diff options
Diffstat (limited to 'lib')
-rw-r--r-- | lib/unicorn.rb | 886 | ||||
-rw-r--r-- | lib/unicorn/app/exec_cgi.rb | 4 | ||||
-rw-r--r-- | lib/unicorn/configurator.rb | 31 | ||||
-rw-r--r-- | lib/unicorn/const.rb | 51 | ||||
-rw-r--r-- | lib/unicorn/http_request.rb | 112 | ||||
-rw-r--r-- | lib/unicorn/http_response.rb | 50 | ||||
-rw-r--r-- | lib/unicorn/http_server.rb | 695 | ||||
-rw-r--r-- | lib/unicorn/launcher.rb | 7 | ||||
-rw-r--r-- | lib/unicorn/preread_input.rb | 30 | ||||
-rw-r--r-- | lib/unicorn/socket_helper.rb | 8 | ||||
-rw-r--r-- | lib/unicorn/tee_input.rb | 126 | ||||
-rw-r--r-- | lib/unicorn/tmpio.rb | 29 | ||||
-rw-r--r-- | lib/unicorn/util.rb | 136 | ||||
-rw-r--r-- | lib/unicorn/worker.rb | 40 |
14 files changed, 1094 insertions, 1111 deletions
diff --git a/lib/unicorn.rb b/lib/unicorn.rb index 735354f..622dc6c 100644 --- a/lib/unicorn.rb +++ b/lib/unicorn.rb @@ -1,835 +1,83 @@ # -*- encoding: binary -*- - require 'fcntl' require 'etc' require 'stringio' require 'rack' -require 'unicorn/socket_helper' -require 'unicorn/const' -require 'unicorn/http_request' -require 'unicorn/configurator' -require 'unicorn/util' -require 'unicorn/tee_input' -require 'unicorn/http_response' +require 'kgio' -# Unicorn module containing all of the classes (include C extensions) for running -# a Unicorn web server. It contains a minimalist HTTP server with just enough -# functionality to service web application requests fast as possible. +# Unicorn module containing all of the classes (include C extensions) for +# running a Unicorn web server. It contains a minimalist HTTP server with just +# enough functionality to service web application requests fast as possible. module Unicorn - - # raised inside TeeInput when a client closes the socket inside the - # application dispatch. This is always raised with an empty backtrace - # since there is nothing in the application stack that is responsible - # for client shutdowns/disconnects. - class ClientShutdown < EOFError - end - - class << self - def run(app, options = {}) - HttpServer.new(app, options).start.join - end - - # This returns a lambda to pass in as the app, this does not "build" the - # app (which we defer based on the outcome of "preload_app" in the - # Unicorn config). The returned lambda will be called when it is - # time to build the app. - def builder(ru, opts) - # allow Configurator to parse cli switches embedded in the ru file - Unicorn::Configurator::RACKUP.update(:file => ru, :optparse => opts) - - # always called after config file parsing, may be called after forking - lambda do || - inner_app = case ru - when /\.ru$/ - raw = File.read(ru) - raw.sub!(/^__END__\n.*/, '') - eval("Rack::Builder.new {(#{raw}\n)}.to_app", TOPLEVEL_BINDING, ru) - else - require ru - Object.const_get(File.basename(ru, '.rb').capitalize) - end - - pp({ :inner_app => inner_app }) if $DEBUG - - # return value, matches rackup defaults based on env - case ENV["RACK_ENV"] - when "development" - Rack::Builder.new do - use Rack::CommonLogger, $stderr - use Rack::ShowExceptions - use Rack::Lint - run inner_app - end.to_app - when "deployment" - Rack::Builder.new do - use Rack::CommonLogger, $stderr - run inner_app - end.to_app - else - inner_app - end - end - end - - # returns an array of strings representing TCP listen socket addresses - # and Unix domain socket paths. This is useful for use with - # Raindrops::Middleware under Linux: http://raindrops.bogomips.org/ - def listener_names - HttpServer::LISTENERS.map { |io| SocketHelper.sock_name(io) } - end + def self.run(app, options = {}) + Unicorn::HttpServer.new(app, options).start.join end - # This is the process manager of Unicorn. This manages worker - # processes which in turn handle the I/O and application process. - # Listener sockets are started in the master process and shared with - # forked worker children. - - class HttpServer < Struct.new(:app, :timeout, :worker_processes, - :before_fork, :after_fork, :before_exec, - :logger, :pid, :listener_opts, :preload_app, - :reexec_pid, :orig_app, :init_listeners, - :master_pid, :config, :ready_pipe, :user) - include ::Unicorn::SocketHelper - - # prevents IO objects in here from being GC-ed - IO_PURGATORY = [] - - # all bound listener sockets - LISTENERS = [] - - # This hash maps PIDs to Workers - WORKERS = {} - - # We use SELF_PIPE differently in the master and worker processes: - # - # * The master process never closes or reinitializes this once - # initialized. Signal handlers in the master process will write to - # it to wake up the master from IO.select in exactly the same manner - # djb describes in http://cr.yp.to/docs/selfpipe.html - # - # * The workers immediately close the pipe they inherit from the - # master and replace it with a new pipe after forking. This new - # pipe is also used to wakeup from IO.select from inside (worker) - # signal handlers. However, workers *close* the pipe descriptors in - # the signal handlers to raise EBADF in IO.select instead of writing - # like we do in the master. We cannot easily use the reader set for - # IO.select because LISTENERS is already that set, and it's extra - # work (and cycles) to distinguish the pipe FD from the reader set - # once IO.select returns. So we're lazy and just close the pipe when - # a (rare) signal arrives in the worker and reinitialize the pipe later. - SELF_PIPE = [] - - # signal queue used for self-piping - SIG_QUEUE = [] - - # constant lookups are faster and we're single-threaded/non-reentrant - REQUEST = HttpRequest.new - - # We populate this at startup so we can figure out how to reexecute - # and upgrade the currently running instance of Unicorn - # This Hash is considered a stable interface and changing its contents - # will allow you to switch between different installations of Unicorn - # or even different installations of the same applications without - # downtime. Keys of this constant Hash are described as follows: - # - # * 0 - the path to the unicorn/unicorn_rails executable - # * :argv - a deep copy of the ARGV array the executable originally saw - # * :cwd - the working directory of the application, this is where - # you originally started Unicorn. - # - # To change your unicorn executable to a different path without downtime, - # you can set the following in your Unicorn config file, HUP and then - # continue with the traditional USR2 + QUIT upgrade steps: - # - # Unicorn::HttpServer::START_CTX[0] = "/home/bofh/1.9.2/bin/unicorn" - START_CTX = { - :argv => ARGV.map { |arg| arg.dup }, - :cwd => lambda { - # favor ENV['PWD'] since it is (usually) symlink aware for - # Capistrano and like systems - begin - a = File.stat(pwd = ENV['PWD']) - b = File.stat(Dir.pwd) - a.ino == b.ino && a.dev == b.dev ? pwd : Dir.pwd - rescue - Dir.pwd - end - }.call, - 0 => $0.dup, - } - - # This class and its members can be considered a stable interface - # and will not change in a backwards-incompatible fashion between - # releases of Unicorn. You may need to access it in the - # before_fork/after_fork hooks. See the Unicorn::Configurator RDoc - # for examples. - class Worker < Struct.new(:nr, :tmp, :switched) - - # worker objects may be compared to just plain numbers - def ==(other_nr) - self.nr == other_nr - end - - # Changes the worker process to the specified +user+ and +group+ - # This is only intended to be called from within the worker - # process from the +after_fork+ hook. This should be called in - # the +after_fork+ hook after any priviledged functions need to be - # run (e.g. to set per-worker CPU affinity, niceness, etc) - # - # Any and all errors raised within this method will be propagated - # directly back to the caller (usually the +after_fork+ hook. - # These errors commonly include ArgumentError for specifying an - # invalid user/group and Errno::EPERM for insufficient priviledges - def user(user, group = nil) - # we do not protect the caller, checking Process.euid == 0 is - # insufficient because modern systems have fine-grained - # capabilities. Let the caller handle any and all errors. - uid = Etc.getpwnam(user).uid - gid = Etc.getgrnam(group).gid if group - Unicorn::Util.chown_logs(uid, gid) - tmp.chown(uid, gid) - if gid && Process.egid != gid - Process.initgroups(user, gid) - Process::GID.change_privilege(gid) - end - Process.euid != uid and Process::UID.change_privilege(uid) - self.switched = true - end - - end - - # Creates a working server on host:port (strange things happen if - # port isn't a Number). Use HttpServer::run to start the server and - # HttpServer.run.join to join the thread that's processing - # incoming requests on the socket. - def initialize(app, options = {}) - self.app = app - self.reexec_pid = 0 - self.ready_pipe = options.delete(:ready_pipe) - self.init_listeners = options[:listeners] ? options[:listeners].dup : [] - self.config = Configurator.new(options.merge(:use_defaults => true)) - self.listener_opts = {} - - # we try inheriting listeners first, so we bind them later. - # we don't write the pid file until we've bound listeners in case - # unicorn was started twice by mistake. Even though our #pid= method - # checks for stale/existing pid files, race conditions are still - # possible (and difficult/non-portable to avoid) and can be likely - # to clobber the pid if the second start was in quick succession - # after the first, so we rely on the listener binding to fail in - # that case. Some tests (in and outside of this source tree) and - # monitoring tools may also rely on pid files existing before we - # attempt to connect to the listener(s) - config.commit!