[v2] hv_netvsc: Mark VF as slave before exposing it to user-mode

Message ID 1698440390-13719-1-git-send-email-longli@linuxonhyperv.com
State New
Headers
Series [v2] hv_netvsc: Mark VF as slave before exposing it to user-mode |

Commit Message

longli@linuxonhyperv.com Oct. 27, 2023, 8:59 p.m. UTC
  From: Long Li <longli@microsoft.com>

When a VF is being exposed form the kernel, it should be marked as "slave"
before exposing to the user-mode. The VF is not usable without netvsc running
as master. The user-mode should never see a VF without the "slave" flag.

This commit moves the code of setting the slave flag to the time before VF is
exposed to user-mode.

Signed-off-by: Long Li <longli@microsoft.com>
---

Change since v1:
Use a new function to handle NETDEV_POST_INIT.

 drivers/net/hyperv/netvsc_drv.c | 24 +++++++++++++++++-------
 1 file changed, 17 insertions(+), 7 deletions(-)
  

Comments

Jakub Kicinski Oct. 30, 2023, 9:25 p.m. UTC | #1
On Fri, 27 Oct 2023 13:59:50 -0700 longli@linuxonhyperv.com wrote:
> When a VF is being exposed form the kernel, it should be marked as "slave"
> before exposing to the user-mode. The VF is not usable without netvsc running
> as master. The user-mode should never see a VF without the "slave" flag.
> 
> This commit moves the code of setting the slave flag to the time before VF is
> exposed to user-mode.

Can you give a real example in the commit message of a flow in user
space which would get confused by seeing the VF netdev without
IFF_SLAVE?

You're only moving setting IFF_SLAVE but not linking the master,
is there no code which would assume that if SLAVE is set there 
is a master?
  
Long Li Oct. 30, 2023, 10:29 p.m. UTC | #2
> Subject: Re: [Patch v2] hv_netvsc: Mark VF as slave before exposing it to user-
> mode
> 
> On Fri, 27 Oct 2023 13:59:50 -0700 longli@linuxonhyperv.com wrote:
> > When a VF is being exposed form the kernel, it should be marked as "slave"
> > before exposing to the user-mode. The VF is not usable without netvsc
> > running as master. The user-mode should never see a VF without the "slave"
> flag.
> >
> > This commit moves the code of setting the slave flag to the time
> > before VF is exposed to user-mode.
> 
> Can you give a real example in the commit message of a flow in user space
> which would get confused by seeing the VF netdev without IFF_SLAVE?

A user-mode program may see the VF netdev show up without SLAVE flag before seeing the NETVSC netdev. It may try to configure the VF before it will be bonded to a NETVSC.

With the IFF_SLAVE correctly set at the time of VF showing up to the user-mode, it can rely on this flag to decide if this device should be ignored. (without implementing some timeout logic to detect a potential NETVSC device that may show up later)

> 
> You're only moving setting IFF_SLAVE but not linking the master, is there no
> code which would assume that if SLAVE is set there is a master?

The same (taking IFF_SLAVE without linking to master) is done in the original code before VF is joined, but it was for another purpose. I think there is a gap between when the VF is acted upon by other parts of the system and when it's bonded.
  

Patch

diff --git a/drivers/net/hyperv/netvsc_drv.c b/drivers/net/hyperv/netvsc_drv.c
index ec77fb9dcf89..fdad58dcc6a8 100644
--- a/drivers/net/hyperv/netvsc_drv.c
+++ b/drivers/net/hyperv/netvsc_drv.c
@@ -2206,9 +2206,6 @@  static int netvsc_vf_join(struct net_device *vf_netdev,
 		goto upper_link_failed;
 	}
 
-	/* set slave flag before open to prevent IPv6 addrconf */
-	vf_netdev->flags |= IFF_SLAVE;
-
 	schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
 
 	call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
@@ -2320,11 +2317,9 @@  static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
 	 */
 	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
 		ndev = hv_get_drvdata(ndev_ctx->device_ctx);
-		if (ether_addr_equal(vf_netdev->perm_addr, ndev->perm_addr)) {
-			netdev_notice(vf_netdev,
-				      "falling back to mac addr based matching\n");
+		if (ether_addr_equal(vf_netdev->perm_addr, ndev->perm_addr) ||
+		    ether_addr_equal(vf_netdev->dev_addr, ndev->perm_addr))
 			return ndev;
-		}
 	}
 
 	netdev_notice(vf_netdev,
@@ -2332,6 +2327,19 @@  static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
 	return NULL;
 }
 
+static int netvsc_prepare_slave(struct net_device *vf_netdev)
+{
+	struct net_device *ndev;
+
+	ndev = get_netvsc_byslot(vf_netdev);
+	if (!ndev)
+		return NOTIFY_DONE;
+
+	/* set slave flag before open to prevent IPv6 addrconf */
+	vf_netdev->flags |= IFF_SLAVE;
+	return NOTIFY_DONE;
+}
+
 static int netvsc_register_vf(struct net_device *vf_netdev)
 {
 	struct net_device_context *net_device_ctx;
@@ -2753,6 +2761,8 @@  static int netvsc_netdev_event(struct notifier_block *this,
 		return NOTIFY_DONE;
 
 	switch (event) {
+	case NETDEV_POST_INIT:
+		return netvsc_prepare_slave(event_dev);
 	case NETDEV_REGISTER:
 		return netvsc_register_vf(event_dev);
 	case NETDEV_UNREGISTER: