# (c) 2012, Jeroen Hoekx # # 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 . # Make coding more python3-ish from __future__ import (absolute_import, division, print_function) __metaclass__ = type import base64 import crypt import glob import hashlib import itertools import json import ntpath import os.path import re import string import sys import time import uuid import yaml from collections import Mapping, MutableMapping import datetime from functools import partial from random import Random, SystemRandom, shuffle, randint from jinja2.filters import environmentfilter, do_groupby as _do_groupby from ansible.errors import AnsibleError, AnsibleFilterError from ansible.module_utils.six import iteritems, string_types, integer_types, reraise from ansible.module_utils.six.moves import reduce, shlex_quote from ansible.module_utils._text import to_bytes, to_text from ansible.module_utils.common.collections import is_sequence from ansible.parsing.ajson import AnsibleJSONEncoder from ansible.parsing.yaml.dumper import AnsibleDumper from ansible.utils.encrypt import passlib_or_crypt from ansible.utils.hashing import md5s, checksum_s from ansible.utils.unicode import unicode_wrap from ansible.utils.vars import merge_hash try: from __main__ import display except ImportError: from ansible.utils.display import Display display = Display() UUID_NAMESPACE_ANSIBLE = uuid.UUID('361E6D51-FAEC-444A-9079-341386DA8E2E') def to_yaml(a, *args, **kw): '''Make verbose, human readable yaml''' default_flow_style = kw.pop('default_flow_style', None) transformed = yaml.dump(a, Dumper=AnsibleDumper, allow_unicode=True, default_flow_style=default_flow_style, **kw) return to_text(transformed) def to_nice_yaml(a, indent=4, *args, **kw): '''Make verbose, human readable yaml''' transformed = yaml.dump(a, Dumper=AnsibleDumper, indent=indent, allow_unicode=True, default_flow_style=False, **kw) return to_text(transformed) def to_json(a, *args, **kw): ''' Convert the value to JSON ''' return json.dumps(a, cls=AnsibleJSONEncoder, *args, **kw) def to_nice_json(a, indent=4, *args, **kw): '''Make verbose, human readable JSON''' try: return json.dumps(a, indent=indent, sort_keys=True, separators=(',', ': '), cls=AnsibleJSONEncoder, *args, **kw) except Exception as e: # Fallback to the to_json filter display.warning(u'Unable to convert data using to_nice_json, falling back to to_json: %s' % to_text(e)) return to_json(a, *args, **kw) def to_bool(a): ''' return a bool for the arg ''' if a is None or isinstance(a, bool): return a if isinstance(a, string_types): a = a.lower() if a in ('yes', 'on', '1', 'true', 1): return True return False def to_datetime(string, format="%Y-%m-%d %H:%M:%S"): return datetime.datetime.strptime(string, format) def strftime(string_format, second=None): ''' return a date string using string. See https://docs.python.org/2/library/time.html#time.strftime for format ''' if second is not None: try: second = int(second) except Exception: raise AnsibleFilterError('Invalid value for epoch value (%s)' % second) return time.strftime(string_format, time.localtime(second)) def quote(a): ''' return its argument quoted for shell usage ''' return shlex_quote(to_text(a)) def fileglob(pathname): ''' return list of matched regular files for glob ''' return [g for g in glob.glob(pathname) if os.path.isfile(g)] def regex_replace(value='', pattern='', replacement='', ignorecase=False): ''' Perform a `re.sub` returning a string ''' value = to_text(value, errors='surrogate_or_strict', nonstring='simplerepr') if ignorecase: flags = re.I else: flags = 0 _re = re.compile(pattern, flags=flags) return _re.sub(replacement, value) def regex_findall(value, regex, multiline=False, ignorecase=False): ''' Perform re.findall and return the list of matches ''' flags = 0 if ignorecase: flags |= re.I if multiline: flags |= re.M return re.findall(regex, value, flags) def regex_search(value, regex, *args, **kwargs): ''' Perform re.search and