On Tue, Dec 13, 2016 at 05:16:26PM -0800, Stephen Hemminger wrote:

On Tue, 13 Dec 2016 01:12:32 +0100 (CET)
Michał Mirosław [off-list ref] wrote:

quoted

This series removes an abuse of VLAN CFI bit in Linux networking stack.
Currently Linux always clears the bit on outgoing traffic and presents
it cleared to userspace (even via AF_PACKET/tcpdump when hw-accelerated).

This uses a new vlan_present bit in struct skbuff, and removes an assumption
that vlan_proto != 0 when VLAN tag is present.

As I can't test most of the driver changes, please look at them carefully.

The series is supposed to be bisect-friendly and that requires temporary
insertion of #define VLAN_TAG_PRESENT in BPF code to be able to split
JIT changes per architecture.

I wonder if CFI can every validly be non-zero in the modern world, on Hyper-V.
There are no token ring devices and that seems to be the only use case where CFI would
be non-zero. Unless someone is planning to reuse it a a protocol bit which seems
like a really bad idea.

Maybe the right thing is to keep hard coded as zero and not start adding
more untestable code conditions.

My recommendation would be get rid of VLAN_TAG_PRESENT, but don't preserve
CFI bit.

According to Wikipedia page [1] on 802.1Q, CFI bit got already changed
to DEI (Drop eligible indicator) in 2011 revision of the IEEE standard.

I can't verify this, though.

Best Regards,
Michał Mirosław

[1] https://en.wikipedia.org/wiki/IEEE_802.1Q#Frame_format