KERNEL: assertion (!sk->sk_forward_alloc)

KERNEL: assertion (!sk->sk_forward_alloc)

[Anton Blanchard]

Hi,

I was running a network test on a machine with a number of e1000s in it
(netperf TCP_RR). The test included a range of message sizes from 64B
up to 64kB. The kernel is 2.6.10-rc2 BK from November 22.

I set tcp_tso_win_divisor to 2:

# cat /proc/sys/net/ipv4/tcp_tso_win_divisor
2

After a while some network assertion errors appeared in the log:

KERNEL: assertion (!sk->sk_forward_alloc) failed at net/core/stream.c (279)
KERNEL: assertion (!sk->sk_forward_alloc) failed at net/ipv4/af_inet.c (180)

Note that this test was run using the send-to-self patch.

Anton

Re: KERNEL: assertion (!sk->sk_forward_alloc)

[David S. Miller]

On Tue, 30 Nov 2004 08:18:57 +1100
Anton Blanchard <anton@samba.org> wrote:

> KERNEL: assertion (!sk->sk_forward_alloc) failed at net/core/stream.c (279)
> KERNEL: assertion (!sk->sk_forward_alloc) failed at net/ipv4/af_inet.c (180)
> 
> Note that this test was run using the send-to-self patch.

Try to reproduce without that patch installed.

If there is any mixup of SKB handling by any element in the
path the packet travels, you're screw up all the TCP
accounting knobs on the socket and easily trigger messages
like the above.

Re: KERNEL: assertion (!sk->sk_forward_alloc)

[Anton Blanchard]

 
Hi Dave,

> Try to reproduce without that patch installed.
> 
> If there is any mixup of SKB handling by any element in the
> path the packet travels, you're screw up all the TCP
> accounting knobs on the socket and easily trigger messages
> like the above.

I'll try. FYI we have 70 gigabit cards in that machine, the reason
send-to-self is so useful is that we only need one expensive machine.
Without the patch we need to find two expensive machines with 35
gigabits in each :)

Anton