Re: [RFC PATCH 4/5] net: filter: run cgroup eBPF programs

[Sargun Dhillon <sargun@sargun.me>]

On Wed, Aug 17, 2016 at 04:00:47PM +0200, Daniel Mack wrote:
> If CONFIG_CGROUP_BPF is enabled, and the cgroup associated with the
> receiving socket has an eBPF programs installed, run them from
> sk_filter_trim_cap().
> 
> eBPF programs used in this context are expected to either return 1 to
> let the packet pass, or != 1 to drop them. The programs have access to
> the full skb, including the MAC headers.
> 
> This patch only implements the call site for ingress packets.
> 
> Signed-off-by: Daniel Mack <daniel@zonque.org>
> ---
>  net/core/filter.c | 44 ++++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 44 insertions(+)
> 
> diff --git a/net/core/filter.c b/net/core/filter.c
> index c5d8332..a1dd94b 100644
> --- a/net/core/filter.c
> +++ b/net/core/filter.c
> @@ -52,6 +52,44 @@
>  #include <net/dst.h>
>  #include <net/sock_reuseport.h>
>  
> +#ifdef CONFIG_CGROUP_BPF
> +static int sk_filter_cgroup_bpf(struct sock *sk, struct sk_buff *skb,
> +				enum bpf_attach_type type)
> +{
> +	struct sock_cgroup_data *skcd = &sk->sk_cgrp_data;
> +	struct cgroup *cgrp = sock_cgroup_ptr(skcd);
> +	struct bpf_prog *prog;
> +	int ret = 0;
> +
> +	rcu_read_lock();
> +
> +	switch (type) {
> +	case BPF_ATTACH_TYPE_CGROUP_EGRESS:
> +		prog = rcu_dereference(cgrp->bpf_egress);
> +		break;
> +	case BPF_ATTACH_TYPE_CGROUP_INGRESS:
> +		prog = rcu_dereference(cgrp->bpf_ingress);
> +		break;
> +	default:
> +		WARN_ON_ONCE(1);
> +		ret = -EINVAL;
> +		break;
> +	}
> +
> +	if (prog) {
> +		unsigned int offset = skb->data - skb_mac_header(skb);
> +
> +		__skb_push(skb, offset);
> +		ret = bpf_prog_run_clear_cb(prog, skb) > 0 ? 0 : -EPERM;
> +		__skb_pull(skb, offset);
> +	}
> +
> +	rcu_read_unlock();
> +
> +	return ret;
> +}
> +#endif /* !CONFIG_CGROUP_BPF */
> +
>  /**
>   *	sk_filter_trim_cap - run a packet through a socket filter
>   *	@sk: sock associated with &sk_buff
> @@ -78,6 +116,12 @@ int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
>  	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
>  		return -ENOMEM;
>  
> +#ifdef CONFIG_CGROUP_BPF
> +	err = sk_filter_cgroup_bpf(sk, skb, BPF_ATTACH_TYPE_CGROUP_INGRESS);
> +	if (err)
> +		return err;
> +#endif
> +
>  	err = security_sock_rcv_skb(sk, skb);
>  	if (err)
>  		return err;
> -- 
> 2.5.5
> 

So, casually looking at this patch, it looks like you're relying on 
sock_cgroup_data, which only points to the default hierarchy. If someone uses 
net_prio or net_classid, cgroup_sk_alloc_disable is called, and this wont work 
anymore. 

Any ideas on how to work around that? Does it make sense to add another pointer 
to sock_cgroup_data, or at least a warning when allocation is disabled?