mirror of
https://salsa.debian.org/freeipa-team/freeipa.git
synced 2024-12-25 16:31:08 -06:00
ae7cd4702d
Operations in FIPS mode make impossible use of NTLMSSP when authenticating to trusted Active Directory domain controllers because RC4 cipher is not allowed. Instead, Kerberos authentication have to be used. We switched to enforce Kerberos authentication when communicating with trusted domains' domain controllers everywhere. Kerberos library uses system wide configuration which in IPA defaults to resolving location of KDCs via DNS SRV records. Once trust is established, SSSD will populate a list of closest DCs and provide them through the KDC locator plugin. But at the time the trust is established performing DNS SRV-based discovery of Kerberos KDCs might fail due to multiple reasons. It might also succeed but point to a DC that doesn't know about the account we have to use to establish trust. One edge case is when DNS SRV record points to an unreachable DC, whether due to a firewall or a network topology limitations. In such case an administrator would pass --server <server> option to 'ipa trust-add' or 'ipa trust-fetch-domains' commands. 'ipa trust-fetch-domains' runs a helper via oddjobd. This helper was already modified to support --server option and generated custom krb5.conf overlay to pin to a specific AD DC. However, this configuration was removed as soon as we finished talking to AD DCs. With switch to always use Kebreros to authenticate in retrieval of the topology information, we have to use the overlay everywhere as well. Convert the code that generated the overlay file into a context that generates the overlay and sets environment. Reuse it in other trust-related places where this matters. Oddjob helper runs as root and can write to /run/ipa for the krb5.conf overlay. Server side of 'ipa trust-add' code calls into ipaserver/dcerpc.py and runs under ipaapi so can only write to /tmp. Since it is a part of the Apache instance, it uses private /tmp mounted on tmpfs. Fixes: https://pagure.io/freeipa/issue/8664 Related: https://pagure.io/freeipa/issue/8655 Signed-off-by: Alexander Bokovoy <abokovoy@redhat.com> Reviewed-By: Christian Heimes <cheimes@redhat.com> Reviewed-By: Rob Crittenden <rcritten@redhat.com>
1760 lines
58 KiB
Python
1760 lines
58 KiB
Python
# Authors: Simo Sorce <ssorce@redhat.com>
|
|
#
|
|
# Copyright (C) 2007-2016 Red Hat, Inc.
|
|
# see file 'COPYING' for use and warranty information
|
|
#
|
|
# This program 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.
|
|
#
|
|
# This program 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 this program. If not, see <http://www.gnu.org/licenses/>.
|
|
#
|
|
|
|
from __future__ import print_function
|
|
|
|
import codecs
|
|
import logging
|
|
import string
|
|
import tempfile
|
|
import subprocess
|
|
import random
|
|
import math
|
|
import os
|
|
import sys
|
|
import errno
|
|
import copy
|
|
import shutil
|
|
import socket
|
|
import re
|
|
import datetime
|
|
import netaddr
|
|
import time
|
|
import textwrap
|
|
import io
|
|
from contextlib import contextmanager
|
|
import locale
|
|
import collections
|
|
import urllib
|
|
|
|
import six
|
|
from six.moves import input
|
|
|
|
try:
|
|
import netifaces
|
|
except ImportError:
|
|
netifaces = None
|
|
|
|
from ipapython.dn import DN
|
|
from ipaplatform.paths import paths
|
|
from ipaplatform.constants import User, Group
|
|
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
# only for OTP password that is manually retyped by user
|
|
TMP_PWD_ENTROPY_BITS = 128
|
|
|
|
|
|
PROTOCOL_NAMES = {
|
|
socket.SOCK_STREAM: 'tcp',
|
|
socket.SOCK_DGRAM: 'udp'
|
|
}
|
|
|
|
InterfaceDetails = collections.namedtuple(
|
|
'InterfaceDetails', [
|
|
'name', # interface name
|
|
'ifnet' # network details of interface
|
|
])
|
|
|
|
|
|
class UnsafeIPAddress(netaddr.IPAddress):
|
|
"""Any valid IP address with or without netmask."""
|
|
|
|
# Use inet_pton() rather than inet_aton() for IP address parsing. We
|
|
# will use the same function in IPv4/IPv6 conversions + be stricter
|
|
# and don't allow IP addresses such as '1.1.1' in the same time
|
|
netaddr_ip_flags = netaddr.INET_PTON
|
|
|
|
def __init__(self, addr):
|
|
if isinstance(addr, UnsafeIPAddress):
|
|
self._net = addr._net
|
|
super(UnsafeIPAddress, self).__init__(addr,
|
|
flags=self.netaddr_ip_flags)
|
|
return
|
|
|
|
elif isinstance(addr, netaddr.IPAddress):
|
|
self._net = None # no information about netmask
|
|
super(UnsafeIPAddress, self).__init__(addr,
|
|
flags=self.netaddr_ip_flags)
|
|
return
|
|
|
|
elif isinstance(addr, netaddr.IPNetwork):
|
|
self._net = addr
|
|
super(UnsafeIPAddress, self).__init__(self._net.ip,
|
|
flags=self.netaddr_ip_flags)
|
|
return
|
|
|
|
# option of last resort: parse it as string
|
|
self._net = None
|
|
addr = str(addr)
|
|
try:
|
|
try:
|
|
addr = netaddr.IPAddress(addr, flags=self.netaddr_ip_flags)
|
|
except netaddr.AddrFormatError:
|
|
# netaddr.IPAddress doesn't handle zone indices in textual
|
|
# IPv6 addresses. Try removing zone index and parse the
|
|
# address again.
|
|
addr, sep, _foo = addr.partition('%')
|
|
if sep != '%':
|
|
raise
|
|
addr = netaddr.IPAddress(addr, flags=self.netaddr_ip_flags)
|
|
if addr.version != 6:
|
|
raise
|
|
except ValueError:
|
|
self._net = netaddr.IPNetwork(addr, flags=self.netaddr_ip_flags)
|
|
addr = self._net.ip
|
|
super(UnsafeIPAddress, self).__init__(addr,
|
|
flags=self.netaddr_ip_flags)
|
|
|
|
def __getstate__(self):
|
|
state = {
|
|
'_net': self._net,
|
|
'super_state': super(UnsafeIPAddress, self).__getstate__(),
|
|
}
|
|
return state
|
|
|
|
def __setstate__(self, state):
|
|
super(UnsafeIPAddress, self).__setstate__(state['super_state'])
|
|
self._net = state['_net']
|
|
|
|
|
|
class CheckedIPAddress(UnsafeIPAddress):
|
|
"""IPv4 or IPv6 address with additional constraints.
|
|
|
|
Reserved or link-local addresses are never accepted.
