Re: [PATCH] io_files_update_prep shouldn't consider all the flags invalid

[Daniele Salvatore Albano <d.albano@gmail.com>]

Sure thing, tomorrow I will put it together review all the other ops
as well, just in case (although I believe you may already have done
it), and test it.

For the test cases, should I submit a separate patch for liburing or
do you prefer to use pull requests on gh?

On Fri, 17 Jul 2020 at 17:21, Jens Axboe <axboe@kernel.dk> wrote:
>
> On 7/17/20 10:13 AM, Daniele Salvatore Albano wrote:
> > On Tue, 14 Jul 2020 at 18:32, Daniele Salvatore Albano
> > <d.albano@gmail.com> wrote:
> >>
> >> Currently when an IORING_OP_FILES_UPDATE is submitted with the
> >> IOSQE_IO_LINK flag it fails with EINVAL even if it's considered a
> >> valid because the expectation is that there are no flags set for the
> >> sqe.
> >>
> >> The patch updates the check to allow IOSQE_IO_LINK and ensure that
> >> EINVAL is returned only for IOSQE_FIXED_FILE and IOSQE_BUFFER_SELECT.
> >>
> >> Signed-off-by: Daniele Albano <d.albano@gmail.com>
> >> ---
> >>  fs/io_uring.c | 9 ++++++++-
> >>  1 file changed, 8 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/fs/io_uring.c b/fs/io_uring.c
> >> index ba70dc62f15f..7058b1a0bd39 100644
> >> --- a/fs/io_uring.c
> >> +++ b/fs/io_uring.c
> >> @@ -5205,7 +5205,14 @@ static int io_async_cancel(struct io_kiocb *req)
> >>  static int io_files_update_prep(struct io_kiocb *req,
> >>                                 const struct io_uring_sqe *sqe)
> >>  {
> >> -       if (sqe->flags || sqe->ioprio || sqe->rw_flags)
> >> +       unsigned flags = 0;
> >> +
> >> +       if (sqe->ioprio || sqe->rw_flags)
> >> +               return -EINVAL;
> >> +
> >> +       flags = READ_ONCE(sqe->flags);
> >> +
> >> +       if (flags & (IOSQE_FIXED_FILE | IOSQE_BUFFER_SELECT))
> >>                 return -EINVAL;
> >>
> >>         req->files_update.offset = READ_ONCE(sqe->off);
> >> --
> >> 2.25.1
> >
> > Hi,
> >
> > Did you get the chance to review this patch? Would you prefer to get
> > the flags loaded before the first branching?
>
> I think it looks fine, but looking a bit further, I think we should
> extend this kind of checking to also include timeout_prep and cancel_prep
> as well. They suffer from the same kind of issue where they disallow all
> flags, and they should just fail on the same as the above.
>
> And we should just use req->flags for this checking, and get rid of the
> sqe->flags reading in those prep functions. Something like this:
>
>
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index 74bc4a04befa..5c87b9a686dd 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -4732,7 +4732,9 @@ static int io_timeout_remove_prep(struct io_kiocb *req,
>  {
>         if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
>                 return -EINVAL;
> -       if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len)
> +       if (req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))
> +               return -EINVAL;
> +       if (sqe->ioprio || sqe->buf_index || sqe->len)
>                 return -EINVAL;
>
>         req->timeout.addr = READ_ONCE(sqe->addr);
> @@ -4910,8 +4912,9 @@ static int io_async_cancel_prep(struct io_kiocb *req,
>  {
>         if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
>                 return -EINVAL;
> -       if (sqe->flags || sqe->ioprio || sqe->off || sqe->len ||
> -           sqe->cancel_flags)
> +       if (req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))
> +               return -EINVAL;
> +       if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags)
>                 return -EINVAL;
>
>         req->cancel.addr = READ_ONCE(sqe->addr);
> @@ -4929,9 +4932,10 @@ static int io_async_cancel(struct io_kiocb *req)
>  static int io_files_update_prep(struct io_kiocb *req,
>                                 const struct io_uring_sqe *sqe)
>  {
> -       if (sqe->flags || sqe->ioprio || sqe->rw_flags)
> +       if (sqe->ioprio || sqe->rw_flags)
> +               return -EINVAL;
> +       if (req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))
>                 return -EINVAL;
> -
>         req->files_update.offset = READ_ONCE(sqe->off);
>         req->files_update.nr_args = READ_ONCE(sqe->len);
>         if (!req->files_update.nr_args)
>
> --
> Jens Axboe
>