[net] hv_netvsc: fix deadlock on hotplug

Message ID 20170906211114.4394-1-sthemmin@microsoft.com
State Superseded
Delegated to: David Miller
Headers show
Series
  • [net] hv_netvsc: fix deadlock on hotplug
Related show

Commit Message

Stephen Hemminger Sept. 6, 2017, 9:11 p.m.
When a virtual device is added dynamically (via host console), then
the vmbus sends an offer message for the primary channel. The processing
of this message for networking causes the network device to then
initialize the sub channels.

The problem is that setting up the sub channels needs to wait until
the subsequent subchannel offers have been processed. These offers
come in on the same ring buffer and work queue as where the primary
offer is being processed; leading to a deadlock.

This did not happen in older kernels, because the sub channel waiting
logic was broken (it wasn't really waiting).

The solution is to do the sub channel setup in its own work queue
context that is scheduled by the primary channel setup; and then
happens later.

Fixes: 732e49850c5e ("netvsc: fix race on sub channel creation")
Reported-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
---
This is version for 4.13 (linux-net)

 drivers/net/hyperv/hyperv_net.h   |   3 +
 drivers/net/hyperv/netvsc.c       |   3 +
 drivers/net/hyperv/netvsc_drv.c   |  18 +-----
 drivers/net/hyperv/rndis_filter.c | 118 +++++++++++++++++++++++++++-----------
 4 files changed, 94 insertions(+), 48 deletions(-)

Patch

diff --git a/drivers/net/hyperv/hyperv_net.h b/drivers/net/hyperv/hyperv_net.h
index 12cc64bfcff8..b82119be14e9 100644
--- a/drivers/net/hyperv/hyperv_net.h
+++ b/drivers/net/hyperv/hyperv_net.h
@@ -200,6 +200,8 @@  int netvsc_recv_callback(struct net_device *net,
 			 const struct ndis_pkt_8021q_info *vlan);
 void netvsc_channel_cb(void *context);
 int netvsc_poll(struct napi_struct *napi, int budget);
+
+void rndis_set_subchannel(struct work_struct *w);
 int rndis_filter_open(struct netvsc_device *nvdev);
 int rndis_filter_close(struct netvsc_device *nvdev);
 int rndis_filter_device_add(struct hv_device *dev,
@@ -766,6 +768,7 @@  struct netvsc_device {
 	u32 num_chn;
 
 	atomic_t open_chn;
+	struct work_struct subchan_work;
 	wait_queue_head_t subchan_open;
 
 	struct rndis_device *extension;
diff --git a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c
index d18c3326a1f7..76ef14cf55a1 100644
--- a/drivers/net/hyperv/netvsc.c
+++ b/drivers/net/hyperv/netvsc.c
@@ -79,6 +79,7 @@  static struct netvsc_device *alloc_net_device(void)
 	net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
 	init_completion(&net_device->channel_init_wait);
 	init_waitqueue_head(&net_device->subchan_open);
+	INIT_WORK(&net_device->subchan_work, rndis_set_subchannel);
 
 	return net_device;
 }
@@ -553,6 +554,8 @@  void netvsc_device_remove(struct hv_device *device)
 	struct netvsc_device *net_device = net_device_ctx->nvdev;
 	int i;
 
+	cancel_work_sync(&net_device->subchan_work);
+
 	netvsc_disconnect_vsp(device);
 
 	RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
diff --git a/drivers/net/hyperv/netvsc_drv.c b/drivers/net/hyperv/netvsc_drv.c
index d91cbc6c3ca4..9c8dd86a663e 100644
--- a/drivers/net/hyperv/netvsc_drv.c
+++ b/drivers/net/hyperv/netvsc_drv.c
@@ -740,24 +740,13 @@  static int netvsc_set_queues(struct net_device *net, struct hv_device *dev,
 			     u32 num_chn)
 {
 	struct netvsc_device_info device_info;
-	int ret;
 
 	memset(&device_info, 0, sizeof(device_info));
 	device_info.num_chn = num_chn;
 	device_info.ring_size = ring_size;
 	device_info.max_num_vrss_chns = num_chn;
 
