# (c) 2012, Daniel Hokka Zakrisson # (c) 2012-2014, Michael DeHaan and others # (c) 2017, Toshio Kuratomi # (c) 2017 Ansible Project # GNU General Public License v3.0+ (see COPYING or https://www.gnu.org/licenses/gpl-3.0.txt) from __future__ import (absolute_import, division, print_function) __metaclass__ = type import glob import imp import os import os.path import sys import warnings from collections import defaultdict from ansible import constants as C from ansible.errors import AnsibleError from ansible.module_utils._text import to_text from ansible.parsing.utils.yaml import from_yaml from ansible.plugins import get_plugin_class, MODULE_CACHE, PATH_CACHE, PLUGIN_PATH_CACHE from ansible.utils.plugin_docs import get_docstring try: from __main__ import display except ImportError: from ansible.utils.display import Display display = Display() def get_all_plugin_loaders(): return [(name, obj) for (name, obj) in globals().items() if isinstance(obj, PluginLoader)] class PluginLoader: ''' PluginLoader loads plugins from the configured plugin directories. It searches for plugins by iterating through the combined list of play basedirs, configured paths, and the python path. The first match is used. ''' def __init__(self, class_name, package, config, subdir, aliases=None, required_base_class=None): aliases = {} if aliases is None else aliases self.class_name = class_name self.base_class = required_base_class self.package = package self.subdir = subdir # FIXME: remove alias dict in favor of alias by symlink? self.aliases = aliases if config and not isinstance(config, list): config = [config] elif not config: config = [] self.config = config if class_name not in MODULE_CACHE: MODULE_CACHE[class_name] = {} if class_name not in PATH_CACHE: PATH_CACHE[class_name] = None if class_name not in PLUGIN_PATH_CACHE: PLUGIN_PATH_CACHE[class_name] = defaultdict(dict) self._module_cache = MODULE_CACHE[class_name] self._paths = PATH_CACHE[class_name] self._plugin_path_cache = PLUGIN_PATH_CACHE[class_name] self._extra_dirs = [] self._searched_paths = set() def __setstate__(self, data): ''' Deserializer. ''' class_name = data.get('class_name') package = data.get('package') config = data.get('config') subdir = data.get('subdir') aliases = data.get('aliases') base_class = data.get('base_class') PATH_CACHE[class_name] = data.get('PATH_CACHE') PLUGIN_PATH_CACHE[class_name] = data.get('PLUGIN_PATH_CACHE') self.__init__(class_name, package, config, subdir, aliases, base_class) self._extra_dirs = data.get('_extra_dirs', []) self._searched_paths = data.get('_searched_paths', set()) def __getstate__(self): ''' Serializer. ''' return dict( class_name=self.class_name, base_class=self.base_class, package=self.package, config=self.config, subdir=self.subdir, aliases=self.aliases, _extra_dirs=self._extra_dirs, _searched_paths=self._searched_paths, PATH_CACHE=PATH_CACHE[self.class_name], PLUGIN_PATH_CACHE=PLUGIN_PATH_CACHE[self.class_name], ) def format_paths(self, paths): ''' Returns a string suitable for printing of the search path ''' # Uses a list to get the order right ret = [] for i in paths: if i not in ret: ret.append(i) return os.pathsep.join(ret) def print_paths(self): return self.format_paths(self._get_paths(subdirs=False)) def _all_directories(self, dir): results = [] results.append(dir) for root, subdirs, files in os.walk(dir, followlinks=True): if '__init__.py' in files: for x in subdirs: results.append(os.path.join(root, x)) return results def _get_package_paths(self, subdirs=True): ''' Gets the path of a Python package ''' if not self.package: return [] if not hasattr(self, 'package_path'): m = __import__(self.package) parts = self.package.split('.')[1:] for parent_mod in parts: m = getattr(m, parent_mod) self.package_path = os.path.dirname(m.__file__) if subdirs: return self._all_directories(self.package_path) return [self.package_path] def _get_paths(self, subdirs=True): ''' Return a list of paths to search for plugins in ''' # FIXME: This is potentially buggy if subdirs is sometimes True and sometimes False. # In current usage, everything calls this with subdirs=True except for module_utils_loader and ansible-doc # which always calls it with subdirs=False. So there currently isn't a problem with this caching. if self._paths is not None: return self._paths ret = self._extra_dirs[:] # look in any configured plugin paths, allow one level deep for subcategories if self.config is not None: for path in self.config: path = os.path.realpath(os.path.expanduser(path)) if subdirs: contents = glob.glob("%s/*" % path) + glob.glob("%s/*/*" % path) for c in contents: if os.path.isdir(c) and c not in ret: ret.append(c) if path not in ret: ret.append(path) # look for any plugins installed in the package subtree # Note package path always gets added last so that every other type of # path is searched before it. ret.extend(self._get_package_paths(subdirs=subdirs)) # HACK: because powershell modules are in the same directory # hierarchy as other modules we have to process them last. This is # because powershell only works on windows but the other modules work # anywhere (possibly including windows if the correct language # interpreter is installed). the non-powershell modules can have any # file extension and thus powershell modules are picked up in that. # The non-hack way to fix this is to have powershell modules be # a different PluginLoader/ModuleLoader. But that requires changing # other things too (known thing to change would be PATHS_CACHE, # PLUGIN_PATHS_CACHE, and MODULE_CACHE. Since those three dicts key # on the class_name and neither regular modules nor powershell modules # would have class_names, they would not work as written. reordered_paths = [] win_dirs = [] for path in ret: if path.endswith('windows'): win_dirs.append(path) else: reordered_paths.append(path) reordered_paths.extend(win_dirs) # cache and return the result self._paths = reordered_paths return reordered_paths def _load_config_defs(self, name, path): ''' Reads plugin docs to find configuration setting definitions, to push to config manager for later use ''' # plugins w/o class name don't support config if self.class_name: type_name = get_plugin_class(self.class_name) # if type name != 'module_doc_fragment': if type_name in C.CONFIGURABLE_PLUGINS: dstring = get_docstring(path, fragment_loader, verbose=False, ignore_errors=True)[0] if dstring and 'options' in dstring and isinstance(dstring['options'], dict): C.config.initialize_plugin_configuration_definitions(type_name, name, dstring['options']) display.debug('Loaded config def from plugin (%s/%s)' % (type_name, name)) def add_directory(self, directory, with_subdir=False): ''' Adds an additional directory to the search path ''' directory = os.path.realpath(directory) if directory is not None: if with_subdir: directory = os.path.join(directory, self.subdir) if directory not in self._extra_dirs: # append the directory and invalidate the path cache self._extra_dirs.append(directory) self._paths = None display.debug('Added %s to loader search path' % (directory)) def _find_plugin(self, name, mod_type='', ignore_deprecated=False, check_aliases=False): ''' Find a plugin named name ''' global _PLUGIN_FILTERS if name in _PLUGIN_FILTERS[self.package]: return None if mod_type: suffix = mod_type elif self.class_name: # Ansible plugins that run in the controller process (most plugins) suffix = '.py' else: # Only Ansible Modules. Ansible modules can be any executable so # they can have any suffix suffix = '' if check_aliases: name = self.aliases.get(name, name) # The particular cache to look for modules within. This matches the # requested mod_type pull_cache = self._plugin_path_cache[suffix] try: return pull_cache[name] except KeyError: # Cache miss. Now let's find the plugin pass # TODO: Instead of using the self._paths cache (PATH_CACHE) and # self._searched_paths we could use an iterator. Before enabling that # we need to make sure we don't want to add additional directories # (add_directory()) once we start using the iterator. Currently, it # looks like _get_paths() never forces a cache refresh so if we expect # additional directories to be added later, it is buggy. for path in (p for p in self._get_paths() if p not in self._searched_paths and os.path.isdir(p)): try: full_paths = (os.path.join(path, f) for f in os.listdir(path)) except OSError as e: display.warning("Error accessing plugin paths: %s" % to_text(e)) for full_path in (f for f in full_paths if os.path.isfile(f) and not f.endswith('__init__.py')): full_name = os.path.basename(full_path) # HACK: We have no way of executing python byte compiled files as ansible modules so specifically exclude them # FIXME: I believe this is only correct for modules and module_utils. # For all other plugins we want .pyc and .pyo should be valid if full_path.endswith(('.pyc', '.pyo')): continue splitname = os.path.splitext(full_name) base_name = splitname[0] try: extension = splitname[1] except IndexError: extension = '' # Module found, now enter it into the caches that match this file if base_name not in self._plugin_path_cache['']: self._plugin_path_cache[''][base_name] = full_path if full_name not in self._plugin_path_cache['']: self._plugin_path_cache[''][full_name] = full_path if base_name not in self._plugin_path_cache[extension]: self._plugin_path_cache[extension][base_name] = full_path if full_name not in self._plugin_path_cache[extension]: self._plugin_path_cache[extension][full_name] = full_path