(self, :skip => [:listeners, :pid]) - self.orig_app = app - end - - # Runs the thing. Returns self so you can run join on it - def start - BasicSocket.do_not_reverse_lookup = true - - # inherit sockets from parents, they need to be plain Socket objects - # before they become UNIXServer or TCPServer - inherited = ENV['UNICORN_FD'].to_s.split(/,/).map do |fd| - io = Socket.for_fd(fd.to_i) - set_server_sockopt(io, listener_opts[sock_name(io)]) - IO_PURGATORY << io - logger.info "inherited addr=#{sock_name(io)} fd=#{fd}" - server_cast(io) - end - - config_listeners = config[:listeners].dup - LISTENERS.replace(inherited) - - # we start out with generic Socket objects that get cast to either - # TCPServer or UNIXServer objects; but since the Socket objects - # share the same OS-level file descriptor as the higher-level *Server - # objects; we need to prevent Socket objects from being garbage-collected - config_listeners -= listener_names - if config_listeners.empty? && LISTENERS.empty? - config_listeners << Unicorn::Const::DEFAULT_LISTEN - init_listeners << Unicorn::Const::DEFAULT_LISTEN - START_CTX[:argv] << "-l#{Unicorn::Const::DEFAULT_LISTEN}" - end - config_listeners.each { |addr| listen(addr) } - raise ArgumentError, "no listeners" if LISTENERS.empty? - - # this pipe is used to wake us up from select(2) in #join when signals - # are trapped. See trap_deferred. - init_self_pipe! - - # setup signal handlers before writing pid file in case people get - # trigger happy and send signals as soon as the pid file exists. - # Note that signals don't actually get handled until the #join method - QUEUE_SIGS.each { |sig| trap_deferred(sig) } - trap(:CHLD) { |_| awaken_master } - self.pid = config[:pid] - - self.master_pid = $$ - build_app! if preload_app - maintain_worker_count - self - end - - # replaces current listener set with +listeners+. This will - # close the socket if it will not exist in the new listener set - def listeners=(listeners) - cur_names, dead_names = [], [] - listener_names.each do |name| - if ?/ == name[0] - # mark unlinked sockets as dead so we can rebind them - (File.socket?(name) ? cur_names : dead_names) << name - else - cur_names << name - end - end - set_names = listener_names(listeners) - dead_names.concat(cur_names - set_names).uniq! - - LISTENERS.delete_if do |io| - if dead_names.include?(sock_name(io)) - IO_PURGATORY.delete_if do |pio| - pio.fileno == io.fileno && (pio.close rescue nil).nil? # true - end - (io.close rescue nil).nil? # true - else - set_server_sockopt(io, listener_opts[sock_name(io)]) - false - end - end - - (set_names - cur_names).each { |addr| listen(addr) } - end - - def stdout_path=(path); redirect_io($stdout, path); end - def stderr_path=(path); redirect_io($stderr, path); end - - def logger=(obj) - HttpRequest::DEFAULTS["rack.logger"] = super - end - - # sets the path for the PID file of the master process - def pid=(path) - if path - if x = valid_pid?(path) - return path if pid && path == pid && x == $$ - if x == reexec_pid && pid =~ /\.oldbin\z/ - logger.warn("will not set pid=#{path} while reexec-ed "\ - "child is running PID:#{x}") - return - end - raise ArgumentError, "Already running on PID:#{x} " \ - "(or pid=#{path} is stale)" - end - end - unlink_pid_safe(pid) if pid - - if path - fp = begin - tmp = "#{File.dirname(path)}/#{rand}.#$$" - File.open(tmp, File::RDWR|File::CREAT|File::EXCL, 0644) - rescue Errno::EEXIST - retry - end - fp.syswrite("#$$\n") - File.rename(fp.path, path) - fp.close - end - super(path) - end - - # add a given address to the +listeners+ set, idempotently - # Allows workers to add a private, per-process listener via the - # after_fork hook. Very useful for debugging and testing. - # +:tries+ may be specified as an option for the number of times - # to retry, and +:delay+ may be specified as the time in seconds - # to delay between retries. - # A negative value for +:tries+ indicates the listen will be - # retried indefinitely, this is useful when workers belonging to - # different masters are spawned during a transparent upgrade. - def listen(address, opt = {}.merge(listener_opts[address] || {})) - address = config.expand_addr(address) - return if String === address && listener_names.include?(address) - - delay = opt[:delay] || 0.5 - tries = opt[:tries] || 5 - begin - io = bind_listen(address, opt) - unless TCPServer === io || UNIXServer === io - IO_PURGATORY << io - io = server_cast(io) - end - logger.info "listening on addr=#{sock_name(io)} fd=#{io.fileno}" - LISTENERS << io - io - rescue Errno::EADDRINUSE => err - logger.error "adding listener failed addr=#{address} (in use)" - raise err if tries == 0 - tries -= 1 - logger.error "retrying in #{delay} seconds " \ - "(#{tries < 0 ? 'infinite' : tries} tries left)" - sleep(delay) - retry - rescue => err - logger.fatal "error adding listener addr=#{address}" - raise err - end - end - - # monitors children and receives signals forever - # (or until a termination signal is sent). This handles signals - # one-at-a-time time and we'll happily drop signals in case somebody - # is signalling us too often. - def join - respawn = true - last_check = Time.now - - proc_name 'master' - logger.info "master process ready" # test_exec.rb relies on this message - if ready_pipe - ready_pipe.syswrite($$.to_s) - ready_pipe.close rescue nil - self.ready_pipe = nil - end - begin - loop do - reap_all_workers - case SIG_QUEUE.shift - when nil - # avoid murdering workers after our master process (or the - # machine) comes out of suspend/hibernation - if (last_check + timeout) >= (last_check = Time.now) - murder_lazy_workers - else - # wait for workers to wakeup on suspend - master_sleep(timeout/2.0 + 1) - end - maintain_worker_count if respawn - master_sleep(1) - when :QUIT # graceful shutdown - break - when :TERM, :INT # immediate shutdown - stop(false) - break - when :USR1 # rotate logs - logger.info "master reopening logs..." - Unicorn::Util.reopen_logs - logger.info "master done reopening logs" - kill_each_worker(:USR1) - when :USR2 # exec binary, stay alive in case something went wrong - reexec - when :WINCH - if Process.ppid == 1 || Process.getpgrp != $$ - respawn = false - logger.info "gracefully stopping all workers" - kill_each_worker(:QUIT) - self.worker_processes = 0 - else - logger.info "SIGWINCH ignored because we're not daemonized" - end - when :TTIN - respawn = true - self.worker_processes += 1 - when :TTOU - self.worker_processes -= 1 if self.worker_processes > 0 - when :HUP - respawn = true - if config.config_file - load_config! - redo # immediate reaping since we may have QUIT workers - else # exec binary and exit if there's no config file - logger.info "config_file not present, reexecuting binary" - reexec - break - end - end - end - rescue Errno::EINTR - retry - rescue => e - logger.error "Unhandled master loop exception #{e.inspect}." - logger.error e.backtrace.join("\n") - retry - end - stop # gracefully shutdown all workers on our way out - logger.info "master complete" - unlink_pid_safe(pid) if pid - end - - # Terminates all workers, but does not exit master process - def stop(graceful = true) - self.listeners = [] - limit = Time.now + timeout - until WORKERS.empty? || Time.now > limit - kill_each_worker(graceful ? :QUIT : :TERM) - sleep(0.1) - reap_all_workers - end - kill_each_worker(:KILL) - end - - private - - # list of signals we care about and trap in master. - QUEUE_SIGS = [ :WINCH, :QUIT, :INT, :TERM, :USR1, :USR2, :HUP, - :TTIN, :TTOU ] - - # defer a signal for later processing in #join (master process) - def trap_deferred(signal) - trap(signal) do |sig_nr| - if SIG_QUEUE.size < 5 - SIG_QUEUE << signal - awaken_master - else - logger.error "ignoring SIG#{signal}, queue=#{SIG_QUEUE.inspect}" - end - end - end - - # wait for a signal hander to wake us up and then consume the pipe - # Wake up every second anyways to run murder_lazy_workers - def master_sleep(sec) - IO.select([ SELF_PIPE[0] ], nil, nil, sec) or return - SELF_PIPE[0].read_nonblock(Const::CHUNK_SIZE, HttpRequest::BUF) - rescue Errno::EAGAIN, Errno::EINTR - end - - def awaken_master - begin - SELF_PIPE[1].write_nonblock('.') # wakeup master process from select - rescue Errno::EAGAIN, Errno::EINTR - # pipe is full, master should wake up anyways - retry - end - end - - # reaps all unreaped workers - def reap_all_workers - begin - loop do - wpid, status = Process.waitpid2(-1, Process::WNOHANG) - wpid or break - if reexec_pid == wpid - logger.error "reaped #{status.inspect} exec()-ed" - self.reexec_pid = 0 - self.pid = pid.chomp('.oldbin') if pid - proc_name 'master' - else - worker = WORKERS.delete(wpid) and worker.tmp.close rescue nil - logger.info "reaped #{status.inspect} " \ - "worker=#{worker.nr rescue 'unknown'}" - end - end - rescue Errno::ECHILD - end - end - - # reexecutes the START_CTX with a new binary - def reexec - if reexec_pid > 0 - begin - Process.kill(0, reexec_pid) - logger.error "reexec-ed child already running PID:#{reexec_pid}" - return - rescue Errno::ESRCH - self.reexec_pid = 0 - end - end - - if pid - old_pid = "#{pid}.oldbin" - prev_pid = pid.dup - begin - self.pid = old_pid # clear the path for a new pid file - rescue ArgumentError - logger.error "old PID:#{valid_pid?(old_pid)} running with " \ - "existing pid=#{old_pid}, refusing rexec" - return - rescue => e - logger.error "error writing pid=#{old_pid} #{e.class} #{e.message}" - return - end - end - - self.reexec_pid = fork do - listener_fds = LISTENERS.map { |sock| sock.fileno } - ENV['UNICORN_FD'] = listener_fds.join(',') - Dir.chdir(START_CTX[:cwd]) - cmd = [ START_CTX[0] ].concat(START_CTX[:argv]) - - # avoid leaking FDs we don't know about, but let before_exec - # unset FD_CLOEXEC, if anything else in the app eventually - # relies on FD inheritence. - (3..1024).each do |io| - next if listener_fds.include?(io) - io = IO.for_fd(io) rescue nil - io or next - IO_PURGATORY << io - io.fcntl(Fcntl::F_SETFD, Fcntl::FD_CLOEXEC) - end - logger.info "executing #{cmd.inspect} (in #{Dir.pwd})" - before_exec.call(self) - exec(*cmd) - end - proc_name 'master (old)' - end - - # forcibly terminate all workers that haven't checked in in timeout - # seconds. The timeout is implemented using an unlinked File - # shared between the parent process and each worker. The worker - # runs File#chmod to modify the ctime of the File. If the ctime - # is stale for >timeout seconds, then we'll kill the corresponding - # worker. - def murder_lazy_workers - WORKERS.dup.each_pair do |wpid, worker| - stat = worker.tmp.stat - # skip workers that disable fchmod or have never fchmod-ed - stat.mode == 0100600 and next - (diff = (Time.now - stat.ctime)) <= timeout and next - logger.error "worker=#{worker.nr} PID:#{wpid} timeout " \ - "(#{diff}s > #{timeout}s), killing" - kill_worker(:KILL, wpid) # take no prisoners for timeout violations - end - end - - def spawn_missing_workers - (0...worker_processes).each do |worker_nr| - WORKERS.values.include?