Skip to contentSkip to conversation

xen-netfront: add basic XDP support

PATCH Superseded Denis Kirjanov 7 participants 15 messages Mar 2, 2020 – Mar 6, 2020

Subsystems (4)

v2 of 14 Superseded Latest is v14 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 xen-netfront: add basic XDP support

the patch adds a basic xdo logic to the netfront driver

XDP redirect is not supported yet

v2:
- avoid data copying while passing to XDP
- tell xen-natback that we need the headroom space

---
 drivers/net/xen-netback/common.h |   1 +
 drivers/net/xen-netback/rx.c     |   9 ++-
 drivers/net/xen-netback/xenbus.c |  21 ++++++
 drivers/net/xen-netfront.c       | 157 +++++++++++++++++++++++++++++++++++++++
 4 files changed, 186 insertions(+), 2 deletions(-)
drivers/net/xen-netback/common.h +1 -0
index 05847eb..0750c6f 100644
@@ -280,6 +280,7 @@ struct xenvif {
 	u8 ip_csum:1;
 	u8 ipv6_csum:1;
 	u8 multicast_control:1;
+	u8 xdp_enabled:1;
 
 	/* Is this interface disabled? True when backend discovers
 	 * frontend is rogue.
drivers/net/xen-netback/rx.c +7 -2
index ef58870..a110a59 100644
@@ -33,6 +33,11 @@
 #include <xen/xen.h>
 #include <xen/events.h>
 
+static inline int xenvif_rx_xdp_offset(struct xenvif *vif)
+{
+	return (vif->xdp_enabled ? XDP_PACKET_HEADROOM : 0);
+}
+
 static bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue)
 {
 	RING_IDX prod, cons;
@@ -356,7 +361,7 @@ static void xenvif_rx_data_slot(struct xenvif_queue *queue,
 				struct xen_netif_rx_request *req,
 				struct xen_netif_rx_response *rsp)
 {
-	unsigned int offset = 0;
+	unsigned int offset = xenvif_rx_xdp_offset(queue->vif);
 	unsigned int flags;
 
 	do {
@@ -389,7 +394,7 @@ static void xenvif_rx_data_slot(struct xenvif_queue *queue,
 			flags |= XEN_NETRXF_extra_info;
 	}
 
-	rsp->offset = 0;
+	rsp->offset = xenvif_rx_xdp_offset(queue->vif);
 	rsp->flags = flags;
 	rsp->id = req->id;
 	rsp->status = (s16)offset;
drivers/net/xen-netback/xenbus.c +21 -0
index 286054b..81a6023 100644
@@ -393,6 +393,20 @@ static void set_backend_state(struct backend_info *be,
 	}
 }
 
+static void read_xenbus_fronetend_xdp(struct backend_info *be,
+				      struct xenbus_device *dev)
+{
+	struct xenvif *vif = be->vif;
+	unsigned int val;
+	int err;
+
+	err = xenbus_scanf(XBT_NIL, dev->otherend,
+			   "feature-xdp", "%u", &val);
+	if (err < 0)
+		return;
+	vif->xdp_enabled = val;
+}
+
 /**
  * Callback received when the frontend's state changes.
  */
@@ -417,6 +431,11 @@ static void frontend_changed(struct xenbus_device *dev,
 		set_backend_state(be, XenbusStateConnected);
 		break;
 
+	case XenbusStateReconfiguring:
+		read_xenbus_fronetend_xdp(be, dev);
+		xenbus_switch_state(dev, XenbusStateReconfigured);
+		break;
+
 	case XenbusStateClosing:
 		set_backend_state(be, XenbusStateClosing);
 		break;
@@ -935,6 +954,8 @@ static int read_xenbus_vif_flags(struct backend_info *be)
 
 	vif->gso_mask = 0;
 
+	vif->xdp_enabled = 0;
+
 	if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv4", 0))
 		vif->gso_mask |= GSO_BIT(TCPV4);
 
drivers/net/xen-netfront.c +157 -0

199 lines not on this page · raw message

Signature · 1 line
1.8.3.1
maintainer
#1 lore raw
Quoted 7 lines · Denis Kirjanov

the patch adds a basic xdo logic to the netfront driver

XDP redirect is not supported yet

v2:
- avoid data copying while passing to XDP
- tell xen-natback that we need the headroom space

Please add the patch history below the "---" delimiter

Quoted 8 lines
---
  drivers/net/xen-netback/common.h |   1 +
  drivers/net/xen-netback/rx.c     |   9 ++-
  drivers/net/xen-netback/xenbus.c |  21 ++++++
  drivers/net/xen-netfront.c       | 157 +++++++++++++++++++++++++++++++++++++++
  4 files changed, 186 insertions(+), 2 deletions(-)

You are modifying xen-netback sources. Please Cc the maintainers.

Quoted 52 lines
drivers/net/xen-netback/common.h +1 -0
index 05847eb..0750c6f 100644
@@ -280,6 +280,7 @@ struct xenvif {
  	u8 ip_csum:1;
  	u8 ipv6_csum:1;
  	u8 multicast_control:1;
+	u8 xdp_enabled:1;
 
  	/* Is this interface disabled? True when backend discovers
  	 * frontend is rogue.
drivers/net/xen-netback/rx.c +7 -2
index ef58870..a110a59 100644
@@ -33,6 +33,11 @@
  #include <xen/xen.h>
  #include <xen/events.h>
 
+static inline int xenvif_rx_xdp_offset(struct xenvif *vif)
+{
+	return (vif->xdp_enabled ? XDP_PACKET_HEADROOM : 0);
+}
+
  static bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue)
  {
  	RING_IDX prod, cons;
@@ -356,7 +361,7 @@ static void xenvif_rx_data_slot(struct xenvif_queue *queue,
  				struct xen_netif_rx_request *req,
  				struct xen_netif_rx_response *rsp)
  {
-	unsigned int offset = 0;
+	unsigned int offset = xenvif_rx_xdp_offset(queue->vif);
  	unsigned int flags;
 
  	do {
@@ -389,7 +394,7 @@ static void xenvif_rx_data_slot(struct xenvif_queue *queue,
  			flags |= XEN_NETRXF_extra_info;
  	}
 
-	rsp->offset = 0;
+	rsp->offset = xenvif_rx_xdp_offset(queue->vif);
  	rsp->flags = flags;
  	rsp->id = req->id;
  	rsp->status = (s16)offset;
drivers/net/xen-netback/xenbus.c +0 -0
index 286054b..81a6023 100644

drivers/net/xen-netback/xenbus.c static void set_backend_state(struct backend_info *be, · line 397

Quoted 5 lines
drivers/net/xen-netback/xenbus.c +2 -0
@@ -393,6 +393,20 @@ static void set_backend_state(struct backend_info *be,
  	}
  }
 
+static void read_xenbus_fronetend_xdp(struct backend_info *be,
+				      struct xenbus_device *dev)

Typo: s/fronetend/frontend/

Quoted 16 lines
drivers/net/xen-netback/xenbus.c +12 -0
+{
+	struct xenvif *vif = be->vif;
+	unsigned int val;
+	int err;
+
+	err = xenbus_scanf(XBT_NIL, dev->otherend,
+			   "feature-xdp", "%u", &val);
+	if (err < 0)
+		return;
+	vif->xdp_enabled = val;
+}
+
  /**
   * Callback received when the frontend's state changes.
   */

drivers/net/xen-netback/xenbus.c static void frontend_changed(struct xenbus_device *dev, · line 438

Quoted 8 lines
drivers/net/xen-netback/xenbus.c +5 -0
@@ -417,6 +431,11 @@ static void frontend_changed(struct xenbus_device *dev,
  		set_backend_state(be, XenbusStateConnected);
  		break;
 
+	case XenbusStateReconfiguring:
+		read_xenbus_fronetend_xdp(be, dev);
+		xenbus_switch_state(dev, XenbusStateReconfigured);
+		break;
+

Where is the reaction to the backend being set to "Reconfigured"?

