netfilter: don't setup nat info for confirmed ct

We cannot setup nat info if the ct has been confirmed already, else,
different cpu may race to handle the same ct. In extreme situation,
we may hit the "BUG_ON(nf_nat_initialized(ct, maniptype))" in the
nf_nat_setup_info.

Also running the following commands will easily hit NF_CT_ASSERT in
nf_conntrack_alter_reply:
  # nft flush ruleset
  # ping -c 2 -W 1 1.1.1.111 &
  # nft add table t
  # nft add chain t c {type nat hook postrouting priority 0 \;}
  # nft add rule t c snat to 4.5.6.7
  WARNING: CPU: 1 PID: 10065 at net/netfilter/nf_conntrack_core.c:1472
  nf_conntrack_alter_reply+0x9a/0x1a0 [nf_conntrack]
  [...]
  Call Trace:
   nf_nat_setup_info+0xad/0x840 [nf_nat]
   ? deactivate_slab+0x65d/0x6c0
   nft_nat_eval+0xcd/0x100 [nft_nat]
   nft_do_chain+0xff/0x5d0 [nf_tables]
   ? mark_held_locks+0x6f/0xa0
   ? __local_bh_enable_ip+0x70/0xa0
   ? trace_hardirqs_on_caller+0x11f/0x190
   ? ipt_do_table+0x310/0x610
   ? trace_hardirqs_on+0xd/0x10
   ? __local_bh_enable_ip+0x70/0xa0
   ? ipt_do_table+0x32b/0x610
   ? __lock_acquire+0x2ac/0x1580
   ? ipt_do_table+0x32b/0x610
   nft_nat_do_chain+0x65/0x80 [nft_chain_nat_ipv4]
   nf_nat_ipv4_fn+0x1ae/0x240 [nf_nat_ipv4]
   nf_nat_ipv4_out+0x4a/0xf0 [nf_nat_ipv4]
   nft_nat_ipv4_out+0x15/0x20 [nft_chain_nat_ipv4]
   nf_hook_slow+0x2c/0xf0
   ip_output+0x154/0x270

So for the confirmed ct, just ignore it and return NF_ACCEPT.

Fixes: 9a08ecf ("netfilter: don't attach a nat extension by default")
Signed-off-by: Liping Zhang <zlpnobody@gmail.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

net/netfilter/nf_nat_core.c

@@ -409,6 +409,10 @@ nf_nat_setup_info(struct nf_conn *ct,
 {
 	struct nf_conntrack_tuple curr_tuple, new_tuple;
 
+	/* Can't setup nat info for confirmed ct. */
+	if (nf_ct_is_confirmed(ct))
+		return NF_ACCEPT;
+
 	NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
 		     maniptype == NF_NAT_MANIP_DST);
 	BUG_ON(nf_nat_initialized(ct, maniptype));