[PATCH RFC v2 0/9] veth: Driver XDP

[PATCH RFC v2 0/9] veth: Driver XDP

[Toshiaki Makita]

From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>

This patch set introduces driver XDP for veth.
Basically this is used in conjunction with redirect action of another XDP
program.

  NIC -----------> veth===veth
 (XDP) (redirect)        (XDP)

In this case xdp_frame can be forwarded to the peer veth without
modification, so we can expect far better performance than generic XDP.

The envisioned use cases are:

* Container managed XDP program
Container host redirects frames to containers by XDP redirect action, and
privileged containers can deploy their own XDP programs.

* XDP program cascading
Two or more XDP programs can be called for each packet by redirecting
xdp frames to veth.

* Internal interface for an XDP bridge
When using XDP redirection to create a virtual bridge, veth can be used
to create an internal interface for the bridge.

With single core and simple XDP programs which only redirect and drop
packets, I got 10.5 Mpps redirect/drop rate with i40e 25G NIC + veth.

XXV710 (i40e) --- (XDP redirect) --> veth===veth (XDP drop)

This changeset is making use of NAPI to implement ndo_xdp_xmit and
XDP_TX/REDIRECT. This is mainly because XDP heavily relies on NAPI
context.

This patchset is based on top of net-next commit 75d4e704fa8d
(netdev-FAQ: clarify DaveM's position for stable backports).
Any feedback is welcome. Thanks!

v2:
- Squash NAPI patch with "Add driver XDP" patch.
- Remove conversion from xdp_frame to skb when NAPI is not enabled.
- Introduce per-queue XDP ring (patch 8).
- Introduce bulk skb xmit when XDP is enabled on the peer (patch 9).

Toshiaki Makita (9):
  net: Export skb_headers_offset_update
  veth: Add driver XDP
  veth: Avoid drops by oversized packets when XDP is enabled
  veth: Add another napi ring for ndo_xdp_xmit and handle xdp_frames
  veth: Add ndo_xdp_xmit
  xdp: Add a flag for disabling napi_direct of xdp_return_frame in
    xdp_mem_info
  veth: Add XDP TX and REDIRECT
  veth: Support per queue XDP ring
  veth: Bulk skb xmit for XDP path

 drivers/net/veth.c     | 734 ++++++++++++++++++++++++++++++++++++++++++++++++-
 include/linux/filter.h |  16 ++
 include/linux/skbuff.h |   1 +
 include/net/xdp.h      |   4 +
 net/core/filter.c      |  11 +-
 net/core/skbuff.c      |   3 +-
 net/core/xdp.c         |   6 +-
 7 files changed, 753 insertions(+), 22 deletions(-)

-- 
2.14.3

[PATCH RFC v2 1/9] net: Export skb_headers_offset_update

[Toshiaki Makita]

From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>

v2:
- Drop skb_copy_header part because it has already been exported now.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
---
 include/linux/skbuff.h | 1 +
 net/core/skbuff.c      | 3 ++-
 2 files changed, 3 insertions(+), 1 deletion(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 164cdedf6012..2bdba543fda7 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -1030,6 +1030,7 @@ static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
 }
 
 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
+void skb_headers_offset_update(struct sk_buff *skb, int off);
 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index c642304f178c..180ab7d7f84f 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -1290,7 +1290,7 @@ struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
 }
 EXPORT_SYMBOL(skb_clone);
 
-static void skb_headers_offset_update(struct sk_buff *skb, int off)
+void skb_headers_offset_update(struct sk_buff *skb, int off)
 {
 	/* Only adjust this if it actually is csum_start rather than csum */
 	if (skb->ip_summed == CHECKSUM_PARTIAL)
@@ -1304,6 +1304,7 @@ static void skb_headers_offset_update(struct sk_buff *skb, int off)
 	skb->inner_network_header += off;
 	skb->inner_mac_header += off;
 }
+EXPORT_SYMBOL(skb_headers_offset_update);
 
 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
 {
-- 
2.14.3

[PATCH RFC v2 2/9] veth: Add driver XDP

[Toshiaki Makita]

From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>

This is basic implementation of veth driver XDP.

Incoming packets are sent from the peer veth device in the form of skb,
so this is generally doing the same thing as generic XDP.

This itself is not so useful, but a starting point to implement other
useful veth XDP features like TX and REDIRECT.

This introduces NAPI when XDP is enabled, because XDP is now heavily
relies on NAPI context. Use ptr_ring to emulate NIC ring. Tx function
enqueues packets to the ring and peer NAPI handler drains the ring.

Currently only one ring is allocated for each veth device, so it does
not scale on multiqueue env. This can be resolved by allocating rings
on the per-queue basis later.

Note that NAPI is not used but netif_rx is used when XDP is not loaded,
so this does not change the default behaviour.

v2:
- Squashed with the patch adding NAPI.
- Implement adjust_tail.
- Don't acquire consumer lock because it is guarded by NAPI.
- Make poll_controller noop since it is unnecessary.
- Register rxq_info on enabling XDP rather than on opening the device.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
---
 drivers/net/veth.c | 357 +++++++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 350 insertions(+), 7 deletions(-)

diff --git a/drivers/net/veth.c b/drivers/net/veth.c
index a69ad39ee57e..317ec92cf816 100644
--- a/drivers/net/veth.c
+++ b/drivers/net/veth.c
@@ -19,10 +19,18 @@
 #include <net/xfrm.h>
 #include <linux/veth.h>
 #include <linux/module.h>
+#include <linux/bpf.h>
+#include <linux/filter.h>
+#include <linux/ptr_ring.h>
+#include <linux/skb_array.h>
+#include <linux/bpf_trace.h>
 
 #define DRV_NAME	"veth"
 #define DRV_VERSION	"1.0"
 
+#define VETH_RING_SIZE		256
+#define VETH_XDP_HEADROOM	(XDP_PACKET_HEADROOM + NET_IP_ALIGN)
+
 struct pcpu_vstats {
 	u64			packets;
 	u64			bytes;
@@ -30,9 +38,15 @@ struct pcpu_vstats {
 };
 
 struct veth_priv {
+	struct napi_struct	xdp_napi;
+	struct net_device	*dev;
+	struct bpf_prog __rcu	*xdp_prog;
 	struct net_device __rcu	*peer;
 	atomic64_t		dropped;
 	unsigned		requested_headroom;
+	bool			rx_notify_masked;
+	struct ptr_ring		xdp_ring;
+	struct xdp_rxq_info	xdp_rxq;
 };
 
 /*
@@ -98,11 +112,43 @@ static const struct ethtool_ops veth_ethtool_ops = {
 	.get_link_ksettings	= veth_get_link_ksettings,
 };
 
-static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
+/* general routines */
+
+static void __veth_xdp_flush(struct veth_priv *priv)
+{
+	/* Write ptr_ring before reading rx_notify_masked */
+	smp_mb();
+	if (!priv->rx_notify_masked) {
+		priv->rx_notify_masked = true;
+		napi_schedule(&priv->xdp_napi);
+	}
+}
+
+static int veth_xdp_rx(struct veth_priv *priv, struct sk_buff *skb)
+{
+	if (unlikely(ptr_ring_produce(&priv->xdp_ring, skb))) {
+		dev_kfree_skb_any(skb);
+		return NET_RX_DROP;
+	}
+
+	return NET_RX_SUCCESS;
+}
+
+static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb, bool xdp)
 {
 	struct veth_priv *priv = netdev_priv(dev);
+
+	return __dev_forward_skb(dev, skb) ?: xdp ?
+		veth_xdp_rx(priv, skb) :
+		netif_rx(skb);
+}
+
+static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
 	struct net_device *rcv;
 	int length = skb->len;
+	bool rcv_xdp = false;
 
 	rcu_read_lock();
 	rcv = rcu_dereference(priv->peer);
@@ -111,7 +157,10 @@ static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 		goto drop;
 	}
 
-	if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) {
+	rcv_priv = netdev_priv(rcv);
+	rcv_xdp = rcu_access_pointer(rcv_priv->xdp_prog);
+
+	if (likely(veth_forward_skb(rcv, skb, rcv_xdp) == NET_RX_SUCCESS)) {
 		struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
 
 		u64_stats_update_begin(&stats->syncp);
@@ -122,14 +171,15 @@ static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 drop:
 		atomic64_inc(&priv->dropped);
 	}
+
+	if (rcv_xdp)
+		__veth_xdp_flush(rcv_priv);
+
 	rcu_read_unlock();
+
 	return NETDEV_TX_OK;
 }
 
-/*
- * general routines
- */
-
 static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev)
 {
 	struct veth_priv *priv = netdev_priv(dev);
@@ -179,18 +229,245 @@ static void veth_set_multicast_list(struct net_device *dev)
 {
 }
 