|
|
"""
|
|
def __init__(self, addr, parse_netmask=True,
|
|
allow_loopback=False, allow_multicast=False):
|
|
try:
|
|
super(CheckedIPAddress, self).__init__(addr)
|
|
except netaddr.core.AddrFormatError as e:
|
|
raise ValueError(e)
|
|
|
|
if isinstance(addr, CheckedIPAddress):
|
|
self.prefixlen = addr.prefixlen
|
|
return
|
|
|
|
if not parse_netmask and self._net:
|
|
raise ValueError(
|
|
"netmask and prefix length not allowed here: {}".format(addr))
|
|
|
|
if self.version not in (4, 6):
|
|
raise ValueError("unsupported IP version {}".format(self.version))
|
|
|
|
if not allow_loopback and self.is_loopback():
|
|
raise ValueError("cannot use loopback IP address {}".format(addr))
|
|
if (not self.is_loopback() and self.is_reserved()) \
|
|
or self in netaddr.ip.IPV4_6TO4:
|
|
raise ValueError(
|
|
"cannot use IANA reserved IP address {}".format(addr))
|
|
|
|
if self.is_link_local():
|
|
raise ValueError(
|
|
"cannot use link-local IP address {}".format(addr))
|
|
if not allow_multicast and self.is_multicast():
|
|
raise ValueError("cannot use multicast IP address {}".format(addr))
|
|
|
|
if self._net is None:
|
|
if self.version == 4:
|
|
self._net = netaddr.IPNetwork(
|
|
netaddr.cidr_abbrev_to_verbose(str(self)))
|
|
elif self.version == 6:
|
|
self._net = netaddr.IPNetwork(str(self) + '/64')
|
|
|
|
self.prefixlen = self._net.prefixlen
|
|
|
|
def __getstate__(self):
|
|
state = {
|
|
'prefixlen': self.prefixlen,
|
|
'super_state': super(CheckedIPAddress, self).__getstate__(),
|
|
}
|
|
return state
|
|
|
|
def __setstate__(self, state):
|
|
super(CheckedIPAddress, self).__setstate__(state['super_state'])
|
|
self.prefixlen = state['prefixlen']
|
|
|
|
def is_network_addr(self):
|
|
return self == self._net.network
|
|
|
|
def is_broadcast_addr(self):
|
|
return self.version == 4 and self == self._net.broadcast
|
|
|
|
def get_matching_interface(self):
|
|
"""Find matching local interface for address
|
|
:return: InterfaceDetails named tuple or None if no interface has
|
|
this address
|
|
"""
|
|
if netifaces is None:
|
|
raise ImportError("netifaces")
|
|
logger.debug("Searching for an interface of IP address: %s", self)
|
|
if self.version == 4:
|
|
family = netifaces.AF_INET
|
|
elif self.version == 6:
|
|
family = netifaces.AF_INET6
|
|
else:
|
|
raise ValueError(
|
|
"Unsupported address family ({})".format(self.version)
|
|
)
|
|
|
|
for interface in netifaces.interfaces():
|
|
for ifdata in netifaces.ifaddresses(interface).get(family, []):
|
|
|
|
# link-local addresses contain '%suffix' that causes parse
|
|
# errors in IPNetwork
|
|
ifaddr = ifdata['addr'].split(u'%', 1)[0]
|
|
|
|
# newer versions of netifaces provide IPv6 netmask in format
|
|
# 'ffff:ffff:ffff:ffff::/64'. We have to split and use prefix
|
|
# or the netmask with older versions
|
|
ifmask = ifdata['netmask'].split(u'/')[-1]
|
|
|
|
ifaddrmask = '{addr}/{netmask}'.format(
|
|
addr=ifaddr,
|
|
netmask=ifmask
|
|
)
|
|
logger.debug(
|
|
"Testing local IP address: %s (interface: %s)",
|
|
ifaddrmask, interface)
|
|
|
|
ifnet = netaddr.IPNetwork(ifaddrmask)
|
|
|
|
if ifnet.ip == self:
|
|
return InterfaceDetails(interface, ifnet)
|
|
return None
|
|
|
|
def set_ip_net(self, ifnet):
|
|
"""Set IP Network details for this address. IPNetwork is valid only
|
|
locally, so this should be set only for local IP addresses
|
|
|
|
:param ifnet: netaddr.IPNetwork object with information about IP
|
|
network where particula address belongs locally
|
|
"""
|
|
assert isinstance(ifnet, netaddr.IPNetwork)
|
|
self._net = ifnet
|
|
|
|
|
|
class CheckedIPAddressLoopback(CheckedIPAddress):
|
|
"""IPv4 or IPv6 address with additional constraints with
|
|
possibility to use a loopback IP.
|
|
Reserved or link-local addresses are never accepted.
|
|
"""
|
|
def __init__(self, addr, parse_netmask=True, allow_multicast=False):
|
|
|
|
super(CheckedIPAddressLoopback, self).__init__(
|
|
addr, parse_netmask=parse_netmask,
|
|
allow_multicast=allow_multicast,
|
|
allow_loopback=True)
|
|
|
|
if self.is_loopback():
|
|
# print is being used instead of a logger, because at this
|
|
# moment, in execution process, there is no logger configured
|
|
print("WARNING: You are using a loopback IP: {}".format(addr),
|
|
file=sys.stderr)
|
|
|
|
|
|
def valid_ip(addr):
|
|
return netaddr.valid_ipv4(addr) or netaddr.valid_ipv6(addr)
|
|
|
|
def format_netloc(host, port=None):
|
|
"""
|
|
Format network location (host:port).
|
|
|
|
If the host part is a literal IPv6 address, it must be enclosed in square
|
|
brackets (RFC 2732).
|
|
"""
|
|
host = str(host)
|
|
try:
|
|
socket.inet_pton(socket.AF_INET6, host)
|
|
host = '[%s]' % host
|
|
except socket.error:
|
|
pass
|
|
if port is None:
|
|
return host
|
|
else:
|
|
return '%s:%s' % (host, str(port))
|
|
|
|
def realm_to_suffix(realm_name):
|
|
'Convert a kerberos realm to a IPA suffix.'
|
|
s = realm_name.split(".")
|
|
suffix_dn = DN(*[('dc', x.lower()) for x in s])
|
|
return suffix_dn
|
|
|
|
def suffix_to_realm(suffix_dn):
|
|
'Convert a IPA suffix to a kerberos realm.'
|
|
assert isinstance(suffix_dn, DN)
|
|
realm = '.'.join([x.value for x in suffix_dn])
|
|
return realm
|
|
|
|
def template_str(txt, vars):
|
|
val = string.Template(txt).substitute(vars)
|
|
|
|
# eval() is a special string one can insert into a template to have the
|
|
# Python interpreter evaluate the string. This is intended to allow
|
|
# math to be performed in templates.
|
|
pattern = re.compile(r'(eval\s*\(([^()]*)\))')
|
|
val = pattern.sub(lambda x: str(eval(x.group(2))), val)
|
|
|
|
return val
|
|
|
|
def template_file(infilename, vars):
|
|
"""Read a file and perform template substitutions"""
|
|
with open(infilename) as f:
|
|
return template_str(f.read(), vars)
|
|
|
|
def copy_template_file(infilename, outfilename, vars):
|
|
"""Copy a file, performing template substitutions"""
|
|
txt = template_file(infilename, vars)
|
|
with open(outfilename, 'w') as file:
|
|
file.write(txt)
|
|
|
|
|
|
def write_tmp_file(txt):
|
|
fd = tempfile.NamedTemporaryFile('w+')
|
|
fd.write(txt)
|
|
fd.flush()
|
|
|
|
return fd
|
|
|
|
|
|
def flush_sync(f):
|
|
"""Flush and fsync file to disk
|
|
|
|
:param f: a file object with fileno and name
|
|
"""
|
|
# flush file buffer to file descriptor
|
|
f.flush()
|
|
# flush Kernel buffer to disk
|
|
os.fsync(f.fileno())
|
|
# sync metadata in directory
|
|
dirname = os.path.dirname(os.path.abspath(f.name))
|
|
dirfd = os.open(dirname, os.O_RDONLY | os.O_DIRECTORY)
|
|
try:
|
|
os.fsync(dirfd)
|
|
finally:
|
|
os.close(dirfd)
|
|
|
|
|
|
def shell_quote(string):
|
|
if isinstance(string, str):
|
|
return "'" + string.replace("'", "'\\''") + "'"
|
|
else:
|
|
return b"'" + string.replace(b"'", b"'\\''") + b"'"
|
|
|
|
|
|
class _RunResult(collections.namedtuple('_RunResult',
|
|
'output error_output returncode')):
|
|
"""Result of ipautil.run"""
|
|
|
|
|
|
class CalledProcessError(subprocess.CalledProcessError):
|
|
"""CalledProcessError with stderr
|
|
|
|
Hold stderr of failed call and print it in repr() to simplify debugging.
|
|
"""
|
|
def __init__(self, returncode, cmd, output=None, stderr=None):
|
|
super(CalledProcessError, self).__init__(returncode, cmd, output)
|
|
self.stderr = stderr
|
|
|
|
def __str__(self):
|
|
args = [
|
|
self.__class__.__name__, '('
|
|
'Command {!s} '.format(self.cmd),
|
|
'returned non-zero exit status {!r}'.format(self.returncode)
|
|
]
|
|
if self.stderr is not None:
|
|
args.append(': {!r}'.format(self.stderr))
|
|
args.append(')')
|
|
return ''.join(args)
|
|
|
|
__repr__ = __str__
|
|
|
|
|
|
def run(args, stdin=None, raiseonerr=True, nolog=(), env=None,
|
|
capture_output=False, skip_output=False, cwd=None,
|
|
runas=None, suplementary_groups=(),
|
|
capture_error=False, encoding=None, redirect_output=False,
|
|
umask=None, nolog_output=False, nolog_error=False):
|
|
"""
|
|
Execute an external command.
|
|
|
|
:param args: List of arguments for the command
|
|
:param stdin: Optional input to the command
|
|
:param raiseonerr: If True, raises an exception if the return code is
|
|
not zero
|
|
:param nolog: Tuple of strings that shouldn't be logged, like passwords.
