13 Aug
2022
13 Aug
'22
12:52 a.m.
Thread fork as we're drifting from documentation research specifically.
One matter that keeps coming to mind for me is the formal history of the runlevel-based
init system. It isn't in Research obviously, nor was it in PWB. The first time it
shows up in the wild is System III, but this version is slightly different than what was
in CB-UNIX at the time, which is what eventually wound up in System V.
The pressing question is whether the version in System III represents an earlier borrowing
from CB or if perhaps the runlevel init started in USG, got bumped over to CB and
improved, then that improved version came back to supported UNIX in 5.0.
As for the notable differences:
SysIII init allows for runlevels 1-9. It will call /etc/rc with the current state, the
prior state, and the number of times the current state has been entered. If the script is
called for instance due to a powerfailure, then the current state is passed suffixed with
an 'x'. The inittab entries are in a different format:
state:id:flags:process
Where state is the runlevel, id is a two-character identifier, flags can be either
'c' (like respawn) or 'o' (like off I think). No flag then indicates
to run once. Flags 't' or 'k' will terminate or kill a process before
it is started again if a given runlevel is entered and it is already running.
This of course is in contrast to SysV init which instead offers runlevels 0-6 as well as
a, b, and c. Init itself can be called with runlevels S|s or Q|q additionally and these
act as calls to enter single user mode or rerun the current init state if I'm
understanding correctly. Neither S nor Q options appear to be valid for the inittab
runlevel. Init tab entries here are:
id:rstate:action:process
Where id and rstate are essentially just the same fields from SysIII swapped. Action
replaces the flags field with the more well known respawn, wait, once, initdefault, etc.
behaviors.
All in all, different enough that inittabs between the two wouldn't be compatible.
SysV also includes the telinit command which appears to be able to handle those a, b, and
c runlevels.
Anywho, that's my understanding of the init changes, with the pertinent question
remaining whether the SysIII-style init ultimately started from the same place as SysV, or
if the general design idea was there between USG and CB, and they got to a similar answer
from different directions. Colon-delimited /etc files aren't uncommon, so while
unlikely, it could be entirely possible the two inittab formats arose relatively
independently, but the truth remains obscure in my mind at least. I don't really
blame Research for not picking up this init system, it seems like there were a few
parallel streams of development around the turn of the 80s, and the easier answer was
probably to just stay the course. I seem to recall reading in a thread somewhere Rob Pike
discussing the resistance in the Research group regarding sucking up each and every little
feature USG tried to promulgate as standard, and this init system got specific mention.
- Matt G.