Re: [RFC PATCH bpf-next 3/4] bpf: Introduce css_descendant open-coded iterator kfuncs

[Alexei Starovoitov]

On Sun, Aug 27, 2023 at 12:21 AM Chuyi Zhou <zhouchuyi@xxxxxxxxxxxxx> wrote:
>
> This Patch adds kfuncs bpf_iter_css_{new,next,destroy} which allow creation
> and manipulation of struct bpf_iter_css in open-coded iterator style. These
> kfuncs actually wrapps css_next_descendant_{pre, post}. BPF programs can
> use these kfuncs through bpf_for_each macro for iteration of all descendant
> css under a root css.
>
> Normally, css_next_descendant_{pre, post} should be called with rcu
> locking. Although we have bpf_rcu_read_lock(), here we still calls
> rcu_read_lock in bpf_iter_css_new and unlock in bpf_iter_css_destroy
> for convenience use.
>
> Signed-off-by: Chuyi Zhou <zhouchuyi@xxxxxxxxxxxxx>
> ---
>  include/uapi/linux/bpf.h       |  5 +++++
>  kernel/bpf/helpers.c           |  3 +++
>  kernel/bpf/task_iter.c         | 39 ++++++++++++++++++++++++++++++++++
>  tools/include/uapi/linux/bpf.h |  5 +++++
>  tools/lib/bpf/bpf_helpers.h    |  6 ++++++
>  5 files changed, 58 insertions(+)
>
> diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> index cfbd527e3733..19f1f1bf9301 100644
> --- a/include/uapi/linux/bpf.h
> +++ b/include/uapi/linux/bpf.h
> @@ -7203,4 +7203,9 @@ struct bpf_iter_process {
>         __u64 __opaque[1];
>  } __attribute__((aligned(8)));
>
> +struct bpf_iter_css {
> +       __u64 __opaque[2];
> +       char __opaque_c[1];

Burning extra 8 bytes for flags seems excessive.
Maybe let's add two iterators for descendant_post/_pre ?
The bpf prog code will be easier to read (no need to guess
what bool flag does).

> +} __attribute__((aligned(8)));
> +
>  #endif /* _UAPI__LINUX_BPF_H__ */
> diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> index 81a2005edc26..47d46a51855f 100644
> --- a/kernel/bpf/helpers.c
> +++ b/kernel/bpf/helpers.c
> @@ -2461,6 +2461,9 @@ BTF_ID_FLAGS(func, bpf_iter_css_task_destroy, KF_ITER_DESTROY)
>  BTF_ID_FLAGS(func, bpf_iter_process_new, KF_ITER_NEW)
>  BTF_ID_FLAGS(func, bpf_iter_process_next, KF_ITER_NEXT | KF_RET_NULL)
>  BTF_ID_FLAGS(func, bpf_iter_process_destroy, KF_ITER_DESTROY)
> +BTF_ID_FLAGS(func, bpf_iter_css_new, KF_ITER_NEW)
> +BTF_ID_FLAGS(func, bpf_iter_css_next, KF_ITER_NEXT | KF_RET_NULL)
> +BTF_ID_FLAGS(func, bpf_iter_css_destroy, KF_ITER_DESTROY)
>  BTF_ID_FLAGS(func, bpf_dynptr_adjust)
>  BTF_ID_FLAGS(func, bpf_dynptr_is_null)
>  BTF_ID_FLAGS(func, bpf_dynptr_is_rdonly)
> diff --git a/kernel/bpf/task_iter.c b/kernel/bpf/task_iter.c
> index a6717a76c1e0..ef9aef62f1ac 100644
> --- a/kernel/bpf/task_iter.c
> +++ b/kernel/bpf/task_iter.c
> @@ -893,6 +893,45 @@ __bpf_kfunc void bpf_iter_process_destroy(struct bpf_iter_process *it)
>         rcu_read_unlock();
>  }
>
> +struct bpf_iter_css_kern {
> +       struct cgroup_subsys_state *root;
> +       struct cgroup_subsys_state *pos;
> +       char flag;
> +} __attribute__((aligned(8)));
> +
> +__bpf_kfunc int bpf_iter_css_new(struct bpf_iter_css *it,
> +               struct cgroup_subsys_state *root, char flag)
> +{
> +       struct bpf_iter_css_kern *kit = (void *)it;
> +
> +       BUILD_BUG_ON(sizeof(struct bpf_iter_css_kern) != sizeof(struct bpf_iter_css));
> +       BUILD_BUG_ON(__alignof__(struct bpf_iter_css_kern) != __alignof__(struct bpf_iter_css));
> +       kit->root = root;
> +       kit->pos = NULL;
> +       kit->flag = flag;
> +       rcu_read_lock();

Same request as in previous patch.
let's make bpf prog do explicit bpf_rcu_read_lock() instead.