xsk: add zero-copy support for Rx

Extend the xsk_rcv to support the new MEM_TYPE_ZERO_COPY memory, and
wireup ndo_bpf call in bind.

Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

include/net/xdp_sock.h

@@ -22,6 +22,7 @@ struct xdp_umem_props {
 
 struct xdp_umem_page {
 	void *addr;
+	dma_addr_t dma;
 };
 
 struct xdp_umem {
@@ -38,6 +39,9 @@ struct xdp_umem {
 	struct work_struct work;
 	struct page **pgs;
 	u32 npgs;
+	struct net_device *dev;
+	u16 queue_id;
+	bool zc;
 };
 
 struct xdp_sock {
@@ -60,6 +64,8 @@ int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp);
 int xsk_rcv(struct xdp_sock *xs, struct xdp_buff *xdp);
 void xsk_flush(struct xdp_sock *xs);
 bool xsk_is_setup_for_bpf_map(struct xdp_sock *xs);
+u64 *xsk_umem_peek_addr(struct xdp_umem *umem, u64 *addr);
+void xsk_umem_discard_addr(struct xdp_umem *umem);
 #else
 static inline int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp)
 {

include/uapi/linux/if_xdp.h

@@ -13,7 +13,9 @@
 #include <linux/types.h>
 
 /* Options for the sxdp_flags field */
-#define XDP_SHARED_UMEM 1
+#define XDP_SHARED_UMEM	(1 << 0)
+#define XDP_COPY	(1 << 1) /* Force copy-mode */
+#define XDP_ZEROCOPY	(1 << 2) /* Force zero-copy mode */
 
 struct sockaddr_xdp {
 	__u16 sxdp_family;

net/xdp/xdp_umem.c

@@ -17,6 +17,81 @@
 
 #define XDP_UMEM_MIN_CHUNK_SIZE 2048
 
+int xdp_umem_assign_dev(struct xdp_umem *umem, struct net_device *dev,
+			u32 queue_id, u16 flags)
+{
+	bool force_zc, force_copy;
+	struct netdev_bpf bpf;
+	int err;
+
+	force_zc = flags & XDP_ZEROCOPY;
+	force_copy = flags & XDP_COPY;
+
+	if (force_zc && force_copy)
+		return -EINVAL;
+
+	if (force_copy)
+		return 0;
+
+	dev_hold(dev);
+
+	if (dev->netdev_ops->ndo_bpf) {
+		bpf.command = XDP_QUERY_XSK_UMEM;
+
+		rtnl_lock();
+		err = dev->netdev_ops->ndo_bpf(dev, &bpf);
+		rtnl_unlock();
+
+		if (err) {
+			dev_put(dev);
+			return force_zc ? -ENOTSUPP : 0;
+		}
+
+		bpf.command = XDP_SETUP_XSK_UMEM;
+		bpf.xsk.umem = umem;
+		bpf.xsk.queue_id = queue_id;
+
+		rtnl_lock();
+		err = dev->netdev_ops->ndo_bpf(dev, &bpf);
+		rtnl_unlock();
+
+		if (err) {
+			dev_put(dev);
+			return force_zc ? err : 0; /* fail or fallback */
+		}
+
+		umem->dev = dev;
+		umem->queue_id = queue_id;
+		umem->zc = true;
+		return 0;
+	}
+
+	dev_put(dev);
+	return force_zc ? -ENOTSUPP : 0; /* fail or fallback */
+}
+
+void xdp_umem_clear_dev(struct xdp_umem *umem)
+{
+	struct netdev_bpf bpf;
+	int err;
+
+	if (umem->dev) {
+		bpf.command = XDP_SETUP_XSK_UMEM;
+		bpf.xsk.umem = NULL;
+		bpf.xsk.queue_id = umem->queue_id;
+
+		rtnl_lock();
+		err = umem->dev->netdev_ops->ndo_bpf(umem->dev, &bpf);
+		rtnl_unlock();
+
+		if (err)
+			WARN(1, "failed to disable umem!\n");
+
+		dev_put(umem->dev);
+		umem->dev = NULL;
+	}
+}
+
 static void xdp_umem_unpin_pages(struct xdp_umem *umem)
 {
 	unsigned int i;
@@ -43,6 +118,8 @@ static void xdp_umem_release(struct xdp_umem *umem)
 	struct task_struct *task;
 	struct mm_struct *mm;
 
