Re: Why does ionice(1) ban the user to set back to 'none' class?

[Bernhard Voelker <mail@bernhard-voelker.de>]

On 06/19/2012 04:11 AM, gnehzuil.lzheng@gmail.com wrote:
> [CC'd to util-linux mailing list]
> 
> On 06/18/2012 09:28 PM, Jens Axboe wrote:
> 
>> On 06/18/2012 01:48 PM, Zheng Liu wrote:
>>> Hi Jens,
>>>
>>> I meet a problem when I use ionice(1) to adjust a process's io priority.
>>> I do the following operations:
>>>
>>> $ ionice -p${pid}
>>> none: prio 0
>>> $ ionice -p${pid} -c2 -n4
>>> $ ionice -p${pid}
>>> best-effort: prio 4
>>> $ ionice -p${pid} -c0 -n0
>>> $ ionice -p${pid}
>>> best-effort: prio 0
>>>
>>> So I cannot set scheduling class back to 'none'.  If I call ioprio_set(2)
>>> directly, it will be fine.  But if I use ionice(1), I cannot change it.  I
>>> read the docs about ionice in [1].  I notice this code:
>>>
>>> switch (ioprio_class) {
>>> 	case IOPRIO_CLASS_NONE:
>>> 		ioprio_class = IOPRIO_CLASS_BE;
>>>                                ^^^^^^^^^^^^^^^^
>>>                                *It means that we cannot set back to none.*
>>> 		break;
>>> 	case IOPRIO_CLASS_RT:
>>> 	case IOPRIO_CLASS_BE:
>>> 		break;
>>> 	case IOPRIO_CLASS_IDLE:
>>> 		ioprio = 7;
>>> 		break;
>>> 	default:
>>> 		printf("bad prio class %d\n", ioprio_class);
>>> 		return 1;
>>> }
>>>
>>> My question is why we need to ban the user to set back to 'none'.  Is there
>>> some reasons?  Thank you.
>>
>> The kernel code does allow it, so it's only in the tool that this
>> restriction exists. I don't think we have a particularly good reason to
>> have it there. This is a tools question, though, not a kernel issue. I
>> suggest you take it up with the util-linux crew. I'm fine with removing
>> this restriction from ionice, especially since you could just roll your
>> own that did it.
>>
> 
> 
> Thanks for your reply.  Hopefully we can get some feedbacks from
> util-linux crew.

That has already been fixed by commit v2.15-45-g5dc9371 more than 2 years ago:

  Author: Jakob Unterwurzacher <jakobunt@gmail.com>  2009-01-17 02:38:32
  Committer: Karel Zak <kzak@redhat.com>  2009-05-27 23:43:18

and the change made it into 2.16.

The current piece of code looks like this:

	switch (ioclass) {
		case IOPRIO_CLASS_NONE:
			if ((set & 1) && !tolerant)
				warnx(_("ignoring given class data for none class"));
			data = 0;
			break;
		case IOPRIO_CLASS_RT:
		case IOPRIO_CLASS_BE:
			break;

We'are currently at version 2.21.2.

Have a nice day,
Berny