[PATCH net-next v3 1/2] virtio-net: support coexistence of XDP and GUEST_CSUM

[Heng Qi <hengqi@linux.alibaba.com>]

We are now re-probing the csum related fields and trying
to have XDP and RX hw checksum capabilities coexist on the
XDP path. For the benefit of:
1. RX hw checksum capability can be used if XDP is loaded.
2. Avoid packet loss when loading XDP in the vm-vm scenario.

Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
---
v2->v3:
  - Use skb_checksum_setup() instead of virtnet_flow_dissect_udp_tcp().
    Essentially equivalent.

 drivers/net/virtio_net.c | 86 ++++++++++++++++++++++++++++++++++------
 1 file changed, 73 insertions(+), 13 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 5a7f7a76b920..0a715e0fbc97 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -1568,6 +1568,44 @@ static void virtio_skb_set_hash(const struct virtio_net_hdr_v1_hash *hdr_hash,
 	skb_set_hash(skb, __le32_to_cpu(hdr_hash->hash_value), rss_hash_type);
 }
 
+static int virtnet_set_csum_after_xdp(struct virtnet_info *vi,
+				      struct sk_buff *skb,
+				      __u8 flags)
+{
+	int err = 0;
+
+	/* When XDP program is loaded, for example, the vm-vm scenario
+	 * on the same host, packets marked as VIRTIO_NET_HDR_F_NEEDS_CSUM
+	 * will travel. Although these packets are safe from the point of
+	 * view of the vm, to avoid modification by XDP and successful
+	 * forwarding in the upper layer, we re-probe the necessary checksum
+	 * related information: skb->csum_{start, offset}, pseudo-header csum
+	 * using skb_chdcksum_setup().
+	 *
+	 * This benefits us:
+	 * 1. XDP can be loaded when there's _F_GUEST_CSUM.
+	 * 2. The device verifies the checksum of packets, especially
+	 *    benefiting for large packets.
+	 * 3. In the same-host vm-vm scenario, packets marked as
+	 *    VIRTIO_NET_HDR_F_NEEDS_CSUM are no longer dropped after being
+	 *    processed by XDP.
+	 */
+	if (flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
+		/* We don't parse SCTP because virtio-net currently doesn't
+		 * support CRC checksum offloading for SCTP.
+		 */
+		err = skb_checksum_setup(skb, true);
+	} else if (flags & VIRTIO_NET_HDR_F_DATA_VALID) {
+		/* We want to benefit from this: XDP guarantees that packets marked
+		 * as VIRTIO_NET_HDR_F_DATA_VALID still have correct csum after they
+		 * are processed.
+		 */
+		skb->ip_summed = CHECKSUM_UNNECESSARY;
+	}
+
+	return err;
+}
+
 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
 			void *buf, unsigned int len, void **ctx,
 			unsigned int *xdp_xmit,
@@ -1576,6 +1614,7 @@ static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
 	struct net_device *dev = vi->dev;
 	struct sk_buff *skb;
 	struct virtio_net_hdr_mrg_rxbuf *hdr;
+	__u8 flags;
 
 	if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
 		pr_debug("%s: short packet %i\n", dev->name, len);
@@ -1584,6 +1623,13 @@ static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
 		return;
 	}
 
+	/* Save the flags of the hdr before XDP processes the data.
+	 * It is ok to use this for both mergeable and small modes.
+	 * Because that's what we do now.
+	 */
+	if (unlikely(vi->xdp_enabled))
+		flags = ((struct virtio_net_hdr_mrg_rxbuf *)buf)->hdr.flags;
+
 	if (vi->mergeable_rx_bufs)
 		skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit,
 					stats);
@@ -1595,23 +1641,37 @@ static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
 	if (unlikely(!skb))
 		return;
 
-	hdr = skb_vnet_hdr(skb);
-	if (dev->features & NETIF_F_RXHASH && vi->has_rss_hash_report)
-		virtio_skb_set_hash((const struct virtio_net_hdr_v1_hash *)hdr, skb);
-
-	if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
-		skb->ip_summed = CHECKSUM_UNNECESSARY;
+	if (unlikely(vi->xdp_enabled)) {
+		/* Required to do this before re-probing and calculating
+		 * the pseudo-header checksum.
+		 */
+		skb->protocol = eth_type_trans(skb, dev);
+		skb_reset_network_header(skb);
+		if (virtnet_set_csum_after_xdp(vi, skb, flags) < 0) {
+			pr_debug("%s: errors occurred in setting partial csum",
+				 dev->name);
+			goto frame_err;
+		}
+	} else {
+		hdr = skb_vnet_hdr(skb);
+		if (dev->features & NETIF_F_RXHASH && vi->has_rss_hash_report)
+			virtio_skb_set_hash((const struct virtio_net_hdr_v1_hash *)hdr, skb);
+
+		if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
+			skb->ip_summed = CHECKSUM_UNNECESSARY;
+
+		if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
+					  virtio_is_little_endian(vi->vdev))) {
+			net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
+					     dev->name, hdr->hdr.gso_type,
+					     hdr->hdr.gso_size);
+			goto frame_err;
+		}
 
-	if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
-				  virtio_is_little_endian(vi->vdev))) {
-		net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
-				     dev->name, hdr->hdr.gso_type,
-				     hdr->hdr.gso_size);
-		goto frame_err;
+		skb->protocol = eth_type_trans(skb, dev);
 	}
 
 	skb_record_rx_queue(skb, vq2rxq(rq->vq));
-	skb->protocol = eth_type_trans(skb, dev);
 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
 
-- 
2.19.1.6.gb485710b