Quoted 128 lines
drivers/net/xen-netback/xenbus.c +2 -0
  	case XenbusStateClosing:
  		set_backend_state(be, XenbusStateClosing);
  		break;
@@ -935,6 +954,8 @@ static int read_xenbus_vif_flags(struct backend_info *be)
 
  	vif->gso_mask = 0;
 
+	vif->xdp_enabled = 0;
+
  	if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv4", 0))
  		vif->gso_mask |= GSO_BIT(TCPV4);
 
drivers/net/xen-netfront.c +73 -0
index 482c6c8..db8a280 100644
@@ -44,6 +44,8 @@
  #include <linux/mm.h>
  #include <linux/slab.h>
  #include <net/ip.h>
+#include <linux/bpf.h>
+#include <linux/bpf_trace.h>
 
  #include <xen/xen.h>
  #include <xen/xenbus.h>
@@ -102,6 +104,8 @@ struct netfront_queue {
  	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
  	struct netfront_info *info;
 
+	struct bpf_prog __rcu *xdp_prog;
+
  	struct napi_struct napi;
 
  	/* Split event channels support, tx_* == rx_* when using
@@ -778,6 +782,40 @@ static int xennet_get_extras(struct netfront_queue *queue,
  	return err;
  }
 
+u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
+		   struct xen_netif_rx_response *rx, struct bpf_prog *prog,
+		   struct xdp_buff *xdp)
+{
+	u32 len = rx->status;
+	u32 act = XDP_PASS;
+
+	xdp->data_hard_start = page_address(pdata);
+	xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM;
+	xdp_set_data_meta_invalid(xdp);
+	xdp->data_end = xdp->data + len;
+	xdp->handle = 0;
+
+	act = bpf_prog_run_xdp(prog, xdp);
+	switch (act) {
+	case XDP_PASS:
+	case XDP_TX:
+	case XDP_DROP:
+		break;
+
+	case XDP_ABORTED:
+		trace_xdp_exception(queue->info->netdev, prog, act);
+		break;
+
+	default:
+		bpf_warn_invalid_xdp_action(act);
+	}
+
+	if (act != XDP_PASS && act != XDP_TX)
+		xdp->data_hard_start = NULL;
+
+	return act;
+}
+
  static int xennet_get_responses(struct netfront_queue *queue,
  				struct netfront_rx_info *rinfo, RING_IDX rp,
  				struct sk_buff_head *list)
@@ -792,6 +830,9 @@ static int xennet_get_responses(struct netfront_queue *queue,
  	int slots = 1;
  	int err = 0;
  	unsigned long ret;
+	struct bpf_prog *xdp_prog;
+	struct xdp_buff xdp;
+	u32 verdict;
 
  	if (rx->flags & XEN_NETRXF_extra_info) {
  		err = xennet_get_extras(queue, extras, rp);
@@ -827,6 +868,22 @@ static int xennet_get_responses(struct netfront_queue *queue,
 
  		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
 
+		rcu_read_lock();
+		xdp_prog = rcu_dereference(queue->xdp_prog);
+		if (xdp_prog) {
+			/* currently only a single page contains data */
+			WARN_ON_ONCE(skb_shinfo(skb)->nr_frags != 1);
+			verdict = xennet_run_xdp(queue,
+				       skb_frag_page(&skb_shinfo(skb)->frags[0]),
+				       rx, xdp_prog, &xdp);
+
+			if (verdict != XDP_PASS && verdict != XDP_TX) {
+				err = -EINVAL;
+				goto next;
+			}
+
+		}
+		rcu_read_unlock();
  		__skb_queue_tail(list, skb);
 
  next:
@@ -1261,6 +1318,105 @@ static void xennet_poll_controller(struct net_device *dev)
  }
  #endif
 
+#define NETBACK_XDP_HEADROOM_DISABLE	0
+#define NETBACK_XDP_HEADROOM_ENABLE	1
+
+static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
+{
+	struct xenbus_transaction xbt;
+	const char *message;
+	int err;
+
+again:
+	err = xenbus_transaction_start(&xbt);
+	if (err) {
+		xenbus_dev_fatal(np->xbdev, err, "starting transaction");
+		goto out;
+	}
+

drivers/net/xen-netfront.c static void xennet_poll_controller(struct net_device *dev) · line 1346

Quoted 10 lines
drivers/net/xen-netfront.c +10 -0
+	err = xenbus_printf(xbt, np->xbdev->nodename, "feature-xdp", "%d", xdp);
+	if (err) {
+		message = "writing feature-xdp";
+		goto abort_transaction;
+	}
+
+	err = xenbus_transaction_end(xbt, 0);
+	if (err) {
+		if (err == -EAGAIN)
+			goto again;

Writing a single node to Xenstore doesn't need a transaction.

Quoted 28 lines
drivers/net/xen-netfront.c +28 -0
+	}
+
+	return 0;
+
+abort_transaction:
+	xenbus_dev_fatal(np->xbdev, err, "%s", message);
+	xenbus_transaction_end(xbt, 1);
+out:
+	return err;
+}
+
+static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
+			struct netlink_ext_ack *extack)
+{
+	struct netfront_info *np = netdev_priv(dev);
+	struct bpf_prog *old_prog;
+	unsigned int i, err;
+
+	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
+	if (!old_prog && !prog)
+		return 0;
+
+	if (prog)
+		bpf_prog_add(prog, dev->real_num_tx_queues);
+
+	for (i = 0; i < dev->real_num_tx_queues; ++i)
+		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
+

drivers/net/xen-netfront.c static void xennet_poll_controller(struct net_device *dev) · line 1384

Quoted 10 lines
drivers/net/xen-netfront.c +10 -0
+	if (old_prog)
+		for (i = 0; i < dev->real_num_tx_queues; ++i)
+			bpf_prog_put(old_prog);
+
+	err = talk_to_netback_xdp(np, old_prog ? NETBACK_XDP_HEADROOM_DISABLE:
+				  NETBACK_XDP_HEADROOM_ENABLE);
+	if (err)
+		return err;
+
+	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);

What is happening in case the backend doesn't support XDP?

Is it really a good idea to communicate xdp_set via a frontend state change? This will be rather slow. OTOH I have no idea how often this might happen.

Juergen

maintainer
#2 lore raw

From: Jürgen Groß <[email protected]>
Date: Mon, 2 Mar 2020 16:29:41 +0100

Quoted 8 lines · Denis Kirjanov
Quoted 5 lines

the patch adds a basic xdo logic to the netfront driver XDP redirect is not supported yet
v2:
- avoid data copying while passing to XDP
- tell xen-natback that we need the headroom space

Please add the patch history below the "---" delimiter

Incorrect, I prefer to have them in the commit message and recorded in the GIT history.

Please don't give out this advice for networking changes.

Thank you.

#3 lore raw
In reply to Jürgen Groß
Quoted 294 lines · Jürgen Groß
Quoted 7 lines · Denis Kirjanov

the patch adds a basic xdo logic to the netfront driver

XDP redirect is not supported yet

v2:
- avoid data copying while passing to XDP
- tell xen-natback that we need the headroom space

Please add the patch history below the "---" delimiter

Quoted 9 lines
---
  drivers/net/xen-netback/common.h |   1 +
  drivers/net/xen-netback/rx.c     |   9 ++-
  drivers/net/xen-netback/xenbus.c |  21 ++++++
  drivers/net/xen-netfront.c       | 157
+++++++++++++++++++++++++++++++++++++++
  4 files changed, 186 insertions(+), 2 deletions(-)

You are modifying xen-netback sources. Please Cc the maintainers.

