net/tunnel: set inner protocol in network gro hooks
authorPaolo Abeni <pabeni@redhat.com>
Tue, 7 Mar 2017 17:33:31 +0000 (18:33 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 22 Mar 2017 11:43:34 +0000 (12:43 +0100)
commitb07eed8f7119d43cbfe4d2d66241bd04099e6585
tree3b38d31cf2ade5d8121cf19f930e07ea951c2080
parentdb6e7796186a8d9a41e9d14b4a824c9cb80aaaf4
net/tunnel: set inner protocol in network gro hooks

[ Upstream commit 294acf1c01bace5cea5d30b510504238bf5f7c25 ]

The gso code of several tunnels type (gre and udp tunnels)
takes for granted that the skb->inner_protocol is properly
initialized and drops the packet elsewhere.

On the forwarding path no one is initializing such field,
so gro encapsulated packets are dropped on forward.

Since commit 38720352412a ("gre: Use inner_proto to obtain
inner header protocol"), this can be reproduced when the
encapsulated packets use gre as the tunneling protocol.

The issue happens also with vxlan and geneve tunnels since
commit 8bce6d7d0d1e ("udp: Generalize skb_udp_segment"), if the
forwarding host's ingress nic has h/w offload for such tunnel
and a vxlan/geneve device is configured on top of it, regardless
of the configured peer address and vni.

To address the issue, this change initialize the inner_protocol
field for encapsulated packets in both ipv4 and ipv6 gro complete
callbacks.

Fixes: 38720352412a ("gre: Use inner_proto to obtain inner header protocol")
Fixes: 8bce6d7d0d1e ("udp: Generalize skb_udp_segment")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/ipv4/af_inet.c
net/ipv6/ip6_offload.c