Date:   Wed, 15 Dec 2021 08:40:44 -0800
From:   Jakub Kicinski <kuba@kernel.org>
To:     Pavel Begunkov <asml.silence@gmail.com>,
        Alexei Starovoitov <ast@kernel.org>
Cc:     netdev@vger.kernel.org, bpf@vger.kernel.org,
        Daniel Borkmann <daniel@iogearbox.net>,
        Andrii Nakryiko <andrii@kernel.org>,
        Martin KaFai Lau <kafai@fb.com>,
        Song Liu <songliubraving@fb.com>, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v3] cgroup/bpf: fast path skb BPF filtering
Message-ID: <20211215084044.064e6861@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
In-Reply-To: <462ce9402621f5e32f08cc8acbf3d9da4d7d69ca.1639579508.git.asml.silence@gmail.com>
References: <462ce9402621f5e32f08cc8acbf3d9da4d7d69ca.1639579508.git.asml.silence@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Wed, 15 Dec 2021 14:49:18 +0000 Pavel Begunkov wrote:
> +static inline bool
> +__cgroup_bpf_prog_array_is_empty(struct cgroup_bpf *cgrp_bpf,
> +				 enum cgroup_bpf_attach_type type)
> +{
> +	struct bpf_prog_array *array = rcu_access_pointer(cgrp_bpf->effective[type]);
> +
> +	return array == &bpf_empty_prog_array.hdr;
> +}
> +
> +#define CGROUP_BPF_TYPE_ENABLED(sk, atype)				       \
> +({									       \
> +	struct cgroup *__cgrp = sock_cgroup_ptr(&(sk)->sk_cgrp_data);	       \
> +									       \
> +	!__cgroup_bpf_prog_array_is_empty(&__cgrp->bpf, (atype));	       \
> +})
> +

> diff --git a/include/linux/bpf.h b/include/linux/bpf.h
> index e7a163a3146b..0d2195c6fb2a 100644
> --- a/include/linux/bpf.h
> +++ b/include/linux/bpf.h
> @@ -1161,6 +1161,19 @@ struct bpf_prog_array {
>  	struct bpf_prog_array_item items[];
>  };
>  
> +struct bpf_empty_prog_array {
> +	struct bpf_prog_array hdr;
> +	struct bpf_prog *null_prog;
> +};
> +
> +/* to avoid allocating empty bpf_prog_array for cgroups that
> + * don't have bpf program attached use one global 'bpf_empty_prog_array'
> + * It will not be modified the caller of bpf_prog_array_alloc()
> + * (since caller requested prog_cnt == 0)
> + * that pointer should be 'freed' by bpf_prog_array_free()
> + */
> +extern struct bpf_empty_prog_array bpf_empty_prog_array;

mumble mumble, this adds more "fun" dependencies [1] Maybe I'm going
about this all wrong, maybe I should be pulling out struct cgroup_bpf
so that cgroup.h does not need bpf-cgroup, not breaking bpf <-> bpf-cgroup.
Alexei, WDYT?

[1] https://lore.kernel.org/all/20211215061916.715513-2-kuba@kernel.org/