virtio/vsock: replace virtio_vsock_pkt with sk_buff

This commit changes virtio/vsock to use sk_buff instead of
virtio_vsock_pkt. Beyond better conforming to other net code, using
sk_buff allows vsock to use sk_buff-dependent features in the future
(such as sockmap) and improves throughput.

This patch introduces the following performance changes:

Tool: Uperf
Env: Phys Host + L1 Guest
Payload: 64k
Threads: 16
Test Runs: 10
Type: SOCK_STREAM
Before: commit b7bfaa76 ("Linux 6.2-rc3")

Before
------
g2h: 16.77Gb/s
h2g: 10.56Gb/s

After
-----
g2h: 21.04Gb/s
h2g: 10.76Gb/s
Signed-off-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/virtio_vsock.h

@@ -7,6 +7,109 @@
 #include <net/sock.h>
 #include <net/af_vsock.h>
 
+#define VIRTIO_VSOCK_SKB_HEADROOM (sizeof(struct virtio_vsock_hdr))
+
+struct virtio_vsock_skb_cb {
+	bool reply;
+	bool tap_delivered;
+};
+
+#define VIRTIO_VSOCK_SKB_CB(skb) ((struct virtio_vsock_skb_cb *)((skb)->cb))
+
+static inline struct virtio_vsock_hdr *virtio_vsock_hdr(struct sk_buff *skb)
+{
+	return (struct virtio_vsock_hdr *)skb->head;
+}
+
+static inline bool virtio_vsock_skb_reply(struct sk_buff *skb)
+{
+	return VIRTIO_VSOCK_SKB_CB(skb)->reply;
+}
+
+static inline void virtio_vsock_skb_set_reply(struct sk_buff *skb)
+{
+	VIRTIO_VSOCK_SKB_CB(skb)->reply = true;
+}
+
+static inline bool virtio_vsock_skb_tap_delivered(struct sk_buff *skb)
+{
+	return VIRTIO_VSOCK_SKB_CB(skb)->tap_delivered;
+}
+
+static inline void virtio_vsock_skb_set_tap_delivered(struct sk_buff *skb)
+{
+	VIRTIO_VSOCK_SKB_CB(skb)->tap_delivered = true;
+}
+
+static inline void virtio_vsock_skb_clear_tap_delivered(struct sk_buff *skb)
+{
+	VIRTIO_VSOCK_SKB_CB(skb)->tap_delivered = false;
+}
+
+static inline void virtio_vsock_skb_rx_put(struct sk_buff *skb)
+{
+	u32 len;
+
+	len = le32_to_cpu(virtio_vsock_hdr(skb)->len);
+
+	if (len > 0)
+		skb_put(skb, len);
+}
+
+static inline struct sk_buff *virtio_vsock_alloc_skb(unsigned int size, gfp_t mask)
+{
+	struct sk_buff *skb;
+
+	if (size < VIRTIO_VSOCK_SKB_HEADROOM)
+		return NULL;
+
+	skb = alloc_skb(size, mask);
+	if (!skb)
+		return NULL;
+
+	skb_reserve(skb, VIRTIO_VSOCK_SKB_HEADROOM);
+	return skb;
+}
+
+static inline void
+virtio_vsock_skb_queue_head(struct sk_buff_head *list, struct sk_buff *skb)
+{
+	spin_lock_bh(&list->lock);
+	__skb_queue_head(list, skb);
+	spin_unlock_bh(&list->lock);
+}
+
+static inline void
+virtio_vsock_skb_queue_tail(struct sk_buff_head *list, struct sk_buff *skb)
+{
+	spin_lock_bh(&list->lock);
+	__skb_queue_tail(list, skb);
+	spin_unlock_bh(&list->lock);
+}
+
+static inline struct sk_buff *virtio_vsock_skb_dequeue(struct sk_buff_head *list)
+{
+	struct sk_buff *skb;
+
+	spin_lock_bh(&list->lock);
+	skb = __skb_dequeue(list);
+	spin_unlock_bh(&list->lock);
+
+	return skb;
+}
+
+static inline void virtio_vsock_skb_queue_purge(struct sk_buff_head *list)
+{
+	spin_lock_bh(&list->lock);
+	__skb_queue_purge(list);
+	spin_unlock_bh(&list->lock);
+}
+
+static inline size_t virtio_vsock_skb_len(struct sk_buff *skb)
+{
+	return (size_t)(skb_end_pointer(skb) - skb->head);
+}
+
 #define VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE	(1024 * 4)
 #define VIRTIO_VSOCK_MAX_BUF_SIZE		0xFFFFFFFFUL
 #define VIRTIO_VSOCK_MAX_PKT_BUF_SIZE		(1024 * 64)
@@ -35,23 +138,10 @@ struct virtio_vsock_sock {
 	u32 last_fwd_cnt;
 	u32 rx_bytes;
 	u32 buf_alloc;
-	struct list_head rx_queue;
+	struct sk_buff_head rx_queue;
 	u32 msg_count;
 };
 
