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

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

At Tue, 22 Oct 2013 15:41:10 -0700,
Stephen Hemminger wrote:
> 
> On Tue, 22 Oct 2013 11:56:31 +0900
> Atzm Watanabe <atzm@stratosphere.co.jp> wrote:
> 
> > Hmm...  I think TPACKET{,2,3}_HDRLEN should not be removed without
> > careful considerations.  Because some userspace programs (e.g libpcap)
> > are using them in order to check mmap ability of the kernel...
> 
> 
> That's bad because it means the library then depends on the headers
> of the machine it was built on, not the machine it is running on. I often
> build software on boxes where /usr/include version of kernel headers is out dated.

Yes, this means that building such library with the new header
(TPACKET{,2,3}_HDRLEN were deleted) will make the binary that does
not support PACKET_MMAP. :-(
So, IMHO, we should be prudent in deleting TPACKET{,2,3}_HDRLEN,
otherwise we will break such userspace programs very easily...

e.g. snippet of the latest libpcap-1.4.0 (pcap-linux.c):
   /* check for memory mapped access avaibility. We assume every needed 
    * struct is defined if the macro TPACKET_HDRLEN is defined, because it
    * uses many ring related structs and macros */
  # ifdef TPACKET_HDRLEN
  #  define HAVE_PACKET_RING
  #  ifdef TPACKET2_HDRLEN
  #   define HAVE_TPACKET2
  #  else
  #   define TPACKET_V1   0
  #  endif /* TPACKET2_HDRLEN */
  # endif /* TPACKET_HDRLEN */

It looks like PACKET_MMAP is used when HAVE_PACKET_RING is defined.