return the list of matches or a backref ''' groups = list() for arg in args: if arg.startswith('\\g'): match = re.match(r'\\g<(\S+)>', arg).group(1) groups.append(match) elif arg.startswith('\\'): match = int(re.match(r'\\(\d+)', arg).group(1)) groups.append(match) else: raise AnsibleFilterError('Unknown argument') flags = 0 if kwargs.get('ignorecase'): flags |= re.I if kwargs.get('multiline'): flags |= re.M match = re.search(regex, value, flags) if match: if not groups: return match.group() else: items = list() for item in groups: items.append(match.group(item)) return items def ternary(value, true_val, false_val): ''' value ? true_val : false_val ''' if bool(value): return true_val else: return false_val def regex_escape(string): '''Escape all regular expressions special characters from STRING.''' return re.escape(string) def from_yaml(data): if isinstance(data, string_types): return yaml.safe_load(data) return data def from_yaml_all(data): if isinstance(data, string_types): return yaml.safe_load_all(data) return data @environmentfilter def rand(environment, end, start=None, step=None, seed=None): if seed is None: r = SystemRandom() else: r = Random(seed) if isinstance(end, integer_types): if not start: start = 0 if not step: step = 1 return r.randrange(start, end + 1, step) elif hasattr(end, '__iter__'): if start or step: raise AnsibleFilterError('start and step can only be used with integer values') return r.choice(end) else: raise AnsibleFilterError('random can only be used on sequences and integers') def randomize_list(mylist, seed=None): try: mylist = list(mylist) if seed: r = Random(seed) r.shuffle(mylist) else: shuffle(mylist) except Exception: pass return mylist def get_hash(data, hashtype='sha1'): try: # see if hash is supported h = hashlib.new(hashtype) except Exception: return None h.update(to_bytes(data, errors='surrogate_or_strict')) return h.hexdigest() def get_encrypted_password(password, hashtype='sha512', salt=None, salt_size=None, rounds=None): passlib_mapping = { 'md5': 'md5_crypt', 'blowfish': 'bcrypt', 'sha256': 'sha256_crypt', 'sha512': 'sha512_crypt', } hashtype = passlib_mapping.get(hashtype, hashtype) try: return passlib_or_crypt(password, hashtype, salt=salt, salt_size=salt_size, rounds=rounds) except AnsibleError as e: reraise(AnsibleFilterError, AnsibleFilterError(str(e), orig_exc=e), sys.exc_info()[2]) def to_uuid(string): return str(uuid.uuid5(UUID_NAMESPACE_ANSIBLE, str(string))) def mandatory(a): from jinja2.runtime import Undefined ''' Make a variable mandatory ''' if isinstance(a, Undefined): if a._undefined_name is not None: name = "'%s' " % to_text(a._undefined_name) else: name = '' raise AnsibleFilterError("Mandatory variable %snot defined." % name) return a def combine(*terms, **kwargs): recursive = kwargs.get('recursive', False) if len(kwargs) > 1 or (len(kwargs) == 1 and 'recursive' not in kwargs): raise AnsibleFilterError("'recursive' is the only valid keyword argument") dicts = [] for t in terms: if isinstance(t, MutableMapping): dicts.append(t) elif isinstance(t, list): dicts.append(combine(*t, **kwargs)) else: raise AnsibleFilterError("|combine expects dictionaries, got " + repr(t)) if recursive: return reduce(merge_hash, dicts) else: return dict(itertools.chain(*map(iteritems, dicts))) def comment(text, style='plain', **kw): # Predefined comment types comment_styles = { 'plain': { 'decoration': '# ' }, 'erlang': { 'decoration': '% ' }, 'c': { 'decoration': '// ' }, 'cblock': { 'beginning': '/*', 'decoration': ' * ', 'end': ' */' }, 'xml': { 'beginning': '' } } # Pointer to the right comment type style_params = comment_styles[style] if 'decoration' in kw: prepostfix = kw['decoration'] else: prepostfix = style_params['decoration'] # Default params p = { 'newline': '\n', 'beginning': '', 'prefix': (prepostfix).rstrip(), 'prefix_count': 1, 'decoration': '', 'postfix': (prepostfix).rstrip(), 'postfix_count': 1, 'end': '' } # Update default params