# # Author:: Adam Jacob () # Author:: Christopher Walters () # Author:: Christopher Brown () # Author:: Tim Hinderliter () # Copyright:: Copyright 2008-2018, Chef Software Inc. # License:: Apache License, Version 2.0 # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. require "chef/config" require "chef/mixin/params_validate" require "chef/mixin/path_sanity" require "chef/log" require "chef/deprecated" require "chef/server_api" require "chef/api_client" require "chef/api_client/registration" require "chef/audit/runner" require "chef/node" require "chef/role" require "chef/file_cache" require "chef/run_context" require "chef/runner" require "chef/run_status" require "chef/cookbook/cookbook_collection" require "chef/cookbook/file_vendor" require "chef/cookbook/file_system_file_vendor" require "chef/cookbook/remote_file_vendor" require "chef/event_dispatch/dispatcher" require "chef/event_loggers/base" require "chef/event_loggers/windows_eventlog" require "chef/exceptions" require "chef/formatters/base" require "chef/formatters/doc" require "chef/formatters/minimal" require "chef/version" require "chef/resource_reporter" require "chef/data_collector" require "chef/audit/audit_reporter" require "chef/run_lock" require "chef/policy_builder" require "chef/request_id" require "chef/platform/rebooter" require "chef/mixin/deprecation" require "ohai" require "rbconfig" class Chef # == Chef::Client # The main object in a Chef run. Preps a Chef::Node and Chef::RunContext, # syncs cookbooks if necessary, and triggers convergence. class Client include Chef::Mixin::PathSanity extend Chef::Mixin::Deprecation # # The status of the Chef run. # # @return [Chef::RunStatus] # attr_reader :run_status # # The node represented by this client. # # @return [Chef::Node] # def node run_status.node end def node=(value) run_status.node = value end # # The ohai system used by this client. # # @return [Ohai::System] # attr_reader :ohai # # The rest object used to communicate with the Chef server. # # @return [Chef::ServerAPI] # attr_reader :rest # # The runner used to converge. # # @return [Chef::Runner] # attr_accessor :runner # # Extra node attributes that were applied to the node. # # @return [Hash] # attr_reader :json_attribs # # The event dispatcher for the Chef run, including any configured output # formatters and event loggers. # # @return [EventDispatch::Dispatcher] # # @see Chef::Formatters # @see Chef::Config#formatters # @see Chef::Config#stdout # @see Chef::Config#stderr # @see Chef::Config#force_logger # @see Chef::Config#force_formatter # TODO add stdout, stderr, and default formatters to Chef::Config so the # defaults aren't calculated here. Remove force_logger and force_formatter # from this code. # @see Chef::EventLoggers # @see Chef::Config#disable_event_logger # @see Chef::Config#event_loggers # @see Chef::Config#event_handlers # attr_reader :events attr_reader :logger # # Creates a new Chef::Client. # # @param json_attribs [Hash] Node attributes to layer into the node when it is # fetched. # @param args [Hash] Options: # @option args [Array] :override_runlist A runlist to # use instead of the node's embedded run list. # @option args [Array] :specific_recipes A list of recipe file paths # to load after the run list has been loaded. # def initialize(json_attribs = nil, args = {}) @json_attribs = json_attribs || {} @logger = args.delete(:logger) || Chef::Log.with_child @ohai = Ohai::System.new(logger: logger) event_handlers = configure_formatters + configure_event_loggers event_handlers += Array(Chef::Config[:event_handlers]) @events = EventDispatch::Dispatcher.new(*event_handlers) # @todo it seems like a bad idea to be deletin' other peoples' hashes. @override_runlist = args.delete(:override_runlist) @specific_recipes = args.delete(:specific_recipes) @run_status = Chef::RunStatus.new(nil, events) if new_runlist = args.delete(:runlist) @json_attribs["run_list"] = new_runlist end end # # Do a full run for this Chef::Client. # # Locks the run while doing its job. # # Fires run_start before doing anything and fires run_completed or # run_failed when finished. Also notifies client listeners of run_started # at the beginning of Compile, and run_completed_successfully or run_failed # when all is