ipv6: make icmp6_send() robust against null skb->dev

syzbot was able to crash one host with the following stack trace :

kasan: GPF could be caused by NULL-ptr deref or user memory access
general protection fault: 0000 [#1] PREEMPT SMP KASAN
CPU: 0 PID: 8625 Comm: syz-executor4 Not tainted 4.20.0+ #8
RIP: 0010:dev_net include/linux/netdevice.h:2169 [inline]
RIP: 0010:icmp6_send+0x116/0x2d30 net/ipv6/icmp.c:426
 icmpv6_send
 smack_socket_sock_rcv_skb
 security_sock_rcv_skb
 sk_filter_trim_cap
 __sk_receive_skb
 dccp_v6_do_rcv
 release_sock

This is because a RX packet found socket owned by user and
was stored into socket backlog. Before leaving RCU protected section,
skb->dev was cleared in __sk_receive_skb(). When socket backlog
was finally handled at release_sock() time, skb was fed to
smack_socket_sock_rcv_skb() then icmp6_send()

We could fix the bug in smack_socket_sock_rcv_skb(), or simply
make icmp6_send() more robust against such possibility.

In the future we might provide to icmp6_send() the net pointer
instead of infering it.

Fixes: d66a8acb ("Smack: Inform peer that IPv6 traffic has been blocked")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Piotr Sawicki <p.sawicki2@partner.samsung.com>
Cc: Casey Schaufler <casey@schaufler-ca.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Acked-by: Casey Schaufler <casey@schaufler-ca.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv6/icmp.c

@@ -423,10 +423,10 @@ static int icmp6_iif(const struct sk_buff *skb)
 static void icmp6_send(struct sk_buff *skb, u8 type, u8 code, __u32 info,
 		       const struct in6_addr *force_saddr)
 {
-	struct net *net = dev_net(skb->dev);
 	struct inet6_dev *idev = NULL;
 	struct ipv6hdr *hdr = ipv6_hdr(skb);
 	struct sock *sk;
+	struct net *net;
 	struct ipv6_pinfo *np;
 	const struct in6_addr *saddr = NULL;
 	struct dst_entry *dst;
@@ -437,12 +437,16 @@ static void icmp6_send(struct sk_buff *skb, u8 type, u8 code, __u32 info,
 	int iif = 0;
 	int addr_type = 0;
 	int len;
-	u32 mark = IP6_REPLY_MARK(net, skb->mark);
+	u32 mark;
 
 	if ((u8 *)hdr < skb->head ||
 	    (skb_network_header(skb) + sizeof(*hdr)) > skb_tail_pointer(skb))
 		return;
 
+	if (!skb->dev)
+		return;
+	net = dev_net(skb->dev);
+	mark = IP6_REPLY_MARK(net, skb->mark);
 	/*
 	 *	Make sure we respect the rules
 	 *	i.e. RFC 1885 2.4(e)