+static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
+				      int buflen)
+{
+	struct sk_buff *skb;
+
+	if (!buflen) {
+		buflen = SKB_DATA_ALIGN(headroom + len) +
+			 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
+	}
+	skb = build_skb(head, buflen);
+	if (!skb)
+		return NULL;
+
+	skb_reserve(skb, headroom);
+	skb_put(skb, len);
+
+	return skb;
+}
+
+static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
+					struct sk_buff *skb)
+{
+	u32 pktlen, headroom, act, metalen;
+	void *orig_data, *orig_data_end;
+	int size, mac_len, delta, off;
+	struct bpf_prog *xdp_prog;
+	struct xdp_buff xdp;
+
+	rcu_read_lock();
+	xdp_prog = rcu_dereference(priv->xdp_prog);
+	if (!xdp_prog) {
+		rcu_read_unlock();
+		goto out;
+	}
+
+	mac_len = skb->data - skb_mac_header(skb);
+	pktlen = skb->len + mac_len;
+	size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) +
+	       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
+	if (size > PAGE_SIZE)
+		goto drop;
+
+	headroom = skb_headroom(skb) - mac_len;
+	if (skb_shared(skb) || skb_head_is_locked(skb) ||
+	    skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) {
+		struct sk_buff *nskb;
+		void *head, *start;
+		struct page *page;
+		int head_off;
+
+		page = alloc_page(GFP_ATOMIC);
+		if (!page)
+			goto drop;
+
+		head = page_address(page);
+		start = head + VETH_XDP_HEADROOM;
+		if (skb_copy_bits(skb, -mac_len, start, pktlen)) {
+			page_frag_free(head);
+			goto drop;
+		}
+
+		nskb = veth_build_skb(head,
+				      VETH_XDP_HEADROOM + mac_len, skb->len,
+				      PAGE_SIZE);
+		if (!nskb) {
+			page_frag_free(head);
+			goto drop;
+		}
+
+		skb_copy_header(nskb, skb);
+		head_off = skb_headroom(nskb) - skb_headroom(skb);
+		skb_headers_offset_update(nskb, head_off);
+		dev_consume_skb_any(skb);
+		skb = nskb;
+	}
+
+	xdp.data_hard_start = skb->head;
+	xdp.data = skb_mac_header(skb);
+	xdp.data_end = xdp.data + pktlen;
+	xdp.data_meta = xdp.data;
+	xdp.rxq = &priv->xdp_rxq;
+	orig_data = xdp.data;
+	orig_data_end = xdp.data_end;
+
+	act = bpf_prog_run_xdp(xdp_prog, &xdp);
+
+	switch (act) {
+	case XDP_PASS:
+		break;
+	default:
+		bpf_warn_invalid_xdp_action(act);
+	case XDP_ABORTED:
+		trace_xdp_exception(priv->dev, xdp_prog, act);
+	case XDP_DROP:
+		goto drop;
+	}
+	rcu_read_unlock();
+
+	delta = orig_data - xdp.data;
+	off = mac_len + delta;
+	if (off > 0)
+		__skb_push(skb, off);
+	else if (off < 0)
+		__skb_pull(skb, -off);
+	skb->mac_header -= delta;
+	off = xdp.data_end - orig_data_end;
+	if (off != 0)
+		__skb_put(skb, off);
+	skb->protocol = eth_type_trans(skb, priv->dev);
+
+	metalen = xdp.data - xdp.data_meta;
+	if (metalen)
+		skb_metadata_set(skb, metalen);
+out:
+	return skb;
+drop:
+	rcu_read_unlock();
+	dev_kfree_skb_any(skb);
+	return NULL;
+}
+
+static int veth_xdp_rcv(struct veth_priv *priv, int budget)
+{
+	int i, done = 0;
+
+	for (i = 0; i < budget; i++) {
+		struct sk_buff *skb = __ptr_ring_consume(&priv->xdp_ring);
+
+		if (!skb)
+			break;
+
+		skb = veth_xdp_rcv_skb(priv, skb);
+
+		if (skb)
+			napi_gro_receive(&priv->xdp_napi, skb);
+
+		done++;
+	}
+
+	return done;
+}
+
+static int veth_poll(struct napi_struct *napi, int budget)
+{
+	struct veth_priv *priv =
+		container_of(napi, struct veth_priv, xdp_napi);
+	int done;
+
+	done = veth_xdp_rcv(priv, budget);
+
+	if (done < budget && napi_complete_done(napi, done)) {
+		/* Write rx_notify_masked before reading ptr_ring */
+		smp_store_mb(priv->rx_notify_masked, false);
+		if (unlikely(!__ptr_ring_empty(&priv->xdp_ring))) {
+			priv->rx_notify_masked = true;
+			napi_schedule(&priv->xdp_napi);
+		}
+	}
+
+	return done;
+}
+
+static int veth_napi_add(struct net_device *dev)
+{
+	struct veth_priv *priv = netdev_priv(dev);
+	int err;
+
+	err = ptr_ring_init(&priv->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
+	if (err)
+		return err;
+
+	netif_napi_add(dev, &priv->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
+	napi_enable(&priv->xdp_napi);
+
+	return 0;
+}
+
+static void veth_napi_del(struct net_device *dev)
+{
+	struct veth_priv *priv = netdev_priv(dev);
+
+	napi_disable(&priv->xdp_napi);
+	netif_napi_del(&priv->xdp_napi);
+	ptr_ring_cleanup(&priv->xdp_ring, __skb_array_destroy_skb);
+}
+
+static int veth_enable_xdp(struct net_device *dev)
+{
+	struct veth_priv *priv = netdev_priv(dev);
+	int err;
+
+	err = xdp_rxq_info_reg(&priv->xdp_rxq, dev, 0);
+	if (err < 0)
+		return err;
+
+	err = xdp_rxq_info_reg_mem_model(&priv->xdp_rxq,
+					 MEM_TYPE_PAGE_SHARED, NULL);
+	if (err < 0)
+		goto err;
+
+	err = veth_napi_add(dev);
+	if (err)
+		goto err;
+
+	return 0;
+err:
+	xdp_rxq_info_unreg(&priv->xdp_rxq);
+
+	return err;
+}
+
+static void veth_disable_xdp(struct net_device *dev)
+{
+	struct veth_priv *priv = netdev_priv(dev);
+
+	veth_napi_del(dev);
+	xdp_rxq_info_unreg(&priv->xdp_rxq);
+}
+
 static int veth_open(struct net_device *dev)
 {
 	struct veth_priv *priv = netdev_priv(dev);
 	struct net_device *peer = rtnl_dereference(priv->peer);
+	int err;
 
 	if (!peer)
 		return -ENOTCONN;
 
+	if (rtnl_dereference(priv->xdp_prog)) {
+		err = veth_enable_xdp(dev);
+		if (err)
+			return err;
+	}
+
 	if (peer->flags & IFF_UP) {
 		netif_carrier_on(dev);
 		netif_carrier_on(peer);
 	}
+
 	return 0;
 }
 
@@ -203,6 +480,9 @@ static int veth_close(struct net_device *dev)
 	if (peer)
 		netif_carrier_off(peer);
 
+	if (rtnl_dereference(priv->xdp_prog))
+		veth_disable_xdp(dev);
+
 	return 0;
 }
 
@@ -228,7 +508,7 @@ static void veth_dev_free(struct net_device *dev)
 static void veth_poll_controller(struct net_device *dev)
 {
 	/* veth only receives frames when its peer sends one
-	 * Since it's a synchronous operation, we are guaranteed
+	 * Since it has nothing to do with disabling irqs, we are guaranteed
 	 * never to have pending data when we poll for it so
 	 * there is nothing to do here.
 	 *
@@ -276,6 +556,65 @@ static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
 	rcu_read_unlock();
 }
 
+static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
+			struct netlink_ext_ack *extack)
+{
+	struct veth_priv *priv = netdev_priv(dev);
+	struct bpf_prog *old_prog;
+	struct net_device *peer;
+	int err;
+
+	old_prog = rtnl_dereference(priv->xdp_prog);
+	peer = rtnl_dereference(priv->peer);
+
+	if (prog) {
+		if (!peer)
+			return -ENOTCONN;
+
+		if (!old_prog && dev->flags & IFF_UP) {
+			err = veth_enable_xdp(dev);
+			if (err)
+				return err;
+		}
+	}
+
+	rcu_assign_pointer(priv->xdp_prog, prog);
+
+	if (old_prog) {
+		bpf_prog_put(old_prog);
+		if (!prog && dev->flags & IFF_UP)
+			veth_disable_xdp(dev);
+	}
+
+	return 0;
+}
+
+static u32 veth_xdp_query(struct net_device *dev)
+{
+	struct veth_priv *priv = netdev_priv(dev);
+	const struct bpf_prog *xdp_prog;
+
+	xdp_prog = rtnl_dereference(priv->xdp_prog);
+	if (xdp_prog)
+		return xdp_prog->aux->id;
+
+	return 0;
+}
+
+static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
+{
+	switch (xdp->command) {
+	case XDP_SETUP_PROG:
+		return veth_xdp_set(dev, xdp->prog, xdp->extack);
+	case XDP_QUERY_PROG:
+		xdp->prog_id = veth_xdp_query(dev);
+		xdp->prog_attached = !!xdp->prog_id;
+		return 0;
+	default:
+		return -EINVAL;
+	}
+}
+
 static const struct net_device_ops veth_netdev_ops = {
 	.ndo_init            = veth_dev_init,
 	.ndo_open            = veth_open,
@@ -290,6 +629,7 @@ static const struct net_device_ops veth_netdev_ops = {
 	.ndo_get_iflink		= veth_get_iflink,
 	.ndo_features_check	= passthru_features_check,
 	.ndo_set_rx_headroom	= veth_set_rx_headroom,
+	.ndo_bpf		= veth_xdp,
 };
 
 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
@@ -451,10 +791,13 @@ static int veth_newlink(struct net *src_net, struct net_device *dev,
 	 */
 
 	priv = netdev_priv(dev);
+	priv->dev = dev;
 	rcu_assign_pointer(priv->peer, peer);
 
 	priv = netdev_priv(peer);
+	priv->dev = peer;
 	rcu_assign_pointer(priv->peer, dev);
+
 	return 0;
 
 err_register_dev:
-- 
2.14.3

[PATCH RFC v2 3/9] veth: Avoid drops by oversized packets when XDP is enabled

[Toshiaki Makita]

From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>

All oversized packets including GSO packets are dropped if XDP is
enabled on receiver side, so don't send such packets from peer.