self._searched_paths.add(path) try: return pull_cache[name] except KeyError: # Didn't find the plugin in this directory. Load modules from the next one pass # if nothing is found, try finding alias/deprecated if not name.startswith('_'): alias_name = '_' + name # We've already cached all the paths at this point if alias_name in pull_cache: if not ignore_deprecated and not os.path.islink(pull_cache[alias_name]): # FIXME: this is not always the case, some are just aliases display.deprecated('%s is kept for backwards compatibility but usage is discouraged. ' 'The module documentation details page may explain more about this rationale.' % name.lstrip('_')) return pull_cache[alias_name] return None def find_plugin(self, name, mod_type='', ignore_deprecated=False, check_aliases=False): ''' Find a plugin named name ''' # Import here to avoid circular import from ansible.vars.reserved import is_reserved_name plugin = self._find_plugin(name, mod_type=mod_type, ignore_deprecated=ignore_deprecated, check_aliases=check_aliases) if plugin and self.package == 'ansible.modules' and is_reserved_name(name): raise AnsibleError( 'Module "%s" shadows the name of a reserved keyword. Please rename or remove this module. Found at %s' % (name, plugin) ) return plugin def has_plugin(self, name): ''' Checks if a plugin named name exists ''' return self.find_plugin(name) is not None __contains__ = has_plugin def _load_module_source(self, name, path): # avoid collisions across plugins full_name = '.'.join([self.package, name]) if full_name in sys.modules: # Avoids double loading, See https://github.com/ansible/ansible/issues/13110 return sys.modules[full_name] with warnings.catch_warnings(): warnings.simplefilter("ignore", RuntimeWarning) with open(path, 'rb') as module_file: module = imp.load_source(full_name, path, module_file) return module def _update_object(self, obj, name, path): # set extra info on the module, in case we want it later setattr(obj, '_original_path', path) setattr(obj, '_load_name', name) def get(self, name, *args, **kwargs): ''' instantiates a plugin of the given name using arguments ''' found_in_cache = True class_only = kwargs.pop('class_only', False) if name in self.aliases: name = self.aliases[name] path = self.find_plugin(name) if path is None: return None if path not in self._module_cache: self._module_cache[path] = self._load_module_source(name, path) found_in_cache = False obj = getattr(self._module_cache[path], self.class_name) if self.base_class: # The import path is hardcoded and should be the right place, # so we are not expecting an ImportError. module = __import__(self.package, fromlist=[self.base_class]) # Check whether this obj has the required base class. try: plugin_class = getattr(module, self.base_class) except AttributeError: return None if not issubclass(obj, plugin_class): return None self._display_plugin_load(self.class_name, name, self._searched_paths, path, found_in_cache=found_in_cache, class_only=class_only) if not class_only: try: obj = obj(*args, **kwargs) except TypeError as e: if "abstract" in e.args[0]: # Abstract Base Class. The found plugin file does not # fully implement the defined interface. return None raise # load plugin config data if not found_in_cache: self._load_config_defs(name, path) self._update_object(obj, name, path) return obj def _display_plugin_load(self, class_name, name, searched_paths, path, found_in_cache=None, class_only=None): msg = 'Loading %s \'%s\' from %s' % (class_name, os.path.basename(name), path) if len(searched_paths) > 1: msg = '%s (searched paths: %s)' % (msg, self.format_paths(searched_paths)) if found_in_cache or class_only: msg = '%s (found_in_cache=%s, class_only=%s)' % (msg, found_in_cache, class_only) display.debug(msg) def all(self, *args, **kwargs): ''' Iterate through all plugins of this type A plugin loader is initialized with a specific type. This function is an iterator returning all of the plugins of that type to the caller. :kwarg path_only: If this is set to True, then we return the paths to where the plugins reside instead of an instance of the plugin. This conflicts with class_only and both should not be set. :kwarg class_only: If this is set to True then we return the python class which implements a plugin rather than an instance of the plugin. This conflicts with path_only and both should not be set. :kwarg _dedupe: By default, we only return one plugin per plugin name. Deduplication happens in the same way as the :meth:`get` and :meth:`find_plugin` methods resolve which plugin should take precedence. If this is set to False, then we return all of the plugins found, including those with duplicate names. In the case of duplicates, the order in which they are returned is the one that would take precedence