Re: [RFC] [PATCH] Don't destroy TCP sockets twice

[Andi Kleen <andi@firstfloor.org>]

On Mon, Aug 09, 2010 at 05:30:02PM -0400, Herbert Xu wrote:
> Andi Kleen <andi@firstfloor.org> wrote:
> > 
> > While working on something else I noticed that tcp_v4/6_destroy_sock()
> > can get called twice on a socket. This happens because when a reset or
> > similar happens tcp_done destroys the connection socket state, and 
> > then eventually when the socket is released it is destroyed again. 
> 
> I'm still having problems understanding why you're getting a call
> to send a FIN after tcp_done.  This shouldn't happen at all because
> tcp_done moves the socket to the TCP_CLOSE state, where FINs should
> not be sent.
> 
> Can you clarify on how we can reproduce this problem in the
>  upstream kernel?

This simple patch demonstrates double destroy. I have patches for showing
the more complicated case too, but they're much more ugly.

-Andi

diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index a778ee0..bfbb06b 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -459,6 +459,8 @@ struct tcp_sock {
 	 * contains related tcp_cookie_transactions fields.
 	 */
 	struct tcp_cookie_values  *cookie_values;
+
+	int destroyed;
 };
 
 static inline struct tcp_sock *tcp_sk(const struct sock *sk)
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 0207662..25bf80a 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -1919,6 +1919,9 @@ void tcp_v4_destroy_sock(struct sock *sk)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 
+	BUG_ON(tp->destroyed != 0);
+	tp->destroyed = 1;
+
 	tcp_clear_xmit_timers(sk);
 
 	tcp_cleanup_congestion_control(sk);
-- 
ak@linux.intel.com -- Speaking for myself only.