ip_queue/libipq panics on SMP kernel

ip_queue/libipq panics on SMP kernel

[Silimite]

Is anyone successfully using the QUEUE (with libipq) stuff on SMP kernels?

I'm just curious because I've never been able to get it to run for
more than a few hours before the kernel freezes/panics.   I've tried
many different kernel versions from the 2.6 series and I have tried on
several different machines running different Linux distros.  I have
had other people report the same problem.

Even a simple program that does nothing but set the verdict to
NF_ACCEPT will eventually kill the whole system.  It appears that you
need a rule loaded into iptables and a program listening and
processing the queued packets.  The time it takes varies anywhere from
a few minutes to several hours before the machine dies (race
condition?).

This has something to do with SMP because if I compile a non-SMP
kernel everything works fine, then if I use the exact same kernel
config except compiled with SMP turned on then ip_queue causes
problems.  This is irregardless of whether or not I'm actually using
multiple processors or even an SMP capable machine.

I also noticed that the libipq header(s) are not C++ friendly.  The
header does not do a "#ifdef __cplusplus" extern "C" check.

s

Re: ip_queue/libipq panics on SMP kernel

[Patrick McHardy]

Silimite wrote:
> Is anyone successfully using the QUEUE (with libipq) stuff on SMP kernels?

Probably, I haven't seen reports about problems yet.

> I'm just curious because I've never been able to get it to run for
> more than a few hours before the kernel freezes/panics.   I've tried
> many different kernel versions from the 2.6 series and I have tried on
> several different machines running different Linux distros.  I have
> had other people report the same problem.
> 
> Even a simple program that does nothing but set the verdict to
> NF_ACCEPT will eventually kill the whole system.  It appears that you
> need a rule loaded into iptables and a program listening and
> processing the queued packets.  The time it takes varies anywhere from
> a few minutes to several hours before the machine dies (race
> condition?).
> 
> This has something to do with SMP because if I compile a non-SMP
> kernel everything works fine, then if I use the exact same kernel
> config except compiled with SMP turned on then ip_queue causes
> problems.  This is irregardless of whether or not I'm actually using
> multiple processors or even an SMP capable machine.

Please capture the Oops using serial console or netconsole and send
it to this list.

Regards
Patrick