failure due to swapper and inet_sock_destruct

failure due to swapper and inet_sock_destruct

[Melkor Ainur]

Hello,

I am using the 2.4.20 kernel in a fairly high stress
bursty network environment. Every so often, I see the
following error message from the kernel:

Attempt to release TCP socket in state 10 cfbfd540

I put some tracking in the kernel, and determined that
this cfbfd540 was specifically the sk1 used in
inet_tcp_listen and looking at the sport of 0x17, it
is the telnet daemon. So it was created by xinetd for
the telnet  listening socket.

The failure often happens during a period of
inactivity. The call to inet_sock_destruct happened in
the swapper context. My first question is whether
anyone has observed this behavior before. Secondly, in
order to debug the problem, I'd like to understand
what  could trigger the swapper to call
inet_sock_destruct. I did further determine that the
call did not come from either tcp_close or
tcp_destroy_sock. Looking at the code, it looks like
the only call to inet_sock_destruct only happens from
sk_free. sk_free looks to be used by sock_put which
happens from sock_wfree, sklist_remove_socket, etc.
Anyway, I'll keep investigating but I could use any
help/advice on what would trigger the destruction of
an sk without a call to close the parent listening
socket. Any reason why it happened in the swapper
context? 

Thanks,
Melkor




 


__________________________________________________
Do you Yahoo!?
Yahoo! Tax Center - File online, calculators, forms, and more
http://tax.yahoo.com

Re: failure due to swapper and inet_sock_destruct

[James Morris]

On Sat, 5 Apr 2003, Melkor Ainur wrote:

> Hello,
> 
> I am using the 2.4.20 kernel in a fairly high stress
> bursty network environment. Every so often, I see the
> following error message from the kernel:
> 
> Attempt to release TCP socket in state 10 cfbfd540
> 

Could you please try the patch from Dave Miller below (which is already in
2.4.21-pre).


- James
-- 
James Morris 
<jmorris@intercode.com.au>

# This is a BitKeeper generated patch for the following project:
# Project Name: Linux kernel tree
# This patch format is intended for GNU patch command version 2.5 or higher.
# This patch includes the following deltas:
#	           ChangeSet	1.930.3.3 -> 1.930.3.4
#	net/ipv4/tcp_minisocks.c	1.12    -> 1.13   
#
# The following is the BitKeeper ChangeSet Log
# --------------------------------------------
# 03/01/30	davem@nuts.ninka.net	1.930.3.4
# [TCP]: In tcp_check_req, handle ACKless packets properly.
# --------------------------------------------
#
diff -Nru a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
--- a/net/ipv4/tcp_minisocks.c	Sun Apr  6 21:41:32 2003
+++ b/net/ipv4/tcp_minisocks.c	Sun Apr  6 21:41:32 2003
@@ -938,6 +938,12 @@
 	if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN))
 		goto embryonic_reset;
 
+	/* ACK sequence verified above, just make sure ACK is
+	 * set.  If ACK not set, just silently drop the packet.
+	 */
+	if (!(flg & TCP_FLAG_ACK))
+		return NULL;
+
 	/* If TCP_DEFER_ACCEPT is set, drop bare ACK. */
 	if (tp->defer_accept && TCP_SKB_CB(skb)->end_seq == req->rcv_isn+1) {
 		req->acked = 1;

Re: failure due to swapper and inet_sock_destruct

[Melkor Ainur]

Hello,

Thanks for the suggestion. I will try it. However,
examining the change, I could only see how it could
lead to a sock_put(child) but not how it could lead to
a sk_free(parent). Could you help me understand?

Melkor

--- James Morris <jmorris@intercode.com.au> wrote:
> On Sat, 5 Apr 2003, Melkor Ainur wrote:
> 
> > Hello,
> > 
> > I am using the 2.4.20 kernel in a fairly high
> stress
> > bursty network environment. Every so often, I see
> the
> > following error message from the kernel:
> > 
> > Attempt to release TCP socket in state 10 cfbfd540
> > 
> 
> Could you please try the patch from Dave Miller
> below (which is already in
> 2.4.21-pre).
> 
> 
> - James
> -- 
> James Morris 
> <jmorris@intercode.com.au>
> 
> # This is a BitKeeper generated patch for the
> following project:
> # Project Name: Linux kernel tree
> # This patch format is intended for GNU patch
> command version 2.5 or higher.
> # This patch includes the following deltas:
> #	           ChangeSet	1.930.3.3 -> 1.930.3.4
> #	net/ipv4/tcp_minisocks.c	1.12    -> 1.13   
> #
> # The following is the BitKeeper ChangeSet Log
> # --------------------------------------------
> # 03/01/30	davem@nuts.ninka.net	1.930.3.4
> # [TCP]: In tcp_check_req, handle ACKless packets
> properly.
> # --------------------------------------------
> #
> diff -Nru a/net/ipv4/tcp_minisocks.c
> b/net/ipv4/tcp_minisocks.c
> --- a/net/ipv4/tcp_minisocks.c	Sun Apr  6 21:41:32
> 2003
> +++ b/net/ipv4/tcp_minisocks.c	Sun Apr  6 21:41:32
> 2003
> @@ -938,6 +938,12 @@
>  	if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN))
>  		goto embryonic_reset;
>  
> +	/* ACK sequence verified above, just make sure ACK
> is
> +	 * set.  If ACK not set, just silently drop the
> packet.
> +	 */
> +	if (!(flg & TCP_FLAG_ACK))
> +		return NULL;
> +
>  	/* If TCP_DEFER_ACCEPT is set, drop bare ACK. */
>  	if (tp->defer_accept && TCP_SKB_CB(skb)->end_seq
> == req->rcv_isn+1) {
>  		req->acked = 1;
> 


__________________________________________________
Do you Yahoo!?
Yahoo! Tax Center - File online, calculators, forms, and more
http://tax.yahoo.com