Drop TSO and SCTP fragmentation features so that veth devices themselves
segment packets with XDP enabled. Also cap MTU accordingly.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
---
 drivers/net/veth.c | 49 +++++++++++++++++++++++++++++++++++++++++++------
 1 file changed, 43 insertions(+), 6 deletions(-)

diff --git a/drivers/net/veth.c b/drivers/net/veth.c
index 317ec92cf816..88d349da72cc 100644
--- a/drivers/net/veth.c
+++ b/drivers/net/veth.c
@@ -533,6 +533,23 @@ static int veth_get_iflink(const struct net_device *dev)
 	return iflink;
 }
 
+static netdev_features_t veth_fix_features(struct net_device *dev,
+					   netdev_features_t features)
+{
+	struct veth_priv *priv = netdev_priv(dev);
+	struct net_device *peer;
+
+	peer = rtnl_dereference(priv->peer);
+	if (peer) {
+		struct veth_priv *peer_priv = netdev_priv(peer);
+
+		if (rtnl_dereference(peer_priv->xdp_prog))
+			features &= ~NETIF_F_GSO_SOFTWARE;
+	}
+
+	return features;
+}
+
 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
 {
 	struct veth_priv *peer_priv, *priv = netdev_priv(dev);
@@ -571,10 +588,19 @@ static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
 		if (!peer)
 			return -ENOTCONN;
 
-		if (!old_prog && dev->flags & IFF_UP) {
-			err = veth_enable_xdp(dev);
-			if (err)
-				return err;
+		if (!old_prog) {
+			if (dev->flags & IFF_UP) {
+				err = veth_enable_xdp(dev);
+				if (err)
+					return err;
+			}
+
+			peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
+			peer->max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
+				peer->hard_header_len -
+				SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
+			if (peer->mtu > peer->max_mtu)
+				dev_set_mtu(peer, peer->max_mtu);
 		}
 	}
 
@@ -582,10 +608,20 @@ static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
 
 	if (old_prog) {
 		bpf_prog_put(old_prog);
-		if (!prog && dev->flags & IFF_UP)
-			veth_disable_xdp(dev);
+		if (!prog) {
+			if (dev->flags & IFF_UP)
+				veth_disable_xdp(dev);
+
+			if (peer) {
+				peer->hw_features |= NETIF_F_GSO_SOFTWARE;
+				peer->max_mtu = ETH_MAX_MTU;
+			}
+		}
 	}
 
+	if ((!!old_prog ^ !!prog) && peer)
+		netdev_update_features(peer);
+
 	return 0;
 }
 
@@ -627,6 +663,7 @@ static const struct net_device_ops veth_netdev_ops = {
 	.ndo_poll_controller	= veth_poll_controller,
 #endif
 	.ndo_get_iflink		= veth_get_iflink,
+	.ndo_fix_features	= veth_fix_features,
 	.ndo_features_check	= passthru_features_check,
 	.ndo_set_rx_headroom	= veth_set_rx_headroom,
 	.ndo_bpf		= veth_xdp,
-- 
2.14.3

[PATCH RFC v2 4/9] veth: Add another napi ring for ndo_xdp_xmit and handle xdp_frames

[Toshiaki Makita]

From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>

This is preparation for XDP TX and ndo_xdp_xmit.
Add another napi ring and handle redirected xdp_frames through it.

v2:
- Use another ring instead of using flag to differentiate skb and
  xdp_frame. This approach makes bulk skb transmit possible in
  veth_xmit later.
- Clear xdp_frame feilds in skb->head.
- Implement adjust_tail.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
---
 drivers/net/veth.c | 125 ++++++++++++++++++++++++++++++++++++++++++++++++-----
 1 file changed, 114 insertions(+), 11 deletions(-)

diff --git a/drivers/net/veth.c b/drivers/net/veth.c
index 88d349da72cc..cb3fa558fbe0 100644
--- a/drivers/net/veth.c
+++ b/drivers/net/veth.c
@@ -46,6 +46,7 @@ struct veth_priv {
 	unsigned		requested_headroom;
 	bool			rx_notify_masked;
 	struct ptr_ring		xdp_ring;
+	struct ptr_ring		xdp_tx_ring;
 	struct xdp_rxq_info	xdp_rxq;
 };
 
@@ -114,6 +115,11 @@ static const struct ethtool_ops veth_ethtool_ops = {
 
 /* general routines */
 
+static void veth_xdp_free(void *frame)
+{
+	xdp_return_frame(frame);
+}
+
 static void __veth_xdp_flush(struct veth_priv *priv)
 {
 	/* Write ptr_ring before reading rx_notify_masked */
@@ -248,6 +254,61 @@ static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
 	return skb;
 }
 
+static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
+					struct xdp_frame *frame)
+{
+	int len = frame->len, delta = 0;
+	struct bpf_prog *xdp_prog;
+	unsigned int headroom;
+	struct sk_buff *skb;
+
+	rcu_read_lock();
+	xdp_prog = rcu_dereference(priv->xdp_prog);
+	if (xdp_prog) {
+		struct xdp_buff xdp;
+		u32 act;
+
+		xdp.data_hard_start = frame->data - frame->headroom;
+		xdp.data = frame->data;
+		xdp.data_end = frame->data + frame->len;
+		xdp.data_meta = frame->data - frame->metasize;
+		xdp.rxq = &priv->xdp_rxq;
+
+		act = bpf_prog_run_xdp(xdp_prog, &xdp);
+
+		switch (act) {
+		case XDP_PASS:
+			delta = frame->data - xdp.data;
+			len = xdp.data_end - xdp.data;
+			break;
+		default:
+			bpf_warn_invalid_xdp_action(act);
+		case XDP_ABORTED:
+			trace_xdp_exception(priv->dev, xdp_prog, act);
+		case XDP_DROP:
+			goto err_xdp;
+		}
+	}
+	rcu_read_unlock();
+
+	headroom = frame->data - delta - (void *)frame;
+	skb = veth_build_skb(frame, headroom, len, 0);
+	if (!skb) {
+		xdp_return_frame(frame);
+		goto err;
+	}
+
+	memset(frame, 0, sizeof(*frame));
+	skb->protocol = eth_type_trans(skb, priv->dev);
+err:
+	return skb;
+err_xdp:
+	rcu_read_unlock();
+	xdp_return_frame(frame);
+
+	return NULL;
+}
+
 static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
 					struct sk_buff *skb)
 {
@@ -352,21 +413,53 @@ static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
 
 static int veth_xdp_rcv(struct veth_priv *priv, int budget)
 {
-	int i, done = 0;
+	int done = 0;
+	bool more;
 
-	for (i = 0; i < budget; i++) {
-		struct sk_buff *skb = __ptr_ring_consume(&priv->xdp_ring);
+	do {
+		int curr_budget, i;
+		bool curr_more;
 
-		if (!skb)
-			break;
+		more = false;
 
-		skb = veth_xdp_rcv_skb(priv, skb);
+		curr_more = true;
+		curr_budget = min(budget - done, budget >> 1);
+		for (i = 0; i < curr_budget; i++) {
+			struct xdp_frame *frame;
+			struct sk_buff *skb;
 
-		if (skb)
-			napi_gro_receive(&priv->xdp_napi, skb);
+			frame = __ptr_ring_consume(&priv->xdp_tx_ring);
+			if (!frame) {
+				curr_more = false;
+				break;
+			}
 
-		done++;
-	}
+			skb = veth_xdp_rcv_one(priv, frame);
+			if (skb)
+				napi_gro_receive(&priv->xdp_napi, skb);
+
+			done++;
+		}
+		more |= curr_more;
+
+		curr_more = true;
+		curr_budget = min(budget - done, budget >> 1);
+		for (i = 0; i < curr_budget; i++) {
+			struct sk_buff *skb = __ptr_ring_consume(&priv->xdp_ring);
+
+			if (!skb) {
+				curr_more = false;
+				break;
+			}
+
+			skb = veth_xdp_rcv_skb(priv, skb);
+			if (skb)
+				napi_gro_receive(&priv->xdp_napi, skb);
+
+			done++;
+		}
+		more |= curr_more;
+	} while (more && done < budget);
 
 	return done;
 }
@@ -382,7 +475,8 @@ static int veth_poll(struct napi_struct *napi, int budget)
 	if (done < budget && napi_complete_done(napi, done)) {
 		/* Write rx_notify_masked before reading ptr_ring */
 		smp_store_mb(priv->rx_notify_masked, false);
-		if (unlikely(!__ptr_ring_empty(&priv->xdp_ring))) {
+		if (unlikely(!__ptr_ring_empty(&priv->xdp_tx_ring) ||
+			     !__ptr_ring_empty(&priv->xdp_ring))) {
 			priv->rx_notify_masked = true;
 			napi_schedule(&priv->xdp_napi);
 		}
@@ -400,10 +494,18 @@ static int veth_napi_add(struct net_device *dev)
 	if (err)
 		return err;
 
+	err = ptr_ring_init(&priv->xdp_tx_ring, VETH_RING_SIZE, GFP_KERNEL);
+	if (err)
+		goto err_xdp_tx_ring;
+
 	netif_napi_add(dev, &priv->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
 	napi_enable(&priv->xdp_napi);
 
 	return 0;
+err_xdp_tx_ring:
+	ptr_ring_cleanup(&priv->xdp_ring, __skb_array_destroy_skb);
+
+	return err;
 }
 
 static void veth_napi_del(struct net_device *dev)
@@ -413,6 +515,7 @@ static void veth_napi_del(struct net_device *dev)
 	napi_disable(&priv->xdp_napi);
 	netif_napi_del(&priv->xdp_napi);
 	ptr_ring_cleanup(&priv->xdp_ring, __skb_array_destroy_skb);
+	ptr_ring_cleanup(&priv->xdp_tx_ring, veth_xdp_free);
 }
 
 static int veth_enable_xdp(struct net_device *dev)
-- 
2.14.3

[PATCH RFC v2 5/9] veth: Add ndo_xdp_xmit

[Toshiaki Makita]

From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>

This allows NIC's XDP to redirect packets to veth. The destination veth
device enqueues redirected packets to the napi ring of its peer, then
they are processed by XDP on its peer veth device.
This can be thought as calling another XDP program by XDP program using
REDIRECT, when the peer enables driver XDP.

