On Mon, Mar 11, 2019 at 09:13:28PM +0000, Al Viro wrote:
> On Mon, Mar 11, 2019 at 08:44:31PM +0100, Christoph Hellwig wrote:
> > On Sun, Mar 10, 2019 at 07:08:17AM +0000, Al Viro wrote:
> > > From: Al Viro <viro@xxxxxxxxxxxxxxxxxx>
> > >
> > > Instead of having aio_complete() set ->ki_res.{res,res2}, do that
> > > explicitly in its callers, drop the reference (as aio_complete()
> > > used to do) and delay the rest until the final iocb_put().
> > >
> > > Signed-off-by: Al Viro <viro@xxxxxxxxxxxxxxxxxx>
> > > ---
> > > fs/aio.c | 45 ++++++++++++++++++++-------------------------
> > > 1 file changed, 20 insertions(+), 25 deletions(-)
> > >
> > > diff --git a/fs/aio.c b/fs/aio.c
> > > index 2249a7a1d6b3..b9c4c1894020 100644
> > > --- a/fs/aio.c
> > > +++ b/fs/aio.c
> > > @@ -1077,24 +1077,10 @@ static inline void iocb_destroy(struct aio_kiocb *iocb)
> > > kmem_cache_free(kiocb_cachep, iocb);
> > > }
> > >
> > > -static inline void iocb_put(struct aio_kiocb *iocb)
> > > -{
> > > - if (refcount_dec_and_test(&iocb->ki_refcnt))
> > > - iocb_destroy(iocb);
> > > -}
> >
> > Maybe iocb_put should just have been added in the place you move
> > it to in patch 1?
>
> Might as well...
Actually, that wouldn't be any better - it would need at least a declaration
before aio_complete(), since in the original patch aio_complete() calls it.
So that wouldn't be less noisy...
