1$NetBSD: NOTES,v 1.3 2006/04/18 11:40:26 salo Exp $
2
3POSIX and init:
4--------------
5
6POSIX.1 does not define 'init' but it mentions it in a few places.
7
8B.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
15B.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
27B.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
33B.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
53On setsid():
54-----------
55
56It appears that neither getty nor login call setsid(), so init must
57do this -- seems reasonable.  B.4.3.2 p 248 implies that this is the
58way that 'init' should work; it says that setsid() should be called
59after forking.
60
61Process group leaders cannot call setsid() -- another reason to
62fork!  Of course setsid() causes the current process to become a
63process group leader, so we can only call setsid() once.  Note that
64the controlling terminal acquires the session leader's process
65group when opened.
66
67Controlling terminals:
68---------------------
69
70B.7.1.1.3 p276: 'POSIX.1 does not specify a mechanism by which to
71allocate a controlling terminal.  This is normally done by a system
72utility (such as 'getty') and is considered ... outside the scope
73of POSIX.1.'  It goes on to say that historically the first open()
74of a tty in a session sets the controlling terminal.  P130 has the
75full details; nothing particularly surprising.
76
77The glossary p12 describes a 'controlling process' as the first
78process in a session that acquires a controlling terminal.  Access
79to the terminal from the session is revoked if the controlling
80process exits (see p50, in the discussion of process termination).
81
82Design notes:
83------------
84
85your generic finite state machine
86we are fascist about which signals we elect to receive,
87          even signals purportedly generated by hardware
88handle 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 :-)
92apply hysteresis to rapidly exiting gettys
93check wait status of children we reap
94          don't wait for stopped children
95don't use SIGCHILD, it's too expensive
96          but it may close windows and avoid races, sigh
97look for EINTR in case we need to change state
98init 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
102necessary 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)
119Given the -s flag, we start at state 1; otherwise state 2
120