Note that when the peer veth device does not set driver xdp, redirected
packets will be dropped because the peer is not ready for NAPI.

v2:
- Drop the part converting xdp_frame into skb when XDP is not enabled.
- Implement bulk interface of ndo_xdp_xmit.
- Implement XDP_XMIT_FLUSH bit and drop ndo_xdp_flush.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
---
 drivers/net/veth.c     | 39 +++++++++++++++++++++++++++++++++++++++
 include/linux/filter.h | 16 ++++++++++++++++
 net/core/filter.c      | 11 +----------
 3 files changed, 56 insertions(+), 10 deletions(-)

diff --git a/drivers/net/veth.c b/drivers/net/veth.c
index cb3fa558fbe0..b809d609a642 100644
--- a/drivers/net/veth.c
+++ b/drivers/net/veth.c
@@ -17,6 +17,7 @@
 #include <net/rtnetlink.h>
 #include <net/dst.h>
 #include <net/xfrm.h>
+#include <net/xdp.h>
 #include <linux/veth.h>
 #include <linux/module.h>
 #include <linux/bpf.h>
@@ -254,6 +255,43 @@ static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
 	return skb;
 }
 
+static int veth_xdp_xmit(struct net_device *dev, int n,
+			 struct xdp_frame **frames, u32 flags)
+{
+	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
+	struct net_device *rcv;
+	int i, drops = 0;
+
+	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
+		return -EINVAL;
+
+	rcv = rcu_dereference(priv->peer);
+	if (unlikely(!rcv))
+		return -ENXIO;
+
+	rcv_priv = netdev_priv(rcv);
+	/* xdp_ring is initialized on receive side? */
+	if (!rcu_access_pointer(rcv_priv->xdp_prog))
+		return -ENXIO;
+
+	spin_lock(&rcv_priv->xdp_tx_ring.producer_lock);
+	for (i = 0; i < n; i++) {
+		struct xdp_frame *frame = frames[i];
+
+		if (unlikely(xdp_ok_fwd_dev(rcv, frame->len) ||
+			     __ptr_ring_produce(&rcv_priv->xdp_tx_ring, frame))) {
+			xdp_return_frame_rx_napi(frame);
+			drops++;
+		}
+	}
+	spin_unlock(&rcv_priv->xdp_tx_ring.producer_lock);
+
+	if (flags & XDP_XMIT_FLUSH)
+		__veth_xdp_flush(rcv_priv);
+
+	return n - drops;
+}
+
 static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
 					struct xdp_frame *frame)
 {
@@ -770,6 +808,7 @@ static const struct net_device_ops veth_netdev_ops = {
 	.ndo_features_check	= passthru_features_check,
 	.ndo_set_rx_headroom	= veth_set_rx_headroom,
 	.ndo_bpf		= veth_xdp,
+	.ndo_xdp_xmit		= veth_xdp_xmit,
 };
 
 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
diff --git a/include/linux/filter.h b/include/linux/filter.h
index 45fc0f5000d8..12777eb70b40 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -19,6 +19,7 @@
 #include <linux/cryptohash.h>
 #include <linux/set_memory.h>
 #include <linux/kallsyms.h>
+#include <linux/if_vlan.h>
 
 #include <net/sch_generic.h>
 
@@ -786,6 +787,21 @@ static inline bool bpf_dump_raw_ok(void)
 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
 				       const struct bpf_insn *patch, u32 len);
 
+static __always_inline int
+xdp_ok_fwd_dev(const struct net_device *fwd, unsigned int pktlen)
+{
+	unsigned int len;
+
+	if (unlikely(!(fwd->flags & IFF_UP)))
+		return -ENETDOWN;
+
+	len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
+	if (pktlen > len)
+		return -EMSGSIZE;
+
+	return 0;
+}
+
 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
  * same cpu context. Further for best results no more than a single map
  * for the do_redirect/do_flush pair should be used. This limitation is
diff --git a/net/core/filter.c b/net/core/filter.c
index 3d9ba7e5965a..05d9e84566a4 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -3216,16 +3216,7 @@ EXPORT_SYMBOL_GPL(xdp_do_redirect);
 
 static int __xdp_generic_ok_fwd_dev(struct sk_buff *skb, struct net_device *fwd)
 {
-	unsigned int len;
-
-	if (unlikely(!(fwd->flags & IFF_UP)))
-		return -ENETDOWN;
-
-	len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
-	if (skb->len > len)
-		return -EMSGSIZE;
-
-	return 0;
+	return xdp_ok_fwd_dev(fwd, skb->len);
 }
 
 static int xdp_do_generic_redirect_map(struct net_device *dev,
-- 
2.14.3

[PATCH RFC v2 6/9] xdp: Add a flag for disabling napi_direct of xdp_return_frame in xdp_mem_info

[Toshiaki Makita]

From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>

We need some mechanism to disable napi_direct on calling
xdp_return_frame_rx_napi() from some context.
When veth gets support of XDP_REDIRECT, it will redirects packets which
are redirected from other devices. On redirection veth will reuse
xdp_mem_info of the redirection source device to make return_frame work.
But in this case .ndo_xdp_xmit() called from veth redirection uses
xdp_mem_info which is not guarded by NAPI, because the .ndo_xdp_xmit is
not called directly from the rxq which owns the xdp_mem_info.

This approach introduces a flag in xdp_mem_info to indicate that
napi_direct should be disabled even when _rx_napi variant is used.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
---
 include/net/xdp.h | 4 ++++
 net/core/xdp.c    | 6 ++++--
 2 files changed, 8 insertions(+), 2 deletions(-)

diff --git a/include/net/xdp.h b/include/net/xdp.h
index 2deea7166a34..ea0c80f6c8ee 100644
--- a/include/net/xdp.h
+++ b/include/net/xdp.h
@@ -41,6 +41,9 @@ enum xdp_mem_type {
 	MEM_TYPE_MAX,
 };
 
+/* XDP flags for xdp_mem_info */
+#define XDP_MEM_RF_NO_DIRECT	BIT(0)	/* don't use napi_direct */
+
 /* XDP flags for ndo_xdp_xmit */
 #define XDP_XMIT_FLUSH		(1U << 0)	/* doorbell signal consumer */
 #define XDP_XMIT_FLAGS_MASK	XDP_XMIT_FLUSH
@@ -48,6 +51,7 @@ enum xdp_mem_type {
 struct xdp_mem_info {
 	u32 type; /* enum xdp_mem_type, but known size type */
 	u32 id;
+	u32 flags;
 };
 
 struct page_pool;
diff --git a/net/core/xdp.c b/net/core/xdp.c
index 9d1f22072d5d..e94f146360b2 100644
--- a/net/core/xdp.c
+++ b/net/core/xdp.c
@@ -327,10 +327,12 @@ static void __xdp_return(void *data, struct xdp_mem_info *mem, bool napi_direct,
 		/* mem->id is valid, checked in xdp_rxq_info_reg_mem_model() */
 		xa = rhashtable_lookup(mem_id_ht, &mem->id, mem_id_rht_params);
 		page = virt_to_head_page(data);
-		if (xa)
+		if (xa) {
+			napi_direct &= !(mem->flags & XDP_MEM_RF_NO_DIRECT);
 			page_pool_put_page(xa->page_pool, page, napi_direct);
-		else
+		} else {
 			put_page(page);
+		}
 		rcu_read_unlock();
 		break;
 	case MEM_TYPE_PAGE_SHARED:
-- 
2.14.3

[PATCH RFC v2 7/9] veth: Add XDP TX and REDIRECT

[Toshiaki Makita]

From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>

This allows further redirection of xdp_frames like

 NIC   -> veth--veth -> veth--veth
 (XDP)          (XDP)         (XDP)

The intermediate XDP, redirecting packets from NIC to the other veth,
reuses xdp_mem_info from NIC so that page recycling of the NIC works on
the destination veth's XDP.
In this way return_frame is not fully guarded by NAPI, since another
NAPI handler on another cpu may use the same xdp_mem_info concurrently.
Thus disable napi_direct by XDP_MEM_RF_NO_DIRECT flag.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
---
 drivers/net/veth.c | 110 +++++++++++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 103 insertions(+), 7 deletions(-)

diff --git a/drivers/net/veth.c b/drivers/net/veth.c
index b809d609a642..a47e1ba7d7e6 100644
--- a/drivers/net/veth.c
+++ b/drivers/net/veth.c
@@ -44,6 +44,7 @@ struct veth_priv {
 	struct bpf_prog __rcu	*xdp_prog;
 	struct net_device __rcu	*peer;
 	atomic64_t		dropped;
+	struct xdp_mem_info	xdp_mem;
 	unsigned		requested_headroom;
 	bool			rx_notify_masked;
 	struct ptr_ring		xdp_ring;
@@ -292,10 +293,42 @@ static int veth_xdp_xmit(struct net_device *dev, int n,
 	return n - drops;
 }
 
+static void veth_xdp_flush(struct net_device *dev)
+{
+	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
+	struct net_device *rcv;
+
+	rcu_read_lock();
+	rcv = rcu_dereference(priv->peer);
+	if (unlikely(!rcv))
+		goto out;
+
+	rcv_priv = netdev_priv(rcv);
+	/* xdp_ring is initialized on receive side? */
+	if (unlikely(!rcu_access_pointer(rcv_priv->xdp_prog)))
+		goto out;
+
+	__veth_xdp_flush(rcv_priv);
+out:
+	rcu_read_unlock();
+}
+
+static int veth_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
+{
+	struct xdp_frame *frame = convert_to_xdp_frame(xdp);
+
+	if (unlikely(!frame))
+		return -EOVERFLOW;
+
+	return veth_xdp_xmit(dev, 1, &frame, 0);
+}
+
 static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
-					struct xdp_frame *frame)
+					struct xdp_frame *frame, bool *xdp_xmit,
+					bool *xdp_redir)
 {
 	int len = frame->len, delta = 0;
+	struct xdp_frame orig_frame;
 	struct bpf_prog *xdp_prog;
 	unsigned int headroom;
 	struct sk_buff *skb;
@@ -319,6 +352,31 @@ static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
 			delta = frame->data - xdp.data;
 			len = xdp.data_end - xdp.data;
 			break;
+		case XDP_TX:
+			orig_frame = *frame;
+			xdp.data_hard_start = frame;
+			xdp.rxq->mem = frame->mem;
+			xdp.rxq->mem.flags |= XDP_MEM_RF_NO_DIRECT;
+			if (unlikely(veth_xdp_tx(priv->dev, &xdp))) {
+				trace_xdp_exception(priv->dev, xdp_prog, act);
+				frame = &orig_frame;
+				goto err_xdp;
+			}
+			*xdp_xmit = true;
+			rcu_read_unlock();
+			goto xdp_xmit;
+		case XDP_REDIRECT:
+			orig_frame = *frame;
+			xdp.data_hard_start = frame;
+			xdp.rxq->mem = frame->mem;
+			xdp.rxq->mem.flags |= XDP_MEM_RF_NO_DIRECT;
+			if (xdp_do_redirect(priv->dev, &xdp, xdp_prog)) {
+				frame = &orig_frame;
+				goto err_xdp;
+			}
+			*xdp_redir = true;
+			rcu_read_unlock();
+			goto xdp_xmit;
 		default:
 			bpf_warn_invalid_xdp_action(act);
 		case XDP_ABORTED:
@@ -343,12 +401,13 @@ static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
 err_xdp:
 	rcu_read_unlock();
 	xdp_return_frame(frame);
-
+xdp_xmit:
 	return NULL;
 }
 
 static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
-					struct sk_buff *skb)
+					struct sk_buff *skb, bool *xdp_xmit,
+					bool *xdp_redir)
 {
 	u32 pktlen, headroom, act, metalen;
 	void *orig_data, *orig_data_end;
@@ -417,6 +476,26 @@ static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
 	switch (act) {
 	case XDP_PASS:
 		break;
+	case XDP_TX:
+		get_page(virt_to_page(xdp.data));
+		dev_consume_skb_any(skb);
+		xdp.rxq->mem = priv->xdp_mem;
+		if (unlikely(veth_xdp_tx(priv->dev, &xdp))) {
+			trace_xdp_exception(priv->dev, xdp_prog, act);
+			goto err_xdp;
+		}
+		*xdp_xmit = true;
+		rcu_read_unlock();
+		goto xdp_xmit;
+	case XDP_REDIRECT:
+		get_page(virt_to_page(xdp.data));
+		dev_consume_skb_any(skb);
+		xdp.rxq->mem = priv->xdp_mem;
+		if (xdp_do_redirect(priv->dev, &xdp, xdp_prog))
+			goto err_xdp;
+		*xdp_redir = true;
+		rcu_read_unlock();
+		goto xdp_xmit;
 	default:
 		bpf_warn_invalid_xdp_action(act);
 	case XDP_ABORTED:
@@ -447,9 +526,15 @@ static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
 	rcu_read_unlock();
 	dev_kfree_skb_any(skb);
 	return NULL;
+err_xdp:
+	rcu_read_unlock();
+	page_frag_free(xdp.data);
+xdp_xmit:
+	return NULL;
 }
 
-static int veth_xdp_rcv(struct veth_priv *priv, int budget)
+static int veth_xdp_rcv(struct veth_priv *priv, int budget, bool *xdp_xmit,
+			bool *xdp_redir)
 {
 	int done = 0;
 	bool more;
@@ -472,7 +557,7 @@ static int veth_xdp_rcv(struct veth_priv *priv, int budget)
 				break;
 			}
 
-			skb = veth_xdp_rcv_one(priv, frame);
+			skb = veth_xdp_rcv_one(priv, frame, xdp_xmit, xdp_redir);
 			if (skb)
 				napi_gro_receive(&priv->xdp_napi, skb);
 
@@ -490,7 +575,7 @@ static int veth_xdp_rcv(struct veth_priv *priv, int budget)
 				break;
 			}
 
-			skb = veth_xdp_rcv_skb(priv, skb);
+			skb = veth_xdp_rcv_skb(priv, skb, xdp_xmit, xdp_redir);
 			if (skb)
 				napi_gro_receive(&priv->xdp_napi, skb);
 
@@ -506,9 +591,11 @@ static int veth_poll(struct napi_struct *napi, int budget)
 {
 	struct veth_priv *priv =
 		container_of(napi, struct veth_priv, xdp_napi);
+	bool xdp_xmit = false;
+	bool xdp_redir = false;
 	int done;
 
-	done = veth_xdp_rcv(priv, budget);
+	done = veth_xdp_rcv(priv, budget, &xdp_xmit, &xdp_redir);
 
 	if (done < budget && napi_complete_done(napi, done)) {
 		/* Write rx_notify_masked before reading ptr_ring */
@@ -520,6 +607,11 @@ static int veth_poll(struct napi_struct *napi, int budget)
 		}
 	}
 
+	if (xdp_xmit)
+		veth_xdp_flush(priv->dev);
+	if (xdp_redir)
+		xdp_do_flush_map();
+
 	return done;
 }
 
@@ -570,6 +662,9 @@ static int veth_enable_xdp(struct net_device *dev)
 	if (err < 0)
 		goto err;
 
+	/* Save original mem info as it can be overwritten */
+	priv->xdp_mem = priv->xdp_rxq.mem;
+
 	err = veth_napi_add(dev);
 	if (err)
 		goto err;
@@ -586,6 +681,7 @@ static void veth_disable_xdp(struct net_device *dev)
 	struct veth_priv *priv = netdev_priv(dev);
 
 	veth_napi_del(dev);
+	priv->xdp_rxq.mem = priv->xdp_mem;
 	xdp_rxq_info_unreg(&priv->xdp_rxq);
 }
 
-- 
2.14.3

[PATCH RFC v2 8/9] veth: Support per queue XDP ring

[Toshiaki Makita]

From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>

Move XDP and napi related fields in veth_priv to newly created veth_rq
structure.

When xdp_frames are enqueued from ndo_xdp_xmit and XDP_TX, rxq is
selected by current cpu.
When skbs are enqueued from the peer device, rxq is determined by its
peer's txq. In this way we can implement bulk packet send using
skb->xmit_more later.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
---
 drivers/net/veth.c | 290 +++++++++++++++++++++++++++++++++++------------------
 1 file changed, 191 insertions(+), 99 deletions(-)

diff --git a/drivers/net/veth.c b/drivers/net/veth.c
index a47e1ba7d7e6..67debd3eafe6 100644
--- a/drivers/net/veth.c
+++ b/drivers/net/veth.c
@@ -38,20 +38,24 @@ struct pcpu_vstats {
 	struct u64_stats_sync	syncp;
 };
 
-struct veth_priv {
+struct veth_rq {
 	struct napi_struct	xdp_napi;
 	struct net_device	*dev;
 	struct bpf_prog __rcu	*xdp_prog;
-	struct net_device __rcu	*peer;
-	atomic64_t		dropped;
 	struct xdp_mem_info	xdp_mem;
-	unsigned		requested_headroom;
 	bool			rx_notify_masked;
 	struct ptr_ring		xdp_ring;
 	struct ptr_ring		xdp_tx_ring;
 	struct xdp_rxq_info	xdp_rxq;
 };
 
+struct veth_priv {
+	struct net_device __rcu	*peer;
+	atomic64_t		dropped;
+	struct veth_rq		*rq;
+	unsigned int		requested_headroom;
+};
+
 /*
  * ethtool interface
  */
@@ -122,19 +126,19 @@ static void veth_xdp_free(void *frame)
 	xdp_return_frame(frame);
 }
 
-static void __veth_xdp_flush(struct veth_priv *priv)
+static void __veth_xdp_flush(struct veth_rq *rq)
 {
 	/* Write ptr_ring before reading rx_notify_masked */
 	smp_mb();
-	if (!priv->rx_notify_masked) {
-		priv->rx_notify_masked = true;
-		napi_schedule(&priv->xdp_napi);
+	if (!rq->rx_notify_masked) {
+		rq->rx_notify_masked = true;
+		napi_schedule(&rq->xdp_napi);
 	}
 }
 
-static int veth_xdp_rx(struct veth_priv *priv, struct sk_buff *skb)
+static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
 {
-	if (unlikely(ptr_ring_produce(&priv->xdp_ring, skb))) {
+	if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
 		dev_kfree_skb_any(skb);
 		return NET_RX_DROP;
 	}
@@ -142,12 +146,11 @@ static int veth_xdp_rx(struct veth_priv *priv, struct sk_buff *skb)
 	return NET_RX_SUCCESS;
 }
 
-static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb, bool xdp)
+static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
+			    struct veth_rq *rq, bool xdp)
 {
-	struct veth_priv *priv = netdev_priv(dev);
-
 	return __dev_forward_skb(dev, skb) ?: xdp ?
-		veth_xdp_rx(priv, skb) :
+		veth_xdp_rx(rq, skb) :
 		netif_rx(skb);
 }
 
@@ -157,6 +160,8 @@ static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 	struct net_device *rcv;
 	int length = skb->len;
 	bool rcv_xdp = false;
+	struct veth_rq *rq;
+	int rxq;
 
 	rcu_read_lock();
 	rcv = rcu_dereference(priv->peer);
@@ -166,9 +171,12 @@ static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 	}
 
 	rcv_priv = netdev_priv(rcv);
-	rcv_xdp = rcu_access_pointer(rcv_priv->xdp_prog);
+	rxq = skb_get_queue_mapping(skb);
+	skb_record_rx_queue(skb, rxq);
+	rq = &rcv_priv->rq[rxq];
+	rcv_xdp = rcu_access_pointer(rq->xdp_prog);
 
-	if (likely(veth_forward_skb(rcv, skb, rcv_xdp) == NET_RX_SUCCESS)) {
+	if (likely(veth_forward_skb(rcv, skb, rq, rcv_xdp) == NET_RX_SUCCESS)) {
 		struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
 
 		u64_stats_update_begin(&stats->syncp);
@@ -181,7 +189,7 @@ static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 	}
 
 	if (rcv_xdp)
-		__veth_xdp_flush(rcv_priv);
+		__veth_xdp_flush(rq);
 
 	rcu_read_unlock();
 
@@ -256,11 +264,17 @@ static struct sk_buff *veth_build_skb(void *head, int headroom, int len,
 	return skb;
 }
 
+static int veth_select_rxq(struct net_device *dev)
+{
+	return smp_processor_id() % dev->real_num_rx_queues;
+}
+
 static int veth_xdp_xmit(struct net_device *dev, int n,
 			 struct xdp_frame **frames, u32 flags)
 {
 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
 	struct net_device *rcv;
+	struct veth_rq *rq;
 	int i, drops = 0;
 
 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
@@ -271,24 +285,25 @@ static int veth_xdp_xmit(struct net_device *dev, int n,
 		return -ENXIO;
 
 	rcv_priv = netdev_priv(rcv);
+	rq = &rcv_priv->rq[veth_select_rxq(rcv)];
 	/* xdp_ring is initialized on receive side? */
-	if (!rcu_access_pointer(rcv_priv->xdp_prog))
+	if (!rcu_access_pointer(rq->xdp_prog))
 		return -ENXIO;
 
-	spin_lock(&rcv_priv->xdp_tx_ring.producer_lock);
+	spin_lock(&rq->xdp_tx_ring.producer_lock);
 	for (i = 0; i < n; i++) {
 		struct xdp_frame *frame = frames[i];
 
 		if (unlikely(xdp_ok_fwd_dev(rcv, frame->len) ||
-			     __ptr_ring_produce(&rcv_priv->xdp_tx_ring, frame))) {
+			     __ptr_ring_produce(&rq->xdp_tx_ring, frame))) {
 			xdp_return_frame_rx_napi(frame);
 			drops++;
 		}
 	}
-	spin_unlock(&rcv_priv->xdp_tx_ring.producer_lock);
+	spin_unlock(&rq->xdp_tx_ring.producer_lock);
 
 	if (flags & XDP_XMIT_FLUSH)
-		__veth_xdp_flush(rcv_priv);
+		__veth_xdp_flush(rq);
 
 	return n - drops;
 }
@@ -297,6 +312,7 @@ static void veth_xdp_flush(struct net_device *dev)
 {
 	struct veth_priv *rcv_priv, *priv = netdev_priv(dev);
 	struct net_device *rcv;
+	struct veth_rq *rq;
 
 	rcu_read_lock();
 	rcv = rcu_dereference(priv->peer);
@@ -304,11 +320,12 @@ static void veth_xdp_flush(struct net_device *dev)
 		goto out;
 
 	rcv_priv = netdev_priv(rcv);
+	rq = &rcv_priv->rq[veth_select_rxq(rcv)];
 	/* xdp_ring is initialized on receive side? */
-	if (unlikely(!rcu_access_pointer(rcv_priv->xdp_prog)))
+	if (unlikely(!rcu_access_pointer(rq->xdp_prog)))
 		goto out;
 
-	__veth_xdp_flush(rcv_priv);
+	__veth_xdp_flush(rq);
 out:
 	rcu_read_unlock();
 }
@@ -323,7 +340,7 @@ static int veth_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
 	return veth_xdp_xmit(dev, 1, &frame, 0);
 }
 
-static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
+static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq,
 					struct xdp_frame *frame, bool *xdp_xmit,
 					bool *xdp_redir)
 {
@@ -334,7 +351,7 @@ static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
 	struct sk_buff *skb;
 
 	rcu_read_lock();
-	xdp_prog = rcu_dereference(priv->xdp_prog);
+	xdp_prog = rcu_dereference(rq->xdp_prog);
 	if (xdp_prog) {
 		struct xdp_buff xdp;
 		u32 act;
@@ -343,7 +360,7 @@ static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
 		xdp.data = frame->data;
 		xdp.data_end = frame->data + frame->len;
 		xdp.data_meta = frame->data - frame->metasize;
-		xdp.rxq = &priv->xdp_rxq;
+		xdp.rxq = &rq->xdp_rxq;
 
 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
 
@@ -357,8 +374,8 @@ static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
 			xdp.data_hard_start = frame;
 			xdp.rxq->mem = frame->mem;
 			xdp.rxq->mem.flags |= XDP_MEM_RF_NO_DIRECT;
-			if (unlikely(veth_xdp_tx(priv->dev, &xdp))) {
-				trace_xdp_exception(priv->dev, xdp_prog, act);
+			if (unlikely(veth_xdp_tx(rq->dev, &xdp))) {
+				trace_xdp_exception(rq->dev, xdp_prog, act);
 				frame = &orig_frame;
 				goto err_xdp;
 			}
@@ -370,7 +387,7 @@ static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
 			xdp.data_hard_start = frame;
 			xdp.rxq->mem = frame->mem;
 			xdp.rxq->mem.flags |= XDP_MEM_RF_NO_DIRECT;
-			if (xdp_do_redirect(priv->dev, &xdp, xdp_prog)) {
+			if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) {
 				frame = &orig_frame;
 				goto err_xdp;
 			}
@@ -380,7 +397,7 @@ static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
 		default:
 			bpf_warn_invalid_xdp_action(act);
 		case XDP_ABORTED:
-			trace_xdp_exception(priv->dev, xdp_prog, act);
+			trace_xdp_exception(rq->dev, xdp_prog, act);
 		case XDP_DROP:
 			goto err_xdp;
 		}
@@ -395,7 +412,7 @@ static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
 	}
 
 	memset(frame, 0, sizeof(*frame));
-	skb->protocol = eth_type_trans(skb, priv->dev);
+	skb->protocol = eth_type_trans(skb, rq->dev);
 err:
 	return skb;
 err_xdp:
@@ -405,9 +422,8 @@ static struct sk_buff *veth_xdp_rcv_one(struct veth_priv *priv,
 	return NULL;
 }
 
-static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
-					struct sk_buff *skb, bool *xdp_xmit,
-					bool *xdp_redir)
+static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq, struct sk_buff *skb,
+					bool *xdp_xmit, bool *xdp_redir)
 {
 	u32 pktlen, headroom, act, metalen;
 	void *orig_data, *orig_data_end;
@@ -416,7 +432,7 @@ static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
 	struct xdp_buff xdp;
 
 	rcu_read_lock();
-	xdp_prog = rcu_dereference(priv->xdp_prog);
+	xdp_prog = rcu_dereference(rq->xdp_prog);
 	if (!xdp_prog) {
 		rcu_read_unlock();
 		goto out;
@@ -467,7 +483,7 @@ static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
 	xdp.data = skb_mac_header(skb);
 	xdp.data_end = xdp.data + pktlen;
 	xdp.data_meta = xdp.data;
-	xdp.rxq = &priv->xdp_rxq;
+	xdp.rxq = &rq->xdp_rxq;
 	orig_data = xdp.data;
 	orig_data_end = xdp.data_end;
 