Quoted 56 lines
drivers/net/xen-netback/common.h +1 -0
index 05847eb..0750c6f 100644
@@ -280,6 +280,7 @@ struct xenvif {
  	u8 ip_csum:1;
  	u8 ipv6_csum:1;
  	u8 multicast_control:1;
+	u8 xdp_enabled:1;
 
  	/* Is this interface disabled? True when backend discovers
  	 * frontend is rogue.
drivers/net/xen-netback/rx.c +6 -1
index ef58870..a110a59 100644
@@ -33,6 +33,11 @@
  #include <xen/xen.h>
  #include <xen/events.h>
 
+static inline int xenvif_rx_xdp_offset(struct xenvif *vif)
+{
+	return (vif->xdp_enabled ? XDP_PACKET_HEADROOM : 0);
+}
+
  static bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue)
  {
  	RING_IDX prod, cons;
@@ -356,7 +361,7 @@ static void xenvif_rx_data_slot(struct xenvif_queue
 *queue,
  				struct xen_netif_rx_request *req,
  				struct xen_netif_rx_response *rsp)
  {
-	unsigned int offset = 0;
+	unsigned int offset = xenvif_rx_xdp_offset(queue->vif);
  	unsigned int flags;
 
  	do {
drivers/net/xen-netback/rx.c +1 -1
@@ -389,7 +394,7 @@ static void xenvif_rx_data_slot(struct xenvif_queue
 *queue,
  			flags |= XEN_NETRXF_extra_info;
  	}
 
-	rsp->offset = 0;
+	rsp->offset = xenvif_rx_xdp_offset(queue->vif);
  	rsp->flags = flags;
  	rsp->id = req->id;
  	rsp->status = (s16)offset;
drivers/net/xen-netback/xenbus.c +0 -0
index 286054b..81a6023 100644

drivers/net/xen-netback/xenbus.c static void set_backend_state(struct backend_info · line 398

Quoted 6 lines
drivers/net/xen-netback/xenbus.c +2 -0
@@ -393,6 +393,20 @@ static void set_backend_state(struct backend_info
 *be,
  	}
  }
 
+static void read_xenbus_fronetend_xdp(struct backend_info *be,
+				      struct xenbus_device *dev)

Typo: s/fronetend/frontend/

Quoted 16 lines
drivers/net/xen-netback/xenbus.c +12 -0
+{
+	struct xenvif *vif = be->vif;
+	unsigned int val;
+	int err;
+
+	err = xenbus_scanf(XBT_NIL, dev->otherend,
+			   "feature-xdp", "%u", &val);
+	if (err < 0)
+		return;
+	vif->xdp_enabled = val;
+}
+
  /**
   * Callback received when the frontend's state changes.
   */

drivers/net/xen-netback/xenbus.c static void frontend_changed(struct xenbus_device · line 439

Quoted 9 lines
drivers/net/xen-netback/xenbus.c +5 -0
@@ -417,6 +431,11 @@ static void frontend_changed(struct xenbus_device
 *dev,
  		set_backend_state(be, XenbusStateConnected);
  		break;
 
+	case XenbusStateReconfiguring:
+		read_xenbus_fronetend_xdp(be, dev);
+		xenbus_switch_state(dev, XenbusStateReconfigured);
+		break;
+

Where is the reaction to the backend being set to "Reconfigured"?

Quoted 134 lines
drivers/net/xen-netback/xenbus.c +0 -0
  	case XenbusStateClosing:
  		set_backend_state(be, XenbusStateClosing);

break;

drivers/net/xen-netback/xenbus.c +2 -0
@@ -935,6 +954,8 @@ static int read_xenbus_vif_flags(struct backend_info
 *be)
 
  	vif->gso_mask = 0;
 
+	vif->xdp_enabled = 0;
+
  	if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv4", 0))
  		vif->gso_mask |= GSO_BIT(TCPV4);
drivers/net/xen-netfront.c +38 -0
index 482c6c8..db8a280 100644
@@ -44,6 +44,8 @@
  #include <linux/mm.h>
  #include <linux/slab.h>
  #include <net/ip.h>
+#include <linux/bpf.h>
+#include <linux/bpf_trace.h>
 
  #include <xen/xen.h>
  #include <xen/xenbus.h>
@@ -102,6 +104,8 @@ struct netfront_queue {
  	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
  	struct netfront_info *info;
 
+	struct bpf_prog __rcu *xdp_prog;
+
  	struct napi_struct napi;
 
  	/* Split event channels support, tx_* == rx_* when using
@@ -778,6 +782,40 @@ static int xennet_get_extras(struct netfront_queue
 *queue,
  	return err;
  }
 
+u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
+		   struct xen_netif_rx_response *rx, struct bpf_prog *prog,
+		   struct xdp_buff *xdp)
+{
+	u32 len = rx->status;
+	u32 act = XDP_PASS;
+
+	xdp->data_hard_start = page_address(pdata);
+	xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM;
+	xdp_set_data_meta_invalid(xdp);
+	xdp->data_end = xdp->data + len;
+	xdp->handle = 0;
+
+	act = bpf_prog_run_xdp(prog, xdp);
+	switch (act) {
+	case XDP_PASS:
+	case XDP_TX:
+	case XDP_DROP:
+		break;
+
+	case XDP_ABORTED:
+		trace_xdp_exception(queue->info->netdev, prog, act);
+		break;
+
+	default:
+		bpf_warn_invalid_xdp_action(act);
+	}
+
+	if (act != XDP_PASS && act != XDP_TX)
+		xdp->data_hard_start = NULL;
+
+	return act;
+}
+
  static int xennet_get_responses(struct netfront_queue *queue,
  				struct netfront_rx_info *rinfo, RING_IDX rp,

struct sk_buff_head *list)

drivers/net/xen-netfront.c +3 -0
@@ -792,6 +830,9 @@ static int xennet_get_responses(struct netfront_queue
 *queue,
  	int slots = 1;
  	int err = 0;
  	unsigned long ret;
+	struct bpf_prog *xdp_prog;
+	struct xdp_buff xdp;
+	u32 verdict;
 
  	if (rx->flags & XEN_NETRXF_extra_info) {
  		err = xennet_get_extras(queue, extras, rp);
drivers/net/xen-netfront.c +16 -0
@@ -827,6 +868,22 @@ static int xennet_get_responses(struct netfront_queue
 *queue,
 
  		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
 
+		rcu_read_lock();
+		xdp_prog = rcu_dereference(queue->xdp_prog);
+		if (xdp_prog) {
+			/* currently only a single page contains data */
+			WARN_ON_ONCE(skb_shinfo(skb)->nr_frags != 1);
+			verdict = xennet_run_xdp(queue,
+				       skb_frag_page(&skb_shinfo(skb)->frags[0]),
+				       rx, xdp_prog, &xdp);
+
+			if (verdict != XDP_PASS && verdict != XDP_TX) {
+				err = -EINVAL;
+				goto next;
+			}
+
+		}
+		rcu_read_unlock();
  		__skb_queue_tail(list, skb);
 

next:

drivers/net/xen-netfront.c +17 -0
@@ -1261,6 +1318,105 @@ static void xennet_poll_controller(struct
 net_device *dev)
  }
  #endif
 
+#define NETBACK_XDP_HEADROOM_DISABLE	0
+#define NETBACK_XDP_HEADROOM_ENABLE	1
+
+static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
+{
+	struct xenbus_transaction xbt;
+	const char *message;
+	int err;
+
+again:
+	err = xenbus_transaction_start(&xbt);
+	if (err) {
+		xenbus_dev_fatal(np->xbdev, err, "starting transaction");
+		goto out;
+	}
+
+	err = xenbus_printf(xbt, np->xbdev->nodename, "feature-xdp", "%d",

drivers/net/xen-netfront.c static void xennet_poll_controller(struct · line 1348

Quoted 10 lines
drivers/net/xen-netfront.c +9 -0
 xdp);
+	if (err) {
+		message = "writing feature-xdp";
+		goto abort_transaction;
+	}
+
+	err = xenbus_transaction_end(xbt, 0);
+	if (err) {
+		if (err == -EAGAIN)
+			goto again;

Writing a single node to Xenstore doesn't need a transaction.