p.update(style_params) p.update(kw) # Compose substrings for the final string str_beginning = '' if p['beginning']: str_beginning = "%s%s" % (p['beginning'], p['newline']) str_prefix = '' if p['prefix']: if p['prefix'] != p['newline']: str_prefix = str( "%s%s" % (p['prefix'], p['newline'])) * int(p['prefix_count']) else: str_prefix = str( "%s" % (p['newline'])) * int(p['prefix_count']) str_text = ("%s%s" % ( p['decoration'], # Prepend each line of the text with the decorator text.replace( p['newline'], "%s%s" % (p['newline'], p['decoration'])))).replace( # Remove trailing spaces when only decorator is on the line "%s%s" % (p['decoration'], p['newline']), "%s%s" % (p['decoration'].rstrip(), p['newline'])) str_postfix = p['newline'].join( [''] + [p['postfix'] for x in range(p['postfix_count'])]) str_end = '' if p['end']: str_end = "%s%s" % (p['newline'], p['end']) # Return the final string return "%s%s%s%s%s" % ( str_beginning, str_prefix, str_text, str_postfix, str_end) def extract(item, container, morekeys=None): from jinja2.runtime import Undefined value = container[item] if value is not Undefined and morekeys is not None: if not isinstance(morekeys, list): morekeys = [morekeys] try: value = reduce(lambda d, k: d[k], morekeys, value) except KeyError: value = Undefined() return value @environmentfilter def do_groupby(environment, value, attribute): """Overridden groupby filter for jinja2, to address an issue with jinja2>=2.9.0,<2.9.5 where a namedtuple was returned which has repr that prevents ansible.template.safe_eval.safe_eval from being able to parse and eval the data. jinja2<2.9.0,>=2.9.5 is not affected, as <2.9.0 uses a tuple, and >=2.9.5 uses a standard tuple repr on the namedtuple. The adaptation here, is to run the jinja2 `do_groupby` function, and cast all of the namedtuples to a regular tuple. See https://github.com/ansible/ansible/issues/20098 We may be able to remove this in the future. """ return [tuple(t) for t in _do_groupby(environment, value, attribute)] def b64encode(string, encoding='utf-8'): return to_text(base64.b64encode(to_bytes(string, encoding=encoding, errors='surrogate_or_strict'))) def b64decode(string, encoding='utf-8'): return to_text(base64.b64decode(to_bytes(string, errors='surrogate_or_strict')), encoding=encoding) def flatten(mylist, levels=None): ret = [] for element in mylist: if element in (None, 'None', 'null'): # ignore undefined items break elif is_sequence(element): if levels is None: ret.extend(flatten(element)) elif levels >= 1: # decrement as we go down the stack ret.extend(flatten(element, levels=(int(levels) - 1))) else: ret.append(element) else: ret.append(element) return ret def subelements(obj, subelements, skip_missing=False): '''Accepts a dict or list of dicts, and a dotted accessor and produces a product of the element and the results of the dotted accessor >>> obj = [{"name": "alice", "groups": ["wheel"], "authorized": ["/tmp/alice/onekey.pub"]}] >>> subelements(obj, 'groups') [({'name': 'alice', 'groups': ['wheel'], 'authorized': ['/tmp/alice/onekey.pub']}, 'wheel')] ''' if isinstance(obj, dict): element_list = list(obj.values()) elif isinstance(obj, list): element_list = obj[:] else: raise AnsibleFilterError('obj must be a list of dicts or a nested dict') if isinstance(subelements, list): subelement_list = subelements[:] elif isinstance(subelements, string_types): subelement_list = subelements.split('.') else: raise AnsibleFilterError('subelements must be a list or a string') results = [] for element in element_list: values = element for subelement in subelement_list: try: values = values[subelement] except KeyError: if skip_missing: values = [] break raise AnsibleFilterError("could not find %r key in iterated item %r" % (subelement, values)) except TypeError: raise AnsibleFilterError("the key %s should point to a dictionary, got '%s'" % (subelement, values)) if not isinstance(values, list): raise AnsibleFilterError("the key %r