Re: [PATCH net-next] net: filter: rename 'struct sk_filter' to 'struct bpf_prog'

[David Miller <davem@davemloft.net>]

From: Alexei Starovoitov <ast@plumgrid.com>
Date: Mon, 28 Jul 2014 18:12:05 -0700

> On Mon, Jul 28, 2014 at 2:45 PM, Pablo Neira Ayuso <pablo@netfilter.org> wrote:
>>> > struct sk_filter_cb {
>>> >         int type;
>>> >         struct module *me;
>>> >         void (*charge)(struct sock *sk, struct sk_filter *fp);
>>> >         void (*uncharge)(struct sock *sk, struct sk_filter *fp);
>>> >         unsigned int (*run_filter)(struct sk_filter *fp, struct sk_buff *skb);
>>> > };
>>>
>>> Pablo,
>>>
>>> I don't think you understand the scope of BPF.
>>> 'struct module *'? to attach nft to sockets? ouch.
>>
>> The idea is that there will be one sk_filter_cb per socket filtering
>> approach. The structure module is just there in case one of the
>> approach is loadable as kernel module, it's the typical code pattern
>> in the kernel. You can git grep for similar code.
> 
> socket filtering is available to unprivileged users.
> So you're proposing to let them increment refcnt of modules?!
> That's not secure.

It's impossible to avoid, and really is nothing new.

Users can open sockets, and that holds a reference to the module
implementing that protocol.  Is that not secure too?

This discussion is degenerating into nonsense, please stop ignoring
Pablo's core points.

Thanks.