Quoted 32 lines
drivers/net/xen-netfront.c +32 -0
+	}
+
+	return 0;
+
+abort_transaction:
+	xenbus_dev_fatal(np->xbdev, err, "%s", message);
+	xenbus_transaction_end(xbt, 1);
+out:
+	return err;
+}
+
+static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
+			struct netlink_ext_ack *extack)
+{
+	struct netfront_info *np = netdev_priv(dev);
+	struct bpf_prog *old_prog;
+	unsigned int i, err;
+
+	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
+	if (!old_prog && !prog)
+		return 0;
+
+	if (prog)
+		bpf_prog_add(prog, dev->real_num_tx_queues);
+
+	for (i = 0; i < dev->real_num_tx_queues; ++i)
+		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
+
+	if (old_prog)
+		for (i = 0; i < dev->real_num_tx_queues; ++i)
+			bpf_prog_put(old_prog);
+
Quoted 8 lines

drivers/net/xen-netfront.c static void xennet_poll_controller(struct net_device *dev) · line 1384

Quoted 6 lines
drivers/net/xen-netfront.c +6 -0
+	err = talk_to_netback_xdp(np, old_prog ? NETBACK_XDP_HEADROOM_DISABLE:
+				  NETBACK_XDP_HEADROOM_ENABLE);
+	if (err)
+		return err;
+
+	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);

What is happening in case the backend doesn't support XDP?

Here we just ask xen-backend to make a headroom, that's it. It's better to send xen-backend changes in a separate patch.

Quoted 4 lines

Is it really a good idea to communicate xdp_set via a frontend state change? This will be rather slow. OTOH I have no idea how often this might happen.

I don't think that it's going to switch often and more likely it's a one shot action.

Quoted 4 lines

Juergen

maintainer
#4 lore raw
Quoted 24 lines · Denis Kirjanov
Quoted 23 lines · Jürgen Groß
Quoted 7 lines · Denis Kirjanov

the patch adds a basic xdo logic to the netfront driver

XDP redirect is not supported yet

v2:
- avoid data copying while passing to XDP
- tell xen-natback that we need the headroom space

Please add the patch history below the "---" delimiter

Quoted 9 lines
---
   drivers/net/xen-netback/common.h |   1 +
   drivers/net/xen-netback/rx.c     |   9 ++-
   drivers/net/xen-netback/xenbus.c |  21 ++++++
   drivers/net/xen-netfront.c       | 157
+++++++++++++++++++++++++++++++++++++++
   4 files changed, 186 insertions(+), 2 deletions(-)

You are modifying xen-netback sources. Please Cc the maintainers.

...

Quoted 41 lines
Quoted 33 lines
Quoted 33 lines
drivers/net/xen-netfront.c +32 -0
+	}
+
+	return 0;
+
+abort_transaction:
+	xenbus_dev_fatal(np->xbdev, err, "%s", message);
+	xenbus_transaction_end(xbt, 1);
+out:
+	return err;
+}
+
+static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
+			struct netlink_ext_ack *extack)
+{
+	struct netfront_info *np = netdev_priv(dev);
+	struct bpf_prog *old_prog;
+	unsigned int i, err;
+
+	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
+	if (!old_prog && !prog)
+		return 0;
+
+	if (prog)
+		bpf_prog_add(prog, dev->real_num_tx_queues);
+
+	for (i = 0; i < dev->real_num_tx_queues; ++i)
+		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
+
+	if (old_prog)
+		for (i = 0; i < dev->real_num_tx_queues; ++i)
+			bpf_prog_put(old_prog);
+
Quoted 8 lines

drivers/net/xen-netfront.c static void xennet_poll_controller(struct net_device *dev) · line 1384

Quoted 6 lines
drivers/net/xen-netfront.c +6 -0
+	err = talk_to_netback_xdp(np, old_prog ? NETBACK_XDP_HEADROOM_DISABLE:
+				  NETBACK_XDP_HEADROOM_ENABLE);
+	if (err)
+		return err;
+
+	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);

What is happening in case the backend doesn't support XDP?

Quoted 2 lines

Here we just ask xen-backend to make a headroom, that's it. It's better to send xen-backend changes in a separate patch.

Okay, but what do you do if the backend doesn't support XDP (e.g. in case its an older kernel)? How do you know it is supporting XDP?

Quoted 8 lines
Quoted 5 lines

Is it really a good idea to communicate xdp_set via a frontend state change? This will be rather slow. OTOH I have no idea how often this might happen.

I don't think that it's going to switch often and more likely it's a one shot action.

What do you do in case of a live migration? You need to tell the backend about XDP again.

Juergen

#5 lore raw
Quoted 71 lines · Jürgen Groß
Quoted 24 lines · Denis Kirjanov
Quoted 23 lines · Jürgen Groß
Quoted 7 lines · Denis Kirjanov

the patch adds a basic xdo logic to the netfront driver

XDP redirect is not supported yet

v2:
- avoid data copying while passing to XDP
- tell xen-natback that we need the headroom space

Please add the patch history below the "---" delimiter

Quoted 9 lines
---
   drivers/net/xen-netback/common.h |   1 +
   drivers/net/xen-netback/rx.c     |   9 ++-
   drivers/net/xen-netback/xenbus.c |  21 ++++++
   drivers/net/xen-netfront.c       | 157
+++++++++++++++++++++++++++++++++++++++
   4 files changed, 186 insertions(+), 2 deletions(-)

You are modifying xen-netback sources. Please Cc the maintainers.

...

Quoted 43 lines
Quoted 43 lines
Quoted 36 lines
drivers/net/xen-netfront.c +12 -0
+	}
+
+	return 0;
+
+abort_transaction:
+	xenbus_dev_fatal(np->xbdev, err, "%s", message);
+	xenbus_transaction_end(xbt, 1);
+out:
+	return err;
+}
+
+static int xennet_xdp_set(struct net_device *dev, struct bpf_prog

*prog,

drivers/net/xen-netfront.c +21 -0
+			struct netlink_ext_ack *extack)
+{
+	struct netfront_info *np = netdev_priv(dev);
+	struct bpf_prog *old_prog;
+	unsigned int i, err;
+
+	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
+	if (!old_prog && !prog)
+		return 0;
+
+	if (prog)
+		bpf_prog_add(prog, dev->real_num_tx_queues);
+
+	for (i = 0; i < dev->real_num_tx_queues; ++i)
+		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
+
+	if (old_prog)
+		for (i = 0; i < dev->real_num_tx_queues; ++i)
+			bpf_prog_put(old_prog);
+
+	err = talk_to_netback_xdp(np, old_prog ?

NETBACK_XDP_HEADROOM_DISABLE:

drivers/net/xen-netfront.c static void xennet_poll_controller(struct net_device *dev) · line 1384

Quoted 5 lines
drivers/net/xen-netfront.c +5 -0
+				  NETBACK_XDP_HEADROOM_ENABLE);
+	if (err)
+		return err;
+
+	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);

What is happening in case the backend doesn't support XDP?

Quoted 5 lines
Quoted 2 lines

Here we just ask xen-backend to make a headroom, that's it. It's better to send xen-backend changes in a separate patch.

Okay, but what do you do if the backend doesn't support XDP (e.g. in case its an older kernel)? How do you know it is supporting XDP?

We can check a xenbus reply to xenbus state change.

Quoted 7 lines
Quoted 7 lines
Quoted 6 lines

Is it really a good idea to communicate xdp_set via a frontend state change? This will be rather slow. OTOH I have no idea how often this might happen.

Quoted 6 lines
Quoted 3 lines

I don't think that it's going to switch often and more likely it's a one shot
action.

What do you do in case of a live migration? You need to tell the backend about XDP again.

Yep I haven't thought about that. Thanks for pointing out.

Quoted 4 lines

Juergen

maintainer
#6 lore raw
Quoted 72 lines · Denis Kirjanov
Quoted 71 lines · Jürgen Groß
Quoted 24 lines · Denis Kirjanov
Quoted 23 lines · Jürgen Groß
Quoted 7 lines · Denis Kirjanov

the patch adds a basic xdo logic to the netfront driver

XDP redirect is not supported yet

v2:
- avoid data copying while passing to XDP
- tell xen-natback that we need the headroom space

Please add the patch history below the "---" delimiter

Quoted 9 lines
---
    drivers/net/xen-netback/common.h |   1 +
    drivers/net/xen-netback/rx.c     |   9 ++-
    drivers/net/xen-netback/xenbus.c |  21 ++++++
    drivers/net/xen-netfront.c       | 157
+++++++++++++++++++++++++++++++++++++++
    4 files changed, 186 insertions(+), 2 deletions(-)

You are modifying xen-netback sources. Please Cc the maintainers.

...

