Re: issues with pppd error handling

[Piotr Figiel <figiel gmail com>]

Hi,

2017-11-29 9:52 GMT+01:00 Beniamino Galvani <bgalvani redhat com>:
> On Mon, Nov 27, 2017 at 03:52:26PM +0100, Piotr Figiel wrote:
> > Is it really needed to receive and process notifications about
> > internal pppd state changes? Since pppd is/should be running in
> > nopersist mode pppd should exit upon connection termination. Actually
> > there were only two 'live' cases where this information is used in NM
> > and both are related with disconnect and cause ppp_failed() to run as
> > mentioned in my previous message.
>
> The phase state change information is needed in several places to
> determine why pppd failed. It is also needed in NMDevicePpp to start
> the IP configuration when pppd enters the 'running' phase.

Yes. I forgot to mention my previous comment ignores pppoe (and also
bases on 1.8 which doesn't have NMDevicePpp). Without the last part of
my patch I think the issue with nm-device-modem.c would be that the
ppp_failed() would be issued multiple times each time disabling the
connection (I don't know for sure but I think it may race with
auto-reconnect). I think more conservative change to avoid that would
be to remove contents of devices/wwan/nm-modem.c:ppp_state_changed()
and rely only on pppd exit to determine connection termination in case
of modem ppp connections (I don't have this tested though but
functionally for ppp with modem it should be the same as the previous
patch).

Best regards, Piotr.