+	xdp_umem_clear_dev(umem);
+
 	if (umem->fq) {
 		xskq_destroy(umem->fq);
 		umem->fq = NULL;

net/xdp/xdp_umem.h

@@ -13,6 +13,9 @@ static inline char *xdp_umem_get_data(struct xdp_umem *umem, u64 addr)
 	return umem->pages[addr >> PAGE_SHIFT].addr + (addr & (PAGE_SIZE - 1));
 }
 
+int xdp_umem_assign_dev(struct xdp_umem *umem, struct net_device *dev,
+			u32 queue_id, u16 flags);
+void xdp_umem_clear_dev(struct xdp_umem *umem);
 bool xdp_umem_validate_queues(struct xdp_umem *umem);
 void xdp_get_umem(struct xdp_umem *umem);
 void xdp_put_umem(struct xdp_umem *umem);

net/xdp/xsk.c

@@ -36,19 +36,28 @@ static struct xdp_sock *xdp_sk(struct sock *sk)
 
 bool xsk_is_setup_for_bpf_map(struct xdp_sock *xs)
 {
-	return !!xs->rx;
+	return READ_ONCE(xs->rx) &&  READ_ONCE(xs->umem) &&
+		READ_ONCE(xs->umem->fq);
 }
 
-static int __xsk_rcv(struct xdp_sock *xs, struct xdp_buff *xdp)
+u64 *xsk_umem_peek_addr(struct xdp_umem *umem, u64 *addr)
+{
+	return xskq_peek_addr(umem->fq, addr);
+}
+EXPORT_SYMBOL(xsk_umem_peek_addr);
+
+void xsk_umem_discard_addr(struct xdp_umem *umem)
+{
+	xskq_discard_addr(umem->fq);
+}
+EXPORT_SYMBOL(xsk_umem_discard_addr);
+
+static int __xsk_rcv(struct xdp_sock *xs, struct xdp_buff *xdp, u32 len)
 {
-	u32 len = xdp->data_end - xdp->data;
 	void *buffer;
 	u64 addr;
 	int err;
 
-	if (xs->dev != xdp->rxq->dev || xs->queue_id != xdp->rxq->queue_index)
-		return -EINVAL;
-
 	if (!xskq_peek_addr(xs->umem->fq, &addr) ||
 	    len > xs->umem->chunk_size_nohr) {
 		xs->rx_dropped++;
@@ -60,25 +69,41 @@ static int __xsk_rcv(struct xdp_sock *xs, struct xdp_buff *xdp)
 	buffer = xdp_umem_get_data(xs->umem, addr);
 	memcpy(buffer, xdp->data, len);
 	err = xskq_produce_batch_desc(xs->rx, addr, len);
-	if (!err)
+	if (!err) {
 		xskq_discard_addr(xs->umem->fq);
-	else
-		xs->rx_dropped++;
+		xdp_return_buff(xdp);
+		return 0;
+	}
 
+	xs->rx_dropped++;
 	return err;
 }
 
-int xsk_rcv(struct xdp_sock *xs, struct xdp_buff *xdp)
+static int __xsk_rcv_zc(struct xdp_sock *xs, struct xdp_buff *xdp, u32 len)
 {
-	int err;
+	int err = xskq_produce_batch_desc(xs->rx, (u64)xdp->handle, len);
 