-struct virtio_vsock_pkt {
-	struct virtio_vsock_hdr	hdr;
-	struct list_head list;
-	/* socket refcnt not held, only use for cancellation */
-	struct vsock_sock *vsk;
-	void *buf;
-	u32 buf_len;
-	u32 len;
-	u32 off;
-	bool reply;
-	bool tap_delivered;
-};
-
 struct virtio_vsock_pkt_info {
 	u32 remote_cid, remote_port;
 	struct vsock_sock *vsk;
@@ -68,7 +158,7 @@ struct virtio_transport {
 	struct vsock_transport transport;
 
 	/* Takes ownership of the packet */
-	int (*send_pkt)(struct virtio_vsock_pkt *pkt);
+	int (*send_pkt)(struct sk_buff *skb);
 };
 
 ssize_t
@@ -149,11 +239,10 @@ virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
 void virtio_transport_destruct(struct vsock_sock *vsk);
 
 void virtio_transport_recv_pkt(struct virtio_transport *t,
-			       struct virtio_vsock_pkt *pkt);
-void virtio_transport_free_pkt(struct virtio_vsock_pkt *pkt);
-void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct virtio_vsock_pkt *pkt);
+			       struct sk_buff *skb);
+void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb);
 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 wanted);
 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit);
-void virtio_transport_deliver_tap_pkt(struct virtio_vsock_pkt *pkt);
-
+void virtio_transport_deliver_tap_pkt(struct sk_buff *skb);
+int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *list);
 #endif /* _LINUX_VIRTIO_VSOCK_H */

net/vmw_vsock/virtio_transport.c

@@ -42,8 +42,7 @@ struct virtio_vsock {
 	bool tx_run;
 
 	struct work_struct send_pkt_work;
-	spinlock_t send_pkt_list_lock;
-	struct list_head send_pkt_list;
+	struct sk_buff_head send_pkt_queue;
 
 	atomic_t queued_replies;
 
@@ -101,41 +100,31 @@ virtio_transport_send_pkt_work(struct work_struct *work)
 	vq = vsock->vqs[VSOCK_VQ_TX];
 
 	for (;;) {
-		struct virtio_vsock_pkt *pkt;
 		struct scatterlist hdr, buf, *sgs[2];
 		int ret, in_sg = 0, out_sg = 0;
+		struct sk_buff *skb;
 		bool reply;
 
-		spin_lock_bh(&vsock->send_pkt_list_lock);
-		if (list_empty(&vsock->send_pkt_list)) {
-			spin_unlock_bh(&vsock->send_pkt_list_lock);
+		skb = virtio_vsock_skb_dequeue(&vsock->send_pkt_queue);
+		if (!skb)
 			break;
-		}
-
-		pkt = list_first_entry(&vsock->send_pkt_list,
-				       struct virtio_vsock_pkt, list);
-		list_del_init(&pkt->list);
-		spin_unlock_bh(&vsock->send_pkt_list_lock);
 
