[REGRESSION (bisected): selftest/net/gre_gso.sh fails with 5.15.199

[REGRESSION (bisected): selftest/net/gre_gso.sh fails with 5.15.199

[Stefan Bader]

[-- Attachment #1.1.1: Type: text/plain, Size: 1467 bytes --]

We noticed that after applying 5.15.199 the gre_gso.sh test no longer 
was able to make any connection. This was bisected down to:

commit 64c71d60a21a9ed0a802483dcd422b5b24eb1abe
Author: Eric Dumazet <edumazet@google.com>
Date:   Wed Jan 7 16:31:09 2026 +0000

     ip6_tunnel: use skb_vlan_inet_prepare() in __ip6_tnl_rcv()

     [ Upstream commit 81c734dae203757fb3c9eee6f9896386940776bd ]

Above patch replaces pskb_inet_may_pull() with skb_vlan_inet_prepare() 
but as

9990ddf47d41 net: tunnel: make skb_vlan_inet_prepare() return drop reasons

has not been backported to 5.15.y (appeared in v6.13 without stable 
reference) skb_vlan_inet_prepare() still returns a boolean which 
basically inverts the meaning. To properly replace the function call in 
5.15.y (and maybe any long-term up to 6.13) as other callers do:

diff --git a/net/ipv6/ip6_tunnel.c b/net/ipv6/ip6_tunnel.c
index 553851e3aca1..7c1b5d01f820 100644
--- a/net/ipv6/ip6_tunnel.c
+++ b/net/ipv6/ip6_tunnel.c
@@ -846,7 +846,7 @@ static int __ip6_tnl_rcv(struct ip6_tnl *tunnel, 
struct sk_buff *skb,

         skb_reset_network_header(skb);

-       if (skb_vlan_inet_prepare(skb, true)) {
+       if (!skb_vlan_inet_prepare(skb, true)) {
                 DEV_STATS_INC(tunnel->dev, rx_length_errors);
                 DEV_STATS_INC(tunnel->dev, rx_errors);
                 goto drop;

With above applied the gre_gso.sh test again runs successful.

-Stefan

[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 53483 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]