[net-next PATCH v4 0/6] XDP for virtio_net

[net-next PATCH v4 0/6] XDP for virtio_net

[John Fastabend]

This implements virtio_net for the mergeable buffers and big_packet
modes. I tested this with vhost_net running on qemu and did not see
any issues. For testing num_buf > 1 I added a hack to vhost driver
to only use 100 bytes per buffer so that packets were pushed across
multiple buffers.

There are some restrictions for XDP to be enabled and work well
(see patch 3) for more details.

  1. LRO must be off
  2. MTU must be less than PAGE_SIZE
  3. queues must be available to dedicate to XDP
  4. num_bufs received in mergeable buffers must be 1
  5. big_packet mode must have all data on single page

Please review any comments/feedback welcome as always.

---

John Fastabend (6):
      net: virtio dynamically disable/enable LRO
      net: xdp: add invalid buffer warning
      virtio_net: Add XDP support
      virtio_net: add dedicated XDP transmit queues
      virtio_net: add XDP_TX support
      virtio_net: xdp, add slowpath case for non contiguous buffers


 drivers/net/virtio_net.c |  376 +++++++++++++++++++++++++++++++++++++++++++++-
 include/linux/filter.h   |    1 
 net/core/filter.c        |    6 +
 3 files changed, 377 insertions(+), 6 deletions(-)

[net-next PATCH v4 1/6] net: virtio dynamically disable/enable LRO

[John Fastabend]

This adds support for dynamically setting the LRO feature flag. The
message to control guest features in the backend uses the
CTRL_GUEST_OFFLOADS msg type.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
---
 drivers/net/virtio_net.c |   45 ++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 44 insertions(+), 1 deletion(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index a21d93a..d814e7cb 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -1419,6 +1419,41 @@ static void virtnet_init_settings(struct net_device *dev)
 	.set_settings = virtnet_set_settings,
 };
 
+static int virtnet_set_features(struct net_device *netdev,
+				netdev_features_t features)
+{
+	struct virtnet_info *vi = netdev_priv(netdev);
+	struct virtio_device *vdev = vi->vdev;
+	struct scatterlist sg;
+	u64 offloads = 0;
+
+	if (features & NETIF_F_LRO)
+		offloads |= (1 << VIRTIO_NET_F_GUEST_TSO4) |
+			    (1 << VIRTIO_NET_F_GUEST_TSO6);
+
+	if (features & NETIF_F_RXCSUM)
+		offloads |= (1 << VIRTIO_NET_F_GUEST_CSUM);
+
+	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
+		sg_init_one(&sg, &offloads, sizeof(uint64_t));
+		if (!virtnet_send_command(vi,
+					  VIRTIO_NET_CTRL_GUEST_OFFLOADS,
+					  VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET,
+					  &sg)) {
+			dev_warn(&netdev->dev,
+				 "Failed to set guest offloads by virtnet command.\n");
+			return -EINVAL;
+		}
+	} else if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) &&
+		   !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
+		dev_warn(&netdev->dev,
+			 "No support for setting offloads pre version_1.\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 static const struct net_device_ops virtnet_netdev = {
 	.ndo_open            = virtnet_open,
 	.ndo_stop   	     = virtnet_close,
@@ -1435,6 +1470,7 @@ static void virtnet_init_settings(struct net_device *dev)
 #ifdef CONFIG_NET_RX_BUSY_POLL
 	.ndo_busy_poll		= virtnet_busy_poll,
 #endif
+	.ndo_set_features	= virtnet_set_features,
 };
 
 static void virtnet_config_changed_work(struct work_struct *work)
@@ -1815,6 +1851,12 @@ static int virtnet_probe(struct virtio_device *vdev)
 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
 		dev->features |= NETIF_F_RXCSUM;
 
+	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) &&
+	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)) {
+		dev->features |= NETIF_F_LRO;
+		dev->hw_features |= NETIF_F_LRO;
+	}
+
 	dev->vlan_features = dev->features;
 
 	/* MTU range: 68 - 65535 */
@@ -2057,7 +2099,8 @@ static int virtnet_restore(struct virtio_device *vdev)
 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
 	VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
 	VIRTIO_NET_F_CTRL_MAC_ADDR, \
-	VIRTIO_NET_F_MTU
+	VIRTIO_NET_F_MTU, \
+	VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
 
 static unsigned int features[] = {
 	VIRTNET_FEATURES,

Re: [net-next PATCH v4 1/6] net: virtio dynamically disable/enable LRO

[David Miller]

From: John Fastabend <john.fastabend@gmail.com>
Date: Fri, 02 Dec 2016 12:49:45 -0800

> +	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
> +		sg_init_one(&sg, &offloads, sizeof(uint64_t));
> +		if (!virtnet_send_command(vi,
> +					  VIRTIO_NET_CTRL_GUEST_OFFLOADS,
> +					  VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET,
> +					  &sg)) {
> +			dev_warn(&netdev->dev,
> +				 "Failed to set guest offloads by virtnet command.\n");
> +			return -EINVAL;
> +		}
> +	} else if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) &&
> +		   !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {

Hmmm, to me this reads as:

	if (X) {
	 ...
	else if (X && ...) {

I don't see how the second basic block can ever execute.  If the virtio
has the VIRTIO_NET_F_CTRL_GUEST_OFFLOADS feature, we will execute only
the first basic block.