|
|
Each tuple consists of a string to be replaced by XXXXXXXX.
|
|
|
|
Example:
|
|
We have a command
|
|
['/usr/bin/setpasswd', '--password', 'Secret123', 'someuser']
|
|
and we don't want to log the password so nolog would be set to:
|
|
('Secret123',)
|
|
The resulting log output would be:
|
|
|
|
/usr/bin/setpasswd --password XXXXXXXX someuser
|
|
|
|
If a value isn't found in the list it is silently ignored.
|
|
:param env: Dictionary of environment variables passed to the command.
|
|
When None, current environment is copied
|
|
:param capture_output: Capture stdout
|
|
:param skip_output: Redirect the output to /dev/null and do not log it
|
|
:param cwd: Current working directory
|
|
:param runas: Name or User object of a user that the command should be
|
|
run as. The spawned process will have both real and effective UID and
|
|
GID set.
|
|
:param suplementary_groups: List of group names or Group object that will
|
|
be used as suplementary groups for subporcess. The option runas must
|
|
be specified together with this option.
|
|
:param capture_error: Capture stderr
|
|
:param nolog_output: do not log stdout even if it is being captured
|
|
:param nolog_error: do not log stderr even if it is being captured
|
|
:param encoding: For Python 3, the encoding to use for output,
|
|
error_output, and (if it's not bytes) stdin.
|
|
If None, the current encoding according to locale is used.
|
|
:param redirect_output: Redirect (error) output to standard (error) output.
|
|
:param umask: Set file-creation mask before running the command.
|
|
|
|
:return: An object with these attributes:
|
|
|
|
`returncode`: The process' exit status
|
|
|
|
`output` and `error_output`: captured output, as strings. Under
|
|
Python 3, these are encoded with the given `encoding`.
|
|
None unless `capture_output` or `capture_error`, respectively, are
|
|
given
|
|
|
|
`raw_output`, `raw_error_output`: captured output, as bytes.
|
|
|
|
`output_log` and `error_log`: The captured output, as strings, with any
|
|
unencodable characters discarded. These should only be used
|
|
for logging or error messages.
|
|
|
|
If skip_output is given, all output-related attributes on the result
|
|
(that is, all except `returncode`) are None.
|
|
|
|
For backwards compatibility, the return value can also be used as a
|
|
(output, error_output, returncode) triple.
|
|
"""
|
|
assert isinstance(suplementary_groups, (tuple, list))
|
|
p_in = None
|
|
p_out = None
|
|
p_err = None
|
|
|
|
if isinstance(nolog, str):
|
|
# We expect a tuple (or list, or other iterable) of nolog strings.
|
|
# Passing just a single string is bad: strings are iterable, so this
|
|
# would result in every individual character of that string being
|
|
# replaced by XXXXXXXX.
|
|
# This is a sanity check to prevent that.
|
|
raise ValueError('nolog must be a tuple of strings.')
|
|
|
|
if skip_output and (capture_output or capture_error):
|
|
raise ValueError('skip_output is incompatible with '
|
|
'capture_output or capture_error')
|
|
|
|
if redirect_output and (capture_output or capture_error):
|
|
raise ValueError('redirect_output is incompatible with '
|
|
'capture_output or capture_error')
|
|
|
|
if skip_output and redirect_output:
|
|
raise ValueError('skip_output is incompatible with redirect_output')
|
|
|
|
if env is None:
|
|
# copy default env
|
|
env = copy.deepcopy(os.environ)
|
|
env["PATH"] = "/bin:/sbin:/usr/kerberos/bin:/usr/kerberos/sbin:/usr/bin:/usr/sbin"
|
|
if stdin:
|
|
p_in = subprocess.PIPE
|
|
if skip_output:
|
|
p_out = p_err = open(os.devnull, 'w')
|
|
elif redirect_output:
|
|
p_out = sys.stdout
|
|
p_err = sys.stderr
|
|
else:
|
|
p_out = subprocess.PIPE
|
|
p_err = subprocess.PIPE
|
|
|
|
if encoding is None:
|
|
encoding = locale.getpreferredencoding()
|
|
|
|
if six.PY3 and isinstance(stdin, str):
|
|
stdin = stdin.encode(encoding)
|
|
|
|
arg_string = nolog_replace(repr(args), nolog)
|
|
logger.debug('Starting external process')
|
|
logger.debug('args=%s', arg_string)
|
|
|
|
if runas is not None:
|
|
runas = User(runas)
|
|
suplementary_groups = [Group(group) for group in suplementary_groups]
|
|
suplementary_gids = [group.gid for group in suplementary_groups]
|
|
|
|
logger.debug(
|
|
'runas=%s (UID %d, GID %s)', runas, runas.uid, runas.pgid
|
|
)
|
|
if suplementary_groups:
|
|
for group in suplementary_groups:
|
|
logger.debug(
|
|
'supplementary_group=%s (GID %d)', group, group.gid
|
|
)
|
|
|
|
if runas is not None or umask is not None:
|
|
# preexec function is not supported in WSGI environment
|
|
def preexec_fn():
|
|
if runas is not None:
|
|
os.setgroups(suplementary_gids)
|
|
os.setregid(runas.pgid, runas.pgid)
|
|
os.setreuid(runas.uid, runas.uid)
|
|
|
|
if umask is not None:
|
|
os.umask(umask)
|
|
else:
|
|
preexec_fn = None
|
|
|
|
try:
|
|
# pylint: disable=subprocess-popen-preexec-fn
|
|
p = subprocess.Popen(args, stdin=p_in, stdout=p_out, stderr=p_err,
|
|
close_fds=True, env=env, cwd=cwd,
|
|
preexec_fn=preexec_fn)
|
|
stdout, stderr = p.communicate(stdin)
|
|
except KeyboardInterrupt:
|
|
logger.debug('Process interrupted')
|
|
p.wait()
|
|
raise
|
|
except BaseException:
|
|
logger.debug('Process execution failed')
|
|
raise
|
|
finally:
|
|
if skip_output:
|
|
p_out.close() # pylint: disable=E1103
|
|
|
|
logger.debug('Process finished, return code=%s', p.returncode)
|
|
|
|
# The command and its output may include passwords that we don't want
|
|
# to log. Replace those.
|
|
if skip_output or redirect_output:
|
|
output_log = None
|
|
error_log = None
|
|
else:
|
|
if six.PY3:
|
|
output_log = stdout.decode(locale.getpreferredencoding(),
|
|
errors='replace')
|
|
else:
|
|
output_log = stdout
|
|
|
|
if six.PY3:
|
|
error_log = stderr.decode(locale.getpreferredencoding(),
|
|
errors='replace')
|
|
else:
|
|
error_log = stderr
|
|
|
|
output_log = nolog_replace(output_log, nolog)
|
|
if nolog_output:
|
|
logger.debug('stdout=<REDACTED>')
|
|
else:
|
|
logger.debug('stdout=%s', output_log)
|
|
|
|
error_log = nolog_replace(error_log, nolog)
|
|
if nolog_error:
|
|
logger.debug('stderr=<REDACTED>')
|
|
else:
|
|
logger.debug('stderr=%s', error_log)
|
|
|
|
if capture_output:
|
|
if six.PY2:
|
|
output = stdout
|
|
else:
|
|
output = stdout.decode(encoding)
|
|
else:
|
|
output = None
|
|
|
|
if capture_error:
|
|
if six.PY2:
|
|
error_output = stderr
|
|
else:
|
|
error_output = stderr.decode(encoding)
|
|
else:
|
|
error_output = None
|
|
|
|
if p.returncode != 0 and raiseonerr:
|
|
raise CalledProcessError(
|
|
p.returncode, arg_string, output_log, error_log
|
|
)
|
|
|
|
result = _RunResult(output, error_output, p.returncode)
|
|
result.raw_output = stdout
|
|
result.raw_error_output = stderr
|
|
result.output_log = output_log
|
|
result.error_log = error_log
|
|
return result
|
|
|
|
|
|
def nolog_replace(string, nolog):
|
|
"""Replace occurences of strings given in `nolog` with XXXXXXXX"""
|
|
for value in nolog:
|
|
if not value or not isinstance(value, str):
|
|
continue
|
|
|
|
quoted = urllib.parse.quote(value)
|
|
shquoted = shell_quote(value)
|
|
for nolog_value in (shquoted, value, quoted):
|
|
string = string.replace(nolog_value, 'XXXXXXXX')
|
|
return string
|
|
|
|
|
|
def install_file(fname, dest):
|
|
# SELinux: use copy to keep the right context
|
|
if os.path.isfile(dest):
|
|
os.rename(dest, dest + ".orig")
|
|
shutil.copy(fname, dest)
|
|
os.remove(fname)
|
|
|
|
|
|
def backup_file(fname):
|
|
if os.path.isfile(fname):
|
|
os.rename(fname, fname + ".orig")
|
|
|
|
|
|
class CIDict(dict):
|
|
"""
|
|
Case-insensitive but case-respecting dictionary.
|
|
|
|
This code is derived from python-ldap's cidict.py module,
|
|
written by stroeder: http://python-ldap.sourceforge.net/
|
|
|
|
This version extends 'dict' so it works properly with TurboGears.
|
|
If you extend UserDict, isinstance(foo, dict) returns false.
|
|
"""
|
|
|
|
def __init__(self, default=None, **kwargs):
|
|
super(CIDict, self).__init__()
|
|
self._keys = {} # mapping of lowercased keys to proper case
|
|
if default:
|
|
self.update(default)
|
|
if kwargs:
|
|
self.update(kwargs)
|
|
|
|
def __getitem__(self, key):
|
|
return super(CIDict, self).__getitem__(key.lower())
|
|
|
|
def __setitem__(self, key, value, seen_keys=None):
|
|
"""cidict[key] = value
|
|
|
|
The ``seen_keys`` argument is used by ``update()`` to keep track of
|
|
duplicate keys. It should be an initially empty set that is
|
|
passed to all calls to __setitem__ that should not set duplicate keys.
|
|
"""
|
|
lower_key = key.lower()
|
|
if seen_keys is not None:
|
|
if lower_key in seen_keys:
|
|
raise ValueError('Duplicate key in update: %s' % key)
|
|
seen_keys.add(lower_key)
|
|
self._keys[lower_key] = key
|
|
return super(CIDict, self).__setitem__(lower_key, value)
|
|
|
|
def __delitem__(self, key):
|
|
lower_key = key.lower()
|
|
del self._keys[lower_key]
|
|
return super(CIDict, self).__delitem__(lower_key)
|
|
|
|
def update(self, new=None, **kwargs):
|
|
"""Update self from dict/iterable new and kwargs
|
|
|
|
Functions like ``dict.update()``.
|
|
|
|
Neither ``new`` nor ``kwargs`` may contain two keys that only differ in
|
|
case, as this situation would result in loss of data.
|
|
"""
|
|
seen = set()
|
|
if new:
|
|
try:
|
|
keys = new.keys
|
|
except AttributeError:
|
|
self.update(dict(new))
|
|
else:
|
|
for key in keys():
|
|
self.__setitem__(key, new[key], seen)
|
|
seen = set()
|
|
for key, value in kwargs.items():
|
|
self.__setitem__(key, value, seen)
|
|
|
|
def __contains__(self, key):
|
|
return super(CIDict, self).__contains__(key.lower())
|
|
|
|
if six.PY2:
|
|
def has_key(self, key):
|
|
# pylint: disable=no-member
|
|
return super(CIDict, self).has_key(key.lower()) # noqa
|
|
# pylint: enable=no-member
|
|
|
|
def get(self, key, failobj=None):
|
|
try:
|
|
return self[key]
|
|
except KeyError:
|
|
return failobj
|
|
|
|
def __iter__(self):
|
|
return six.itervalues(self._keys)
|
|
|
|
def keys(self):
|
|
if six.PY2:
|
|
return list(self.iterkeys())
|
|
else:
|
|
return self.iterkeys()
|
|
|
|
def items(self):
|
|
if six.PY2:
|
|
return list(self.iteritems())
|
|
else:
|
|
return self.iteritems()
|
|
|
|
def values(self):
|
|
if six.PY2:
|
|
return list(self.itervalues())
|
|
else:
|
|
return self.itervalues()
|
|
|
|
def copy(self):
|
|
"""Returns a shallow copy of this CIDict"""
|
|
return CIDict(list(self.items()))
|
|
|
|
def iteritems(self):
|
|
return ((k, self[k]) for k in six.itervalues(self._keys))
|
|
|
|
def iterkeys(self):
|
|
return six.itervalues(self._keys)
|
|
|
|
def itervalues(self):
|
|
return (v for k, v in six.iteritems(self))
|
|
|
|
def setdefault(self, key, value=None):
|
|
try:
|
|
return self[key]
|
|
except KeyError:
|
|
self[key] = value
|
|
return value
|
|
|
|
def pop(self, key, *args):
|
|
try:
|
|
value = self[key]
|
|
del self[key]
|
|
return value
|
|
except KeyError:
|
|
if len(args) == 1:
|
|
return args[0]
|
|
raise
|
|
|
|
def popitem(self):
|
|
(lower_key, value) = super(CIDict, self).popitem()
|
|
key = self._keys[lower_key]
|
|
del self._keys[lower_key]
|
|
|
|
return (key, value)
|
|
|
|
def clear(self):
|
|
self._keys.clear()
|
|
return super(CIDict, self).clear()
|
|
|
|
def viewitems(self):
|
|
raise NotImplementedError('CIDict.viewitems is not implemented')
|
|
|
|
def viewkeys(self):
|
|
raise NotImplementedError('CIDict.viewkeys is not implemented')
|
|
|
|
def viewvvalues(self):
|
|
raise NotImplementedError('CIDict.viewvvalues is not implemented')
|
|
|
|
|
|
class GeneralizedTimeZone(datetime.tzinfo):
|
|
"""This class is a basic timezone wrapper for the offset specified
|
|
in a Generalized Time. It is dst-ignorant."""
|
|
def __init__(self,offsetstr="Z"):
|
|
super(GeneralizedTimeZone, self).__init__()
|
|
|
|
self.name = offsetstr
|
|
self.houroffset = 0
|
|
self.minoffset = 0
|
|
|
|
if offsetstr == "Z":
|
|
self.houroffset = 0
|
|
self.minoffset = 0
|
|
else:
|
|
if (len(offsetstr) >= 3) and re.match(r'[-+]\d\d', offsetstr):
|
|
self.houroffset = int(offsetstr[0:3])
|
|
offsetstr = offsetstr[3:]
|
|
if (len(offsetstr) >= 2) and re.match(r'\d\d', offsetstr):
|
|
self.minoffset = int(offsetstr[0:2])
|
|
offsetstr = offsetstr[2:]
|
|
if len(offsetstr) > 0:
|
|
raise ValueError()
|
|
if self.houroffset < 0:
|
|
self.minoffset *= -1
|
|
|
|
def utcoffset(self, dt):
|
|
return datetime.timedelta(hours=self.houroffset, minutes=self.minoffset)
|
|
|
|
def dst(self):
|
|
return datetime.timedelta(0)
|
|
|
|
def tzname(self):
|
|
return self.name
|
|
|
|
|
|
def parse_generalized_time(timestr):
|
|
"""Parses are Generalized Time string (as specified in X.680),
|
|
returning a datetime object. Generalized Times are stored inside
|
|
the krbPasswordExpiration attribute in LDAP.
|
|
|
|
This method doesn't attempt to be perfect wrt timezones. If python
|
|
can't be bothered to implement them, how can we..."""
|
|
|
|
if len(timestr) < 8:
|
|
return None
|
|
try:
|
|
date = timestr[:8]
|
|
time = timestr[8:]
|
|
|
|
year = int(date[:4])
|
|
month = int(date[4:6])
|
|
day = int(date[6:8])
|
|
|
|
hour = min = sec = msec = 0
|
|
tzone = None
|
|
|
|
if (len(time) >= 2) and re.match(r'\d', time[0]):
|
|
hour = int(time[:2])
|
|
time = time[2:]
|
|
if len(time) >= 2 and (time[0] == "," or time[0] == "."):
|
|
hour_fraction = "."
|
|
time = time[1:]
|
|
while (len(time) > 0) and re.match(r'\d', time[0]):
|
|
hour_fraction += time[0]
|
|
time = time[1:]
|
|
total_secs = int(float(hour_fraction) * 3600)
|
|
min, sec = divmod(total_secs, 60)
|
|
|
|
if (len(time) >= 2) and re.match(r'\d', time[0]):
|
|
min = int(time[:2])
|
|
time = time[2:]
|
|
if len(time) >= 2 and (time[0] == "," or time[0] == "."):
|
|
min_fraction = "."