Quoted 43 lines
Quoted 43 lines
Quoted 36 lines
drivers/net/xen-netfront.c +12 -0
+	}
+
+	return 0;
+
+abort_transaction:
+	xenbus_dev_fatal(np->xbdev, err, "%s", message);
+	xenbus_transaction_end(xbt, 1);
+out:
+	return err;
+}
+
+static int xennet_xdp_set(struct net_device *dev, struct bpf_prog

*prog,

drivers/net/xen-netfront.c +21 -0
+			struct netlink_ext_ack *extack)
+{
+	struct netfront_info *np = netdev_priv(dev);
+	struct bpf_prog *old_prog;
+	unsigned int i, err;
+
+	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
+	if (!old_prog && !prog)
+		return 0;
+
+	if (prog)
+		bpf_prog_add(prog, dev->real_num_tx_queues);
+
+	for (i = 0; i < dev->real_num_tx_queues; ++i)
+		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
+
+	if (old_prog)
+		for (i = 0; i < dev->real_num_tx_queues; ++i)
+			bpf_prog_put(old_prog);
+
+	err = talk_to_netback_xdp(np, old_prog ?

NETBACK_XDP_HEADROOM_DISABLE:

drivers/net/xen-netfront.c static void xennet_poll_controller(struct net_device *dev) · line 1384

Quoted 5 lines
drivers/net/xen-netfront.c +5 -0
+				  NETBACK_XDP_HEADROOM_ENABLE);
+	if (err)
+		return err;
+
+	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);

What is happening in case the backend doesn't support XDP?

Quoted 6 lines
Quoted 5 lines
Quoted 2 lines

Here we just ask xen-backend to make a headroom, that's it. It's better to send xen-backend changes in a separate patch.

Okay, but what do you do if the backend doesn't support XDP (e.g. in case its an older kernel)? How do you know it is supporting XDP?

We can check a xenbus reply to xenbus state change.

Using the frontend state for that purpose is rather dangerous.

In case the backend doesn't support the "Reconfiguring" state you might end up with a broken network.

I'd rather let the backend advertise the support via a "feature-xdp" node in Xenstore and do the activation via another node.

Juergen

maintainer
#7 lore raw
Quoted 80 lines · Jürgen Groß
Quoted 72 lines · Denis Kirjanov
Quoted 71 lines · Jürgen Groß
Quoted 24 lines · Denis Kirjanov
Quoted 23 lines · Jürgen Groß
Quoted 7 lines · Denis Kirjanov

the patch adds a basic xdo logic to the netfront driver

XDP redirect is not supported yet

v2:
- avoid data copying while passing to XDP
- tell xen-natback that we need the headroom space

Please add the patch history below the "---" delimiter

Quoted 9 lines
---
    drivers/net/xen-netback/common.h |   1 +
    drivers/net/xen-netback/rx.c     |   9 ++-
    drivers/net/xen-netback/xenbus.c |  21 ++++++
    drivers/net/xen-netfront.c       | 157
+++++++++++++++++++++++++++++++++++++++
    4 files changed, 186 insertions(+), 2 deletions(-)

You are modifying xen-netback sources. Please Cc the maintainers.

...

Quoted 43 lines
Quoted 43 lines
Quoted 36 lines
drivers/net/xen-netfront.c +12 -0
+	}
+
+	return 0;
+
+abort_transaction:
+	xenbus_dev_fatal(np->xbdev, err, "%s", message);
+	xenbus_transaction_end(xbt, 1);
+out:
+	return err;
+}
+
+static int xennet_xdp_set(struct net_device *dev, struct bpf_prog

*prog,

drivers/net/xen-netfront.c +21 -0
+			struct netlink_ext_ack *extack)
+{
+	struct netfront_info *np = netdev_priv(dev);
+	struct bpf_prog *old_prog;
+	unsigned int i, err;
+
+	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
+	if (!old_prog && !prog)
+		return 0;
+
+	if (prog)
+		bpf_prog_add(prog, dev->real_num_tx_queues);
+
+	for (i = 0; i < dev->real_num_tx_queues; ++i)
+		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
+
+	if (old_prog)
+		for (i = 0; i < dev->real_num_tx_queues; ++i)
+			bpf_prog_put(old_prog);
+
+	err = talk_to_netback_xdp(np, old_prog ?

NETBACK_XDP_HEADROOM_DISABLE:

drivers/net/xen-netfront.c static void xennet_poll_controller(struct net_device *dev) · line 1384

Quoted 5 lines
drivers/net/xen-netfront.c +5 -0
+				  NETBACK_XDP_HEADROOM_ENABLE);
+	if (err)
+		return err;
+
+	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);

What is happening in case the backend doesn't support XDP?

Quoted 6 lines
Quoted 5 lines
Quoted 2 lines

Here we just ask xen-backend to make a headroom, that's it. It's better to send xen-backend changes in a separate patch.

Okay, but what do you do if the backend doesn't support XDP (e.g. in case its an older kernel)? How do you know it is supporting XDP?

We can check a xenbus reply to xenbus state change.

Quoted 7 lines

Using the frontend state for that purpose is rather dangerous.

In case the backend doesn't support the "Reconfiguring" state you might end up with a broken network.

I'd rather let the backend advertise the support via a "feature-xdp" node in Xenstore and do the activation via another node.

Yes, that's how feature negotiation is supposed to work. If XDP somehow doesn't fit into this model, we would like to know why at the very least.

Wei.

Quoted 3 lines

Juergen

#8 lore raw
In reply to Denis Kirjanov

Hi Denis,

There's a bunch of things still missing from my remarks on V1. XDP is not supposed to allocate and free pages constantly as that's one of the things that's making it fast.

You are also missing proper support for XDP_REDIRECT, ndo_xdp_xmit. We usually require the whole functionality to merge the driver.

Quoted 1 line · Denis Kirjanov

[...]

Quoted 8 lines
drivers/net/xen-netfront.c +8 -0
+u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
+		   struct xen_netif_rx_response *rx, struct bpf_prog *prog,
+		   struct xdp_buff *xdp)
+{
+	u32 len = rx->status;
+	u32 act = XDP_PASS;
+
+	xdp->data_hard_start = page_address(pdata);

drivers/net/xen-netfront.c static int xennet_get_extras(struct netfront_queue *queue, · line 802

Quoted 10 lines
drivers/net/xen-netfront.c +10 -0
+	xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM;
+	xdp_set_data_meta_invalid(xdp);
+	xdp->data_end = xdp->data + len;
+	xdp->handle = 0;
+
+	act = bpf_prog_run_xdp(prog, xdp);
+	switch (act) {
+	case XDP_PASS:
+	case XDP_TX:
+	case XDP_DROP:

Maybe i am missing something on the usage, but XDP_TX and XDROP are handled similarly?
XDP_TX is supposed to sent the packet out of the interface it arrived.

Quoted 169 lines
drivers/net/xen-netfront.c +135 -0
+		break;
+
+	case XDP_ABORTED:
+		trace_xdp_exception(queue->info->netdev, prog, act);
+		break;
+
+	default:
+		bpf_warn_invalid_xdp_action(act);
+	}
+
+	if (act != XDP_PASS && act != XDP_TX)
+		xdp->data_hard_start = NULL;
+
+	return act;
+}
+
 static int xennet_get_responses(struct netfront_queue *queue,
 				struct netfront_rx_info *rinfo, RING_IDX rp,
 				struct sk_buff_head *list)
@@ -792,6 +830,9 @@ static int xennet_get_responses(struct netfront_queue *queue,
 	int slots = 1;
 	int err = 0;
 	unsigned long ret;
+	struct bpf_prog *xdp_prog;
+	struct xdp_buff xdp;
+	u32 verdict;
 
 	if (rx->flags & XEN_NETRXF_extra_info) {
 		err = xennet_get_extras(queue, extras, rp);
@@ -827,6 +868,22 @@ static int xennet_get_responses(struct netfront_queue *queue,
 
