linux 2.4.20 kernel assertion failure config

[John Bicket]

Hi, 
I'm trying to track down this bug that Doug was talking about (see below), running with
the click patches to the kernel in 2.4.20.

This is the error that it the kernel logs:

Mar 27 15:25:38 rn-5 kernel: KERNEL: assertion (atomic_read(&sk->wmem_alloc) == 0) failed at af_inet.c(174)

I looked at it for a bit, and I'm no skbuff master (yet), but it appears that 
when inet_sock_destruct(struct sock *sk) is called, there are still packets that are still allocated
out of the socket's buffer.

My intuition is that there is something going on in an expensive push or expensive uniquify where
we don't entirely clean up after click takes over a packet from the kernel.

I wonder if the packet's destructor isn't getting called?

Any help would be appreciated.

Thanks,

--jbicket


>From: "Douglas S. J. De Couto" <decouto at lcs.mit.edu>
>Date: Wed Feb 5, 2003  10:17:16 US/Eastern
>To: click at amsterdam.lcs.mit.edu
>Subject: linux 2.4.20 kernel assertion failure config
>Reply-To: decouto at lcs.mit.edu
>
>
>
>arp_demux :: Classifier(12/0806 20/0001, // arp queries
>			12/0800);        // ZIP packets
>
>to_host :: ToHost(grid0);
>
>from_host :: FromHost(grid0, 18.26.7.6/255.255.255.0)
>	  -> arp_demux
>	  -> ARPResponder(0.0.0.0/0 1:1:1:1:1:1)
>	  -> to_host;
>
>arp_demux [1]
>	  -> Unstrip(3)
>	  -> Discard;
>
>----
>
>Run this config in kernel click; try to ping 18.2.6.7.1, for example.
>When the arg to Unstrip is 2, no errors; When the arg to Unstrip is 3,
>a expensive push is incurred, and the kernel prints an error when ping
>quits.
>
>d
>
>
>-- 
>Douglas S. J. De Couto    decouto at lcs.mit.edu
>
>
--
Douglas S. J. De Couto     <decouto at lcs.mit.edu>


----- End forwarded message -----
 LocalWords:  Couto arp demux arp ToHost FromHost ARPResponder arg arg