On Thu, 3 Dec 2020 10:25:41 -0800
sdf@xxxxxxxxxx wrote:
> > +BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
> > + u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
> > +{
> > + int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
> > + struct net_device *dev = skb->dev;
> > + int len;
> > + int mtu;
> > +
> > + if (flags & ~(BPF_MTU_CHK_SEGS))
> > + return -EINVAL;
> > +
> > + dev = __dev_via_ifindex(dev, ifindex);
> > + if (!dev)
> > + return -ENODEV;
> > +
> > + mtu = READ_ONCE(dev->mtu);
> > +
> > + /* TC len is L2, remove L2-header as dev MTU is L3 size */
>
> [..]
> > + len = skb->len - ETH_HLEN;
> Any reason not to do s/ETH_HLEN/dev->hard_header_len/ (or min_header_len?)
> thought this patch?
Will fix in V9.
There is a very small (performance) overhead, but mostly because
net_device struct layout have placed mtu and hard_header_len on
different cache-lines. (This is something that should be fixed
separately).
--
Best regards,
Jesper Dangaard Brouer
MSc.CS, Principal Kernel Engineer at Red Hat
LinkedIn: http://www.linkedin.com/in/brouer
