16.08.2015 13:30, Erich Titl пишет:
> Hi Andrew
>
> Am 16.08.2015 um 12:25 schrieb Andrew:
>> Hi.
>>
>> 16.08.2015 13:08, Erich Titl пишет:
>>> Hi Andrew
>>>
>>> Am 15.08.2015 um 17:50 schrieb Andrew:
>>>> After month of uptime I've got crash again. It seems that crash caused
>>>> by shaper removing from ppp interface before it's deletion, which 
>>>> causes
>>>> race condition.
>>>
>>> Why does this cause a race condition? You should be able to remove a
>>> shaper without the system crashing.
>>>
>> Interface is shutted down and destroyed, and before this - shaper qdiscs
>> were deleted by script.
>
> Appears logical
>
> This isn't required, in any case on ip-up we
>> also tried to delete root/ingress qdisc before creating new shaper 
>> classes.
>
> Mhhh.... I don't recall the samantics, but why should we delete the 
> root/ingress qdisc for a single inerface?
>
I mean PPP interface. It'll be safer to clean qdisc before creating a 
new one - it will not hurt anything if there's no qdisc atached, and it 
will clean attached earlier qdisc if it's present (for ex., if accel-ppp 
interface caching is used).

>>
>> I think that either accel-ppp doesn't wait for ip-down script
>> termination before deleting interface, or kernel cleans structures
>> asynchronously without locks - so in some cases kernel tries to remove
>> shaper twice simultaneously, that causes bug.
>>> Did you get a response from kernel.org?
>>>
>> No, still no response. Here is bug:
>> https://bugzilla.kernel.org/show_bug.cgi?id=100971
>>>>
>>>> I removed shaper delete from scripts, and I think that this'll fix 
>>>> bug.
>>>
>>> No it just avoids it.
>> In any case, alternative - wait for fix in upstream - isn't a good idea
>> (no response for month).
>> Other accel-ppp user said that he has no troubles with it - he has no
>> shaper cleanup on shutdown.
>
> Or maybe a bug in accel-ppp?
>
Userland tools shouldn't cause kernel errors - so it's kernel bug.
> cheers
>
> ET


------------------------------------------------------------------------------

_______________________________________________
leaf-devel mailing list
leaf-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/leaf-devel

Reply via email to