Maybe I misunderstand the logic for whatever reason.

Re: [net-next PATCH v4 1/6] net: virtio dynamically disable/enable LRO

[Michael S. Tsirkin]

On Fri, Dec 02, 2016 at 12:49:45PM -0800, John Fastabend wrote:
> This adds support for dynamically setting the LRO feature flag. The
> message to control guest features in the backend uses the
> CTRL_GUEST_OFFLOADS msg type.
> 
> Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
> ---
>  drivers/net/virtio_net.c |   45 ++++++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 44 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index a21d93a..d814e7cb 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -1419,6 +1419,41 @@ static void virtnet_init_settings(struct net_device *dev)
>  	.set_settings = virtnet_set_settings,
>  };
>  
> +static int virtnet_set_features(struct net_device *netdev,
> +				netdev_features_t features)
> +{
> +	struct virtnet_info *vi = netdev_priv(netdev);
> +	struct virtio_device *vdev = vi->vdev;
> +	struct scatterlist sg;
> +	u64 offloads = 0;
> +
> +	if (features & NETIF_F_LRO)
> +		offloads |= (1 << VIRTIO_NET_F_GUEST_TSO4) |
> +			    (1 << VIRTIO_NET_F_GUEST_TSO6);
> +
> +	if (features & NETIF_F_RXCSUM)
> +		offloads |= (1 << VIRTIO_NET_F_GUEST_CSUM);
> +
> +	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
> +		sg_init_one(&sg, &offloads, sizeof(uint64_t));
> +		if (!virtnet_send_command(vi,
> +					  VIRTIO_NET_CTRL_GUEST_OFFLOADS,
> +					  VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET,
> +					  &sg)) {
> +			dev_warn(&netdev->dev,
> +				 "Failed to set guest offloads by virtnet command.\n");
> +			return -EINVAL;
> +		}
> +	} else if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) &&
> +		   !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {

No need for VIRTIO_F_VERSION_1 here.

> +		dev_warn(&netdev->dev,
> +			 "No support for setting offloads pre version_1.\n");
> +		return -EINVAL;
> +	}
> +
> +	return 0;
> +}
> +
>  static const struct net_device_ops virtnet_netdev = {
>  	.ndo_open            = virtnet_open,
>  	.ndo_stop   	     = virtnet_close,
> @@ -1435,6 +1470,7 @@ static void virtnet_init_settings(struct net_device *dev)
>  #ifdef CONFIG_NET_RX_BUSY_POLL
>  	.ndo_busy_poll		= virtnet_busy_poll,
>  #endif
> +	.ndo_set_features	= virtnet_set_features,
>  };
>  
>  static void virtnet_config_changed_work(struct work_struct *work)
> @@ -1815,6 +1851,12 @@ static int virtnet_probe(struct virtio_device *vdev)
>  	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
>  		dev->features |= NETIF_F_RXCSUM;
>  
> +	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) &&
> +	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)) {
> +		dev->features |= NETIF_F_LRO;
> +		dev->hw_features |= NETIF_F_LRO;
> +	}
> +
>  	dev->vlan_features = dev->features;
>  
>  	/* MTU range: 68 - 65535 */
> @@ -2057,7 +2099,8 @@ static int virtnet_restore(struct virtio_device *vdev)
>  	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
>  	VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
>  	VIRTIO_NET_F_CTRL_MAC_ADDR, \
> -	VIRTIO_NET_F_MTU
> +	VIRTIO_NET_F_MTU, \
> +	VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
>  
>  static unsigned int features[] = {
>  	VIRTNET_FEATURES,

Re: [net-next PATCH v4 1/6] net: virtio dynamically disable/enable LRO

[John Fastabend]

