[CRIU] [PATCH 13/14] restore: collect signals from zombies (v2)

Pavel Emelyanov xemul at parallels.com
Fri Jan 25 06:06:16 EST 2013


On 01/18/2013 04:20 PM, Andrew Vagin wrote:
> On Fri, Jan 18, 2013 at 02:36:36PM +0400, Pavel Emelyanov wrote:
>>> +
>>> +	for_each_pstree_item(pi) {
>>> +		if (pi->state == TASK_HELPER)
>>> +			continue;
>>> +		ret = (int) futex_get(&task_entries->nr_in_progress);
>>> +		if (ret < 0)
>>> +			goto out;
>>> +		futex_set_and_wake(&task_entries->pid, pi->pid.virt);
>>> +		pr_info("Wait %d\n", pi->pid.virt);
>>> +		futex_wait_while(&task_entries->nr_in_progress, ret);
>>> +	}
>>> +
>>
>>> @@ -749,6 +755,10 @@ long __export_restore_task(struct task_restore_core_args *args)
>>>  	act.rt_sa_restorer = cr_restore_rt;
>>>  	sys_sigaction(SIGCHLD, &act, NULL, sizeof(rt_sigset_t));
>>>  
>>> +	restore_finish_stage(CR_STATE_RESTORE_PREP_ZOMBIES);
>>> +	futex_wait_until(&task_entries->pid, my_pid);
>>> +	restore_finish_stage(CR_STATE_RESTORE_ZOMBIES);
>>
>> Why do we need to serialize the alive tasks restoration?
> 
> Alive tasks handles SIGCHLD in sigzombie_handler.
> crtools doesn't know who is alive and who is not alive.

I don't like this. Let's orchestrate restoration of zombies only.

>>
>>> +
>>>  	sys_sigaction(SIGCHLD, &args->sigchld_act, NULL, sizeof(rt_sigset_t));
>>>  
>>>  	ret = restore_signals(args->siginfo, args->siginfo_nr, 1);
>>>
>>
>>
> .
> 




More information about the CRIU mailing list