Re: [PATCH for-next v4] cgroup/bpf: fast path skb BPF filtering

[Pavel Begunkov]

On 1/27/22 05:08, Alexei Starovoitov wrote:
> On Wed, Jan 26, 2022 at 1:29 PM Pavel Begunkov <asml.silence@xxxxxxxxx> wrote:
> > On 1/26/22 20:30, Martin KaFai Lau wrote:
> > > On Wed, Jan 26, 2022 at 12:22:13AM +0000, Pavel Begunkov wrote:
> > > >    #define BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb)                        \
> > > >    ({                                                                       \
> > > >       int __ret = 0;                                                        \
> > > > -    if (cgroup_bpf_enabled(CGROUP_INET_INGRESS))                  \
> > > > +    if (cgroup_bpf_enabled(CGROUP_INET_INGRESS) && sk &&                  \
> > >   From reading sk_filter_trim_cap() where this will be called, sk cannot be NULL.
> > > If yes, the new sk test is not needed.
> >
> > Well, there is no sane way to verify how it's used considering
> >
> > EXPORT_SYMBOL(__cgroup_bpf_run_filter_skb);
>
> BPF_CGROUP_RUN_PROG_INET_INGRESS() is used in one place.
> Are you folks saying that you want to remove !sk check
> from __cgroup_bpf_run_filter_skb()?
> Seems like micro optimization, but sure why not.
>
> EXPORT_SYMBOL() is there because of "ipv6 as a module" mess.
> So it's not a concern.

Ok. I'd usually assume symbols for internal use only to be GPL.


> Pls tag the subj of the patch with [PATCH bpf-next]
> instead of "for-next".

--
Pavel Begunkov