Re: [PATCH bpf-next v5 2/3] libbpf: add low level TC-BPF API

[Andrii Nakryiko <andrii.nakryiko@gmail.com>]

On Fri, Apr 30, 2021 at 11:32 PM Kumar Kartikeya Dwivedi
<memxor@gmail.com> wrote:
>
> On Sat, May 01, 2021 at 01:05:40AM IST, Andrii Nakryiko wrote:
> > On Wed, Apr 28, 2021 at 9:26 AM Kumar Kartikeya Dwivedi
> > <memxor@gmail.com> wrote:
> > >
> > > This adds functions that wrap the netlink API used for adding,
> > > manipulating, and removing traffic control filters.
> > >
> > > An API summary:
> > >
> > > A bpf_tc_hook represents a location where a TC-BPF filter can be
> > > attached. This means that creating a hook leads to creation of the
> > > backing qdisc, while destruction either removes all filters attached to
> > > a hook, or destroys qdisc if requested explicitly (as discussed below).
> > >
> > > The TC-BPF API functions operate on this bpf_tc_hook to attach, replace,
> > > query, and detach tc filters.
> > >
> > > All functions return 0 on success, and a negative error code on failure.
> > >

[...]

> > >
> > > Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
> > > Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
> > > ---
> >
> > API looks good to me (except the flags field that just stands out).
> > But I'll defer to Daniel to make the final call.
> >
> > >  tools/lib/bpf/libbpf.h   |  41 ++++
> > >  tools/lib/bpf/libbpf.map |   5 +
> > >  tools/lib/bpf/netlink.c  | 463 ++++++++++++++++++++++++++++++++++++++-
> > >  3 files changed, 508 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
> > > index bec4e6a6e31d..3de701f46a33 100644
> > > --- a/tools/lib/bpf/libbpf.h
> > > +++ b/tools/lib/bpf/libbpf.h
> > > @@ -775,6 +775,47 @@ LIBBPF_API int bpf_linker__add_file(struct bpf_linker *linker, const char *filen
> > >  LIBBPF_API int bpf_linker__finalize(struct bpf_linker *linker);
> > >  LIBBPF_API void bpf_linker__free(struct bpf_linker *linker);
> > >
> > > +enum bpf_tc_attach_point {
> > > +       BPF_TC_INGRESS = 1 << 0,
> > > +       BPF_TC_EGRESS  = 1 << 1,
> > > +       BPF_TC_CUSTOM  = 1 << 2,
> > > +};
> > > +
> > > +enum bpf_tc_attach_flags {
> > > +       BPF_TC_F_REPLACE = 1 << 0,
> > > +};
> > > +
> > > +struct bpf_tc_hook {
> > > +       size_t sz;
> > > +       int ifindex;
> > > +       enum bpf_tc_attach_point attach_point;
> > > +       __u32 parent;
> > > +       size_t :0;
> > > +};
> > > +
> > > +#define bpf_tc_hook__last_field parent
> > > +
> > > +struct bpf_tc_opts {
> > > +       size_t sz;
> > > +       int prog_fd;
> > > +       __u32 prog_id;
> > > +       __u32 handle;
> > > +       __u32 priority;
> > > +       size_t :0;
> > > +};
> > > +
> > > +#define bpf_tc_opts__last_field priority
> > > +
> > > +LIBBPF_API int bpf_tc_hook_create(struct bpf_tc_hook *hook, int flags);
> > > +LIBBPF_API int bpf_tc_hook_destroy(struct bpf_tc_hook *hook);
> > > +LIBBPF_API int bpf_tc_attach(const struct bpf_tc_hook *hook,
> > > +                            struct bpf_tc_opts *opts,
> > > +                            int flags);
> >
> > why didn't you put flags into bpf_tc_opts? they are clearly optional
> > and fit into "opts" paradigm...
> >
>
> I can move this into opts, but during previous discussion it was kept outside
> opts by Daniel, so I kept that unchanged.

for bpf_tc_attach() I see no reason to keep flags separate. For
bpf_tc_hook_create()... for extensibility it would need it's own opts
for hook creation. But if flags is 99% the only thing we'll need, then
we can always add extra bpf_tc_hook_create_opts() later.

>
> > > +LIBBPF_API int bpf_tc_detach(const struct bpf_tc_hook *hook,
> > > +                            const struct bpf_tc_opts *opts);
> > > +LIBBPF_API int bpf_tc_query(const struct bpf_tc_hook *hook,
> > > +                           struct bpf_tc_opts *opts);
> > > +
> > >  #ifdef __cplusplus
> > >  } /* extern "C" */
> > >  #endif

[...]

> > > +               return -EINVAL;
> > > +
> > > +       return tc_qdisc_create_excl(hook, flags);
> > > +}
> > > +
> > > +static int tc_cls_detach(const struct bpf_tc_hook *hook,
> > > +                        const struct bpf_tc_opts *opts, bool flush);
> > > +
> > > +int bpf_tc_hook_destroy(struct bpf_tc_hook *hook)
> > > +{
> > > +       if (!hook || !OPTS_VALID(hook, bpf_tc_hook) ||
> > > +           OPTS_GET(hook, ifindex, 0) <= 0)
> > > +               return -EINVAL;
> > > +
> > > +       switch ((int)OPTS_GET(hook, attach_point, 0)) {
> >
> > int casting. Did the compiler complain about that or what?
> >
>
> It complains on -Wswitch, as we switch on values apart from the enum values, but
> I'll see if I can remove it.

ah, because of BPF_TC_INGRESS|BPF_TC_EGRESS? That sucks, of course. An
alternative I guess is just declaring BPF_TC_INGRESS_EGRESS =
BPF_TC_INGRESS | BPF_TC_EGRESS, but I don't know how awful that would
be.

>
> > > +               case BPF_TC_INGRESS:
> > > +               case BPF_TC_EGRESS:
> > > +                       return tc_cls_detach(hook, NULL, true);
> > > +               case BPF_TC_INGRESS|BPF_TC_EGRESS:
> > > +                       return tc_qdisc_delete(hook);
> > > +               case BPF_TC_CUSTOM:
> > > +                       return -EOPNOTSUPP;
> > > +               default:
> > > +                       return -EINVAL;
> > > +       }
> > > +}
> > > +

[...]