svn commit: r274653 - head/usr.sbin/freebsd-update

[Konstantin Belousov]

On Tue, Nov 18, 2014 at 08:43:59AM -0600, Mark Felder wrote:
> On Tue, Nov 18, 2014, at 08:24, Konstantin Belousov wrote:
> > Why do you suppose that md-backed volumes are not persistent ?
> > vnode-backed devices are stable.
> >
> 
> You're right, a vnode-backed filesystem would definitely be persistent.
> Clearly I've done a poor job of researching this thoroughly. I've now
> read md(4) in its entirety and remember mounting disk images like
> Linux's "mount -o loop" in the past. I don't see a way to reliably
> detect an mfs filesystem because it uses md and masquerades as ufs in
> df's output. Do you have any suggestions on how to detect this reliably?
> 
> On the other hand, anyone could write a filesystem we aren't
> blacklisting and fall into the same trap. I thought this was going to
> help a few people since Oliver reported it and seemed to have a valid
> use-case for a tmpfs mounted /var but still wanted to use
> freebsd-update. Instead I'm beginning to think we should just throw this
> away and add an entry into BUGS in the man page and call it a day. We
> can't keep everyone from shooting their feet off and we probably
> shouldn't waste our time trying.

Up to you.  You could somewhat improve this by parsing mdconfig -lv
output and see if the device is swap/memory or vnode backed.

In either case, if the code to detect transient /var is kept around,
there should be a way to turn heuristic off and just do what user
asked for.  Some sort of the force flag.