Re: Filesystem full? I've got 4.2TB free!

[Stephen Borrill <netbsd%precedence.co.uk@localhost>]

On Wed, 22 Apr 2009, Stephen Borrill wrote:
> On Wed, 22 Apr 2009, Andy Ruhl wrote:
> > On Wed, Apr 22, 2009 at 3:11 AM, Stephen Borrill
> > <netbsd%precedence.co.uk@localhost> wrote:
> > > Filesystem in question is:
> > >
> > > Filesystem  Size Used      Avail %Cap    iUsed   iAvail %iCap Mounted on
> > > /dev/dk0    4.4T  39G       4.2T   0%   224044 592051282   0% /backup
> >
> > For the heck of it, what does du -sh /backup show?
>
> It's consistent with the size used reported by df.
>
> I've been playing about with more and I'm getting more and more confused now 
> :-/
>
> I can touch /backup/dellboy/foo, but I cannot mkdir /backup/dellboy/foo.
> rm -r /backup/dellboy then running rsync to recreate fails again, but not 
> always at the same directory depth

Antti had the great idea of running rump_ffs under gdb and putting a 
breakpoint on ffs_alloc(). The problem is this clause here:

        if (flags & B_CONTIG) {
                /*
                 * XXX ump->um_lock handling is "suspect" at best.
                 * For the case where ffs_hashalloc() fails early
                 * in the B_CONTIG case we reach here with um_lock
                 * already unlocked, so we can't release it again
                 * like in the normal error path.  See kern/39206.
                 *
                 *
                 * Fail silently - it's up to our caller to report
                 * errors.
                 */
                return (ENOSPC);
        }


--
Stephen