Re: [PATCH] ipv6: don't call addrconf_dst_alloc again when enable lo

[chenweilong <chenweilong@huawei.com>]

On 2014/1/2 16:23, Hannes Frederic Sowa wrote:
> On Thu, Jan 02, 2014 at 03:58:22PM +0800, chenweilong wrote:
>> On 2014/1/2 14:54, Hannes Frederic Sowa wrote:
>>> On Thu, Jan 02, 2014 at 01:48:46PM +0800, chenweilong wrote:
>>>> On 2013/12/31 11:57, Hannes Frederic Sowa wrote:
>>>>> On Wed, Jun 19, 2013 at 11:05:32PM -0700, David Miller wrote:
>>>>>> From: Gao feng <gaofeng@cn.fujitsu.com>
>>>>>> Date: Sun, 16 Jun 2013 11:14:30 +0800
>>>>>>
>>>>>>> If we disable all of the net interfaces, and enable
>>>>>>> un-lo interface before lo interface, we already allocated
>>>>>>> the addrconf dst in ipv6_add_addr. So we shouldn't allocate
>>>>>>> it again when we enable lo interface.
>>>>>>>
>>>>>>> Otherwise the message below will be triggered.
>>>>>>> unregister_netdevice: waiting for sit1 to become free. Usage count = 1
>>>>>>>
>>>>>>> This problem is introduced by commit 25fb6ca4ed9cad72f14f61629b68dc03c0d9713f
>>>>>>> "net IPv6 : Fix broken IPv6 routing table after loopback down-up"
>>>>>>>
>>>>>>> Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
>>>>>>
>>>>>> This is the second such regression added by that commit :-/
>>>>>>
>>>>>> Applied and queue up for -stable, thanks.
>>>>>
>>>>> Hmm, and this change also has a regression and breaks the original fix. :/
>>>>>
>>>>> https://bugzilla.kernel.org/show_bug.cgi?id=67951
>>>>>
>>>>> I tried to track it down but it seems pretty complicated. Maybe we have to
>>>>> special-case the take-down of the loopback device.
>>>>>
>>>>>
>>>>>
>>>>
>>>> When I did the tests,If 'ifconfig lo down',all IPv6 connection broken,
>>>> but IPv4 connection were still OK.
>>>>
>>>> Is it designed like that or a bug?
>>>
>>> This seems to solve the loopback up/down problem, but there are still
>>> some issues with up/down of interfaces and routing table interactions.
>>>
>>> We enable routes over interfaces when interface is actually down and
>>> kick manually specified on-link routes when we actually should try to
>>> keep them and just disable them.
>>>
>>> diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
>>> index 6c16345..61d752a 100644
>>> --- a/net/ipv6/addrconf.c
>>> +++ b/net/ipv6/addrconf.c
>>> @@ -2629,8 +2629,10 @@ static void init_loopback(struct net_device *dev)
>>>  			if (sp_ifa->flags & (IFA_F_DADFAILED | IFA_F_TENTATIVE))
>>>  				continue;
>>>  
>>> -			if (sp_ifa->rt)
>>> +			if (sp_ifa->rt) {
>>> +				ip6_ins_rt(sp_ifa->rt);
>>>  				continue;
>>> +			}
>>>  
>>>  			sp_rt = addrconf_dst_alloc(idev, &sp_ifa->addr, false);
>>>  
>>>
>>> .
>>>
>> I test the patch,it has the problem Gao feng reported.
> 
> Thanks for testing. I wonder why.
> 
>> How about this:
>>
>> diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
>> index d5fa5b8..5e2db6e 100644
>> --- a/net/ipv6/addrconf.c
>> +++ b/net/ipv6/addrconf.c
>> @@ -2609,10 +2609,13 @@ static void init_loopback(struct net_device *dev)
>>
>>                         if (sp_ifa->flags & (IFA_F_DADFAILED | IFA_F_TENTATIVE))
>>                                 continue;
>>
>> -                       if (sp_ifa->rt)
>> -                               continue;
>> +                       if (sp_ifa->rt && sp_ifa->rt->dst.dev == dev){
>> +                               ip6_del_rt(sp_ifa->rt);
>> +                       }
>>
> 
> It could work, but it looks like a band-aid for another problem to me. I am
> not sure if it is in init_loopback, yet.
> 
> 
> .
> 
Fix some sapce errors in my post.

diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
index 62d1799..d2f8c0a 100644
--- a/net/ipv6/addrconf.c
+++ b/net/ipv6/addrconf.c
@@ -2422,8 +2422,9 @@ static void init_loopback(struct net_device *dev)
 			if (sp_ifa->flags & (IFA_F_DADFAILED | IFA_F_TENTATIVE))
 				continue;

-			if (sp_ifa->rt)
-				continue;
+			if (sp_ifa->rt && sp_ifa->rt->dst.dev == dev) {
+				ip6_del_rt(sp_ifa->rt);
+			}

 			sp_rt = addrconf_dst_alloc(idev, &sp_ifa->addr, 0);