Re: [PATCH RFC v4 bpf-next 08/11] tun: Support xdp in the Tx path for skb

[David Ahern <dsahern@gmail.com>]

On 3/1/20 8:28 PM, Jason Wang wrote:
>> +static u32 tun_do_xdp_tx_generic(struct tun_struct *tun,
>> +                 struct net_device *dev,
>> +                 struct sk_buff *skb)
>> +{
>> +    struct bpf_prog *xdp_prog;
>> +    u32 act = XDP_PASS;
>> +
>> +    xdp_prog = rcu_dereference(tun->xdp_egress_prog);
>> +    if (xdp_prog) {
>> +        struct xdp_txq_info txq = { .dev = dev };
>> +        struct xdp_buff xdp;
>> +
>> +        skb = tun_prepare_xdp_skb(skb);
>> +        if (!skb) {
>> +            act = XDP_DROP;
>> +            goto out;
>> +        }
>> +
>> +        xdp.txq = &txq;
>> +
>> +        act = do_xdp_generic_core(skb, &xdp, xdp_prog);
>> +        switch (act) {
>> +        case XDP_TX:    /* for Tx path, XDP_TX == XDP_PASS */
>> +            act = XDP_PASS;
>> +            break;
> 
> 
> Jute a note here, I agree for TX XDP it may be better to do this.
> 
> But for offloaded program we need different semantic. Or we can deal
> this with attach types?
> 

This path is for XDP_FLAGS_DRV_MODE. Offloaded programs
(XDP_FLAGS_HW_MODE) will be run in vhost context.