Re: Running /etc/rc.d/nfsd onestart complains about $mountd not enabled

[Adrian Christiansen <adrian.christiansen%gmail.com@localhost>]

2015-07-19 5:57 GMT+02:00 Robert Elz <kre%munnari.oz.au@localhost>:
>     Date:        Sat, 18 Jul 2015 21:28:01 -0400
>     From:        Greg Troxel <gdt%ir.bbn.com@localhost>
>     Message-ID:  <rmid1zp6in2.fsf%fnord.ir.bbn.com@localhost>
>
>   | Well, if you just want nfsd, it seems to work, and the mountd comment is
>   | a warning.  So what's actually wrong?
>
>   | I'm not sure what you are suggesting should be different.

Sorry for having used Linux for too long and being confused about
NetBSD. But I'm working on it!

What I wanted to happen was for nfsd's rc.d script to check if rpcbind
and mountd was running, NOT if they were marked as YES in rc.conf.
This is my laptop and I don't want it to be running nfsd all the time
on every boot.

Actually!, if you do /etc/rc.d/nfsd onestart with rpcbind=YES and
mountd=YES inside rc.conf, it will execute nfsd without rpcbind and
mountd even running! THIS IS WRONG! That is the only check done by
rc.subr, is it in rc.conf, if so I guess they are running.

You can try this too, no need to reboot.
The same goes for /etc/rc.d/nfsd start if nfs_server=YES is also in rc.conf.

>
> It wasn't me, and this behaviour rarely bothers me, so it isn't important,
> but I think the issue for "onestart" is that the rc.d dependancy checking
> checks whether the dependency is turned on in rc.conf, where for onestart
> that's generally exactly what you don't want - you don't want things just
> magically starting because the system happens to have rebooted.
>
> For onestart what should really be being checked is whether or not the
> dependency is actually running, not whether or not it is marked to start at
> boot.  At boot time doing those checks would needlessly slow the boot
> process - if it was marked to start at boot, it was started, and if not,
> it was not - but that relationship isn't true for onestart.
>
> What's more, the cost of actually checking the presence of the required
> processes is negligible for the onestart case, where only one process is
> being started, so we could easily afford to do it (even if it meant calling
> the dependant rc.d script with the "status" arg and parsing the result).
>
> Whether dependencies that aren't running should be automatically started
> if required (using onestart) or just warned as missing is a whole other
> question of course - I suspect opinions on that one will differ.
>
> kre
>

From using FreeBSD and Linux this was the behaviour I was expecting
from onestart.