complete. # # Phase 1: Setup # -------------- # Gets information about the system and the run we are doing. # # 1. Run ohai to collect system information. # 2. Register / connect to the Chef server (unless in solo mode). # 3. Retrieve the node (or create a new one). # 4. Merge in json_attribs, Chef::Config.environment, and override_run_list. # # @see #run_ohai # @see #load_node # @see #build_node # @see Chef::Config#lockfile # @see Chef::RunLock#acquire # # Phase 2: Compile # ---------------- # Decides *what* we plan to converge by compiling recipes. # # 1. Sync required cookbooks to the local cache. # 2. Load libraries from all cookbooks. # 3. Load attributes from all cookbooks. # 4. Load LWRPs from all cookbooks. # 5. Load resource definitions from all cookbooks. # 6. Load recipes in the run list. # 7. Load recipes from the command line. # # @see #setup_run_context Syncs and compiles cookbooks. # @see Chef::CookbookCompiler#compile # # Phase 3: Converge # ----------------- # Brings the system up to date. # # 1. Converge the resources built from recipes in Phase 2. # 2. Save the node. # 3. Reboot if we were asked to. # # @see #converge_and_save # @see Chef::Runner # # Phase 4: Audit # -------------- # Runs 'control_group' audits in recipes. This entire section can be enabled or disabled with config. # # 1. 'control_group' DSL collects audits during Phase 2 # 2. Audits are run using RSpec # 3. Errors are collected and reported using the formatters # # @see #run_audits # @see Chef::Audit::Runner#run # # @raise [Chef::Exceptions::RunFailedWrappingError] If converge or audit failed. # # @see Chef::Config#enforce_path_sanity # @see Chef::Config#solo # @see Chef::Config#audit_mode # # @return Always returns true. # def run start_profiling run_error = nil runlock = RunLock.new(Chef::Config.lockfile) # TODO feels like acquire should have its own block arg for this runlock.acquire # don't add code that may fail before entering this section to be sure to release lock begin runlock.save_pid request_id = Chef::RequestID.instance.request_id run_context = nil events.run_start(Chef::VERSION) logger.info("*** Chef #{Chef::VERSION} ***") logger.info("Platform: #{RUBY_PLATFORM}") logger.info "Chef-client pid: #{Process.pid}" logger.debug("Chef-client request_id: #{request_id}") enforce_path_sanity run_ohai generate_guid register unless Chef::Config[:solo_legacy_mode] register_data_collector_reporter load_node build_node run_status.run_id = request_id run_status.start_clock logger.info("Starting Chef Run for #{node.name}") run_started do_windows_admin_check Chef.resource_handler_map.lock! Chef.provider_handler_map.lock! run_context = setup_run_context load_required_recipe(@rest, run_context) unless Chef::Config[:solo_legacy_mode] if Chef::Config[:audit_mode] != :audit_only converge_error = converge_and_save(run_context) end if Chef::Config[:why_run] == true # why_run should probably be renamed to why_converge logger.debug("Not running controls in 'why-run' mode - this mode is used to see potential converge changes") elsif Chef::Config[:audit_mode] != :disabled audit_error = run_audits(run_context) end # Raise converge_error so run_failed reporters/events are processed. raise converge_error if converge_error run_status.stop_clock logger.info("Chef Run complete in #{run_status.elapsed_time} seconds") run_completed_successfully events.run_completed(node, run_status) # keep this inside the main loop to get exception backtraces end_profiling # rebooting has to be the last thing we do, no exceptions. Chef::Platform::Rebooter.reboot_if_needed!(node) rescue Exception => run_error # CHEF-3336: Send the error first in case something goes wrong below and we don't know why logger.trace("Re-raising exception: #{run_error.class} - #{run_error.message}\n#{run_error.backtrace.join("\n ")}") # If we failed really early, we may not have a run_status yet. Too early for these to be of much use. if run_status run_status.stop_clock run_status.exception = run_error run_failed end events.run_failed(run_error, run_status) ensure Chef::RequestID.instance.reset_request_id @run_status = nil runlock.release end # Raise audit, converge, and other errors here so that we exit # with the proper exit status code and everything