Re: [PATCH 6/6] xfs: refactor xfs_file_fallocate

[Brian Foster <bfoster@redhat.com>]

On Wed, Aug 21, 2024 at 02:57:56PM +0200, Christoph Hellwig wrote:
> On Wed, Aug 21, 2024 at 08:44:31AM -0400, Brian Foster wrote:
> > > +	error = xfs_reflink_unshare(XFS_I(inode), offset, len);
> > > +	if (error)
> > > +		return error;
> > > +
> > 
> > Doesn't unshare imply alloc?
> 
> Yes, ooks like that got lost and no test noticed it
> 
> > > -	if (xfs_file_sync_writes(file))
> > > +	if (!error && xfs_file_sync_writes(file))
> > >  		error = xfs_log_force_inode(ip);
> > 
> > I'd think if you hit -ENOSPC or something after doing a partial alloc to
> > a sync inode, you'd still want to flush the changes that were made..?
> 
> Persistence behavior on error is always undefined.  And that's also
> what the current code does, as it jumps past the log force from all
> error exits.
> 

Ok, if this preserves existing behavior then I'm not too worried about
it. Thanks.

Brian