On Sat, Aug 1, 2020 at 6:31 AM Willem de Bruijn <willemdebruijn.ker...@gmail.com> wrote: > > The kernel interface cannot be changed. If packet sockets used to pass > the first byte up to userspace, they have to continue to do so. > > So I think you can limit the header_ops to only dev_hard_header.
Actually if we want to keep the kernel interface unchanged, we shouldn't implement header_ops for dev_hard_header either, because this changes the way the user space program sends DGRAM packets, too. Before the change the userspace program needs to add the 1-byte header before sending, and after the change the userspace program will let the kernel add the header via dev_hard_header. > Fixes should be small and targeted. Any larger refactoring is > best addressed in a separate net-next patch. I guess the best way for this fix patch would be just add a 0-byte packet check before the driver reads skb->data[0]. Thanks! I'll add the check and re-send the patch.
