Hi there,
I am running 2.6.10 from kernel.org on Debian Sid ppc/x86, the same
issue occurs with 2.6.9. Though, 2.6.8.1 and previous are fine.
When my ISP connection via PPPoE (kernel side) goes down, reconnection
does not occur, and the kernel displays continuous:
kernel: unregister_netdevice: waiting for ppp0 to become free. Usage count = 1
This is that on both machine: x86 and ppc.
The only solution is a reboot, because pppd is not killable, and the
kernel keeps on with this message.
Does anyone have a solution? I am completely clueless. :( I thought
about stalled sockets, but how can I destroy them?
What is blocking ppp0?
What system outputs do you need to solve this?
Cheers.
On 2005-01-21 at 15:44:44, Aur?lien G?R?ME wrote:
> I am running 2.6.10 from kernel.org on Debian Sid ppc/x86, the same
> issue occurs with 2.6.9. Though, 2.6.8.1 and previous are fine.
>
> When my ISP connection via PPPoE (kernel side) goes down, reconnection
> does not occur, and the kernel displays continuous:
>
> kernel: unregister_netdevice: waiting for ppp0 to become free. Usage count = 1
BTW, I have seen many cases when this symptom annoyed me too, the last
one is that my shutdown scripts tried unloading the network driver
modules. Is your setup doing this by any chance? In my case,
apparently there were conntrack entries keeping the device in use,
which is almost useless when preparing to shutdown :)
OTOH, I couldn't find a way to flush those conntracks, so I worked
around it by not rmmoding ethernet drivers.
In your case, it's probably conntrack too, I'd presume you are using
that PPPoE machine as a masquerading gateway, which by definition needs
connection tracking... I'm not sure either if this is a "real" change,
I only vaguely recollect as some moons earlier this wasn't a problem in
2.6.
--
Janos | romfs is at http://romfs.sourceforge.net/ | Don't talk about silence.
Janos Farkas wrote:
> On 2005-01-21 at 15:44:44, Aur?lien G?R?ME wrote:
>
>>I am running 2.6.10 from kernel.org on Debian Sid ppc/x86, the same
>>issue occurs with 2.6.9. Though, 2.6.8.1 and previous are fine.
>>
>>When my ISP connection via PPPoE (kernel side) goes down, reconnection
>>does not occur, and the kernel displays continuous:
>>
>>kernel: unregister_netdevice: waiting for ppp0 to become free. Usage count = 1
>
>
> BTW, I have seen many cases when this symptom annoyed me too, the last
> one is that my shutdown scripts tried unloading the network driver
> modules. Is your setup doing this by any chance? In my case,
> apparently there were conntrack entries keeping the device in use,
> which is almost useless when preparing to shutdown :)
>
> OTOH, I couldn't find a way to flush those conntracks, so I worked
> around it by not rmmoding ethernet drivers.
>
> In your case, it's probably conntrack too, I'd presume you are using
> that PPPoE machine as a masquerading gateway, which by definition needs
> connection tracking... I'm not sure either if this is a "real" change,
> I only vaguely recollect as some moons earlier this wasn't a problem in
> 2.6.
>
Here's some data I've collected on a possible related or identical one.
So far it seems related to interface removal. I've seen this recently on
vlan interfaces, since ifupdown on my systems kill vlan devices on ifdown.
In my cases I've not been able to reproduce without ipv6 loaded in the
kernel, and it only seems to happen when lo/loopback is taken offline
first (ifdown -a here does it in that order..)
A shutdown rrmod-type takedown of a NIC would probably be run after
"ifdown -a", hence after lo is down, thats why I suspect it to be the
same problem.
I still have a binary search to do for pinpointing what broke it, though.
On Wed, Jan 26, 2005 at 10:44:22AM +0100, Janos Farkas wrote:
> On 2005-01-21 at 15:44:44, Aur?lien G?R?ME wrote:
> > I am running 2.6.10 from kernel.org on Debian Sid ppc/x86, the same
> > issue occurs with 2.6.9. Though, 2.6.8.1 and previous are fine.
> >
> > When my ISP connection via PPPoE (kernel side) goes down, reconnection
> > does not occur, and the kernel displays continuous:
> >
> > kernel: unregister_netdevice: waiting for ppp0 to become free. Usage count = 1
>
> BTW, I have seen many cases when this symptom annoyed me too, the last
> one is that my shutdown scripts tried unloading the network driver
> modules. Is your setup doing this by any chance? In my case,
> apparently there were conntrack entries keeping the device in use,
> which is almost useless when preparing to shutdown :)
Actually, it happens after my ISP's LCP echos are not coming anymore,
and then when pppd try to reconnect.
> OTOH, I couldn't find a way to flush those conntracks, so I worked
> around it by not rmmoding ethernet drivers.
I have conntrack modules loaded too. I will try removing 8139too
(on x86) and sungem (on ppc) if the issue occurs again. By the way,
I also have IPv6 loaded, since I use a dual-stacked connection with
native IPv6. How lucky I am! :)
> In your case, it's probably conntrack too, I'd presume you are using
> that PPPoE machine as a masquerading gateway, which by definition needs
> connection tracking... I'm not sure either if this is a "real" change,
> I only vaguely recollect as some moons earlier this wasn't a problem in
> 2.6.
Yep, 2.6.8.1 works fine, this issue appears on 2.6.9 and 2.6.10. I
switched to a Debian 2.6.10 kernel for security reasons, and
the issue has not come yet. I had a glance at the changelog and
saw some network related patches. This is the -as patchset, see
<http://kerneltrap.org/node/4545> about it.
Cheers.
On Mer, 2005-01-26 at 20:31, Aurélien GÉRÔME wrote:
> Yep, 2.6.8.1 works fine, this issue appears on 2.6.9 and 2.6.10. I
> switched to a Debian 2.6.10 kernel for security reasons, and
> the issue has not come yet. I had a glance at the changelog and
> saw some network related patches. This is the -as patchset, see
> <http://kerneltrap.org/node/4545> about it.
2.6.9 changed the tty hangup code and this did introduce a problem with
serial ppp ignoring hangups. That was fixed in 2.6.9-ac and the fix is
in 2.6.10. If you are using serial ppp the 2.6.8 to 9 breakage is known
but the 2.6.10 case should be fine.
Alan