Re: [PATCH] iomap: replace folio_batch allocation with stack allocation

[Brian Foster <bfoster@redhat.com>]

On Wed, Nov 12, 2025 at 12:39:40AM -0800, Christoph Hellwig wrote:
> On Tue, Nov 11, 2025 at 12:50:47PM -0500, Brian Foster wrote:
> >  		if (imap.br_state == XFS_EXT_UNWRITTEN &&
> >  		    offset_fsb < eof_fsb) {
> > -			loff_t len = min(count,
> > -					 XFS_FSB_TO_B(mp, imap.br_blockcount));
> > +			loff_t foffset = offset, fend;
> >  
> > -			end = iomap_fill_dirty_folios(iter, offset, len);
> > +			fend = offset +
> > +			       min(count, XFS_FSB_TO_B(mp, imap.br_blockcount));
> > +			iomap_flags |= iomap_fill_dirty_folios(iter, &foffset,
> > +							       fend);
> >  			end_fsb = min_t(xfs_fileoff_t, end_fsb,
> > -					XFS_B_TO_FSB(mp, end));
> > +					XFS_B_TO_FSB(mp, foffset));
> 
> Maybe it's just me, but I found the old calling convention a lot more
> logic.  Why not keep it and extend it for passing the flags as an
> in/out argument?  That would also keep the churn down a bit.
> 

Hmm.. well I never really loved the flag return (or the end return), but
I wanted to make the iomap helper more consistent with the underlying
filemap helper because I think that reduces unnecessary complexity. I
suppose we could also make the flags an out param and either return void
or just pass through the filemap helper return (i.e. folio count)...

Brian

> Otherwise the change looks good.
>