-		virtio_transport_deliver_tap_pkt(pkt);
+		virtio_transport_deliver_tap_pkt(skb);
+		reply = virtio_vsock_skb_reply(skb);
 
-		reply = pkt->reply;
-
-		sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr));
+		sg_init_one(&hdr, virtio_vsock_hdr(skb), sizeof(*virtio_vsock_hdr(skb)));
 		sgs[out_sg++] = &hdr;
-		if (pkt->buf) {
-			sg_init_one(&buf, pkt->buf, pkt->len);
+		if (skb->len > 0) {
+			sg_init_one(&buf, skb->data, skb->len);
 			sgs[out_sg++] = &buf;
 		}
 
-		ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, pkt, GFP_KERNEL);
+		ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, skb, GFP_KERNEL);
 		/* Usually this means that there is no more space available in
 		 * the vq
 		 */
 		if (ret < 0) {
-			spin_lock_bh(&vsock->send_pkt_list_lock);
-			list_add(&pkt->list, &vsock->send_pkt_list);
-			spin_unlock_bh(&vsock->send_pkt_list_lock);
+			virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
 			break;
 		}
 
@@ -164,32 +153,32 @@ virtio_transport_send_pkt_work(struct work_struct *work)
 }
 
 static int
-virtio_transport_send_pkt(struct virtio_vsock_pkt *pkt)
+virtio_transport_send_pkt(struct sk_buff *skb)
 {
+	struct virtio_vsock_hdr *hdr;
 	struct virtio_vsock *vsock;
-	int len = pkt->len;
+	int len = skb->len;
+
+	hdr = virtio_vsock_hdr(skb);
 
 	rcu_read_lock();
 	vsock = rcu_dereference(the_virtio_vsock);
 	if (!vsock) {
-		virtio_transport_free_pkt(pkt);
+		kfree_skb(skb);
 		len = -ENODEV;
 		goto out_rcu;
 	}
 
-	if (le64_to_cpu(pkt->hdr.dst_cid) == vsock->guest_cid) {
-		virtio_transport_free_pkt(pkt);
+	if (le64_to_cpu(hdr->dst_cid) == vsock->guest_cid) {
+		kfree_skb(skb);
 		len = -ENODEV;
 		goto out_rcu;
 	}
 
-	if (pkt->reply)
+	if (virtio_vsock_skb_reply(skb))
 		atomic_inc(&vsock->queued_replies);
 
-	spin_lock_bh(&vsock->send_pkt_list_lock);
-	list_add_tail(&pkt->list, &vsock->send_pkt_list);
-	spin_unlock_bh(&vsock->send_pkt_list_lock);
-
+	virtio_vsock_skb_queue_tail(&vsock->send_pkt_queue, skb);
 	queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
 
 out_rcu:
@@ -201,9 +190,7 @@ static int
 virtio_transport_cancel_pkt(struct vsock_sock *vsk)
 {
 	struct virtio_vsock *vsock;
-	struct virtio_vsock_pkt *pkt, *n;
 	int cnt = 0, ret;
-	LIST_HEAD(freeme);
 
 	rcu_read_lock();
 	vsock = rcu_dereference(the_virtio_vsock);
@@ -212,20 +199,7 @@ virtio_transport_cancel_pkt(struct vsock_sock *vsk)
 		goto out_rcu;
 	}
 