-	err = __xsk_rcv(xs, xdp);
-	if (likely(!err))
+	if (err) {
 		xdp_return_buff(xdp);
+		xs->rx_dropped++;
+	}
 
 	return err;
 }
 
+int xsk_rcv(struct xdp_sock *xs, struct xdp_buff *xdp)
+{
+	u32 len;
+
+	if (xs->dev != xdp->rxq->dev || xs->queue_id != xdp->rxq->queue_index)
+		return -EINVAL;
+
+	len = xdp->data_end - xdp->data;
+
+	return (xdp->rxq->mem.type == MEM_TYPE_ZERO_COPY) ?
+		__xsk_rcv_zc(xs, xdp, len) : __xsk_rcv(xs, xdp, len);
+}
+
 void xsk_flush(struct xdp_sock *xs)
 {
 	xskq_produce_flush_desc(xs->rx);
@@ -87,12 +112,29 @@ void xsk_flush(struct xdp_sock *xs)
 
 int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp)
 {
+	u32 len = xdp->data_end - xdp->data;
+	void *buffer;
+	u64 addr;
 	int err;
 
-	err = __xsk_rcv(xs, xdp);
-	if (!err)
+	if (!xskq_peek_addr(xs->umem->fq, &addr) ||
+	    len > xs->umem->chunk_size_nohr) {
+		xs->rx_dropped++;
+		return -ENOSPC;
+	}
+
+	addr += xs->umem->headroom;
+
+	buffer = xdp_umem_get_data(xs->umem, addr);
+	memcpy(buffer, xdp->data, len);
+	err = xskq_produce_batch_desc(xs->rx, addr, len);
+	if (!err) {
+		xskq_discard_addr(xs->umem->fq);
 		xsk_flush(xs);
+		return 0;
+	}
 
+	xs->rx_dropped++;
 	return err;
 }
 
@@ -291,6 +333,7 @@ static int xsk_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 	struct sock *sk = sock->sk;
 	struct xdp_sock *xs = xdp_sk(sk);
 	struct net_device *dev;
+	u32 flags, qid;
 	int err = 0;
 
 	if (addr_len < sizeof(struct sockaddr_xdp))
@@ -315,16 +358,26 @@ static int xsk_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 		goto out_unlock;
 	}
 
-	if ((xs->rx && sxdp->sxdp_queue_id >= dev->real_num_rx_queues) ||
-	    (xs->tx && sxdp->sxdp_queue_id >= dev->real_num_tx_queues)) {
+	qid = sxdp->sxdp_queue_id;
+
+	if ((xs->rx && qid >= dev->real_num_rx_queues) ||
+	    (xs->tx && qid >= dev->real_num_tx_queues)) {
 		err = -EINVAL;
 		goto out_unlock;
 	}
 
-	if (sxdp->sxdp_flags & XDP_SHARED_UMEM) {
+	flags = sxdp->sxdp_flags;
+
+	if (flags & XDP_SHARED_UMEM) {
 		struct xdp_sock *umem_xs;
 		struct socket *sock;
 
+		if ((flags & XDP_COPY) || (flags & XDP_ZEROCOPY)) {
+			/* Cannot specify flags for shared sockets. */
+			err = -EINVAL;
+			goto out_unlock;
+		}
+
 		if (xs->umem) {
 			/* We have already our own. */
 			err = -EINVAL;
@@ -343,8 +396,7 @@ static int xsk_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 			err = -EBADF;
 			sockfd_put(sock);
 			goto out_unlock;
-		} else if (umem_xs->dev != dev ||
-			   umem_xs->queue_id != sxdp->sxdp_queue_id) {
+		} else if (umem_xs->dev != dev || umem_xs->queue_id != qid) {
 			err = -EINVAL;
 			sockfd_put(sock);
 			goto out_unlock;
@@ -360,6 +412,10 @@ static int xsk_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 		/* This xsk has its own umem. */
 		xskq_set_umem(xs->umem->fq, &xs->umem->props);
 		xskq_set_umem(xs->umem->cq, &xs->umem->props);
+
+		err = xdp_umem_assign_dev(xs->umem, dev, qid, flags);
+		if (err)
+			goto out_unlock;
 	}
 
 	xs->dev = dev;