(worker_nr) and next - worker = Worker.new(worker_nr, Unicorn::Util.tmpio) - before_fork.call(self, worker) - WORKERS[fork { - ready_pipe.close if ready_pipe - self.ready_pipe = nil - worker_loop(worker) - }] = worker - end - end - - def maintain_worker_count - (off = WORKERS.size - worker_processes) == 0 and return - off < 0 and return spawn_missing_workers - WORKERS.dup.each_pair { |wpid,w| - w.nr >= worker_processes and kill_worker(:QUIT, wpid) rescue nil - } - end - - # if we get any error, try to write something back to the client - # assuming we haven't closed the socket, but don't get hung up - # if the socket is already closed or broken. We'll always ensure - # the socket is closed at the end of this function - def handle_error(client, e) - msg = case e - when EOFError,Errno::ECONNRESET,Errno::EPIPE,Errno::EINVAL,Errno::EBADF - Const::ERROR_500_RESPONSE - when HttpParserError # try to tell the client they're bad - Const::ERROR_400_RESPONSE + # This returns a lambda to pass in as the app, this does not "build" the + # app (which we defer based on the outcome of "preload_app" in the + # Unicorn config). The returned lambda will be called when it is + # time to build the app. + def self.builder(ru, opts) + # allow Configurator to parse cli switches embedded in the ru file + Unicorn::Configurator::RACKUP.update(:file => ru, :optparse => opts) + + # always called after config file parsing, may be called after forking + lambda do || + inner_app = case ru + when /\.ru$/ + raw = File.read(ru) + raw.sub!(/^__END__\n.*/, '') + eval("Rack::Builder.new {(#{raw}\n)}.to_app", TOPLEVEL_BINDING, ru) else - logger.error "Read error: #{e.inspect}" - logger.error e.backtrace.join("\n") - Const::ERROR_500_RESPONSE - end - client.write_nonblock(msg) - client.close - rescue - nil - end - - # once a client is accepted, it is processed in its entirety here - # in 3 easy steps: read request, call app, write app response - def process_client(client) - client.fcntl(Fcntl::F_SETFD, Fcntl::FD_CLOEXEC) - response = app.call(env = REQUEST.read(client)) - - if 100 == response[0].to_i - client.write(Const::EXPECT_100_RESPONSE) - env.delete(Const::HTTP_EXPECT) - response = app.call(env) - end - HttpResponse.write(client, response, HttpRequest::PARSER.headers?) - rescue => e - handle_error(client, e) - end - - # gets rid of stuff the worker has no business keeping track of - # to free some resources and drops all sig handlers. - # traps for USR1, USR2, and HUP may be set in the after_fork Proc - # by the user. - def init_worker_process(worker) - QUEUE_SIGS.each { |sig| trap(sig, nil) } - trap(:CHLD, 'DEFAULT') - SIG_QUEUE.clear - proc_name "worker[#{worker.nr}]" - START_CTX.clear - init_self_pipe! - WORKERS.values.each { |other| other.tmp.close rescue nil } - WORKERS.clear - LISTENERS.each { |sock| sock.fcntl(Fcntl::F_SETFD, Fcntl::FD_CLOEXEC) } - worker.tmp.fcntl(Fcntl::F_SETFD, Fcntl::FD_CLOEXEC) - after_fork.call(self, worker) # can drop perms - worker.user(*user) if user.kind_of?(Array) && ! worker.switched - self.timeout /= 2.0 # halve it for select() - build_app! unless preload_app - end - - def reopen_worker_logs(worker_nr) - logger.info "worker=#{worker_nr} reopening logs..." - Unicorn::Util.reopen_logs - logger.info "worker=#{worker_nr} done reopening logs" - init_self_pipe! - end - - # runs inside each forked worker, this sits around and waits - # for connections and doesn't die until the parent dies (or is - # given a INT, QUIT, or TERM signal) - def worker_loop(worker) - ppid = master_pid - init_worker_process(worker) - nr = 0 # this becomes negative if we need to reopen logs - alive = worker.tmp # tmp is our lifeline to the master process - ready = LISTENERS - - # closing anything we IO.select on will raise EBADF - trap(:USR1) { nr = -65536; SELF_PIPE[0].close rescue nil } - trap(:QUIT) { alive = nil; LISTENERS.each { |s| s.close rescue nil } } - [:TERM, :INT].each { |sig| trap(sig) { exit!(0) } } # instant shutdown - logger.info "worker=#{worker.nr} ready" - m = 0 - - begin - nr < 0 and reopen_worker_logs(worker.nr) - nr = 0 - - # we're a goner in timeout seconds anyways if alive.chmod - # breaks, so don't trap the exception. Using fchmod() since - # futimes() is not available in base Ruby and I very strongly - # prefer temporary files to be unlinked for security, - # performance and reliability reasons, so utime is out. No-op - # changes with chmod doesn't update ctime on all filesystems; so - # we change our counter each and every time (after process_client - # and before IO.select). - alive.chmod(m = 0 == m ? 1 : 0) - - ready.each do |sock| - begin - process_client(sock.accept_nonblock) - nr += 1 - alive.chmod(m = 0 == m ? 1 : 0) - rescue Errno::EAGAIN, Errno::ECONNABORTED - end - break if nr < 0 - end - - # make the following bet: if we accepted clients this round, - # we're probably reasonably busy, so avoid calling select() - # and do a speculative accept_nonblock on ready listeners - # before we sleep again in select(). - redo unless nr == 0 # (nr < 0) => reopen logs - - ppid == Process.ppid or return - alive.chmod(m = 0 == m ? 1 : 0) - begin - # timeout used so we can detect parent death: - ret = IO.select(LISTENERS, nil, SELF_PIPE, timeout) or redo - ready = ret[0] - rescue Errno::EINTR - ready = LISTENERS - rescue Errno::EBADF - nr < 0 or return - end - rescue => e - if alive - logger.error "Unhandled listen loop exception #{e.inspect}." - logger.error e.backtrace.join("\n") - end - end while alive - end - - # delivers a signal to a worker and fails gracefully if the worker - # is no longer running. - def kill_worker(signal, wpid) - begin - Process.kill(signal, wpid) - rescue Errno::ESRCH - worker = WORKERS.delete(wpid) and worker.tmp.close rescue nil - end - end - - # delivers a signal to each worker - def kill_each_worker(signal) - WORKERS.keys.each { |wpid| kill_worker(signal, wpid) } - end - - # unlinks a PID file at given +path+ if it contains the current PID - # still potentially racy without locking the directory (which is - # non-portable and may interact badly with other programs), but the - # window for hitting the race condition is small - def unlink_pid_safe(path) - (File.read(path).to_i == $$ and File.unlink(path)) rescue nil - end - - # returns a PID if a given path contains a non-stale PID file, - # nil otherwise. - def valid_pid?(path) - wpid = File.read(path).to_i - wpid <= 0 and return nil - begin - Process.kill(0, wpid) - wpid - rescue Errno::ESRCH - # don't unlink stale pid files, racy without non-portable locking... - end - rescue Errno::ENOENT - end - - def load_config! - loaded_app = app - begin - logger.info "reloading config_file=#{config.config_file}" - config[:listeners].replace(init_listeners) - config.reload - config.commit!(self) - kill_each_worker(:QUIT) - Unicorn::Util.reopen_logs - self.app = orig_app - build_app! if preload_app - logger.info "done reloading config_file=#{config.config_file}" - rescue StandardError, LoadError, SyntaxError => e - logger.error "error reloading config_file=#{config.config_file}: " \ - "#{e.class} #{e.message} #{e.backtrace}" - self.app = loaded_app - end - end - - # returns an array of string names for the given listener array - def listener_names(listeners = LISTENERS) - listeners.map { |io| sock_name(io) } - end - - def build_app! - if app.respond_to?(:arity) && app.arity == 0 - if defined?(Gem) && Gem.respond_to?(:refresh) - logger.info "Refreshing Gem list" - Gem.refresh - end - self.app = app.call + require ru + Object.const_get(File.basename(ru, '.rb').capitalize) + end + + pp({ :inner_app => inner_app }) if $DEBUG + + # return value, matches rackup defaults based on env + case ENV["RACK_ENV"] + when "development" + Rack::Builder.new do + use Rack::CommonLogger, $stderr + use Rack::ShowExceptions + use Rack::Lint + run inner_app + end.to_app + when "deployment" + Rack::Builder.new do + use Rack::CommonLogger, $stderr + run inner_app + end.to_app + else + inner_app end end + end - def proc_name(tag) - $0 = ([ File.basename(START_CTX[0]), tag - ]).concat(START_CTX[:argv]).join(' ') - end - - def redirect_io(io, path) - File.open(path, 'ab') { |fp| io.reopen(fp) } if path - io.sync = true - end - - def init_self_pipe! - SELF_PIPE.each { |io| io.close rescue nil } - SELF_PIPE.replace(IO.pipe) - SELF_PIPE.each { |io| io.fcntl(Fcntl::F_SETFD, Fcntl::FD_CLOEXEC) } + # returns an array of strings representing TCP listen socket addresses + # and Unix domain socket paths. This is useful for use with + # Raindrops::Middleware under Linux: http://raindrops.bogomips.org/ + def self.listener_names + Unicorn::HttpServer::LISTENERS.map do |io| + Unicorn::SocketHelper.sock_name(io) end - end end + +# raised inside TeeInput when a client closes the socket inside the +# application dispatch. This is always raised with an empty backtrace +# since there is nothing in the application stack that is responsible +# for client shutdowns/disconnects. +class Unicorn::ClientShutdown < EOFError; end + +require 'unicorn/const' +require 'unicorn/socket_helper' +require 'unicorn/tee_input' +require 'unicorn/http_request' +require 'unicorn/configurator' +require 'unicorn/tmpio' +require 'unicorn/util' +require 'unicorn/http_response' +require 'unicorn/worker' +require 'unicorn/http_server' diff --git a/lib/unicorn/app/exec_cgi.rb b/lib/unicorn/app/exec_cgi.rb index 412c1d9..fea22f6 100644 --- a/lib/unicorn/app/exec_cgi.rb +++ b/lib/unicorn/app/exec_cgi.rb @@ -43,7 +43,7 @@ module Unicorn::App # Calls the app def call(env) - out, err = Unicorn::Util.tmpio, Unicorn::Util.tmpio + out, err = Unicorn::TmpIO.new, Unicorn::TmpIO.new inp = force_file_input(env) pid = fork { run_child(inp, out, err, env) } inp.close @@ -124,7 +124,7 @@ module Unicorn::App if inp.respond_to?