[Devel] Re: current state of netns

Denis V. Lunev den at sw.ru
Thu Oct 18 03:11:22 PDT 2007


Daniel Lezcano wrote:
> Eric W. Biederman wrote:
>> "Denis V. Lunev" <den at sw.ru> writes:
>>
>>> Hello, Eric!
>>>
>>> I see that you quite busy and there is no reaction from Dave for your 
>>> latest
>>> portion of netns patches. Right now, me and Pavel are working 
>>> exclusively for
>>> mainstream.
>>>
>>> May be we could bring a torch from your hands and start to push Dave 
>>> Miller even
>>> with IPv4 staff. 3 weeks passed, no reaction for you latest code. 
>>> Looks like it
>>> has been missed somehow... I even have to stop my fingers every day from
>>> touching a generic structures like flowi :)
>>
>> Short summary. - The merge window opened late.
>> - All of the netns code needs to be to Dave Miller before the merge 
>> window.
>> - My last round of changes were not bug fixes and were sent after Dave
>>   had stopped accepting feature additions for 2.6.24
>>
>> Therefore after the merge window when Dave Miller is ready to queue up
>> more networking patches I expect progress can be made again.
>>
>> I think the only thing that is happening is unfortunate timing.
>>
>> I'm not really opposed to people taking my patches or something like
>> them cleaning them up and running with them, I just think the current
>> slow down bad timing.  We have achieved the hard part which is to
>> get the core network namespace infrastructure accepted.
>>
>> On another note.  While I think using CONFIG_NET_NS is nice.  I really
>> only introduced it so that production kernels can avoid enabling an
>> experimental feature.  So far it still looks sane to me to remove
>> CONFIG_NET_NS when things are solid and we can remove the experimental
>> tag.
>>
>> As for ipv4 and ipv6.  However we do that we want to very carefully
>> sequence the patches so that we increasingly make the network
>> namespace infrastructure fine grained.  Similar to make locks fine
>> grained.  I did that for my core network namespaces patches but that
>> careful ordering still needs to happen for my ipv4 patches.
> 
> Denis, Pavel,
> 
> this is great to have you with us for netns. Do you mind if we follow 
> the rule : "patches sent to netdev@ are coming from Eric's git tree, any 
> enhancements are posted to Eric/containers" ?
> So at least, we have the patches stacked and that give us time to review 
> and to test.
> 
> Eric, what do you think about that ?

NETNS49 is outdated in comparison to mainstream now in respect to 
locking and rtnl handling. Some other small features also counts.

I think a re-base should be scheduled for some time.

> By the way, Benjamin and I, we are making ipv6 per namespace. We will 
> send a first patchset for addrconf, ndisc, ip_fib6, fib6_rules probably 
> at the end of the week or at the begin of the next week.
> 
> We are also planning to choose a small patch subset from Eric's tree for 
> ipv4 to be proposed to containers@ before sending it to netdev@ (we 
> should be here very careful and send ipv4, piece by piece, and ensure at 
> all cost init_net_ns will not be broken).
> 
> I don't have a clear idea when the merge window will be closed. I guess, 
> we should resend af_netlink, af_unix and af_packet before sending 
> anything new, like af_inet.

IMHO we should update them to new rtnl and resend.

> Can we coordinate our effort, what do you plan to do ?

Right now there is no concrete plan. We thought about IPv6 also. But, as 
I see two major areas
- prepare finegrained pieces for mainstream
- re-base current infrastructure

Making IPv6 in NETNS49 does not seems good to me. Serious amount of job 
will be lost... Kernel is changed even by us :) And even _NAMESPACE_ 
code is changed during committing process.

At least some areas for generic optimizations are revealed.

Regards,
	Den




More information about the Devel mailing list