Merge branch 'hv_netvsc-prevent-packet-loss-during-vf-add-remove'

Long Li says:

====================
hv_netvsc: Prevent packet loss during VF add/remove

This patch set fixes issues with packet loss on VF add/remove.
====================

Link: https://lore.kernel.org/r/1610153623-17500-1-git-send-email-longli@linuxonhyperv.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

2 files changed, 43 insertions, 8 deletions

diff --git a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c
index 2350342b961f..3a3db2f0134d 100644
--- a/drivers/net/hyperv/netvsc.c
+++ b/drivers/net/hyperv/netvsc.c
@@ -37,6 +37,10 @@ void netvsc_switch_datapath(struct net_device *ndev, bool vf)
 	struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
 	struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
 
+	/* Block sending traffic to VF if it's about to be gone */
+	if (!vf)
+		net_device_ctx->data_path_is_vf = vf;
+
 	memset(init_pkt, 0, sizeof(struct nvsp_message));
 	init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
 	if (vf)
@@ -50,8 +54,11 @@ void netvsc_switch_datapath(struct net_device *ndev, bool vf)
 
 	vmbus_sendpacket(dev->channel, init_pkt,
 			       sizeof(struct nvsp_message),
-			       VMBUS_RQST_ID_NO_RESPONSE,
-			       VM_PKT_DATA_INBAND, 0);
+			       (unsigned long)init_pkt,
+			       VM_PKT_DATA_INBAND,
+			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
+	wait_for_completion(&nv_dev->channel_init_wait);
+	net_device_ctx->data_path_is_vf = vf;
 }
 
 /* Worker to setup sub channels on initial setup
@@ -754,8 +761,31 @@ static void netvsc_send_completion(struct net_device *ndev,
 				   const struct vmpacket_descriptor *desc,
 				   int budget)
 {
-	const struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
+	const struct nvsp_message *nvsp_packet;
 	u32 msglen = hv_pkt_datalen(desc);
+	struct nvsp_message *pkt_rqst;
+	u64 cmd_rqst;
+
+	/* First check if this is a VMBUS completion without data payload */
+	if (!msglen) {
+		cmd_rqst = vmbus_request_addr(&incoming_channel->requestor,
+					      (u64)desc->trans_id);
+		if (cmd_rqst == VMBUS_RQST_ERROR) {
+			netdev_err(ndev, "Invalid transaction id\n");
+			return;
+		}
+
+		pkt_rqst = (struct nvsp_message *)(uintptr_t)cmd_rqst;
+		switch (pkt_rqst->hdr.msg_type) {
+		case NVSP_MSG4_TYPE_SWITCH_DATA_PATH:
+			complete(&net_device->channel_init_wait);
+			break;
+
+		default:
+			netdev_err(ndev, "Unexpected VMBUS completion!!\n");
+		}
+		return;
+	}
 
 	/* Ensure packet is big enough to read header fields */
 	if (msglen < sizeof(struct nvsp_message_header)) {
@@ -763,6 +793,7 @@ static void netvsc_send_completion(struct net_device *ndev,
 		return;
 	}
 
+	nvsp_packet = hv_pkt_data(desc);
 	switch (nvsp_packet->hdr.msg_type) {
 	case NVSP_MSG_TYPE_INIT_COMPLETE:
 		if (msglen < sizeof(struct nvsp_message_header) +
diff --git a/drivers/net/hyperv/netvsc_drv.c b/drivers/net/hyperv/netvsc_drv.c
index f32f28311d57..75b4d6703cf1 100644
--- a/drivers/net/hyperv/netvsc_drv.c
+++ b/drivers/net/hyperv/netvsc_drv.c
@@ -539,7 +539,8 @@ static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
 	 */
 	vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
 	if (vf_netdev && netif_running(vf_netdev) &&
-	    netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net))
+	    netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net) &&
+	    net_device_ctx->data_path_is_vf)
 		return netvsc_vf_xmit(net, vf_netdev, skb);
 
 	/* We will atmost need two pages to describe the rndis
@@ -2381,12 +2382,15 @@ static int netvsc_register_vf(struct net_device *vf_netdev)
  * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
  * interface, there is only the CHANGE event and no UP or DOWN event.
  */
-static int netvsc_vf_changed(struct net_device *vf_netdev)
+static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
 {
 	struct net_device_context *net_device_ctx;
 	struct netvsc_device *netvsc_dev;
 	struct net_device *ndev;
-	bool vf_is_up = netif_running(vf_netdev);
+	bool vf_is_up = false;
+
+	if (event != NETDEV_GOING_DOWN)
+		vf_is_up = netif_running(vf_netdev);
 
 	ndev = get_netvsc_byref(vf_netdev);
 	if (!ndev)
@@ -2399,7 +2403,6 @@ static int netvsc_vf_changed(struct net_device *vf_netdev)
 
 	if (net_device_ctx->data_path_is_vf == vf_is_up)
 		return NOTIFY_OK;
-	net_device_ctx->data_path_is_vf = vf_is_up;
 
 	netvsc_switch_datapath(ndev, vf_is_up);
 	netdev_info(ndev, "Data path switched %s VF: %s\n",
@@ -2716,7 +2719,8 @@ static int netvsc_netdev_event(struct notifier_block *this,
 	case NETDEV_UP:
 	case NETDEV_DOWN:
 	case NETDEV_CHANGE:
-		return netvsc_vf_changed(event_dev);
+	case NETDEV_GOING_DOWN:
+		return netvsc_vf_changed(event_dev, event);
 	default:
 		return NOTIFY_DONE;
 	}