net: xfrmi: implement header_ops->parse_protocol for AF_PACKET
authorJason A. Donenfeld <Jason@zx2c4.com>
Tue, 30 Jun 2020 01:06:25 +0000 (19:06 -0600)
committerDavid S. Miller <davem@davemloft.net>
Tue, 30 Jun 2020 19:29:39 +0000 (12:29 -0700)
The xfrm interface uses skb->protocol to determine packet type, and
bails out if it's not set. For AF_PACKET injection, we need to support
its call chain of:

    packet_sendmsg -> packet_snd -> packet_parse_headers ->
      dev_parse_header_protocol -> parse_protocol

Without a valid parse_protocol, this returns zero, and xfrmi rejects the
skb. So, this wires up the ip_tunnel handler for layer 3 packets for
that case.

Reported-by: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/xfrm/xfrm_interface.c

index c407ecbc5d462b952aa60cf41c98e8a5c7e39ddf..b615729812e5acd1c1be39f402db5a9a6061e25d 100644 (file)
@@ -37,6 +37,7 @@
 #include <net/ip.h>
 #include <net/ipv6.h>
 #include <net/ip6_route.h>
+#include <net/ip_tunnels.h>
 #include <net/addrconf.h>
 #include <net/xfrm.h>
 #include <net/net_namespace.h>
@@ -581,6 +582,7 @@ static const struct net_device_ops xfrmi_netdev_ops = {
 static void xfrmi_dev_setup(struct net_device *dev)
 {
        dev->netdev_ops         = &xfrmi_netdev_ops;
+       dev->header_ops         = &ip_tunnel_header_ops;
        dev->type               = ARPHRD_NONE;
        dev->mtu                = ETH_DATA_LEN;
        dev->min_mtu            = ETH_MIN_MTU;