|
|
time = time[1:]
|
|
while (len(time) > 0) and re.match(r'\d', time[0]):
|
|
min_fraction += time[0]
|
|
time = time[1:]
|
|
sec = int(float(min_fraction) * 60)
|
|
|
|
if (len(time) >= 2) and re.match(r'\d', time[0]):
|
|
sec = int(time[:2])
|
|
time = time[2:]
|
|
if len(time) >= 2 and (time[0] == "," or time[0] == "."):
|
|
sec_fraction = "."
|
|
time = time[1:]
|
|
while (len(time) > 0) and re.match(r'\d', time[0]):
|
|
sec_fraction += time[0]
|
|
time = time[1:]
|
|
msec = int(float(sec_fraction) * 1000000)
|
|
|
|
if (len(time) > 0):
|
|
tzone = GeneralizedTimeZone(time)
|
|
|
|
return datetime.datetime(year, month, day, hour, min, sec, msec, tzone)
|
|
|
|
except ValueError:
|
|
return None
|
|
|
|
|
|
def ipa_generate_password(entropy_bits=256, uppercase=1, lowercase=1, digits=1,
|
|
special=1, min_len=0):
|
|
"""
|
|
Generate token containing at least `entropy_bits` bits and with the given
|
|
character restraints.
|
|
|
|
:param entropy_bits:
|
|
The minimal number of entropy bits attacker has to guess:
|
|
128 bits entropy: secure
|
|
256 bits of entropy: secure enough if you care about quantum
|
|
computers
|
|
|
|
Integer values specify minimal number of characters from given
|
|
character class and length.
|
|
Value None prevents given character from appearing in the token.
|
|
|
|
Example:
|
|
TokenGenerator(uppercase=3, lowercase=3, digits=0, special=None)
|
|
|
|
At least 3 upper and 3 lower case ASCII chars, may contain digits,
|
|
no special chars.
|
|
"""
|
|
special_chars = '!$%&()*+,-./:;<>?@[]^_{|}~'
|
|
pwd_charsets = {
|
|
'uppercase': {
|
|
'chars': string.ascii_uppercase,
|
|
'entropy': math.log(len(string.ascii_uppercase), 2)
|
|
},
|
|
'lowercase': {
|
|
'chars': string.ascii_lowercase,
|
|
'entropy': math.log(len(string.ascii_lowercase), 2)
|
|
},
|
|
'digits': {
|
|
'chars': string.digits,
|
|
'entropy': math.log(len(string.digits), 2)
|
|
},
|
|
'special': {
|
|
'chars': special_chars,
|
|
'entropy': math.log(len(special_chars), 2)
|
|
},
|
|
}
|
|
req_classes = dict(
|
|
uppercase=uppercase,
|
|
lowercase=lowercase,
|
|
digits=digits,
|
|
special=special
|
|
)
|
|
# 'all' class is used when adding entropy to too-short tokens
|
|
# it contains characters from all allowed classes
|
|
pwd_charsets['all'] = {
|
|
'chars': ''.join([
|
|
charclass['chars'] for charclass_name, charclass
|
|
in pwd_charsets.items()
|
|
if req_classes[charclass_name] is not None
|
|
])
|
|
}
|
|
pwd_charsets['all']['entropy'] = math.log(
|
|
len(pwd_charsets['all']['chars']), 2)
|
|
rnd = random.SystemRandom()
|
|
|
|
todo_entropy = entropy_bits
|
|
password = u''
|
|
# Generate required character classes:
|
|
# The order of generated characters is fixed to comply with check in
|
|
# NSS function sftk_newPinCheck() in nss/lib/softoken/fipstokn.c.
|
|
for charclass_name in ['digits', 'uppercase', 'lowercase', 'special']:
|
|
charclass = pwd_charsets[charclass_name]
|
|
todo_characters = req_classes[charclass_name]
|
|
if todo_characters is None:
|
|
continue
|
|
while todo_characters > 0:
|
|
password += rnd.choice(charclass['chars'])
|
|
todo_entropy -= charclass['entropy']
|
|
todo_characters -= 1
|
|
|
|
# required character classes do not provide sufficient entropy
|
|
# or does not fulfill minimal length constraint
|
|
allchars = pwd_charsets['all']
|
|
while todo_entropy > 0 or len(password) < min_len:
|
|
password += rnd.choice(allchars['chars'])
|
|
todo_entropy -= allchars['entropy']
|
|
|
|
return password
|
|
|
|
|
|
def user_input(prompt, default = None, allow_empty = True):
|
|
if default is None:
|
|
while True:
|
|
try:
|
|
ret = input("%s: " % prompt)
|
|
if allow_empty or ret.strip():
|
|
return ret.strip()
|
|
except EOFError:
|
|
if allow_empty:
|
|
return ''
|
|
raise RuntimeError("Failed to get user input")
|
|
|
|
if isinstance(default, str):
|
|
while True:
|
|
try:
|
|
ret = input("%s [%s]: " % (prompt, default))
|
|
if not ret and (allow_empty or default):
|
|
return default
|
|
elif ret.strip():
|
|
return ret.strip()
|
|
except EOFError:
|
|
return default
|
|
|
|
if isinstance(default, bool):
|
|
choice = "yes" if default else "no"
|
|
while True:
|
|
try:
|
|
ret = input("%s [%s]: " % (prompt, choice))
|
|
ret = ret.strip()
|
|
if not ret:
|
|
return default
|
|
elif ret.lower()[0] == "y":
|
|
return True
|
|
elif ret.lower()[0] == "n":
|
|
return False
|
|
except EOFError:
|
|
return default
|
|
|
|
if isinstance(default, int):
|
|
while True:
|
|
try:
|
|
ret = input("%s [%s]: " % (prompt, default))
|
|
ret = ret.strip()
|
|
if not ret:
|
|
return default
|
|
ret = int(ret)
|
|
except ValueError:
|
|
pass
|
|
except EOFError:
|
|
return default
|
|
else:
|
|
return ret
|
|
|
|
return None
|
|
|
|
|
|
def host_port_open(host, port, socket_type=socket.SOCK_STREAM,
|
|
socket_timeout=None, log_errors=False,
|
|
log_level=logging.DEBUG):
|
|
"""
|
|
host: either hostname or IP address;
|
|
if hostname is provided, port MUST be open on ALL resolved IPs
|
|
|
|
returns True is port is open, False otherwise
|
|
"""
|
|
port_open = True
|
|
|
|
# port has to be open on ALL resolved IPs
|
|
for res in socket.getaddrinfo(host, port, socket.AF_UNSPEC, socket_type):
|
|
af, socktype, proto, _canonname, sa = res
|
|
s = None
|
|
try:
|
|
s = socket.socket(af, socktype, proto)
|
|
|
|
if socket_timeout is not None:
|
|
s.settimeout(socket_timeout)
|
|
|
|
s.connect(sa)
|
|
|
|
if socket_type == socket.SOCK_DGRAM:
|
|
s.send(b'')
|
|
s.recv(512)
|
|
except socket.error:
|
|
port_open = False
|
|
if log_errors:
|
|
msg = ('Failed to connect to port %(port)s %(proto)s on '
|
|
'%(addr)s' % dict(port=port,
|
|
proto=PROTOCOL_NAMES[socket_type],
|
|
addr=sa[0]))
|
|
logger.log(log_level, msg)
|
|
finally:
|
|
if s is not None:
|
|
s.close()
|
|
|
|
return port_open
|
|
|
|
|
|
def check_port_bindable(port, socket_type=socket.SOCK_STREAM):
|
|
"""Check if a port is free and not bound by any other application
|
|
|
|
:param port: port number
|
|
:param socket_type: type (SOCK_STREAM for TCP, SOCK_DGRAM for UDP)
|
|
|
|
Returns True if the port is free, False otherwise
|
|
"""
|
|
if socket_type == socket.SOCK_STREAM:
|
|
proto = 'TCP'
|
|
elif socket_type == socket.SOCK_DGRAM:
|
|
proto = 'UDP'
|
|
else:
|
|
raise ValueError(socket_type)
|
|
|
|
# Detect dual stack or IPv4 single stack
|
|
try:
|
|
s = socket.socket(socket.AF_INET6, socket_type)
|
|
anyaddr = '::'
|
|
logger.debug(
|
|
"check_port_bindable: Checking IPv4/IPv6 dual stack and %s",
|
|
proto
|
|
)
|
|
except socket.error:
|
|
s = socket.socket(socket.AF_INET, socket_type)
|
|
anyaddr = ''
|
|
logger.debug("check_port_bindable: Checking IPv4 only and %s", proto)
|
|
|
|
# Attempt to bind
|
|
try:
|
|
if socket_type == socket.SOCK_STREAM:
|
|
# reuse TCP sockets in TIME_WAIT state
|
|
s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, True)
|
|
s.bind((anyaddr, port))
|
|
except socket.error as e:
|
|
logger.debug(
|
|
"check_port_bindable: failed to bind to port %i/%s: %s",
|
|
port, proto, e
|
|
)
|
|
return False
|
|
else:
|
|
logger.debug(
|
|
"check_port_bindable: bind success: %i/%s", port, proto
|
|
)
|
|
return True
|
|
finally:
|
|
s.close()
|
|
|
|
|
|
def config_replace_variables(filepath, replacevars=dict(), appendvars=dict(),
|
|
removevars=None):
|
|
"""
|
|
Take a key=value based configuration file, and write new version
|
|
with certain values replaced, appended, or removed.
