>> static int netif_receive_skb_internal(struct sk_buff *skb)
>> {
>> int ret;
>> @@ -4258,6 +4336,21 @@ static int netif_receive_skb_internal(struct sk_buff
>> *skb)
>>
>> rcu_read_lock();
>>
>> + if (static_key_false(&generic_xdp_needed)) {
>> + struct bpf_prog *xdp_prog =
>> rcu_dereference(skb->dev->xdp_prog);
>> +
>> + if (xdp_prog) {
>> + u32 act = netif_receive_generic_xdp(skb, xdp_prog);
>
> That's indeed the best attachment point in the stack.
> I was trying to see whether it can be lowered into something like
> dev_gro_receive(), but not everyone calls it.
It would be a helpful (follow-on) optimization for packets that do
pass through it. It allows skb recycling with napi_reuse_skb and can
be used to protect if a vulnerability in the gro stack pops up.
> Another option to put it into eth_type_trans() itself, then
> there are no problems with gro, l2 headers, and adjust_head,
> but changing all drivers is too much.
>
>> +
>> + if (act != XDP_PASS) {
>> + rcu_read_unlock();
>> + if (act == XDP_TX)
>> + dev_queue_xmit(skb);
>
> It should be fine. For cls_bpf we do recursion check __bpf_tx_skb()
> but I forgot specific details. May be here it's fine as-is.
> Daniel, do we need recursion check here?
That limiter is for egress redirecting to egress, I believe. This
ingress to egress will go through netif_rx and a softirq if looping.
Another point on redirect is clearing skb state. queue_mapping and
sender_cpu will be dirty, but should be able to handle it. It seems
possible to attach to a virtual device, such as a tunnel. In that case
the packet may have gone through a complex receive path before
reaching the tunnel, including tc ingress, so even more skb fields may
be set (e.g., priority). The same holds for act_mirred or
__bpf_redirect, so I assume that this is safe.
