[systemd-devel] [PATCH] test: bump KDBUS_CONN_MAX_MSGS_PER_USER value

[David Herrmann]

Hi

On Tue, Feb 17, 2015 at 12:35 PM, Lukasz Skalski <l.skalski at samsung.com> wrote:
> On 02/17/2015 12:20 PM, Djalal Harouni wrote:
>> Hi Lukasz,
>>
>
> Hi Djalal,
>
>> On Tue, Feb 17, 2015 at 11:37:53AM +0100, Lukasz Skalski wrote:
>>> diff --git a/test/test-message.c b/test/test-message.c
>>> index 03ac71e..0cae942 100644
>>> --- a/test/test-message.c
>>> +++ b/test/test-message.c
>>> @@ -28,7 +28,7 @@
>>>   * maximum number of queued messages from the same indvidual user after the
>>>   * the un-accounted value has been hit
>>>   */
>>> -#define KDBUS_CONN_MAX_MSGS_PER_USER                16
>>> +#define KDBUS_CONN_MAX_MSGS_PER_USER                128
>> Actually, the limit bump is just a temporary solution, to inspect or fix
>> other bugs related to Gnome. David is working on this, I will let him
>> decide if he wants to apply this patch or let it as it...
>>
>
> Yes, I know that it is temporary solution, but at this moment, after
> bumping KDBUS_CONN_MAX_MSGS_PER_USER value we have some new errors in
> tests - for example in test-message.c:736 we get EXFULL errno
> (queue.c:287 says that we can't take more than half of the remaining
> space) instead of ENOBUFS. IMO we should align tests with latest changes.

For the archives (as said on IRC): The old quota checks were easily
exploitable, but on the other hand way too restrictive. On some fast
machines, 32 queued messages exceeds the polkit queues on normal
boots. That's why I bumped the limit for now.

The patch you posted is technically correct (and I have the same
queued locally since yesterday). However, the limit is so high, that
you basically exceed the 10M pool size and thus cannot really test for
the quota limits.

I'm reworking the whole quota logic. I hope I can push it later today.

Thanks
David