Re: [PATCH v2 4/4] io_uring: pre-increment f_pos on rw

[Dylan Yudaken <dylany@fb.com>]

On Mon, 2022-02-21 at 18:00 +0000, Pavel Begunkov wrote:
> On 2/21/22 14:16, Dylan Yudaken wrote:
> > In read/write ops, preincrement f_pos when no offset is specified,
> > and
> > then attempt fix up the position after IO completes if it completed
> > less
> > than expected. This fixes the problem where multiple queued up IO
> > will all
> > obtain the same f_pos, and so perform the same read/write.
> > 
> > This is still not as consistent as sync r/w, as it is able to
> > advance the
> > file offset past the end of the file. It seems it would be quite a
> > performance hit to work around this limitation - such as by keeping
> > track
> > of concurrent operations - and the downside does not seem to be too
> > problematic.
> > 
> > The attempt to fix up the f_pos after will at least mean that in
> > situations
> > where a single operation is run, then the position will be
> > consistent.
> > 
> > Co-developed-by: Jens Axboe <axboe@kernel.dk>
> > Signed-off-by: Jens Axboe <axboe@kernel.dk>
> > Signed-off-by: Dylan Yudaken <dylany@fb.com>
> > ---
> >   fs/io_uring.c | 81 ++++++++++++++++++++++++++++++++++++++++++----
> > -----
> >   1 file changed, 68 insertions(+), 13 deletions(-)
> > 
> > diff --git a/fs/io_uring.c b/fs/io_uring.c
> > index abd8c739988e..a951d0754899 100644
> > --- a/fs/io_uring.c
> > +++ b/fs/io_uring.c
> > @@ -3066,21 +3066,71 @@ static inline void io_rw_done(struct kiocb
> > *kiocb, ssize_t ret)
> 
> [...]
> 
> > +                       return false;
> >                 }
> >         }
> > -       return is_stream ? NULL : &kiocb->ki_pos;
> > +       *ppos = is_stream ? NULL : &kiocb->ki_pos;
> > +       return false;
> > +}
> > +
> > +static inline void
> > +io_kiocb_done_pos(struct io_kiocb *req, struct kiocb *kiocb, u64
> > actual)
> 
> That's a lot of inlining, I wouldn't be surprised if the compiler
> will even refuse to do that.
> 
> io_kiocb_done_pos() {
>         // rest of it
> }
> 
> inline io_kiocb_done_pos() {
>         if (!(flags & CUR_POS));
>                 return;
>         __io_kiocb_done_pos();
> }
> 
> io_kiocb_update_pos() is huge as well

Good idea, will split the slower paths out.

> 
> > +{
> > +       u64 expected;
> > +
> > +       if (likely(!(req->flags & REQ_F_CUR_POS)))
> > +               return;
> > +
> > +       expected = req->rw.len;
> > +       if (actual >= expected)
> > +               return;
> > +
> > +       /*
> > +        * It's not definitely safe to lock here, and the
> > assumption is,
> > +        * that if we cannot lock the position that it will be
> > changing,
> > +        * and if it will be changing - then we can't update it
> > anyway
> > +        */
> > +       if (req->file->f_mode & FMODE_ATOMIC_POS
> > +               && !mutex_trylock(&req->file->f_pos_lock))
> > +               return;
> > +
> > +       /*
> > +        * now we want to move the pointer, but only if everything
> > is consistent
> > +        * with how we left it originally
> > +        */
> > +       if (req->file->f_pos == kiocb->ki_pos + (expected -
> > actual))
> > +               req->file->f_pos = kiocb->ki_pos;
> 
> I wonder, is it good enough / safe to just assign it considering that
> the request was executed outside of locks? vfs_seek()?

No I do not think so - in the case of multiple r/w the same thing will
happen, even with no vfs_seek().

> 
> > +
> > +       /* else something else messed with f_pos and we can't do
> > anything */
> > +
> > +       if (req->file->f_mode & FMODE_ATOMIC_POS)
> > +               mutex_unlock(&req->file->f_pos_lock);
> >   }
> 
> Do we even care about races while reading it? E.g.
> pos = READ_ONCE();

I think so - if I remove all the locks the test cases fail.

> 
> >   
> > -       ppos = io_kiocb_update_pos(req, kiocb);
> > -
> >         ret = rw_verify_area(READ, req->file, ppos, req->result);
> >         if (unlikely(ret)) {
> >                 kfree(iovec);
> > +               io_kiocb_done_pos(req, kiocb, 0);
> 
> Why do we update it on failure?
> 
> [...]
> 
> > -       ppos = io_kiocb_update_pos(req, kiocb);
> > -
> >         ret = rw_verify_area(WRITE, req->file, ppos, req->result);
> >         if (unlikely(ret))
> >                 goto out_free;
> > @@ -3858,6 +3912,7 @@ static int io_write(struct io_kiocb *req,
> > unsigned int issue_flags)
> >                 return ret ?: -EAGAIN;
> >         }
> >   out_free:
> > +       io_kiocb_done_pos(req, kiocb, 0);
> 
> Looks weird. It appears we don't need it on failure and
> successes are covered by kiocb_done() / ->ki_complete
> 
> >         /* it's reportedly faster than delegating the null check to
> > kfree() */
> >         if (iovec)
> >                 kfree(iovec);
>