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

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon, Jun 26, 2023 at 08:03:00PM +0800, Heng Qi wrote:
> 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@xxxxxxxxxxxxxxxxx>
> Reviewed-by: Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx>
> ---
> 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,

why do you want tp avoid forwarding? did you mean
"and to allow forwarding"?

> we re-probe the necessary checksum
> +	 * related information: skb->csum_{start, offset}, pseudo-header csum
> +	 * using skb_chdcksum_setup().

typo

> +	 *
> +	 * This benefits us:

Drop "This benefits us:" - benefits compared to what?

> +	 * 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.

please rewrite so the text makes sense in the final C file,
not for someone reading the diff. In that cotext it does not
matter that we used to drop packets or that we used to
disable _F_GUEST_CSUM unless you explain
why they had to be dropped previously.


> +	 */
> +	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.

what does this refer to? where does it exclude 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.

drop "We want to benefit from this: "

> +		 */
> +		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.

What does the last sentence mean?
Instead please explain why this is necessary. what can change the
header?

> +	 */
> +	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.

What if checksum was disabled on device? No need for all the
elaborate hacks then, right?
What about disabling by ethtool?


> +		 */
> +		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





[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux