[CRIU] [PATCH CRIU 08/14] restore/net/ipv6: net device and def/all configuration

[Pavel Emelyanov]

On 04/04/2016 05:46 PM, Pavel Tikhomirov wrote:
> 
> 
> On 04/04/2016 05:40 PM, Pavel Emelyanov wrote:
>> On 04/04/2016 04:58 PM, Pavel Tikhomirov wrote:
>>>
>>>
>>> On 04/04/2016 04:38 PM, Pavel Emelyanov wrote:
>>>> On 04/04/2016 01:30 PM, Pavel Tikhomirov wrote:
>>>>>
>>>>>
>>>>> On 04/04/2016 01:16 PM, Pavel Emelyanov wrote:
>>>>>>
>>>>>>> @@ -163,9 +163,11 @@ static int net_conf_op(char *tgt, int *conf, int n, int op, char *proto,
>>>>>>>     			continue;
>>>>>>>     		}
>>>>>>>     		/*
>>>>>>> -		 * If dev conf value is the same as default skip restoring it
>>>>>>> +		 * If dev conf value is the same as default skip restoring it,
>>>>>>> +		 * mtu may be changed by disable_ipv6 so we can not skip
>>>>>>> +		 * it's restore
>>>>>>>     		 */
>>>>>>
>>>>>> No, this means that we should restore mtu after disable_ipv6. And we still can check
>>>>>> for it to match default value.
>>>>>
>>>>> Imagine situation:
>>>>> 1)restore disable_ipv6 for default
>>>>> 2)restore mtu for default
>>>>> 3)create eth0
>>>>> 4)restore disable_ipv6 for eth0
>>>>> 5)restore mtu for eth0
>>>>>
>>>>> After step 3 both default and eth0 mtu are set to default value. But
>>>>> after step 4 eth0's one can possibly change. So do you mean we need to
>>>>> first restore both default/disable_ipv6 and eth0/disable_ipv6 and only
>>>>> after that restore default/mtu ?
>>>>
>>>> No. By step 5 will eth0's mtu match the default mtu?
>>>
>>> By step 5 we will have eth0/mtu(in image) equal to default/mtu(in image)
>>> equal to default/mtu(sysctl) but not equal to eth0/mtu(sysctl).
>>
>> How can this happen? Anything in the image should match the same in sysctl-s
>> on the dump side, so if eth0.mtu == default.mtu in image, then the same is
>> true for sysctls on dump side. Why should it differ on restore?
> 
>  From comment to 06/14 patch:
> 
>  >In addrconf_sysctl_disable->addrconf_disable_ipv6->dev_disable_change:
>  >On addrconf_notify + NETDEV_UP if idev->cnf.mtu6 differs from dev->mtu,
>  >sysctl mtu6 is overwritten.
>  >
>  >So changing disable_ipv6 sysctl for some device may change mtu sysctl
>  >for it...
> 
> eth0.mtu sysctl changed on step 4.

Argh... Let's come with examples. Would you show the combination of eth0/default
mtu/disable_ipv6 -s _in_ _images_ that would be restored wrongly if we don't
make the check you suggest?

-- Pavel