[CRIU] [PATCH v2 1/8] add netns protobuf entry and image, also add conf to net device entry

Pavel Emelyanov xemul at parallels.com
Fri Sep 26 14:22:18 PDT 2014


On 09/27/2014 01:06 AM, Pavel Emelyanov wrote:
> On 09/26/2014 04:45 PM, Pavel Tikhomirov wrote:
>>
>> On 09/26/2014 09:53 AM, Pavel Emelyanov wrote:
>>> On 09/25/2014 12:00 PM, Pavel Tikhomirov wrote:
>>>
>>>> diff --git a/protobuf-desc.c b/protobuf-desc.c
>>>> index b97418b..61891e0 100644
>>>> --- a/protobuf-desc.c
>>>> +++ b/protobuf-desc.c
>>>> @@ -94,6 +94,7 @@ void cr_pb_init(void)
>>>>   	CR_PB_DESC(IPCNS_MSG_ENT,	IpcMsg,		ipc_msg);
>>>>   	CR_PB_DESC(REMAP_FPATH,		RemapFilePath,	remap_file_path);
>>>>   	CR_PB_DESC(NETDEV,		NetDevice,	net_device);
>>>> +	CR_PB_DESC(NETNS,		Netns,		netns);
>>> If you put the PB_NETNS between the *AUTOGEN* comments in the enum
>>> this line would not be needed.
>>>
>>>>   	CR_PB_MDESC_INIT(cr_pb_descs[PB_PAGEMAP_HEAD],	PagemapHead,	pagemap_head);
>>>>   
>>>>   #include "protobuf-desc-gen.h"
>>>> diff --git a/protobuf/netdev.proto b/protobuf/netdev.proto
>>>> index 4fa23d3..afe76a1 100644
>>>> --- a/protobuf/netdev.proto
>>>> +++ b/protobuf/netdev.proto
>>>> @@ -23,4 +23,11 @@ message net_device_entry {
>>>>   	optional tun_link_entry tun	= 6;
>>>>   
>>>>   	optional bytes address		= 7;
>>>> +
>>>> +	repeated int32 conf		= 8;
>>>> +}
>>>> +
>>>> +message netns_entry {
>>>> +	repeated int32 def_conf		= 1;
>>>> +	repeated int32 all_conf		= 2;
>>> The "all" is not required, it's not status sysctl, but the behavior one.
>>> When you write to it all devices get their setting updated :)
>>
>> Hm.., try:
>> sudo sysctl net.ipv4.conf.lo.disable_policy=0
>> sysctl net.ipv4.conf.all.disable_policy=1
>> cat /proc/sys/net/ipv4/conf/lo/disable_policy
>>  >0
>> I hope that "all" just act like a mask.
> 
> It's not. "All" should propagate the set up value to all other devices.
> "Default" is what newly created device would accept. If this doesn't
> work, report this on the netdev@, things probably got broken :(

Wait a second, it looks like I'm wrong.



More information about the CRIU mailing list