-	ret = rndis_filter_device_add(dev, &device_info);
-	if (ret)
-		return ret;
-
-	ret = netif_set_real_num_tx_queues(net, num_chn);
-	if (ret)
-		return ret;
-
-	ret = netif_set_real_num_rx_queues(net, num_chn);
-
-	return ret;
+	return rndis_filter_device_add(dev, &device_info);
 }
 
 static int netvsc_set_channels(struct net_device *net,
@@ -1573,8 +1562,6 @@  static int netvsc_probe(struct hv_device *dev,
 
 	/* RCU not necessary here, device not registered */
 	nvdev = net_device_ctx->nvdev;
-	netif_set_real_num_tx_queues(net, nvdev->num_chn);
-	netif_set_real_num_rx_queues(net, nvdev->num_chn);
 
 	/* MTU range: 68 - 1500 or 65521 */
 	net->min_mtu = NETVSC_MTU_MIN;
@@ -1616,11 +1603,10 @@  static int netvsc_remove(struct hv_device *dev)
 	 * removed. Also blocks mtu and channel changes.
 	 */
 	rtnl_lock();
+	unregister_netdevice(net);
 	rndis_filter_device_remove(dev, ndev_ctx->nvdev);
 	rtnl_unlock();
 
-	unregister_netdev(net);
-
 	hv_set_drvdata(dev, NULL);
 
 	free_netdev(net);
diff --git a/drivers/net/hyperv/rndis_filter.c b/drivers/net/hyperv/rndis_filter.c
index d6308ffda53e..015a8e2b733c 100644
--- a/drivers/net/hyperv/rndis_filter.c
+++ b/drivers/net/hyperv/rndis_filter.c
@@ -1037,8 +1037,6 @@  static void netvsc_sc_open(struct vmbus_channel *new_sc)
 
 	/* Set the channel before opening.*/
 	nvchan->channel = new_sc;
-	netif_napi_add(ndev, &nvchan->napi,
-		       netvsc_poll, NAPI_POLL_WEIGHT);
 
 	ret = vmbus_open(new_sc, nvscdev->ring_size * PAGE_SIZE,
 			 nvscdev->ring_size * PAGE_SIZE, NULL, 0,
@@ -1046,12 +1044,90 @@  static void netvsc_sc_open(struct vmbus_channel *new_sc)
 	if (ret == 0)
 		napi_enable(&nvchan->napi);
 	else
-		netif_napi_del(&nvchan->napi);
+		netdev_notice(ndev, "sub channel open failed: %d\n", ret);
 