first, followed by the others in decreasing precedence order. This should only be used by subclasses which want to manage their own deduplication of the plugins. :*args: Any extra arguments are passed to each plugin when it is instantiated. :**kwargs: Any extra keyword arguments are passed to each plugin when it is instantiated. ''' # TODO: Change the signature of this method to: # def all(return_type='instance', args=None, kwargs=None): # if args is None: args = [] # if kwargs is None: kwargs = {} # return_type can be instance, class, or path. # These changes will mean that plugin parameters won't conflict with our params and # will also make it impossible to request both a path and a class at the same time. # # Move _dedupe to be a class attribute, CUSTOM_DEDUPE, with subclasses for filters and # tests setting it to True global _PLUGIN_FILTERS dedupe = kwargs.pop('_dedupe', True) path_only = kwargs.pop('path_only', False) class_only = kwargs.pop('class_only', False) # Having both path_only and class_only is a coding bug if path_only and class_only: raise AnsibleError('Do not set both path_only and class_only when calling PluginLoader.all()') all_matches = [] found_in_cache = True for i in self._get_paths(): all_matches.extend(glob.glob(os.path.join(i, "*.py"))) loaded_modules = set() for path in sorted(all_matches, key=os.path.basename): name = os.path.splitext(path)[0] basename = os.path.basename(name) if basename == '__init__' or basename in _PLUGIN_FILTERS[self.package]: continue if dedupe and basename in loaded_modules: continue loaded_modules.add(basename) if path_only: yield path continue if path not in self._module_cache: module = self._load_module_source(name, path) self._module_cache[path] = module found_in_cache = False try: obj = getattr(self._module_cache[path], self.class_name) except AttributeError as e: display.warning("Skipping plugin (%s) as it seems to be invalid: %s" % (path, to_text(e))) continue if self.base_class: # The import path is hardcoded and should be the right place, # so we are not expecting an ImportError. module = __import__(self.package, fromlist=[self.base_class]) # Check whether this obj has the required base class. try: plugin_class = getattr(module, self.base_class) except AttributeError: continue if not issubclass(obj, plugin_class): continue self._display_plugin_load(self.class_name, basename, self._searched_paths, path, found_in_cache=found_in_cache, class_only=class_only) if not class_only: try: obj = obj(*args, **kwargs) except TypeError as e: display.warning("Skipping plugin (%s) as it seems to be incomplete: %s" % (path, to_text(e))) # load plugin config data if not found_in_cache: self._load_config_defs(basename, path) self._update_object(obj, basename, path) yield obj class Jinja2Loader(PluginLoader): """ PluginLoader optimized for Jinja2 plugins The filter and test plugins are Jinja2 plugins encapsulated inside of our plugin format. The way the calling code is setup, we need to do a few things differently in the all() method """ def find_plugin(self, name): # Nothing using Jinja2Loader use this method. We can't use the base class version because # we deduplicate differently than the base class raise AnsibleError('No code should call find_plugin for Jinja2Loaders (Not implemented)') def get(self, name, *args, **kwargs): # Nothing using Jinja2Loader use this method. We can't use the base class version because # we deduplicate differently than the base class raise AnsibleError('No code should call find_plugin for Jinja2Loaders (Not implemented)') def all(self, *args, **kwargs): """ Differences with :meth:`PluginLoader.all`: * We do not deduplicate ansible plugin names. This is because we don't care about our plugin names, here. We care about the names of the actual jinja2 plugins which are inside of our plugins. * We reverse the order of the list of plugins compared to other PluginLoaders. This is because of how calling code chooses to sync the plugins from the list. It adds all the Jinja2 plugins from one of our Ansible plugins into a dict. Then it adds the Jinja2 plugins from the next Ansible plugin, overwriting any Jinja2 plugins that had the same name. This is an encapsulation violation (the PluginLoader should not know about what calling code does with the data) but we're pushing the common code here. We'll fix this in the future by moving more of the common code into this PluginLoader. * We return a list. We could iterate the list instead but that's extra work for no gain because the API receiving this doesn't care. It just needs an iterable """ # We don't deduplicate ansible plugin names. Instead, calling code deduplicates jinja2 # plugin names. kwargs['_dedupe'] = False # We have to instantiate a list of all plugins so that we