gets raised # as a RunFailedWrappingError if run_error || converge_error || audit_error error = if Chef::Config[:audit_mode] == :disabled run_error || converge_error else e = if run_error == converge_error Chef::Exceptions::RunFailedWrappingError.new(converge_error, audit_error) else Chef::Exceptions::RunFailedWrappingError.new(run_error, converge_error, audit_error) end e.fill_backtrace e end Chef::Application.debug_stacktrace(error) raise error end true end # # Private API # TODO make this stuff protected or private # # @api private def configure_formatters formatters_for_run.map do |formatter_name, output_path| if output_path.nil? Chef::Formatters.new(formatter_name, STDOUT_FD, STDERR_FD) else io = File.open(output_path, "a+") io.sync = true Chef::Formatters.new(formatter_name, io, io) end end end # @api private def formatters_for_run if Chef::Config.formatters.empty? [default_formatter] else Chef::Config.formatters end end # @api private def default_formatter if !Chef::Config[:force_logger] || Chef::Config[:force_formatter] [:doc] else [:null] end end # @api private def configure_event_loggers if Chef::Config.disable_event_logger [] else Chef::Config.event_loggers.map do |evt_logger| case evt_logger when Symbol Chef::EventLoggers.new(evt_logger) when Class evt_logger.new else end end end end # A rest object with validate_utf8 set to false. This will not throw exceptions # on non-UTF8 strings in JSON but will sanitize them so that e.g. POSTs will # never fail. Cannot be configured on a request-by-request basis, so we carry # around another rest object for it. # # @api private def rest_clean(client_name = node_name, config = Chef::Config) @rest_clean ||= Chef::ServerAPI.new(config[:chef_server_url], client_name: client_name, signing_key_filename: config[:client_key], validate_utf8: false) end # Resource reporters send event information back to the chef server for # processing. Can only be called after we have a @rest object # @api private def register_reporters [ Chef::ResourceReporter.new(rest_clean), Chef::Audit::AuditReporter.new(rest_clean), ].each do |r| events.register(r) end end # # Callback to fire notifications that the Chef run is starting # # @api private # def run_started self.class.run_start_notifications.each do |notification| notification.call(run_status) end events.run_started(run_status) end # # Callback to fire notifications that the run completed successfully # # @api private # def run_completed_successfully success_handlers = self.class.run_completed_successfully_notifications success_handlers.each do |notification| notification.call(run_status) end end # # Callback to fire notifications that the Chef run failed # # @api private # def run_failed failure_handlers = self.class.run_failed_notifications failure_handlers.each do |notification| notification.call(run_status) end end # # Instantiates a Chef::Node object, possibly loading the node's prior state # when using chef-client. Sets Chef.node to the new node. # # @return [Chef::Node] The node object for this Chef run # # @see Chef::PolicyBuilder#load_node # # @api private # def load_node policy_builder.load_node run_status.node = policy_builder.node Chef.set_node(policy_builder.node) node end # # Mutates the `node` object to prepare it for the chef run. # # @return [Chef::Node] The updated node object # # @see Chef::PolicyBuilder#build_node # # @api private # def build_node policy_builder.build_node run_status.node = node node end # # Sync cookbooks to local cache. # # TODO this appears to be unused. # # @see Chef::PolicyBuilder#sync_cookbooks # # @api private # def sync_cookbooks policy_builder.sync_cookbooks end # # Sets up the run context. # # @see Chef::PolicyBuilder#setup_run_context # # @return The newly set up run context # # @api private def setup_run_context run_context = policy_builder.setup_run_context(specific_recipes) assert_cookbook_path_not_empty(run_context) run_status.run_context = run_context run_context end # # Adds a required recipe as specified by the Chef Server # # @return The modified run context # # @api private # # TODO: @rest doesn't appear to be used anywhere outside # of client.register except for here. If it's common practice # to create your own rest client, perhaps we should do