Subject: Re: df output with a really large filesystems
To: Bill Studenmund <>
From: David Laight <>
List: tech-userlevel
Date: 03/31/2004 21:37:27
On Mon, Mar 29, 2004 at 06:38:25PM -0800, Bill Studenmund wrote:
> On Fri, Mar 26, 2004 at 12:17:59AM +0900, enami tsugutomo wrote:
> > > BTW, our filer only has one shelf of disks (to make the 1.1T fs).
> > > i guess if i had more shelves i would have a problem?
> > 
> > Ya, if the # of blocks doesn't fit the `long' variable, it is
> > impossible for userland program to calculate correct value.  For nfs,
> > the limit is 2T since it uses 512 as block size (f_bsize).
> Interesting. Looks like statfs got missed. stat(2) already has:
>          int64_t   st_blocks;  /* blocks allocated for file */
> So we're ok if a file is big, but not the fs.
> Someone want to fix this fast so it makes 2.0? :-)

The first thing to fix is the netbsd disklabel.  Although the kernel
uses 64bit numbers for disk sectors, and both the ffsv1 and ffsv2
filesystems can handle disks with > 2^32 sectors, there is no way
to describe such a disk volums.


David Laight: