[CRIU] [RFC PATCH 3/5] cr-restore: restore root sid before mount namespace

[Stanislav Kinsburskiy]

23.11.2015 20:22, Pavel Emelyanov пишет:
>
>> Note: there is another way, how to set process as a group leader. We can fork
>> a child with CLONE_VM, and use it to mount all the mount points.
>>
>> Signed-off-by: Stanislav Kinsburskiy <skinsbursky at virtuozzo.com>
>> ---
>>   cr-restore.c |    8 +++++++-
>>   1 file changed, 7 insertions(+), 1 deletion(-)
>>
>> diff --git a/cr-restore.c b/cr-restore.c
>> index 06fceb6..9aec077 100644
>> --- a/cr-restore.c
>> +++ b/cr-restore.c
>> @@ -1441,7 +1441,10 @@ static int create_children_and_session(void)
>>   			return ret;
>>   	}
>>   
>> -	restore_sid();
>> +	if (getpid() == 1)
>> +		pr_info("Skipping restore_sid() for init\n");
>> +	else
>> +		restore_sid();
>>   
>>   	pr_info("Restoring children in our session:\n");
>>   	list_for_each_entry(child, &current->children, sibling) {
>> @@ -1524,6 +1527,9 @@ static int restore_task_with_children(void *_arg)
>>   		if (restore_finish_stage(CR_STATE_RESTORE_NS) < 0)
>>   			goto err;
>>   
>> +		pr_info("Calling restore_sid() for init\n");
>> +		restore_sid();
> In the patch comment you say that process has to be group leader, but
> make it session leader instead. Why?

Making process a session leader makes it a group leader.
While restoring process group leader before session leads to session 
restore failure, because of the following hunk:

SYSCALL_DEFINE0(setsid)
{
         struct task_struct *group_leader = current->group_leader;
         struct pid *sid = task_pid(group_leader);
         pid_t session = pid_vnr(sid);
         int err = -EPERM;

<snip>

         /* Fail if a process group id already exists that equals the
          * proposed session id.
          */
         if (pid_task(sid, PIDTYPE_PGID))
                 goto out;

<snip>
out:
<snip>
         return err;
}