On Tue, 14 Jul 2020 at 18:32, Daniele Salvatore Albano
<d.albano@xxxxxxxxx> 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@xxxxxxxxx>
> ---
> 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?
Thanks!
Daniele