On 16-12-03 09:36 PM, Michael S. Tsirkin wrote:
> On Fri, Dec 02, 2016 at 12:49:45PM -0800, John Fastabend wrote:
>> This adds support for dynamically setting the LRO feature flag. The
>> message to control guest features in the backend uses the
>> CTRL_GUEST_OFFLOADS msg type.
>>
>> Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
>> ---
>>  drivers/net/virtio_net.c |   45 ++++++++++++++++++++++++++++++++++++++++++++-
>>  1 file changed, 44 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
>> index a21d93a..d814e7cb 100644
>> --- a/drivers/net/virtio_net.c
>> +++ b/drivers/net/virtio_net.c
>> @@ -1419,6 +1419,41 @@ static void virtnet_init_settings(struct net_device *dev)
>>  	.set_settings = virtnet_set_settings,
>>  };
>>  
>> +static int virtnet_set_features(struct net_device *netdev,
>> +				netdev_features_t features)
>> +{
>> +	struct virtnet_info *vi = netdev_priv(netdev);
>> +	struct virtio_device *vdev = vi->vdev;
>> +	struct scatterlist sg;
>> +	u64 offloads = 0;
>> +
>> +	if (features & NETIF_F_LRO)
>> +		offloads |= (1 << VIRTIO_NET_F_GUEST_TSO4) |
>> +			    (1 << VIRTIO_NET_F_GUEST_TSO6);
>> +
>> +	if (features & NETIF_F_RXCSUM)
>> +		offloads |= (1 << VIRTIO_NET_F_GUEST_CSUM);
>> +
>> +	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) {
>> +		sg_init_one(&sg, &offloads, sizeof(uint64_t));
>> +		if (!virtnet_send_command(vi,
>> +					  VIRTIO_NET_CTRL_GUEST_OFFLOADS,
>> +					  VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET,
>> +					  &sg)) {
>> +			dev_warn(&netdev->dev,
>> +				 "Failed to set guest offloads by virtnet command.\n");
>> +			return -EINVAL;
>> +		}
>> +	} else if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) &&
>> +		   !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
> 
> No need for VIRTIO_F_VERSION_1 here.
> 

Yep Dave pointed it out as well. Also I found a bug on driver unload
path where XDP tx queues are not cleaned up correctly so will need a v5
for that.

Thanks,
John

[net-next PATCH v4 2/6] net: xdp: add invalid buffer warning

[John Fastabend]

This adds a warning for drivers to use when encountering an invalid
buffer for XDP. For normal cases this should not happen but to catch
this in virtual/qemu setups that I may not have expected from the
emulation layer having a standard warning is useful.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
---
 include/linux/filter.h |    1 +
 net/core/filter.c      |    6 ++++++
 2 files changed, 7 insertions(+)

diff --git a/include/linux/filter.h b/include/linux/filter.h
index 7f246a2..90dfc3c 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -595,6 +595,7 @@ int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
 				       const struct bpf_insn *patch, u32 len);
 void bpf_warn_invalid_xdp_action(u32 act);
+void bpf_warn_invalid_xdp_buffer(void);
 
 #ifdef CONFIG_BPF_JIT
 extern int bpf_jit_enable;
diff --git a/net/core/filter.c b/net/core/filter.c
index 698a262..7926dd0 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -2783,6 +2783,12 @@ void bpf_warn_invalid_xdp_action(u32 act)
 }
 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
 
+void bpf_warn_invalid_xdp_buffer(void)
+{
+	WARN_ONCE(1, "Illegal XDP buffer encountered, expect throughput degradation\n");
+}
+EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_buffer);
+
 static u32 sk_filter_convert_ctx_access(enum bpf_access_type type, int dst_reg,
 					int src_reg, int ctx_off,
 					struct bpf_insn *insn_buf,

[net-next PATCH v4 3/6] virtio_net: Add XDP support

[John Fastabend]

From: John Fastabend <john.fastabend@gmail.com>

This adds XDP support to virtio_net. Some requirements must be
met for XDP to be enabled depending on the mode. First it will
only be supported with LRO disabled so that data is not pushed
across multiple buffers. Second the MTU must be less than a page
size to avoid having to handle XDP across multiple pages.

If mergeable receive is enabled this patch only supports the case
where header and data are in the same buf which we can check when
a packet is received by looking at num_buf. If the num_buf is
greater than 1 and a XDP program is loaded the packet is dropped
and a warning is thrown. When any_header_sg is set this does not
happen and both header and data is put in a single buffer as expected
so we check this when XDP programs are loaded.  Subsequent patches
will process the packet in a degraded mode to ensure connectivity
and correctness is not lost even if backend pushes packets into
multiple buffers.

If big packets mode is enabled and MTU/LRO conditions above are
met then XDP is allowed.

This patch was tested with qemu with vhost=on and vhost=off where
mergeable and big_packet modes were forced via hard coding feature
negotiation. Multiple buffers per packet was forced via a small
test patch to vhost.c in the vhost=on qemu mode.

