ip6_tunnel: clear skb2->cb[] in ip4ip6_err()

Oskar Kjos reported the following problem.

ip4ip6_err() calls icmp_send() on a cloned skb whose cb[] was written
by the IPv6 receive path as struct inet6_skb_parm. icmp_send() passes
IPCB(skb2) to __ip_options_echo(), which interprets that cb[] region
as struct inet_skb_parm (IPv4). The layouts differ: inet6_skb_parm.nhoff
at offset 14 overlaps inet_skb_parm.opt.rr, producing a non-zero rr
value. __ip_options_echo() then reads optlen from attacker-controlled
packet data at sptr[rr+1] and copies that many bytes into dopt->__data,
a fixed 40-byte stack buffer (IP_OPTIONS_DATA_FIXED_SIZE).

To fix this we clear skb2->cb[], as suggested by Oskar Kjos.

Also add minimal IPv4 header validation (version == 4, ihl >= 5).

Fixes: c4d3efa ("[IPV6] IP6TUNNEL: Add support to IPv4 over IPv6 tunnel.")
Reported-by: Oskar Kjos <oskar.kjos@hotmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/20260326155138.2429480-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Author: edumazet

net/ipv6/ip6_tunnel.c

@@ -601,11 +601,16 @@ ip4ip6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 	if (!skb2)
 		return 0;
 
+	/* Remove debris left by IPv6 stack. */
+	memset(IPCB(skb2), 0, sizeof(*IPCB(skb2)));
+
 	skb_dst_drop(skb2);
 
 	skb_pull(skb2, offset);
 	skb_reset_network_header(skb2);
 	eiph = ip_hdr(skb2);
+	if (eiph->version != 4 || eiph->ihl < 5)
+		goto out;
 
 	/* Try to guess incoming interface */
 	rt = ip_route_output_ports(dev_net(skb->dev), &fl4, NULL, eiph->saddr,