Re: [PATCH bpf-next v2 1/3] bpf: Expose bpf_get_socket_cookie to tracing programs

[Daniel Borkmann]

On 12/8/20 8:30 PM, Florent Revest wrote:
> On Fri, 2020-12-04 at 20:03 +0100, Daniel Borkmann wrote:
> > On 12/4/20 7:56 PM, Daniel Borkmann wrote:
> > > On 12/3/20 10:33 PM, Florent Revest wrote:
> > > > This creates a new helper proto because the existing
> > > > bpf_get_socket_cookie_sock_proto has a ARG_PTR_TO_CTX argument
> > > > and only
> > > > works for BPF programs where the context is a sock.
> > > >
> > > > This helper could also be useful to other BPF program types such
> > > > as LSM.
> > > >
> > > > Signed-off-by: Florent Revest <revest@xxxxxxxxxx>
> > > > ---
> > > >    include/uapi/linux/bpf.h       | 7 +++++++
> > > >    kernel/trace/bpf_trace.c       | 4 ++++
> > > >    net/core/filter.c              | 7 +++++++
> > > >    tools/include/uapi/linux/bpf.h | 7 +++++++
> > > >    4 files changed, 25 insertions(+)
> > > >
> > > > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > > > index c3458ec1f30a..3e0e33c43998 100644
> > > > --- a/include/uapi/linux/bpf.h
> > > > +++ b/include/uapi/linux/bpf.h
> > > > @@ -1662,6 +1662,13 @@ union bpf_attr {
> > > >     *     Return
> > > >     *         A 8-byte long non-decreasing number.
> > > >     *
> > > > + * u64 bpf_get_socket_cookie(void *sk)
> > > > + *     Description
> > > > + *         Equivalent to **bpf_get_socket_cookie**\ () helper
> > > > that accepts
> > > > + *         *sk*, but gets socket from a BTF **struct sock**.
> > > > + *     Return
> > > > + *         A 8-byte long non-decreasing number.
> > >
> > > I would not mention this here since it's not fully correct and we
> > > should avoid users taking non-decreasing granted in their progs.
> > > The only assumption you can make is that it can be considered a
> > > unique number. See also [0] with reverse counter..
> > >
> > >     [0]
> > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=92acdc58ab11af66fcaef485433fde61b5e32fac
>
> Ah this is a good point, thank you! I will send a v3 with an extra
> patch that s/non-decreasing/unique/ in the other descriptions. I had
> not given it any extra thought, I just stupidly copied/pasted existing
> descriptions. :)
>
> > One more thought, in case you plan to use this from sleepable
> > context, you would need to use sock_gen_cookie() variant in the BPF
> > helper instead.
>
> Out of curiosity, why don't we just always call sock_gen_cookie? Is it
> to avoid the performance impact of increasing the preempt counter and
> introducing a memory barriers ?

Yes, all the other contexts where the existing helpers are used already have
preemption disabled, so the extra preempt_{disable,enable}() is unnecessary
overhead given we want the cookie generation be efficient.

Thanks,
Daniel