Re: current hang: arp_drain: locked; punting

[buhrow%lothlorien.nfbcal.org@localhost (Brian Buhrow)]

        Hello.  I'm not sure I entirely know how to reproduce the problem, but
if you use a machine which you expect to see a lot of arp traffic, I think
you'll encounter the problem.  In our case, we had about 200 machines all
trying to do arp requests at the same time.  When the problem is occurring,
the NetBSD machine doesn't pass packets, arp or otherwise.  It just emits
the arp_drain message over and over again.  Eventually, it clears itself
and  all returns to normal.  Another factor, which I believe contributes to
the problem, and in fact might be the activating force, is if you have a
lot of public addresses behind a box and someone begins scanning those
addresses.  If there are no real machines behind those addresses, the arp
requests pile up unanswered on the routing box and, when it comes time to
clear out the unanswered slots, you run into this problem.
-Brian
On Oct 21, 10:46am, Thomas Klausner wrote:
} Subject: Re: current hang: arp_drain: locked; punting
} On Mon, Oct 20, 2008 at 08:40:00PM -0700, Brian Buhrow wrote:
} >     Hello.  I've seen this hang as far back as 3.1.  In 3.1, the system
} > eventually recovers, but when there is a lot of arp traffic, you can run
} > into this.  I mention this because I suspect it's an old bug.
} 
} Thanks for this comment.
} Is there a PR about this?
} 
} I wouldn't know how to reproduce this, since it just happened. :|
}  Thomas
>-- End of excerpt from Thomas Klausner