Re: [PATCH v2] packet: Deliver VLAN TPID to userspace

[Atzm Watanabe <atzm@stratosphere.co.jp>]

At Wed, 11 Dec 2013 13:19:21 -0000,
David Laight wrote:
> 
> > From: Atzm Watanabe
> ...
> > > You've defined a new header variant, but all the code seems to rely
> > > on the fact that the 'new' variant is identical to the old one
> > > with the addition of an extra field at the end.
> > >
> > > In which case it ought to be valid to just extend the old header variant.
> > > If the header really can change format there ought to be a discriminating
> > > member somewhere - which you don't seem to have changed.
> > 
> > Yes.  I think that struct tpacket3_hdr can grow safely until 48 bytes,
> > so I'd just like to extend tpacket_hdr_variant1 like you said.
> > 
> > But in the past discussions, there were mentions that a new member
> > cannot be added into struct tpacket_hdr_variant1, and possibly the
> > variant which includes a new member should be separated from the old
> > one.
> > 
> >   http://patchwork.ozlabs.org/patch/284671/
> >   http://patchwork.ozlabs.org/patch/285382/
> 
> I don't remember it being mentioned earlier that there are pad bytes
> following the header.
> 
> Might be worth explicitly defining the pad bytes and zeroing them.
> That could make additional changes easier.

Agreed.
But I think that struct tpacket{2,3}_hdr can be padded but struct
tpacket_hdr is not, because its size may differ depending on
architectures.
So I'll try to pad only to struct tpacket{2,3}_hdr.


Thank you.