Home | History | Annotate | Line # | Download | only in init 
      1 $NetBSD: NOTES,v 1.3 2006/04/18 11:40:26 salo Exp $
      2 
      3 POSIX and init:
      4 --------------
      5 
      6 POSIX.1 does not define 'init' but it mentions it in a few places.
      7 
      8 B.2.2.2, p205 line 873:
      9 
     10 	This is part of the extensive 'job control' glossary entry.
     11 	This specific reference says that 'init' must by default provide
     12 	protection from job control signals to jobs it starts --
     13 	it sets SIGTSTP, SIGTTIN and SIGTTOU to SIG_IGN.
     14 
     15 B.2.2.2, p206 line 889:
     16 
     17 	Here is a reference to 'vhangup'.  It says, 'POSIX.1 does
     18 	not specify how controlling terminal access is affected by
     19 	a user logging out (that is, by a controlling process
     20 	terminating).'  vhangup() is recognized as one way to handle
     21 	the problem.  I'm not clear what happens in Reno; I have
     22 	the impression that when the controlling process terminates,
     23 	references to the controlling terminal are converted to
     24 	references to a 'dead' vnode.  I don't know whether vhangup()
     25 	is required.
     26 
     27 B.2.2.2, p206 line 921:
     28 
     29 	Orphaned process groups bear indirectly on this issue.  A
     30 	session leader's process group is considered to be orphaned;
     31 	that is, it's immune to job control signals from the terminal.
     32 
     33 B.2.2.2, p233 line 2055:
     34 
     35 	'Historically, the implementation-dependent process that
     36 	inherits children whose parents have terminated without
     37 	waiting on them is called "init" and has a process ID of 1.'
     38 
     39 	It goes on to note that it used to be the case that 'init'
     40 	was responsible for sending SIGHUP to the foreground process
     41 	group of a tty whose controlling process has exited, using
     42 	vhangup().  It is now the responsibility of the kernel to
     43 	do this when the controlling process calls _exit().  The
     44 	kernel is also responsible for sending SIGCONT to stopped
     45 	process groups that become orphaned.  This is like old BSD
     46 	but entire process groups are signaled instead of individual
     47 	processes.
     48 
     49 	In general it appears that the kernel now automatically
     50 	takes care of orphans, relieving 'init' of any responsibility.
     51 	Specifics are listed on the _exit() page (p50).
     52 
     53 On setsid():
     54 -----------
     55 
     56 It appears that neither getty nor login call setsid(), so init must
     57 do this -- seems reasonable.  B.4.3.2 p 248 implies that this is the
     58 way that 'init' should work; it says that setsid() should be called
     59 after forking.
     60 
     61 Process group leaders cannot call setsid() -- another reason to
     62 fork!  Of course setsid() causes the current process to become a
     63 process group leader, so we can only call setsid() once.  Note that
     64 the controlling terminal acquires the session leader's process
     65 group when opened.
     66 
     67 Controlling terminals:
     68 ---------------------
     69 
     70 B.7.1.1.3 p276: 'POSIX.1 does not specify a mechanism by which to
     71 allocate a controlling terminal.  This is normally done by a system
     72 utility (such as 'getty') and is considered ... outside the scope
     73 of POSIX.1.'  It goes on to say that historically the first open()
     74 of a tty in a session sets the controlling terminal.  P130 has the
     75 full details; nothing particularly surprising.
     76 
     77 The glossary p12 describes a 'controlling process' as the first
     78 process in a session that acquires a controlling terminal.  Access
     79 to the terminal from the session is revoked if the controlling
     80 process exits (see p50, in the discussion of process termination).
     81 
     82 Design notes:
     83 ------------
     84 
     85 your generic finite state machine
     86 we are fascist about which signals we elect to receive,
     87 	even signals purportedly generated by hardware
     88 handle fatal errors gracefully if possible (we reboot if we goof!!)
     89 	if we get a segmentation fault etc., print a message on the console
     90 	and spin for a while before rebooting
     91 	(this at least decreases the amount of paper consumed :-)
     92 apply hysteresis to rapidly exiting gettys
     93 check wait status of children we reap
     94 	don't wait for stopped children
     95 don't use SIGCHILD, it's too expensive
     96 	but it may close windows and avoid races, sigh
     97 look for EINTR in case we need to change state
     98 init is responsible for utmp and wtmp maintenance (ick)
     99 	maybe now we can consider replacements?  maintain them in parallel
    100 	init only removes utmp and closes out wtmp entries...
    101 
    102 necessary states and state transitions (gleaned from the man page):
    103 	1: single user shell (with password checking?); on exit, go to 2
    104 	2: run rc script, on exit 0 check if init.root sysctl != "/", if it
    105            differs then fork + chroot into the value of init.root and run
    106            /etc/rc inside the chroot: on exit 0, go to 3; on exit N (error),
    107            go to 1 (applies also to /etc/rc when init.root == "/")
    108 	3: read ttys file: on completion, go to 4.  If we did chroot in
    109 	   state 2, we chroot after forking each getty to the same dir
    110 	   (init.root is not re-read)
    111 	4: multi-user operation: on SIGTERM, go to 7; on SIGHUP, go to 5;
    112 		on SIGTSTP, go to 6
    113 	5: clean up mode (re-read ttys file, killing off controlling processes
    114 		on lines that are now 'off', starting them on lines newly 'on')
    115 		on completion, go to 4
    116 	6: boring mode (no new sessions); signals as in 4
    117 	7: death: send SIGHUP to all controlling processes, reap for 30 seconds,
    118 		then go to 1 (warn if not all processes died, i.e. wait blocks)
    119 Given the -s flag, we start at state 1; otherwise state 2
    120