 	atomic_inc(&nvscdev->open_chn);
 	wake_up(&nvscdev->subchan_open);
 }
 
+
+/* Open sub-channels after completing the handling of the device probe.
+ * This breaks overlap of processing the host message for the
+ * new primary channel with the initialization of sub-channels.
+ */
+void rndis_set_subchannel(struct work_struct *w)
+{
+	struct netvsc_device *nvdev
+		= container_of(w, struct netvsc_device, subchan_work);
+	struct nvsp_message *init_packet = &nvdev->channel_init_pkt;
+	struct net_device_context *ndev_ctx;
+	struct rndis_device *rdev;
+	struct net_device *ndev;
+	struct hv_device *hv_dev;
+	int i, ret;
+
+	if (!rtnl_trylock()) {
+		schedule_work(w);
+		return;
+	}
+
+	rdev = nvdev->extension;
+	if (!rdev)
+		goto unlock;	/* device was removed */
+
+	ndev = rdev->ndev;
+	ndev_ctx = netdev_priv(ndev);
+	hv_dev = ndev_ctx->device_ctx;
+
+	memset(init_packet, 0, sizeof(struct nvsp_message));
+	init_packet->hdr.msg_type = NVSP_MSG5_TYPE_SUBCHANNEL;
+	init_packet->msg.v5_msg.subchn_req.op = NVSP_SUBCHANNEL_ALLOCATE;
+	init_packet->msg.v5_msg.subchn_req.num_subchannels =
+						nvdev->num_chn - 1;
+	ret = vmbus_sendpacket(hv_dev->channel, init_packet,
+			       sizeof(struct nvsp_message),
+			       (unsigned long)init_packet,
+			       VM_PKT_DATA_INBAND,
+			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
+	if (ret) {
+		netdev_err(ndev, "sub channel allocate send failed: %d\n", ret);
+		goto failed;
+	}
+
+	wait_for_completion(&nvdev->channel_init_wait);
+	if (init_packet->msg.v5_msg.subchn_comp.status != NVSP_STAT_SUCCESS) {
+		netdev_err(ndev, "sub channel request failed\n");
+		goto failed;
+	}
+
+	nvdev->num_chn = 1 +
+		init_packet->msg.v5_msg.subchn_comp.num_subchannels;
+
+	/* wait for all sub channels to open */
+	wait_event(nvdev->subchan_open,
+		   atomic_read(&nvdev->open_chn) == nvdev->num_chn);
+
+	/* ignore failues from setting rss parameters, still have channels */
+	rndis_filter_set_rss_param(rdev, netvsc_hash_key,
+				   nvdev->num_chn);
+
+	netif_set_real_num_tx_queues(ndev, nvdev->num_chn);
+	netif_set_real_num_rx_queues(ndev, nvdev->num_chn);
+
+	rtnl_unlock();
+	return;
+
+failed:
+	/* fallback to only primary channel */
+	for (i = 1; i < nvdev->num_chn; i++)
+		netif_napi_del(&nvdev->chan_table[i].napi);
+
+	nvdev->max_chn = 1;
+	nvdev->num_chn = 1;
+unlock:
+	rtnl_unlock();
+}
+
 int rndis_filter_device_add(struct hv_device *dev,
 			    struct netvsc_device_info *device_info)
 {
@@ -1061,7 +1137,6 @@  int rndis_filter_device_add(struct hv_device *dev,
 	struct rndis_device *rndis_device;
 	struct ndis_offload hwcaps;
 	struct ndis_offload_params offloads;
-	struct nvsp_message *init_packet;
 	struct ndis_recv_scale_cap rsscap;
 	u32 rsscap_size = sizeof(struct ndis_recv_scale_cap);
 	unsigned int gso_max_size = GSO_MAX_SIZE;
@@ -1226,37 +1301,16 @@  int rndis_filter_device_add(struct hv_device *dev,
 
 	vmbus_set_sc_create_callback(dev->channel, netvsc_sc_open);
 
-	init_packet = &net_device->channel_init_pkt;
-	memset(init_packet, 0, sizeof(struct nvsp_message));
-	init_packet->hdr.msg_type = NVSP_MSG5_TYPE_SUBCHANNEL;
-	init_packet->msg.v5_msg.subchn_req.op = NVSP_SUBCHANNEL_ALLOCATE;
-	init_packet->msg.v5_msg.subchn_req.num_subchannels =
-						net_device->num_chn - 1;
-	ret = vmbus_sendpacket(dev->channel, init_packet,
-			       sizeof(struct nvsp_message),
-			       (unsigned long)init_packet,
-			       VM_PKT_DATA_INBAND,
-			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
-	if (ret)
-		goto out;
 
-	wait_for_completion(&net_device->channel_init_wait);
-	if (init_packet->msg.v5_msg.subchn_comp.status != NVSP_STAT_SUCCESS) {
-		ret = -ENODEV;
-		goto out;
-	}
+	for (i = 1; i < net_device->num_chn; i++)
+		netif_napi_add(net, &net_device->chan_table[i].napi,
+			       netvsc_poll, NAPI_POLL_WEIGHT);
 
-	net_device->num_chn = 1 +
-		init_packet->msg.v5_msg.subchn_comp.num_subchannels;
-
-	/* wait for all sub channels to open */
-	wait_event(net_device->subchan_open,
-		   atomic_read(&net_device->open_chn) == net_device->num_chn);
+	if (net_device->num_chn > 1)
+		schedule_work(&net_device->subchan_work);
 
-	/* ignore failues from setting rss parameters, still have channels */
-	rndis_filter_set_rss_param(rndis_device, netvsc_hash_key,
-				   net_device->num_chn);
 out:
+	/* if unavailable, just proceed with one queue */
 	if (ret) {
 		net_device->max_chn = 1;
 		net_device->num_chn = 1;
@@ -1277,10 +1331,10 @@  void rndis_filter_device_remove(struct hv_device *dev,
 	/* Halt and release the rndis device */
 	rndis_filter_halt_device(rndis_dev);
 
-	kfree(rndis_dev);
 	net_dev->extension = NULL;
 
 	netvsc_device_remove(dev);
+	kfree(rndis_dev);
 }
 
 int rndis_filter_open(struct netvsc_device *nvdev)