Suggested-by: Shrijeet Mukherjee <shrijeet@gmail.com>
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
---
 drivers/net/virtio_net.c |  175 +++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 170 insertions(+), 5 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index d814e7cb..80b1cfc 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -22,6 +22,7 @@
 #include <linux/module.h>
 #include <linux/virtio.h>
 #include <linux/virtio_net.h>
+#include <linux/bpf.h>
 #include <linux/scatterlist.h>
 #include <linux/if_vlan.h>
 #include <linux/slab.h>
@@ -81,6 +82,8 @@ struct receive_queue {
 
 	struct napi_struct napi;
 
+	struct bpf_prog __rcu *xdp_prog;
+
 	/* Chain pages by the private ptr. */
 	struct page *pages;
 
@@ -324,6 +327,38 @@ static struct sk_buff *page_to_skb(struct virtnet_info *vi,
 	return skb;
 }
 
+static u32 do_xdp_prog(struct virtnet_info *vi,
+		       struct bpf_prog *xdp_prog,
+		       struct page *page, int offset, int len)
+{
+	int hdr_padded_len;
+	struct xdp_buff xdp;
+	u32 act;
+	u8 *buf;
+
+	buf = page_address(page) + offset;
+
+	if (vi->mergeable_rx_bufs)
+		hdr_padded_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
+	else
+		hdr_padded_len = sizeof(struct padded_vnet_hdr);
+
+	xdp.data = buf + hdr_padded_len;
+	xdp.data_end = xdp.data + (len - vi->hdr_len);
+
+	act = bpf_prog_run_xdp(xdp_prog, &xdp);
+	switch (act) {
+	case XDP_PASS:
+		return XDP_PASS;
+	default:
+		bpf_warn_invalid_xdp_action(act);
+	case XDP_TX:
+	case XDP_ABORTED:
+	case XDP_DROP:
+		return XDP_DROP;
+	}
+}
+
 static struct sk_buff *receive_small(struct virtnet_info *vi, void *buf, unsigned int len)
 {
 	struct sk_buff * skb = buf;
@@ -340,14 +375,32 @@ static struct sk_buff *receive_big(struct net_device *dev,
 				   void *buf,
 				   unsigned int len)
 {
+	struct bpf_prog *xdp_prog;
 	struct page *page = buf;
-	struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE);
+	struct sk_buff *skb;
 
+	rcu_read_lock();
+	xdp_prog = rcu_dereference(rq->xdp_prog);
+	if (xdp_prog) {
+		struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
+		u32 act;
+
+		if (unlikely(hdr->hdr.gso_type || hdr->hdr.flags))
+			goto err_xdp;
+		act = do_xdp_prog(vi, xdp_prog, page, 0, len);
+		if (act == XDP_DROP)
+			goto err_xdp;
+	}
+	rcu_read_unlock();
+
+	skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE);
 	if (unlikely(!skb))
 		goto err;
 
 	return skb;
 
+err_xdp:
+	rcu_read_unlock();
 err:
 	dev->stats.rx_dropped++;
 	give_pages(rq, page);
@@ -365,11 +418,42 @@ static struct sk_buff *receive_mergeable(struct net_device *dev,
 	u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
 	struct page *page = virt_to_head_page(buf);
 	int offset = buf - page_address(page);
-	unsigned int truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
+	struct sk_buff *head_skb, *curr_skb;
+	struct bpf_prog *xdp_prog;
+	unsigned int truesize;
+
+	rcu_read_lock();
+	xdp_prog = rcu_dereference(rq->xdp_prog);
+	if (xdp_prog) {
+		u32 act;
+
+		/* No known backend devices should send packets with
+		 * more than a single buffer when XDP conditions are
+		 * met. However it is not strictly illegal so the case
+		 * is handled as an exception and a warning is thrown.
+		 */
+		if (unlikely(num_buf > 1)) {
+			bpf_warn_invalid_xdp_buffer();
+			goto err_xdp;
+		}
 
-	struct sk_buff *head_skb = page_to_skb(vi, rq, page, offset, len,
-					       truesize);
-	struct sk_buff *curr_skb = head_skb;
+		/* Transient failure which in theory could occur if
+		 * in-flight packets from before XDP was enabled reach
+		 * the receive path after XDP is loaded. In practice I
+		 * was not able to create this condition.
+		 */
+		if (unlikely(hdr->hdr.gso_type || hdr->hdr.flags))
+			goto err_xdp;
+
+		act = do_xdp_prog(vi, xdp_prog, page, offset, len);
+		if (act == XDP_DROP)
+			goto err_xdp;
+	}
+	rcu_read_unlock();
+
+	truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
+	head_skb = page_to_skb(vi, rq, page, offset, len, truesize);
+	curr_skb = head_skb;
 
 	if (unlikely(!curr_skb))
 		goto err_skb;
@@ -423,6 +507,8 @@ static struct sk_buff *receive_mergeable(struct net_device *dev,
 	ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
 	return head_skb;
 
+err_xdp:
+	rcu_read_unlock();
 err_skb:
 	put_page(page);
 	while (--num_buf) {
@@ -1328,6 +1414,13 @@ static int virtnet_set_channels(struct net_device *dev,
 	if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
 		return -EINVAL;
 
