Re: [PATCH v2 net-next 2/2] tc: make ingress and egress qdiscs consistent

[Daniel Borkmann <daniel@iogearbox.net>]

On 04/08/2015 03:34 PM, Jiri Pirko wrote:
> Wed, Apr 08, 2015 at 03:27:57PM CEST, daniel@iogearbox.net wrote:
>> On 04/08/2015 03:14 PM, Thomas Graf wrote:
>>> On 04/08/15 at 08:58am, Jamal Hadi Salim wrote:
>>>> On 04/08/15 08:31, Daniel Borkmann wrote:
>>>>> That means the tc's cls_u32
>>>>> sample selectors a la ip, ip6, udp, tcp, icmp don't work on ingress
>>>>> either,so in u32 speak you would need to do that by hand, but that
>>>>> doesn't work as you don't have the Ethernet type context available.
>>>>> Am I missing something? :)
>>>>
>>>> u32 works fine. I am sure i have tests which run these on both
>>>> in/egress.
>>>
>>> His point is that an u32 filter written for egress won't work at
>>> ingress because the offsets are different. This has always been the
>>> case and we can't break this behaviour either. I'm sure you have
>>> these weird negative offset u32 egress filters in your repertoire
>>> as well ;-)
>>
>> Okay, you can use negative offsets in cls_u32 to accomodate for
>> that; so yeah, you'd need to implement your filter differently
>> on ingress. That should also work on cls_bpf et al.
>
> That is certainly doable. But is that what we want? I don't think so. I
> would like to have the same for in/eg.

I mean it's certainly a non-obvious hack, where user space has to
fix up something that the kernel should have gotten right in the
first place. :/