Re: bridge new vlan and iftagged "none"

hi Roland,

Ronald Klop:
> With VLANFILTER disabled epair4a did receive traffic and also
> broadcasts on vlan 3. I don't know if this is expected.

> Interestingly, with VLANFILTER disabled the "untagged 3" interfaces
> also saw broadcast traffic which was not destined for vlan 3.

with vlanfilter disabled, all interfaces will receive traffic on all
vlans.  the vlan tag may be added/removed based on the interface's
untagged setting, but no filtering will be done.

so the "untagged 3" interface should see broadcast traffic for (e.g.)
vlan 1, but that traffic should have a .1q tag attached with the vlan
id set to 1.

note that bridge always behaved like this, that hasn't changed with
the vlan filtering feature except for the new 'untagged' behaviour.

the bug i noticed here (which the patch is meant to fix) is that when
vlanfilter is enabled, traffic originated from the host, rather than a
bridge interface, might be sent to a bridge port when it should have
been filtered.  this mostly affects BUM traffic since the incoming
filtering prevents ND/ARP from working.

> With the story above is the patch still needed? I will test anyway to
> see what happens. It is a RPI4, so compiling is a bit slow.

i was able to reproduce the problem here so it's not too important to
test that if it's a hassle.

to answer your other question:

> I realize that I can now configure this to sent "tagged 1" traffic
> between genet0 and the switch and even further into my network. Would
> that have /any/ influence on performance?

there is overhead on the wire of 4 bytes per frame for the .1q tag.
the actual impact of this depends on your average frame size, but in
most cases it's so small it's negligible.

there is also some overhead involved in processing the tag itself.  on
the switch side, and on modern host network interfaces, this is done in
hardware so there should be no impact on performance, but i don't think
genet(4) supports this, meaning the kernel has to do tag processing in
software.  however, even on the sort of hardware that uses genet (i.e.,
slow CPUs), i don't think you'd notice this overhead in practice on a
1Gbps interface.