cinit - braindumps - shutdown
=============================
Nico Schottelius <nico-cinit@@@@schottelius.org>
v0.1, for cinit-0.3, 2007-01-11
:Author Initials: NS

Just some braindumps. Do not read.

Introduction
------------
Shutting down the system.

When?
-----
- reboot
- halt
- power off
- further needed?

How / What?
------------
- Stop services in the correct order
   * reuse the service tree / needs / needed_by!
   * call off, if existent
   * if respawning, send SIGTERM?
      - not if off is existent?
   * give some variables to off?
      - $RESPAWN_PID: if respawning
- send SIGTERM to other apps?
   * or put that into special? special/post_svc_shutdown
- call special/{halt,poweroff,reboot}
   * no, implemented that directly into cinit!
- implement timely controlled shutdown:
   * cinit.shutdown forks (?) itself and waits, then sends a signal
     to cinit to initiate the shutdown process
   * create mappers for target OS
      - Linux
      - {free,net,open}bsd
      - others may contribute theirself
- also implement poweroff
   * kill -TERM 1
- also implement reboot
   * kill -HUP 1
- also implement halt
   * kill -USR1 1


The implementation
-----------------
x shutdown is initiated
x disable communication (nobody can influence us anymore)
x install special signal handler?
   - no. we do not care about zombies anymore, we are gone soon anyway
   - simply reset to the default
- begin to shutdown services
   * where to start?
   * begin shutdown at the end of the tree, viewing/sorting it by the 'needed_by'
      -> where to begin sort?
      -> worst implementation: scan all services (easy for now)
      -> result: list of services, that noone needs :-)
   * stop those services and step above
      - stop a service, as soon as noone needs it anymore
         * check all needed_by links
   * more or less the inverted gen_svc_tree
- how to shutdown a service
   * if existing, call off
      - if off does not exist:
         * if respawn: send sigterm to the process
         * also implement sigkill at this stage?
            - think we can delay it to the end
            - otherwise users can use off!
      - implement environment for off ($cinit_respawn_pid, $cinit_service_name)
      - do *NOT* care about the return value, we can't do anything for it
         anymore anyway
- after all services are shutdown, kill all remaining services
   * SIGTERM
   * wait(defined_time_in_seconds)
   * SIGKILL
- call the last command
   * to cleanup the system
      - umount -a (if needed)
      - swapoff (if needed)
      - sync (if needed)
- switch off