+	/* For now we don't support modifying channels while XDP is loaded
+	 * also when XDP is loaded all RX queues have XDP programs so we only
+	 * need to check a single RX queue.
+	 */
+	if (vi->rq[0].xdp_prog)
+		return -EINVAL;
+
 	get_online_cpus();
 	err = virtnet_set_queues(vi, queue_pairs);
 	if (!err) {
@@ -1454,6 +1547,69 @@ static int virtnet_set_features(struct net_device *netdev,
 	return 0;
 }
 
+static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog)
+{
+	unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr);
+	struct virtnet_info *vi = netdev_priv(dev);
+	struct bpf_prog *old_prog;
+	int i;
+
+	if ((dev->features & NETIF_F_LRO) && prog) {
+		netdev_warn(dev, "can't set XDP while LRO is on, disable LRO first\n");
+		return -EINVAL;
+	}
+
+	if (vi->mergeable_rx_bufs && !vi->any_header_sg) {
+		netdev_warn(dev, "XDP expects header/data in single page\n");
+		return -EINVAL;
+	}
+
+	if (dev->mtu > max_sz) {
+		netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz);
+		return -EINVAL;
+	}
+
+	if (prog) {
+		prog = bpf_prog_add(prog, vi->max_queue_pairs - 1);
+		if (IS_ERR(prog))
+			return PTR_ERR(prog);
+	}
+
+	for (i = 0; i < vi->max_queue_pairs; i++) {
+		old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
+		rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
+		if (old_prog)
+			bpf_prog_put(old_prog);
+	}
+
+	return 0;
+}
+
+static bool virtnet_xdp_query(struct net_device *dev)
+{
+	struct virtnet_info *vi = netdev_priv(dev);
+	int i;
+
+	for (i = 0; i < vi->max_queue_pairs; i++) {
+		if (vi->rq[i].xdp_prog)
+			return true;
+	}
+	return false;
+}
+
+static int virtnet_xdp(struct net_device *dev, struct netdev_xdp *xdp)
+{
+	switch (xdp->command) {
+	case XDP_SETUP_PROG:
+		return virtnet_xdp_set(dev, xdp->prog);
+	case XDP_QUERY_PROG:
+		xdp->prog_attached = virtnet_xdp_query(dev);
+		return 0;
+	default:
+		return -EINVAL;
+	}
+}
+
 static const struct net_device_ops virtnet_netdev = {
 	.ndo_open            = virtnet_open,
 	.ndo_stop   	     = virtnet_close,
@@ -1471,6 +1627,7 @@ static int virtnet_set_features(struct net_device *netdev,
 	.ndo_busy_poll		= virtnet_busy_poll,
 #endif
 	.ndo_set_features	= virtnet_set_features,
+	.ndo_xdp		= virtnet_xdp,
 };
 
 static void virtnet_config_changed_work(struct work_struct *work)
@@ -1532,12 +1689,20 @@ static void virtnet_free_queues(struct virtnet_info *vi)
 
 static void free_receive_bufs(struct virtnet_info *vi)
 {
+	struct bpf_prog *old_prog;
 	int i;
 
+	rtnl_lock();
 	for (i = 0; i < vi->max_queue_pairs; i++) {
 		while (vi->rq[i].pages)
 			__free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
+
+		old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
+		RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL);
+		if (old_prog)
+			bpf_prog_put(old_prog);
 	}
+	rtnl_unlock();
 }
 
 static void free_receive_page_frags(struct virtnet_info *vi)

[net-next PATCH v4 4/6] virtio_net: add dedicated XDP transmit queues

[John Fastabend]

XDP requires using isolated transmit queues to avoid interference
with normal networking stack (BQL, NETDEV_TX_BUSY, etc). This patch
adds a XDP queue per cpu when a XDP program is loaded and does not
expose the queues to the OS via the normal API call to
netif_set_real_num_tx_queues(). This way the stack will never push
an skb to these queues.

However virtio/vhost/qemu implementation only allows for creating
TX/RX queue pairs at this time so creating only TX queues was not
possible. And because the associated RX queues are being created I
went ahead and exposed these to the stack and let the backend use
them. This creates more RX queues visible to the network stack than
TX queues which is worth mentioning but does not cause any issues as
far as I can tell.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
---
 drivers/net/virtio_net.c |   30 ++++++++++++++++++++++++++++--
 1 file changed, 28 insertions(+), 2 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 80b1cfc..b67203e 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -114,6 +114,9 @@ struct virtnet_info {
 	/* # of queue pairs currently used by the driver */
 	u16 curr_queue_pairs;
 
+	/* # of XDP queue pairs currently used by the driver */
+	u16 xdp_queue_pairs;
+
 	/* I like... big packets and I cannot lie! */
 	bool big_packets;
 