(:size) && inp.size == 0 ::File.open('/dev/null', 'rb') else - tmp = Unicorn::Util.tmpio + tmp = Unicorn::TmpIO.new buf = inp.read(CHUNK_SIZE) begin diff --git a/lib/unicorn/configurator.rb b/lib/unicorn/configurator.rb index fb37c56..dd515a7 100644 --- a/lib/unicorn/configurator.rb +++ b/lib/unicorn/configurator.rb @@ -8,14 +8,22 @@ require 'logger' # example configuration files. An example config file for use with # nginx is also available at # http://unicorn.bogomips.org/examples/nginx.conf -class Unicorn::Configurator < Struct.new(:set, :config_file, :after_reload) +class Unicorn::Configurator + attr_accessor :set, :config_file, :after_reload + + # :stopdoc: # used to stash stuff for deferred processing of cli options in # config.ru after "working_directory" is bound. Do not rely on # this being around later on... - RACKUP = {} # :nodoc: + RACKUP = { + :daemonize => false, + :host => Unicorn::Const::DEFAULT_HOST, + :port => Unicorn::Const::DEFAULT_PORT, + :set_listener => false, + :options => { :listeners => [] } + } # Default settings for Unicorn - # :stopdoc: DEFAULTS = { :timeout => 60, :logger => Logger.new($stderr), @@ -58,6 +66,9 @@ class Unicorn::Configurator < Struct.new(:set, :config_file, :after_reload) parse_rackup_file + RACKUP[:set_listener] and + set[:listeners] << "#{RACKUP[:host]}:#{RACKUP[:port]}" + # unicorn_rails creates dirs here after working_directory is bound after_reload.call if after_reload @@ -495,23 +506,15 @@ private /^#\\(.*)/ =~ File.read(ru) or return RACKUP[:optparse].parse!($1.split(/\s+/)) - # XXX ugly as hell, WILL FIX in 2.x (along with Rainbows!/Zbatery) - host, port, set_listener, options, daemonize = - eval("[ host, port, set_listener, options, daemonize ]", - TOPLEVEL_BINDING) - - # XXX duplicate code from bin/unicorn{,_rails} - set[:listeners] << "#{host}:#{port}" if set_listener - - if daemonize + if RACKUP[:daemonize] # unicorn_rails wants a default pid path, (not plain 'unicorn') if after_reload spid = set[:pid] pid('tmp/pids/unicorn.pid') if spid.nil? || spid == :unset end unless RACKUP[:daemonized] - Unicorn::Launcher.daemonize!(options) - RACKUP[:ready_pipe] = options.delete(:ready_pipe) + Unicorn::Launcher.daemonize!(RACKUP[:options]) + RACKUP[:ready_pipe] = RACKUP[:options].delete(:ready_pipe) end end end diff --git a/lib/unicorn/const.rb b/lib/unicorn/const.rb index 2f0b106..e7908d6 100644 --- a/lib/unicorn/const.rb +++ b/lib/unicorn/const.rb @@ -1,36 +1,37 @@ # -*- encoding: binary -*- -module Unicorn +# Frequently used constants when constructing requests or responses. +# Many times the constant just refers to a string with the same +# contents. Using these constants gave about a 3% to 10% performance +# improvement over using the strings directly. Symbols did not really +# improve things much compared to constants. +module Unicorn::Const - # Frequently used constants when constructing requests or responses. Many times - # the constant just refers to a string with the same contents. Using these constants - # gave about a 3% to 10% performance improvement over using the strings directly. - # Symbols did not really improve things much compared to constants. - module Const + # The current version of Unicorn, currently 2.0.0pre3 + UNICORN_VERSION = "2.0.0pre3" - # The current version of Unicorn, currently 1.1.5 - UNICORN_VERSION="1.1.5" + # default TCP listen host address (0.0.0.0, all interfaces) + DEFAULT_HOST = "0.0.0.0" - DEFAULT_HOST = "0.0.0.0" # default TCP listen host address - DEFAULT_PORT = 8080 # default TCP listen port - DEFAULT_LISTEN = "#{DEFAULT_HOST}:#{DEFAULT_PORT}" + # default TCP listen port (8080) + DEFAULT_PORT = 8080 - # The basic max request size we'll try to read. - CHUNK_SIZE=(16 * 1024) + # default TCP listen address and port (0.0.0.0:8080) + DEFAULT_LISTEN = "#{DEFAULT_HOST}:#{DEFAULT_PORT}" - # Maximum request body size before it is moved out of memory and into a - # temporary file for reading (112 kilobytes). - MAX_BODY=1024 * 112 + # The basic request body size we'll try to read at once (16 kilobytes). + CHUNK_SIZE = 16 * 1024 - # common errors we'll send back - ERROR_400_RESPONSE = "HTTP/1.1 400 Bad Request\r\n\r\n" - ERROR_500_RESPONSE = "HTTP/1.1 500 Internal Server Error\r\n\r\n" - EXPECT_100_RESPONSE = "HTTP/1.1 100 Continue\r\n\r\n" + # Maximum request body size before it is moved out of memory and into a + # temporary file for reading (112 kilobytes). + MAX_BODY = 1024 * 112 - # A frozen format for this is about 15% faster - REMOTE_ADDR="REMOTE_ADDR".freeze - RACK_INPUT="rack.input".freeze - HTTP_EXPECT="HTTP_EXPECT" - end + # :stopdoc: + # common errors we'll send back + ERROR_400_RESPONSE = "HTTP/1.1 400 Bad Request\r\n\r\n" + ERROR_500_RESPONSE = "HTTP/1.1 500 Internal Server Error\r\n\r\n" + EXPECT_100_RESPONSE = "HTTP/1.1 100 Continue\r\n\r\n" + HTTP_EXPECT = "HTTP_EXPECT" + # :startdoc: end diff --git a/lib/unicorn/http_request.rb b/lib/unicorn/http_request.rb index 65870ed..2dcd839 100644 --- a/lib/unicorn/http_request.rb +++ b/lib/unicorn/http_request.rb @@ -2,70 +2,68 @@ require 'unicorn_http' -module Unicorn - class HttpRequest +# TODO: remove redundant names +Unicorn.const_set(:HttpRequest, Unicorn::HttpParser) +class Unicorn::HttpParser - # default parameters we merge into the request env for Rack handlers - DEFAULTS = { - "rack.errors" => $stderr, - "rack.multiprocess" => true, - "rack.multithread" => false, - "rack.run_once" => false, - "rack.version" => [1, 1], - "SCRIPT_NAME" => "", + # default parameters we merge into the request env for Rack handlers + DEFAULTS = { + "rack.errors" => $stderr, + "rack.multiprocess" => true, + "rack.multithread" => false, + "rack.run_once" => false, + "rack.version" => [1, 1], + "SCRIPT_NAME" => "", - # this is not in the Rack spec, but some apps may rely on it - "SERVER_SOFTWARE" => "Unicorn #{Const::UNICORN_VERSION}" - } + # this is not in the Rack spec, but some apps may rely on it + "SERVER_SOFTWARE" => "Unicorn #{Unicorn::Const::UNICORN_VERSION}" + } - NULL_IO = StringIO.new("") - LOCALHOST = '127.0.0.1' + NULL_IO = StringIO.new("") - # Being explicitly single-threaded, we have certain advantages in - # not having to worry about variables being clobbered :) - BUF = "" - PARSER = HttpParser.new - REQ = {} + # :stopdoc: + # A frozen format for this is about 15% faster + REMOTE_ADDR = 'REMOTE_ADDR'.freeze + RACK_INPUT = 'rack.input'.freeze + TeeInput = Unicorn::TeeInput + # :startdoc: - # Does the majority of the IO processing. It has been written in - # Ruby using about 8 different IO processing strategies. - # - # It is currently carefully constructed to make sure that it gets - # the best possible performance for the common case: GET requests - # that are fully complete after a single read(2) - # - # Anyone who thinks they can make it faster is more than welcome to - # take a crack at it. - # - # returns an environment hash suitable for Rack if successful - # This does minimal exception trapping and it is up to the caller - # to handle any socket errors (e.g. user aborted upload). - def read(socket) - REQ.clear - PARSER.reset + # Does the majority of the IO processing. It has been written in + # Ruby using about 8 different IO processing strategies. + # + # It is currently carefully constructed to make sure that it gets + # the best possible performance for the common case: GET requests + # that are fully complete after a single read(2) + # + # Anyone who thinks they can make it faster is more than welcome to + # take a crack at it. + # + # returns an environment hash suitable for Rack if successful + # This does minimal exception trapping and it is up to the caller + # to handle any socket errors (e.g. user aborted upload). + def read(socket) + reset + e = env - # From http://www.ietf.org/rfc/rfc3875: - # "Script authors should be aware that the REMOTE_ADDR and - # REMOTE_HOST meta-variables (see sections 4.1.8 and 4.1.9) - # may not identify the ultimate source of the request. They - # identify the client for the immediate request to the server; - # that client may be a proxy, gateway, or other intermediary - # acting on behalf of the actual source client." - REQ[Const::REMOTE_ADDR] = - TCPSocket === socket ? socket.peeraddr[-1] : LOCALHOST + # From http://www.ietf.org/rfc/rfc3875: + # "Script authors should be aware that the REMOTE_ADDR and + # REMOTE_HOST meta-variables (see sections 4.1.8 and 4.1.9) + # may not identify the ultimate source of the request. They + # identify the client for the immediate request to the server; + # that client may be a proxy, gateway, or other intermediary + # acting on behalf of the actual source client." + e[REMOTE_ADDR] = socket.kgio_addr - # short circuit the common case with small GET requests first - if PARSER.headers(REQ, socket.readpartial(Const::CHUNK_SIZE, BUF)).nil? - # Parser is not done, queue up more data to read and continue parsing - # an Exception thrown from the PARSER will throw us out of the loop - begin - BUF << socket.readpartial(Const::CHUNK_SIZE) - end while PARSER.headers(REQ, BUF).nil? - end - REQ[Const::RACK_INPUT] = 0 == PARSER.content_length ? - NULL_IO : Unicorn::TeeInput.new(socket, REQ, PARSER, BUF) - REQ.update(DEFAULTS) + # short circuit the common case with small GET requests first + socket.kgio_read!(16384, buf) + if parse.nil? + # Parser is not done, queue up more data to read and continue parsing + # an Exception thrown from the parser will throw us out of the loop + begin + buf << socket.kgio_read!(16384) + end while parse.nil? end - + e[RACK_INPUT] = 0 == content_length ? NULL_IO : TeeInput.new(socket, self) + e.merge!