|
|
|
|
All (key,value) pairs from replacevars and appendvars that were not found
|
|
in the configuration file, will be added there.
|
|
|
|
All entries in set removevars are removed.
|
|
|
|
It is responsibility of a caller to ensure that replacevars and
|
|
appendvars do not overlap.
|
|
|
|
It is responsibility of a caller to back up file.
|
|
|
|
returns dictionary of affected keys and their previous values
|
|
|
|
One have to run restore_context(filepath) afterwards or
|
|
security context of the file will not be correct after modification
|
|
"""
|
|
pattern = re.compile(r'''
|
|
(^
|
|
\s*
|
|
(?P<option> [^\#;]+?)
|
|
(\s*=\s*)
|
|
(?P<value> .+?)?
|
|
(\s*((\#|;).*)?)?
|
|
$)''', re.VERBOSE)
|
|
orig_stat = os.stat(filepath)
|
|
old_values = dict()
|
|
temp_filename = None
|
|
with tempfile.NamedTemporaryFile(mode="w", delete=False) as new_config:
|
|
temp_filename = new_config.name
|
|
with open(filepath, 'r') as f:
|
|
for line in f:
|
|
new_line = line
|
|
m = pattern.match(line)
|
|
if m:
|
|
option, value = m.group('option', 'value')
|
|
if option is not None:
|
|
if replacevars and option in replacevars:
|
|
# replace value completely
|
|
new_line = u"%s=%s\n" % (option, replacevars[option])
|
|
old_values[option] = value
|
|
if appendvars and option in appendvars:
|
|
# append new value unless it is already existing in the original one
|
|
if not value:
|
|
new_line = u"%s=%s\n" % (option, appendvars[option])
|
|
elif value.find(appendvars[option]) == -1:
|
|
new_line = u"%s=%s %s\n" % (option, value, appendvars[option])
|
|
old_values[option] = value
|
|
if removevars and option in removevars:
|
|
old_values[option] = value
|
|
new_line = None
|
|
if new_line is not None:
|
|
new_config.write(new_line)
|
|
# Now add all options from replacevars and appendvars that were not found in the file
|
|
new_vars = replacevars.copy()
|
|
new_vars.update(appendvars)
|
|
newvars_view = set(new_vars.keys()) - set(old_values.keys())
|
|
append_view = (set(appendvars.keys()) - newvars_view)
|
|
for item in newvars_view:
|
|
new_config.write("%s=%s\n" % (item,new_vars[item]))
|
|
for item in append_view:
|
|
new_config.write("%s=%s\n" % (item,appendvars[item]))
|
|
new_config.flush()
|
|
# Make sure the resulting file is readable by others before installing it
|
|
os.fchmod(new_config.fileno(), orig_stat.st_mode)
|
|
os.fchown(new_config.fileno(), orig_stat.st_uid, orig_stat.st_gid)
|
|
|
|
# At this point new_config is closed but not removed due to 'delete=False' above
|
|
# Now, install the temporary file as configuration and ensure old version is available as .orig
|
|
# While .orig file is not used during uninstall, it is left there for administrator.
|
|
install_file(temp_filename, filepath)
|
|
|
|
return old_values
|
|
|
|
def inifile_replace_variables(filepath, section, replacevars=dict(), appendvars=dict()):
|
|
"""
|
|
Take a section-structured key=value based configuration file, and write new version
|
|
with certain values replaced or appended within the section
|
|
|
|
All (key,value) pairs from replacevars and appendvars that were not found
|
|
in the configuration file, will be added there.
|
|
|
|
It is responsibility of a caller to ensure that replacevars and
|
|
appendvars do not overlap.
|
|
|
|
It is responsibility of a caller to back up file.
|
|
|
|
returns dictionary of affected keys and their previous values
|
|
|
|
One have to run restore_context(filepath) afterwards or
|
|
security context of the file will not be correct after modification
|
|
"""
|
|
pattern = re.compile(r'''
|
|
(^
|
|
\[
|
|
(?P<section> .+) \]
|
|
(\s+((\#|;).*)?)?
|
|
$)|(^
|
|
\s*
|
|
(?P<option> [^\#;]+?)
|
|
(\s*=\s*)
|
|
(?P<value> .+?)?
|
|
(\s*((\#|;).*)?)?
|
|
$)''', re.VERBOSE)
|
|
def add_options(config, replacevars, appendvars, oldvars):
|
|
# add all options from replacevars and appendvars that were not found in the file
|
|
new_vars = replacevars.copy()
|
|
new_vars.update(appendvars)
|
|
newvars_view = set(new_vars.keys()) - set(oldvars.keys())
|
|
append_view = (set(appendvars.keys()) - newvars_view)
|
|
for item in newvars_view:
|
|
config.write("%s=%s\n" % (item,new_vars[item]))
|
|
for item in append_view:
|
|
config.write("%s=%s\n" % (item,appendvars[item]))
|
|
|
|
orig_stat = os.stat(filepath)
|
|
old_values = dict()
|
|
temp_filename = None
|
|
with tempfile.NamedTemporaryFile(mode='w', delete=False) as new_config:
|
|
temp_filename = new_config.name
|
|
with open(filepath, 'r') as f:
|
|
in_section = False
|
|
finished = False
|
|
line_idx = 1
|
|
for line in f:
|
|
line_idx = line_idx + 1
|
|
new_line = line
|
|
m = pattern.match(line)
|
|
if m:
|
|
sect, option, value = m.group('section', 'option', 'value')
|
|
if in_section and sect is not None:
|
|
# End of the searched section, add remaining options
|
|
add_options(new_config, replacevars, appendvars, old_values)
|
|
finished = True
|
|
if sect is not None:
|
|
# New section is found, check whether it is the one we are looking for
|
|
in_section = (str(sect).lower() == str(section).lower())
|
|
if option is not None and in_section:
|
|
# Great, this is an option from the section we are loking for
|
|
if replacevars and option in replacevars:
|
|
# replace value completely
|
|
new_line = u"%s=%s\n" % (option, replacevars[option])
|
|
old_values[option] = value
|
|
if appendvars and option in appendvars:
|
|
# append a new value unless it is already existing in the original one
|
|
if not value:
|
|
new_line = u"%s=%s\n" % (option, appendvars[option])
|
|
elif value.find(appendvars[option]) == -1:
|
|
new_line = u"%s=%s %s\n" % (option, value, appendvars[option])
|
|
old_values[option] = value
|
|
new_config.write(new_line)
|
|
# We have finished parsing the original file.
|
|
# There are two remaining cases:
|
|
# 1. Section we were looking for was not found, we need to add it.
|
|
if not (in_section or finished):
|
|
new_config.write("[%s]\n" % (section))
|
|
# 2. The section is the last one but some options were not found, add them.
|
|
if in_section or not finished:
|
|
add_options(new_config, replacevars, appendvars, old_values)
|
|
|
|
new_config.flush()
|
|
# Make sure the resulting file is readable by others before installing it
|
|
os.fchmod(new_config.fileno(), orig_stat.st_mode)
|
|
os.fchown(new_config.fileno(), orig_stat.st_uid, orig_stat.st_gid)
|
|
|
|
# At this point new_config is closed but not removed due to 'delete=False' above
|
|
# Now, install the temporary file as configuration and ensure old version is available as .orig
|
|
# While .orig file is not used during uninstall, it is left there for administrator.
|
|
install_file(temp_filename, filepath)
|
|
|
|
return old_values
|
|
|
|
def backup_config_and_replace_variables(
|
|
fstore, filepath, replacevars=dict(), appendvars=dict()):
|
|
"""
|
|
Take a key=value based configuration file, back up it, and
|
|
write new version with certain values replaced or appended
|
|
|
|
All (key,value) pairs from replacevars and appendvars that
|
|
were not found in the configuration file, will be added there.