@@ -479,9 +495,9 @@ static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
 	case XDP_TX:
 		get_page(virt_to_page(xdp.data));
 		dev_consume_skb_any(skb);
-		xdp.rxq->mem = priv->xdp_mem;
-		if (unlikely(veth_xdp_tx(priv->dev, &xdp))) {
-			trace_xdp_exception(priv->dev, xdp_prog, act);
+		xdp.rxq->mem = rq->xdp_mem;
+		if (unlikely(veth_xdp_tx(rq->dev, &xdp))) {
+			trace_xdp_exception(rq->dev, xdp_prog, act);
 			goto err_xdp;
 		}
 		*xdp_xmit = true;
@@ -490,8 +506,8 @@ static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
 	case XDP_REDIRECT:
 		get_page(virt_to_page(xdp.data));
 		dev_consume_skb_any(skb);
-		xdp.rxq->mem = priv->xdp_mem;
-		if (xdp_do_redirect(priv->dev, &xdp, xdp_prog))
+		xdp.rxq->mem = rq->xdp_mem;
+		if (xdp_do_redirect(rq->dev, &xdp, xdp_prog))
 			goto err_xdp;
 		*xdp_redir = true;
 		rcu_read_unlock();
@@ -499,7 +515,7 @@ static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
 	default:
 		bpf_warn_invalid_xdp_action(act);
 	case XDP_ABORTED:
-		trace_xdp_exception(priv->dev, xdp_prog, act);
+		trace_xdp_exception(rq->dev, xdp_prog, act);
 	case XDP_DROP:
 		goto drop;
 	}
@@ -515,7 +531,7 @@ static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
 	off = xdp.data_end - orig_data_end;
 	if (off != 0)
 		__skb_put(skb, off);
-	skb->protocol = eth_type_trans(skb, priv->dev);
+	skb->protocol = eth_type_trans(skb, rq->dev);
 
 	metalen = xdp.data - xdp.data_meta;
 	if (metalen)
@@ -533,7 +549,7 @@ static struct sk_buff *veth_xdp_rcv_skb(struct veth_priv *priv,
 	return NULL;
 }
 
-static int veth_xdp_rcv(struct veth_priv *priv, int budget, bool *xdp_xmit,
+static int veth_xdp_rcv(struct veth_rq *rq, int budget, bool *xdp_xmit,
 			bool *xdp_redir)
 {
 	int done = 0;
@@ -551,15 +567,15 @@ static int veth_xdp_rcv(struct veth_priv *priv, int budget, bool *xdp_xmit,
 			struct xdp_frame *frame;
 			struct sk_buff *skb;
 
-			frame = __ptr_ring_consume(&priv->xdp_tx_ring);
+			frame = __ptr_ring_consume(&rq->xdp_tx_ring);
 			if (!frame) {
 				curr_more = false;
 				break;
 			}
 
-			skb = veth_xdp_rcv_one(priv, frame, xdp_xmit, xdp_redir);
+			skb = veth_xdp_rcv_one(rq, frame, xdp_xmit, xdp_redir);
 			if (skb)
-				napi_gro_receive(&priv->xdp_napi, skb);
+				napi_gro_receive(&rq->xdp_napi, skb);
 
 			done++;
 		}
@@ -568,16 +584,16 @@ static int veth_xdp_rcv(struct veth_priv *priv, int budget, bool *xdp_xmit,
 		curr_more = true;
 		curr_budget = min(budget - done, budget >> 1);
 		for (i = 0; i < curr_budget; i++) {
-			struct sk_buff *skb = __ptr_ring_consume(&priv->xdp_ring);
+			struct sk_buff *skb = __ptr_ring_consume(&rq->xdp_ring);
 
 			if (!skb) {
 				curr_more = false;
 				break;
 			}
 
-			skb = veth_xdp_rcv_skb(priv, skb, xdp_xmit, xdp_redir);
+			skb = veth_xdp_rcv_skb(rq, skb, xdp_xmit, xdp_redir);
 			if (skb)
-				napi_gro_receive(&priv->xdp_napi, skb);
+				napi_gro_receive(&rq->xdp_napi, skb);
 
 			done++;
 		}
@@ -589,26 +605,26 @@ static int veth_xdp_rcv(struct veth_priv *priv, int budget, bool *xdp_xmit,
 
 static int veth_poll(struct napi_struct *napi, int budget)
 {
-	struct veth_priv *priv =
-		container_of(napi, struct veth_priv, xdp_napi);
+	struct veth_rq *rq =
+		container_of(napi, struct veth_rq, xdp_napi);
 	bool xdp_xmit = false;
 	bool xdp_redir = false;
 	int done;
 
-	done = veth_xdp_rcv(priv, budget, &xdp_xmit, &xdp_redir);
+	done = veth_xdp_rcv(rq, budget, &xdp_xmit, &xdp_redir);
 
 	if (done < budget && napi_complete_done(napi, done)) {
 		/* Write rx_notify_masked before reading ptr_ring */
-		smp_store_mb(priv->rx_notify_masked, false);
-		if (unlikely(!__ptr_ring_empty(&priv->xdp_tx_ring) ||
-			     !__ptr_ring_empty(&priv->xdp_ring))) {
-			priv->rx_notify_masked = true;
-			napi_schedule(&priv->xdp_napi);
+		smp_store_mb(rq->rx_notify_masked, false);
+		if (unlikely(!__ptr_ring_empty(&rq->xdp_tx_ring) ||
+			     !__ptr_ring_empty(&rq->xdp_ring))) {
+			rq->rx_notify_masked = true;
+			napi_schedule(&rq->xdp_napi);
 		}
 	}
 
 	if (xdp_xmit)
-		veth_xdp_flush(priv->dev);
+		veth_xdp_flush(rq->dev);
 	if (xdp_redir)
 		xdp_do_flush_map();
 
@@ -618,22 +634,36 @@ static int veth_poll(struct napi_struct *napi, int budget)
 static int veth_napi_add(struct net_device *dev)
 {
 	struct veth_priv *priv = netdev_priv(dev);
-	int err;
+	int err, i;
 
-	err = ptr_ring_init(&priv->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
-	if (err)
-		return err;
+	for (i = 0; i < dev->real_num_rx_queues; i++) {
+		struct veth_rq *rq = &priv->rq[i];
 
-	err = ptr_ring_init(&priv->xdp_tx_ring, VETH_RING_SIZE, GFP_KERNEL);
-	if (err)
-		goto err_xdp_tx_ring;
+		err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL);
+		if (err)
+			goto err_xdp_ring;
+
+		err = ptr_ring_init(&rq->xdp_tx_ring, VETH_RING_SIZE,
+				    GFP_KERNEL);
+		if (err)
+			goto err_xdp_tx_ring;
+	}
+
+	for (i = 0; i < dev->real_num_rx_queues; i++) {
+		struct veth_rq *rq = &priv->rq[i];
 
-	netif_napi_add(dev, &priv->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
-	napi_enable(&priv->xdp_napi);
+		netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT);
+		napi_enable(&rq->xdp_napi);
+	}
 
 	return 0;
 err_xdp_tx_ring:
-	ptr_ring_cleanup(&priv->xdp_ring, __skb_array_destroy_skb);
+	ptr_ring_cleanup(&priv->rq[i].xdp_ring, __skb_array_destroy_skb);
+err_xdp_ring:
+	for (i--; i >= 0; i--) {
+		ptr_ring_cleanup(&priv->rq[i].xdp_ring, __skb_array_destroy_skb);
+		ptr_ring_cleanup(&priv->rq[i].xdp_tx_ring, veth_xdp_free);
+	}
 
 	return err;
 }
@@ -641,37 +671,56 @@ static int veth_napi_add(struct net_device *dev)
 static void veth_napi_del(struct net_device *dev)
 {
 	struct veth_priv *priv = netdev_priv(dev);
+	int i;
 
-	napi_disable(&priv->xdp_napi);
-	netif_napi_del(&priv->xdp_napi);
-	ptr_ring_cleanup(&priv->xdp_ring, __skb_array_destroy_skb);
-	ptr_ring_cleanup(&priv->xdp_tx_ring, veth_xdp_free);
+	for (i = 0; i < dev->real_num_rx_queues; i++) {
+		struct veth_rq *rq = &priv->rq[i];
+
+		napi_disable(&rq->xdp_napi);
+		napi_hash_del(&rq->xdp_napi);
+	}
+	synchronize_net();
+
+	for (i = 0; i < dev->real_num_rx_queues; i++) {
+		struct veth_rq *rq = &priv->rq[i];
+
+		netif_napi_del(&rq->xdp_napi);
+		ptr_ring_cleanup(&rq->xdp_ring, __skb_array_destroy_skb);
+		ptr_ring_cleanup(&rq->xdp_tx_ring, veth_xdp_free);
+	}
 }
 
 static int veth_enable_xdp(struct net_device *dev)
 {
 	struct veth_priv *priv = netdev_priv(dev);
-	int err;
+	int err, i;
 
-	err = xdp_rxq_info_reg(&priv->xdp_rxq, dev, 0);
-	if (err < 0)
-		return err;
+	for (i = 0; i < dev->real_num_rx_queues; i++) {
+		struct veth_rq *rq = &priv->rq[i];
 
-	err = xdp_rxq_info_reg_mem_model(&priv->xdp_rxq,
-					 MEM_TYPE_PAGE_SHARED, NULL);
-	if (err < 0)
-		goto err;
+		err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i);
+		if (err < 0)
+			goto err_rxq_reg;
+
+		err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
+						 MEM_TYPE_PAGE_SHARED, NULL);
+		if (err < 0)
+			goto err_reg_mem;
 
-	/* Save original mem info as it can be overwritten */
-	priv->xdp_mem = priv->xdp_rxq.mem;
+		/* Save original mem info as it can be overwritten */
+		rq->xdp_mem = rq->xdp_rxq.mem;
+	}
 
 	err = veth_napi_add(dev);
 	if (err)
-		goto err;
+		goto err_rxq_reg;
 
 	return 0;
-err:
-	xdp_rxq_info_unreg(&priv->xdp_rxq);
+err_reg_mem:
+	xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
+err_rxq_reg:
+	for (i--; i >= 0; i--)
+		xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq);
 
 	return err;
 }
@@ -679,10 +728,15 @@ static int veth_enable_xdp(struct net_device *dev)
 static void veth_disable_xdp(struct net_device *dev)
 {
 	struct veth_priv *priv = netdev_priv(dev);
+	int i;
 
 	veth_napi_del(dev);
-	priv->xdp_rxq.mem = priv->xdp_mem;
-	xdp_rxq_info_unreg(&priv->xdp_rxq);
+	for (i = 0; i < dev->real_num_rx_queues; i++) {
+		struct veth_rq *rq = &priv->rq[i];
+
+		rq->xdp_rxq.mem = rq->xdp_mem;
+		xdp_rxq_info_unreg(&rq->xdp_rxq);
+	}
 }
 
 static int veth_open(struct net_device *dev)
@@ -694,7 +748,7 @@ static int veth_open(struct net_device *dev)
 	if (!peer)
 		return -ENOTCONN;
 
