])
for line in out.split('\n'):
if line:
words = line.split()
if re.match('^\S', line) and len(words) > 3:
current_if = self.parse_interface_line(words)
interfaces[ current_if['device'] ] = current_if
elif words[0].startswith('options='):
self.parse_options_line(words, current_if, ips)
elif words[0] == 'nd6':
self.parse_nd6_line(words, current_if, ips)
elif words[0] == 'ether':
self.parse_ether_line(words, current_if, ips)
elif words[0] == 'media:':
self.parse_media_line(words, current_if, ips)
elif words[0] == 'status:':
self.parse_status_line(words, current_if, ips)
elif words[0] == 'lladdr':
self.parse_lladdr_line(words, current_if, ips)
elif words[0] == 'inet':
self.parse_inet_line(words, current_if, ips)
elif words[0] == 'inet6':
self.parse_inet6_line(words, current_if, ips)
else:
self.parse_unknown_line(words, current_if, ips)
return interfaces, ips
def parse_interface_line(self, words):
device = words[0][0:-1]
current_if = {'device': device, 'ipv4': [], 'ipv6': [], 'type': 'unknown'}
current_if['flags'] = self.get_options(words[1])
current_if['mtu'] = words[3]
current_if['macaddress'] = 'unknown' # will be overwritten later
return current_if
def parse_options_line(self, words, current_if, ips):
# Mac has options like this...
current_if['options'] = self.get_options(words[0])
def parse_nd6_line(self, words, current_if, ips):
# FreBSD has options like this...
current_if['options'] = self.get_options(words[1])
def parse_ether_line(self, words, current_if, ips):
current_if['macaddress'] = words[1]
def parse_media_line(self, words, current_if, ips):
# not sure if this is useful - we also drop information
current_if['media'] = words[1]
if len(words) > 2:
current_if['media_select'] = words[2]
if len(words) > 3:
current_if['media_type'] = words[3][1:]
if len(words) > 4:
current_if['media_options'] = self.get_options(words[4])
def parse_status_line(self, words, current_if, ips):
current_if['status'] = words[1]
def parse_lladdr_line(self, words, current_if, ips):
current_if['lladdr'] = words[1]
def parse_inet_line(self, words, current_if, ips):
address = {'address': words[1]}
# deal with hex netmask
if words[3].startswith('0x'):
address['netmask'] = socket.inet_ntoa(struct.pack('!L', int(words[3], base=16)))
else:
# otherwise assume this is a dotted quad
address['netmask'] = words[3]
# calculate the network
address_bin = struct.unpack('!L', socket.inet_aton(address['address']))[0]
netmask_bin = struct.unpack('!L', socket.inet_aton(address['netmask']))[0]
address['network'] = socket.inet_ntoa(struct.pack('!L', address_bin & netmask_bin))
# broadcast may be given or we need to calculate
if len(words) > 5:
address['broadcast'] = words[5]
else:
address['broadcast'] = socket.inet_ntoa(struct.pack('!L', address_bin | (~netmask_bin & 0xffffffff)))
# add to our list of addresses
if not words[1].startswith('127.'):
ips['all_ipv4_addresses'].append(address['address'])
current_if['ipv4'].append(address)
def parse_inet6_line(self, words, current_if, ips):
address = {'address': words[1]}
if (len(words) >= 4) and (words[2] == 'prefixlen'):
address['prefix'] = words[3]
if (len(words) >= 6) and (words[4] == 'scopeid'):
address['scope'] = words[5]
if not address['address'] == '::1' and not address['address'] == 'fe80::1%lo0':
ips['all_ipv6_addresses'].append(address['address'])
current_if['ipv6'].append(address)
def parse_unknown_line(self, words, current_if, ips):
# we are going to ignore unknown lines here - this may be
# a bad idea - but you can override it in your subclass
pass
def get_options(self, option_string):
start = option_string.find('<') + 1
end = option_string.rfind('>')
if (start > 0) and (end > 0) and (end > start + 1):
option_csv = option_string[start:end]
return option_csv.split(',')
else:
return []
def merge_default_interface(self, defaults, interfaces, ip_type):
if not 'interface' in defaults.keys():
return
ifinfo = interfaces[defaults['interface']]
# copy all the interface values across except addresses
for item in ifinfo.keys():
if item != 'ipv4' and item != 'ipv6':
defaults[item] = ifinfo[item]
if len(ifinfo[ip_type]) > 0:
for item in ifinfo[ip_type][0].keys():
defaults[item] = ifinfo[ip_type][0][item]
class DarwinNetwork(GenericBsdIfconfigNetwork, Network):
"""
This is the Mac OS X/Darwin Network Class.
It uses the GenericBsdIfconfigNetwork unchanged
"""
platform = 'Darwin'
# media line is different to the default FreeBSD one
def parse_media_line(self, words, current_if, ips):
# not sure if this is useful - we also drop information
current_if['media'] = 'Unknown' # Mac does not give us this
current_if['media_select'] = words[1]
if len(words) > 2:
current_if['media_type'] = words[2][1:]
if len(words) > 3:
current_if['media_options'] = self.get_options(words[3])
class FreeBSDNetwork(GenericBsdIfconfigNetwork, Network):
"""
This is the FreeBSD Network Class.
It uses the GenericBsdIfconfigNetwork unchanged
"""
platform = 'FreeBSD'
class OpenBSDNetwork(GenericBsdIfconfigNetwork, Network):
"""
This is the OpenBSD Network Class.
It uses the GenericBsdIfconfigNetwork.
"""
platform = 'OpenBSD'
# Return macaddress instead of lladdr
def parse_lladdr_line(self, words, current_if, ips):
current_if['macaddress'] = words[1]
class Virtual(Facts):
"""
This is a generic Virtual subclass of Facts. This should be further
subclassed to implement per platform. If you subclass this,
you should define:
- virtualization_type
- virtualization_role
- container (e.g. solaris zones, freebsd jails, linux containers)
All subclasses MUST define platform.
"""
def __new__(cls, *arguments, **keyword):
subclass = cls
for sc in Virtual.__subclasses__():
if sc.platform == platform.system():
subclass = sc
return super(cls, subclass).__new__(subclass, *arguments, **keyword)
def __init__(self):
Facts.__init__(self)
def populate(self):
return self.facts
class LinuxVirtual(Virtual):
"""
This is a Linux-specific subclass of Virtual. It defines
- virtualization_type
- virtualization_role
"""
platform = 'Linux'
def __init__(self):
Virtual.__init__(self)
def populate(self):
self.get_virtual_facts()
return self.facts
# For more information, check: http://people.redhat.com/~rjones/virt-what/
def get_virtual_facts(self):
if os.path.exists("/proc/xen"):
self.facts['virtualization_type'] = 'xen'
self.facts['virtualization_role'] = 'guest'
try:
for line in open('/proc/xen/capabilities'):
if "control_d" in line:
self.facts['virtualization_role'] = 'host'
except IOError:
pass
return
if os.path.exists('/proc/vz'):
self.facts['virtualization_type'] = 'openvz'
if os.path.exists('/proc/bc'):
self.facts['virtualization_role'] = 'host'
else:
self.facts['virtualization_role'] = 'guest'
return
product_name = get_file_content('/sys/devices/virtual/dmi/id/product_name')
if product_name in ['KVM', 'Bochs']:
self.facts['virtualization_type'] = 'kvm'
self.facts['virtualization_role'] = 'guest'
return
if product_name == 'VMware Virtual Platform':
self.facts['virtualization_type'] = 'VMware'
self.facts['virtualization_role'] = 'guest'
return
bios_vendor = get_file_content('/sys/devices/virtual/dmi/id/bios_vendor')
if bios_vendor == 'Xen':
self.facts['virtualization_type'] = 'xen'
self.facts['virtualization_role'] = 'guest'
return
if bios_vendor == 'innotek GmbH':
self.facts['virtualization_type'] = 'virtualbox'
self.facts['virtualization_role'] = 'guest'
return
sys_vendor = get_file_content('/sys/devices/virtual/dmi/id/sys_vendor')
# FIXME: This does also match hyperv
if sys_vendor == 'Microsoft Corporation':
self.facts['virtualization_type'] = 'VirtualPC'
self.facts['virtualization_role'] = 'guest'
return
if sys_vendor == 'Parallels Software International Inc.':
self.facts['virtualization_type'] = 'parallels'
self.facts['virtualization_role'] = 'guest'
return
if os.path.exists('/proc/self/status'):
for line in open('/proc/self/status').readlines():
if re.match('^VxID: \d+', line):
self.facts['virtualization_type'] = 'linux_vserver'
if re.match('^VxID: 0', line):
self.facts['virtualization_role'] = 'host'
else:
self.facts['virtualization_role'] = 'guest'
return
if os.path.exists('/proc/cpuinfo'):
for line in open('/proc/cpuinfo').readlines():
if re.match('^model name.*QEMU Virtual CPU', line):
self.facts['virtualization_type'] = 'kvm'
elif re.match('^vendor_id.*User Mode Linux', line):
self.facts['virtualization_type'] = 'uml'
elif re.match('^model name.*UML', line):
self.facts['virtualization_type'] = 'uml'
elif re.match('^vendor_id.*PowerVM Lx86', line):
self.facts['virtualization_type'] = 'powervm_lx86'
elif re.match('^vendor_id.*IBM/S390', line):