|
|
The file must exist before this function is called.
|
|
|
|
It is responsibility of a caller to ensure that replacevars and
|
|
appendvars do not overlap.
|
|
|
|
returns dictionary of affected keys and their previous values
|
|
|
|
One have to run restore_context(filepath) afterwards or
|
|
security context of the file will not be correct after modification
|
|
"""
|
|
# Backup original filepath
|
|
fstore.backup_file(filepath)
|
|
old_values = config_replace_variables(filepath, replacevars, appendvars)
|
|
|
|
return old_values
|
|
|
|
|
|
def wait_for_open_ports(host, ports, timeout=0):
|
|
"""
|
|
Wait until the specified port(s) on the remote host are open. Timeout
|
|
in seconds may be specified to limit the wait. If the timeout is
|
|
exceeded, socket.timeout exception is raised.
|
|
"""
|
|
timeout = float(timeout)
|
|
if not isinstance(ports, (tuple, list)):
|
|
ports = [ports]
|
|
|
|
logger.debug('wait_for_open_ports: %s %s timeout %d', host, ports, timeout)
|
|
op_timeout = time.time() + timeout
|
|
|
|
for port in ports:
|
|
logger.debug('waiting for port: %s', port)
|
|
log_error = True
|
|
while True:
|
|
port_open = host_port_open(host, port, log_errors=log_error)
|
|
log_error = False # Log only first err so that the log is readable
|
|
|
|
if port_open:
|
|
logger.debug('SUCCESS: port: %s', port)
|
|
break
|
|
if timeout and time.time() > op_timeout: # timeout exceeded
|
|
raise socket.timeout("Timeout exceeded")
|
|
time.sleep(1)
|
|
|
|
|
|
def wait_for_open_socket(socket_name, timeout=0):
|
|
"""
|
|
Wait until the specified socket on the local host is open. Timeout
|
|
in seconds may be specified to limit the wait.
|
|
"""
|
|
timeout = float(timeout)
|
|
op_timeout = time.time() + timeout
|
|
|
|
while True:
|
|
try:
|
|
s = socket.socket(socket.AF_UNIX)
|
|
s.connect(socket_name)
|
|
s.close()
|
|
break
|
|
except socket.error as e:
|
|
if e.errno in (2,111): # 111: Connection refused, 2: File not found
|
|
if timeout and time.time() > op_timeout: # timeout exceeded
|
|
raise e
|
|
time.sleep(1)
|
|
else:
|
|
raise e
|
|
|
|
|
|
def dn_attribute_property(private_name):
|
|
'''
|
|
Create a property for a dn attribute which assures the attribute
|
|
is a DN or None. If the value is not None the setter converts it to
|
|
a DN. The getter assures it's either None or a DN instance.
|
|
|
|
The private_name parameter is the class internal attribute the property
|
|
shadows.
|
|
|
|
For example if a class has an attribute called base_dn, then:
|
|
|
|
base_dn = dn_attribute_property('_base_dn')
|
|
|
|
Thus the class with have an attriubte called base_dn which can only
|
|
ever be None or a DN instance. The actual value is stored in _base_dn.
|
|
'''
|
|
|
|
def setter(self, value):
|
|
if value is not None:
|
|
value = DN(value)
|
|
setattr(self, private_name, value)
|
|
|
|
def getter(self):
|
|
value = getattr(self, private_name)
|
|
if value is not None:
|
|
assert isinstance(value, DN)
|
|
return value
|
|
|
|
return property(getter, setter)
|
|
|
|
def posixify(string):
|
|
"""
|
|
Convert a string to a more strict alpha-numeric representation.
|
|
|
|
- Alpha-numeric, underscore, dot and dash characters are accepted
|
|
- Space is converted to underscore
|
|
- Other characters are omitted
|
|
- Leading dash is stripped
|
|
|
|
Note: This mapping is not one-to-one and may map different input to the
|
|
same result. When using posixify, make sure the you do not map two different
|
|
entities to one unintentionally.
|
|
"""
|
|
|
|
def valid_char(char):
|
|
return char.isalnum() or char in ('_', '.', '-')
|
|
|
|
# First replace space characters
|
|
replaced = string.replace(' ','_')
|
|
omitted = ''.join(filter(valid_char, replaced))
|
|
|
|
# Leading dash is not allowed
|
|
return omitted.lstrip('-')
|
|
|
|
@contextmanager
|
|
def private_ccache(path=None):
|
|
|
|
if path is None:
|
|
dir_path = tempfile.mkdtemp(prefix='krbcc')
|
|
path = os.path.join(dir_path, 'ccache')
|
|
else:
|
|
dir_path = None
|
|
|
|
original_value = os.environ.get('KRB5CCNAME', None)
|
|
|
|
os.environ['KRB5CCNAME'] = path
|
|
|
|
try:
|
|
yield path
|
|
finally:
|
|
if original_value is not None:
|
|
os.environ['KRB5CCNAME'] = original_value
|
|
else:
|
|
os.environ.pop('KRB5CCNAME', None)
|
|
|
|
if os.path.exists(path):
|
|
os.remove(path)
|
|
if dir_path is not None:
|
|
try:
|
|
os.rmdir(dir_path)
|
|
except OSError:
|
|
pass
|
|
|
|
|
|
@contextmanager
|
|
def private_krb5_config(realm, server, dir="/run/ipa"):
|
|
"""Generate override krb5 config file for a trusted domain DC access
|
|
Provide a context where environment variable KRB5_CONFIG is set
|
|
with the overlay on top of paths.KRB5_CONF. Overlay's file path
|
|
is passed to the context in case it is needed for something else
|
|
|
|
:param realm: realm of the trusted AD domain
|
|
:param server: server to override KDC to
|
|
:param dir: path where to create a temporary krb5.conf overlay
|
|
"""
|
|
cfg = paths.KRB5_CONF
|
|
tcfg = None
|
|
if server:
|
|
content = textwrap.dedent(u"""
|
|
[realms]
|
|
%s = {
|
|
kdc = %s
|
|
}
|
|
""") % (
|
|
realm.upper(),
|
|
server,
|
|
)
|
|
|
|
(fd, tcfg) = tempfile.mkstemp(dir=dir, prefix="krb5conf", text=True)
|
|
|
|
with io.open(fd, mode='w', encoding='utf-8') as o:
|
|
o.write(content)
|
|
cfg = ":".join([tcfg, cfg])
|
|
|
|
original_value = os.environ.get('KRB5_CONFIG', None)
|
|
|
|
os.environ['KRB5_CONFIG'] = cfg
|
|
|
|
try:
|
|
yield tcfg
|
|
except GeneratorExit:
|
|
pass
|
|
finally:
|
|
if original_value is not None:
|
|
os.environ['KRB5_CONFIG'] = original_value
|
|
else:
|
|
os.environ.pop('KRB5_CONFIG', None)
|
|
|
|
if tcfg is not None and os.path.exists(tcfg):
|
|
os.remove(tcfg)
|
|
|
|
|
|
if six.PY2:
|
|
def fsdecode(value):
|
|
"""
|
|
Decode argument using the file system encoding, as returned by
|
|
`sys.getfilesystemencoding()`.
|
|
"""
|
|
if isinstance(value, bytes):
|
|
return value.decode(sys.getfilesystemencoding())
|
|
elif isinstance(value, str):
|
|
return value
|
|
else:
|
|
raise TypeError("expect {0} or {1}, not {2}".format(
|
|
bytes.__name__,
|
|
str.__name__,
|
|
type(value).__name__))
|
|
else:
|
|
fsdecode = os.fsdecode #pylint: disable=no-member
|
|
|
|
|
|
def unescape_seq(seq, *args):
|
|
"""
|
|
unescape (remove '\\') all occurences of sequence in input strings.
