Re: [PATCH 13/14] bpf: Add dispatchers to kallsyms

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

On Sat, Feb 8, 2020 at 7:43 AM Jiri Olsa <jolsa@kernel.org> wrote:
>
> Adding dispatchers to kallsyms. It's displayed as
>   bpf_dispatcher_<NAME>
>
> where NAME is the name of dispatcher.
>
> Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> ---
>  include/linux/bpf.h     | 19 ++++++++++++-------
>  kernel/bpf/dispatcher.c |  6 ++++++
>  2 files changed, 18 insertions(+), 7 deletions(-)
>
> diff --git a/include/linux/bpf.h b/include/linux/bpf.h
> index b91bac10d3ea..837cdc093d2c 100644
> --- a/include/linux/bpf.h
> +++ b/include/linux/bpf.h
> @@ -520,6 +520,7 @@ struct bpf_dispatcher {
>         int num_progs;
>         void *image;
>         u32 image_off;
> +       struct bpf_ksym ksym;
>  };
>
>  static __always_inline unsigned int bpf_dispatcher_nop_func(
> @@ -535,13 +536,17 @@ struct bpf_trampoline *bpf_trampoline_lookup(u64 key);
>  int bpf_trampoline_link_prog(struct bpf_prog *prog);
>  int bpf_trampoline_unlink_prog(struct bpf_prog *prog);
>  void bpf_trampoline_put(struct bpf_trampoline *tr);
> -#define BPF_DISPATCHER_INIT(name) {                    \
> -       .mutex = __MUTEX_INITIALIZER(name.mutex),       \
> -       .func = &name##_func,                           \
> -       .progs = {},                                    \
> -       .num_progs = 0,                                 \
> -       .image = NULL,                                  \
> -       .image_off = 0                                  \
> +#define BPF_DISPATCHER_INIT(_name) {                           \
> +       .mutex = __MUTEX_INITIALIZER(_name.mutex),              \
> +       .func = &_name##_func,                                  \
> +       .progs = {},                                            \
> +       .num_progs = 0,                                         \
> +       .image = NULL,                                          \
> +       .image_off = 0,                                         \
> +       .ksym = {                                               \
> +               .name = #_name,                                 \
> +               .lnode = LIST_HEAD_INIT(_name.ksym.lnode),      \
> +       },                                                      \
>  }
>
>  #define DEFINE_BPF_DISPATCHER(name)                                    \
> diff --git a/kernel/bpf/dispatcher.c b/kernel/bpf/dispatcher.c
> index b3e5b214fed8..8771d2cc5840 100644
> --- a/kernel/bpf/dispatcher.c
> +++ b/kernel/bpf/dispatcher.c
> @@ -152,6 +152,12 @@ void bpf_dispatcher_change_prog(struct bpf_dispatcher *d, struct bpf_prog *from,
>         if (!changed)
>                 goto out;
>
> +       if (!prev_num_progs)
> +               bpf_image_ksym_add(d->image, &d->ksym);
> +
> +       if (!d->num_progs)
> +               bpf_ksym_del(&d->ksym);
> +
>         bpf_dispatcher_update(d, prev_num_progs);

On slightly unrelated note: seems like bpf_dispatcher_update won't
propagate any lower-level errors back, which seems pretty bad as a
bunch of stuff can go wrong.

Björn, was it a conscious decision or this just slipped through the cracks?

Jiri, reason I started looking at this was twofold:
1. you add/remove symbol before dispatcher is updated, which is
different order from BPF trampoline updates. I think updating symbols
after successful update makes more sense, no?
2. I was wondering if bpf_dispatcher_update() could return 0/1 (and <0
on error, of course), depending on whether dispatcher is present or
not. Though I'm not hard set on this.

>  out:
>         mutex_unlock(&d->mutex);
> --
> 2.24.1
>