 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
 
+		rcu_read_lock();
+		xdp_prog = rcu_dereference(queue->xdp_prog);
+		if (xdp_prog) {
+			/* currently only a single page contains data */
+			WARN_ON_ONCE(skb_shinfo(skb)->nr_frags != 1);
+			verdict = xennet_run_xdp(queue,
+				       skb_frag_page(&skb_shinfo(skb)->frags[0]),
+				       rx, xdp_prog, &xdp);
+
+			if (verdict != XDP_PASS && verdict != XDP_TX) {
+				err = -EINVAL;
+				goto next;
+			}
+
+		}
+		rcu_read_unlock();
 		__skb_queue_tail(list, skb);
 
 next:
@@ -1261,6 +1318,105 @@ static void xennet_poll_controller(struct net_device *dev)
 }
 #endif
 
+#define NETBACK_XDP_HEADROOM_DISABLE	0
+#define NETBACK_XDP_HEADROOM_ENABLE	1
+
+static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
+{
+	struct xenbus_transaction xbt;
+	const char *message;
+	int err;
+
+again:
+	err = xenbus_transaction_start(&xbt);
+	if (err) {
+		xenbus_dev_fatal(np->xbdev, err, "starting transaction");
+		goto out;
+	}
+
+	err = xenbus_printf(xbt, np->xbdev->nodename, "feature-xdp", "%d", xdp);
+	if (err) {
+		message = "writing feature-xdp";
+		goto abort_transaction;
+	}
+
+	err = xenbus_transaction_end(xbt, 0);
+	if (err) {
+		if (err == -EAGAIN)
+			goto again;
+	}
+
+	return 0;
+
+abort_transaction:
+	xenbus_dev_fatal(np->xbdev, err, "%s", message);
+	xenbus_transaction_end(xbt, 1);
+out:
+	return err;
+}
+
+static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
+			struct netlink_ext_ack *extack)
+{
+	struct netfront_info *np = netdev_priv(dev);
+	struct bpf_prog *old_prog;
+	unsigned int i, err;
+
+	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
+	if (!old_prog && !prog)
+		return 0;
+
+	if (prog)
+		bpf_prog_add(prog, dev->real_num_tx_queues);
+
+	for (i = 0; i < dev->real_num_tx_queues; ++i)
+		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
+
+	if (old_prog)
+		for (i = 0; i < dev->real_num_tx_queues; ++i)
+			bpf_prog_put(old_prog);
+
+	err = talk_to_netback_xdp(np, old_prog ? NETBACK_XDP_HEADROOM_DISABLE:
+				  NETBACK_XDP_HEADROOM_ENABLE);
+	if (err)
+		return err;
+
+	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
+
+	return 0;
+}
+
+static u32 xennet_xdp_query(struct net_device *dev)
+{
+	struct netfront_info *np = netdev_priv(dev);
+	unsigned int num_queues = dev->real_num_tx_queues;
+	unsigned int i;
+	struct netfront_queue *queue;
+	const struct bpf_prog *xdp_prog;
+
+	for (i = 0; i < num_queues; ++i) {
+		queue = &np->queues[i];
+		xdp_prog = rtnl_dereference(queue->xdp_prog);
+		if (xdp_prog)
+			return xdp_prog->aux->id;
+	}
+
+	return 0;
+}
+
+static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
+{
+	switch (xdp->command) {
+	case XDP_SETUP_PROG:
+		return xennet_xdp_set(dev, xdp->prog, xdp->extack);
+	case XDP_QUERY_PROG:
+		xdp->prog_id = xennet_xdp_query(dev);
+		return 0;
+	default:
+		return -EINVAL;
+	}
+}
+
 static const struct net_device_ops xennet_netdev_ops = {
 	.ndo_open            = xennet_open,
 	.ndo_stop            = xennet_close,
@@ -1272,6 +1428,7 @@ static void xennet_poll_controller(struct net_device *dev)
 	.ndo_fix_features    = xennet_fix_features,
 	.ndo_set_features    = xennet_set_features,
 	.ndo_select_queue    = xennet_select_queue,
+	.ndo_bpf            = xennet_xdp,
 #ifdef CONFIG_NET_POLL_CONTROLLER
 	.ndo_poll_controller = xennet_poll_controller,
 #endif
Signature · 1 line
1.8.3.1

Cheers
/Ilias

#9 lore raw
Quoted 6 lines · Ilias Apalodimas

Hi Denis,

There's a bunch of things still missing from my remarks on V1. XDP is not supposed to allocate and free pages constantly as that's one of the
things that's making it fast.

Hi Ilias,

I've removed the copying to an allocated page so there is no page allocation/free logic added.

Quoted 4 lines

You are also missing proper support for XDP_REDIRECT, ndo_xdp_xmit. We usually
require the whole functionality to merge the driver.

I wanted to minimize changes and send follow-up patches

Quoted 19 lines · Denis Kirjanov
Quoted 1 line

[...]

Quoted 14 lines
drivers/net/xen-netfront.c +14 -0
+u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
+		   struct xen_netif_rx_response *rx, struct bpf_prog *prog,
+		   struct xdp_buff *xdp)
+{
+	u32 len = rx->status;
+	u32 act = XDP_PASS;
+
+	xdp->data_hard_start = page_address(pdata);
+	xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM;
+	xdp_set_data_meta_invalid(xdp);
+	xdp->data_end = xdp->data + len;
+	xdp->handle = 0;
+
+	act = bpf_prog_run_xdp(prog, xdp);
Quoted 8 lines

drivers/net/xen-netfront.c static int xennet_get_extras(struct netfront_queue *queue, · line 802

Quoted 4 lines
drivers/net/xen-netfront.c +4 -0
+	switch (act) {
+	case XDP_PASS:
+	case XDP_TX:
+	case XDP_DROP:

Maybe i am missing something on the usage, but XDP_TX and XDROP are handled similarly?
XDP_TX is supposed to sent the packet out of the interface it arrived.

Yep, you're right. I'll add it to the next version.

Thanks!

Quoted 178 lines
Quoted 171 lines
drivers/net/xen-netfront.c +19 -0
+		break;
+
+	case XDP_ABORTED:
+		trace_xdp_exception(queue->info->netdev, prog, act);
+		break;
+
+	default:
+		bpf_warn_invalid_xdp_action(act);
+	}
+
+	if (act != XDP_PASS && act != XDP_TX)
+		xdp->data_hard_start = NULL;
+
+	return act;
+}
+
 static int xennet_get_responses(struct netfront_queue *queue,
 				struct netfront_rx_info *rinfo, RING_IDX rp,
 				struct sk_buff_head *list)
@@ -792,6 +830,9 @@ static int xennet_get_responses(struct netfront_queue
 *queue,
 	int slots = 1;
 	int err = 0;
 	unsigned long ret;
+	struct bpf_prog *xdp_prog;
+	struct xdp_buff xdp;
+	u32 verdict;
 
 	if (rx->flags & XEN_NETRXF_extra_info) {
drivers/net/xen-netfront.c +16 -0
 		err = xennet_get_extras(queue, extras, rp);
@@ -827,6 +868,22 @@ static int xennet_get_responses(struct netfront_queue
 *queue,
 
 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
 
+		rcu_read_lock();
+		xdp_prog = rcu_dereference(queue->xdp_prog);
+		if (xdp_prog) {
+			/* currently only a single page contains data */
+			WARN_ON_ONCE(skb_shinfo(skb)->nr_frags != 1);
+			verdict = xennet_run_xdp(queue,
+				       skb_frag_page(&skb_shinfo(skb)->frags[0]),
+				       rx, xdp_prog, &xdp);
+
+			if (verdict != XDP_PASS && verdict != XDP_TX) {
+				err = -EINVAL;
+				goto next;
+			}
+
+		}
+		rcu_read_unlock();
 		__skb_queue_tail(list, skb);
 

next:

drivers/net/xen-netfront.c +99 -0
@@ -1261,6 +1318,105 @@ static void xennet_poll_controller(struct
 net_device *dev)
 }
 #endif
 