can reverse it. We reverse it so # that calling code will deduplicate this correctly. plugins = [p for p in super(Jinja2Loader, self).all(*args, **kwargs)] plugins.reverse() return plugins def _load_plugin_filter(): filters = defaultdict(frozenset) if C.PLUGIN_FILTERS_CFG is None: filter_cfg = '/etc/ansible/plugin_filters.yml' user_set = False else: filter_cfg = C.PLUGIN_FILTERS_CFG user_set = True if os.path.exists(filter_cfg): with open(filter_cfg, 'rb') as f: try: filter_data = from_yaml(f.read()) except Exception as e: display.warning(u'The plugin filter file, {0} was not parsable.' u' Skipping: {1}'.format(filter_cfg, to_text(e))) return filters try: version = filter_data['filter_version'] except KeyError: display.warning(u'The plugin filter file, {0} was invalid.' u' Skipping.'.format(filter_cfg)) return filters # Try to convert for people specifying version as a float instead of string version = to_text(version) version = version.strip() if version == u'1.0': # Modules and action plugins share the same blacklist since the difference between the # two isn't visible to the users filters['ansible.modules'] = frozenset(filter_data['module_blacklist']) filters['ansible.plugins.action'] = filters['ansible.modules'] else: display.warning(u'The plugin filter file, {0} was a version not recognized by this' u' version of Ansible. Skipping.') else: if user_set: display.warning(u'The plugin filter file, {0} does not exist.' u' Skipping.'.format(filter_cfg)) # Specialcase the stat module as Ansible can run very few things if stat is blacklisted. if 'stat' in filters['ansible.modules']: raise AnsibleError('The stat module was specified in the module blacklist file, {0}, but' ' Ansible will not function without the stat module. Please remove stat' ' from the blacklist.'.format(filter_cfg)) return filters # TODO: All of the following is initialization code It should be moved inside of an initialization # function which is called at some point early in the ansible and ansible-playbook CLI startup. _PLUGIN_FILTERS = _load_plugin_filter() # doc fragments first fragment_loader = PluginLoader( 'ModuleDocFragment', 'ansible.utils.module_docs_fragments', os.path.join(os.path.dirname(__file__), 'module_docs_fragments'), '', ) action_loader = PluginLoader( 'ActionModule', 'ansible.plugins.action', C.DEFAULT_ACTION_PLUGIN_PATH, 'action_plugins', required_base_class='ActionBase', ) cache_loader = PluginLoader( 'CacheModule', 'ansible.plugins.cache', C.DEFAULT_CACHE_PLUGIN_PATH, 'cache_plugins', ) callback_loader = PluginLoader( 'CallbackModule', 'ansible.plugins.callback', C.DEFAULT_CALLBACK_PLUGIN_PATH, 'callback_plugins', ) connection_loader = PluginLoader( 'Connection', 'ansible.plugins.connection', C.DEFAULT_CONNECTION_PLUGIN_PATH, 'connection_plugins', aliases={'paramiko': 'paramiko_ssh'}, required_base_class='ConnectionBase', ) shell_loader = PluginLoader( 'ShellModule', 'ansible.plugins.shell', 'shell_plugins', 'shell_plugins', ) module_loader = PluginLoader( '', 'ansible.modules', C.DEFAULT_MODULE_PATH, 'library', ) module_utils_loader = PluginLoader( '', 'ansible.module_utils', C.DEFAULT_MODULE_UTILS_PATH, 'module_utils', ) # NB: dedicated loader is currently necessary because PS module_utils expects "with subdir" lookup where # regular module_utils doesn't. This can be revisited once we have more granular loaders. ps_module_utils_loader = PluginLoader( '', 'ansible.module_utils', C.DEFAULT_MODULE_UTILS_PATH, 'module_utils', ) lookup_loader = PluginLoader( 'LookupModule', 'ansible.plugins.lookup', C.DEFAULT_LOOKUP_PLUGIN_PATH, 'lookup_plugins', required_base_class='LookupBase', ) filter_loader = Jinja2Loader( 'FilterModule', 'ansible.plugins.filter', C.DEFAULT_FILTER_PLUGIN_PATH, 'filter_plugins', ) test_loader = Jinja2Loader( 'TestModule', 'ansible.plugins.test', C.DEFAULT_TEST_PLUGIN_PATH, 'test_plugins' ) strategy_loader = PluginLoader( 'StrategyModule', 'ansible.plugins.strategy', C.DEFAULT_STRATEGY_PLUGIN_PATH, 'strategy_plugins', required_base_class='StrategyBase', ) terminal_loader = PluginLoader( 'TerminalModule', 'ansible.plugins.terminal', C.DEFAULT_TERMINAL_PLUGIN_PATH, 'terminal_plugins', required_base_class='TerminalBase' ) vars_loader = PluginLoader( 'VarsModule', 'ansible.plugins.vars', C.DEFAULT_VARS_PLUGIN_PATH, 'vars_plugins', ) cliconf_loader = PluginLoader( 'Cliconf', 'ansible.plugins.cliconf', C.DEFAULT_CLICONF_PLUGIN_PATH, 'cliconf_plugins', required_base_class='CliconfBase' ) netconf_loader = PluginLoader( 'Netconf', 'ansible.plugins.netconf', C.DEFAULT_NETCONF_PLUGIN_PATH, 'netconf_plugins', required_base_class='NetconfBase' ) inventory_loader = PluginLoader( 'InventoryModule', 'ansible.plugins.inventory', C.DEFAULT_INVENTORY_PLUGIN_PATH, 'inventory_plugins' )