Re: [gentoo-dev] Re: rfc: improve file system mounting and unmounting in OpenRC

[William Hubbs]

On Wed, Aug 05, 2015 at 10:18:13AM -0400, Ian Stakenvicius wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA256
>
> On 05/08/15 10:01 AM, William Hubbs wrote:
> > On Wed, Aug 05, 2015 at 04:50:40AM +0000, Duncan wrote:
> >> Ian Stakenvicius posted on Tue, 04 Aug 2015 17:17:51 -0400 as
> >> excerpted:
> >>
> >>> So what you are suggesting here now is that you want to (A)
> >>> potentially break mounting with the need to externally manage
> >>> mounts via services in openrc instead of just using /etc/fstab,
> >>> and (B) also break services if something doesn't start, which
> >>> is one of the reasons why you wanted to go through with this
> >>> per-mount service in the first place.  My point is that no, we
> >>> should keep localmount as succeeding even if one of the
> >>> dependent services fails to mount, *just like it does right
> >>> now*, *for the same reasons* as it succeeds on failure right
> >>> now.
> >>
> >> +1
> >>
> >> IMO, localmount must continue to succeed /by/ /default/, even if
> >> some mounts fail, because it's basically legacy, and must
> >> maintain legacy behavior.  Turning it into a wrapper "internally"
> >> is fine, but the overall localmount must still succeed, as too
> >> much depends on that behavior as it is.
> >
> > Here's what I'm trying to deal with. Consider what happens if
> > service a still has "need localmount" and service b has "need
> > mount.foo".
> >
> > Mounting a file system twice causes failures the second time it is
> > mounted, so I  either have to add special handling in the new
> > mount script for file systems that are already mounted or come up
> > with a way to make sure localmount runs after all instances of the
> > new mount script.
>
> Yes.  I sincerely hope this was always part of the plan???
>
> If localmount is kept as a 'mount -a' then it won't remount anything
> that's already been mounted, so nothing to worry about there.  If it's
> a wrapper to the individual local mount.*'s then it just has to
> depend() on them all via a need/want (so that they will be brought in
> even if they aren't in the runlevel) and/or the mount.*'s need a
> 'before localmount' so that they're definitely started/attempted
> first.  Am I missing something that makes this harder than it seems?

It isn't localmount that would have the issue, but mount.* because they
are lexically after localmount, so you would end up with localmount
doing a mount -a then mount.* coming later trying to mount file systems
again that were mounted by localmount.

>
>
> >
> > The issue with making sure all mount scripts run before localmount
> > would complicate things more for users because they would have to
> > add the mount.foo symlinks to the appropriate runlevels (boot for
> > local file systems and default for network ones).
>
> Likely true as things stand now, which is why I expect major changes
> would be needed in openrc internals to make it so this is not
> necessary.  At minimum, "want" dependencies would need to be used so
> that if mount.foo's don't exist they won't cause depend() failures.
>
> (You thought this system was going to let you get away with -not-
> using 'want' for mounting, didn't you?  :)
 
 Not exactly, I just don't want to use want along with fstab scanning in
 netmount to try to make netmount start network file system daemons. I
 never was opposed to the want dependency on its own.

> Of course this still doesn't handle how depend() sections will be
> dynamically generated from /etc/fstab in the first place, nor how the
> cache will be affected by this; so there's still lots of work to do to
> sort that out.

There is no plan to dynamically generate depend() functions.

William