-	if (rtnl_dereference(priv->xdp_prog)) {
+	if (rtnl_dereference(priv->rq[0].xdp_prog)) {
 		err = veth_enable_xdp(dev);
 		if (err)
 			return err;
@@ -717,7 +771,7 @@ static int veth_close(struct net_device *dev)
 	if (peer)
 		netif_carrier_off(peer);
 
-	if (rtnl_dereference(priv->xdp_prog))
+	if (rtnl_dereference(priv->rq[0].xdp_prog))
 		veth_disable_xdp(dev);
 
 	return 0;
@@ -780,7 +834,7 @@ static netdev_features_t veth_fix_features(struct net_device *dev,
 	if (peer) {
 		struct veth_priv *peer_priv = netdev_priv(peer);
 
-		if (rtnl_dereference(peer_priv->xdp_prog))
+		if (rtnl_dereference(peer_priv->rq[0].xdp_prog))
 			features &= ~NETIF_F_GSO_SOFTWARE;
 	}
 
@@ -816,9 +870,9 @@ static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
 	struct veth_priv *priv = netdev_priv(dev);
 	struct bpf_prog *old_prog;
 	struct net_device *peer;
-	int err;
+	int err, i;
 
-	old_prog = rtnl_dereference(priv->xdp_prog);
+	old_prog = rtnl_dereference(priv->rq[0].xdp_prog);
 	peer = rtnl_dereference(priv->peer);
 
 	if (prog) {
@@ -826,6 +880,9 @@ static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
 			return -ENOTCONN;
 
 		if (!old_prog) {
+			if (dev->real_num_rx_queues < peer->real_num_tx_queues)
+				return -ENOSPC;
+
 			if (dev->flags & IFF_UP) {
 				err = veth_enable_xdp(dev);
 				if (err)
@@ -841,7 +898,8 @@ static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
 		}
 	}
 
-	rcu_assign_pointer(priv->xdp_prog, prog);
+	for (i = 0; i < dev->real_num_rx_queues; i++)
+		rcu_assign_pointer(priv->rq[i].xdp_prog, prog);
 
 	if (old_prog) {
 		bpf_prog_put(old_prog);
@@ -867,7 +925,7 @@ static u32 veth_xdp_query(struct net_device *dev)
 	struct veth_priv *priv = netdev_priv(dev);
 	const struct bpf_prog *xdp_prog;
 
-	xdp_prog = rtnl_dereference(priv->xdp_prog);
+	xdp_prog = rtnl_dereference(priv->rq[0].xdp_prog);
 	if (xdp_prog)
 		return xdp_prog->aux->id;
 
@@ -960,13 +1018,31 @@ static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
 	return 0;
 }
 
+static int veth_alloc_queues(struct net_device *dev)
+{
+	struct veth_priv *priv = netdev_priv(dev);
+
+	priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL);
+	if (!priv->rq)
+		return -ENOMEM;
+
+	return 0;
+}
+
+static void veth_free_queues(struct net_device *dev)
+{
+	struct veth_priv *priv = netdev_priv(dev);
+
+	kfree(priv->rq);
+}
+
 static struct rtnl_link_ops veth_link_ops;
 
 static int veth_newlink(struct net *src_net, struct net_device *dev,
 			struct nlattr *tb[], struct nlattr *data[],
 			struct netlink_ext_ack *extack)
 {
-	int err;
+	int err, i;
 	struct net_device *peer;
 	struct veth_priv *priv;
 	char ifname[IFNAMSIZ];
@@ -1019,6 +1095,12 @@ static int veth_newlink(struct net *src_net, struct net_device *dev,
 		return PTR_ERR(peer);
 	}
 
+	err = veth_alloc_queues(peer);
+	if (err) {
+		put_net(net);
+		goto err_peer_alloc_queues;
+	}
+
 	if (!ifmp || !tbp[IFLA_ADDRESS])
 		eth_hw_addr_random(peer);
 
@@ -1047,6 +1129,10 @@ static int veth_newlink(struct net *src_net, struct net_device *dev,
 	 * should be re-allocated
 	 */
 
+	err = veth_alloc_queues(dev);
+	if (err)
+		goto err_alloc_queues;
+
 	if (tb[IFLA_ADDRESS] == NULL)
 		eth_hw_addr_random(dev);
 
@@ -1066,22 +1152,28 @@ static int veth_newlink(struct net *src_net, struct net_device *dev,
 	 */
 
 	priv = netdev_priv(dev);
-	priv->dev = dev;
+	for (i = 0; i < dev->real_num_rx_queues; i++)
+		priv->rq[i].dev = dev;
 	rcu_assign_pointer(priv->peer, peer);
 
 	priv = netdev_priv(peer);
-	priv->dev = peer;
+	for (i = 0; i < peer->real_num_rx_queues; i++)
+		priv->rq[i].dev = peer;
 	rcu_assign_pointer(priv->peer, dev);
 
 	return 0;
 
 err_register_dev:
+	veth_free_queues(dev);
+err_alloc_queues:
 	/* nothing to do */
 err_configure_peer:
 	unregister_netdevice(peer);
 	return err;
 
 err_register_peer:
+	veth_free_queues(peer);
+err_peer_alloc_queues:
 	free_netdev(peer);
 	return err;
 }
-- 
2.14.3

[PATCH RFC v2 9/9] veth: Bulk skb xmit for XDP path

[Toshiaki Makita]

From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>

Aquire txq lock instead of rxq ptr_ring lock so we avoid per-packet
lock when skb->xmit_more is true. We ensure that rxqs are always not
less than txqs and txq to rxq is one to one mapping, so we can
completely remove rxq side lock.

Since we removed rxq side lock, this change does not increase the number
of locking even when bulk sending is not possible, e.g. non-GSO packets.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
---
 drivers/net/veth.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/drivers/net/veth.c b/drivers/net/veth.c
index 67debd3eafe6..376d70f983e5 100644
--- a/drivers/net/veth.c
+++ b/drivers/net/veth.c
@@ -138,7 +138,7 @@ static void __veth_xdp_flush(struct veth_rq *rq)
 
 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb)
 {
-	if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) {
+	if (unlikely(__ptr_ring_produce(&rq->xdp_ring, skb))) {
 		dev_kfree_skb_any(skb);
 		return NET_RX_DROP;
 	}
@@ -188,7 +188,7 @@ static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 		atomic64_inc(&priv->dropped);
 	}
 
-	if (rcv_xdp)
+	if (rcv_xdp && !skb->xmit_more)
 		__veth_xdp_flush(rq);
 
 	rcu_read_unlock();
@@ -829,15 +829,21 @@ static netdev_features_t veth_fix_features(struct net_device *dev,
 {
 	struct veth_priv *priv = netdev_priv(dev);
 	struct net_device *peer;
+	bool xdp = false;
 
 	peer = rtnl_dereference(priv->peer);
 	if (peer) {
 		struct veth_priv *peer_priv = netdev_priv(peer);
 
 		if (rtnl_dereference(peer_priv->rq[0].xdp_prog))
-			features &= ~NETIF_F_GSO_SOFTWARE;
+			xdp = true;
 	}
 
+	if (xdp)
+		features &= ~(NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX);
+	else
+		features |= NETIF_F_LLTX;
+
 	return features;
 }
 
-- 
2.14.3

Re: [PATCH RFC v2 2/9] veth: Add driver XDP

[Toshiaki Makita]

On 2018/06/11 1:02, Toshiaki Makita wrote:
> From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
> 
> This is basic implementation of veth driver XDP.
> 
> Incoming packets are sent from the peer veth device in the form of skb,
> so this is generally doing the same thing as generic XDP.
> 
> This itself is not so useful, but a starting point to implement other
> useful veth XDP features like TX and REDIRECT.
> 
> This introduces NAPI when XDP is enabled, because XDP is now heavily
> relies on NAPI context. Use ptr_ring to emulate NIC ring. Tx function
> enqueues packets to the ring and peer NAPI handler drains the ring.
> 
> Currently only one ring is allocated for each veth device, so it does
> not scale on multiqueue env. This can be resolved by allocating rings
> on the per-queue basis later.
> 
> Note that NAPI is not used but netif_rx is used when XDP is not loaded,
> so this does not change the default behaviour.
> 
> v2:
> - Squashed with the patch adding NAPI.
> - Implement adjust_tail.
> - Don't acquire consumer lock because it is guarded by NAPI.
> - Make poll_controller noop since it is unnecessary.
> - Register rxq_info on enabling XDP rather than on opening the device.
> 
> Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
> ---
...
> @@ -203,6 +480,9 @@ static int veth_close(struct net_device *dev)
>  	if (peer)
>  		netif_carrier_off(peer);
>  
> +	if (rtnl_dereference(priv->xdp_prog))
> +		veth_disable_xdp(dev);

Found that it is not safe to disable napi at this point, because packets
still can come from the peer and the ring may be filled after cleanup.
I'll fix the problem in v3, but wait for feedback some more.

-- 
Toshiaki Makita