[LEDE-DEV] [PATCH netifd 3/3] interface-event: Don't dequeue hotplug event in case of interface reload event

Hans Dedecker dedeckeh at gmail.com
Wed Sep 14 05:44:53 PDT 2016


On Wed, Sep 14, 2016 at 1:15 PM, Felix Fietkau <nbd at nbd.name> wrote:
> On 2016-09-13 14:33, Hans Dedecker wrote:
>> Dropping hotplug event in case of interface reload results into hotplug scripts
>> not being being run for the interface and thus external actors not being informed
>> about the actual state of the interface.
>>
>> This is clearly visible if the interface auto parameter is set to disabled for
>> multiple interfaces resulting into no hotplug down event for all interfaces.
>>
>> Therefore don't flush the interface hotplug queue in case an interface reload
>> event is observed.
> I have some doubt about this. What use is the reload event to hotplug
> scripts? As far as I can tell there is no guarantee that the interface
> is in a well defined state when this event is emitted. For reloads that
> imply an up/down cycle, the regular events should suffice. For reloads
> that affect the state without making the interface go through up/down,
> we could maybe have a separate event. Am I missing something?
Problem is related to the interface reload event flushing the pending
event in the interface hotplug queue.
Assume there are three interfaces configured A, B and C; for each of
them we set the auto parameter to 0 followed by a network reload.
This will bring the interfaces in reload config state; each interface
will be brought down resulting into an IFEV_DOWN event.
As a result for interface A a hotplug event with action ifdown will be
launched; for interfaces B and C the hotplug event will be queued as
hotplug event for interface A is running.
Once interfaces B and C are marked as down in netifd the function
interface_do_reload will launch the IFEV_RELOAD event resulting into
drop of the queued hotplug event of both interface B and C in the
function interface_event_cb (hotplug event for interface A is still
running at that moment)
Thus hotplug scripts have seen an ifdown event for interface A but not
for interface B and C; meaning state driven hotplug actions for
interface B and C are not performed.
Therefore I removed the flushing of the hotplug queue based on the
reload event which fixes the above described problem;  did also tests
in other scenarios where a down/up is triggered by reload config and
did not observe any issues.


Hans

>
> - Felix
>



More information about the Lede-dev mailing list