(DEFAULTS) end end diff --git a/lib/unicorn/http_response.rb b/lib/unicorn/http_response.rb index 6f1cd48..5725e25 100644 --- a/lib/unicorn/http_response.rb +++ b/lib/unicorn/http_response.rb @@ -5,19 +5,12 @@ require 'time' # You use it by simply doing: # # status, headers, body = rack_app.call(env) -# HttpResponse.write(socket, [ status, headers, body ]) +# http_response_write(socket, [ status, headers, body ]) # # Most header correctness (including Content-Length and Content-Type) -# is the job of Rack, with the exception of the "Connection: close" -# and "Date" headers. +# is the job of Rack, with the exception of the "Date" and "Status" header. # -# A design decision was made to force the client to not pipeline or -# keepalive requests. HTTP/1.1 pipelining really kills the -# performance due to how it has to be handled and how unclear the -# standard is. To fix this the HttpResponse always gives a -# "Connection: close" header which forces the client to close right -# away. The bonus for this is that it gives a pretty nice speed boost -# to most clients since they can close their connection immediately. +# TODO: allow keepalive module Unicorn::HttpResponse # Every standard HTTP code mapped to the appropriate message. @@ -25,41 +18,28 @@ module Unicorn::HttpResponse hash[code] = "#{code} #{msg}" hash } - - # Rack does not set/require a Date: header. We always override the - # Connection: and Date: headers no matter what (if anything) our - # Rack application sent us. - SKIP = { 'connection' => true, 'date' => true, 'status' => true } + CRLF = "\r\n" # writes the rack_response to socket as an HTTP response - def self.write(socket, rack_response, have_header = true) + def http_response_write(socket, rack_response) status, headers, body = rack_response + status = CODES[status.to_i] || status - if have_header - status = CODES[status.to_i] || status - out = [] - - # Don't bother enforcing duplicate supression, it's a Hash most of - # the time anyways so just hope our app knows what it's doing + if headers + buf = "HTTP/1.1 #{status}\r\n" \ + "Date: #{Time.now.httpdate}\r\n" \ + "Status: #{status}\r\n" \ + "Connection: close\r\n" headers.each do |key, value| - next if SKIP.include?(key.downcase) + next if %r{\A(?:Date\z|Status\z|Connection\z)}i =~ key if value =~ /\n/ # avoiding blank, key-only cookies with /\n+/ - out.concat(value.split(/\n+/).map! { |v| "#{key}: #{v}\r\n" }) + buf << value.split(/\n+/).map! { |v| "#{key}: #{v}\r\n" }.join('') else - out << "#{key}: #{value}\r\n" + buf << "#{key}: #{value}\r\n" end end - - # Rack should enforce Content-Length or chunked transfer encoding, - # so don't worry or care about them. - # Date is required by HTTP/1.1 as long as our clock can be trusted. - # Some broken clients require a "Status" header so we accomodate them - socket.write("HTTP/1.1 #{status}\r\n" \ - "Date: #{Time.now.httpdate}\r\n" \ - "Status: #{status}\r\n" \ - "Connection: close\r\n" \ - "#{out.join('')}\r\n") + socket.write(buf << CRLF) end body.each { |chunk| socket.write(chunk) } diff --git a/lib/unicorn/http_server.rb b/lib/unicorn/http_server.rb new file mode 100644 index 0000000..69b7cc8 --- /dev/null +++ b/lib/unicorn/http_server.rb @@ -0,0 +1,695 @@ +# -*- encoding: binary -*- + +# This is the process manager of Unicorn. This manages worker +# processes which in turn handle the I/O and application process. +# Listener sockets are started in the master process and shared with +# forked worker children. +class Unicorn::HttpServer + attr_accessor :app, :request, :timeout, :worker_processes, + :before_fork, :after_fork, :before_exec, + :listener_opts, :preload_app, + :reexec_pid, :orig_app, :init_listeners, + :master_pid, :config, :ready_pipe, :user + attr_reader :pid, :logger + + # :stopdoc: + include Unicorn::SocketHelper + include Unicorn::HttpResponse + + # backwards compatibility with 1.x + Worker = Unicorn::Worker + + # prevents IO objects in here from being GC-ed + IO_PURGATORY = [] + + # all bound listener sockets + LISTENERS = [] + + # This hash maps PIDs to Workers + WORKERS = {} + + # We use SELF_PIPE differently in the master and worker processes: + # + # * The master process never closes or reinitializes this once + # initialized. Signal handlers in the master process will write to + # it to wake up the master from IO.select in exactly the same manner + # djb describes in http://cr.yp.to/docs/selfpipe.html + # + # * The workers immediately close the pipe they inherit from the + # master and replace it with a new pipe after forking. This new + # pipe is also used to wakeup from IO.select from inside (worker) + # signal handlers. However, workers *close* the pipe descriptors in + # the signal handlers to raise EBADF in IO.select instead of writing + # like we do in the master. We cannot easily use the reader set for + # IO.select because LISTENERS is already that set, and it's extra + # work (and cycles) to distinguish the pipe FD from the reader set + # once IO.select returns. So we're lazy and just close the pipe when + # a (rare) signal arrives in the worker and reinitialize the pipe later. + SELF_PIPE = [] + + # signal queue used for self-piping + SIG_QUEUE = [] + + # list of signals we care about and trap in master. + QUEUE_SIGS = [ :WINCH, :QUIT, :INT, :TERM, :USR1, :USR2, :HUP, :TTIN, :TTOU ] + + # :startdoc: + # We populate this at startup so we can figure out how to reexecute + # and upgrade the currently running instance of Unicorn + # This Hash is considered a stable interface and changing its contents + # will allow you to switch between different installations of Unicorn + # or even different installations of the same applications without + # downtime. Keys of this constant Hash are described as follows: + # + # * 0 - the path to the unicorn/unicorn_rails executable + # * :argv - a deep copy of the ARGV array the executable originally saw + # * :cwd - the working directory of the application, this is where + # you originally started Unicorn. + # + # To change your unicorn executable to a different path without downtime, + # you can set the following in your Unicorn config file, HUP and then + # continue with the traditional USR2 + QUIT upgrade steps: + # + # Unicorn::HttpServer::START_CTX[0] = "/home/bofh/1.9.2/bin/unicorn" + START_CTX = { + :argv => ARGV.map { |arg| arg.dup }, + :cwd => lambda { + # favor ENV['PWD'] since it is (usually) symlink aware for + # Capistrano and like systems + begin + a = File.stat(pwd = ENV['PWD']) + b = File.stat(Dir.pwd) + a.ino == b.ino && a.dev == b.dev ? pwd : Dir.pwd + rescue + Dir.pwd + end + }.call, + 0 => $0.dup, + } + + # Creates a working server on host:port (strange things happen if + # port isn't a Number). Use HttpServer::run to start the server and + # HttpServer.run.join to join the thread that's processing + # incoming requests on the socket. + def initialize(app, options = {}) + @app = app + @request = Unicorn::HttpRequest.new + self.reexec_pid = 0 + options = options.dup + self.ready_pipe = options.delete(:ready_pipe) + self.init_listeners = options[:listeners] ? options[:listeners].dup : [] + options[:use_defaults] = true + self.config = Unicorn::Configurator.new(options) + self.listener_opts = {} + + # we try inheriting listeners first, so we bind them later. + # we don't write the pid file until we've bound listeners in case + # unicorn was started twice by mistake. Even though our #pid= method + # checks for stale/existing pid files, race conditions are still + # possible (and difficult/non-portable to avoid) and can be likely + # to clobber the pid if the second start was in quick succession + # after the first, so we rely on the listener binding to fail in + # that case. Some tests (in and outside of this source tree) and + # monitoring tools may also rely on pid files existing before we + # attempt to connect to the listener(s) + config.commit!(self, :skip => [:listeners, :pid]) + self.orig_app = app + end + + # Runs the thing. Returns self so you can run join on it + def start + BasicSocket.do_not_reverse_lookup = true + + # inherit sockets from parents, they need to be plain Socket objects + # before they become Kgio::UNIXServer or Kgio::TCPServer + inherited = ENV['UNICORN_FD'].to_s.split(/,/).map do |fd| + io = Socket.for_fd(fd.to_i) + set_server_sockopt(io, listener_opts[sock_name(io)]) + IO_PURGATORY << io + logger.info "inherited addr=#{sock_name(io)} fd=#{fd}" + server_cast(io) + end + + config_listeners = config[:listeners].dup + LISTENERS.replace(inherited) + + # we start out with generic Socket objects that get cast to either + # Kgio::TCPServer or Kgio::UNIXServer objects; but since the Socket + # objects share the same OS-level file descriptor as the higher-level + # *Server objects; we need to prevent Socket objects from being + # garbage-collected + config_listeners -= listener_names + if config_listeners.empty? && LISTENERS.empty? + config_listeners << Unicorn::Const::DEFAULT_LISTEN + init_listeners << Unicorn::Const::DEFAULT_LISTEN + START_CTX[:argv] << "-l#{Unicorn::Const::DEFAULT_LISTEN}" + end + config_listeners.each { |addr| listen(addr) } + raise ArgumentError, "no listeners" if LISTENERS.empty? + + # this pipe is used to wake us up from select(2) in #join when signals + # are trapped. See trap_deferred. + init_self_pipe! + + # setup signal handlers before writing pid file in case people get + # trigger happy and send signals as soon as the pid file exists. + # Note that signals don't actually get handled until the #join method + QUEUE_SIGS.each { |sig| trap(sig) { SIG_QUEUE << sig; awaken_master } } + trap(:CHLD) { awaken_master } + self.pid = config[:pid] + + self.master_pid = $$ + build_app! if preload_app + maintain_worker_count + self + end + + # replaces current listener set with +listeners+. This will + # close the socket if it will not exist in the new listener set + def listeners=(listeners) + cur_names, dead_names = [], [] + listener_names.each do |name| + if ?/ == name[0] + # mark unlinked sockets as dead so we can rebind them + (File.socket?(name) ? cur_names : dead_names) << name + else + cur_names << name + end + end + set_names = listener_names(listeners) + dead_names.concat(cur_names - set_names).uniq! + + LISTENERS.delete_if do |io| + if dead_names.include?(sock_name(io)) + IO_PURGATORY.delete_if do |pio| + pio.fileno == io.fileno && (pio.close rescue nil).nil? # true + end + (io.close rescue nil).nil? # true + else + set_server_sockopt(io, listener_opts[sock_name(io)]) + false + end + end + + (set_names - cur_names).each { |addr| listen(addr) } + end + + def stdout_path=(path); redirect_io($stdout, path); end + def stderr_path=(path); redirect_io($stderr, path); end + + def logger=(obj) + Unicorn::HttpRequest::DEFAULTS["rack.logger"] = @logger = obj + end + + # sets the path for the PID file of the master process + def pid=(path) + if path + if x = valid_pid?