Re: [PATCH RFC bpf-next v3 06/16] bpf/helpers: introduce bpf_timer_set_sleepable_cb() kfunc

[Alexei Starovoitov]

On Wed, Feb 21, 2024 at 8:25 AM Benjamin Tissoires <bentiss@xxxxxxxxxx> wrote:
>
> In this patch, bpf_timer_set_sleepable_cb() is functionally equivalent
> to bpf_timer_set_callback(), to the exception that it enforces
> the timer to be started with BPF_F_TIMER_SLEEPABLE.
>
> But given that bpf_timer_set_callback() is a helper when
> bpf_timer_set_sleepable_cb() is a kfunc, we need to teach the verifier
> about its attached callback.
> Marking that callback as sleepable will be done in a separate patch
>
> Signed-off-by: Benjamin Tissoires <bentiss@xxxxxxxxxx>
>
> ---
>
> new in v3 (split from v2 02/10)
> ---
>  kernel/bpf/helpers.c  | 48 ++++++++++++++++++++++++++++++++++++++++++++++--
>  kernel/bpf/verifier.c | 31 +++++++++++++++++++++++++++++--
>  2 files changed, 75 insertions(+), 4 deletions(-)
>
> diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> index f9add0abe40a..2c6dc3d0ffff 100644
> --- a/kernel/bpf/helpers.c
> +++ b/kernel/bpf/helpers.c
> @@ -1108,6 +1108,7 @@ struct bpf_hrtimer {
>         void __rcu *callback_fn;
>         void *value;
>         struct semaphore sleepable_lock;
> +       bool is_sleepable;
>  };
>
>  /* the actual struct hidden inside uapi struct bpf_timer */
> @@ -1270,8 +1271,8 @@ static const struct bpf_func_proto bpf_timer_init_proto = {
>         .arg3_type      = ARG_ANYTHING,
>  };
>
> -BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callback_fn,
> -          struct bpf_prog_aux *, aux)
> +static int __bpf_timer_set_callback(struct bpf_timer_kern *timer, void *callback_fn,
> +                                   struct bpf_prog_aux *aux, bool is_sleepable)
>  {
>         struct bpf_prog *prev, *prog = aux->prog;
>         struct bpf_hrtimer *t;
> @@ -1311,12 +1312,19 @@ BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callb
>                 t->prog = prog;
>         }
>         rcu_assign_pointer(t->callback_fn, callback_fn);
> +       t->is_sleepable = is_sleepable;
>         up(&t->sleepable_lock);
>  out:
>         __bpf_spin_unlock_irqrestore(&timer->lock);
>         return ret;
>  }
>
> +BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callback_fn,
> +          struct bpf_prog_aux *, aux)
> +{
> +       return __bpf_timer_set_callback(timer, callback_fn, aux, false);
> +}
> +
>  static const struct bpf_func_proto bpf_timer_set_callback_proto = {
>         .func           = bpf_timer_set_callback,
>         .gpl_only       = true,
> @@ -1342,6 +1350,11 @@ BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, fla
>                 goto out;
>         }
>
> +       if (t->is_sleepable && !(flags & BPF_F_TIMER_SLEEPABLE)) {
> +               ret = -EINVAL;
> +               goto out;
> +       }
> +
>         if (flags & BPF_F_TIMER_ABS)
>                 mode = HRTIMER_MODE_ABS_SOFT;
>         else
> @@ -2606,6 +2619,36 @@ __bpf_kfunc void bpf_throw(u64 cookie)
>         WARN(1, "A call to BPF exception callback should never return\n");
>  }
>
> +/**
> + * bpf_timer_set_sleepable_cb() - Configure the timer to call %callback_fn
> + * static function in a sleepable context.
> + * @timer: The bpf_timer that needs to be configured
> + * @callback_fn: a static bpf function
> + *
> + * @returns %0 on success. %-EINVAL if %timer was not initialized with
> + * bpf_timer_init() earlier. %-EPERM if %timer is in a map that doesn't
> + * have any user references.
> + * The user space should either hold a file descriptor to a map with timers
> + * or pin such map in bpffs. When map is unpinned or file descriptor is
> + * closed all timers in the map will be cancelled and freed.
> + *
> + * This kfunc is equivalent to %bpf_timer_set_callback except that it tells
> + * the verifier that the target callback is run in a sleepable context.
> + */
> +__bpf_kfunc int bpf_timer_set_sleepable_cb(struct bpf_timer_kern *timer,
> +                                          int (callback_fn)(void *map, int *key, struct bpf_timer *timer))
> +{
> +       struct bpf_throw_ctx ctx = {};
> +
> +       arch_bpf_stack_walk(bpf_stack_walker, &ctx);
> +       WARN_ON_ONCE(!ctx.aux);

Sorry. Why such complexity?
Please see how do_misc_fixups() handles BPF_FUNC_timer_set_callback.