Re: [PATCH bpf-next 5/6] bpf: support epoll from bpf struct_ops links.

[Andrii Nakryiko <andrii.nakryiko@gmail.com>]

On Mon, Apr 29, 2024 at 2:36 PM Kui-Feng Lee <thinker.li@gmail.com> wrote:
>
> Add epoll support to bpf struct_ops links to trigger EPOLLHUP event upon
> detachment.
>
> Signed-off-by: Kui-Feng Lee <thinker.li@gmail.com>
> ---
>  include/linux/bpf.h         |  2 ++
>  kernel/bpf/bpf_struct_ops.c | 14 ++++++++++++++
>  kernel/bpf/syscall.c        | 15 +++++++++++++++
>  3 files changed, 31 insertions(+)
>
> diff --git a/include/linux/bpf.h b/include/linux/bpf.h
> index eeeed4b1bd32..a4550b927352 100644
> --- a/include/linux/bpf.h
> +++ b/include/linux/bpf.h
> @@ -1574,6 +1574,7 @@ struct bpf_link {
>         const struct bpf_link_ops *ops;
>         struct bpf_prog *prog;
>         struct work_struct work;
> +       wait_queue_head_t wait_hup;

let's keep it struct_ops-specific, there is no need to pay for this
for all existing BPF link types. We can always generalize later, if
necessary.

pw-bot: cr


>  };
>
>  struct bpf_link_ops {
> @@ -1587,6 +1588,7 @@ struct bpf_link_ops {
>                               struct bpf_link_info *info);
>         int (*update_map)(struct bpf_link *link, struct bpf_map *new_map,
>                           struct bpf_map *old_map);
> +       __poll_t (*poll)(struct file *file, struct poll_table_struct *pts);
>  };
>
>  struct bpf_tramp_link {

[...]