Re: [PATCH net-next] rocker: check for BRIDGE_FLAGS_SELF in bridge setlink handler

sorry, this email might have gone with non-text (though i selected
'plain text' in my google mail web client). Retrying...apologize for
multiple emails.

On Mon, Mar 9, 2015 at 3:56 PM, Roopa Prabhu [off-list ref] wrote:

On Mon, Mar 9, 2015 at 8:59 AM, Arad, Ronen [off-list ref] wrote:

quoted

quoted

-----Original Message-----
From: netdev-owner@vger.kernel.org [mailto:netdev-owner@vger.kernel.org]
On
Behalf Of Jiri Pirko

quoted

I also think that is is fine that in case either master or self are not
set,
driver ndo setlink is called as well (in case the offload bit is on). It
makes it transparent for user in case he does not care what in under. In
case he cares, he can specify master or self to achieve exactly what he
needs. I like it that way.

It would be nice to have switch driver implementation that could work
Consistently with and without the bridge driver. I expect use case where a
bridge is only used when L3 is needed or to leverage L2 protocols that are
implemented by the bridge driver (e.g. STP, IGMP snooping).

quoted

If my driver can process setlink requests using SELF flag today, I'd like
it
to work the same when setlink is propagated down from a bridge master or
from a team/bond master.

quoted

There is current issue with notification.
When SELF flag is set, the bridge driver does not offload to the port
driver
and notification is issued by rtnetlink.c by calling the port getlink ndo.

quoted

This call, however, is done with zero filter_mask such that VLAN
information
is not included in the notification.

quoted

When SELF flag is not set, bridge driver offloads to the port and issues
the
notification. In that case it sets the filter to VLAN_COMPRESSED.
I don't see how I can get my driver to behave consistently with and
without
a bridge.
The closest I can get notification with and without a bridge is for the
driver to examine the flags. If SELF is set, the driver knows it got
invoked
directly from rtnetlink and it should notify VLAN setting as the
subsequent
notification triggered by rtnetlink won't.
When the switch driver does not see SELF set it knows that it was invoked
by
The bridge driver which already takes care of notification including VLAN
information.

You are right about self being handled from rtnetlink.c and the notification
triggered from rtnetlink.c
can be different from the one sent from the bridge driver.
In the specific case of vlan information not being present in the self
notifications generated from rtnetlink.c, that is
something that will need to be fixed eventually.
Since rtnetlink.c generates these notifications and not the port driver,
like you say, port driver could generate the appropriate notification or
rtnetlink.c will have to get the size of the skb from the port driver before
doing a bridge_getlink call into the port driver.

This is another good example, where the port driver might need to look at
the self flag after-all.

Thanks,
Roopa