+#define NETBACK_XDP_HEADROOM_DISABLE	0
+#define NETBACK_XDP_HEADROOM_ENABLE	1
+
+static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
+{
+	struct xenbus_transaction xbt;
+	const char *message;
+	int err;
+
+again:
+	err = xenbus_transaction_start(&xbt);
+	if (err) {
+		xenbus_dev_fatal(np->xbdev, err, "starting transaction");
+		goto out;
+	}
+
+	err = xenbus_printf(xbt, np->xbdev->nodename, "feature-xdp", "%d",
 xdp);
+	if (err) {
+		message = "writing feature-xdp";
+		goto abort_transaction;
+	}
+
+	err = xenbus_transaction_end(xbt, 0);
+	if (err) {
+		if (err == -EAGAIN)
+			goto again;
+	}
+
+	return 0;
+
+abort_transaction:
+	xenbus_dev_fatal(np->xbdev, err, "%s", message);
+	xenbus_transaction_end(xbt, 1);
+out:
+	return err;
+}
+
+static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
+			struct netlink_ext_ack *extack)
+{
+	struct netfront_info *np = netdev_priv(dev);
+	struct bpf_prog *old_prog;
+	unsigned int i, err;
+
+	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
+	if (!old_prog && !prog)
+		return 0;
+
+	if (prog)
+		bpf_prog_add(prog, dev->real_num_tx_queues);
+
+	for (i = 0; i < dev->real_num_tx_queues; ++i)
+		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
+
+	if (old_prog)
+		for (i = 0; i < dev->real_num_tx_queues; ++i)
+			bpf_prog_put(old_prog);
+
+	err = talk_to_netback_xdp(np, old_prog ? NETBACK_XDP_HEADROOM_DISABLE:
+				  NETBACK_XDP_HEADROOM_ENABLE);
+	if (err)
+		return err;
+
+	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
+
+	return 0;
+}
+
+static u32 xennet_xdp_query(struct net_device *dev)
+{
+	struct netfront_info *np = netdev_priv(dev);
+	unsigned int num_queues = dev->real_num_tx_queues;
+	unsigned int i;
+	struct netfront_queue *queue;
+	const struct bpf_prog *xdp_prog;
+
+	for (i = 0; i < num_queues; ++i) {
+		queue = &np->queues[i];
+		xdp_prog = rtnl_dereference(queue->xdp_prog);
+		if (xdp_prog)
+			return xdp_prog->aux->id;
+	}
+
+	return 0;
+}
+
+static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
+{
+	switch (xdp->command) {
+	case XDP_SETUP_PROG:
+		return xennet_xdp_set(dev, xdp->prog, xdp->extack);
+	case XDP_QUERY_PROG:
+		xdp->prog_id = xennet_xdp_query(dev);
+		return 0;
+	default:
+		return -EINVAL;
+	}
+}
+
 static const struct net_device_ops xennet_netdev_ops = {
drivers/net/xen-netfront.c +1 -0
 	.ndo_open            = xennet_open,
 	.ndo_stop            = xennet_close,
@@ -1272,6 +1428,7 @@ static void xennet_poll_controller(struct net_device
 *dev)
 	.ndo_fix_features    = xennet_fix_features,
 	.ndo_set_features    = xennet_set_features,
 	.ndo_select_queue    = xennet_select_queue,
+	.ndo_bpf            = xennet_xdp,
 #ifdef CONFIG_NET_POLL_CONTROLLER
 	.ndo_poll_controller = xennet_poll_controller,
Quoted 4 lines

#endif
--
1.8.3.1

Cheers
/Ilias

Signature · 2 lines
Regards / Mit besten Grüßen,
Denis
#10 lore raw
Quoted 8 lines · Denis Kirjanov
Quoted 6 lines · Ilias Apalodimas

Hi Denis,

There's a bunch of things still missing from my remarks on V1. XDP is not supposed to allocate and free pages constantly as that's one of the
things that's making it fast.

Quoted 5 lines

Hi Ilias,

I've removed the copying to an allocated page so there is no page allocation/free logic added.

i
Yea that has been removed. I am not familiar with the driver though, so i'll give you an example.
Let's say the BPF program says the packet must be dropped. What will happen to the page with the packet payload?
Ideally on XDP we want that page recycled back into the device descriptors, so the driver won't have to allocate and map a fresh page.

Quoted 8 lines
Quoted 5 lines

You are also missing proper support for XDP_REDIRECT, ndo_xdp_xmit. We usually
require the whole functionality to merge the driver.

I wanted to minimize changes and send follow-up patches

Adding XDP_REDIRECT is pretty trivial and the ndo_xdp_xmit should be very similar to XDP_TX. So assuming you'll fix XDP_TX adding the .ndo one will be relatively small amount of code.

Quoted 28 lines
Quoted 19 lines · Denis Kirjanov
Quoted 1 line

[...]

Quoted 14 lines
drivers/net/xen-netfront.c +14 -0
+u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
+		   struct xen_netif_rx_response *rx, struct bpf_prog *prog,
+		   struct xdp_buff *xdp)
+{
+	u32 len = rx->status;
+	u32 act = XDP_PASS;
+
+	xdp->data_hard_start = page_address(pdata);
+	xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM;
+	xdp_set_data_meta_invalid(xdp);
+	xdp->data_end = xdp->data + len;
+	xdp->handle = 0;
+
+	act = bpf_prog_run_xdp(prog, xdp);
Quoted 8 lines

drivers/net/xen-netfront.c static int xennet_get_extras(struct netfront_queue *queue, · line 802

Quoted 4 lines
drivers/net/xen-netfront.c +4 -0
+	switch (act) {
+	case XDP_PASS:
+	case XDP_TX:
+	case XDP_DROP:

Maybe i am missing something on the usage, but XDP_TX and XDROP are handled similarly?
XDP_TX is supposed to sent the packet out of the interface it arrived.

Quoted 4 lines

Yep, you're right. I'll add it to the next version.

Thanks!

Cheers
/Ilias

#11 lore raw
Quoted 14 lines · On Thu, 5 Mar 2020 15:31:14 +0200 Ilias Apalodimas…
Quoted 12 lines · Denis Kirjanov
Quoted 5 lines · Ilias Apalodimas

Hi Denis,

There's a bunch of things still missing from my remarks on V1. XDP is not supposed to allocate and free pages constantly as that's one of the things that's making it fast.

Hi Ilias,

I've removed the copying to an allocated page so there is no page allocation/free logic added.

Quoted 6 lines

Yea that has been removed. I am not familiar with the driver though, so i'll give you an example.
Let's say the BPF program says the packet must be dropped. What will happen to the page with the packet payload?
Ideally on XDP we want that page recycled back into the device descriptors, so the driver won't have to allocate and map a fresh page.

I agree. The main point with XDP is that we can do something faster-than the normal network stack. Especially in case of XDP_DROP, we do driver specific recycling tricks, to avoid any allocations and reinsert the RX-frame in RX-ring, and avoid overhead of SKB allocations.

Looking closer at your patch it seem you run XDP after the SKB alloc?!?

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

You are also missing proper support for XDP_REDIRECT, ndo_xdp_xmit. We usually require the whole functionality to merge the driver.

I agree, we have unfortunately seen drivers not getting completed if we don't require full-XDP feature set.

Quoted 7 lines
Quoted 3 lines

I wanted to minimize changes and send follow-up patches

Adding XDP_REDIRECT is pretty trivial and the ndo_xdp_xmit should be very similar to XDP_TX. So assuming you'll fix XDP_TX adding the .ndo one will be relatively small amount of code.

You can have a patchset with more patches, if you prefer splitting this up in multiple patches.

