Re: stat64 for over 2TB file returned invalid st_blocks

[Andreas Dilger <adilger@clusterfs.com>]

On Dec 06, 2005  09:48 -0500, Trond Myklebust wrote:
> On Tue, 2005-12-06 at 08:30 -0600, Dave Kleikamp wrote:
> > I think it looks good.  The only issue I have is that I agree with
> > Andreas that i_blocks should be of type sector_t.  I find the case of
> > accessing very large files over nfs with CONFIG_LBD disabled to be very
> > unlikely.
> 
> NO! sector_t is a block-device specific type. It does not belong in the
> generic inode.

sector_t would imply "units of 512-byte sectors", and this is exactly
what i_blocks is actually measuring, so I don't really understand your
objection.

If you have objection to the use of sector_t, it could be some other type
that is defined virtually identically as CONFIG_LBD sector_t, except that
it might be desirable to allow it to be configured separately for network
filesystems that have large files.  I'm sure the embedded linux folks
wouldn't be thrilled at an extra 4 bytes in every inode and 64-bit math
if they don't really use it.

Even in HPC very few users have many-TB files, and Lustre is one of the few
filesystems that can actually do this today.  We of course would enable
this option for our kernels, but I don't want to force it upon everyone.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.