Re: [fuse-devel] How to hang FUSE

[Miklos S. <mi...@sz...>]

> > > > > 
> > > > > You still need an -n option because the /old etc is read-only.
> > > > 
> > > > Yeah, the _real_ problem is /etc/mtab.  It usually works fine if you
> > > > symlink /proc/mounts to /etc/mtab, but some things (like "user" mounts
> > > > or automatic removal of loop devices) won't work correctly.
> > > You also can't image a system using anaconda -- if you have a little
> > > %post to create the link, you'll cause severe anaconda indigestion 
> > > as it wants to muck around with mtab after the %post scriptlets run.
> > 
> > What is anaconda and why is it mucking with mtab?
> anaconda is the Fedora/Red Hat system installer and since it's
> regularly mounting/unmounting filesystems it naturally wants
> to keep mtab up-to-date, even in the target system.

OK, that's during installation/upgrade.  I can see that keeping mtab
up to date is needed for some packages which rely on that file
(although arguably that sort of dependency is not very good).

But after installation, at boot time, mtab could be switched to a
symlink and thinkgs should work fine afterwards.

> > Interested in exactly why.  I'm very much hoping the /proc/mounts
> > approach is feasible, otherwise we are in big trouble over
> > unprivileged mounts.
> It's been a couple years since we worked with it, but IIRC the main
> problem was that /proc/mounts doesn't carry all the data that is
> found in /etc/mtab.

Like the "loop" and "user" options.  And lots of filesystems show only
some or none of their mount options in /proc/mounts.

Any others that you can remember?

Miklos