-	spin_lock_bh(&vsock->send_pkt_list_lock);
-	list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
-		if (pkt->vsk != vsk)
-			continue;
-		list_move(&pkt->list, &freeme);
-	}
-	spin_unlock_bh(&vsock->send_pkt_list_lock);
-
-	list_for_each_entry_safe(pkt, n, &freeme, list) {
-		if (pkt->reply)
-			cnt++;
-		list_del(&pkt->list);
-		virtio_transport_free_pkt(pkt);
-	}
+	cnt = virtio_transport_purge_skbs(vsk, &vsock->send_pkt_queue);
 
 	if (cnt) {
 		struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
@@ -246,38 +220,28 @@ virtio_transport_cancel_pkt(struct vsock_sock *vsk)
 
 static void virtio_vsock_rx_fill(struct virtio_vsock *vsock)
 {
-	int buf_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE;
-	struct virtio_vsock_pkt *pkt;
-	struct scatterlist hdr, buf, *sgs[2];
+	int total_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE + VIRTIO_VSOCK_SKB_HEADROOM;
+	struct scatterlist pkt, *p;
 	struct virtqueue *vq;
+	struct sk_buff *skb;
 	int ret;
 
 	vq = vsock->vqs[VSOCK_VQ_RX];
 
 	do {
-		pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
-		if (!pkt)
+		skb = virtio_vsock_alloc_skb(total_len, GFP_KERNEL);
+		if (!skb)
 			break;
 
-		pkt->buf = kmalloc(buf_len, GFP_KERNEL);
-		if (!pkt->buf) {
-			virtio_transport_free_pkt(pkt);
+		memset(skb->head, 0, VIRTIO_VSOCK_SKB_HEADROOM);
+		sg_init_one(&pkt, virtio_vsock_hdr(skb), total_len);
+		p = &pkt;
+		ret = virtqueue_add_sgs(vq, &p, 0, 1, skb, GFP_KERNEL);
+		if (ret < 0) {
+			kfree_skb(skb);
 			break;
 		}
 
-		pkt->buf_len = buf_len;
-		pkt->len = buf_len;
-
-		sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr));
-		sgs[0] = &hdr;
-
-		sg_init_one(&buf, pkt->buf, buf_len);
-		sgs[1] = &buf;
-		ret = virtqueue_add_sgs(vq, sgs, 0, 2, pkt, GFP_KERNEL);
-		if (ret) {
-			virtio_transport_free_pkt(pkt);
-			break;
-		}
 		vsock->rx_buf_nr++;
 	} while (vq->num_free);
 	if (vsock->rx_buf_nr > vsock->rx_buf_max_nr)
@@ -299,12 +263,12 @@ static void virtio_transport_tx_work(struct work_struct *work)
 		goto out;
 
 	do {
-		struct virtio_vsock_pkt *pkt;
+		struct sk_buff *skb;
 		unsigned int len;
 
 		virtqueue_disable_cb(vq);
-		while ((pkt = virtqueue_get_buf(vq, &len)) != NULL) {
-			virtio_transport_free_pkt(pkt);
+		while ((skb = virtqueue_get_buf(vq, &len)) != NULL) {
+			consume_skb(skb);
 			added = true;
 		}
 	} while (!virtqueue_enable_cb(vq));
@@ -529,7 +493,7 @@ static void virtio_transport_rx_work(struct work_struct *work)
 	do {
 		virtqueue_disable_cb(vq);
 		for (;;) {
-			struct virtio_vsock_pkt *pkt;
+			struct sk_buff *skb;
 			unsigned int len;
 
 			if (!virtio_transport_more_replies(vsock)) {
@@ -540,23 +504,22 @@ static void virtio_transport_rx_work(struct work_struct *work)
 				goto out;
 			}
 
-			pkt = virtqueue_get_buf(vq, &len);
-			if (!pkt) {
+			skb = virtqueue_get_buf(vq, &len);
+			if (!skb)
 				break;
-			}
 
 			vsock->rx_buf_nr--;
 
 			/* Drop short/long packets */
-			if (unlikely(len < sizeof(pkt->hdr) ||
-				     len > sizeof(pkt->hdr) + pkt->len)) {
-				virtio_transport_free_pkt(pkt);
+			if (unlikely(len < sizeof(struct virtio_vsock_hdr) ||
+				     len > virtio_vsock_skb_len(skb))) {
+				kfree_skb(skb);
 				continue;
 			}
 
