Re: [PATCH 2/3] xfs: use bio_reuse in the zone GC code

[Christoph Hellwig <hch@lst.de>]

On Tue, Jan 13, 2026 at 11:41:08AM -0700, Keith Busch wrote:
> On Tue, Jan 13, 2026 at 10:32:08AM -0800, Darrick J. Wong wrote:
> > On Tue, Jan 13, 2026 at 11:26:35AM -0700, Keith Busch wrote:
> > > On Tue, Jan 13, 2026 at 08:19:02AM +0100, Christoph Hellwig wrote:
> > > > @@ -825,10 +823,7 @@ xfs_zone_gc_write_chunk(
> > > >  	WRITE_ONCE(chunk->state, XFS_GC_BIO_NEW);
> > > >  	list_move_tail(&chunk->entry, &data->writing);
> > > >  
> > > > -	bio_reset(&chunk->bio, mp->m_rtdev_targp->bt_bdev, REQ_OP_WRITE);
> > > > -	bio_add_folio_nofail(&chunk->bio, chunk->scratch->folio, chunk->len,
> > > > -			offset_in_folio(chunk->scratch->folio, bvec_paddr));
> > > > -
> > > > +	bio_reuse(&chunk->bio);
> > > 
> > > bio_reuse() uses the previous bio->bi_opf value, so don't you need to
> > > explicitly override it to REQ_OP_WRITE here? Or maybe bio_reuse() should
> > > take the desired op as a parameter so it doesn't get doubly initialized
> > > by the caller.
> > 
> > xfs_zone_gc_submit_write changes bi_opf to REQ_OP_ZONE_APPEND, so I
> > don't think it's necessary to reset it in bio_reuse.
> 
> Only for sequential zones. Conventional zones still expect someone set
> the REQ_OP_WRITE, as the previous use of the bio in this path would have
> been for a READ operation.

Yes.  This is indeed broken for conventional zones / devices.  I'll
update it, and passing in the op as suggest might make it easier to
use.