Linux Ethernet Bridging - Dec 2016 - [Bridge] [PATCH net-next] net: remove abuse of VLAN DEI/CFI bit

On Sat,  3 Dec 2016 10:22:28 +0100 (CET)
Micha? Miros?aw <mirq-linux at rere.qmqm.pl> wrote:
> This All-in-one patch removes abuse of VLAN CFI bit, so it can be passed
> intact through linux networking stack.
> 
> Signed-off-by: Micha? Miros?aw <michal.miroslaw at atendesoftware.pl>
> ---
> 
> Dear NetDevs
> 
> I guess this needs to be split to the prep..convert[]..finish sequence,
> but if you like it as is, then it's ready.
> 
> The biggest question is if the modified interface and vlan_present
> is the way to go. This can be changed to use vlan_proto != 0 instead
> of an extra flag bit.
> 
> As I can't test most of the driver changes, please look at them
carefully.
> OVS and bridge eyes are especially welcome.
> 
> Best Regards,
> Micha??Miros?aw
Is the motivation to support 802.1ad Drop Eligability Indicator (DEI)?

If so then you need to be more verbose in the commit log, and lots more
work is needed. You need to rename fields and validate every place a
driver is using DEI bit to make sure it really does the right thing
on that hardware. It is not just a mechanical change.

On Tue, Dec 13, 2016 at 05:21:18PM -0800, Stephen Hemminger
wrote:
> On Sat,  3 Dec 2016 10:22:28 +0100 (CET)
> Micha? Miros?aw <mirq-linux at rere.qmqm.pl> wrote:
> > This All-in-one patch removes abuse of VLAN CFI bit, so it can be
passed
> > intact through linux networking stack.
> > 
> > Signed-off-by: Micha? Miros?aw <michal.miroslaw at
atendesoftware.pl>
> > ---
> > 
> > Dear NetDevs
> > 
> > I guess this needs to be split to the prep..convert[]..finish
sequence,
> > but if you like it as is, then it's ready.
> > 
> > The biggest question is if the modified interface and vlan_present
> > is the way to go. This can be changed to use vlan_proto != 0 instead
> > of an extra flag bit.
> > 
> > As I can't test most of the driver changes, please look at them
carefully.
> > OVS and bridge eyes are especially welcome.
> > 
> > Best Regards,
> > Micha??Miros?aw
> Is the motivation to support 802.1ad Drop Eligability Indicator (DEI)?
> 
> If so then you need to be more verbose in the commit log, and lots more
> work is needed. You need to rename fields and validate every place a
> driver is using DEI bit to make sure it really does the right thing
> on that hardware. It is not just a mechanical change.
My main motivation is to be able to see the bit intact in tcpdump and be
able to pass it untouched through at least a veth pair. It would be great
if all devices didn't do something stupid with the bit, but it's not
something I am able to make happen.

Best Regards,
Micha? Miros?aw

On Tue, Dec 13, 2016 at 05:21:18PM -0800, Stephen Hemminger
wrote:
> On Sat,  3 Dec 2016 10:22:28 +0100 (CET)
> Micha? Miros?aw <mirq-linux at rere.qmqm.pl> wrote:
> 
> > This All-in-one patch removes abuse of VLAN CFI bit, so it can be
passed
> > intact through linux networking stack.
> > 
> > Signed-off-by: Micha? Miros?aw <michal.miroslaw at
atendesoftware.pl>
> > ---
> > 
> > Dear NetDevs
> > 
> > I guess this needs to be split to the prep..convert[]..finish
sequence,
> > but if you like it as is, then it's ready.
> > 
> > The biggest question is if the modified interface and vlan_present
> > is the way to go. This can be changed to use vlan_proto != 0 instead
> > of an extra flag bit.
> > 
> > As I can't test most of the driver changes, please look at them
carefully.
> > OVS and bridge eyes are especially welcome.
> > 
> > Best Regards,
> > Micha??Miros?aw
> 
> Is the motivation to support 802.1ad Drop Eligability Indicator (DEI)?
> 
> If so then you need to be more verbose in the commit log, and lots more
> work is needed. You need to rename fields and validate every place a
> driver is using DEI bit to make sure it really does the right thing
> on that hardware. It is not just a mechanical change.
There are not many mentions of CFI bit in the Linux tree. Places that
used it as VLAN_TAG_PRESENT are fixed with this patchset. Other uses are:

 - VLAN code: ignored
 - ebt_vlan: ignored
 - OVS: cleared because of netlink API assumptions
 - DSA: transferred to/from (E)DSA tag
 - drivers: gianfar: uses properly in filtering rules
 - drivers: cnic: false-positive (uses only VLAN ID, CFI bit marks the field
'valid')
 - drivers: qedr: false-positive (like cnic)

So unless there is something hidden in the hardware, no driver does anything
special with the CFI bit.

After this patchset only OVS will need further modifications to be able to
support handling of DEI bit.

Best Regards,
Micha??Miros?aw