-			pkt->len = len - sizeof(pkt->hdr);
-			virtio_transport_deliver_tap_pkt(pkt);
-			virtio_transport_recv_pkt(&virtio_transport, pkt);
+			virtio_vsock_skb_rx_put(skb);
+			virtio_transport_deliver_tap_pkt(skb);
+			virtio_transport_recv_pkt(&virtio_transport, skb);
 		}
 	} while (!virtqueue_enable_cb(vq));
 
@@ -610,7 +573,7 @@ static int virtio_vsock_vqs_init(struct virtio_vsock *vsock)
 static void virtio_vsock_vqs_del(struct virtio_vsock *vsock)
 {
 	struct virtio_device *vdev = vsock->vdev;
-	struct virtio_vsock_pkt *pkt;
+	struct sk_buff *skb;
 
 	/* Reset all connected sockets when the VQs disappear */
 	vsock_for_each_connected_socket(&virtio_transport.transport,
@@ -637,23 +600,16 @@ static void virtio_vsock_vqs_del(struct virtio_vsock *vsock)
 	virtio_reset_device(vdev);
 
 	mutex_lock(&vsock->rx_lock);
-	while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_RX])))
-		virtio_transport_free_pkt(pkt);
+	while ((skb = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_RX])))
+		kfree_skb(skb);
 	mutex_unlock(&vsock->rx_lock);
 
 	mutex_lock(&vsock->tx_lock);
-	while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_TX])))
-		virtio_transport_free_pkt(pkt);
+	while ((skb = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_TX])))
+		kfree_skb(skb);
 	mutex_unlock(&vsock->tx_lock);
 
-	spin_lock_bh(&vsock->send_pkt_list_lock);
-	while (!list_empty(&vsock->send_pkt_list)) {
-		pkt = list_first_entry(&vsock->send_pkt_list,
-				       struct virtio_vsock_pkt, list);
-		list_del(&pkt->list);
-		virtio_transport_free_pkt(pkt);
-	}
-	spin_unlock_bh(&vsock->send_pkt_list_lock);
+	virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue);
 
 	/* Delete virtqueues and flush outstanding callbacks if any */
 	vdev->config->del_vqs(vdev);
@@ -690,8 +646,7 @@ static int virtio_vsock_probe(struct virtio_device *vdev)
 	mutex_init(&vsock->tx_lock);
 	mutex_init(&vsock->rx_lock);
 	mutex_init(&vsock->event_lock);
-	spin_lock_init(&vsock->send_pkt_list_lock);
-	INIT_LIST_HEAD(&vsock->send_pkt_list);
+	skb_queue_head_init(&vsock->send_pkt_queue);
 	INIT_WORK(&vsock->rx_work, virtio_transport_rx_work);
 	INIT_WORK(&vsock->tx_work, virtio_transport_tx_work);
 	INIT_WORK(&vsock->event_work, virtio_transport_event_work);

net/vmw_vsock/vsock_loopback.c

@@ -16,7 +16,7 @@ struct vsock_loopback {
 	struct workqueue_struct *workqueue;
 
 	spinlock_t pkt_list_lock; /* protects pkt_list */
-	struct list_head pkt_list;
+	struct sk_buff_head pkt_queue;
 	struct work_struct pkt_work;
 };
 
@@ -27,13 +27,13 @@ static u32 vsock_loopback_get_local_cid(void)
 	return VMADDR_CID_LOCAL;
 }
 