(path) + return path if pid && path == pid && x == $$ + if x == reexec_pid && pid =~ /\.oldbin\z/ + logger.warn("will not set pid=#{path} while reexec-ed "\ + "child is running PID:#{x}") + return + end + raise ArgumentError, "Already running on PID:#{x} " \ + "(or pid=#{path} is stale)" + end + end + unlink_pid_safe(pid) if pid + + if path + fp = begin + tmp = "#{File.dirname(path)}/#{rand}.#$$" + File.open(tmp, File::RDWR|File::CREAT|File::EXCL, 0644) + rescue Errno::EEXIST + retry + end + fp.syswrite("#$$\n") + File.rename(fp.path, path) + fp.close + end + @pid = path + end + + # add a given address to the +listeners+ set, idempotently + # Allows workers to add a private, per-process listener via the + # after_fork hook. Very useful for debugging and testing. + # +:tries+ may be specified as an option for the number of times + # to retry, and +:delay+ may be specified as the time in seconds + # to delay between retries. + # A negative value for +:tries+ indicates the listen will be + # retried indefinitely, this is useful when workers belonging to + # different masters are spawned during a transparent upgrade. + def listen(address, opt = {}.merge(listener_opts[address] || {})) + address = config.expand_addr(address) + return if String === address && listener_names.include?(address) + + delay = opt[:delay] || 0.5 + tries = opt[:tries] || 5 + begin + io = bind_listen(address, opt) + unless Kgio::TCPServer === io || Kgio::UNIXServer === io + IO_PURGATORY << io + io = server_cast(io) + end + logger.info "listening on addr=#{sock_name(io)} fd=#{io.fileno}" + LISTENERS << io + io + rescue Errno::EADDRINUSE => err + logger.error "adding listener failed addr=#{address} (in use)" + raise err if tries == 0 + tries -= 1 + logger.error "retrying in #{delay} seconds " \ + "(#{tries < 0 ? 'infinite' : tries} tries left)" + sleep(delay) + retry + rescue => err + logger.fatal "error adding listener addr=#{address}" + raise err + end + end + + # monitors children and receives signals forever + # (or until a termination signal is sent). This handles signals + # one-at-a-time time and we'll happily drop signals in case somebody + # is signalling us too often. + def join + respawn = true + last_check = Time.now + + proc_name 'master' + logger.info "master process ready" # test_exec.rb relies on this message + if ready_pipe + ready_pipe.syswrite($$.to_s) + ready_pipe.close rescue nil + self.ready_pipe = nil + end + begin + reap_all_workers + case SIG_QUEUE.shift + when nil + # avoid murdering workers after our master process (or the + # machine) comes out of suspend/hibernation + if (last_check + @timeout) >= (last_check = Time.now) + sleep_time = murder_lazy_workers + else + # wait for workers to wakeup on suspend + sleep_time = @timeout/2.0 + 1 + end + maintain_worker_count if respawn + master_sleep(sleep_time) + when :QUIT # graceful shutdown + break + when :TERM, :INT # immediate shutdown + stop(false) + break + when :USR1 # rotate logs + logger.info "master reopening logs..." + Unicorn::Util.reopen_logs + logger.info "master done reopening logs" + kill_each_worker(:USR1) + when :USR2 # exec binary, stay alive in case something went wrong + reexec + when :WINCH + if Process.ppid == 1 || Process.getpgrp != $$ + respawn = false + logger.info "gracefully stopping all workers" + kill_each_worker(:QUIT) + self.worker_processes = 0 + else + logger.info "SIGWINCH ignored because we're not daemonized" + end + when :TTIN + respawn = true + self.worker_processes += 1 + when :TTOU + self.worker_processes -= 1 if self.worker_processes > 0 + when :HUP + respawn = true + if config.config_file + load_config! + else # exec binary and exit if there's no config file + logger.info "config_file not present, reexecuting binary" + reexec + end + end + rescue Errno::EINTR + rescue => e + logger.error "Unhandled master loop exception #{e.inspect}." + logger.error e.backtrace.join("\n") + end while true + stop # gracefully shutdown all workers on our way out + logger.info "master complete" + unlink_pid_safe(pid) if pid + end + + # Terminates all workers, but does not exit master process + def stop(graceful = true) + self.listeners = [] + limit = Time.now + timeout + until WORKERS.empty? || Time.now > limit + kill_each_worker(graceful ? :QUIT : :TERM) + sleep(0.1) + reap_all_workers + end + kill_each_worker(:KILL) + end + + private + + # wait for a signal hander to wake us up and then consume the pipe + def master_sleep(sec) + IO.select([ SELF_PIPE[0] ], nil, nil, sec) or return + SELF_PIPE[0].kgio_tryread(11) + end + + def awaken_master + SELF_PIPE[1].kgio_trywrite('.') # wakeup master process from select + end + + # reaps all unreaped workers + def reap_all_workers + begin + wpid, status = Process.waitpid2(-1, Process::WNOHANG) + wpid or return + if reexec_pid == wpid + logger.error "reaped #{status.inspect} exec()-ed" + self.reexec_pid = 0 + self.pid = pid.chomp('.oldbin') if pid + proc_name 'master' + else + worker = WORKERS.delete(wpid) and worker.tmp.close rescue nil + m = "reaped #{status.inspect} worker=#{worker.nr rescue 'unknown'}" + status.success? ? logger.info(m) : logger.error(m) + end + rescue Errno::ECHILD + break + end while true + end + + # reexecutes the START_CTX with a new binary + def reexec + if reexec_pid > 0 + begin + Process.kill(0, reexec_pid) + logger.error "reexec-ed child already running PID:#{reexec_pid}" + return + rescue Errno::ESRCH + self.reexec_pid = 0 + end + end + + if pid + old_pid = "#{pid}.oldbin" + prev_pid = pid.dup + begin + self.pid = old_pid # clear the path for a new pid file + rescue ArgumentError + logger.error "old PID:#{valid_pid?(old_pid)} running with " \ + "existing pid=#{old_pid}, refusing rexec" + return + rescue => e + logger.error "error writing pid=#{old_pid} #{e.class} #{e.message}" + return + end + end + + self.reexec_pid = fork do + listener_fds = LISTENERS.map { |sock| sock.fileno } + ENV['UNICORN_FD'] = listener_fds.join(',') + Dir.chdir(START_CTX[:cwd]) + cmd = [ START_CTX[0] ].concat(START_CTX[:argv]) + + # avoid leaking FDs we don't know about, but let before_exec + # unset FD_CLOEXEC, if anything else in the app eventually + # relies on FD inheritence. + (3..1024).each do |io| + next if listener_fds.include?(io) + io = IO.for_fd(io) rescue next + IO_PURGATORY << io + io.fcntl(Fcntl::F_SETFD, Fcntl::FD_CLOEXEC) + end + logger.info "executing #{cmd.inspect} (in #{Dir.pwd})" + before_exec.call(self) + exec(*cmd) + end + proc_name 'master (old)' + end + + # forcibly terminate all workers that haven't checked in in timeout + # seconds. The timeout is implemented using an unlinked File + # shared between the parent process and each worker. The worker + # runs File#chmod to modify the ctime of the File. If the ctime + # is stale for >timeout seconds, then we'll kill the corresponding + # worker. + def murder_lazy_workers + t = @timeout + next_sleep = 1 + WORKERS.dup.each_pair do |wpid, worker| + stat = worker.tmp.stat + # skip workers that disable fchmod or have never fchmod-ed + stat.mode == 0100600 and next + diff = Time.now - stat.ctime + if diff <= t + tmp = t - diff + next_sleep < tmp and next_sleep = tmp + next + end + logger.error "worker=#{worker.nr} PID:#{wpid} timeout " \ + "(#{diff}s > #{t}s), killing" + kill_worker(:KILL, wpid) # take no prisoners for timeout violations + end + next_sleep + end + + def spawn_missing_workers + (0...worker_processes).each do |worker_nr| + WORKERS.values.include?(worker_nr) and next + worker = Worker.new(worker_nr, Unicorn::TmpIO.new) + before_fork.call(self, worker) + WORKERS[fork { + ready_pipe.close if ready_pipe + self.ready_pipe = nil + worker_loop(worker) + }] = worker + end + end + + def maintain_worker_count + (off = WORKERS.size - worker_processes) == 0 and return + off < 0 and return spawn_missing_workers + WORKERS.dup.each_pair { |wpid,w| + w.nr >= worker_processes and kill_worker(:QUIT, wpid) rescue nil + } + end + + # if we get any error, try to write something back to the client + # assuming we haven't closed the socket, but don't get hung up + # if the socket is already closed or broken. We'll always ensure + # the socket is closed at the end of this function + def handle_error(client, e) + msg = case e + when EOFError,Errno::ECONNRESET,Errno::EPIPE,Errno::EINVAL,Errno::EBADF + Unicorn::Const::ERROR_500_RESPONSE + when HttpParserError # try to tell the client they're bad + Unicorn::Const::ERROR_400_RESPONSE + else + logger.error "Read error: #{e.inspect}" + logger.error e.backtrace.join("\n") + Unicorn::Const::ERROR_500_RESPONSE + end + client.kgio_trywrite(msg) + client.close + rescue + nil + end + + # once a client is accepted, it is processed in its entirety here + # in 3 easy steps: read request, call app, write app response + def process_client(client) + r = @app.call(env = @request.read(client)) + + if 100 == r[0].to_i + client.write(Unicorn::Const::EXPECT_100_RESPONSE) + env.delete(Unicorn::Const::HTTP_EXPECT) + r = @app.call(env) + end + # r may be frozen or const, so don't modify it + @request.headers? or r = [ r[0], nil, r[2] ] + http_response_write(client, r) + rescue => e + handle_error(client, e) + end + + # gets rid of stuff the worker has no business keeping track of + # to free some resources and drops all sig handlers. + # traps for USR1, USR2, and HUP may be set in the after_fork Proc + # by the user. + def init_worker_process(worker) + QUEUE_SIGS.each { |sig| trap(sig, nil) } + trap(:CHLD, 'DEFAULT') + SIG_QUEUE.clear + proc_name "worker[#{worker.nr}]" + START_CTX.clear + init_self_pipe! + WORKERS.values.each { |other| other.tmp.close rescue nil } + WORKERS.clear + LISTENERS.each { |sock| sock.fcntl(Fcntl::F_SETFD, Fcntl::FD_CLOEXEC) } + worker.tmp.fcntl(Fcntl::F_SETFD, Fcntl::FD_CLOEXEC) + after_fork.call(self, worker) # can drop perms + worker.user(*user) if user.kind_of?