@@ -1552,7 +1555,8 @@ static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog)
 	unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr);
 	struct virtnet_info *vi = netdev_priv(dev);
 	struct bpf_prog *old_prog;
-	int i;
+	u16 xdp_qp = 0, curr_qp;
+	int i, err;
 
 	if ((dev->features & NETIF_F_LRO) && prog) {
 		netdev_warn(dev, "can't set XDP while LRO is on, disable LRO first\n");
@@ -1569,12 +1573,34 @@ static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog)
 		return -EINVAL;
 	}
 
+	curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs;
+	if (prog)
+		xdp_qp = nr_cpu_ids;
+
+	/* XDP requires extra queues for XDP_TX */
+	if (curr_qp + xdp_qp > vi->max_queue_pairs) {
+		netdev_warn(dev, "request %i queues but max is %i\n",
+			    curr_qp + xdp_qp, vi->max_queue_pairs);
+		return -ENOMEM;
+	}
+
+	err = virtnet_set_queues(vi, curr_qp + xdp_qp);
+	if (err) {
+		dev_warn(&dev->dev, "XDP Device queue allocation failure.\n");
+		return err;
+	}
+
 	if (prog) {
 		prog = bpf_prog_add(prog, vi->max_queue_pairs - 1);
-		if (IS_ERR(prog))
+		if (IS_ERR(prog)) {
+			virtnet_set_queues(vi, curr_qp);
 			return PTR_ERR(prog);
+		}
 	}
 
+	vi->xdp_queue_pairs = xdp_qp;
+	netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp);
+
 	for (i = 0; i < vi->max_queue_pairs; i++) {
 		old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
 		rcu_assign_pointer(vi->rq[i].xdp_prog, prog);

[net-next PATCH v4 5/6] virtio_net: add XDP_TX support

[John Fastabend]

This adds support for the XDP_TX action to virtio_net. When an XDP
program is run and returns the XDP_TX action the virtio_net XDP
implementation will transmit the packet on a TX queue that aligns
with the current CPU that the XDP packet was processed on.

Before sending the packet the header is zeroed.  Also XDP is expected
to handle checksum correctly so no checksum offload  support is
provided.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
---
 drivers/net/virtio_net.c |   63 ++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 60 insertions(+), 3 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index b67203e..137caba 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -330,12 +330,43 @@ static struct sk_buff *page_to_skb(struct virtnet_info *vi,
 	return skb;
 }
 
+static void virtnet_xdp_xmit(struct virtnet_info *vi,
+			     unsigned int qnum, struct xdp_buff *xdp)
+{
+	struct send_queue *sq = &vi->sq[qnum];
+	struct virtio_net_hdr_mrg_rxbuf *hdr;
+	unsigned int num_sg, len;
+	void *xdp_sent;
+	int err;
+
+	/* Free up any pending old buffers before queueing new ones. */
+	while ((xdp_sent = virtqueue_get_buf(sq->vq, &len)) != NULL) {
+		struct page *page = virt_to_head_page(xdp_sent);
+
+		put_page(page);
+	}
+
+	/* Zero header and leave csum up to XDP layers */
+	hdr = xdp->data;
+	memset(hdr, 0, vi->hdr_len);
+
+	num_sg = 1;
+	sg_init_one(sq->sg, xdp->data, xdp->data_end - xdp->data);
+	err = virtqueue_add_outbuf(sq->vq, sq->sg, num_sg,
+				   xdp->data, GFP_ATOMIC);
+	if (unlikely(err))
+		put_page(virt_to_head_page(xdp->data));
+	else
+		virtqueue_kick(sq->vq);
+}
+
 static u32 do_xdp_prog(struct virtnet_info *vi,
 		       struct bpf_prog *xdp_prog,
 		       struct page *page, int offset, int len)
 {
 	int hdr_padded_len;
 	struct xdp_buff xdp;
+	unsigned int qp;
 	u32 act;
 	u8 *buf;
 
@@ -353,9 +384,15 @@ static u32 do_xdp_prog(struct virtnet_info *vi,
 	switch (act) {
 	case XDP_PASS:
 		return XDP_PASS;
+	case XDP_TX:
+		qp = vi->curr_queue_pairs -
+			vi->xdp_queue_pairs +
+			smp_processor_id();
+		xdp.data = buf + (vi->mergeable_rx_bufs ? 0 : 4);
+		virtnet_xdp_xmit(vi, qp, &xdp);
+		return XDP_TX;
 	default:
 		bpf_warn_invalid_xdp_action(act);
-	case XDP_TX:
 	case XDP_ABORTED:
 	case XDP_DROP:
 		return XDP_DROP;
@@ -391,8 +428,16 @@ static struct sk_buff *receive_big(struct net_device *dev,
 		if (unlikely(hdr->hdr.gso_type || hdr->hdr.flags))
 			goto err_xdp;
 		act = do_xdp_prog(vi, xdp_prog, page, 0, len);
-		if (act == XDP_DROP)
+		switch (act) {
+		case XDP_PASS:
+			break;
+		case XDP_TX:
+			rcu_read_unlock();
+			goto xdp_xmit;
+		case XDP_DROP:
+		default:
 			goto err_xdp;
+		}
 	}
 	rcu_read_unlock();
 
