kernel BUG in __skb_gso_segment

[Tudor Ambarus]

Hi,

There's a bug [1] reported by syzkaller in linux-5.15.y that I'd like
to squash. The commit in stable that introduces the bug is:
b99c71f90978 net: skip virtio_net_hdr_set_proto if protocol already set
The upstream commit for this is:
1ed1d592113959f00cc552c3b9f47ca2d157768f

I discovered that in mainline this bug was squashed by the following
commits:
e9d3f80935b6 ("net/af_packet: make sure to pull mac header")
dfed913e8b55 ("net/af_packet: add VLAN support for AF_PACKET SOCK_RAW GSO")

I'm seeking for some guidance on how to fix linux-5.15.y. From what I
understand, the bug in stable is triggered because we end up with a
header offset of 18, that eventually triggers the GSO crash in
__skb_pull. If I revert the commit in culprit from linux-5.15.y, we'll
end up with a header offset of 14, the bug is not hit and the packet is 
dropped at validate_xmit_skb() time. I'm wondering if reverting it is
the right thing to do, as the commit is marked as a fix. Backporting the
2 commits from mainline is not an option as they introduce new support.
Would such a patch be better than reverting the offending commit?

diff --git a/include/linux/virtio_net.h b/include/linux/virtio_net.h
index a960de68ac69..188b6f05e5bd 100644
--- a/include/linux/virtio_net.h
+++ b/include/linux/virtio_net.h
@@ -25,7 +25,7 @@ static inline bool virtio_net_hdr_match_proto(__be16 
protocol, __u8 gso_type)
 static inline int virtio_net_hdr_set_proto(struct sk_buff *skb,
                                           const struct virtio_net_hdr 
*hdr)
 {
-       if (skb->protocol)
+       if (skb->protocol && skb->protocol != htons(ETH_P_ALL))
                return 0;

        switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {


Thanks!
ta

[1] 
https://syzkaller.appspot.com/bug?id=ce5575575f074c33ff80d104f5baee26f22e95f5