(Array) && ! worker.switched + self.timeout /= 2.0 # halve it for select() + build_app! unless preload_app + end + + def reopen_worker_logs(worker_nr) + logger.info "worker=#{worker_nr} reopening logs..." + Unicorn::Util.reopen_logs + logger.info "worker=#{worker_nr} done reopening logs" + init_self_pipe! + end + + # runs inside each forked worker, this sits around and waits + # for connections and doesn't die until the parent dies (or is + # given a INT, QUIT, or TERM signal) + def worker_loop(worker) + ppid = master_pid + init_worker_process(worker) + nr = 0 # this becomes negative if we need to reopen logs + alive = worker.tmp # tmp is our lifeline to the master process + ready = LISTENERS + + # closing anything we IO.select on will raise EBADF + trap(:USR1) { nr = -65536; SELF_PIPE[0].close rescue nil } + trap(:QUIT) { alive = nil; LISTENERS.each { |s| s.close rescue nil } } + [:TERM, :INT].each { |sig| trap(sig) { exit!(0) } } # instant shutdown + logger.info "worker=#{worker.nr} ready" + m = 0 + + begin + nr < 0 and reopen_worker_logs(worker.nr) + nr = 0 + + # we're a goner in timeout seconds anyways if alive.chmod + # breaks, so don't trap the exception. Using fchmod() since + # futimes() is not available in base Ruby and I very strongly + # prefer temporary files to be unlinked for security, + # performance and reliability reasons, so utime is out. No-op + # changes with chmod doesn't update ctime on all filesystems; so + # we change our counter each and every time (after process_client + # and before IO.select). + alive.chmod(m = 0 == m ? 1 : 0) + + ready.each do |sock| + if client = sock.kgio_tryaccept + process_client(client) + nr += 1 + alive.chmod(m = 0 == m ? 1 : 0) + end + break if nr < 0 + end + + # make the following bet: if we accepted clients this round, + # we're probably reasonably busy, so avoid calling select() + # and do a speculative non-blocking accept() on ready listeners + # before we sleep again in select(). + redo unless nr == 0 # (nr < 0) => reopen logs + + ppid == Process.ppid or return + alive.chmod(m = 0 == m ? 1 : 0) + + # timeout used so we can detect parent death: + ret = IO.select(LISTENERS, nil, SELF_PIPE, timeout) and ready = ret[0] + rescue Errno::EINTR + ready = LISTENERS + rescue Errno::EBADF + nr < 0 or return + rescue => e + if alive + logger.error "Unhandled listen loop exception #{e.inspect}." + logger.error e.backtrace.join("\n") + end + end while alive + end + + # delivers a signal to a worker and fails gracefully if the worker + # is no longer running. + def kill_worker(signal, wpid) + Process.kill(signal, wpid) + rescue Errno::ESRCH + worker = WORKERS.delete(wpid) and worker.tmp.close rescue nil + end + + # delivers a signal to each worker + def kill_each_worker(signal) + WORKERS.keys.each { |wpid| kill_worker(signal, wpid) } + end + + # unlinks a PID file at given +path+ if it contains the current PID + # still potentially racy without locking the directory (which is + # non-portable and may interact badly with other programs), but the + # window for hitting the race condition is small + def unlink_pid_safe(path) + (File.read(path).to_i == $$ and File.unlink(path)) rescue nil + end + + # returns a PID if a given path contains a non-stale PID file, + # nil otherwise. + def valid_pid?(path) + wpid = File.read(path).to_i + wpid <= 0 and return + Process.kill(0, wpid) + wpid + rescue Errno::ESRCH, Errno::ENOENT + # don't unlink stale pid files, racy without non-portable locking... + end + + def load_config! + loaded_app = app + logger.info "reloading config_file=#{config.config_file}" + config[:listeners].replace(init_listeners) + config.reload + config.commit!(self) + kill_each_worker(:QUIT) + Unicorn::Util.reopen_logs + self.app = orig_app + build_app! if preload_app + logger.info "done reloading config_file=#{config.config_file}" + rescue StandardError, LoadError, SyntaxError => e + logger.error "error reloading config_file=#{config.config_file}: " \ + "#{e.class} #{e.message} #{e.backtrace}" + self.app = loaded_app + end + + # returns an array of string names for the given listener array + def listener_names(listeners = LISTENERS) + listeners.map { |io| sock_name(io) } + end + + def build_app! + if app.respond_to?(:arity) && app.arity == 0 + if defined?(Gem) && Gem.respond_to?(:refresh) + logger.info "Refreshing Gem list" + Gem.refresh + end + self.app = app.call + end + end + + def proc_name(tag) + $0 = ([ File.basename(START_CTX[0]), tag + ]).concat(START_CTX[:argv]).join(' ') + end + + def redirect_io(io, path) + File.open(path, 'ab') { |fp| io.reopen(fp) } if path + io.sync = true + end + + def init_self_pipe! + SELF_PIPE.each { |io| io.close rescue nil } + SELF_PIPE.replace(Kgio::Pipe.new) + SELF_PIPE.each { |io| io.fcntl(Fcntl::F_SETFD, Fcntl::FD_CLOEXEC) } + end +end + diff --git a/lib/unicorn/launcher.rb b/lib/unicorn/launcher.rb index 0d415dd..662b603 100644 --- a/lib/unicorn/launcher.rb +++ b/lib/unicorn/launcher.rb @@ -20,6 +20,7 @@ module Unicorn::Launcher # to pickup code changes if the original deployment directory # is a symlink or otherwise got replaced. def self.daemonize!(options) + cfg = Unicorn::Configurator $stdin.reopen("/dev/null") # We only start a new process group if we're not being reexecuted @@ -52,9 +53,9 @@ module Unicorn::Launcher end end # $stderr/$stderr can/will be redirected separately in the Unicorn config - Unicorn::Configurator::DEFAULTS[:stderr_path] ||= "/dev/null" - Unicorn::Configurator::DEFAULTS[:stdout_path] ||= "/dev/null" - Unicorn::Configurator::RACKUP[:daemonized] = true + cfg::DEFAULTS[:stderr_path] ||= "/dev/null" + cfg::DEFAULTS[:stdout_path] ||= "/dev/null" + cfg::RACKUP[:daemonized] = true end end diff --git a/lib/unicorn/preread_input.rb b/lib/unicorn/preread_input.rb new file mode 100644 index 0000000..ec83cb2 --- /dev/null +++ b/lib/unicorn/preread_input.rb @@ -0,0 +1,30 @@ +# -*- encoding: binary -*- + +module Unicorn +# This middleware is used to ensure input is buffered to memory +# or disk (depending on size) before the application is dispatched +# by entirely consuming it (from TeeInput) beforehand. +# +# Usage (in config.ru): +# +# require 'unicorn/preread_input' +# if defined?(Unicorn) +# use Unicorn::PrereadInput +# end +# run YourApp.new +class PrereadInput + def initialize(app) + @app = app + end + + def call(env) + buf = "" + input = env["rack.input"] + if buf = input.read(16384) + true while input.read(16384, buf) + input.rewind + end + @app.call(env) + end +end +end diff --git a/lib/unicorn/socket_helper.rb b/lib/unicorn/socket_helper.rb index 1d03eab..7364937 100644 --- a/lib/unicorn/socket_helper.rb +++ b/lib/unicorn/socket_helper.rb @@ -126,12 +126,12 @@ module Unicorn end old_umask = File.umask(opt[:umask] || 0) begin - UNIXServer.new(address) + Kgio::UNIXServer.new(address) ensure File.umask(old_umask) end elsif address =~ /^(\d+\.\d+\.\d+\.\d+):(\d+)$/ - TCPServer.new($1, $2.to_i) + Kgio::TCPServer.new($1, $2.to_i) else raise ArgumentError, "Don't know how to bind: #{address}" end @@ -166,9 +166,9 @@ module Unicorn def server_cast(sock) begin Socket.unpack_sockaddr_in(sock.getsockname) - TCPServer.for_fd(sock.fileno) + Kgio::TCPServer.for_fd(sock.fileno) rescue ArgumentError - UNIXServer.for_fd(sock.fileno) + Kgio::UNIXServer.for_fd(sock.fileno) end end diff --git a/lib/unicorn/tee_input.rb b/lib/unicorn/tee_input.rb index 540cfe0..a3e01d2 100644 --- a/lib/unicorn/tee_input.rb +++ b/lib/unicorn/tee_input.rb @@ -11,8 +11,8 @@ # # When processing uploads, Unicorn exposes a TeeInput object under # "rack.input" of the Rack environment. -class Unicorn::TeeInput < Struct.new(:socket, :req, :parser, - :buf, :len, :tmp, :buf2) +class Unicorn::TeeInput + attr_accessor :tmp, :socket, :parser, :env, :buf, :len, :buf2 # The maximum size (in +bytes+) to buffer in memory before # resorting to a temporary file. Default is 112 kilobytes. @@ -25,16 +25,19 @@ class Unicorn::TeeInput < Struct.new(:socket, :req, :parser, # Initializes a new TeeInput object. You normally do not have to call # this unless you are writing an HTTP server. - def initialize(*args) - super(*args) - self.len = parser.content_length - self.tmp = len && len < @@client_body_buffer_size ? - StringIO.new("") : Unicorn::Util.tmpio - self.buf2 = "" - if buf.size > 0 - parser.filter_body(buf2, buf) and finalize_input - tmp.write(buf2) - tmp.rewind + def initialize(socket, request) + @socket = socket + @parser = request + @buf = request.buf + @env = request.env + @len = request.content_length + @tmp = @len && @len < @@client_body_buffer_size ? + StringIO.new("") : Unicorn::TmpIO.new + @buf2 = "" + if @buf.size > 0 + @parser.filter_body(@buf2, @buf) and finalize_input + @tmp.write(@buf2) + @tmp.rewind end end @@ -55,16 +58,16 @@ class Unicorn::TeeInput < Struct.new(:socket, :req, :parser, # earlier. Most applications should only need to call +read+ with a # specified +length+ in a loop until it returns +nil+. def size - len and return len + @len and return @len if socket - pos = tmp.pos - while tee(@@io_chunk_size, buf2) + pos = @tmp.pos + while tee(@@io_chunk_size, @buf2) end - tmp.seek(pos) + @tmp.seek(pos) end - self.len = tmp.size + @len = @tmp.size end # :call-seq: @@ -87,22 +90,22 @@ class Unicorn::TeeInput < Struct.new(:socket, :req, :parser, # any data and only block when nothing is available (providing # IO#readpartial semantics). def read(*args) - socket or return tmp.read(*args) + @socket or return @tmp.read(*args) length = args.shift if nil == length - rv = tmp.read || "" - while tee(@@io_chunk_size, buf2) - rv << buf2 + rv = @tmp.read || "" + while tee(@@io_chunk_size, @buf2) + rv << @buf2 end rv else rv = args.shift || "" - diff = tmp.size - tmp.pos + diff = @tmp.size - @tmp.pos if 0 == diff ensure_length(tee(length, rv), length) else - ensure_length(tmp.read(diff > length ? length : diff, rv), length) + ensure_length(@tmp.read(diff > length ? length : diff, rv), length) end end end @@ -117,27 +120,27 @@ class Unicorn::TeeInput < Struct.new(:socket, :req, :parser, # This takes zero arguments for strict Rack::Lint compatibility, # unlike IO#gets. def