@@ -407,6 +452,7 @@ static struct sk_buff *receive_big(struct net_device *dev,
 err:
 	dev->stats.rx_dropped++;
 	give_pages(rq, page);
+xdp_xmit:
 	return NULL;
 }
 
@@ -425,6 +471,8 @@ static struct sk_buff *receive_mergeable(struct net_device *dev,
 	struct bpf_prog *xdp_prog;
 	unsigned int truesize;
 
+	head_skb = NULL;
+
 	rcu_read_lock();
 	xdp_prog = rcu_dereference(rq->xdp_prog);
 	if (xdp_prog) {
@@ -449,8 +497,16 @@ static struct sk_buff *receive_mergeable(struct net_device *dev,
 			goto err_xdp;
 
 		act = do_xdp_prog(vi, xdp_prog, page, offset, len);
-		if (act == XDP_DROP)
+		switch (act) {
+		case XDP_PASS:
+			break;
+		case XDP_TX:
+			rcu_read_unlock();
+			goto xdp_xmit;
+		case XDP_DROP:
+		default:
 			goto err_xdp;
+		}
 	}
 	rcu_read_unlock();
 
@@ -528,6 +584,7 @@ static struct sk_buff *receive_mergeable(struct net_device *dev,
 err_buf:
 	dev->stats.rx_dropped++;
 	dev_kfree_skb(head_skb);
+xdp_xmit:
 	return NULL;
 }
 

Re: [net-next PATCH v4 5/6] virtio_net: add XDP_TX support

[John Fastabend]

On 16-12-02 12:51 PM, John Fastabend wrote:
> This adds support for the XDP_TX action to virtio_net. When an XDP
> program is run and returns the XDP_TX action the virtio_net XDP
> implementation will transmit the packet on a TX queue that aligns
> with the current CPU that the XDP packet was processed on.
> 
> Before sending the packet the header is zeroed.  Also XDP is expected
> to handle checksum correctly so no checksum offload  support is
> provided.
> 
> Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
> ---
>  drivers/net/virtio_net.c |   63 ++++++++++++++++++++++++++++++++++++++++++++--
>  1 file changed, 60 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index b67203e..137caba 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -330,12 +330,43 @@ static struct sk_buff *page_to_skb(struct virtnet_info *vi,
>  	return skb;
>  }
>  
> +static void virtnet_xdp_xmit(struct virtnet_info *vi,
> +			     unsigned int qnum, struct xdp_buff *xdp)
> +{
> +	struct send_queue *sq = &vi->sq[qnum];
> +	struct virtio_net_hdr_mrg_rxbuf *hdr;
> +	unsigned int num_sg, len;
> +	void *xdp_sent;
> +	int err;
> +
> +	/* Free up any pending old buffers before queueing new ones. */
> +	while ((xdp_sent = virtqueue_get_buf(sq->vq, &len)) != NULL) {
> +		struct page *page = virt_to_head_page(xdp_sent);
> +
> +		put_page(page);
> +	}
> +
> +	/* Zero header and leave csum up to XDP layers */
> +	hdr = xdp->data;
> +	memset(hdr, 0, vi->hdr_len);
> +
> +	num_sg = 1;
> +	sg_init_one(sq->sg, xdp->data, xdp->data_end - xdp->data);
> +	err = virtqueue_add_outbuf(sq->vq, sq->sg, num_sg,
> +				   xdp->data, GFP_ATOMIC);
> +	if (unlikely(err))
> +		put_page(virt_to_head_page(xdp->data));
> +	else
> +		virtqueue_kick(sq->vq);
> +}
> +

Hi Michael,

Any idea why the above pattern

> +	err = virtqueue_add_outbuf(sq->vq, sq->sg, num_sg,
> +				   xdp->data, GFP_ATOMIC);
> +	if (unlikely(err))
> +		put_page(virt_to_head_page(xdp->data));
> +	else
> +		virtqueue_kick(sq->vq);
> +}

would cause a hang but if I call the virtqueue_kick as below
even in the error case everything seems to be fine.

	err = virtqueue_add_outbuf(sq->vq, sq->sg, num_sg,
				   xdp->data, GFP_ATOMIC);
	if (unlikely(err))
		put_page(virt_to_head_page(xdp->data));

	virtqueue_kick(sq->vq);


I'll take a look through the virtio code but thought I might ask in
case you know off-hand or it could be something else entirely.

I noticed virtio_input.c uses the second pattern and virtio_net.c
uses the above pattern but I'm guessing it never gets exercised due
to stack backoff.