that # here but it seems more appropriate to reuse one that we # know is already created. for ease of testing, we'll pass # the existing rest client in as a parameter # def load_required_recipe(rest, run_context) required_recipe_contents = rest.get("required_recipe") logger.info("Required Recipe found, loading it") Chef::FileCache.store("required_recipe", required_recipe_contents) required_recipe_file = Chef::FileCache.load("required_recipe", false) # TODO: add integration tests with resource reporting turned on # (presumably requires changes to chef-zero) # # Chef::Recipe.new takes a cookbook name and a recipe name along # with the run context. These names are eventually used in the # resource reporter, and if the cookbook name cannot be found in the # cookbook collection then we will fail with an exception. Cases where # we currently also fail: # - specific recipes # - chef-apply would fail if resource reporting was enabled # recipe = Chef::Recipe.new(nil, nil, run_context) recipe.from_file(required_recipe_file) run_context rescue Net::HTTPClientException => e case e.response when Net::HTTPNotFound logger.trace("Required Recipe not configured on the server, skipping it") else raise end end # # The PolicyBuilder strategy for figuring out run list and cookbooks. # # @return [Chef::PolicyBuilder::Policyfile, Chef::PolicyBuilder::ExpandNodeObject] # # @api private # def policy_builder @policy_builder ||= Chef::PolicyBuilder::Dynamic.new(node_name, ohai.data, json_attribs, override_runlist, events) end # # Save the updated node to Chef. # # Does not save if we are in solo mode or using override_runlist. # # @see Chef::Node#save # @see Chef::Config#solo # # @api private # def save_updated_node if Chef::Config[:solo_legacy_mode] # nothing to do elsif policy_builder.temporary_policy? logger.warn("Skipping final node save because override_runlist was given") else logger.debug("Saving the current state of node #{node_name}") node.save end end # # Run ohai plugins. Runs all ohai plugins unless minimal_ohai is specified. # # Sends the ohai_completed event when finished. # # @see Chef::EventDispatcher# # @see Chef::Config#minimal_ohai # # @api private # def run_ohai filter = Chef::Config[:minimal_ohai] ? %w{fqdn machinename hostname platform platform_version ohai_time os os_version init_package} : nil ohai.all_plugins(filter) events.ohai_completed(node) rescue Ohai::Exceptions::CriticalPluginFailure => e logger.error("Critical Ohai plugins failed: #{e.message}") exit(false) end # # Figure out the node name we are working with. # # It tries these, in order: # - Chef::Config.node_name # - ohai[:fqdn] # - ohai[:machinename] # - ohai[:hostname] # # @raise [Chef::Exceptions::CannotDetermineNodeName] If the node name is not # set and cannot be determined via ohai. # # @see Chef::Config#node_name # # @api private # def node_name name = Chef::Config[:node_name] || ohai[:fqdn] || ohai[:machinename] || ohai[:hostname] Chef::Config[:node_name] = name raise Chef::Exceptions::CannotDetermineNodeName unless name name end # # Determine our private key and set up the connection to the Chef server. # # Skips registration and fires the `skipping_registration` event if # Chef::Config.client_key is unspecified or already exists. # # If Chef::Config.client_key does not exist, we register the client with the # Chef server and fire the registration_start and registration_completed events. # # @return [Chef::ServerAPI] The server connection object. # # @see Chef::Config#chef_server_url # @see Chef::Config#client_key # @see Chef::ApiClient::Registration#run # @see Chef::EventDispatcher#skipping_registration # @see Chef::EventDispatcher#registration_start # @see Chef::EventDispatcher#registration_completed # @see Chef::EventDispatcher#registration_failed # # @api private # def register(client_name = node_name, config = Chef::Config) if !config[:client_key] events.skipping_registration(client_name, config) logger.trace("Client key is unspecified - skipping registration") elsif File.exists?