Re: [PATCH 2/2] memcg: Close the race between migration and installing bprm->mm as mm

[Eric W. Biederman]

Oleg Nesterov <oleg@xxxxxxxxxx> writes:

> On 05/04, Eric W. Biederman wrote:
>>
>> --- a/fs/exec.c
>> +++ b/fs/exec.c
>> @@ -1044,6 +1044,8 @@ static int exec_mmap(struct mm_struct *mm)
>>  		return 0;
>>  	}
>>  	mmdrop(active_mm);
>> +	/* The tsk may have migrated before the new mm was attached */
>> +	mm_sync_memcg_from_task(tsk);
>>  	return 0;
>>  }
>
> call_usermodehelper(). perhaps it makes sense to change flush_old_exec()
> to clear PF_KTHREAD/etc before exec_mmap().

Yes.  That does look like something to be fixed.  In practice while it
is rare for anything to be migrated during exec I expect it to be even
rarer for kernel threads.  But that is a legitimate issue.

I expect all of these lines after exec_mmap actually belong
in exec_mmap.  Which would then make rearraning the lines easy to ensure
mm_sync_memcg_from_task sees PF_KTHREAD cleared.
	/*
	 * After clearing bprm->mm (to mark that current is using the
	 * prepared mm now), we have nothing left of the original
	 * process. If anything from here on returns an error, the check
	 * in search_binary_handler() will SEGV current.
	 */
	bprm->mm = NULL;

	set_fs(USER_DS);
	current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
					PF_NOFREEZE | PF_NO_SETAFFINITY);
	flush_thread();
	current->personality &= ~bprm->per_clear;

Eric