gets - socket or return tmp.gets + @socket or return @tmp.gets sep = $/ or return read - orig_size = tmp.size - if tmp.pos == orig_size - tee(@@io_chunk_size, buf2) or return nil - tmp.seek(orig_size) + orig_size = @tmp.size + if @tmp.pos == orig_size + tee(@@io_chunk_size, @buf2) or return nil + @tmp.seek(orig_size) end sep_size = Rack::Utils.bytesize(sep) - line = tmp.gets # cannot be nil here since size > pos + line = @tmp.gets # cannot be nil here since size > pos sep == line[-sep_size, sep_size] and return line - # unlikely, if we got here, then tmp is at EOF + # unlikely, if we got here, then @tmp is at EOF begin - orig_size = tmp.pos - tee(@@io_chunk_size, buf2) or break - tmp.seek(orig_size) - line << tmp.gets + orig_size = @tmp.pos + tee(@@io_chunk_size, @buf2) or break + @tmp.seek(orig_size) + line << @tmp.gets sep == line[-sep_size, sep_size] and return line - # tmp is at EOF again here, retry the loop + # @tmp is at EOF again here, retry the loop end while true line @@ -163,51 +166,32 @@ class Unicorn::TeeInput < Struct.new(:socket, :req, :parser, # the offset (zero) of the +ios+ pointer. Subsequent reads will # start from the beginning of the previously-buffered input. def rewind - tmp.rewind # Rack does not specify what the return value is here + @tmp.rewind # Rack does not specify what the return value is here end private - def client_error(e) - case e - when EOFError - # in case client only did a premature shutdown(SHUT_WR) - # we do support clients that shutdown(SHUT_WR) after the - # _entire_ request has been sent, and those will not have - # raised EOFError on us. - socket.close if socket - raise Unicorn::ClientShutdown, "bytes_read=#{tmp.size}", [] - when Unicorn::HttpParserError - e.set_backtrace([]) - end - raise e - end - # tees off a +length+ chunk of data from the input into the IO # backing store as well as returning it. +dst+ must be specified. # returns nil if reading from the input returns nil def tee(length, dst) - unless parser.body_eof? - if parser.filter_body(dst, socket.readpartial(length, buf)).nil? - tmp.write(dst) - tmp.seek(0, IO::SEEK_END) # workaround FreeBSD/OSX + MRI 1.8.x bug + unless @parser.body_eof? + r = @socket.kgio_read(length, @buf) or eof! + unless @parser.filter_body(dst, @buf) + @tmp.write(dst) + @tmp.seek(0, IO::SEEK_END) # workaround FreeBSD/OSX + MRI 1.8.x bug return dst end end finalize_input - rescue => e - client_error(e) end def finalize_input - while parser.trailers(req, buf).nil? - # Don't worry about raising ClientShutdown here on EOFError, tee() - # will catch EOFError when app is processing it, otherwise in - # initialize we never get any chance to enter the app so the - # EOFError will just get trapped by Unicorn and not the Rack app - buf << socket.readpartial(@@io_chunk_size) + while @parser.trailers(@env, @buf).nil? + r = @socket.kgio_read(@@io_chunk_size) or eof! + @buf << r end - self.socket = nil + @socket = nil end # tee()s into +dst+ until it is of +length+ bytes (or until @@ -220,13 +204,21 @@ private # len is nil for chunked bodies, so we can't ensure length for those # since they could be streaming bidirectionally and we don't want to # block the caller in that case. - return dst if dst.nil? || len.nil? + return dst if dst.nil? || @len.nil? - while dst.size < length && tee(length - dst.size, buf2) - dst << buf2 + while dst.size < length && tee(length - dst.size, @buf2) + dst << @buf2 end dst end + def eof! + # in case client only did a premature shutdown(SHUT_WR) + # we do support clients that shutdown(SHUT_WR) after the + # _entire_ request has been sent, and those will not have + # raised EOFError on us. + @socket.close if @socket + raise Unicorn::ClientShutdown, "bytes_read=#{@tmp.size}", [] + end end diff --git a/lib/unicorn/tmpio.rb b/lib/unicorn/tmpio.rb new file mode 100644 index 0000000..a3c530d --- /dev/null +++ b/lib/unicorn/tmpio.rb @@ -0,0 +1,29 @@ +# -*- encoding: binary -*- +# :stopdoc: +require 'tmpdir' + +# some versions of Ruby had a broken Tempfile which didn't work +# well with unlinked files. This one is much shorter, easier +# to understand, and slightly faster. +class Unicorn::TmpIO < File + + # creates and returns a new File object. The File is unlinked + # immediately, switched to binary mode, and userspace output + # buffering is disabled + def self.new + fp = begin + super("#{Dir::tmpdir}/#{rand}", RDWR|CREAT|EXCL, 0600) + rescue Errno::EEXIST + retry + end + unlink(fp.path) + fp.binmode + fp.sync = true + fp + end + + # for easier env["rack.input"] compatibility with Rack <= 1.1 + def size + stat.size + end +end diff --git a/lib/unicorn/util.rb b/lib/unicorn/util.rb index e9dd57f..82329eb 100644 --- a/lib/unicorn/util.rb +++ b/lib/unicorn/util.rb @@ -1,101 +1,67 @@ # -*- encoding: binary -*- -require 'fcntl' -require 'tmpdir' - -module Unicorn - - class TmpIO < ::File +module Unicorn::Util + +# :stopdoc: + def self.is_log?(fp) + append_flags = File::WRONLY | File::APPEND + + ! fp.closed? && + fp.sync && + fp.path[0] == ?/ && + (fp.fcntl(Fcntl::F_GETFL) & append_flags) == append_flags + rescue IOError, Errno::EBADF + false + end - # for easier env["rack.input"] compatibility - def size - # flush if sync - stat.size + def self.chown_logs(uid, gid) + ObjectSpace.each_object(File) do |fp| + fp.chown(uid, gid) if is_log?(fp) end end - - module Util - class << self - - def is_log?(fp) - append_flags = File::WRONLY | File::APPEND - - ! fp.closed? && - fp.sync && - fp.path[0] == ?/ && - (fp.fcntl(Fcntl::F_GETFL) & append_flags) == append_flags - rescue IOError, Errno::EBADF - false +# :startdoc: + + # This reopens ALL logfiles in the process that have been rotated + # using logrotate(8) (without copytruncate) or similar tools. + # A +File+ object is considered for reopening if it is: + # 1) opened with the O_APPEND and O_WRONLY flags + # 2) opened with an absolute path (starts with "/") + # 3) the current open file handle does not match its original open path + # 4) unbuffered (as far as userspace buffering goes, not O_SYNC) + # Returns the number of files reopened + def self.reopen_logs + to_reopen = [] + nr = 0 + ObjectSpace.each_object(File) { |fp| is_log?(fp) and to_reopen << fp } + + to_reopen.each do |fp| + orig_st = begin + fp.stat + rescue IOError, Errno::EBADF + next end - def chown_logs(uid, gid) - ObjectSpace.each_object(File) do |fp| - fp.chown(uid, gid) if is_log?(fp) - end + begin + b = File.stat(fp.path) + next if orig_st.ino == b.ino && orig_st.dev == b.dev + rescue Errno::ENOENT end - # This reopens ALL logfiles in the process that have been rotated - # using logrotate(8) (without copytruncate) or similar tools. - # A +File+ object is considered for reopening if it is: - # 1) opened with the O_APPEND and O_WRONLY flags - # 2) opened with an absolute path (starts with "/") - # 3) the current open file handle does not match its original open path - # 4) unbuffered (as far as userspace buffering goes, not O_SYNC) - # Returns the number of files reopened - def reopen_logs - to_reopen = [] - nr = 0 - ObjectSpace.each_object(File) { |fp| is_log?(fp) and to_reopen << fp } - - to_reopen.each do |fp| - orig_st = begin - fp.stat - rescue IOError, Errno::EBADF - next - end - - begin - b = File.stat(fp.path) - next if orig_st.ino == b.ino && orig_st.dev == b.dev - rescue Errno::ENOENT - end - - begin - File.open(fp.path, 'a') { |tmpfp| fp.reopen(tmpfp) } - fp.sync = true - new_st = fp.stat - - # this should only happen in the master: - if orig_st.uid != new_st.uid || orig_st.gid != new_st.gid - fp.chown(orig_st.uid, orig_st.gid) - end + begin + File.open(fp.path, 'a') { |tmpfp| fp.reopen(tmpfp) } + fp.sync = true + new_st = fp.stat - nr += 1 - rescue IOError, Errno::EBADF - # not much we can do... - end + # this should only happen in the master: + if orig_st.uid != new_st.uid || orig_st.gid != new_st.gid + fp.chown(orig_st.uid, orig_st.gid) end - nr + nr += 1 + rescue IOError, Errno::EBADF + # not much we can do... end - - # creates and returns a new File object. The File is unlinked - # immediately, switched to binary mode, and userspace output - # buffering is disabled - def tmpio - fp = begin - TmpIO.open("#{Dir::tmpdir}/#{rand}", - File::RDWR|File::CREAT|File::EXCL, 0600) - rescue Errno::EEXIST - retry - end - File.unlink(fp.path) - fp.binmode - fp.sync = true - fp - end - end - + nr end end diff --git a/lib/unicorn/worker.rb b/lib/unicorn/worker.rb new file mode 100644 index 0000000..fd8d20e --- /dev/null +++ b/lib/unicorn/worker.rb @@ -0,0 +1,40 @@ +# -*- encoding: binary -*- + +# This class and its members can be considered a stable interface +# and will not change in a backwards-incompatible fashion between +# releases of Unicorn. You may need to access it in the +# before_fork/after_fork hooks. See the Unicorn::Configurator RDoc +# for examples. +class Unicorn::Worker < Struct.new(:nr, :tmp, :switched) + + # worker objects may be compared to just plain numbers + def ==(other_nr) + self.nr == other_nr + end + + # Changes the worker process to the specified +user+ and +group+ + # This is only intended to be called from within the worker + # process from the +after_fork+ hook. This should be called in + # the +after_fork+ hook after any priviledged functions need to be + # run (e.g. to set per-worker CPU affinity, niceness, etc) + # + # Any and all errors raised within this method will be propagated + # directly back to the caller (usually the +after_fork+ hook. + # These errors commonly include ArgumentError for specifying an + # invalid user/group and Errno::EPERM for insufficient priviledges + def user(user, group = nil) + # we do not protect the caller, checking Process.euid == 0 is + # insufficient because modern systems have fine-grained + # capabilities. Let the caller handle any and all errors. + uid = Etc.getpwnam(user).uid + gid = Etc.getgrnam(group).gid if group + Unicorn::Util.chown_logs(uid, gid) + tmp.chown(uid, gid) + if gid && Process.egid != gid + Process.initgroups(user, gid) + Process::GID.change_privilege(gid) + end + Process.euid != uid and Process::UID.change_privilege(uid) + self.switched = true + end +end |