Signature · 4 lines
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer
#12 lore raw
In reply to Denis Kirjanov

drivers/net/xen-netback/xenbus.c static void set_backend_state(struct backend_info *be, · line 397

Quoted 10 lines · On Mon, 2 Mar 2020 17:21:14 +0300 Denis Kirjanov <[email protected]> wrote:
drivers/net/xen-netback/xenbus.c +2 -0
index 286054b..81a6023 100644
@@ -393,6 +393,20 @@ static void set_backend_state(struct backend_info *be,
 	}
 }
 
+static void read_xenbus_fronetend_xdp(struct backend_info *be,
+				      struct xenbus_device *dev)

Strange spelling of this function name: 'fronetend'

Quoted 16 lines
drivers/net/xen-netback/xenbus.c +12 -0
+{
+	struct xenvif *vif = be->vif;
+	unsigned int val;
+	int err;
+
+	err = xenbus_scanf(XBT_NIL, dev->otherend,
+			   "feature-xdp", "%u", &val);
+	if (err < 0)
+		return;
+	vif->xdp_enabled = val;
+}
+
 /**
  * Callback received when the frontend's state changes.
  */

drivers/net/xen-netback/xenbus.c static void frontend_changed(struct xenbus_device *dev, · line 435

Quoted 5 lines
drivers/net/xen-netback/xenbus.c +2 -0
@@ -417,6 +431,11 @@ static void frontend_changed(struct xenbus_device *dev,
 		set_backend_state(be, XenbusStateConnected);
 		break;
 
+	case XenbusStateReconfiguring:
+		read_xenbus_fronetend_xdp(be, dev);
                            ^^^^^^^^
Quoted 3 lines
drivers/net/xen-netback/xenbus.c +3 -0
+		xenbus_switch_state(dev, XenbusStateReconfigured);
+		break;
+
Signature · 4 lines
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer
#13 lore raw
In reply to Denis Kirjanov
Quoted 4 lines · On Mon, 2 Mar 2020 17:21:14 +0300 Denis Kirjanov <[email protected]> wrote:
drivers/net/xen-netfront.c +0 -0
index 482c6c8..db8a280 100644

[...]

Quoted 52 lines
drivers/net/xen-netfront.c +37 -0
@@ -778,6 +782,40 @@ static int xennet_get_extras(struct netfront_queue *queue,
 	return err;
 }
 
+u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
+		   struct xen_netif_rx_response *rx, struct bpf_prog *prog,
+		   struct xdp_buff *xdp)
+{
+	u32 len = rx->status;
+	u32 act = XDP_PASS;
+
+	xdp->data_hard_start = page_address(pdata);
+	xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM;
+	xdp_set_data_meta_invalid(xdp);
+	xdp->data_end = xdp->data + len;
+	xdp->handle = 0;
+
+	act = bpf_prog_run_xdp(prog, xdp);
+	switch (act) {
+	case XDP_PASS:
+	case XDP_TX:
+	case XDP_DROP:
+		break;
+
+	case XDP_ABORTED:
+		trace_xdp_exception(queue->info->netdev, prog, act);
+		break;
+
+	default:
+		bpf_warn_invalid_xdp_action(act);
+	}
+
+	if (act != XDP_PASS && act != XDP_TX)
+		xdp->data_hard_start = NULL;
+
+	return act;
+}
+
 static int xennet_get_responses(struct netfront_queue *queue,
 				struct netfront_rx_info *rinfo, RING_IDX rp,
 				struct sk_buff_head *list)
@@ -792,6 +830,9 @@ static int xennet_get_responses(struct netfront_queue *queue,
 	int slots = 1;
 	int err = 0;
 	unsigned long ret;
+	struct bpf_prog *xdp_prog;
+	struct xdp_buff xdp;
+	u32 verdict;
 
 	if (rx->flags & XEN_NETRXF_extra_info) {
 		err = xennet_get_extras(queue, extras, rp);

drivers/net/xen-netfront.c static int xennet_get_responses(struct netfront_queue *queue, · line 877

Quoted 10 lines
drivers/net/xen-netfront.c +7 -0
@@ -827,6 +868,22 @@ static int xennet_get_responses(struct netfront_queue *queue,
 
 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
 
+		rcu_read_lock();
+		xdp_prog = rcu_dereference(queue->xdp_prog);
+		if (xdp_prog) {
+			/* currently only a single page contains data */
+			WARN_ON_ONCE(skb_shinfo(skb)->nr_frags != 1);
+			verdict = xennet_run_xdp(queue,
+				       skb_frag_page(&skb_shinfo(skb)->frags[0]),

This looks really weird, skb_shinfo(skb)->frags[0], you already have an SKB and you are sending fragment-0 to the xennet_run_xdp() function.

XDP meant to run before an SKB is allocated...

Quoted 10 lines
drivers/net/xen-netfront.c +9 -0
+				       rx, xdp_prog, &xdp);
+
+			if (verdict != XDP_PASS && verdict != XDP_TX) {
+				err = -EINVAL;
+				goto next;
+			}
+
+		}
+		rcu_read_unlock();
 		__skb_queue_tail(list, skb);
Signature · 4 lines
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer
#14 lore raw
Quoted 77 lines · Jesper Dangaard Brouer
Quoted 4 lines · On Mon, 2 Mar 2020 17:21:14 +0300 Denis Kirjanov <[email protected]> wrote:
drivers/net/xen-netfront.c +0 -0
index 482c6c8..db8a280 100644

[...]

Quoted 55 lines
drivers/net/xen-netfront.c +34 -0
@@ -778,6 +782,40 @@ static int xennet_get_extras(struct netfront_queue
 *queue,
 	return err;
 }
 
+u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
+		   struct xen_netif_rx_response *rx, struct bpf_prog *prog,
+		   struct xdp_buff *xdp)
+{
+	u32 len = rx->status;
+	u32 act = XDP_PASS;
+
+	xdp->data_hard_start = page_address(pdata);
+	xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM;
+	xdp_set_data_meta_invalid(xdp);
+	xdp->data_end = xdp->data + len;
+	xdp->handle = 0;
+
+	act = bpf_prog_run_xdp(prog, xdp);
+	switch (act) {
+	case XDP_PASS:
+	case XDP_TX:
+	case XDP_DROP:
+		break;
+
+	case XDP_ABORTED:
+		trace_xdp_exception(queue->info->netdev, prog, act);
+		break;
+
+	default:
+		bpf_warn_invalid_xdp_action(act);
+	}
+
+	if (act != XDP_PASS && act != XDP_TX)
+		xdp->data_hard_start = NULL;
+
+	return act;
+}
+
 static int xennet_get_responses(struct netfront_queue *queue,
 				struct netfront_rx_info *rinfo, RING_IDX rp,
drivers/net/xen-netfront.c +3 -0
 				struct sk_buff_head *list)
@@ -792,6 +830,9 @@ static int xennet_get_responses(struct netfront_queue
 *queue,
 	int slots = 1;
 	int err = 0;
 	unsigned long ret;
+	struct bpf_prog *xdp_prog;
+	struct xdp_buff xdp;
+	u32 verdict;
 
 	if (rx->flags & XEN_NETRXF_extra_info) {
drivers/net/xen-netfront.c +0 -0
 		err = xennet_get_extras(queue, extras, rp);
@@ -827,6 +868,22 @@ static int xennet_get_responses(struct netfront_queue
 *queue,

drivers/net/xen-netfront.c static int xennet_get_responses(struct netfront_queue · line 878

Quoted 10 lines
drivers/net/xen-netfront.c +7 -0
 
 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
 
+		rcu_read_lock();
+		xdp_prog = rcu_dereference(queue->xdp_prog);
+		if (xdp_prog) {
+			/* currently only a single page contains data */
+			WARN_ON_ONCE(skb_shinfo(skb)->nr_frags != 1);
+			verdict = xennet_run_xdp(queue,
+				       skb_frag_page(&skb_shinfo(skb)->frags[0]),

This looks really weird, skb_shinfo(skb)->frags[0], you already have an SKB and you are sending fragment-0 to the xennet_run_xdp() function.

Quoted 1 line

XDP meant to run before an SKB is allocated...

Agreed. But the xen networking works that data communicated using shared pages between backend and frontend parts.
So xen-netfront just allocates skbs with attached pages to them and shares those pages with xen-netback

Quoted 20 lines

drivers/net/xen-netfront.c static int xennet_get_responses(struct netfront_queue *queue, · line 887

Quoted 11 lines
drivers/net/xen-netfront.c +9 -0
+				       rx, xdp_prog, &xdp);
+
+			if (verdict != XDP_PASS && verdict != XDP_TX) {
+				err = -EINVAL;
+				goto next;
+			}
+
+		}
+		rcu_read_unlock();
 		__skb_queue_tail(list, skb);

--
Best regards,
Jesper Dangaard Brouer
MSc.CS, Principal Kernel Engineer at Red Hat
LinkedIn: http://www.linkedin.com/in/brouer