On Tue, Jun 12, 2007 at 04:39:46PM +1000, Nathan Scott wrote:
> On Tue, 2007-06-12 at 16:08 +1000, David Chinner wrote:
> >
> > If you use this method of setting the extent size hint, then you will
> > *always* get the XFS_DIFLAG_EXTSIZE flag set when you have an extent
> > size hint, regardless of whether it is a realtime file or not.
>
> The extsize flag is relatively recent though, and traditionally
> realtime files could have had their extsize explicitly set with
> no associated extsize flag (thats just how it was implemented,
> originally, in realtime).
*nod*
We've got recent bugs reported because of this assumption and lack
of checking of the extent size hint flag where it needs to be
checked.
Either we have a flag to indicate the di_extsize field is valid or
we don't - it's too confusing to have different interfaces just
because an inode has a different, unrelated flag set on it.
Now that we have a flag, we can't remove support for it.....
> But, not many people use realtime, even fewer would be using the
> extent size option with realtime (like, none?, on Linux anyway)
> ... so, you could pretty much make whatever rule you like.
I sorta left that unsaid, but that is yet another reason I think
the change should stand.
Cheers,
Dave.
--
Dave Chinner
Principal Engineer
SGI Australian Software Group
|