self.facts['virtualization_type'] = 'ibm_systemz'
else:
continue
self.facts['virtualization_role'] = 'guest'
return
# Beware that we can have both kvm and virtualbox running on a single system
if os.path.exists("/proc/modules"):
modules = []
for line in open("/proc/modules").readlines():
data = line.split(" ", 1)
modules.append(data[0])
if 'kvm' in modules:
self.facts['virtualization_type'] = 'kvm'
self.facts['virtualization_role'] = 'host'
return
if 'vboxdrv' in modules:
self.facts['virtualization_type'] = 'virtualbox'
self.facts['virtualization_role'] = 'host'
return
class SunOSVirtual(Virtual):
"""
This is a SunOS-specific subclass of Virtual. It defines
- virtualization_type
- virtualization_role
- container
"""
platform = 'SunOS'
def __init__(self):
Virtual.__init__(self)
def populate(self):
self.get_virtual_facts()
return self.facts
def get_virtual_facts(self):
rc, out, err = module.run_command("/usr/sbin/prtdiag")
for line in out.split('\n'):
if 'VMware' in line:
self.facts['virtualization_type'] = 'vmware'
self.facts['virtualization_role'] = 'guest'
if 'Parallels' in line:
self.facts['virtualization_type'] = 'parallels'
self.facts['virtualization_role'] = 'guest'
if 'VirtualBox' in line:
self.facts['virtualization_type'] = 'virtualbox'
self.facts['virtualization_role'] = 'guest'
if 'HVM domU' in line:
self.facts['virtualization_type'] = 'xen'
self.facts['virtualization_role'] = 'guest'
# Check if it's a zone
if os.path.exists("/usr/bin/zonename"):
rc, out, err = module.run_command("/usr/bin/zonename")
if out.rstrip() != "global":
self.facts['container'] = 'zone'
# Check if it's a branded zone (i.e. Solaris 8/9 zone)
if os.path.isdir('/.SUNWnative'):
self.facts['container'] = 'zone'
# If it's a zone check if we can detect if our global zone is itself virtualized.
# Relies on the "guest tools" (e.g. vmware tools) to be installed
if 'container' in self.facts and self.facts['container'] == 'zone':
rc, out, err = module.run_command("/usr/sbin/modinfo")
for line in out.split('\n'):
if 'VMware' in line:
self.facts['virtualization_type'] = 'vmware'
self.facts['virtualization_role'] = 'guest'
if 'VirtualBox' in line:
self.facts['virtualization_type'] = 'virtualbox'
self.facts['virtualization_role'] = 'guest'
def get_file_content(path, default=None):
data = default
if os.path.exists(path) and os.access(path, os.R_OK):
data = open(path).read().strip()
if len(data) == 0:
data = default
return data
def ansible_facts():
facts = {}
facts.update(Facts().populate())
facts.update(Hardware().populate())
facts.update(Network().populate())
facts.update(Virtual().populate())
return facts
# ===========================================
def run_setup(module):
setup_options = {}
facts = ansible_facts()
for (k, v) in facts.items():
setup_options["ansible_%s" % k.replace('-', '_')] = v
# if facter is installed, and we can use --json because
# ruby-json is ALSO installed, include facter data in the JSON
if os.path.exists("/usr/bin/facter"):
rc, out, err = module.run_command("/usr/bin/facter --json")
facter = True
try:
facter_ds = json.loads(out)
except:
facter = False
if facter:
for (k,v) in facter_ds.items():
setup_options["facter_%s" % k] = v
# ditto for ohai, but just top level string keys
# because it contains a lot of nested stuff we can't use for
# templating w/o making a nicer key for it (TODO)
if os.path.exists("/usr/bin/ohai"):
rc, out, err = module.run_command("/usr/bin/ohai")
12 years ago
ohai = True
try:
ohai_ds = json.loads(out)
except:
ohai = False
if ohai:
for (k,v) in ohai_ds.items():
if type(v) == str or type(v) == unicode:
k2 = "ohai_%s" % k.replace('-', '_')
setup_options[k2] = v
setup_result = { 'ansible_facts': {} }
for (k,v) in setup_options.items():
if module.params['filter'] == '*' or fnmatch.fnmatch(k, module.params['filter']):
setup_result['ansible_facts'][k] = v
# hack to keep --verbose from showing all the setup module results
setup_result['verbose_override'] = True
return setup_result
def main():
global module
module = AnsibleModule(
argument_spec = dict(
filter=dict(default="*", required=False),
),
supports_check_mode = True,
)
data = run_setup(module)
module.exit_json(**data)
# this is magic, see lib/ansible/module_common.py
#<<INCLUDE_ANSIBLE_MODULE_COMMON>>
main()