should point to a list, got %r" % (subelement, values)) for value in values: results.append((element, value)) return results def dict_to_list_of_dict_key_value_elements(mydict): ''' takes a dictionary and transforms it into a list of dictionaries, with each having a 'key' and 'value' keys that correspond to the keys and values of the original ''' if not isinstance(mydict, Mapping): raise AnsibleFilterError("dict2items requires a dictionary, got %s instead." % type(mydict)) ret = [] for key in mydict: ret.append({'key': key, 'value': mydict[key]}) return ret def list_of_dict_key_value_elements_to_dict(mylist, key_name='key', value_name='value'): ''' takes a list of dicts with each having a 'key' and 'value' keys, and transforms the list into a dictionary, effectively as the reverse of dict2items ''' if not is_sequence(mylist): raise AnsibleFilterError("items2dict requires a list, got %s instead." % type(mylist)) return dict((item[key_name], item[value_name]) for item in mylist) def random_mac(value): ''' takes string prefix, and return it completed with random bytes to get a complete 6 bytes MAC address ''' if not isinstance(value, string_types): raise AnsibleFilterError('Invalid value type (%s) for random_mac (%s)' % (type(value), value)) value = value.lower() mac_items = value.split(':') if len(mac_items) > 5: raise AnsibleFilterError('Invalid value (%s) for random_mac: 5 colon(:) separated items max' % value) err = "" for mac in mac_items: if len(mac) == 0: err += ",empty item" continue if not re.match('[a-f0-9]{2}', mac): err += ",%s not hexa byte" % mac err = err.strip(',') if len(err): raise AnsibleFilterError('Invalid value (%s) for random_mac: %s' % (value, err)) # Generate random int between x1000000000 and xFFFFFFFFFF v = randint(68719476736, 1099511627775) # Select first n chars to complement input prefix remain = 2 * (6 - len(mac_items)) rnd = ('%x' % v)[:remain] return value + re.sub(r'(..)', r':\1', rnd) class FilterModule(object): ''' Ansible core jinja2 filters ''' def filters(self): return { # jinja2 overrides 'groupby': do_groupby, # base 64 'b64decode': b64decode, 'b64encode': b64encode, # uuid 'to_uuid': to_uuid, # json 'to_json': to_json, 'to_nice_json': to_nice_json, 'from_json': json.loads, # yaml 'to_yaml': to_yaml, 'to_nice_yaml': to_nice_yaml, 'from_yaml': from_yaml, 'from_yaml_all': from_yaml_all, # path 'basename': partial(unicode_wrap, os.path.basename), 'dirname': partial(unicode_wrap, os.path.dirname), 'expanduser': partial(unicode_wrap, os.path.expanduser), 'expandvars': partial(unicode_wrap, os.path.expandvars), 'realpath': partial(unicode_wrap, os.path.realpath), 'relpath': partial(unicode_wrap, os.path.relpath), 'splitext': partial(unicode_wrap, os.path.splitext), 'win_basename': partial(unicode_wrap, ntpath.basename), 'win_dirname': partial(unicode_wrap, ntpath.dirname), 'win_splitdrive': partial(unicode_wrap, ntpath.splitdrive), # file glob 'fileglob': fileglob, # types 'bool': to_bool, 'to_datetime': to_datetime, # date formating 'strftime': strftime, # quote string for shell usage 'quote': quote, # hash filters # md5 hex digest of string 'md5': md5s, # sha1 hex digeset of string 'sha1': checksum_s, # checksum of string as used by ansible for checksumming files 'checksum': checksum_s, # generic hashing 'password_hash': get_encrypted_password, 'hash': get_hash, # regex 'regex_replace': regex_replace, 'regex_escape': regex_escape, 'regex_search': regex_search, 'regex_findall': regex_findall, # ? : ; 'ternary': ternary, # random stuff 'random': rand, 'shuffle': randomize_list, # undefined 'mandatory': mandatory, # comment-style decoration 'comment': comment, # debug 'type_debug': lambda o: o.__class__.__name__, # Data structures 'combine': combine, 'extract': extract, 'flatten': flatten, 'dict2items': dict_to_list_of_dict_key_value_elements, 'items2dict': list_of_dict_key_value_elements_to_dict, 'subelements': subelements, # Misc 'random_mac': random_mac, }