[LEDE-DEV] [OpenWrt-Devel] [PATCH netifd 2/3] interface: Only teardown interfaces having no proto task when l3_dev link lost

[Felix Fietkau]

On 2016-07-14 16:53, Yousong Zhou wrote:
> On 14/07/2016, Felix Fietkau <nbd at nbd.name> wrote:
>> On 2016-07-14 13:28, Hans Dedecker wrote:
>>> On Thu, Jul 14, 2016 at 12:01 PM, Yousong Zhou <yszhou4tech at gmail.com>
>>> wrote:
>>>> On 14 July 2016 at 16:14, Hans Dedecker <dedeckeh at gmail.com> wrote:
>>>>> Commit c6858766 adds teardown support when l3_dev link is lost
>>>>> especially for shell protocols
>>>>> that have no proto task. However shell protocols which have a proto task
>>>>> running like ppp will
>>>>> also be teared down which is not always the expected action.
>>>>> As an example the PPP daemon can be put into persist state trying to
>>>>> re-establish the link via
>>>>> a hold off mechanism which is not possible when the daemon is terminated
>>>>> by the proto shell
>>>>> teardown.
>>>>> Therefore restrict the teardown action for shell protocols having no
>>>>> proto task.
>>>>>
>>>>
>>>> How about adding an extra flag like managed-link, persistent-link,
>>>> on-demand-link?  It looks to me doing teardown at link-down is more
>>>> common a case.
>>> Initially I was thinking about adding another flag like you propose
>>> but then I was doubting if the change in behavior for shell protocols
>>> having a proto task task was on purpose or not. In case of PPP and
>>> link failure you don't want an immediate restart by netifd in some
>>> cases (see https://github.com/lede-project/source/pull/200) as PPP
>>> daemon can take care of the link re-negotiation based on a holdoff
>>> timeout.
>>> Additionally if the wan link loses connectivity a link down
>>> notification will be received on the main device which will teardown
>>> the protocol. Anyway I'm open for suggestions which way to go forward.
>> Yousong,
>>
>> please provide some more details on where your commit c6858766 is
>> actually needed/useful. In all the use cases I can think of, handling
>> setup/teardown based on the l2 dev should be enough.
>>
>> - Felix
>>
> 
> The issue them was that when l2tp-xxx went down, netifd has no proto
> task state to be notified of, and main_dev state seemed unchanged.  If
> I remeber and understand the code correcly other pppd shell protos do
> teardown because of proto task event, not any device link state, and I
> thought it's reasonable and  should not hurt to do an explicit
> teardown on link down.
It seems to me that we should do less magic here and make the behavior
opt-in via an explicit flag that needs to be enabled by the proto handler.

- Felix