(config[:client_key]) events.skipping_registration(client_name, config) logger.trace("Client key #{config[:client_key]} is present - skipping registration") else events.registration_start(node_name, config) logger.info("Client key #{config[:client_key]} is not present - registering") Chef::ApiClient::Registration.new(node_name, config[:client_key]).run events.registration_completed end # We now have the client key, and should use it from now on. @rest = Chef::ServerAPI.new(config[:chef_server_url], client_name: client_name, signing_key_filename: config[:client_key]) # force initialization of the rest_clean API object rest_clean(client_name, config) register_reporters rescue Exception => e # TODO this should probably only ever fire if we *started* registration. # Move it to the block above. # TODO: munge exception so a semantic failure message can be given to the # user events.registration_failed(client_name, e, config) raise end # # Converges all compiled resources. # # Fires the converge_start, converge_complete and converge_failed events. # # If the exception `:end_client_run_early` is thrown during convergence, it # does not mark the run complete *or* failed, and returns `nil` # # @param run_context The run context. # # @return The thrown exception, if we are in audit mode. `nil` means the # converge was successful or ended early. # # @raise Any converge exception, unless we are in audit mode, in which case # we *return* the exception. # # @see Chef::Runner#converge # @see Chef::Config#audit_mode # @see Chef::EventDispatch#converge_start # @see Chef::EventDispatch#converge_complete # @see Chef::EventDispatch#converge_failed # # @api private # def converge(run_context) converge_exception = nil catch(:end_client_run_early) do begin events.converge_start(run_context) logger.debug("Converging node #{node_name}") @runner = Chef::Runner.new(run_context) @runner.converge events.converge_complete rescue Exception => e events.converge_failed(e) raise e if Chef::Config[:audit_mode] == :disabled converge_exception = e end end converge_exception end # # Converge the node via and then save it if successful. # # @param run_context The run context. # # @return The thrown exception, if we are in audit mode. `nil` means the # converge was successful or ended early. # # @raise Any converge or node save exception, unless we are in audit mode, # in which case we *return* the exception. # # @see #converge # @see #save_updated_mode # @see Chef::Config#audit_mode # # @api private # # We don't want to change the old API on the `converge` method to have it perform # saving. So we wrap it in this method. # TODO given this seems to be pretty internal stuff, how badly do we need to # split this stuff up? # def converge_and_save(run_context) converge_exception = converge(run_context) unless converge_exception begin save_updated_node rescue Exception => e raise e if Chef::Config[:audit_mode] == :disabled converge_exception = e end end converge_exception end # # Run the audit phase. # # Triggers the audit_phase_start, audit_phase_complete and # audit_phase_failed events. # # @param run_context The run context. # # @return Any thrown exceptions. `nil` if successful. # # @see Chef::Audit::Runner#run # @see Chef::EventDispatch#audit_phase_start # @see Chef::EventDispatch#audit_phase_complete # @see Chef::EventDispatch#audit_phase_failed # # @api private # def run_audits(run_context) begin events.audit_phase_start(run_status) logger.info("Starting audit phase") auditor = Chef::Audit::Runner.new(run_context) auditor.run if auditor.failed? audit_exception = Chef::Exceptions::AuditsFailed.new(auditor.num_failed, auditor.num_total) @events.audit_phase_failed(audit_exception, Chef::Audit::Logger.read_buffer) else @events.audit_phase_complete(Chef::Audit::Logger.read_buffer) end rescue Exception => e logger.error("Audit phase failed with error message: #{e.message}") @events.audit_phase_failed(e, Chef::Audit::Logger.read_buffer) audit_exception = e end audit_exception end # # Expands the run list. # # @return [Chef::RunListExpansion] The expanded run list. # # @see Chef::PolicyBuilder#expand_run_list # def expanded_run_list policy_builder.expand_run_list end # # Check if the user has Administrator privileges on windows. # # Throws an error if the user is not an admin, and # `Chef::Config.fatal_windows_admin_check` is true. # # @raise [Chef::Exceptions::WindowsNotAdmin] If the user is not an admin. # # @see Chef::platform#windows? # @see