|
|
|
|
:param seq: sequence to unescape
|
|
:param args: input string to process
|
|
|
|
:returns: tuple of strings with unescaped sequences
|
|
"""
|
|
unescape_re = re.compile(r'\\{}'.format(seq))
|
|
|
|
return tuple(re.sub(unescape_re, seq, a) for a in args)
|
|
|
|
|
|
def escape_seq(seq, *args):
|
|
"""
|
|
escape (prepend '\\') all occurences of sequence in input strings
|
|
|
|
:param seq: sequence to escape
|
|
:param args: input string to process
|
|
|
|
:returns: tuple of strings with escaped sequences
|
|
"""
|
|
|
|
return tuple(a.replace(seq, u'\\{}'.format(seq)) for a in args)
|
|
|
|
|
|
def decode_json(data):
|
|
"""Decode JSON bytes to string with proper encoding
|
|
|
|
Only for supporting Py 3.5
|
|
|
|
Py 3.6 supports bytes as parameter for json.load, we can drop this when
|
|
there is no need for python 3.5 anymore
|
|
|
|
Code from:
|
|
https://bugs.python.org/file43513/json_detect_encoding_3.patch
|
|
|
|
:param data: JSON bytes
|
|
:return: return JSON string
|
|
"""
|
|
|
|
def detect_encoding(b):
|
|
bstartswith = b.startswith
|
|
if bstartswith((codecs.BOM_UTF32_BE, codecs.BOM_UTF32_LE)):
|
|
return 'utf-32'
|
|
if bstartswith((codecs.BOM_UTF16_BE, codecs.BOM_UTF16_LE)):
|
|
return 'utf-16'
|
|
if bstartswith(codecs.BOM_UTF8):
|
|
return 'utf-8-sig'
|
|
|
|
if len(b) >= 4:
|
|
if not b[0]:
|
|
# 00 00 -- -- - utf-32-be
|
|
# 00 XX -- -- - utf-16-be
|
|
return 'utf-16-be' if b[1] else 'utf-32-be'
|
|
if not b[1]:
|
|
# XX 00 00 00 - utf-32-le
|
|
# XX 00 XX XX - utf-16-le
|
|
return 'utf-16-le' if b[2] or b[3] else 'utf-32-le'
|
|
elif len(b) == 2:
|
|
if not b[0]:
|
|
# 00 XX - utf-16-be
|
|
return 'utf-16-be'
|
|
if not b[1]:
|
|
# XX 00 - utf-16-le
|
|
return 'utf-16-le'
|
|
# default
|
|
return 'utf-8'
|
|
|
|
if isinstance(data, str):
|
|
return data
|
|
|
|
return data.decode(detect_encoding(data), 'surrogatepass')
|
|
|
|
|
|
class APIVersion(tuple):
|
|
"""API version parser and handler
|
|
|
|
The class is used to parse ipapython.version.API_VERSION and plugin
|
|
versions.
|
|
"""
|
|
__slots__ = ()
|
|
|
|
def __new__(cls, version):
|
|
major, dot, minor = version.partition(u'.')
|
|
major = int(major)
|
|
minor = int(minor) if dot else 0
|
|
return tuple.__new__(cls, (major, minor))
|
|
|
|
def __str__(self):
|
|
return '{}.{}'.format(*self)
|
|
|
|
def __repr__(self):
|
|
return "<APIVersion('{}.{}')>".format(*self)
|
|
|
|
def __getnewargs__(self):
|
|
return (str(self),)
|
|
|
|
@property
|
|
def major(self):
|
|
return self[0]
|
|
|
|
@property
|
|
def minor(self):
|
|
return self[1]
|
|
|
|
|
|
def remove_keytab(keytab_path):
|
|
"""
|
|
Remove Kerberos keytab and issue a warning if the procedure fails
|
|
|
|
:param keytab_path: path to the keytab file
|
|
"""
|
|
try:
|
|
logger.debug("Removing service keytab: %s", keytab_path)
|
|
os.remove(keytab_path)
|
|
except OSError as e:
|
|
if e.errno != errno.ENOENT:
|
|
logger.warning("Failed to remove Kerberos keytab '%s': %s",
|
|
keytab_path, e)
|
|
logger.warning("You may have to remove it manually")
|
|
|
|
|
|
def remove_ccache(ccache_path=None, run_as=None):
|
|
"""
|
|
remove Kerberos credential cache, essentially a wrapper around kdestroy.
|
|
|
|
:param ccache_path: path to the ccache file
|
|
:param run_as: run kdestroy as this user
|
|
"""
|
|
logger.debug("Removing service credentials cache")
|
|
kdestroy_cmd = [paths.KDESTROY]
|
|
if ccache_path is not None:
|
|
logger.debug("Ccache path: '%s'", ccache_path)
|
|
kdestroy_cmd.extend(['-c', ccache_path])
|
|
|
|
try:
|
|
run(kdestroy_cmd, runas=run_as, env={})
|
|
except CalledProcessError as e:
|
|
logger.warning(
|
|
"Failed to clear Kerberos credentials cache: %s", e)
|
|
|
|
|
|
def remove_file(filename):
|
|
"""Remove a file and log any exceptions raised.
|
|
"""
|
|
try:
|
|
os.unlink(filename)
|
|
except Exception as e:
|
|
# ignore missing file
|
|
if getattr(e, 'errno', None) != errno.ENOENT:
|
|
logger.error('Error removing %s: %s', filename, str(e))
|
|
|
|
|
|
def rmtree(path):
|
|
"""
|
|
Remove a directory structure and log any exceptions raised.
|
|
"""
|
|
try:
|
|
if os.path.exists(path):
|
|
shutil.rmtree(path)
|
|
except Exception as e:
|
|
logger.error('Error removing %s: %s', path, str(e))
|
|
|
|
|
|
def datetime_from_utctimestamp(t, units=1):
|
|
"""
|
|
Convert a timestamp or a time.struct_time to a datetime.datetime
|
|
object down to seconds, with UTC timezone
|
|
|
|
The conversion is safe for year 2038 problem
|
|
|
|
:param t: int or float timestamp in (milli)seconds since UNIX epoch
|
|
or time.struct_time
|
|
:param units: normalizing factor for the timestamp
|
|
(1 for seconds, 1000 for milliseconds)
|
|
defaults to 1
|
|
:return: datetime.datetime object in UTC timezone
|
|
"""
|
|
if isinstance(t, time.struct_time):
|
|
v = int(time.mktime(t))
|
|
elif isinstance(t, (float, int)):
|
|
v = int(t)
|
|
else:
|
|
raise TypeError(t)
|
|
epoch = datetime.datetime(1970, 1, 1, tzinfo=datetime.timezone.utc)
|
|
return epoch + datetime.timedelta(seconds=v // units)
|
|
|
|
|
|
class Sleeper:
|
|
"""Helper for time.sleep() loop with timeout
|
|
|
|
A sleeper object sleeps *sleep* seconds when it is called. Close to its
|
|
deadline it sleeps shorter to not oversleep the *timeout* deadline. A
|
|
sleeper object is *True* and returns *True* before it reaches *timeout*
|
|
deadline. After its deadline a sleeper raises the exception object/class
|
|
in *raises*. If *raises* is not given, it returns False instead.
|
|
|
|
sleep = Sleeper(sleep=1, timeout=60, raises=TimeoutError)
|
|
while True:
|
|
do_something()
|
|
sleep()
|
|
|
|
sleep = Sleeper(sleep=0.5, timeout=60)
|
|
while True:
|
|
do_something
|
|
if not sleep():
|
|
log.info("timeout")
|
|
break
|
|
|
|
longsleep = Sleeper(sleep=1, timeout=sys.maxsize)
|
|
"""
|
|
multiplier = 2
|
|
|
|
def __init__(self, *, sleep, timeout, raises=None):
|
|
if timeout <= 0:
|
|
raise ValueError(f"invalid timeout {timeout}")
|
|
if sleep < 0.01:
|
|
raise ValueError(f"sleep duration {sleep} is too short.")
|
|
|
|
self.timeout = timeout
|
|
self.sleep = sleep
|
|
self.raises = raises
|
|
|
|
self.deadline = time.monotonic() + self.timeout
|
|
|
|
def __bool__(self):
|
|
return time.monotonic() < self.deadline
|
|
|
|
def __call__(self):
|
|
now = time.monotonic()
|
|
if now >= self.deadline:
|
|
if self.raises is not None:
|
|
raise self.raises
|
|
else:
|
|
return False
|
|
# don't sleep over deadline
|
|
dur = min(self.deadline - now, self.sleep)
|
|
time.sleep(dur)
|
|
return True
|