[CRIU] [PATCH 15/37] files: save mnt_id for reg files

Pavel Emelyanov xemul at parallels.com
Mon Apr 21 02:05:26 PDT 2014


On 04/18/2014 10:45 PM, Andrew Vagin wrote:
> On Fri, Apr 18, 2014 at 07:30:04PM +0400, Pavel Emelyanov wrote:
>> On 04/18/2014 06:42 PM, Andrey Vagin wrote:
>>> Signed-off-by: Andrey Vagin <avagin at openvz.org>
>>> ---
>>>  files-reg.c            | 6 ++++++
>>>  protobuf/regfile.proto | 1 +
>>>  2 files changed, 7 insertions(+)
>>>
>>> diff --git a/files-reg.c b/files-reg.c
>>> index 5b6c4a3..3d316ec 100644
>>> --- a/files-reg.c
>>> +++ b/files-reg.c
>>> @@ -22,6 +22,7 @@
>>>  #include "util.h"
>>>  #include "fs-magic.h"
>>>  #include "asm/atomic.h"
>>> +#include "namespaces.h"
>>>  
>>>  #include "protobuf.h"
>>>  #include "protobuf/regfile.pb-c.h"
>>> @@ -557,6 +558,11 @@ int dump_one_reg_file(int lfd, u32 id, const struct fd_parms *p)
>>>  	} else
>>>  		link = p->link;
>>>  
>>> +	if (p->mnt_id >= 0 && (root_ns_mask & CLONE_NEWNS)) {
>>> +		rfe.mnt_id = p->mnt_id;
>>> +		rfe.has_mnt_id = true;
>>> +	}
>>
>> I would also expect a check like this
>>
>> if (p->mnt_id < 0 && (root_ns_mask & CLONE_NEWNS)) {
>> 	pr_err("Mount namespaces w/o mount ids\n");
>> 	return -1;
>> }
>>
>> Do we have such?
> 
> I think you don't want this, because it will break dumping and restoring
> containers on 3.11.

Fair enough.

>>
>>
>> And also
>>
>> if (find_mount_by_id(p->mnt_id) == NULL) {
>> 	pr_err("No mount for this file in this namespace\n");
>> 	return -1;
>> }
>>
>> How about this?
> 
> This looks reasonable. Thanks.

Cool.

>>
>>> +
>>>  	pr_info("Dumping path for %d fd via self %d [%s]\n",
>>>  			p->fd, lfd, &link->name[1]);
>>>  
>>> diff --git a/protobuf/regfile.proto b/protobuf/regfile.proto
>>> index d975ab0..e91dc44 100644
>>> --- a/protobuf/regfile.proto
>>> +++ b/protobuf/regfile.proto
>>> @@ -6,4 +6,5 @@ message reg_file_entry {
>>>  	required uint64		pos	= 3;
>>>  	required fown_entry	fown	= 5;
>>>  	required string		name	= 6;
>>> +	optional sint32		mnt_id	= 7 [default = -1];
>>>  }
>>>
>>
>>
> .
> 




More information about the CRIU mailing list