Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Andrea Mayer <andrea.mayer@uniroma2.it>
To:     "David S. Miller" <davem@davemloft.net>,
        David Ahern <dsahern@kernel.org>,
        Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>,
        Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>,
        Jakub Kicinski <kuba@kernel.org>,
        Shuah Khan <shuah@kernel.org>,
        Shrijeet Mukherjee <shrijeet@gmail.com>,
        netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
        linux-kselftest@vger.kernel.org
Cc:     Stefano Salsano <stefano.salsano@uniroma2.it>,
        Paolo Lungaroni <paolo.lungaroni@cnit.it>,
        Ahmed Abdelsalam <ahabdels.dev@gmail.com>,
        Andrea Mayer <andrea.mayer@uniroma2.it>
Subject: [RFC,net-next, 1/4] vrf: push mac header for tunneled packets when sniffer is attached
Date:   Wed, 21 Oct 2020 20:41:13 +0200
Message-Id: <20201021184116.2722-2-andrea.mayer@uniroma2.it>
In-Reply-To: <20201021184116.2722-1-andrea.mayer@uniroma2.it>
References: <20201021184116.2722-1-andrea.mayer@uniroma2.it>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

Before this patch, a sniffer attached to a VRF used as the receiving
interface of L3 tunneled packets detects them as malformed packets and
it complains about that (i.e.: tcpdump shows bogus packets).

The reason is that a tunneled L3 packet does not carry any L2
information and when the VRF is set as the receiving interface of a
decapsulated L3 packet, no mac header is currently set or valid.
Therefore the purpose of this patch consists of adding a MAC header to
any packet which is directly received on the VRF interface ONLY IF:

 i) a sniffer is attached on the VRF and ii) the mac header is not set.

In this case, the mac address of the VRF is copied in both the
destination and the source address of the ethernet header. The protocol
type is set either to IPv4 or IPv6, depending on which L3 packet is
received.

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
---
 drivers/net/vrf.c | 78 +++++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 72 insertions(+), 6 deletions(-)

diff --git a/drivers/net/vrf.c b/drivers/net/vrf.c
index 60c1aadece89..abb09c6036a6 100644
--- a/drivers/net/vrf.c
+++ b/drivers/net/vrf.c
@@ -1263,6 +1263,61 @@ static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
 	skb_dst_set(skb, &rt6->dst);
 }
 
+static int vrf_prepare_mac_header(struct sk_buff *skb,
+				  struct net_device *vrf_dev, u16 proto)
+{
+	struct ethhdr *eth;
+	int err;
+
+	/* in general, we do not know if there is enough space in the head of
+	 * the packet for hosting the mac header.
+	 */
+	err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev));
+	if (unlikely(err))
+		/* no space in the skb head */
+		return -ENOBUFS;
+
+	__skb_push(skb, ETH_HLEN);
+	eth = (struct ethhdr *)skb->data;
+
+	skb_reset_mac_header(skb);
+
+	/* we set the ethernet destination and the source addresses to the
+	 * address of the VRF device.
+	 */
+	ether_addr_copy(eth->h_dest, vrf_dev->dev_addr);
+	ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
+	eth->h_proto = htons(proto);
+
+	/* the destination address of the Ethernet frame corresponds to the
+	 * address set on the VRF interface; therefore, the packet is intended
+	 * to be processed locally.
+	 */
+	skb->protocol = eth->h_proto;
+	skb->pkt_type = PACKET_HOST;
+
+	skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
+
+	skb_pull_inline(skb, ETH_HLEN);
+
+	return 0;
+}
+
+/* prepare and add the mac header to the packet if it was not previously set.
+ * In this way, packet sniffers such as tcpdump can parse the packet correctly.
+ * If the mac header was previously set, the original mac header is left
+ * untouched and the function returns immediately.
+ */
+static int vrf_add_mac_header_if_unset(struct sk_buff *skb,
+				       struct net_device *vrf_dev,
+				       u16 proto)
+{
+	if (skb_mac_header_was_set(skb))
+		return 0;
+
+	return vrf_prepare_mac_header(skb, vrf_dev, proto);
+}
+
 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
 				   struct sk_buff *skb)
 {
@@ -1289,9 +1344,15 @@ static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
 		skb->skb_iif = vrf_dev->ifindex;
 
 		if (!list_empty(&vrf_dev->ptype_all)) {
-			skb_push(skb, skb->mac_len);
-			dev_queue_xmit_nit(skb, vrf_dev);
-			skb_pull(skb, skb->mac_len);
+			int err;
+
+			err = vrf_add_mac_header_if_unset(skb, vrf_dev,
+							  ETH_P_IPV6);
+			if (likely(!err)) {
+				skb_push(skb, skb->mac_len);
+				dev_queue_xmit_nit(skb, vrf_dev);
+				skb_pull(skb, skb->mac_len);
+			}
 		}
 
 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
@@ -1334,9 +1395,14 @@ static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
 	vrf_rx_stats(vrf_dev, skb->len);
 
 	if (!list_empty(&vrf_dev->ptype_all)) {
-		skb_push(skb, skb->mac_len);
-		dev_queue_xmit_nit(skb, vrf_dev);
-		skb_pull(skb, skb->mac_len);
+		int err;
+
+		err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP);
+		if (likely(!err)) {
+			skb_push(skb, skb->mac_len);
+			dev_queue_xmit_nit(skb, vrf_dev);
+			skb_pull(skb, skb->mac_len);
+		}
 	}
 
 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
-- 
2.20.1