Re: [PATCH net] ip6_gre: validate ERSPAN skb dst before PMTU update

[Ido Schimmel <idosch@nvidia.com>]

On Wed, Sep 30, 2026 at 04:53:20PM +0900, Daehyeon Ko wrote:
> A tc tunnel_key action can attach a METADATA_IP_TUNNEL dst to an skb,
> and mirred preserves it when redirecting the skb to a native ip6erspan
> device. ip6erspan_tunnel_xmit() treats any non-NULL dst as a route for
> PMTU updates. A metadata dst has no output device, so
> dst_dev(dst)->mtu dereferences NULL.
> 
> On v7.2 with KASAN, an initial UID/GID 65534 process with CapEff 0
> created user and network namespaces, used namespace-local CAP_NET_ADMIN,
> and triggered:
> 
>   KASAN: null-ptr-deref
>   RIP: 0010:ip6erspan_tunnel_xmit+0x10fc/0x2cc0
>   Kernel panic - not syncing: Fatal exception in interrupt
> 
> The collect-metadata path already skips this block because its metadata
> describes the outer tunnel. Native mode must still propagate PMTU updates
> for real route destinations. Use skb_valid_dst(), matching IPv4 tunnel
> PMTU handling, to exclude DST_METADATA without suppressing valid routes.
> 
> With this change, the same trigger processed three packets without a
> sanitizer report, oops, or panic.
> 
> Fixes: 5a963eb61b7c ("ip6_gre: Add ERSPAN native tunnel support")
> Cc: stable@vger.kernel.org
> Assisted-by: Codex:GPT-5
> Signed-off-by: Daehyeon Ko <4ncienth@gmail.com>
> ---
> Affected since v4.16-rc1; present in v7.2 and all four pinned
> 2026-09-30 upstream snapshots.
> 
> Tested with CONFIG_USER_NS=y, CONFIG_NET_NS=y, CONFIG_IPV6_GRE=y,
> CONFIG_NET_CLS_ACT=y, CONFIG_NET_ACT_TUNNEL_KEY=y, and
> CONFIG_NET_ACT_MIRRED=y. Config SHA-256:
> f992c9fdb881ca95c467896791c52554719f6204cf06268cd92ac37c14c14f83
> 
> The reproducer is available privately on request and is omitted from this
> public AI-assisted report.
> ---
>  net/ipv6/ip6_gre.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/net/ipv6/ip6_gre.c b/net/ipv6/ip6_gre.c
> index e61cb10b50dc..3ee7fad6089a 100644
> --- a/net/ipv6/ip6_gre.c
> +++ b/net/ipv6/ip6_gre.c
> @@ -1059,7 +1059,7 @@ static netdev_tx_t ip6erspan_tunnel_xmit(struct sk_buff *skb,
>  			 htonl(atomic_fetch_inc(&t->o_seqno)));
>  
>  	/* TooBig packet may have updated dst->dev's mtu */
> -	if (!t->parms.collect_md && dst) {
> +	if (!t->parms.collect_md && skb_valid_dst(skb)) {
>  		mtu = READ_ONCE(dst_dev(dst)->mtu);
>  		if (dst_mtu(dst) > mtu)
>  			dst->ops->update_pmtu(dst, NULL, skb, mtu, false);

This hunk seems obsolete and it's better to remove it like commit
fe1a4ca0a2b7 ("ip6_gre: process toobig in a better way") did in
__gre6_xmit().

Nowadays the ICMP error handler sets the PMTU in the underlay route and
ip6_tnl_xmit() propagates it to the overlay route.