Chef::Config#fatal_windows_admin_check # # @api private # def do_windows_admin_check if Chef::Platform.windows? logger.trace("Checking for administrator privileges....") if !has_admin_privileges? message = "chef-client doesn't have administrator privileges on node #{node_name}." if Chef::Config[:fatal_windows_admin_check] logger.fatal(message) logger.fatal("fatal_windows_admin_check is set to TRUE.") raise Chef::Exceptions::WindowsNotAdmin, message else logger.warn("#{message} This might cause unexpected resource failures.") end else logger.trace("chef-client has administrator privileges on node #{node_name}.") end end end # Notification registration class<] # # @api private # def run_start_notifications @run_start_notifications ||= [] end # # Listeners to be run when the client run completes successfully. # # @return [Array] # # @api private # def run_completed_successfully_notifications @run_completed_successfully_notifications ||= [] end # # Listeners to be run when the client run fails. # # @return [Array] # # @api private # def run_failed_notifications @run_failed_notifications ||= [] end end # # IO stream that will be used as 'STDOUT' for formatters. Formatters are # configured during `initialize`, so this provides a convenience for # setting alternative IO stream during tests. # # @api private # STDOUT_FD = STDOUT # # IO stream that will be used as 'STDERR' for formatters. Formatters are # configured during `initialize`, so this provides a convenience for # setting alternative IO stream during tests. # # @api private # STDERR_FD = STDERR private attr_reader :override_runlist attr_reader :specific_recipes def profiling_prereqs! require "ruby-prof" rescue LoadError raise "You must have the ruby-prof gem installed in order to use --profile-ruby" end def start_profiling return unless Chef::Config[:profile_ruby] profiling_prereqs! RubyProf.start end def end_profiling return unless Chef::Config[:profile_ruby] profiling_prereqs! path = Chef::FileCache.create_cache_path("graph_profile.out", false) File.open(path, "w+") do |file| RubyProf::GraphPrinter.new(RubyProf.stop).print(file, {}) end logger.warn("Ruby execution profile dumped to #{path}") end def empty_directory?(path) !File.exists?(path) || (Dir.entries(path).size <= 2) end def is_last_element?(index, object) object.kind_of?(Array) ? index == object.size - 1 : true end def assert_cookbook_path_not_empty(run_context) if Chef::Config[:solo_legacy_mode] # Check for cookbooks in the path given # Chef::Config[:cookbook_path] can be a string or an array # if it's an array, go through it and check each one, raise error at the last one if no files are found cookbook_paths = Array(Chef::Config[:cookbook_path]) logger.trace "Loading from cookbook_path: #{cookbook_paths.map { |path| File.expand_path(path) }.join(', ')}" if cookbook_paths.all? { |path| empty_directory?(path) } msg = "None of the cookbook paths set in Chef::Config[:cookbook_path], #{cookbook_paths.inspect}, contain any cookbooks" logger.fatal(msg) raise Chef::Exceptions::CookbookNotFound, msg end else logger.warn("Node #{node_name} has an empty run list.") if run_context.node.run_list.empty? end end def has_admin_privileges? require "chef/win32/security" Chef::ReservedNames::Win32::Security.has_admin_privileges? end # Ensure that we have a GUID for this node # If we've got the proper configuration, we'll simply set that. # If we're registed with the data collector, we'll migrate that UUID into our configuration and use that # Otherwise, we'll create a new GUID and save it def generate_guid Chef::Config[:chef_guid] ||= if File.exists?(Chef::Config[:chef_guid_path]) File.read(Chef::Config[:chef_guid_path]) else uuid = UUIDFetcher.node_uuid File.open(Chef::Config[:chef_guid_path], "w+") do |fh| fh.write(uuid) end uuid end end class UUIDFetcher extend Chef::DataCollector::Messages::Helpers end # Register the data collector reporter to send event information to the # data collector server def register_data_collector_reporter events.register(Chef::DataCollector::Reporter.new) if Chef::DataCollector.register_reporter? end end end # HACK cannot load this first, but it must be loaded. require "chef/cookbook_loader" require "chef/cookbook_version" require "chef/cookbook/synchronizer"