Re: [PATCH] net: expose vlan structure to user space

[David Acker <dacker@roinet.com>]

David Acker wrote:
>  From Dave Acker <dacker@roinet.com>
> 
> The purpose is to allow a user space program to use the vlan_ethhdr
> structure when directly handling 802.1Q packets.  This can be done
> by using a raw socket like:
> int s = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
> This socket should see VLAN packets from the base interface unchanged
> (i.e. vlan tag still in packet).
> Currently other user space programs that understand VLANs seem to all
> create their own definitions for this structure.
> This patch does NOT expose the defines related to the sizes of VLAN 
> packets and
> their fields; it only exposes the vlan_ethhdr structure.  It was determined
> that these sizes are not useful since the packet could be as large the MTU
> of the underlying physical network.  Also, you can have nested VLAN headers
> such that the actual data offset is greater than the size of the 
> vlan_ethhdr
> structure.
> -Dave Acker
> 

Hello folks.  Although some conversation appeared on earlier postings of 
this patch, I though I would reply here since the post I am replying to 
here gets the diff format right (from the top level instead of deep in 
the source).  Last I heard:

David S. Miller wrote:
 > I think this really belongs in a portable header file
 > in glibc somewhere.

David, while I can try to work with the glibc folks, I don't understand 
how this patch would be different from what if_ether.h exposes to user 
space today.  Why is one appropriate and the other not?  I am not trying 
to be a pain; I am just trying to understand the rationale so that I can 
explain it to others.

Thanks,
Dave Acker