Thanks,
John

[net-next PATCH v4 6/6] virtio_net: xdp, add slowpath case for non contiguous buffers

[John Fastabend]

virtio_net XDP support expects receive buffers to be contiguous.
If this is not the case we enable a slowpath to allow connectivity
to continue but at a significan performance overhead associated with
linearizing data. To make it painfully aware to users that XDP is
running in a degraded mode we throw an xdp buffer error.

To linearize packets we allocate a page and copy the segments of
the data, including the header, into it. After this the page can be
handled by XDP code flow as normal.

Then depending on the return code the page is either freed or sent
to the XDP xmit path. There is no attempt to optimize this path.

This case is being handled simple as a precaution in case some
unknown backend were to generate packets in this form. To test this
I had to hack qemu and force it to generate these packets. I do not
expect this case to be generated by "real" backends.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
---
 drivers/net/virtio_net.c |   77 +++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 75 insertions(+), 2 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 137caba..13f463d 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -456,6 +456,64 @@ static struct sk_buff *receive_big(struct net_device *dev,
 	return NULL;
 }
 
+/* The conditions to enable XDP should preclude the underlying device from
+ * sending packets across multiple buffers (num_buf > 1). However per spec
+ * it does not appear to be illegal to do so but rather just against convention.
+ * So in order to avoid making a system unresponsive the packets are pushed
+ * into a page and the XDP program is run. This will be extremely slow and we
+ * push a warning to the user to fix this as soon as possible. Fixing this may
+ * require resolving the underlying hardware to determine why multiple buffers
+ * are being received or simply loading the XDP program in the ingress stack
+ * after the skb is built because there is no advantage to running it here
+ * anymore.
+ */
+static struct page *xdp_linearize_page(struct receive_queue *rq,
+				       u16 num_buf,
+				       struct page *p,
+				       int offset,
+				       unsigned int *len)
+{
+	struct page *page = alloc_page(GFP_ATOMIC);
+	unsigned int page_off = 0;
+
+	if (!page)
+		return NULL;
+
+	memcpy(page_address(page) + page_off, page_address(p) + offset, *len);
+	page_off += *len;
+
+	while (--num_buf) {
+		unsigned int buflen;
+		unsigned long ctx;
+		void *buf;
+		int off;
+
+		ctx = (unsigned long)virtqueue_get_buf(rq->vq, &buflen);
+		if (unlikely(!ctx))
+			goto err_buf;
+
+		/* guard against a misconfigured or uncooperative backend that
+		 * is sending packet larger than the MTU.
+		 */
+		if ((page_off + buflen) > PAGE_SIZE)
+			goto err_buf;
+
+		buf = mergeable_ctx_to_buf_address(ctx);
+		p = virt_to_head_page(buf);
+		off = buf - page_address(p);
+
+		memcpy(page_address(page) + page_off,
+		       page_address(p) + off, buflen);
+		page_off += buflen;
+	}
+
+	*len = page_off;
+	return page;
+err_buf:
+	__free_pages(page, 0);
+	return NULL;
+}
+
 static struct sk_buff *receive_mergeable(struct net_device *dev,
 					 struct virtnet_info *vi,
 					 struct receive_queue *rq,
@@ -476,6 +534,7 @@ static struct sk_buff *receive_mergeable(struct net_device *dev,
 	rcu_read_lock();
 	xdp_prog = rcu_dereference(rq->xdp_prog);
 	if (xdp_prog) {
+		struct page *xdp_page;
 		u32 act;
 
 		/* No known backend devices should send packets with
@@ -485,7 +544,15 @@ static struct sk_buff *receive_mergeable(struct net_device *dev,
 		 */
 		if (unlikely(num_buf > 1)) {
 			bpf_warn_invalid_xdp_buffer();
-			goto err_xdp;
+
+			/* linearize data for XDP */
+			xdp_page = xdp_linearize_page(rq, num_buf,
+						      page, offset, &len);
+			if (!xdp_page)
+				goto err_xdp;
+			offset = 0;
+		} else {
+			xdp_page = page;
 		}
 
 		/* Transient failure which in theory could occur if
@@ -496,15 +563,21 @@ static struct sk_buff *receive_mergeable(struct net_device *dev,
 		if (unlikely(hdr->hdr.gso_type || hdr->hdr.flags))
 			goto err_xdp;
 
-		act = do_xdp_prog(vi, xdp_prog, page, offset, len);
+		act = do_xdp_prog(vi, xdp_prog, xdp_page, offset, len);
 		switch (act) {
 		case XDP_PASS:
+			if (unlikely(xdp_page != page))
+				__free_pages(xdp_page, 0);
 			break;
 		case XDP_TX:
+			if (unlikely(xdp_page != page))
+				goto err_xdp;
 			rcu_read_unlock();
 			goto xdp_xmit;
 		case XDP_DROP:
 		default:
+			if (unlikely(xdp_page != page))
+				__free_pages(xdp_page, 0);
 			goto err_xdp;
 		}
 	}