[oe] [PATCH] new interface is already up but udev must call ifup anyway

[Koen Kooi]

On 20-04-09 11:33, Stanislav Brabec wrote:
> Nicola Mfb wrote in Fri 03/06 2009 at 23:31 +0100:
>> 2009/3/6 Stanislav Brabec<utx at penguin.cz>
>>
>>> Few weeks ago yet another change happened in the dark, and network did
>>> not start to work automatically after inserting my WLAN card.
>>>
>>> Debugging this problem, I found that interface is already up but not
>>> configured when /etc/udev/scripts/network.sh is called. This script
>>> thinks, that card is already configured and quits. This behavior is
>>> intentional and was introduced three years ago (and working).
>>>
>>> The fix actually reverts following change:
>>> Author: Matthias Hentges<oe at hentges.net>
>>> Date:   Thu Apr 20 16:01:09 2006 +0000
>>>     udev: network.sh: Do not ifup an already configured network device
>>> again.
>>>
>>> That is why I am asking:
>>>
>>> Is anybody aware of change, that made new interfaces up without
>>> configuring them? Was it an intention or not?
>>>
>>
>> I had the same problem with bnep bluetooth networking, and another issue
>> when spawing dhcp, take a look at:
>> http://lists.openmoko.org/pipermail/devel/2009-February/004895.html
>
> I researched this problem a bit. It seems, that it's caused by
> wpa_supplicant. New wpa_supplicant quickly responds to the device
> addition. It turns the device up and sets up the wireless link, but not
> network.
>
> It causes several problems:
> - link is up ->  no ifup
>
> After applying mentioned patch:
> - AP lookup takes about 20 seconds. Too much for dhcp client =>
>    it fails. Surprisingly avahi succeeds.
>
> Proposed solution:
> Either:
> wpa_wupplicant should perform (or trigger somehow) wlan network hotplug
> completely (i. e. call ifup after network association instead of udev
> device addition)
> or:
> Revert previous behavior - "passive" wpa_supplicant.
> or:
> Apply mentioned patch. Ugly, it fixes only symptom, not the cause.

With my distro hat on: shouldn't something like connman take care of 
wireless? Which would probably mean going back to a passive wpa_supplicant.

regards,

Koen