On Sun, May 04, 2008 at 06:16:11PM -0700, Dan Price wrote:

> > I agree that this should work, but zoneadm isn't ready yet:
> > 
> > - it uses stat() not stat64() on the special device
> > - for other than hsfs, it insists upon a raw device
> 
> (Where is that hsfs special-case code?)

zoneadm looks for /usr/lib/fs/<type>/fsck, and assumes that an fsck pass
is not necessary if it doesn't exist.

> > The former is (presumably) a simple fix. The latter is trickier: we
> > don't have a raw device to fsck. Perhaps zoneadm shouldn't be insisting
> > on a fsck pass anyway, or maybe this just shouldn't be allowed. What do
> > you think?
> 
> Sorry-- I may be *totally* confused, but...
> 
> Why couldn't one fsck an e.g. ufs image living in a file?  fsck
> seems to be OK with that at least for UFS.

It works for UFS, but zoneadm insists upon a 'raw' device, and does fsck
on that.

pcfs refuses to fsck a file. udfs asks you, then refuses to do so
anyway. I'm not sure fixing either of these are in scope.

So I can change zoneadm to allow no raw device if special is a file,
and zoneadmd to fsck the file, but this won't work for pcfs or udfs.

Seem reasonable?

> > > Two final questions: (1) is there any potential interaction with fsck(1m)
> > > here?  In other words, if the file we're looping back represents an FS
> > > which requires an fsck stage, how will that work?
> > 
> > Exactly like it does with lofiadm+mount today (or any direct mount
> > invocation): it'll mount without checking.
> 
> I guess what I meant was that with lofiadm + mounting today, presumably
> I can specify fsck'ing in /etc/vfstab when I place a vfstab entry
> there.  Maybe no one does that, or even uses vfstab, since the lofi
> devices don't persist across reboot.
> 
> On the other hand, the man page suggests that one write a script to
> persist lofi configurations (blech).  I wish lofiadm just supported that
> natively.

I missed that part of the manpage, but I'm dubious as to its utility
anyway. /etc/vfstab is getting less and less relevant.

> > >  (2) Will the lofi nodes created in this manner be visible in the
> > >  output of lofiadm(1m)?
> > 
> > Yes. I did toy with an alternative where it won't appear in /devices/,
> > but it was not nice implementation-wise, and after some thought felt
> > that listing the lofi entries was the right thing to do anyway.
> 
> Ok, thanks.  You might wish to amend the proposal to specifically state
> that.

I think this record will do...

cheers
john

Reply via email to