Re: [PATCH 2/3] shmem: respect MAX_LFS_FILESIZE

[Hugh Dickins <hugh@veritas.com>]

On Thu, 9 Apr 2009, Andrew Morton wrote:
> On Mon, 6 Apr 2009 21:56:13 +0100 (BST)
> Hugh Dickins <hugh@veritas.com> wrote:
> 
> > Question: couldn't the 32-bit kernel's MAX_LFS_FILESIZE be almost doubled?
> > It limits the pagecache index to a signed long, but we use an unsigned long.
> 
> I expect it would be OK, yes.  The only failure mode I can think of is
> if someone is using signed long as a pagecache index and I'd be pretty
> surprised if we've made that mistake anywhere.  The potential for goofs
> is higher down in filesystems, but they shouldn't be using pagecache
> indices much at all.
> 
> Of course it does invite people to write applications which then fail
> on older kernels, but such is life.

Hmm, that's a very good point, and I doubt Ned Kelly can have the
last word on it.  Good filesystems go to a great deal of trouble over
the compatibility issues of new features: it would be rather sad to
blow that all away with a careless doubling of MAX_LFS_FILESIZE.

Or I'm talking nonsense: we already have this issue, when using
a 32-bit kernel to look at big files created with a 64-bit kernel.

But even so, I think I'll leave this change to someone braver.

Hugh