-static int vsock_loopback_send_pkt(struct virtio_vsock_pkt *pkt)
+static int vsock_loopback_send_pkt(struct sk_buff *skb)
 {
 	struct vsock_loopback *vsock = &the_vsock_loopback;
-	int len = pkt->len;
+	int len = skb->len;
 
 	spin_lock_bh(&vsock->pkt_list_lock);
-	list_add_tail(&pkt->list, &vsock->pkt_list);
+	skb_queue_tail(&vsock->pkt_queue, skb);
 	spin_unlock_bh(&vsock->pkt_list_lock);
 
 	queue_work(vsock->workqueue, &vsock->pkt_work);
@@ -44,21 +44,8 @@ static int vsock_loopback_send_pkt(struct virtio_vsock_pkt *pkt)
 static int vsock_loopback_cancel_pkt(struct vsock_sock *vsk)
 {
 	struct vsock_loopback *vsock = &the_vsock_loopback;
-	struct virtio_vsock_pkt *pkt, *n;
-	LIST_HEAD(freeme);
 
-	spin_lock_bh(&vsock->pkt_list_lock);
-	list_for_each_entry_safe(pkt, n, &vsock->pkt_list, list) {
-		if (pkt->vsk != vsk)
-			continue;
-		list_move(&pkt->list, &freeme);
-	}
-	spin_unlock_bh(&vsock->pkt_list_lock);
-
-	list_for_each_entry_safe(pkt, n, &freeme, list) {
-		list_del(&pkt->list);
-		virtio_transport_free_pkt(pkt);
-	}
+	virtio_transport_purge_skbs(vsk, &vsock->pkt_queue);
 
 	return 0;
 }
@@ -121,20 +108,18 @@ static void vsock_loopback_work(struct work_struct *work)
 {
 	struct vsock_loopback *vsock =
 		container_of(work, struct vsock_loopback, pkt_work);
-	LIST_HEAD(pkts);
+	struct sk_buff_head pkts;
+	struct sk_buff *skb;
+
+	skb_queue_head_init(&pkts);
 
 	spin_lock_bh(&vsock->pkt_list_lock);
-	list_splice_init(&vsock->pkt_list, &pkts);
+	skb_queue_splice_init(&vsock->pkt_queue, &pkts);
 	spin_unlock_bh(&vsock->pkt_list_lock);
 
-	while (!list_empty(&pkts)) {
-		struct virtio_vsock_pkt *pkt;
-
-		pkt = list_first_entry(&pkts, struct virtio_vsock_pkt, list);
-		list_del_init(&pkt->list);
-
-		virtio_transport_deliver_tap_pkt(pkt);
-		virtio_transport_recv_pkt(&loopback_transport, pkt);
+	while ((skb = __skb_dequeue(&pkts))) {
+		virtio_transport_deliver_tap_pkt(skb);
+		virtio_transport_recv_pkt(&loopback_transport, skb);
 	}
 }
 
@@ -148,7 +133,7 @@ static int __init vsock_loopback_init(void)
 		return -ENOMEM;
 
 	spin_lock_init(&vsock->pkt_list_lock);
-	INIT_LIST_HEAD(&vsock->pkt_list);
+	skb_queue_head_init(&vsock->pkt_queue);
 	INIT_WORK(&vsock->pkt_work, vsock_loopback_work);
 
 	ret = vsock_core_register(&loopback_transport.transport,
@@ -166,19 +151,13 @@ static int __init vsock_loopback_init(void)
 static void __exit vsock_loopback_exit(void)
 {
 	struct vsock_loopback *vsock = &the_vsock_loopback;
-	struct virtio_vsock_pkt *pkt;
 
 	vsock_core_unregister(&loopback_transport.transport);
 
 	flush_work(&vsock->pkt_work);
 
 	spin_lock_bh(&vsock->pkt_list_lock);
-	while (!list_empty(&vsock->pkt_list)) {
-		pkt = list_first_entry(&vsock->pkt_list,
-				       struct virtio_vsock_pkt, list);
-		list_del(&pkt->list);
-		virtio_transport_free_pkt(pkt);
-	}
+	virtio_vsock_skb_queue_purge(&vsock->pkt_queue);
 	spin_unlock_bh(&vsock->pkt_list_lock);
 
 	destroy_workqueue(vsock->workqueue);