Re: refactoring ip_output() and the L2 _output()

[Ryota Ozaki <ozaki-r%netbsd.org@localhost>]

On Tue, May 26, 2015 at 11:00 PM, Ryota Ozaki <ozaki-r%netbsd.org@localhost> wrote:
> On Tue, May 26, 2015 at 7:55 PM, Mihai Chelaru <chelaru%gmail.com@localhost> wrote:
>> On 26/05/15 05:42, Ryota Ozaki wrote:
>>>
>>> Hi,
>>>
>>> The mail subject may recall someone an old thread:
>>> https://mail-index.netbsd.org/tech-net/2013/02/01/msg003847.html
>>>
>>> Yes, I'm taking over the task :)
>>>
>>>
>>> http://www.netbsd.org/~ozaki-r/refactor_l2_output.diff
>>>
>>> The patch is basically the same as dyoung's patch with some
>>> tweaks to make it work on -current; it gets rid of route
>>> lookups from L2 output routines such as ether_output
>>> and puts them between ip_output and L2 output routines.
>>>
>>> One known issue of the change is that it adds an mbuf flag
>>> to tell ether_output that mbuf is to be set MPLS ether
>>> type. I don't think it's the best way but I don't have
>>> another better approach for now.
>>>
>>> Any comments or suggestions are welcome.
>>>
>>> Thanks,
>>>    ozaki-r
>>>
>>
>> Hi there,
>>
>> I have some questions/observations:
>>
>> * This cod is generic enough to be linked apart from ip_output.c Maybe it
>> can be reused by other protocols as well ? (hint nd6_output)
>
> I think so, but not tried yet. I'll try tomorrow.

Hmm, I was wrong. This is not so easy to reuse it for nd6_output.
Doing next hop determination in nd6_output is not so bad as
compared to doing it in ether_output. So I want to keep as is
for now and would revisit if we need.

  ozaki-r

>
>> * did this patch passed the current net tests ?
>
> Yes!
>
>>
>> Now about MPLS:
>>
>> * I'm kinda reluctant in using flags to describe protocol specifics, maybe
>> should use a tag for now instead of flags even if this interface is slower ?
>
> Sure. So we can postpone to add something. We would have a better solution then.
>
>> * I think flagging/tagging should also be used in mpls LSE - probably in
>> send_frame.
>
> I'm not sure. Without flagging/tagging for LSE, all tests of MPLS pass.
>
>> * and also in ip6 output path. I'll try to find time this week to write some
>> tests for ip6/mpls - but you can test using route.
>> * mpls/gre needs also to be hacked because it uses the same test (have to
>> write a test for mpls+gre, too)
>
> Such tests are welcome! We need much more tests for networking.
>
> Thanks,
>   ozaki-r