Re: [RFC patch net-next-2.6] net: allow multiple rx_handler registration

["Nicolas de Pesloüan" <nicolas.2p.debian@gmail.com>]

Le 01/07/2011 18:49, Ben Greear a écrit :
> On 07/01/2011 09:45 AM, Nicolas de Pesloüan wrote:
>> Le 01/07/2011 17:01, Michał Mirosław a écrit :
>>
>>>>> We could introduce a catch-all macvlan/vlan device that would take
>>>>> addresses/VLANs which are not covered by other configured
>>>>> macvlans/vlans. This would allow clearer configuration and would make
>>>>> the evaluation order explicit. As a bonus, this will give another
>>>>> device to put tcpdump on. ;-)
>>>>
>>>> 'Sounds like what I had in mind in
>>>> http://marc.info/?l=linux-netdev&m=130622112921245&w=2 .
>>>
>>> Almost. My idea assumes that eth0.any won't strip VLAN headers (so its
>>> just looks like a filtered eth0).
>>
>> I originally thought unstripped packets should go to eth0.
>>
>> But, if eth0.any get untagged packets, we face two problems:
>>
>> 1/ We need a way to retrieve the original tag.
>> 2/ We need a way to force the tag on output (or we consider eth0.any a
>> pure tcpdump device, which is less useful).
>>
>> But if eth0.any get the exact same packets as those delivered to eth0,
>> this seems useless.
>>
>> Or maybe, eth0.any should get only packets that weren't delivered to any
>> eth0.XXXX devices... and should be named eth0.unmatched instead of
>> eth0.any :-)
>>
>> Do we need eth0.untagged too (which would only get packets that were
>> originally *not* tagged)?
>>
>> eth0 - Get everything, untouched. (I know several people except tagged
>> packets to be untagged here, but I disagree with this part. eth0 is the
>> raw device and should deliver raw packets, possibly retagging packets
>> that were untagged by hw-accel).
>> eth0.100 - Get VLAN 100 packet, untagged.
>> eth0.untagged - Get only non-tagged packets, untouched.
>> eth0.unmatched - Get only tagged packets, untouched.
>
> Lets let the current vlan tagging changes settle a while before
> adding yet more cruft in this area.

Agreed.

> Packet filters should be able to filter on tags or not, so I don't
> think these extra interfaces would be useful or needed. We may
> need to fix up the sk-filter logic a bit to deal with the
> stripped tags, however.

Agreed too.

	Nicolas.