EF> I know /etc/init called /bin/sh /etc/rc back in the day,
KRE> It still does, normally.
Yes, that's what I found when reading init.c.
> Not really.
Well, it's /sbin/init now.
> "Executing a script" is just a shorthand way of saying "exexute
> the interpreter for which this script is designed to use and tell it
> to run this script".
Yes. /etc/rc is a Bourne Shell script. The interpreter can be /bin/sh,
/rescue/sh (I read the code as that /rescue/init uses the latter) or some
other Bourne Shell.
> [...] but that argv[0] there is just what becomes $0 which /etc/rc
> only uses in error/log messages.
Right, sorry. Of course it's INIT_BSHELL what's executed.
EF> So I guess the manpage should be re-phrased.
KRE> I doubt there's a real need for that.
I would find it useful to read from the manpage, not the source, what actually
happens.
> If you give boot the -a switch
Well, then I could also just walk by the machine and do everything in
single user mode.
This is about a backup server (actually two of them) which, for obvious
reasons, is located in a different (part of of) the building. I need to
phyically go there with a monitor and a keyboard for single user mode.
I was trying to avoid that.
> If you tell it /rescue/init instead, then that init will use /rescue/sh
> as the shell to run the rc script.
OK, so my deduction was correct.
> ps: updating the fundamental stuff on an active system, however you do
> it, is always going to be a very dangerous operation.
Yes, sure. Been there, done that (for certain values of "active").
The scheme I use is to create (when, say, updating from -8 to -10)
/8 and /10, also (for the classical scheme I use) /{usr,var}/{8,10},
null-mount /usr/10 to /10/usr etc., copy /etc (and a few other files)
to /10, unpack the sets there (that being the main reason for the null
mounts), run /10/usr/sbin/etcupdate -d /10 (my enhanced version that
playbacks an input file), /10/usr/sbin/postinstall -d /10 fix xyz, go
single user with /rescue/sh (which is from -8), run a script that,
according to a configuration file, moves (or not moves) things from /foo
to /8/foo and from /10/foo to /foo (which takes a only second), and
/rescue/reboot.
In case anything failes on -10, I go back single user with /rescue/sh,
run my script which then moves things from /foo to /10/foo and from
/8/foo to /foo and I'm back to the state before.
I even use that to switch development servers to a different version
and back to test things on other versions. /rescue always stays at
the lowest version installed.
