[PATCH v2 2/2] xfs: fold xfs_create_tmpfile() into xfs_create()

[Brian Foster]

On Thu, Apr 10, 2014 at 03:29:01AM -0700, Christoph Hellwig wrote:
> > -	struct xfs_trans_res	tres;
> > +	struct xfs_trans_res	*tres;
> >  	uint			resblks;
> >  
> >  	trace_xfs_create(dp, name);
> > @@ -1181,14 +1181,21 @@ xfs_create(
> >  	if (is_dir) {
> >  		rdev = 0;
> >  		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
> > -		tres.tr_logres = M_RES(mp)->tr_mkdir.tr_logres;
> > -		tres.tr_logcount = XFS_MKDIR_LOG_COUNT;
> > +		tres = &M_RES(mp)->tr_mkdir;
> 
> The (nice) reservation cleanup should be a patch of it's own. 
> 

Ok.

> > +	} else {
> > +		/*
> > +		 * If we don't have a name, we're in the ->tmpfile() path. We
> > +		 * have a unique transaction here since we modify the unlinked
> > +		 * list rather than create a directory entry.
> > +		 */
> 
> How is that transaction more "uniqueue" than the others?  Seems like
> this comment generally doesn't add a whole lot of value.
> 

It's just as unique as the others. ;) I wasn't intending to call out
this transaction as special in any way. Rather, I was just trying to
document why there is a separate transaction depending on the existence
of the name. I can drop the comment.

> > +	if (name) {
> > +		xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
> > +		unlock_dp_on_error = true;
> > +
> > +		error = xfs_dir_canenter(tp, dp, name, resblks);
> > +		if (error)
> > +			goto out_trans_cancel;
> > +	}
> 
> So we get another special case in this function.  Can't say I like that
> too much, on the other hand I don't really like the duplicate code
> either.  So I'm not excited about this, but also not strongly against it.
> 

Indeed. I debated whether it would be reasonable to make this function
slightly longer and more complex on its own. When I realized
xfs_create_tmpfile() was 90% duplicate, it seemed worth the tradeoff for
a 100+ line function.

Brian