ip_tunnel: fix kernel panic with icmp_dest_unreach

Daniel Petre reported crashes in icmp_dst_unreach() with following call
graph:

#3 [ffff88003fc03938] __stack_chk_fail at ffffffff81037f77
#4 [ffff88003fc03948] icmp_send at ffffffff814d5fec
#5 [ffff88003fc03ae8] ipv4_link_failure at ffffffff814a1795
#6 [ffff88003fc03af8] ipgre_tunnel_xmit at ffffffff814e7965
#7 [ffff88003fc03b78] dev_hard_start_xmit at ffffffff8146e032
#8 [ffff88003fc03bc8] sch_direct_xmit at ffffffff81487d66
#9 [ffff88003fc03c08] __qdisc_run at ffffffff81487efd
torvalds#10 [ffff88003fc03c48] dev_queue_xmit at ffffffff8146e5a7
torvalds#11 [ffff88003fc03c88] ip_finish_output at ffffffff814ab596

Daniel found a similar problem mentioned in
 http://lkml.indiana.edu/hypermail/linux/kernel/1007.0/00961.html

And indeed this is the root cause : skb->cb[] contains data fooling IP
stack.

We must clear IPCB in ip_tunnel_xmit() sooner in case dst_link_failure()
is called. Or else skb->cb[] might contain garbage from GSO segmentation
layer.

A similar fix was tested on linux-3.9, but gre code was refactored in
linux-3.10. I'll send patches for stable kernels as well.

Many thanks to Daniel for providing reports, patches and testing !

Reported-by: Daniel Petre <daniel.petre@rcs-rds.ro>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv4/ip_tunnel.c

@@ -503,6 +503,7 @@ void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev,
 
 	inner_iph = (const struct iphdr *)skb_inner_network_header(skb);
 
+	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 	dst = tnl_params->daddr;
 	if (dst == 0) {
 		/* NBMA tunnel */
@@ -658,7 +659,6 @@ void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev,
 
 	skb_dst_drop(skb);
 	skb_dst_set(skb, &rt->dst);
-	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 
 	/* Push down and install the IP header. */
 	skb_push(skb, sizeof(struct iphdr));