Message-ID: <CAKH8qBsB2epy_s-NFs3TEuNBhXHfq-WuNvAOM9t16SUe=V8RNA@mail.gmail.com>
Date:   Wed, 14 Dec 2022 10:43:51 -0800
From:   Stanislav Fomichev <sdf@...gle.com>
To:     Martin KaFai Lau <martin.lau@...ux.dev>
Cc:     ast@...nel.org, daniel@...earbox.net, andrii@...nel.org,
        song@...nel.org, yhs@...com, john.fastabend@...il.com,
        kpsingh@...nel.org, haoluo@...gle.com, jolsa@...nel.org,
        David Ahern <dsahern@...il.com>,
        Jakub Kicinski <kuba@...nel.org>,
        Willem de Bruijn <willemb@...gle.com>,
        Jesper Dangaard Brouer <brouer@...hat.com>,
        Anatoly Burakov <anatoly.burakov@...el.com>,
        Alexander Lobakin <alexandr.lobakin@...el.com>,
        Magnus Karlsson <magnus.karlsson@...il.com>,
        Maryam Tahhan <mtahhan@...hat.com>, xdp-hints@...-project.net,
        netdev@...r.kernel.org, bpf@...r.kernel.org
Subject: Re: [PATCH bpf-next v4 05/15] bpf: XDP metadata RX kfuncs

On Tue, Dec 13, 2022 at 5:54 PM Martin KaFai Lau <martin.lau@...ux.dev> wrote:
>
> On 12/12/22 6:35 PM, Stanislav Fomichev wrote:
> > diff --git a/include/linux/bpf.h b/include/linux/bpf.h
> > index ca22e8b8bd82..de6279725f41 100644
> > --- a/include/linux/bpf.h
> > +++ b/include/linux/bpf.h
> > @@ -2477,6 +2477,8 @@ void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
> >                                      struct net_device *netdev);
> >   bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev);
> >
> > +void *bpf_dev_bound_resolve_kfunc(struct bpf_prog *prog, u32 func_id);
> > +
>
> This probably requires an inline version for !CONFIG_NET.

Yeah, not sure why my confings didn't catch this :-(

> > diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
> > index d434a994ee04..c3e501e3e39c 100644
> > --- a/kernel/bpf/core.c
> > +++ b/kernel/bpf/core.c
> > @@ -2097,6 +2097,13 @@ bool bpf_prog_map_compatible(struct bpf_map *map,
> >       if (fp->kprobe_override)
> >               return false;
> >
> > +     /* When tail-calling from a non-dev-bound program to a dev-bound one,
> > +      * XDP metadata helpers should be disabled. Until it's implemented,
> > +      * prohibit adding dev-bound programs to tail-call maps.
> > +      */
> > +     if (bpf_prog_is_dev_bound(fp->aux))
> > +             return false;
> > +
> >       spin_lock(&map->owner.lock);
> >       if (!map->owner.type) {
> >               /* There's no owner yet where we could check for
> > diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c
> > index f714c941f8ea..3b6c9023f24d 100644
> > --- a/kernel/bpf/offload.c
> > +++ b/kernel/bpf/offload.c
> > @@ -757,6 +757,29 @@ void bpf_dev_bound_netdev_unregister(struct net_device *dev)
> >       up_write(&bpf_devs_lock);
> >   }
> >
> > +void *bpf_dev_bound_resolve_kfunc(struct bpf_prog *prog, u32 func_id)
> > +{
> > +     const struct xdp_metadata_ops *ops;
> > +     void *p = NULL;
> > +
> > +     down_read(&bpf_devs_lock);
> > +     if (!prog->aux->offload || !prog->aux->offload->netdev)
>
> This happens when netdev is unregistered in the middle of bpf_prog_load and the
> bpf_offload_dev_match() will eventually fail during dev_xdp_attach()? A comment
> will be useful.

Right, that's the expectation - we load/verify the prog but it's
essentially un-attach-able.
Will try to clarify here.

> > +             goto out;
> > +
> > +     ops = prog->aux->offload->netdev->xdp_metadata_ops;
> > +     if (!ops)
> > +             goto out;
> > +
> > +     if (func_id == xdp_metadata_kfunc_id(XDP_METADATA_KFUNC_RX_TIMESTAMP))
> > +             p = ops->xmo_rx_timestamp;
> > +     else if (func_id == xdp_metadata_kfunc_id(XDP_METADATA_KFUNC_RX_HASH))
> > +             p = ops->xmo_rx_hash;
> > +out:
> > +     up_read(&bpf_devs_lock);
> > +
> > +     return p;
> > +}
> > +
> >   static int __init bpf_offload_init(void)
> >   {
> >       int err;
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index 203d8cfeda70..e61fe0472b9b 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> > @@ -15479,12 +15479,35 @@ static int fixup_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
> >                           struct bpf_insn *insn_buf, int insn_idx, int *cnt)
> >   {
> >       const struct bpf_kfunc_desc *desc;
> > +     void *xdp_kfunc;
> >
> >       if (!insn->imm) {
> >               verbose(env, "invalid kernel function call not eliminated in verifier pass\n");
> >               return -EINVAL;
> >       }
> >
> > +     *cnt = 0;
> > +
> > +     if (xdp_is_metadata_kfunc_id(insn->imm)) {
> > +             if (!bpf_prog_is_dev_bound(env->prog->aux)) {
>
> The "xdp_is_metadata_kfunc_id() && (!bpf_prog_is_dev_bound() ||
> bpf_prog_is_offloaded())" test should have been done much earlier in
> add_kfunc_call(). Then the later stage of the verifier does not have to keep
> worrying about it like here.
>
> nit. may be rename xdp_is_metadata_kfunc_id() to bpf_dev_bound_kfunc_id() and
> hide the "!bpf_prog_is_dev_bound() || bpf_prog_is_offloaded()" test into
> bpf_dev_bound_kfunc_check(&env->log, env->prog).
>
> The change in fixup_kfunc_call could then become:
>
>         if (bpf_dev_bound_kfunc_id(insn->imm)) {
>                 xdp_kfunc = bpf_dev_bound_resolve_kfunc(env->prog, insn->imm);
>                 /* ... */
>         }

Makes sense, ty!


> > +                     verbose(env, "metadata kfuncs require device-bound program\n");
> > +                     return -EINVAL;
> > +             }
> > +
> > +             if (bpf_prog_is_offloaded(env->prog->aux)) {
> > +                     verbose(env, "metadata kfuncs can't be offloaded\n");
> > +                     return -EINVAL;
> > +             }
> > +
> > +             xdp_kfunc = bpf_dev_bound_resolve_kfunc(env->prog, insn->imm);
> > +             if (xdp_kfunc) {
> > +                     insn->imm = BPF_CALL_IMM(xdp_kfunc);
> > +                     return 0;
> > +             }
> > +
> > +             /* fallback to default kfunc when not supported by netdev */
> > +     }
> > +
>
>

Powered by blists - more mailing lists